{"Bibliographic":{"Title":"Proceedings of the Fourteenth Annual Climate Diagnostics Workshop : held at the Scripps Institute of Oceanography, University of California, San Diego, La Jolla, California, October 16-20, 1989.","Authors":"","Publication date":"1990","Publisher":""},"Administrative":{"Date created":"08-20-2023","Language":"English","Rights":"CC 0","Size":"0000872800"},"Pages":["Proceedings of the\nOF\nCOMMUNITY\nFourteenth Annual\nwith\nAvenue\nClimate Diagnostics\nSTATES\nOF\nWorkshop\nScripps Institution of Oceanography\nUniversity of California, San Diego\nLa Jolla, California\nOctober 16-20, 1989\nProperty of\nNOAA Miami Library\n#301 Rickenbacker Causeway\n33149\nMiami, Florida\nU.S. DEPARTMENT OF COMMERCE\nNational Oceanic and Atmospheric Administration\nNational Weather Service\nClimate Analysis Center / NMC","NOAA MIAMI BRANCH LIBRARY\nOAOO QC980.C3 1989\nClimate Diag Proceedings of the Fourtee\n3 9282 0000 8260 1\nPublications listed below are available in both hard copy and\nform from the National Technical Information Service, U.S. Department of\nCommerce, Sills Building, 5285 Port Royal Road, Springfield, VA 22161. Prices\non request. Order by accession number (given in parentheses). When articles\nin\nany of the Workshops are referenced, please include the appropriate number at\nthe end of the reference. This allows various journals to accept these\nreferences more readily.\nProceedings of the First Climate Diagnostics Workshop\n(PB264993)\nProceedings of the Second Climate Diagnostics Workshop\n(PB282151)\nProceedings of the Third Climate Diagnostics Workshop\n(PB298355)\nProceedings of the Fourth Climate Diagnostics Workshop\n(PB80-201130)\nProceedings of the Fifth Climate Diagnostics Workshop\n(PB81-222200)\nProceedings of the Sixth Climate Diagnostics Workshop\n(PB82-219486)\nProceedings of the Seventh Climate Diagnostics Workshop\n(PB83-208033)\nProceedings of the Eighth Climate Diagnostics Workshop\n(PB84-192418)\nProceedings of the Ninth Climate Diagnostics Workshop\n(PB85-183911)\nProceedings of the Tenth\nkshop\n*\nProceeding\nlorkshop\n(PB87-175741)\nProceeding\nkshop\n(PB88-191747)\nProceedin\nWorkshop\n(PB89-178115)\n*\nProceedings\np was combined with the\nFirst WMO\nof Monthly and Seasonal\nVariations\nnge Forecasting Research\nReport Seri\nJanuary 1986.\n,","QC\n980\nC3\n1989\nAND NOAA ATMOSPHERIC\nProceedings of the\nFourteenth Annual\nClimate Diagnostics\nOF\nWorkshop\nHeld at the\nScripps Institution of Oceanography\nUniversity of California, San Diego\nLa Jolla, California\nOctober 16-20, 1989\nMarch 1990\nProperty of\nNOAA Miami Library\n#301 Rickenbacker Causeway\nMiami, Florida 33149\n008028\nU.S. DEPARTMENT OF COMMERCE\nRobert A. Mosbacher, Secretary\nNational Oceanic and Atmospheric Administration\nJohn A. Knauss, Under Secretary\nNational Weather Service\nDr. Elbert W. Friday, Jr., Assistant Administrator","","Fourteenth Annual\nClimate Diagnostics Workshop\nLa Jolla, CA\nOctober 16-20, 1989\nPREFACE\nThe Fourteenth Annual Climate Diagnostics Workshop was hosted by the\nExperimental Forecast Center at the Scripps Institution of Oceanography, in\nLa Jolla, CA. The success of the Workshop was enhanced by the work of John\nRoads and the Scripps organizing committee, who with the Climate Analysis\nCenter jointly planned and organized the Workshop. Thanks to the Session\nChairmen: Nick Graham, Vern Kousky, Phil Arkin, Gene Rasmusson, Tim\nBarnett, Chet Ropelewski, Art Miller, Ed Epstein, Bob Livezey, Richard\nSommerville, Tony Barnston, Shyh-Chin Chen and Dan Cayan for their effective\nand enthusiastic management of the sessions. The excellent facilities,\nlocal arrangements, and hospitality provided by the Scripps staff made the\nWorkshop a pleasant and intellectually stimulating experience for all\nparticipants.\nA highlight of the meeting was the talk by Professor Edward Lorenz on\n\"Climate Change as Related to Chaos\" at the evening dinner for the Workshop\nparticipants.\nAt the Climate Analysis Center, John Janowiak served as Chairman of the\norganizing committee; Gail Lucas and Elaine Michaelides provided secretarial\nservices; and Luke Mannello served as editor/compiler of the manuscripts\nand was responsible for the timely publication of these Proceedings.\nThese Proceedings provide a published record of the presentations made\nat the Workshop. This Workshop had sessions on the following topics:\nobserved atmospheric responses to the 1986-88 Pacific warm/cold evolution,\nnumerical modeling of the ENSO phenomenon, atmospheric oscillations,\natmospheric variability, ocean-atmosphere interaction from an observational\nperspective, climate monitoring, climatic variability from a modeling\nperspective, prediction on climatic time scales, aspects of predictability\non climatic time scales, long-range prediction activities at operational\nforecast centers, and regional climatic studies.\nI would like to sincerely thank all participants for their\ncontributions and their participation in another successful Workshop.\nDavid R. Rodenhuis, Director\nClimate Analysis Center\nJanuary 1990\niii","","A NICE MIXTURE OF\nBUSINESS AND\nLA JOLLA","vi","List of Attendees to the\nFourteenth Climate Diagnostics Workshop\nLa Jolla, CA\nOctober 16-20, 1989\nPatricio Aceituno\nUniversidad de Chile\nGuillermo F. Almeyda\nDesert Research Institute - Reno, NV\nPhillip A. Arkin\nNOAA/NWS/CAC\nTimothy P. Barnett\nScripps Institution of Oceanography\nTony Barnston\nNOAA/NWS/CAC\nDave Baumhefner\nNCAR - Boulder, CO\nGeorge J. Boer\nCanadian Climate Centre\nArthur Brown\nScripps Institution of Oceanography\nDan Cayan\nScripps Institution of Oceanography\nDouglas Chan\nAtmospheric Environment Service - Canada\nFung-Chiau Chang\nLaboratory for Atmospheres/NASA\nMuthuvel Chelliah\nNOAA/NWS/CAC\nShyh-Chin Chen\nScripps Institution of Oceanography\nTsing-Chang Chen\nIowa State University\nWen-Yuan Chen\nNOAA/NWS/CAC\nBeth Chertock\nNOAA/ERL/WPL - Boulder, CO\nJohn R. Christy\nUniversity of Alabama\nPao-Shin Chu\nUniversity of Hawaii\nKenneth F. Dewey\nUniversity of Nebraska\nArthur V. Douglas\nCreighton University\nHugh W. Ellasaesser\nLawrence Livermore National Laboratory\nEdward S. Epstein\nNOAA/NWS/CAC\nFrancis M. Fujioka\nU.S. Forest Service/USDA\nKenneth Gage\nAeronomy Laboratory, NOAA/ERL\nLawrence Gates\nLawrence Livermore National Laboratory\nMichael Ghil\nUniversity of California\nDonald Gilman\nNOAA/NWS/CAC (retired)\nSteven J. Goodman\nES44/NASA - Huntsville, AL\nNicholas E. Graham\nScripps Institution of Oceanography\nStanley L. Grotch\nLawrence Livermore National Laboratory\nGeorgiy V. Gruza\nHydromet Center, USSR\nJean-Francois Gueremy\nFlorida State University\nDavid S. Gutzler\nAtmospheric and Environmental Research, Inc.\nKenneth Hadeen\nNOAA/NESDIS/NCDC\nMichael Halpert\nNWS/NWS/CAC\nJohn Hanzis\nMitre Corporation\nMike Harrison\nMET Office - Berkshire, U.K.\nGary A. Herbert\nNOAA Research Lab-Boulder, CO\nManfred Holl\nCenter for Analysis and Prediction, Monterey\nJohn D. Horel\nUniversity of Utah\nJohn Janowiak\nNOAA/NWS/CAC\nC-Y Jim Kao\nLos Alamos National Laboratory\nWilliam H. Klein\nUniversity of Maryland\nVernon Kousky\nNOAA/NWS/CAC\nGeorge Kukla\nLamont Observatory\nYochanan Kushnir\nNOAA/GFDL\nWilliam K. Lau\nNASA/Goddard Space Flight Center\nSid Levitus\nPrinceton University, NJ\nBrant Liebmann\nCIRES/Boulder, CO\nRichard S. Lindzen\nMassachusetts Institute of Technology\nvii","Robert E. Livezey\nNOAA/NWS/CAC\nEdward N. Lorenz\nMassachusetts Institute of Technology\nDale A. Lowry\nNOAA/NCPO\nUte Luksch\nMax-Planck Institute fur Meteorologie\nJose A. Maliekal\nUniversity of Hawaii\nLeo Mapanao\nOntario, Canada\nJohn R. McAfee\nNOAA/ERL - Boulder, CO\nMorris McCutchan\nForest Service, USDA\nCarlos R. Mechoso\nUniversity of California\nGerald Meehl\nNCAR - Boulder, CO\nArthur J. Miller\nScripps Institution of Technology\nKingtse C. Mo\nNOAA/NWS/CAC\nMark L. Morrissey\nUniversity of Hawaii\nSubramaniam Moten\nFlorida State University\nTom Murphree\nNaval Post Graduate School, Monterey\nJerome Namias\nScripps Institution of Oceanography\nDavid Neelin\nDepartment of Atmospheric Sciences\nReginald E. Newell\nMassachusetts Institute of Technology\nSharon E. Nicholson\nFlorida State University\nArthur O'Brien\nMitre Corporation\nEdward O'Lenic\nNOAA/NWS/CAC\nJan Paegle\nUniversity of Utah\nChung-Kyu Park\nNASA/Goddard Space Flight Center\nTom Phillips\nScripps Institution of Oceanography\nYakov P. Popov\nHydromet Center, USSR\nDavid Portman\nAtmospheric and Environmental Research, Inc.\nV. T. Radiuhin\nWorld Data Center-B, USSR\nEugene M. Rasmusson\nUniversity of Maryland\nViacheslav N. Razuvayev\nWorld Data Center-B, USSR\nGeorge C. Reid\nAeronomy Laboratory, NOAA\nMichael B. Richmond\nIllinois State Water Survey\nJohn 0. Roads\nScripps Institution of Oceanography\nMargaret K. Robinson\nCompass Systems, Inc.\nJeffrey C. Rogers\nOhio State University\nChester F. Ropelewski\nNOAA/NWS/CAC\nRichard D. Rosen\nAtmospheric and Environmental Research, Inc.\nDavid P. Rowell\nMET Office - Berkshire, U.K.\nDavid A. Salstein\nAtmospheric and Environmental Research, Inc.\nNeil Sargent\nCanadian Climate Centre\nJae-Kyung E. Schemm\nNASA/Goddard Space Flight Center\nFrederick Semazzi\nNational Science Foundation\nDennis J. Shea\nNCAR - Boulder, CO\nJoseph Sirutis\nPrinceton University, NJ\nRichard C. Somerville\nScripps Institution of Oceanography\nKenneth R. Sperber\nLawrence Livermore National Laboratory\nWilliam A. Sprigg\nNOAA Climate Office - Rockville, MD\nDuane E. Stevens\nColorado State University\nY. C. Sud\nNASA/Goddard Space Flight Center\nKyozo Ueyoshi\nScripps Institution of Oceanography\nSushel Unninayar\nNational Science Foundation\nHuug van den Dool\nUniversity of Maryland\nA. James Wagner\nNOAA/NWS/CAC\nX. L. Wang\nUniversity of Maryland\nKlaus Wolter\nCIRES - Boulder, CO\nStanley F. Woronko\nCanadian Climate Centre\nviii","14th Annual Climate Diagnostics Workshop\nScripps Institution of Oceanography\nLa Jolla, California\nOctober 16 - 20, 1989\nPROGRAM\nSESSION 1. Observed Atmospheric Responses to the 1986-88\nPacific Warm/Cold Evolution.\nChair: D. R. Rodenhuis\nPage\nAtmospheric circulation changes associated with extremes of the Southern\n1\nVern E. Kousky and C. F. Ropelewski\nOscillation during 1986-1989\nVariations in tropical rainfall during 1986-88, as inferred from satellite\n8\nobservations of cloud-top temperature\nJohn E. Janowiak, P. A. Arkin, and D. Davidowicz\nComparison of the zonal propogation of equatorial OLR fields during the 3\n12\nJose A. Maliekal, M. A. Lander, M. L. Morrissey\nrecent E1 Niños\nMichael Ghil\n18\nMultiple flow regimes and ENSO\nSESSION 2. Numerical Modeling of the ENSO Phenomenon\nChair: Nick Graham\nDavid Neelin\n22\nBifurcations and E1 Niño in a hybrid coupled GCM\nSymmetry and asymmetry in the modelled atmospheric response to E1 Niño and\n*\nGeorge J. Boer\nLa Niña\nComparison of 30 years of data with Robinson climatology clearly distin-\n28\nMargaret K. Robinson\nguishes E1 Niño events\nForecast experiments of the 1982 - 83 E1 Niño with a coupled air-sea model\n35\nJoseph Sirutis and K. Miyakoda\nENSO and CO2 climate change in a coupled ocean-atmosphere GCM\n41\nGerald A. Meehl\nBoundary layer development over the southeast Pacific Ocean under ENSO and\n47\nGuillermo F. Almeyda, J. W. Tellford and S. K. Chai\nanti-ENSO episodes.\nSouthern Oscillation in the OSU coupled upper ocean/atmosphere GCM.\n55\nKenneth R. Sperber, S. Hameed, W. L. Gates and G. L. Potter\nPresentation was made at the Workshop; however, paper was not submitted.\nix","SESSION 3.\nAtmospheric Oscillations\nChair: Vern Kousky\nPage\nVector index of the North Atlantic Oscillation and its association with the\n61\nsurface air temperature and circulation\nAndrei V. Nemirovsky and V. T. Radiuhin\nMomentum exchange among atmosphere, ocean, and solid earth on 40 - 50 day\n67\ntime scales\nDavid S. Gutzler and R. M. Ponte\nHeat and moisture fluxes on the time scale of 20 - 60 days over the Indian\n73\nJean-Francois Gueremy\nmonsoon area\nOnset of the Australian summer monsoon and associations with the 40-50 day\n78\noscillation\nBrant Liebmann, H. Hendon and G. Holland\nSESSION 4.\nAtmospheric Variability: Tropics\nChair: Phil Arkin\nBiennial variability in tropical SST and winds\n84\nChester F. Ropelewski, M. S. Halpert and E. M. Rasmusson\nENSO-related variability as revealed by singular spectrum analysis (SSA)\n91\nXueliang Wang and E. M. Rasmusson\nOn the interaction of multiple time scales in the tropical climate system\n96\nTim P. Barnett\nTropical ocean-atmosphere interactions at the intraseasonal time scale.\n102\nDuane E. Stevens and Q. Hu\nRelationship between summertime ridge position and continental rainfall\n108\nregimes, 1929-1988\nArthur V. Douglas and P. J. Englehart\nDo the upper air data for the United States indicate CO2 induced\nwarming?\n115\nJ. Gavin, G. Kukla and T. Karl\nDescriptive statistics of surface winds in the core of the equatorial\n121\nwestern Pacific\nJames Frederick and P-S. Chu\nImpact of the diurnal cycle upon intraseasonal fluctuations in the tropics\n127\nA. N. Hahmann, J. D. Horel and J. Paegle\nGlobal tropical forcing of the extratropical planetary wave field\n#A-1\nTom Murphree\nIntraseasonal oscillations in the upper troposphere observed with the\n132\nChristmas Island wind profiler\nJohn R. McAfee,\nK. S. Gage, B. B. Balsley, W. L. Ecklund, D. A. Carter and G. C. Reid\n# Manuscript arrived late; see Appendix.","SESSION 5.\nAtmospheric Variability: Global Scale\nChair: E. M. Rasmusson\nPage\nThe sensitivity of the high-latitude response to tropical heating anomalies\n#A-7\nin real-data extended range forecasts\nLawrence E. Buja, J. Paegle, and D. P. Baumhefner\nOn the annual variation of subtropical jet streams\n138\nTsing-Chang Chen and M-C. Yen\nPossible factors controlling global night marine air temperature variations\n145\nover the past century\nReginald E. Newell, W. Zhongxiang and J. Hsiung\nAtlantic sea level pressure teleconnections and associated storm track and\n151\nclimate variability\nJeffrey C. Rogers\nLow frequency variability of subtropical jets.\n156\nChung-Kyu Park and S. Schubert\nClimate variability as measured by observations of lightning activity\n164\nSteven J. Goodman\nGlobal monitoring of net solar irradiance at the ocean surface and valida-\n172\ntion of General Circulation Model simulations\nBeth Chertock\nNatural variability in summertime droughts over the United States Great\n178\nPlains\nFong-Chiau Chang and K-M. Lau\nSESSION 6. Ocean-Atmosphere Interaction from an\nObservational Perspective\nChair: Tim Barnett\nSeasonal relations between tropical Pacific SSTs and Northern Hemisphere\n184\n700 mb heights\nNicholas E. Graham\nVariability of latent and sensible heat fluxes over the northern oceans\n192\nDaniel R. Cayan\nUnstable modes in coupled ocean - atmosphere interactions in intraseasonal\n198\nand interannual time scales\nWilliam K-M. Lau, and A. Hirst\nTemporal variability of the thermohaline structure of the North Atlantic\n205\nOcean\nSidney Levitus\n# Manuscript arrived late; see Appendix.\nxi","SESSION 7.\nClimate Monitoring\nChair: Chet Ropelewski\nPage\nThe development and utilization of an AVHRR - based vegetation index for\n211\nclimate monitoring\nMichael S. Halpert, P. A. Arkin, C. F. Ropelewski and R. Tomlinson\nInterannual variability and trends indicated by OLR\n216\nMuthuvel Chelliah and P. A. Arkin\nVerification of satellite-based rainfall estimates for climate models\n222\nMark L. Morrissey and J. S. Greene\nTrimmed COADS - problems and solutions\nKlaus Wolter\n228\nGlobal tropospheric temperature monitoring with passive microwave radio-\n234\nmeters\nRoy W. Spencer and J. R. Christy\nGlobal tropospheric temperature variations since 1979 as measured from the\n240\nmicrowave sounding unit\nJohn R. Christy and R. W. Spencer\nSESSION 8. Climatic Variability from a Modeling Perspective\nChair: Art Miller\nA GCM study on the 1988 U.S. drought\n245\nKingtse Mo, J. R. Zimmerman, E. Kalnay and M. Kanamitsu\nComparisons of model-produced diabatic heating component fields\n250\nDavid A. Salstein and R. D. Rosen\nZero potential vorticity gradient basic states in the neighborhood of the\n256\nequator\nRichard S. Lindzen, V. Wirth and S. Nigam\nModel simulation of atmospheric circulation and rainfall during June 1988\n260\nand the low-frequency vacillations in a perpetual run\nY. C. Sud, W. K. - M Lau, G. K. Walker and W. E. Smith\nA 3-D dynamical/chemical GCM for simulating the anthropogenical effects on\n265\nC. Y. Jim Kao, G. A. Glatzmaier, R. C. Malone and R. P. Turco\nozone\nPreliminary coupling experiments of an atmospheric and an oceanic GCM\n271\nM. Fisher, M. Ghil, D. Halpern, C. Mechoso and J. Spahr\nWinter simulation in a simplified NMC 4-layer global spectral model\n277\nShyh-Chin Chen and J. 0. Roads\nA comparison of general circulation model and observed regional climates:\n282\nDaily and seasonal variability\nD. A. Portman, W. C. Wang and T. R. Karl\nxii","Page\nThe response of a GCM atmosphere to extratropical SST anomalies\n289\nYochanan Kushnir and N-C. Lau\nModeling North Pacific SST response to atmospheric anomalies\n295\nUte Luksch and H. von Storch\nSESSION 9. Prediction on Climatic Time Scales\nChair: E. S. Epstein\nCloud - radiation interactions: Effects of cirrus optical thickness\n*\nRichard C. Somerville\nfeedbacks\nVariability of skill of long-range forecasts and implications for their use\n301\nRobert E. Livezey\nand value\nRandom and systematic error in extended forecasts\nGeorge J. Boer\nPrediction of ENSO episodes using canonical correlation analysis.\n307\nAnthony G. Barnston and C. F. Ropelewski\nJohn 0. Roads and N. Maisel\nNumerical model precipitation forecasts\n313\nDynamic forecast of the next month's flow\nHuug M. van den Dool\n319\nMonte Carlo forecasting with dynamic stratification\n#A-13\nS. Schubert, M. Suarez and J-K. - E. Schemm\nDynamical prediction in the extended range employing low-frequency PNA mode\n324\nW. Y. Chen\nSea-surface temperature hindcasting experiments using an isopycnic coordi-\n328\n-\nnate model of the tropical Pacific\nArthur J. Miller, J. M. Oberhuber, N. E. Graham and T. P. Barnett\nStatistical prediction of January-February mean North American climate from\n333\nthe 11-year solar cycle and ENSO, for west QB0 phase years\nAnthony G. Barnston and R. E. Livezey\nPresentation was made at the Workship; however, paper was not submitted.\n*\n# Manuscript arrived late; see Appendix.\nxiii","SESSION 10.\nAspects of Predictability on Climatic Time Scales\nChair: Bob Livezey\nPage\nPrediction of forecast skill using Monte Carlo ensembles of 30 - day\n339\nforecasts\nDavid P. Baumhefner\nPotential for long - range predictability of monthly-mean surface tempera-\n344\ntures over North America\nDennis J. Shea and R. Madden\n6-10 day probability forecasts of daily precipitation frequency\n350\nWilliam H. Klein and E. S. Epstein\nStructure of recent surface air temperature (SAT) trends\n357\nG. V. Gruza, E. Ya. Ran'kova and E. V. Rocheva\nExperience on probabilistic long-range forecasting in the U.S.S.R.\n364\nG. V. Gruza and E. Ya. Ran'kova\nThe distribution of the correlation between large - scale averages and\n370\ncomponent members\nStanley L. Grotch\nSESSION 11 Long-Range Prediction Activities at Operational\nForecast Centers\nChair: Richard Somerville\nMedium and long-range forecasting at the National Meteorological Center\n375\nA. James Wagner\nModernization of long-range prediction operations at NMC\n381\nEdward A. 0' Lenic\nTowards a system for integrating statistical and numerical predictions in\n*\nobjective monthly forecasts: overview of the concepts\nStanley Woronko\nThe status of long-range forecasting at the U. K. Meteorological Office\n385\nM. S. J. Harrison,\nA. Dickinson, S. F. Milton, D. S. Richardson, A. W. Colman and A. Woodcock\nPresentation was made at the Workshop; however, paper was not submitted.\n*\nxiv","Regional Climatic Studies: Modelling and Prediction\nSESSION 12.\nChair: Tony Barnston\nPage\nrelationship between global SST and Sahel rainfall-investigations using\n391\nThe\nD. P. Rowell, K. Maskell and J. A. Owen\na GCM\nThe relationship of rainfall variability in East Africa to ENSO and the\n397\nSharon E. Nicholson and B. S. Nyenzi\nAtlantic and Indian Oceans\nSpecifying monthly mean anomalies of\nfire-weather elements in the\n403\nWilliam H. Klein and B. T. Whistler\nUnited States\nLong-range fire weather forecasting\n410\nFrancis M. Fujioka and M. H. McCutchan\nObserved fire weather variability\nKyozo Ueyoshi and J. 0. Roads\n416\nSESSION 13A.\nRegional Climatic Studies: Part 1\nChair: Shyh-Chin Chen\nThe present day agroclimatic changes\nG. V. Menzhulin\n*\nLarge-scale influences vs. local land-atmosphere interaction as factors in\n422\nSahel rainfall variability\nSharon E. Nicholson, A. R. Lare and H. Zheng\nSouthward displacement of the tropical Atlantic windfield associated with\n428\nthe Sahel rainfall trend 1948-83\nStefan Hastenrath\nThe anomalously warm winter of 1988-89 in the U.S.S.R.\n432\nV. N. Razuvayev and 0. N. Bulygina\nReasons for the reversal of circulation and U. S. anomaly patterns between\n438\nfall and winter, 1988-89\nJerome Namias\nClimate data sets\nKenneth D. Hadeen\n443\n* Presentation was made at the Workshop; however, paper was not submitted.\nXV","SESSION 13B.\nRegional Climatic Studies: Part 2\nChair: Dan Cayan\nPage\nOLR anomaly patterns associated to dry and wet winters in the subtropical\n*\nwest coast of South America\nPatricio Aceituno and A. del Rio\nRainfall anomalies in the subtropical west coast of South America: regional\nJose Rutllant and P. Aceituno\ncirculation aspects\nWet/dry spells viewed from SAI time series in the Canadian prairies\n449\nLeo 0. Mapanao\nInter - annual variability in North American moisture fluxes: streamflow\n455\nrecords meet atmospheric analyses\nNeil E. Sargent\nA climatic cloud atlas for North America\n460\nJohn E. Hanzis and A. E. O'Brien\nPresentation was made at the Workshop; however, paper was not submitted.\n*\nxvi","Author Index\nFourteenth Climate Diagnostics Workshop\nAuthor\nPage\nAuthor\nPage\nAlmeyda, G. F.\n47\nHirst, A.\n198\nArkin, P. A.\n8, 211, 216\nHolland, G.\n78\nBalsley, B. B.\n132\nHorel, J. D.\n127\nBarnett, T. P.\n96, 328\nHsiung, J.\n145\nBarnston, A. G.\n307, 333\nHu, Q.\n102\nBaumhefner, D.\n339, A-7\nJanowiak, J.\n8\nBuja, L. E.\nA-7\nKalnay, E.\n245\nBulygina, 0. N.\n432\nKanamitsu, M.\n245\nCarter, D. A.\n132\nKao, C. Y.\n265\nCayan, D.\n192\nKarl, T. R.\n115, 282\nChai, S. K.\n47\nKlein, W. H.\n350, 403\nChang, F-C.\n178\nKousky, V.\n1\nChelliah, M.\n216\nKukla, G.\n115\nChen, S-C.\n277\nKushnir, Y.\n289\nChen, T-C.\n138\nLander, M. A.\n12\nChen, W-Y.\n324\nLare, A.\n422\nChertock, B.\n172\nLau, N-C.\n289\nChristy, J. R.\n234, 240\nLau, W. K-M.\n178,198,260\nChu, P-S.\n121\nLevitus, S.\n205\nColman, A. W.\n385\nLiebmann, B.\n78\nDavidowicz, D.\n8\nLindzen, R. S.\n256\nDickinson, A.\n385\nLivezey, R. E.\n301, 333\nDouglas, A. V.\n108\nLuksch, U.\n295\nEcklund, W.\n132\nMadden, R.\n344\nEnglehart, P. J.\n108\nMaisel, N.\n313\nEpstein, E. S.\n350\nMaliekal, J. A.\n12\nFisher, M.\n271\nMalone, R. C.\n265\nFrederick, J.\n121\nMapanao, L.\n449\nFujioka, F. M.\n410\nMaskell, K.\n391\nGage, K. S.\n132\nMcAfee, J. R.\n132\nGates, W. L.\n55\nMcCutchan, M. H.\n410\nGavin, J.\n115\nMechoso, C. R.\n271\nGhil, M.\n18, 271\nMeehl, G.\n41\nGlatzmaier, G. A.\n265\nMiller, A. J.\n328\nGoodman, S. J.\n164\nMilton, S. F.\n385\nGraham, N. E.\n184, 328\nMiyakoda, K.\n35\nGreene, J. S.\n222\nMo, K. C.\n245\nGrotch, S. L.\n370\nMorrissey, M. L.\n12, 222\nGruza, G. V.\n357, 364\nMurphree, T.\nA-1\nGueremy, J-F.\n73\nNamias, J.\n438\nGutzler, D. S.\n67\nNeelin, D.\n22\nHadeen, K.\n443\nNemirovsky, A. V.\n61\nHahmann, A.\n127\nNewell, R. E.\n145\nHalpern, D.\n271\nNicholson, S. E.\n397, 422\nHalpert, M. S.\n84, 211\nNigam, S.\n256\nHameed, S.\n55\nNyenzi, B. S.\n397\nHanzis, J.\n460\nOberhuber, J. M.\n328\nHarrison, M.\n385\nO'Brien, A. E.\n460\nHastenrath, S.\n428\nO'Lenic, E. A.\n381\nHendon, H.\n78\nOwen, J. A.\n391\nxvii","Author\nPage\nAuthor\nPage\nPaegle, J.\n127, A-7\nSpahr, J. A.\n271\nPark, C-K.\n156\nSpencer, R. W.\n234, 240\nPonte, R. M.\n67\nSperber, K. R.\n55\nPortman, D. A.\n282\nStevens, Duane E.\n102\nPotter, G. L.\n55\nSuarez, M.\nA-13\nRadiuhin, V. T.\n61\nSud, Y. C.\n260\nRan'kova, E. Y.\n357, 364\nTellford, J. W.\n47\nRasmusson, E. M.\n84, 91\nTomlinson, R.\n211\nRazuvayev, V. N.\n432\nTurco, R.\n265\nReid, G. C.\n132\nUeyoshi, K.\n416\nRichardson, D. S.\n385\nvan den Dool, H. M.\n319\nRoads, J. 0.\n277, 313, 416\nvon Storch, H.\n295\nRobinson, M. K.\n28\nWagner, A. J.\n375\nRocheva, E. V.\n357\nWalker, G. K.\n260\nRogers, J. C.\n151\nWang, W. C.\n282\nRopelewski, C.\n84, 211, 307\nWang, X. L.\n91\nRosen, R. D.\n250\nWhistler, B. T.\n403\nRowell, D. P.\n391\nWirth, V.\n256\nSalstein, D. A.\n250\nWolter, K.\n228\nSargent, N.\n455\nWoodcock, A.\n385\nSchemm, J-K. E.\nA-13\nYen, M-C.\n138\nSchubert, S.\n156, A-13\nZheng, H.\n422\nShea, D. J.\n344\nZhongxiang, W.\n145\nSirutis, J.\n35\nZimmerman, J. R.\n245\nSmith, W. E.\n260\nxviii","Atmospheric Circulation Changes Associated with Extremes of the\nSouthern Oscillation during 1986-1989\nVernon E. Kousky\nand\nChester F. Ropelewski\nNOAA/NMC/NWS\nWashington, DC 20233\n1. INTRODUCTION\nIn a recent paper Kousky and Leetmaa (1989) discussed the\nevolution of selected atmospheric and oceanic anomaly fields for\nthe period leading up to and including the Pacific warm episode\n(ENSO) of 1986-1987. A weakening of low-level, atmospheric,\nequatorial easterly winds in the central Pacific, during the\nlatter half of 1986, was accompanied by a gradual shift of the\nwarmest water from the extreme western Pacific to the region\nalong the equator near the date line. At the same time atmos-\npheric convection became enhanced and the oceanic thermocline\ndeepened in the region of the warmest surface water. Oceanic\nKelvin waves, initiated in the central Pacific due to the weaken-\ning easterlies, propagated eastward to the South American coast\nand appear to have been important in depressing the thermocline\nalong the coast resulting in positive sea surface temperature\n(SST) anomalies and El Niño conditions. The precipitation and\natmospheric circulation anomalies observed during the episode\nagree well with those generally found associated with ENSO (e.g.,\nRopelewski and Halpert, 1987; Arkin 1982)\nBeginning in early 1988 there was a rapid transition to a\ncold episode; the first in nearly 15 years. Many of the major\nfeatures of the atmospheric and oceanic anomaly fields for these\nextremes in the Southern Oscillation (SO) have been presented in\nseasonal climate summaries (e.g., Bergman, 1987; Kousky, 1987;\nWagner, 1987; Arkin, 1988; O'Lenic, 1988; Ropelewski, 1988;\nKousky, 1989; Arkin, 1989).\nIn this paper we will discuss the atmospheric and oceanic\ncirculation changes related to these recent extremes of the so,\ngiving emphasis to those features which have not as yet been\nwell-documented.\n2.\nDATA\nThe data presented here have been taken from the atmospheric\ncirculation archives maintained at the Climate Analysis Center\n1","(CAC) in Washington, DC.\nZONALLY AVERAGED MERIDIONAL CIRCULATION CHANGES\n3.\nBjerknes (1966) suggested that the winter Hadley circulation\nis enhanced and the subtropical jet streams are strengthened\nduring periods of anomalously warm SST in the central equatorial\nPacific. Later Bjerknes (1969) determined that in addition to\nthe changes in the Hadley circulation, changes also occur in the\nzonal (or what Bjerknes referred to as the Walker) circulation in\nthe Pacific.\nThe extremes in the SO that occurred during the last three\nyears offer the opportunity to examine the strength of the Hadley\nand Walker circulations during both warm and cold episodes.\nFigures 1a and 1b show time-latitude cross sections of the zonal-\nly averaged vertical motion for December 1986-February 1987 and\nDecember 1988-February 1989, respectively. Both the upward\nmotion (negative values of omega) in the equatorial belt and the\nsinking motion in the Northern Hemisphere subtropics were\nstronger during the 1986-87 warm episode (ENSO) : The more\nintense Hadley circulation during December 1986-February 1987 is\nconsistent with 200 mb westerly wind anomalies observed near 30°N\nduring that season (Kousky, 1987) During December 1988-February\n1989 200 mb easterly anomalies prevailed between 20 and 30°N\n(Arkin, 1989) associated with the weaker Hadley circulation shown\nin Fig. 1b.\nThe zonally averaged vertical motions for the June-August\n1987 and 1988 seasons are shown in Fig. 2. During that time of\nyear the greatest ascending vertical motion occurs slightly north\nof the equator and the descending motion, associated with the\nHadley circulation, occurs in the Southern Hemisphere. Although\nnot as evident as in Fig. 1, the Hadley circulation is again\nstronger during the warm episode. Westerly wind anomalies\ndominated the Southern Hemisphere from 20 to 30°S during June-\nAugust 1987, with the largest anomalies being observed in the\nPacific sector (Arkin, 1988) During June-August 1988, on the\nother hand, westerly anomalies continued to be observed in the\nSouthern Hemisphere subtropics, although mainly confined to the\nAtlantic and Indian Oceans (Ropelewski, 1988) Weak easterly\nanomalies were found in the subtropical South Pacific.\nZONAL CIRCULATION CHANGES IN THE TROPICS\n4.\nThe east-west (zonal) circulation in the equatorial belt\nalso shows substantial differences between the warm and cold\nextremes of the SO. Figures 3a and 3b show the time-longitude\nsections of the 5°N-5°S averaged vertical motion for December\n1986-February 1987 and December 1988-February 1989, respectively.\nThe position of the main axis of rising motion in the Pacific\nsector is considerably different in the two diagrams. During the\n2","warm episode (Fig. 3a), the strongest rising motion was located\njust east of the date line and the sinking motion in the eastern\nPacific was fairly weak. In contrast, during the cold episode\n(Fig. 3b) the main axis of rising motion was between 110 and\n140°E and sinking motion was stronger and more extensive in the\neastern Pacific. (Note: the NMC model resolution changed between\nthese two periods, which affected the vertical motion pattern in\nthe vicinity of the Andes and also in the region of high terrain\nin Africa. Thus, caution should be exercised in interpreting\nvertical motion changes in these regions.)\n5.\nMID-TROPOSPHERIC TEMPERATURE VARIATIONS\nTo illustrate the global pattern of mid-tropospheric temper-\nature anomalies associated with the recent extremes of the so, we\nhave computed the 500 mb temperature anomalies for selected sea-\nsons. Sea surface temperature anomalies in the tropical Pacific,\nduring the 1986-87 ENSO, reached their maximum positive values\nduring June-September 1987. Figure 4a shows the 500 mb tempera-\nture anomaly pattern for the June-August 1987 season. Positive\nanomalies of between 0.5 and 1°c dominated the tropics. Negative\nanomalies prevailed in the Southern Hemisphere subtropics, primarily\nin the Indian-Pacific sectors.\nLarge negative 500 mb temperature anomalies occurred in the\ntropics during December 1988-February 1989 in association with\nthe cold episode in the tropical Pacific (Fig. 4b). Anomalies of\nfrom -1 to -3°C were observed throughout the tropical and sub-\ntropical Pacific. At the same time positive anomalies prevailed\nat the mid-latitudes in both hemispheres with negative anomalies\nat high latitudes. This pattern of temperature anomalies was\nassociated with upper tropospheric easterly anomalies at subtrop-\nical latitudes in both hemispheres. There is some tendency for\nthe mid-tropospheric temperature anomalies to lag the SST anoma-\nlies, which suggests that atmospheric circulation anomalies would\nalso lag the SST anomalies.\n6. CONCLUSION\nThe above patterns of anomalous temperature are consistent\nwith the results in sections 3 showing changes in the Hadley\ncirculation. The warmer tropical troposphere during ENSO is\naccompanied by an enhanced Hadley circulation and stronger than\nnormal subtropical jet streams, especially in the winter hemis-\nphere. These results are consistent with those presented by\nBjerknes (1966) Conversely, colder than normal conditions in\nthe tropical troposphere occurring during the cold (high Southern\nOscillation Index) phase of the SO are related to a weaker than\nnormal Hadley circulation and also weaker than normal subtropical\njet streams.\nDuring ENSO the main region of upward vertical motion within\n3","the tropics is shifted eastward from Indonesia into the central\nPacific while during cold episodes sinking motion dominates the\ncentral and eastern equatorial Pacific. In agreement with\nprevious works (Bjerknes, 1966, 1969) we find an enhanced Hadley\ncirculation and weakened east-west (Walker) circulation during a\nwarm episode and weakened Hadley circulation and strengthened\nWalker circulation during a cold episode.\nACKNOWLEDGMENTS\nThe author wishes to thank John Kopman who helped in prepar-\ning the figures, Mike Halpert who provided the software used in\nproducing the plots and Kathy Stevenson who helped in typing and\nediting the final version.\nREFERENCES\nArkin, P. A. , 1987: The global climate for June-August 1987:\nMature phase of an ENSO warm episode persists. J. Climate,\n1, 306-324.\nArkin, P. A. , 1989: The global climate for December 1988-February\n1989: Cold episode continues. J. Climate, 2, 737-757.\nBergman, K. H. , 1987: The global climate of September-November\n1986: A moderate ENSO warming develops in the tropical\nPacific. Mon. Wea. Rev. 115, 2524-2541.\nBjerknes, J. 1966: A possible response of the atmospheric Hadley\ncirculation to equatorial anomalies of ocean temperature.\nTellus, 18, 820-828.\nBjerknes, J. , 1969: Atmospheric teleconnections from the\nequatorial Pacific. Mon. Wea. Rev. , 97, 163-172.\nKousky, V. E. , 1987: The global climate for December 1986-\nFebruary 1987: El Niño returns to the tropical Pacific.\nMon. Wea. Rev., 115, 2822-2838.\nKousky, V. E. , 1989: The global climate for September-November\n1988: High Southern Oscillation index and cold episode\ncharacteristics continued. J. Climate, 2, 173-192.\nKousky, V. E. , and A. Leetmaa, 1989: The 1987-87 Pacific warm\nepisode: Evolution of oceanic and atmospheric anomaly\nfields. J. Climate, 2, 254-267.\nO'Lenic, E. A. 1988: The global climate for September-November\n1987: Mature phase of tropical warm event continues amid\nglobal warmth and dryness. J. Climate, 1, 548-568.\n4","Ropelewski, C. F. . , 1988: The global of climate the Southern for June-August oscillation, monsoon 1988\nA swing to in the the positive United States, phase and abundant rain in\n:\ndrought areas. J. Climate, 1, 1153-1174. -\nWagner, Moderately A. J. . , 1987: strong The mature and global seasonal phase Climate of temperature ENSO of March-May with 3166-3187. highly anomalies over\n1987:\npersistent the United monthly States. Mon. Wea. Rev. , 115,\n100\n100\nb\na\n10.0\n40\n200\n200\n0.0\n0,0\n20.0\n20.0\n250\n250\n10.0\n300\n300\n-10.0\n10.0\n-19.0\n0.0\n10.0\n10.0\n500\n500\n0.0\n-10.0\n0\n700\n700\n2)\n850\n850\n1000\n1000\n60N\n90N\n90S\n60S\n30S\n30N\n60N\n90S\n60S\n30S\n0\n30N\n0\n90N\nLatitude vs motion pressure (omega cross X 10 section 5 mb s-1) of 1988-February zonally for a) averaged December 1989.\nFIGURE 1\nvertical 1986-February interval 1987 10. and Values b) December less than -10 - X 10 -5 mb S\nContour\nhave been shaded.\n1\n5","100\n100\nV\nb\na\n10,0\n10.0\n0.0\n0\n200\n200\n0.0\n20.0\n10.10\n250\n250\n20,-\n0\n10.0\n300\n300\n0\n500\n500\n700\n700\n850\n850\nIA\n1000\n1000\n90S\n60S\n30S\n0\n30N\n60N\n90N\n30S\n0\n30N\n60N\n90N\n90S\n60S\nLatitude vs pressure cross section of zonally averaged\nFIGURE 2\nvertical motion (omega X 10 5 mb s-1) for a) June-July\nContour interval 10.\n1987 and b) ) June-July 1988.\n1\nValues less than - -10 X 10 5 mb S\nhave been shaded.\n100\nTUTT\n100\nb\na\nv\n20.0\n0.0\n200\n010 20.0\n200\n20\nOpo 0\n2\n250\n-20\n250\n401\n300\n300\n20.0,\n0\n500\n500\n0\n20.0\n0.0\n20\n700\n700\n20\n0,0\n850\n850\nNA\nJAN\nME\n1000\n1000\n00\n30E 60E 90E 20E 50E 180 50W I 120W90W GOW 30W 00\n30E 60E 90E 20E 150E 180 150W 1 20W 90W 60W 30W 00\n00\nLongitude vs pressure cross section of vertical motion\nFIGURE 3\n5\nmb s-1) averaged over the latitude band\n(omega X 10\n5°N-5°S for a) December 1986-February 1987, and b)\nDecember 1988 - February 1989. Contour interval is 20\nX\n10-5 mb S - 1 .\n5\n-1\nValues less than -20 X 10\nmb\nS\nare\nshaded.\n6","60E5.0mE-00100E\n-10230,80 040E -05 140W. 0 12045 00.000W -5.80W 0.000 1040W15 029 400 -5 DED 0.0 -20-ED a 8 SEO\n60N\nVIS\n5.00\n3.5\n160N\n5.0\n5.0\n0/0\n20\n8\n5%\n5.0\n50N\n5\n50N\n0.0\nDON'T\n0\n-5.\nse\n508\nSTREET\nF15.\n-BON\n4030N\n-5.\n20N\n5.0\nID\n0\n20N\n0.\n5.0\nLON\n5.0\n5.\nCON\nEQ\nEQ\n10S\nota\nLOVE\n0.0\n10S\n5.0\n20S\n10. 0\n22 S\n0.\n00\n0.0\n5.0\n305\no\n385\n5.\na\nIfree\n40S\n0.0\n0.0\n40S\n-10.0\nROS\n-10.0\n508\n5.0\n5.0\n0.0\n.0\n60S\n6000 80E 1500 10205.0 140E 160E5.0 180 1600. 0 01.40W 5.120W La.ww 080W -56.0W 401.0 200v 000\n60S\n20E\n-ELED\n- 15. 000E 15 1006 MDL 40E -2060E -20.1030 -51.02.00$ 05 GOW 0. 020W -5.100 0 -20 800 W- 40 60W40 0 4038 0-58000 20 CCCO 0 26E0 5.04EED. 0\nREN\n60N\n15.0\n-200.00\n50N\n-20.\n+10.0\n15.0\n-15.9\n4UN\n40N\n15.0\n-18.8\n$800\n5:\n15.0\n5.0\n2019\n20N\n10.0\n-5.0\n-5:\n5.0\n10ND\nION,\n0\n10.0\n-10.05D\nEQ\n30.\n1500\n-5rd\nrus\nDOS\n0.50\nO\n20$\n0,0\n10.0\n-5.1\n20S\n-10.00\n20-B\n5.0\nbos\n3850\nQ\n100.50\n4050\n-5.0\n5.0\n0.0\n2.0\n50$\n305\nillition\nI\nb\n20.0\n60S\n60S\n5.20E0.0 80E 100E 120E0.0140E 5.00E Ea 0 0.060W 140W -520W 1-050.00 BOW 60W 40016.0 20W 0010.020E\n40E\nFIGURE\n4 500 mb temperature anomalies, computed with respect to\nthe October 1978-September 1985 base period, for a)\nJune-August 1987, and b) December 1988 -February - 1989.\nValues have been multiplied by 10 and contour interval\nis 0.5°c. Values greater than 0.5°c have been shaded.\n7","Variations in tropical rainfall during 1986-88 as inferred\nfrom satellite observations of cloud-top temperature\nJohn E. Janowiak\nPhillip A. Arkin\nDiane B. Davidowicz\nClimate Analysis Center, NMC, NWS, NOAA\nWashington, D. C. 20233\n1. INTRODUCTION\nLarge-scale changes in tropical rainfall distribution during 1986-88\nare shown. The relationship between these rainfall variations and the\nchanging ENSO conditions during that period are briefly described. A\nmoderate \"warm\" episode in the central equatorial Pacific, which persisted\nfrom mid-1986 through late 1987, was observed, followed by a \"cold\" episode\nin early 1989. This study is a by-product of work related to the Global\nPrecipitation Climatology Project (GPCP), a WMO project whose goal is the\ncreation of a ten year monthly climatology of rainfall during the period\n1986-95. The rainfall \"observations\" used here are indirect in that they\nare inferred from an estimation method that employs remotely sensed cloud-\ntop temperatures from geostationary and polar orbiting satellites.\n2. RAINFALL ESTIMATION METHOD\nDirect measurements of rainfall (raingage, radar) are not available in\nmany parts of the tropics, particularly over the oceans. Therefore, the\nonly way to measure rainfall in these regions at the present time is by\nusing indirect observations, from space, of parameters related to rainfall.\nWe use infrared radiometric data that can detect regions of deep\nconvection in the tropics where rainfall is primarily convective and the\nannual cycle of surface temperature is small. Such information is not\nuseful outside of the tropics in the winter hemisphere, as cold surface\ntemperatures cannot be discriminated from cold cloud using the infrared\ndata alone.\nRichards and Arkin (1981) developed a quantitative rainfall estimation\nmethod using infrared brightness temperature data for use in the tropics.\nThe details of the method are not discussed here and the reader is referred\nto their paper for a complete discussion of the procedure. This method has\nbeen coined the GOES Precipitation Index (GPI) Data from the GOES,\nMETEOSAT, and GMS geostationary satellites were used, when available, to\nproduce the estimates used in this paper. In regions for which\ngeostationary satellite data were not available, estimates based on OLR flux\nfrom the NOAA polar orbiting satellites were used. The technique for\nestimating rainfall from OLR flux is not discussed here, but the method will\nbe documented in the near future.\n8","3. TROPICAL RAINFALL VARIATIONS DURING 1986-88\nAs mentioned above, substantial rainfall variations occurred in the\ntropics during 1986-88. Figure 1 shows a time-longitude section of\nestimated rainfall, averaged from 6.25 0 N to 6.25°S, for the period January,\n1986 through December, 1989. During early-mid 1986, rainfall in the\nvicinity of the equator and 180° was estimated to be about 5 mm/day. From\nlate 1986 through mid-1987, rainfall in that region increased dramatically\nto a maximum of about 20 mm/day. Note also the 6 mm/day maximum that\nappeared in the equatorial region centered at 120°W in early 1987 as\nenhanced convection propagated eastward toward the South American coast.\nDuring the \"cold\" ENSO episode, which began in early 1988, rainfall in the\nregion decreased to a minimum in December, 1988 of 2 mm/day. Figure 2 shows\nthe monthly mean SST for the Southern Hemisphere summer periods (DJF) of\n1986-87, 1987-88, and 1988-89. Note that the longitudinal position of the\n28° C isotherm (a \"rule of thumb\" indicator for the outbreak of convection\nover the ocean) at the equator is about 150°W during DJF 1986-87, and moves\nwestward the following two seasons to about 165 E during DJF 1988-89. This\nwestward shift of relatively warm SST is consistent with the westward shift\nof maximum rainfall at the equator as shown in Fig. 1.\nEnhanced rainfall was observed during the \"warm\" event in the\nequatorial mid-Pacific, southeast Brazil and equatorial Africa, while\ndecreased rainfall was observed in northeast Brazil, southern Africa, and\nIndia. Rainfall variations of the opposite sign were observed in the same\nareas during the \"cold\" ENSO event. The differences in rainfall\ndistribution between the \"warm\" ENSO episode (1986-87) and the \"cold\" ENSO\nepisode (1988-89) in the tropics are quite consistent with the expected\ndifferences as pointed out by Ropelewski and Halpert (1987, 1989) among\nothers.\nReferences\nArkin, P. A. and F. Richards, 1981: On the relationship between satellite-\nobserved cloud cover and precipitation. Mon. Wea. Rev., , 109, 1081-1093.\nRopelewski, C. F. and M. S. Halpert, 1987: Global and regional scale\nprecipitation patterns associated with the E1 Nino/Southern Oscillation.\nMon. Wea. Rev., 115, 1606-1626.\n, 1989: Precipitation patterns associated with the high index phase\nof the Southern Oscillation. J. Climate, 2, 268-284.\n9","8\n8.\n0\n6.\n8\n0\n2,0\n0\n2.02.0\n(\n00\n8.\nP.O\n15.0\n1986\n10\n15.0\n0\n10.0\n1010\n2010\n8,0\n6,0\n8,0\n4\n8\n0\n1987\n15.0\n2,0\n4,0\n8.08.0\n90.050\n10.\n15\n0.0\n15.0\n8\n0\n2.0\n4\n8,\n4,0\n0\nM.O\n6.0\n1988\n60\n4\n10.\n6.0\n2,0\n2010\n0\nA\nT\n00\n30E\n60E\n90E\n120E\n150E\n180\n150W\n120W\n90W\n60W\n30W\n00\nFigure 1. Time-longitude section of monthly estimated rainfall for each\nmonth from January, 1986 through December, 1988, averaged from 6.25°N to\n6.25°S. Units are mm/day. Contour interval is 2 mm/day for contours < 10 :\nmm/day and 5 mm/day for contours >10 mm/day.\n10","90E\n1804\n90W\n0\n110\n501\nSON\n0\nDJF 86-87\n40N\n40N\n16\n10\n2000\n14\n10\n20N\n20N\n&\nEQ.\nEQ.\n24\n20\n28\n26\n26\n20S\n20S\n02\n24\n24\n22\n22\n18\n2\n22\n20\n18\n20\n16\nis\n40S\n10\n16\n10\n40S\n14\n16\n14\n50S\nSOS\n90E\n180\n90W\n0\n110\n90E\n180\n90W\n0\n10\nL'O\n2\nSON\nSON\nT2\n0\nDJF 87-88\n401\n40N\n16\n16\n10\nII\n10\n20N\n20\nto\n20N\n28\n24\nEQ.\n28\nEQ.\n24\n28\n30\ne\n20\n28\n26\n205\n26\n2\n20S\nSpr\n18\n38\n24\n24\n22\n18\n22\n2\n22\n20\nor\n20\n20\nto\n16\n40S\n40\n16\n16\n10\n16\n16\nTO\n40S\n1416\n16\n505\nSOS\nSOE\n180\nSOW\n0\n110\nyur\n18U\nyun\n6\n100\n50N\n22\nSON\n12\nDJF 88-89\n40N\n40N\n16\n10\n14\n10-\nis\n10\n70\n20N\n20N\n586\n222-28\nEQ.\n28\nEQ.\n24\n( 27\n28\n28\n123\n20S\n212\n26\n22\n20S\n24\n18\n24\n22\n20\n18\n2\n20\n00\n18\n20\n20\n18\nis\n40S\n10\n10\nto\n16\n16\n40S\n14\n16\n16\n50S\nSOS\n90E\n180\n90W\n0\n110\nFigure 2. Seasonally averaged SST (°C) for the DJF seasons of 1986-87, 1987-\n88, and 1988-89. Temperatures > 20°C are contoured every degree with odd\ncontours dashed. Thick line in tropical Pacific ocean is the 28°C isotherm.\n11","Comparison of the zonal propagation of equatorial\nOLR fields during the three recent El Ninos\nJose A Maliekal, Mark A Lander and Mark L Morrissey\nDepartment of Meteorology\nUniversity of Hawaii\nHonolulu, Hawaii 96822\nWyrtki proposed a comprehensive mode of air-sea coupling to explain the El Nino\nSouthern Oscillation (ENSO) phenomenon. When the Southern Oscillation (SO) index is\nhigh (which is usually the case prior to El Nino) the Trades are stronger than normal, and\nthe accompanying wind stress piles up water in the western Pacific. A subsequent drop in\nthe SO index weakens the Trades allowing the accumulated water in the western Pacific to\nbe transported eastward in the form of internal Kelvin waves. These waves upon arrival off\nthe Peru coast depress the otherwise shallow thermocline producing high SST there. The\nwarm water spreads westward, eventually reaching the equatorial eastern and central\nPacific. This warm pool in the central Pacific causes the rising branch of the Walker\ncirculation cell to shift to central pacific, producing rainfall in an otherwise dry region.\nRasmusson and Carpenter's composite ENSO concurres with Wyrtki's model, albeit, they\ncontend that ENSO cannot be adequately described in terms of a standing oscillation alone.\nHowever, Wyrtki's hypothesis that the SO index had to rise and the Trades had to intensify\nto trigger El Nino was negated by the extraodinarily intense 1982-83 event; prior to which\nneither of these conditions were met.\nRecent studies have claimed that:\n(a) the wind anomaly associated with ENSO originates in the Indian Ocean and migrates\neastward,\n(b) anomalous low level convergence associated with ENSO, which later propagates\neastward, first appears in the vicinity of Indonesia, and not in the Indian Ocean,\n(c) the atmospheric interannual fluctuations associated with ENSO exhibit an eastward\npropagation as well as a standing wave character,\n(d) the dominant ENSO feature is a planetary scale standing wave and the zonal\npropagation is nothing more than an embellishment on the standing wave.\nAnother hypothesis evolved from the observational studies of the equatorial western\nPacific surface wind is that the equatorial oceanic Kelvin waves responsible for ENSO\ninitiation are produced by short period (less than a month) wind fluctuations.\nThese hypotheses hinge upon the nature of air-sea coupling at interannual, and\nshorter time scales. Important unanswered questions are: does zonal propagation (along the\nequatorial plane) play a decisive role in the onset and maintenance of ENSO?; What role,\nif any, do transient small scale fluctuations play in the onset and maintenance of ENSO?\nIn this article, we investigate these issues using equatorial (5S-5N) Outgoing\nLongwave Radiation (OLR) field for the three most recent ENSO events, 1976-77, 1982-\n83 and 1986-87. Time-longitude plots, autocorrelation analysis (in the space domain),\nunivariate spectrum analysis, and multivariate spectrum analysis are used.\nWe formed daily space series of OLR along the equatorial zonal plain. The\ncomponent due to the annual march of sun is the removed by subtracting the 12 year mean\ndaily series from the daily space series. Seasonal averages of OLR values are then formed.\nTransition seasons are of two month duration (April-May for spring and October-November\nfor Fall) and the cardinal seasons are of four month duration (June-September for summer\n12","and December-March for winter). We adopted this rather unconventional break-up to\ncapture changes, if any, that may occur during the short transition period when troughs of\nlow pressure straddle the equator. To ameliorate the effects of unequal averaging, space\nseries are centered (by removing the zonal mean values) and standardized (by dividing with\nthe standard deviation). This normalized series is smoothed by a median filter, which is\nboth robust and resistant.\nPrior to the 1976-77 ENSO (Fig. 1), the western Pacific remained cloud free, but the\nmaritime continent remained cloudy. This pattern flip-flopped in the spring of 1976. Dry\nconditions continued to persist over the maritime continent throughout 1977 with the central\nPacific experiencing the opposite polarity. No systematic eastward propagation is evident\neither prior to or during the ENSO event. The pattern of equatorial convection prior to and\nduring the 1982-83 ENSO event is in stark contrast with that of the 1976-77 event. The\norigin of the anomalous convection associated with the 1982-83 ENSO can be traced back\nto the maritime continent, beginning in the summer of 1981. Once started over the\nmaritime continent, the convection propagated systematically all the way to the Atlantic.\nEastward propagation is systematic and vivid. Throughout the year of 1985, the maritime\ncontinent and the adjoing western Pacific were both convectively active. By the winter of\n1986-87, the conditions over the maritime continent resembled that of ENSO, however the\nequatorial convection was confined to a small region near the date line. During the\nsubsequent months, regions further to the east became convectively active with classical\nENSO conditions. No systematic eastward propagation is suggested.\nAs a next step, daily spatial correlations as a function of separation distances (in 2.5\ndegrees of longitude) are computed using space series of OLR along the equatorial plain.\nThese daily correlation functions are then averaged to form mean seasonal values, which\nare in Fig. 2. Once again, the 1982-83 event stands out. The zonal correlation function of\nOLR showed no appreciable changes prior to and during the 1976-77 ENSO event. The\nmark of the 1982-83 event, began to exhibit itself by the fall of 1982. The implied zonal\nscale of the large scale convection is approximately 15,000 km. As the event progressed to\nmaturity, the zonal length scale increased to about 30,000 km by the spring of 1983. Similar\nstructure is seen during the spring of 1987.\nTo understand the nature of transient smaller scale events prior to and during ENSO\nevents, four sub-regions along the equatorial belt are studied. They are: the Indian Ocean;\nthe maritime continent; the western Pacific Ocean and the central Pacific Ocean. Power\nspectrum of area averaged OLR values are estimated parametrically using autoregressive\n(AR) models. The order of the AR models are determined by Akaike's Information\nCriterion (AIC).\nIn most cases (Fig. 3) the power decreases monotonically from the high values at low\nfrequencies to low values at high frequencies. An increase in synoptic activity over the\nwestern Pacific during the year 1975 (the year before 1976-77 event), an increase in activity\nat 0.025 cycles per day (CPD) over the Indian Ocean during the year 1976 and an increase\nin activity at 0.10 CPD over the central Pacific are the exceptions associated with the 1976-\n77 event. The exceptions associated with the 1982-83 event are: an increase in activity at\n0.025 CPD over the Indian Ocean during 1981 and an increase in activity at 0.10 CPD over\nthe central Pacific during 1983. The 1986-87 event largely resembles the 1976-77 event. We\nconfirmed the validity of our spectra by examining, how well the estimated AR model tracks\nthe original data and the one-step ahead prediction errors.\nMultivariate spectral analysis is then employed to identify whether convection in one\nregion could be identified as driving convection over the other regions. For this purpose,\nwe estimate the spectral coherence and partial coherence, the coherence remaining between\n13","two OLR time series after the influence of third OLR time series is removed from each of\nthe first two. The spectral coherence and the partial spectral coherence are estimated\nparameterically, using multivariate AR models. The order of the AR models are determined\nby AIC. In order to reduce bias, the multivariate spectra were computed after aligning the\nOLR series so that the cross correlation function is highest at zero lag. An estimate of the\nnumber of degrees of freedom of an AR modelled time series is N/p, where N is the\nlength of the time series and p is the model order. In the present study N=365, and p in\nmost cases is less than 4. A conservative estimate of the 95% confidence level for zero\nspectral coherence and zero partial spectral coherence is with 40 degrees of freedom is 0.08.\nPartial spectral coherence is not very much different from spectral coherence in most\ncases (Fig. 4). During 1981 and 1983, there is significant coherence between Indian Ocean\nand western Pacific Ocean; however, the partial coherence between these two regions is not\nsignificantly different from zero. Therefore, the coherent fluctuations in the India Ocean\nand western Pacific during these two years is a result of the influences exerted by the\nmaritime continent. This result further supports the earlier suggestion that the 1982-1983\nevent may have had its origin over the maritime continent.\nThese results seem to suggest that:\n(a) different ENSO events are initiated differently,\n(b) eastward propagation of anomalies is associated with major ENSO events only,\n(c) the maritime continent played critical role in the initiation of 1982-83 ENSO event,\n(d) during minor ENSO events equatorial convection flip-flopps between maritime\ncontinent/western Pacific Ocean and central Pacific Ocean.\n14","Zonal OLR (5S-5N) 1975-1977\nZonal OLR (5S-5N) 1981-1984\nZonal OLR (5S-5N) 1985-1988\n75\n75\nas\n75\n75-76\n76\n76\n76\n76-77\n07\n77\n83-84\n40\nDOE\n120\n100\n160\n120\nsow\n40\nSOE\n120\n160\n160\n120\n40\nDOE\n120\n160\n160\n120\nBOW\n40\nFig 1. Time-longitude plots of seasonally averaged OLR standardized with respect to zonal\nmean and standard deviation. The tick marks correspond to two standard deviation.\nZCF of OLR (EQ) 1975-1977\nZCF of OLA (EQ) 1961-1984\nZCF of OLR (EQ) 1985-1968\nSpring\nSummer\nFall\nWinter\nSpring\nSummer\nFail\nWinter\nSpring\nSummer\nFall\nWinter\nFig 2. Seasonally averaged correlation against separation distance. Each row represent one\nyear and columns represent same season for different years.\n15","Power Spectrum of OLA (5S-5N) :1975-77\nPower Spectrum of OLA (5S-5N) :1981-83\n:\n2\n21\n1.\n=\na\n3\n2\n:\n21\n2\n2\n2\n2\n:\n:\n=\n2\n=\n1:\n1:\n3\n3\n:\nPower Spectrum of OLR (5S-5N) :1985-87\n:\n20\n2\n2\n3\n9\n:\n:\n:\n:\n2\n2\n22\n12\n:\n2\n2\n:\n9\n00\nFig 3. Power spectrum of daily area averaged OLR for four regions. Power is plotted on\na logarithmic scale and frequency is in cycles per day. The four regions are: (a) Indian\nOcean (IO); (b) maritime continent (MC); (c) western Pacific Ocean (WP); and (d) central\nPacific Ocean (CP).\n16","WP CPIMC\nMC WP 10\nas\nes\n:\n:\n2\n:\nOLA(5S-5N) 1982\n10 VPIMC\nMC IWP\nMC WP 110\nWP CPIMC\n:\n:\n2\n:\n:\n:\n:\n:\n:\n04\nMC. WP / CP\n10 MCIWP\n:\n:\n2\n2\n:\n:\n2\n:\nFig 4. Spectral coherence (full lines) and partial spectral coherence (dashed lines) of daily\narea averaged OLR between different regions. The caption IO MC I WP stands for spectral\ncoherence between Indian Ocean and Maritime Continent and partial spectral coherence\nafter the influence of WP is removed from the Indian Ocean and Maritime Continent.\nOLA(SS-SM) :1983\n10 WP IMC\nMC CP |WP\n:\n:\n:\n:\n:\n:\n:\n:\n:\n3\n2\n2\nMC WP 110\nCPIMC\n:\n:\n:\n2\n:\n:\n:\n:\n3\n:\nas\nMC. WPICP\n10 MCIWP\n:\n21\n:\n:\n:\n2\n:\n=\n2\nOLA(SS-SN):1981\n10 WP IMC\nMC CPIWP\n:\n:\n:\n:\n2\n:\n:\n:\n:\n:\n:\n:\n:\n:\n:\n:\n2\n:\n:\n:\na\n:\n10 MC|WP\nMC. WP CP\nas\n:\n:\n:\n3\n:\n:\n:\n2\n:\n2\nI\n:\nMC WP 110\nWP CPIMC\n:\n:\n:\n:\n:\n:\nOLR(SS-SH) 1976\n10 IMC\nMC CPIWP\nMC WP 110\nCPIMC\n:\n:\n:\n:\na\n:\n:\n2\n:\n2\n2\nMC. WP|CP\n10 MC|WP\n:\n:\n:\n:\n=\n:\n:\n:\n:\nOLR(SS-SM):1977\n10 VP / MC\nMC CP |WP\nas\n:\n:\n:\n:\n:\n3\n2\n:\n00\n00\nMC WP 110\nCPIMC\n:\n:\n:\n:\n:\n:\n:\n0.4\n10 MCIWP\nMC. WP|CP\n:\n:\n:\n2\n:\n:\n:\n:\n:\n:\nOLA(SS-SN) 1975\n10 WP IMC\nMC CP|WP\n:\n:\n:\n:\n:\n:\n:\n:\n:\n:\n:\n:\n:\n:\n:\n:\n:\nMC, WP CP\n10 MCIWP\n:\n2\n:\n:\n:\n:\n2\n3\n:\n2\n:","Multiple Flow Regimes and ENSO\nMichael Ghil\nClimate Dynamics Center, Department of Atmospheric Sciences and\nInstitute of Geophysics and Planetary Physics\nUniversity of California, Los Angeles, CA 90024-1565\nEl Niño/Southern Oscillation (ENSO) accounts for the largest interannual\ncontribution to climatic change of the coupled atmosphere-upper ocean system.\nExtratropical atmospheric response to warm (El Niño) or cold (La Niña) sea-\nsurface temperature (SST) anomalies is, however, far from unique (Namias and\nCayan, Oceanus, 1984)\nWe proceed, therefore, to a study of extratropical low-frequency variabil-\nity (LFV) in three steps: 1) classifying intraseasonal, persistent anomalies\nof the Northern Hemisphere (NH) and Southern Hemisphere (SH) circulation into\nmultiple flow regimes; 2) determining the Markov chain of transitions between\nthese regimes; and 3) evaluating the effect of opposite phases of ENSO on the\nrelative importance of the distinct regimes and the likelihood of transitions\nbetween them in a given year. The present paper reviews results from joint\nwork with M. Kimoto, K. Mo, A. Orrhede and R. Vautard, as indicated by the\nreferences.\nMultiple flow regimes\nRecent classification schemes for atmospheric anomaly maps are based\neither on persistence or on similarity. When persistence is the basic criter-\nion, sequences of maps characterized by small day-to-day change are grouped\ninto events, and the latter into regimes. When similarity is given exclusive\nattention, one concentrates on the agglomeration of anomaly maps in certain\nparts of phase space, independently of the moment at which they have occurred.\nThe former approach considers neighborhood in both phase space and time, the\nlatter in phase space only. In either case, neighborhood in phase space can be\ndefined by small angular distance when viewed from the time mean of the entire\nsequence of maps, or by small Euclidean (root-mean-square) distance.\nThe persistence point of view is represented, when considering angular\ndistance, by the pattern-correlation method developed by Horel (MWR, 1985) for\nNH flows and applied by Mo and Ghil (JAS, 1987) to SH flows. For Euclidean\ndistance, this approach was introduced first by Legras and Ghil (JAS, 1985),\nand developed further by Vautard and Legras (JAS, 1988), both in the context of\nmodel results; it is being applied to atmospheric observations in the North\nAtlantic-European sector by Vautard (MWR, 1989, submitted)\n.\nThe pure-similarity point of view is represented, when considering angular\ndistance only, by the cluster analysis of NH flows in Mo and Ghil (JGR, 1988)\nKimoto (Ph.D. thesis, UCLA, 1989) has carried out a classification of NH anomaly\nmaps by multivariate probability density function (pdf) estimation using both\nEuclidean and angular distance, justifying objectively the intuitive preference\nfor the latter. A similar classification based on multivariate pdf estimation\nof the eddy component only of NH flow anomalies was also carried out by Molteni\net al. (QJRMS, 1989, in press)\n18","It is remarkable that the distinct NH flow regimes obtained by most of\nthese rather different classification approaches exhibit substantial similari-\nties. Qualitatively, there is a tendency for localized features, in either the\nPacific or the Atlantic sector. This localization confirms synoptic experience\nand intuition, but is not built into any of these classification schemes of\nhemispheric anomaly maps, as it is into the one-point correlation maps of\nWallace and Gutzler (MWR, 1981) or into the rotated principal component analy-\nsis of Horel (MWR, 1981) and of Barnston and Livezey (MWR, 1987)\nQuantitatively, the multivariate pdf maxima of Kimoto (1989) are in good\nagreement with the centers of the clusters of Mo and Ghil (1988) for the entire\nNH data set, while the minimum-tendency loci of Vautard (1989) are in very\nsatisfactory agreement with the Atlantic-sector results of Kimoto (1989) The\nrespective regimes are not in exact, one-to-one correspondence, nor are the\nsimilarities, where they exist, perfect. But, given the differences in both\nmethods and data sets, it is the similarities, as far as they go, which are\nstriking, not the differences.\nTransition between regimes\nHaving obtained a reasonable classification of flow patterns, one can\nstudy the transitions among them using a Markov-chain description (Spekat et\nal., Meteor. Rundsch., 1983) Such a description has been given for SH flows,\nbased on their pattern-correlation classification (Mo and Ghil, 1987) For NH\nflows, Mo and Ghil (1988) studied the Markov chain of transitions between their\nclusters, and Kimoto (1989) that between his pdf maxima. Preferred transitions\noccur, by-and-large, between regimes lcoalized in the same sector, but some\ninteresting sector-to-sector transitions are also statistically significant\n(Kimoto, 1989) To sharpen the statistical-significance estimates for the\ntransition matrices of NH and SH Markov chains, Vautard, Mo and Ghil (JAS,\n1989, submitted) have carried out Monte Carlo simulations, confirming overall\nthe earlier results of Mo and Ghil (1987) for SH regimes and of Mo and Ghil\n(1988) for NH regimes.\nOver the smaller, North Atlantic-European region, Vautard (1989) has also\nstudied preferred precursors and successors, using a combination of minimum-\ntendency and pdf maximization techniques. Synoptic case studies for some of\nVautard's preferred transitions have been carried out by Orrhede (M.S. thesis,\nUCLA, 1989) These studies indicate certain synoptic developments well-known\nto European forecasters, as well as certain preferred sequences which have not\nattracted sufficient operational attention up till now.\nENSO effects\nTo estimate the effect of opposite phases of ENSO on the relative impor-\ntance of extratropical flow regimes, it is convenient to use the Southern\nOscillation Index (SOI) as an indicator of tropical air-sea interaction. SOI\nis a measure of Tahiti-minus-Darwin sea-level pressures and is anticorrelated\nwith SST anomalies in the eastern equatorial Pacific.\nFig. 1 is a bar diagram of the number of days associated with one of seven\nof Kimoto's (1989) Pacific regimes (upper panel) or one of his six Atlantic\nregimes (lower panel) This frequency distribution is shown separately for\n19","normal SOI values, within one standard deviation, + o, of the mean ; high\nvalues, and low values, SOI - o. For each set of frequencies,\nrepresented in the figure by an open rectangle, the expected values for a\nrandom distribution conserving the same total number of days in each regime,\nand the associated 95% confidence interval, are shown by a full circle brack-\neted by an \"error bar\". The latter data were obtained by using 100 Monte Carlo\nsimulations of the entire sequence of maps, for each regime (see Kimoto, 1989,\npp. 78-79 and 209-210, for details).\nThe figure indicates clearly that the tropical index does not determine\nwhich regime will appear (cf. also Namias and Cayan, 1984), but only modifies\nthe probability of appearance for certain flow regimes. For example, in the\nupper panel, P1 resembles the Pacific-North American (PNA) pattern of Wallace\nand Gutzler (MWR, 1981), while P7 is dominated by a ridge over Alaska; both\ntheir frequency distributions are significantly skewed at the 95% level: P1 is\npreferred by negative SOI and P7 is inhibited by positive SOI. This result is\nconsistent with that of previous investigators (e.g., Horel and Wallace, MWR,\n1981), and represents a more objective description of the complex relation\nbetween the tropical, interannual signal and the extratropical, intraseasonal\ncirculations than a simple one-to-one association between El Niño and PNA, for\ninstance.\nThe effect of tropical SST anomalies in the Pacific sector is weaker on the\nAtlantic regimes (lower panel) Still, the Atlantic block associated with A2 is\nfavored by negative SOI, although not at the 95% level, while the zonal flow in\nA6 is inhibited by negative SOI. Thus, on the whole, positive SST anomalies in\nthe tropics favor blocked flow, while negative ones favor zonal flow, separately\nin the Pacific and Atlantic sector.\nConcluding remarks\nIntraseasonal variability is significantly larger than interannual variabil-\nity in the NH extratropics (Kimoto, 1989; Mo and Ghil, 1989, in preparation).\nHence it is necessary to examine carefully the former before studying the\nlatter.\nThe efects of ENSO on the probability distribution of persistent anomalies\nin the NH circulation are small, but statistically significant for some flow\nregimes. Transition probabilities to or from these regimes should also be\naffected. The use of SOI as a predictand for long-range forecasting in the\nintraseasonal range is thus worthwhile, but has to be subject to considerable\ncaution, due to the nonlinearities apparent in tropical-extratropical inter-\nactions.\n20","a)\nP1\nP2\nP3\nP4\nP5\nP6\nP7\n300\n300\n200\n200\n100\n100\no\nb)\nA1\nA2\nA3\nA4\nA5\nA6\n300\n300\n200\n200\n100\n100\no\n0\n1. Relation between SOI and regional regimes. Number of be-\nFig.\ndays that\nlong to each regional regime (P1 - P7 for the Pacific sector; A1 - A6 for the\nAtlantic-Eurasian sector) is counted for SOI - o, u - o < SOI < + o, and\nSOI, separately. Confidence intervals are estimated by a Monte Carlo method\n(from Kimoto, 1989)\n.\n21","Bifurcations and El Niño in a hybrid coupled GCM\nDavid Neelin\nDepartment of Atmospheric Sciences\nUniversity of California, Los Angeles\nLos Angeles, CA, 90024\nIntroduction\nA hybrid coupled general circulation model for El Niño studies has been described in\nNeelin (1989; N hereafter). The model consists of an ocean general circulation model in\nan idealized Pacific ocean basin, coupled to a two-level, steady state atmospheric model\nwith parameterized moist processes. The combination of an ocean GCM with a simpler\nsteady atmospheric model is refered to as a hybrid coupled GCM or HGCM. The choice of\ncoupling to a steady atmospheric model permits the examination of a deterministic coupled\nsystem unperturbed by atmospheric variability, which nonetheless has the full nonlinearity\nof an ocean GCM.\nInterannual oscillations are found in this model which resemble the El Niño-Southern\noscillation (ENSO) cycle in a number of ways. The period is three to four years for a\nreasonable range of values of the coupling. The temperature anomalies and associated\ncurrent and heat content anomalies are equatorially-trapped and exhibit both eastward\nand westward propagation during different parts of the cycle. Not all aspects of this\noscillation are realistic; during the cold phase, the equatorial cold tongue extends further\nwest than in observations, cutting the Indonesion warm pool into two lobes. The warm\nphase anomalies also tend to occur in the central Pacific rather than in the east, especially\nat low coupling.\nBifurcation diagram of the coupled system\nIt is shown numerically in N that the oscillation arises by a Hopf bifurcation of the\nclimatological stationary point of the coupled system. The bifurcation can arise by varying\neither parameters affecting the climatology or by changing the strength of the wind stress\nanomalies. The atmospheric model is coupled in a flux-corrected mode, so it is possible to\nmodify the strength of the stress anomalies independently from the climatological stress.\nThe magnitude of the modified stress anomaly pattern relative to that of the unmodified\natmospheric model is termed the relative coupling coefficient, denoted u. It provides a\ncrude, but understandable, parameter by which to examine the range of behaviour of the\ncoupled system.\nBelow the bifurcation, which occurs for a value of H between 0.6 and 0.7, the\nclimatology is stable and the El Niño-period oscillation is the most slowly-decaying mode.\nAbove the bifurcation, the climatology is unstable to the El Niño-period oscillation which\nequilibrates into a periodic limit cycle of succesive warm and cold phases. The nature of the\nbifurcation implies that the period-about 37 months-and spatial form change relatively\nlittle, at first, as the coupling is increased while the amplitude of the limit cycle increases.\nFor higher values of the coupling, the period of the El Niño-limit cycle increases to about\nfour years at u = 1.0, corresponding to the unmodified atmospheric model. Furthermore,\n22","the limit cycle itself goes unstable to a higher frequency coupled Kelvin wave mode and\nthe motion becomes doubly-periodic with a 5-6 month oscillation superimposed during the\nwarm phase of the El Niño-period oscillation. There is some doubt as to whether this\ncoupled Kelvin wave mode corresponds to any observed feature and so it may be that the\nlower values of the coupling are more physically relevant. Nonetheless, it can be instructive\nto examine the response of the model at slightly higher coupling where the nonlinear effects\nmay be expected to be even stronger.\nFig. 1 shows the results of a run with a relative coupling coefficient u = 1.1, i.e. wind\nstress anomalies 10% stronger than those of the unmodified atmospheric model. Otherwise,\neverything is as in Run-II of N. The run is begun from the climatological stationary point\nwith a wind stress perturbation applied during the first month. The solution evolves toward\na doubly-periodic attractor as in the u = 1.0 run, but the period has increased from just\nunder 4 years to about 5 years. If the coupling is further increased to u = 1.2, the system\napparently undergoes a global bifurcation in which the El Niño-limit cycle disappears and\nis replaced by a spiral out from the unstable climatological stationary point to a Kelvin\nwave limit cycle about a different stationary point which corresponds to an eternal warm\nphase. Furthermore, this same stationary point (and the Kelvin wave oscillation about\nit) exists concurrently with the El Niño-limit cycle at lower values of the coupling. This\nis demonstrated by a second run at u = 1.1 but beginning from warm initial conditions\n(chosen to be close to the stationary point of the u = 1.2 run). The results of this run are\nshown in Fig. 2. Although the run was only continued for 9 years, it is clear that the El\nNiño-like oscillation is completely absent and that, aside from the 6 month Kelvin wave\noscillation, the time series is stationary over this interval.\nThese numerical results are not conclusive, but do admit of a straight-forward\ndescription in geometric terms in the model parameter-phase-space. A simple bifurcation\ndiagram consistent with these results is drawn schematically in Fig. 3a). The coupled\nKelvin wave oscillation has been omitted for clarity. The variables 41 and 42 may be\ntaken to be the linear normal modes which are neutrally stable at the primary Hopf\nbifurcation, although in practice it is not possible to carry out such a computation for the\nHGCM. The climatological stationary point is labelled CLIM. Above the Hopf bifurcation,\nlabelled H, it is unstable, giving rise to the El Niño-limit cycle. As the limit cycle increases\nin amplitude with higher coupling, the warm phase approaches a \"warm\" saddle node\nstationary point of the coupled system, marked WSN, whose presence may be inferred by\nthe dramatic increase in the length of the warm phase as u is increased beyond about\n0.8. Since the flow goes to zero at the stationary point, and since the flow is continuous in\nphase space, trajectories passing close to it necessarily move very slowly. As the increase in\ncoupling brings the warm phase of the limit cycle close to the saddle node, the oscillation\ntakes on the asymmetric nature characteristic of a relaxation oscillation. It appears that\neventually the limit cycle and the saddle node annihilate each other in a global bifurcation,\nleaving only the stable warm stationary point, marked WS, and the unstable climatological\nstationary point at high coupling.\nAt u = 1.1, there are two attractors-the limit cycle and the warm stationary point.\nThe range over which this two-attractor configuration exists has not yet been shown,\nalthough it is known not to exist at zero coupling. The saddle node bifurcation marked\n23","\"?\" in Fig. 3a) is a plausible but undemonstrated structure consistent with this. The\nsection through phase space at u = 1.1 in Fig. 3b) shows the El Niño-limit cycle together\nwith the stable warm stationary point, the unstable climatological stationary point and the\nhypothesized warm saddle node. Because of the two basins of attraction, this configuration\nin phase space may be expected to give a very complex time series in the presence of\natmospheric noise or, possibly, other sources of variability not currently present in the\nmodel, such as the seasonal cycle.\nAlthough a number of questions remain to be resolved, these results appear to be\npotentially significant for the predictability of El Niño and the examination of similar\nbifurcation diagrams in simpler models is being undertaken.\nThe fast-wave limit\nIt has been suggested (Schopf and Suarez, 1988) that the delay time due to the transit\ntime of equatorial waves across the basin is significant for the time scale and existence of\nEl Niño oscillations. Battisti and Hirst (1989) have shown that this effect appears to be\nimportant in the model of Cane and Zebiak (1985). On the other hand, simple modeling\nresults with a similar shallow water model with a mixed layer suggest that if the coupling\noccurs across a sufficient portion of the basin to permit propagation of SST anomalies,\na slow SST mode can arise which does not depend on wave transit times and can exist\neven in the limit where the oceanic adjustment process occurs rapidly compared to the\nevolution of interannual mode. This is referred to as the fast-wave limit (Neelin, 1989; in\npreparation).\nThe relative importance of these two mechanisms was tested in the HGCM by means\nof distorted physics experiments. By introducing artificial factors multiplying the time\nderivatives of the momentum equations, it is possible to modify the equatorial wavespeed\nwithout changing the dynamics which prevails once the ocean comes into balance with\nthe wind stress. Doubling the wavespeed causes only a small reduction in the period of\nthe El Niño cycle. Quadrupling the wavespeed causes no further change. The form of\nthe oscillation changes very little in these experiments. Thus the time delay associated\nwith wave transit across the basin plays only a secondary role in determining the time\nscale of the interannual oscillation in the HGCM and is not essential to its existence. The\npossibility remains that for other parameter settings, or in presence of the seasonal cycle,\nwave transit times might play a more important role.\nReferences\nNeelin, J.D., 1989: A hybrid coupled general circulation model for El Niño studies.\nJ. Atmos. Sci. (accepted).\nBattisti, D.S. and A.C. Hirst, 1989: Interannual variability in the tropical\natmosphere/ocean system: influence of the basic state and ocean geometry. (submitted to\nJ. Atmos. Sci.).\nCane, M.A. and S.E. Zebiak, 1985: A theory for El Niño and the Southern Oscillation.\nScience, 228, 1084-1087.\nSchopf, P.S. and M.J. Suarez, 1988: Vacillations in a coupled ocean-atmosphere model.\nJ. Atmos. Sci., 45, 549-566.\n24","TIME (YEARS)\n3\nTIME (YEARS)\n12\nI\n27.5\n1. Evolution of sea surface temperature along the equator for a the run unstable of the\ncoupled Figure model at relative coupling coefficient fl. = 1.1. Initial condition is from\nclimatological stationary point with an initial wind stress perturbation.\n25","TIME (YEARS)\n1\nI\n2\nI\nFigure 2. As for Figure 1, but beginning from a warm initial condition.\n26","WS\n?\nWSN\na)\nLC\non\nH\nCLIM\n4\n1\nY\n2\n0.6\n0.7\n0.8\n1.0\n1.1\n1.2\nRELATIVE COUPLING COEFFICIENT, u\nb)\nCLIM\nWSN\nWS\nY\n2\nY\n1\nFigure 3. a) Bifurcation diagram of the hybrid coupled general circulation model as\na function of the relative coupling coefficient. b) Cross-section of the bifurcation diagram\nat fl = 1.1.\n27","Comparison of 30 Years of Data with Robinson\nClimatology Clearly Distinguishes E1 Nino Events\nMargaret K. Robinson\nCompass Systems, Inc.\nBetween 1979 and 1985 the National Science Foundation funded a subjective\nanalysis of southern hemisphere ocean temperatures and salinities that began\nwith the sparce observations and synthesized the results from over 400 published\npapers by 155 authors from 9 countries. This work followed completion of the\nnorthern hemisphere analysis and was funded to provide a consistent world-wide\nocean numerical climatology usable both on computers as a high resolution\nspatial and temporal grid and for the production of colored atlases. With\navailable southern hemisphere data, objective methods alone did not produce\nacceptable subsurface fields. The northern hemisphere values are published in\nthe Robinson (1976) and Robinson, Bauer and Schroeder (1979) atlases.\nTo demonstrate the validity of this climatology, based on data collected be- -\ntween 1928 and 1981, this paper presents a comparison of 30 years of raw data\nwith our climatology along the eastern Pacific Ocean coast from the equator to\n8°S extending 5° of longitude from the coast. Comparisons of data were also\nmade in 10° by 10° areas surrounding Tahiti, Fiji and New Caledonia and a 5°\nlatitude strip along the east coast of Australia between 25° and 30°S. Figure 1\nshows areas where mean anomalies are computed between individual observations\nand the 1° square atlas values in a given month at the surface, 60M and 150M.\nThe comparisons show positive anomalies in the E1 Nino years 1957, 1965,\n1972, 1976 and 1983 in the eastern Pacific coastal area. In each of these years\nthe large positive anomalies extended below 60M and strong events could be de-\ntected at 150M.\n28","The 1957 E1 Nino was poorly defined because of few observations, but at 60M\nin December and in January 1958 mean positive anomalies approached 5°C. In 1965\nmean anomalies reached 4°C at the surface and 60M and 2°C at 150M. Observed\nanomalies in some cells as shown by the range bracket on the time series plots\nreached 7°C at the upper 2 levels and 3°C at 150M. All values were back to\nnormal by July 1966.\nThe 1972 E1 Nino extended from February to December with mean anomalies at\nthe surface and 60M, ranging from 2° to 5°C and observed anomalies were again up\nto 7°C. At 150M mean anomalies ranged between 1° and 3°C, observed values to\n5°C. 1972 had the highest observed anomalies up to that date.\nIn November 1982 the disastrous 1983 E1 Nino was first observed, with mean\nsurface anomalies of 5° to 8°C in all months to August 1983. There were no data\nin September or October, then by November anomalies dropped to 1°C and in 1984\nbecame negative. At 60M mean anomalies exceed 10°C, observed anomalies showed\n13°C. At 150M in December to February mean anomalies ranged between 2° and 5°C,\nwith individual values up to 8°C.\nIn 1985 on a visit to Peru I saw the enormous damage these warm waters had\ncaused. Millions of anchovies had died and washed up on the beaches. Guano\nbirds in great numbers either starved or died from eating rotten fish. The\nfishing industry itself was ruined. The fish meal factories went bankrupt and\nthe poultry industry, dependent on fish meal for food, collapsed.\nThe observed data in the Tahiti area, 10° -20'S, 145° to 155°W, collected\nprior to 1980 were so few that no inferences about E1 Nino effects in the area\ncan be made. The data collected between 1980 and 1984 are mainly colder than\nthe atlas values with mean negative anomalies at 150M ranging up to 3°C in 1983.\n29","A similar situation was seen in the Fiji Island area, 10°-20°S, 175°W to\n175°E. Data collected increased beginning in 1979. Here again negative anom-\nalies at 150M increased from 1980 to a maximum in April 1983 of 4°C. Small\nnegative anomalies occurred at the surface in 1983 and these increased at 60M\nto 2°C.\nThe New Caledonia area, 10°-20°S, 160°-170°E, shows an increased data col-\nlection effort beginning in mid-1979. Since 1979 it shows small negative anom-\nalies predominant at the surface, increasing at 60M to a maximum of 1°C at 60M\nand up to 3°C at 150M. A large value of 5°C in December appears to be a sus-\npicious value from the standard deviation curve.\nThe range of values along the Australian coast from 25° to 35°S extending\n5° of longitude offshore is too large to draw conclusions about any relationship\nbetween this area and the ENSO elsewhere. The narrow East Australian Current\ndata should have been segregated from the offshore gyre. In spite of the high\nvariability among the observed data due to combining data from two different\nwater masses, the mean anomalies do not depart unduly from the atlas means when\ncompared to data in individual 1° squares.\nThe Southern Hemisphere Atlas values are only available on tape. Northern\nHemisphere Atlases are available both in published form and on tape. Since 1982\nNOAA has used our sea surface temperature values in the \"Oceanographic Monthly\nSummary\" series as a base for computing monthly anomalies in coastal areas. The\nprimary data source of these atlases was bathythermograph observations collected\nbetween 1941 and 1966 and data collected by other oceanographic instruments- --\nhydrocasts, XBTs, AXBTs, CTDs and STDs--between 1925 and 1980 in both real time\nand in delayed mode were also used to produce the atlases. These data were\navailable as the MOODS2 tapes.\nThe atlases describe the seasonally heated ocean layer and allow the study\nand evaluation of climatic changes.\n30","(PLOTTING SHEET No 2601\nPILOT CHART OF THE SOUTH PACIFIC OCEAN\n(PLOTTING SHEET\nNo.2601","TIMES\n3.990 -\n030 T\n030 - T\n930\n3 930\n930\n0.030\n32","12\nneed\nnew\n3 920 for 3 930\n020 4'\n3\n3 930\nof\n3 930\n33","n\n34","Forecast experiments of the 1982-83 El-Niño\nwith a coupled air-sea model\nJ. Sirutis and K. Miyakoda\nGeophysical Fluid Dynamics Laboratory/NOAA\nPrinceton University\nPrinceton, New Jersey 08540\n1. Introduction\nRecently the focus of our interest in the area of extended weather prediction has ex-\npanded from a time scale of one month to a season (three to six months). A coupled air-sea\nmodel has been developed for this purpose. As part of a model shake down, some attempts\nhave been made at forecasting the El Niño of 1982-83. Some of the results of these forecast\nexperiments will be described in this paper.\n2. Model Description\nBoth the atmospheric and ocean models used in the coupled system are complete glo-\nbal GCMs with full physics.\nThe ocean model is an adaptation of the model developed by GFDL's oceanographic\ngroup. The major departures of the present model from the original are the resolution, and\nthe treatment of the sub-grid scale turbulent processes. The model has 15 levels in the verti-\ncal, and a horizontal resolution of one degree everywhere except in the equatorial belt (10N-\n10S) where the resolution is 1 X 1/3 degrees. The horizontal diffusion is a non-linear Smagor-\ninsky type, and the Mellor-Yamada level 2.5 turbulent closure scheme is used in modeling\nthe vertical mixing.\nThe atmospheric portion of the coupled system is a spectral model with a rhomboidal\ntruncation of 21 waves, and 9 levels in the vertical. The model has all of the usual physical\nparameterization packages, including radiation, moist convective adjustment and a hydrologic\ncycle over land. The treatment of the vertical sub-grid scale turbulent processes is the same\nas in the ocean model.\nThe two models interact with each other at twelve hour intervals. The ocean model\nprovides the atmospheric model with 12 hour mean sea surface temperatures, while the at-\nmospheric model provides the ocean model with 12 hour mean wind stress, heat flux and net\ndownward solar radiation at the surface.\n3. Experiments and Initial Conditions:\nSome results from three experiments will be presented here.\nA) Sst nudging: This is a five year integration, starting from 1 March 1979, with\nnewtonian nudging of the model sst toward the observed value. The purpose of\nthis integration was twofold: 1) to generate initial conditions for the ocean model;\n2) to generate a heat flux correction term for the coupled model.\nB) Forecast: The starting time for the forecast is 1 January 1982. The length of the\nforecast is 12 months. The initial conditions for the atmospheric portion of the cou-\npled model were generated from NMC analysis. The initial conditions for the\n35","ocean came from the sst nudging integration.\nC) Forecast with heat flux correction: Same as above, except that an annual\nmean correction term has been added to the ocean models heat flux equation.\n4. Results\nFigure 1 shows the sst anomaly in the Pacific Ocean for December 1982, the last\nmonth of the forecast. The observed distribution is shown in the upper left. The anomaly for\nthe sst nudging experiment is shown in the upper right. As expected, the distribution is very\nsimilar to the observed. The anomaly for the forecast is shown in the lower left. Here several\nthings can be noticed. One, that the model has an overall cooling bias. Two, that despite\nthis cooling bias, the model does show skill at forecasting the El Niño, as is evidenced by\nthe warm anomalies in the eastern Pacific to the north and south of the equator. However,\nthe El Niño in the forecast is obscured along the equator by cold water which is the result of\nunrealistically strong upwelling in the model off the coast of South America. An attempt to\nremedy this model bias was made by adding a flux correction term to the ocean models heat\nflux equation. The results of this forecast are shown in the lower right of fig. 1. As can be\nseen, the cooling in the model has been substantially reduced, and the simulation of the El\nNiño is somewhat better than in the previous forecast.\nThe zonal component of the wind at 1000 hPa across the Pacific Ocean for December\n1982 is shown in Figure 2. Focusing on the tropics, we can see that in the NMC analysis (\nupper left), the westerlies in the tropics, intruding into the Pacific basin from the west, ex-\ntend across the entire basin. In the sst nudging experiment ( upper right), the westerlies in\nthe tropics extend eastward to about 135W, with easterly flow in the eastern portion of the\nbasin. In the two forecasts ( lower left and lower right) the intrusion of the westerlies into\nthe Pacific basin extends only to about 170E-180E; the rest is easterlies.\nIn general, the coupled model, in this case, does not do a very good job of simulating\nthe winds across the Pacific basin. The intrusion of the westerlies into the Pacific basin in the\ntropics is too weak, while the easterlies in the eastern part of the basin are too strong.\nFor a more quantitative evaluation of the models forecast ability, verification scores\nfor T at 1000 hPa and sst are show in Figure 3. The verification domain is the Pacific Ocean\nin the tropics, 10S to 10N and 130E to 70W. The verification scores for the sst nudging ex-\nperiment are included here as a reference, since they probably represent an upper limit on\nhow well the model can be expected to do.\nCorrelation coefficients for sst are show in the lower left. As expected, the sst nudg-\ning experiment shows the highest correlation with observations. The correlation increases\nwith time as the El Niño develops and the size of the anomalies increases. The correlation\ncoefficients for the two forecasts are close to each other at the start of the forecast period.\nStarting in May, the two curves diverge, with the forecast without the flux correction show-\ning somewhat more skill than the forecast with the flux correction.\nThe rms errors for sst ( lower right) represent mainly the models cooling bias. Unsur-\nprisingly, the sst nudging case has the lowest rms errors ( about 0.4C), followed by the fore-\ncast with the heat flux correction (about 0.9C), followed, in turn, by the forecast without the\nflux correction which has the highest rms error. The rms errors for the first two cases are fair-\nly constant with time. In the forecast without the flux correction the rms error grows with\ntime, reaching a value of about 2.5C at the end of the forecast period.\nThe correlation coefficients for T 1000 hPa (upper left) are close to each other for all\n36","three cases. The rms errors for T 1000 hPa for the sst nudging experiment and the forecast\nwith the flux correction are also close to each other, and smaller than for the forecast without\nthe flux correction\n5. Summary and Conclusions\nSome attempts at forecasting the El Niño, of 1982-1983 have been made, as part of\nthe development and testing of a coupled air-sea model designed for forecasts of a seasonal\ntime scale. The model, despite its cooling bias, does exhibit some skill in its ability to fore-\ncast the El Niño 12 months ahead. It is most interesting to note that the El Niño signal\nemerges even in the case of the forecast without any heat flux correction.\n37","upper\nright - sst nudging experiment; lower left - forecast;; and lower right - forecast with the heat flux correction. The contour interval is\n80\n80\nFigure 1. Sst anomaly in the Pacific Ocean for December 1982 , the last month of the forecast. Upper left - the observed distribution;\n100\n100\n120\n120\nForecast with heat flux correction\n....\n10\n140\n140\n0.0\nlongitude\nlongitude\ndays\nto\n10\nSst nudging\n160\n160\n-\none degree. Positive values are solid; negative values are dashed; the zero line is the heavy solid line.\n5.0\n50\n180\n180\n160\n160\n140\n140\nn\n&\n40\n20\n0\n-20\n-40\n40\n20\n0\n-20\n-40\n80\n80\n100\n100\nObserved (CAC) - December 1982 sst anomaly\n120\n120\n....\n10\n2X\n140\n140\n50 40 10 again 00\nlongitude\nlongitude\ndepress\n30\n160\n160\n10\nForecast\n180\n180\n160\n160\n140\n140\nc\n40\n0\n20\n0\n-20\n-40\n40\n20\n-20\n-40","Figure 2. U at 1000 hPa in the Pacific Ocean for December 1982 , the last month of the forecast. Upper left - the observed distribution; upper\nright - sst nudging experiment; lower left - forecast; and lower right - forecast with the heat flux correction. The contour interval\n80\n80\n100\nw\n100\nis two m/sec. Positive values (westerlies) are solid; negative values (easterlies) are dashed; the zero line is the heavy solid line.\n120\n120\nForecast with heat flux correction\n140\n140\nlongitude\nlongitude\nSst nudging\n160\n160\n180\n180\n160\n160\n-40 I\n140\n140\nD\nX\n40\n20\no\n-20\n40\n20\n0\n20\n-40\n8\n,4\n80\n80\n100\n100\nNMC analysis - December 1982 u 1000 hPa\nC\n120\n120\n140\n140\nlongitude\nlongitude\n160\n160\nForecast\n180\n180\n160\n160\n140\n140\n-40 /\na\n40\n20\n0\n-20\n40\n40\n20\n0\n-20","Figure 3. Verification scores for monthly mean temperature at 1000 hPa and sea surface temperature. The verification domain is the tropical Pacific\nOcean , 10S to 10N and 130E to 70W. Upper left - correlation coefficients for T 1000 hPa; upper right - rms error for T 1000 hPa; lower\nleft - correlation coefficient for sst; and lower right - rms error for sst. The solid line is the forecast; the dashed line is the forecast with the\nPoliget\nQdfcst\nnudge\n12.\n12.\nfcst\nfcst\n11.\n11.\nT1000 Pacific (10S to 10N, 130E to 70W)\n10.\n10.\nSST Pacific (10S to 10N, 130E to 70W)\n9.\n9.\n8.\n8.\n7.\n7.\nMonths\nMonths\n6.\n6.\n5.\n5.\n4.\n4.\n3.\n3.\n2.\n2.\n1.\n1.\nO.\no.\n5.0\n4.5\n4.0\n3.5\n3.0\n2.5\n2.0\n1.5\n1.0\n0.5\n0.0\n5.0\n4.5\n4.0\n3.5\n3.0\n2.5\n2.0\n1.5\n1.0\n0.5\n0.0\nheat flux correction; and the dotted line is the sst nudging experiment.\nQdfcst\nQdfcst\nnudge\nnudge\n12.\n12.\nfest\nfcst\n11.\n11.\nT1000 Pacific (10S to 10N, 130E to 70W)\n10.\n10.\nSST Pacific (10S to 10N, 130E to 70W)\n9.\n9.\n8.\n8.\n7.\n7.\nMonths\nMonths\n6.\n6.\n5.\n5.\n4.\n4.\n3.\n3.\n2.\n1.\n1.\nd.\nd.\n1.0\n0.9\n0.8\n0.7\n0.6\n0.5\n0.4\n0.3\n0.2\n-0.2\n1.0\n0.9\n0.8\n0.7\n0.6\n0.5\n0.4\n0.3\n0.2\n0.1\n0.0\n-0.1\n-0.2\n0.1\n0.0\n-0.1","ENSO AND CO2 CLIMATE CHANGE\nIN A COUPLED OCEAN-ATMOSPHERE GCM\nGerald A. Meehl\nNational Center for Atmospheric Research *\nP. O. Box 3000\nBoulder, Colorado 80307\nIntroduction: Only recently have global coupled ocean-atmosphere general circulation\nmodels (GCMs) been capable of internally generating El Niño-Southern Oscillation (ENSO)\nphenomena (Sperber et al., 1987; Meehl, 1989; Philander et al., 1989). ENSO is a ba-\nsic mode of weather and climate variability with effects that are felt all over the planet.\nThese global coupled climate models can provide the first look at how ENSO character-\nistics we now observe, including associated extreme events that are of interest for climate\nimpacts studies, may change with increased CO2.\nThe model considered here includes a global spectral atmospheric GCM with rhom-\nboidal 15 (R15) resolution (about 4.5° latitude by 7.5° longitude), realistic geometry, nine\nlayers in the vertical, and parameterized land-surface processes. The global ocean GCM\nhas a latitude-longitude resolution of 5°, realistic geography, topography, four layers, and\na simple thermodynamic sea-ice formulation. The results described here are from two 30-\nyear experiments (Washington and Meehl, 1989). One is run with present amounts of\nCO2 (1xCO2 ), and one with instantaneously doubled CO2 (2xCO2 ). Some of the re-\nsults are derived from the last ten years of each integration. Recently, the experiments\nhave been extended five more years, and some results are for the last fifteeen years (years\n21-35).\nFigure 1 shows the model interannual variability in the tropical Pacific. Area aver-\nages of sea-surface temperature (SST) anomalies for the NINO3 area (90°W to 150°W,\n10°N to 10°S) for 1xCO2 (Fig. 1a) and 2xCO2 (Fig. 1d) show that SSTs in this region\noscillate about their respective means. However, the CO2 SSTs have experienced a\nmean warming of about a degree. Therefore, compared to the 1xCO2 mean, SSTs re-\nmain relatively warmer throughout the period.\nFigure 1b and e show sea-level pressure (SLP) anomalies for the NINO3 area and\nfor an area in the western tropical Pacific (\"Indonesia\"-110°E to 155°E, 10°N to 13°S)\nfor 1xCO2 and 2xCO2 , respectively. Both show the opposition of anomalies characteris-\ntic of the Southern Oscillation.\nA Southern Oscillation Index (SOI) computed for model grid points near the actual\nlocations of Darwin, Australia, and Tahiti, French Polynesia, (the stations used to com-\npute the observed SOI) shows oscillatory behavior (Fig. 1d and e) with periods of nega-\n* The National Center for Atmospheric Research is sponsored by the National Science\nFoundation.\n41","tive SOI generally corresponding to relative maxima of SST in the eastern tropical Pacific\n(NINO3 area, Fig. 1a and d), and positive SOI occurring with relative minima of NINO3\nSST, as observed. Since the SOI is computed as a relative difference, there is no indica-\ntion of the mean climate change due to doubled CO2 that was apparent in SST (higher in\nthe tropics) and SLP (lower in the tropics). Clearly, the Southern Oscillation is still func-\ntioning in the xCO2 experiment in the tropics, but it is oscillating about a new mean.\nWarm and cold events in the coupled model have been previously defined as an\nSST anomaly in the NINO3 area greater than +0.25 and SOI less than -0.5 and NINO3\nSST anomaly less than -0.25 and SOI greater than +0.5, respectively (Meehl, 1989).\nIn that paper, it was shown that warm and cold events in the model have similar sea-\nsonal timing to those observed with initiation in northern spring in the eastern Pacific.\nBy the December-January-February (DJF) season going from the year of initiation (year\n0) to the year after (year+1) the SST anomaly has moved out near the Dateline. This\nDJF season in the so-called mature phase of the ENSO cycle (Rasmusson and Carpen-\nter, 1982), is often associated with teleconnections to the northern extratropics. For the\n1xCO2 experiment in the 15-year period from years 21 to 35, there are four such DJF\nwarm-event seasons (years 23, 24, 27, 35).\nUsing the same definition of warm and cold events applied to the 2xCO2 exper-\niment in years 21 to 35 and using the 1xCO2 long-term mean as a reference state, it is\napparent from Fig. 1 that there can be no cold events defined as such since the mean\nhas shifted and all years are relatively warmer than the 1xCO2 climate. Four warm-event\nDJF seasons in the 2xCO2 experiment, years 26, 29, 32, and 34, still satisfy the definition\nsince the 2: CO2 SOI oscillates about zero (Fig. 1f).\nComposites are computed for the four warm-event DJF seasons in each of the\ntwo experiments. The averages of the 1xCO2 warm events and CO2 warm events\nare subtracted from the CO2 and 2xCO2 15-year mean northern winter SSTs, re-\nspectively. In this way, the changes in ENSO phenomena can be assessed in relation to\nthe mean climate conditions for the 1xCO2 ENSOs using the 1xCO2 long-term mean,\nand for the 2xCO2 ENSOs using the 2xCO2 long-term mean. Observed SST anoma-\nlies for the recent warm event during December 1987 are shown in Fig. 2a. The typical\nobserved warm-event pattern of positive SST anomalies in the central equatorial Pacific\nand negative SST anomalies to the north and southwest is also present in the 1xCO2 and\nCO2 composites (Fig. 2). Even though the mean SST has warmed in the 2xCO2 case\n(Fig. 1d), the amplitude and pattern of the SST anomalies for the CO2 ENSO events\n(Fig. 2b) are similar to those for the xCO2 ENSO events (Fig. 2a).\nSLP anomalies for the same season are shown in Fig. 3. There is good correspon-\ndence between the amplitude and pattern of the SLP anomalies in the tropics between\nthe 1xCO2 and 2xCO2 ENSOs with relatively low SLP over the warm water in the east-\nern Pacific and high SLP in the western Pacific and Indian Ocean, as observed (Fig. 3a,\nvan Loon, 1986). But the extratropical response in the 2xCO2 warm events is funda-\nmentally different from both the observed and the 1xCO2 composites. Relatively high\n42","pressure dominates most of the high northern latitudes in the CO2 case compared to\nlow pressure in the observed and 1xCO2 case, while relatively low pressure appears near\n60-70°S in the 2xCO2 case (Fig. 3c) where positive SLP anomalies had appeared in the\nobserved and 1xCO2 case (Fig. 3a and b). Even though these averages are formed from\nonly four samples in each case, the same pattern is evident in each of the individual sam-\nples.\nExperiments with linear baroclinic models have demonstrated that the response to\ntropical heating associated with ENSO can be altered by changing the basic atmospheric\nstate in the extratropics. Results presented here suggest this to be the case in the present\ncoupled GCM where the increase of CO2 has changed the basic atmospheric state from\nthat simulated in the CO2 case (e.g., warmer tropical troposphere, warmer northern\nhemisphere high-latitude lower troposphere, stronger subtropical upper-level westerlies,\nweaker lower tropospheric westerlies in the northern midlatitudes, etc.-see Washington\nand Meehl, 1989).\nThese experiments with a global coupled ocean-atmosphere GCM raise the possi-\nbility that, in an increased CO2 climate, ENSO phenomena in the tropics will occur in a\nsimilar fashion to what we now observe but with a shift in the mean and will have altered\nteleconnections to the extratropics compared to the present-day climate.\nReferences\nMeehl, G. A., Seasonal cycle forcing of El Niño-Southern Oscillation in a global coupled\nocean-atmosphere GCM. J. Climate in press (1989).\nPhilander, S. G. H., Lau, N. C., Pacanowski, R. C. and Nath, M. J., Two different simu-\nlations of Southern Oscillation and El Niño with coupled ocean-atmosphere general\ncirculation models. Phil. Trans. Roy. Soc. in press (1989).\nRasmusson, E. M. and Carpenter, T. H., Variations in tropical sea surface temperature\nand surface wind fields associated with the Southern Oscillation/El Niño. Mon. Wea\nRev. 110, 354-384 (1982).\nSperber, K. R., Hameed, S., Gates, W. L., and Potter, G. L., Southern Oscillation simu-\nlated in a global climate model. Nature 329, 140-142 (1987).\nVan Loon, H., The characteristics of sea level pressure and sea surface temperature during\nthe development of a warm event in the Southern Oscillation. Namias Symposium\nScripps Institution of Oceanography Reference Series 86-17 (Scripps Institution of\nOceanography, La Jolla, CA) 160-173 (1986).\nWashington, W. M. and Meehl, G. A., Climate sensitivity due to increased CO2: Experi-\nments with a coupled atmosphere and ocean general circulation model. Clim. Dyn.\n4, 1-38 (1989).\n43","d)\n2XCO2 NINO 3 SST ANOMALIES (IXCO2 MEAN) (3 SEASON RUNNING MEAN)\n2\na)\nMEAN-096\nIXCO2 NINO 3 SST ANOMALIES (1XCO2 MEAN) (3 SEASON RUNNING MEAN)\nSD-039\nMEAN-OOL\nSD-034\no\no\n-11\n-1\nOJF\nin\nDJF\nJM\nDJF\nJM\nDJF\nJJA\nDJF\nam\nDJF\nJAA\nDJF\nJJA\nDJF\nJUA\nDJF\nDJF\nJJA\nDJF\nJJA\nDJF\nJJA\nDJF\nJJA\nDJF\nJUM\nDJF\nJUA\n21\n-\n22\nI\n23\n24\n25\n26\n27\n20\nI\n29\nI\n30\nI\nI\n21\n-\n22\n-\n23\n24\n25\n26\n27\n28\n-\n29\nI\n30\nb)\ne)\nNNO3\nINDONESIA\nINDONESIA SLP ANOMALIES 2XCO2 (IXCO2 MEAN 113 SEASON RUNNING MEAN)\nSLP ANOMALIES IXCO2\nNINO\nI\no\nIN\no\n-1\nDJF JM DJF JUA DJF JUA DJF JUA DJF JUM DJF JUA DJF JJA DJF JM DJF JUA DJF JJA\n-1\n22\n23\n24\n25\n26\n27\n28\n29\n30\nI\n-2\nDJF\nJAA\nDJF\nJJA\nDJF\nDJF\nJUA\nJJA\nDJF\nJJA\nDJF\nJJA\nDJF\nJJA\nDJF\nJJA\nDJF\nJJA\nDJF\nJJA\n21\n22\n23\n24\n25\nI\n26\nI\n27\n28\n29\n30\nc)\nf)\nIX CO2 CAC METHOD,SOI, TAHITI MINUS DARWIN (MODEL)\n2XCO2 CAC METHOD, SOI, TAHITI MINUS DARWIN (MODEL) (3 SEASON RUNNING MEAN)\n2\no\n+\n-2\nJUA\nDJF\nJM\nDJF\nJUM\nDJF\nDJF\nJM\nDJF\nJUA\nDJF\nin\nDJF\nJM\nDJF\nJJM\nDJF\nis\nDJF\nJJA\nDJF\nJJA\nDJF\nJM\nDJF\nJJA\nDJF\nJUA\nOJF\nJUA\n21\n22\n23\n24\n25\n26\n27\n29\n30\nI\nI\n21\n22\nI\n23\n24\n25\n26\n27\n28\n29\n30\nFig. 1: a) Area-averaged seasonal SST anomalies for the NINO3 area in the east-\nern tropical Pacific for the 1xCO2 coupled model experiment; mean and\nstandard deviation are given in the right-hand corner; positive anomalies are\nhatched, negative stippled; b) area-averaged SLP anomalies for the NINO3\narea (solid line) and the Indonesia area (dashed line) for the 1x CO2 ex-\nperiment; c) Southern Oscillation Index (SOI) for grid points in the model\nnearest to Darwin and Tahiti for the 1xCO2 experiment: hatching indicates\npositive values, stippling negative; d) same as (a) except for the 2x CO2 ex-\nperiment minus the 1xCO2 long-term mean; e) same as (b) except for the\n2xCO2 experiment: f) same as (c) except for the 2 xCO2 experiment.\n44","a)\nSST ANOMALIES, DECEMBER, 1987 (WARM EVENT)\nDOE\n100\n90W\nDON\nSON\nSON\n60N\ng\nZabs\nB\n2\nSO\nSON\n30N\no\nSA\nTO\n0\no\n2.\nO\nO\nA\n305\n303\ndi\no\nC2\n2\ne\nals\nCO\n605\n2\n605\n905\nBOS\nO\nDOE\n180\nBOW\n0\nb)\nSST ANOMALIES, IXCO2 DJF WARM EVENT COMPOSITE MINUS IXCO 2 DJF MEAN\n90E\n180\n90W\niBON\n90\nSON\nSON\nBON\nSON\no\no\n0.0\n.03\nSOS\n305\n0.0\n0.0\nBOS\nSOS\n905\nSOS\nc)\nSST ANOMALIES 2XCO2 DJF WARM EVENT COMPOSITE MINUS 2XCO 2 DJF MEAN\nDOE\nNO\nBOW\nSON\nDON\nGOA\n60N\nSON\nSON\n0.0\n0\no\nthe\nSOS\n305\n605\n605\n0.0\nn\n905\nBOS\nFig. 2: a) Observed warm-event SST anomalies from December 1987 from Cli-\nmate Analysis Center; solid lines (stippling) are positive values, dashed\nlines are negative anomalies; b) SST anomalies from the coupled model\nfor 1xCO2 warm-event composite minus 1xCO2 15-year mean, DJF sea-\nson for mature phase (end of year zero-beginning of year plus one), four\nevents from a 15-year period; c) same as (b) except for 2xCO2 warm-event\ncomposites minus 2xCO2 long-term mean.\n45","a)\nDJF\nMEAN DEVIATION OF SLP IN WARM EVENTS\nDOE\n180\nBON\n90W\n90N\no\no\no\n60N\n3\n60N\n=2\n5\n4.\nCO\n2\no\n30M\n30N\n05\no\nQ5\nP\n05\n0.5\nO\n05\no\no\n2\n05\n-05\n-2\n30S\n305\n2\nO\no\n2\n605\n3\n605\n24\nDOS\n905\nO\nDOE\n180\nDOW\n0\nb)\nSLP ANOMALIES. IXCO2 2 DJF WARM EVENT COMPOSITE MINUS IXCO 2 DJF MEAN\nDON,\nDOE\n180\n90W\nBON\nGON\n60\na\nSON\nA\nH\n30M\n0\n0\n2\n7\nC\n0\n0\nBOS\n305\nT.P\nH\nGOS\n60S\nH\n905\n2\nBOS\nc)\nSLP ANOMALIES, 2XCO2 2 DJF WARM EVENT COMPOSITE MINUS 2XCO DJF MEAN\n2\nDONG\nDOE\nNO\nBOW\nSON\nIt\nCON\n8\nGON\nit\n2\nBON\n30N\nH\n0\no\nE.\nH\nBOS\n305\n1\n++\n&\n605\n605\ntit\nL\n905\no\n2\n905\nFig. 3: a) Observed SLP anomalies for DJF season during mature phase of warm\nevents minus long-term mean that does not include warm-event seasons\nfrom Van Loon (1986); stippling indicates positive anomalies; b) SLP anoma-\nlies for 1xCO2 warm-event composites minus 1x CO2 long-term mean for\nDJF during mature phase; c) same as (b) except for 2> CO2 warm-event\ncomposites minus 2 x CO2 long-term mean.\n46","BOUNDARY LAYER DEVELOPMENT OVER THE SOUTHEAST PACIFIC OCEAN\nUNDER ENSO AND ANTI-ENSO EPISODES\nGuillermo F. Almeyda, James W. Telford and Steve K. Chai\nAir Motions Laboratory, ASC\nDesert Research Institute\nReno, NV 89506\nI).\n- INTRODUCTION\nA theoretical model for the marine boundary layer has been developed and\napplied to the southeast Pacific ocean under the El Niño-Southern Oscillation\n(ENSO) and ANTI-ENSO episodes. This model simulates the marine boundary layer\ndevelopment associated with the physical processes taken place at the ocean\nsurface such as evaporation and at the top boundary of the marine boundary layer\nsuch as entrainment. This model is useful for examining the response of the\nmarine boundary to the warm (ENSO) and cold (ANTI-ENSO) episodes of any strength\npredicted by other climate models. This information is important because the\nmarine boundary layer will directly influence the weather condition in the nearby\narea.\nIS THE ENSO EVENT PREDICTABLE ?\nMuch effort in this direction has been performed by the oceanographic\ncommunity; however, with very limited results. Wyrtki, et al. (1976) predicted in\nOctober 1974 an El Niño event for early 1975, based on his theory (Wyrtki, 1975)\nand the southern oscillation index (Quinn, 1974). The event did not occurred as\npredicted, but a year later. More recently, Cane, et al. (1986) forecasted an\nENSO 1986/1987, based on another simple model (Cane and Zebiak, 1985, Zebiak and\nCane, 1987). The event did occur but the forecasted sea surface temperature (SST)\nanomalies for the central Pacific did not match the observed data, thus giving an\nimpression of pure coincidence (Kerr, 1987).\nOn the other hand, from the atmospheric point of view, the future state of\nthe atmosphere can only be predicted up to seven days ahead, by numerical weather\nprediction models. These models do not predict any ENSO or ANTI-ENSO events, but\nthe physical foundations are sufficient as to be complemented and adapted to\nconsider climate behavior. For a long range forecast of the climate range type,\nit becomes necessary to include the effects of the ocean. A new generation of\ncoupled ocean-atmosphere models (for example the six predictive levels) are under\nintensive studies to understand the effects of ENSO and ANTI-ENSO events; for\ninstance climate anomalies and climate changes at a regional and global scale.\nOCEANOGRAPHIC AND CLIMATE STATES\nThe actual state of the ocean surface and atmosphere, as derived from\nsatellites and vessels information (CAC, October, 1989), shows that, after\na\nminimum peak of the SST (Cooling episode) in the central and eastern Pacific in\nthe tropical region associated with strong easterly winds and a positive maximum\npeak in the southern oscillation index (SOI), there is a tendency toward\nnormalization and possibly a warming trend of the SST anomalies (to be\naccomplished possibly by yearly steps) associated with a weakening of the\neasterlies and a trend toward normal and negative values of the SOI. The\noceanographic models run lately (October, 1989) suggests that it is unlikely to\ndevelop an ENSO event until April 1990 at least.\nII). - MARINE BOUNDARY LAYER MODEL\nDue to the impossibility so far of making long range forecast of the\nocean-atmosphere system, and that there are serious attempts to predict ENSO\nevents with very restrictive models (Barnett, et al. 1988), we have attempted to\nnarrow our studies to the marine boundary layer and develop a very simple\nthermodynamic model which is able to simulate the development of this boundary\nlayer, providing that the sea surface temperature and winds are known; and use\nthis model to simulate the development of the marine boundary layer over the\nsoutheast Pacific ocean under the two peaks of the ocean-atmosphere natural\noscillation.\nA variety of models (Mixed layer models, large eddy models, and higher order\nclosure models) have been developed recently in order to advance our knowledge on\nthe physical processes, the structure and the prediction of the evolution, of the\n47","marine boundary layer (Lilly, 1968, Deardorff, 1976, Sommeria, 1976, Chen and\nCotton, 1983 Randal, 1980, Stage and Bussinger, 1981). We have developed a mixed\nlayer type of model by assuming a quasi-steady state condition. We demonstrate\nthat the vapor flux from the ocean controls the entrainment rate of overlying air\nin the development of the marine boundary layer. The energy and mass conservation\nlaws have been used in the development of the equations.\nENTRAINMENT AND MIXING\nThe entrainment and mixing processes, important observed phenomena at the\ntop boundary of the marine boundary layer, are treated theoretically to determine\nthe buoyancy of a resulting mixed parcel.\nA clear warmer and drier parcel (M1) of overlying air entrains into the\nmarine boundary layer top and mixes with a saturated or cloudy air parcel (M2).\nThe sensible heat from the entrained air parcel is used to evaporate the water\ndroplets in M2 so that the resultant temperature of the mixture could be warmer\nthan, equal to, or colder than the surrounding saturated or cloudy air.\nBased upon the first law of thermodynamics and the ideal gas laws, the\nmixing process has been modeled in a series of steps so that the virtual\npotential temperature of the mixture can be determined as function of the\nproportional composition (M1/M2) and the relative humidity of the clear overlying\nair as shown in Figure 1.\nThe shaded area represents the conditions for active entrainment episode,\nwhile the region above represents positive buoyancy of the mixture, therefore,\nentrainment is suppressed. The active entrainment area increases as the inversion\nbecomes weaker.\nASSUMPTIONS\nBased on the entrainment instability criteria described in Figure 1, we have\nmade the following assumptions:\na) - After an initial condition is established where the air is warmer than the\nocean, and after the first marine boundary layer has been formed mainly by\nmechanical mixing, the surface air temperature must remain in equal to the sea\nsurface temperature (SST).\nb) - The top boundary must be saturated with some haze droplets (noncloudy case).\nc) - The stratus cloud top temperature must be maintained, by the entrainment\nprocess, to be equal to the wet-bulb temperature of the overlying air\n(nonprecipitating cloudy case).\nTHE EQUATIONS\na) For graphical display.\n- Potential temperature (0)\nk\n0 = T (P/P)\n(1)\n,\nwhere P is pressure, PO is 1000mb\nT is the temperature and k = 0.286\n- Wet bulb temperature (Iw)\n(Cw/Cpv-1)\n[\n10]-|\n[ Cpa+r1Cpv Cpa+rCpv\nL\nL\nTw =\nTo\n- (T-To)\n+\nCw -Cpv\nCw-Cpv\n(2)\nwhere To is 273.16 K, LO is the latent heat for vaporization at To CDa and CDV\nare the specific heat at constant pressure for air and vapor, respectively, Cw is\nthe specific heat for water and r is the vapor mixing ratio.\n- Critical potential temperature (0c)\n0 c = (Tw/T)A.\n(3)\nThis parameter is defined as the temperature of an overlying air parcel which is\nsaturated by evaporating additional water. i.e., cools to its wet bulb\ntemperature, and then brought dry adiabatically to the standard pressure of 1000\nmb.\n48","b). Adiabatic compression expansion.\nThe Telford and Chai's equation (Telford and Chai, 1984) for a moist parcel:\n(Stage A: above condensation level, Stage B : below condensation level).\n4 L(TA)\n(4)\nrw,\nwhere Pa is the dry air pressure, Ra is the gas constant for dry air, and rw is\nthe water mixing ratio.\n- Total mixing ratio (r)\n(5)\nr = rs + rw,\nwhere rs is the saturated mixing ratio.\nc). Boundary Fluxes\nAt the surface boundary:\n- Sensible heat transfer (Sheat)\n(P2-P3) (Cpa +rCpv) (Ts-Ti)\n(6)\nSheat =\ngt (1 + r)\nwhere g is the acceleration due to gravity, P2 and P3 are the top boundary\npressures before and after entrainment, respectively, Ts is the sea surface and\nTi is the temperature of the entrained layer air when brought adiabatically to\nsea surface.\n- Latent heat transfer (Vflux)\nor\nU\n(7)\nVflux =\nwhere U is the wind speed and the subscript of o denotes some reference state.\n- Mass transfer of vapor per unit area\n(8)\nSmv = Vflux t,\nwhere t is time.\nAt the top Boundary:\n- Height of entrained air layer\nsz - P2-P3\n(9)\nwhere P is the density of the moist air.\nd). Time step\n(10)\nNUMERICAL SOLUTIONS\nThe mathematical procedures for handling the entrainment process consist in\ntaking a layer of clear warmer and drier air mass right above the top of the\nwell-mixed marine boundary layer, and then (1) compress it adiabatically to the\nsea surface pressure, (2) release sensible heat to the ocean to acquire the SST,\n(3) add water vapor so that the mixing ratio equals to that of the mixed layer,\n49","and (4) expand this layer adiabatically to its original pressure.\nThe first three steps have analytical solutions, however, the last one\nrequires some iterations. This numerical procedure modifies the amount of\nentrained air by adjusting the pressure at the new top to satisfy the conditions\nfor the noncloudy or the non-precipitating cloudy cases.\nThe new pressure at the top of the layer resulting from the numerical\niteration ensures the temperature requirements and determines the amount of\nentrained overlying air during a time step of one hour.\nIII) - APPLICATION AND RESULTS\nThe model has been applied to simulate the development of the marine\nboundary layer off the coast of South America (Chile-Perú-Ecuador) as shown in\nFigure 2 (see dashed line), because the Pacific ocean in the tropical region\nundergoes periods of warming (Rassmuson and Wallace 1983) and cooling (ANTI-\nENSO) as part of a natural oscillation of the ocean and atmosphere system (Graham\nand White, 1988). In addition the atmosphere in the southeastern Pacific off the\nsouth American coast is characterized by the Hadley and Walker circulations\ndriven by thermal differences. The south Pacific anticyclone and the associated\nsteady trade wind regime are the most important climate factors in this region.\nThe air usually subsides strongly in the poleward region of the Haddley\ncirculation near the center of the south Pacific anticyclone, acting also as a\nperturbing factor. In contact with the ocean, this dry, warm and very stable air\nmass becomes modified by the evaporating and cooler ocean surface. As it moves\nnorthward along the trade wind over the ocean (see dashed line on Figure 2) the\nmarine boundary layer develops due to evaporation and entrainment processes.\nMARINE BOUNDARY LAYER SIMULATED FOR THE ENSO 1982/1983\nThe marine boundary layer as simulated by the model for the ENSO 1982/1983\n(March-April-May, 1983) as shown in Figure 3, begins to grow (initial point at\n30s, 75W) rapidly before cloud forms as seen by the changes in magnitude of the\ntop boundary layer height and capped by an inversion which becomes stronger. At\n6.5 km north of the initial point, the condensation level is reached where the\ninversion jump is at its maximum of about 9 K.\nThe marine boundary layer continues to grow and a layer of stratus cloud\nstarts to develop, increasing in depth as the air approaches the equatorial\nregion (cloud depth of 3.7 km). The cloud base initially lowers from 800 to 200 m\nand then remains constant thereafter (see the lower curve), while the height of\nthe top boundary (the upper curve) continues to rise steadily. The inversion jump\nbecomes weaker after cloud formation to 1 K when the air reaches the\nintertropical convergence zone.\nMARINE BOUNDARY LAYER FOR A COLD EPISODE (ANTI-ENSO 1985)\nFigure 4 illustrates the result for the cold episode during October, 1985,\nin the same region as shown in Figure 2. There is a dramatic contrast to the ENSO\ncase. The growing tendency is the same as the air moves equatorward, but the\nheights and the cloud depth are quite different. Near the equatorial region, the\ncloud top height is only 2.2 km, the cloud thickness is 2.1 km and the cloud base\nheight is 100 m, in contrast to 3.7 km, 3.4 km and 200 m for the ENSO case. In\naddition the inversion jump remains strong at around 7 K (drought condition) due\nto the measured mixing ratio of the upper air.\nIV). - CONCLUSIONS\nA simple model has been developed to simulate the development of the marine\nboundary layer and applied to the southeast Pacific ocean under a warm (ENSO) and\na cold (ANTI-ENSO) episodes as part of a natural oscillation.\nThe structure and evolution of the marine layer, when the air flows at\nconstant speed over the ocean whose surface temperature is either constant or\nchanging, is partially explained as caused by the evaporation of the ocean and\nthe entrainment of warm and dry air from above the well-mixed marine boundary\nlayer.\nThe marine boundary layer off the coast of Chile, Perú and Ecuador simulated\nunder the El Niño- southern oscillation event is characterized by a thicker cloud\nlayer (with the potential for precipitation) capped by a weaker inversion near\nthe equatorial region; in contrast to a thinner stratus cloud, capped by a\nstronger inversion (drought condition) for the same region as simulated for the\ncold episode.\n50","313.0\nPRESSURE = 900 MB\nR(G/KG)\nRH(%)\nCLOUDY AIR TEMPERATURE = 288 K\n311.5.\n22.3\n100.\n20.1\n90.\nCLEAR AIR TEMPERATURE = 298 K\n17.9\n80.\nII 1 G/KG\n70.\nWATER MIXING RATIO\n15.6\n310.0.\n60.\n13.4\n11.2\n50.\n8.9\n40.\n308.5\n6.7\n30.\n4.5\n20.\n2.2\n10.\n.0\n0.\n307.0\n305.5.\n304.0.\n302.5.\n301.0.\n299.5.\n298.0\nCLEAR AIR\nCLOUDY AIR\n296.5\n.4\n5\n.6\n7\n8\n.9\n1.0\n.0\n1\n.2\n3\nPROPORTIONAL COMPOSITION\nFigure 1. - Potential virtual temperature lines of a mixed parcel as a function of\nthe proportional composition for different relative humidíties of the clear air.\nThe cloud top temperature is 288 K with a liquid water mixing ratio of 1 g/kg.\nThe overlying air is 10 k warmer in temperature. The mixing is taking place at\n900 mb pressure level.\n51","110\n100\n90\n80\n70W\n0\n26\n27\n6 July 76\nL\n22\n26\n26\n26\n24\n10\n22\n24\n20\n20\n22\n19\n22\n20\n16\n20\n14\n20\n30\n20\n18\n12\n12\n4OS\nFigure 2. - The region of application of the model (see dashed line) .\n52","5.3\n5.0\n4.7\n4.4\nCHILE-PERU-ECUADOR\n4.1\nENSO 1982-1983\nMAM 1983\n3.8\nU = 5 M/S\n3.5\ntop\n3.2\ncloud\n2.9\n2.6\n2.3\n2.0\n1.7\n1.4\n1.1\n.8\n5\ncloud base\n2\n0. 2. 4. 6. 8. 10. 12. 14. 16. 18. 20. 22. 24. 26. 28. 30. 32.\n1\nDISTANCE SOUTH-NORTH * 100 (KM)\nFigure (March-April-May - 3. - The simulated The marine boundary layer obtained for a\nthe Climate 1983). wind speed is 5 m/s. The SST warm episode\nsubroutine The Diagnostics sea breeze Bulletin, effects has and been interpolated neglected. using has been a curve obtained fitting from\n52","5.3\n5.0\n4.7\n4.4\n4.1\nCHILE-PERU-ECUADOR\nANTI-ENSO 1985\n3.8\nOCT 1985\nU = 5 M/S\n3.5\n3.2\n2.9\n2.6\n2.3\n2.0\n1.7\n1.4\n1.1\n8\nSTRATUS CLOUD\n5\n2\n1\n0. 2. 4. 6. 8. 10. 12. 14. 16. 18. 20. 22. 24. 26. 28. 30. 32.\nDISTANCE SOUTH-NORTH* 100 (KM)\nFigure 4. - The simulated marine boundary layer as obtained from the model for a\ncold episode (October, 1985). Wind speed is 5 m/s.\n54","Southern Oscillation in the OSU Coupled Upper Ocean/Atmosphere GCM\nKenneth R. Sperber\n1 Lawrence Livermore National Laboratory\nP.O. Box 808, L-264, Livermore, CA 94550\nSultan Hameed\nInstitute for Atmospheric Sciences, State University of New York\nStony Brook, NY 11794-2300\nW. Lawrence Gates1 and Gerald L. Potter1\n1. Introduction: With the advent of coupled ocean/atmosphere global climate models it\nhas become possible to study their inherent natural variability. This natural variability arises\nas a consequence of the encoded physics and dynamics since the only imposed constraint on\nthe simulation is the diurnal and annual cycle of solar radiation. One of the dominent modes of\ninterannual variability is the Southern Oscillation (SO). The Oregon State University coupled\nupper ocean/atmosphere global climate model has been shown to qualitatively simulate many\naspects of the SO (Sperber et al. 1987; Sperber, 1989). Sea-level pressure (SLP), sea-surface\ntemperature (SST), thermocline layer temperature, surface air temperature, precipitation,\nand mixed layer currents are among the simulated fields which reproduce SO characteristics.\nThus the nature of ocean/atmosphere feedback processes can be investigated.\nThe primary signature of the SO is a large-scale variation of SLP and SST across the trop-\nical Pacific Ocean. This variation results in anomalous weather conditions over much of the\nglobe. Of immediate interest is the evolution of SST anomalies during extreme phases of the\nSO. During the low phase the southeast tradewind system is weaker than normal or may even\nreverse. Associated with this breakdown of the trans-Pacific pressure gradient are anoma-\nlously warm SST anomalies, usually on the order of several °C. These extreme warmings, El\nNiño events, are typically characterized by westward propagation of SST anomalies from the\ncoast of South America beginning early in the calander year of onset. Towards the end of the\nyear these anomalies merge with those propagating from the western Pacific (Rasmusson and\nCarpenter, 1982). Thus, these events usually occur in phase with the seasonal cycle. Atypical\nevents, such as the 82/83 and 86/87 El Niños, evolve out of phase with the seasonal cycle.\nThat is, eastward propagation of SST anomalies from the western Pacific dominates with the\nwarm anomalies in the eastern Pacific not developing until late northern summer and early\nfall.\n2. Evolution of Trans-Pacific SST Anomalies: GCM simulated composite monthly SST\nanomaly maps for warm episodes (low Southern Oscillation Index) have been generated.\nThese provide the time sequence of the spatial evolution of SST anomalies in the region\n130°E-70°W, 30°S-30°N. Due to space limitations it is not possible to present the monthly\ncomposites for the whole event. Rather, averages of selected months that capture the essence\nof the temporal evolution will be presented. These are shown in Figures 1a-d. This may\nbe compared with the temporal evolution of SST anomalies presented in Rasmusson and\nCarpenter (1982) based upon the composite evolution for the 1951, 1953, 1957, 1965, 1969,\n55","and 1972 El Niños. Following the notation of Rasmusson and Carpenter (1982) we denote\nmonths of the year during which the maximum SST anomaly in the eastern Pacific occurs\nwith (0) and months during the subsequent year with (+1) respectively.\nIn the early spring [March(0) and April(0)], see Figure 1a, the simulated SST anomalies\nadjacent to South America reach their maximum values corresponding to the timing found\nby Rasmusson and Carpenter (1982). Subsequent evolution of the model warm event in the\neastern Pacific departs from the typical scenario previously outlined. Rather development\nmimics the evolution of the atypical event. During early summer, [May(0), June(0), and\nJuly(0)], see Figure 1b, the spatial extent of the .2°C isotherm in the western Pacific increases\nwith anomalies .4°C now located near 150°W. In the monthly anomalies, not shown, the SST\nanomalies in the eastern Pacific became negative during May(0) and June(0) with out of phase\npositive anomalies developing during July(0). Rapid growth of anomalous conditions occurs\nduring late summer and early fall. By August(0) the eastern and western .2°C isotherms merge\nresulting in SST anomalies of .2°C that span the tropical basin from South America to\nIndonesia. From September(0) through November(0) the strength of the anomalies continue\nto enhance throughout the basin. The resulting mean August(0), September(0), October(0),\nand November(0) anomalous distribution is given in Figure 1c. The largest SST anomalies of\nthe warm event, ~.9°C, occur near the equator at 115°W during October(0). Thus, the model\nwarm event reaches its mature phase somewhat earlier than seen in observations which show\nthe greatest extent of the warm anomalies between November (0) and January (+1) after onset\n(Rasmusson and Carpenter, 1982; Philander, 1983). The location of the simulated maximum\nis adjacent to the location of the negative node of the annually averaged SLP isocorrelation\nseen in Sperber et al. (1987) which characterizes the primary atmospheric signature of the\nSO. The subsequent decay of the positive anomalies in the model also occurs earlier than has\nbeen observed. In the December(0), January(+1), and February(+1) composite, Figure 1d,\nthe model warm event has decayed with anomalies of ~ .2°C spanning most of the tropical\nand subtropical Pacific.\nThat the composite model event decays somewhat earlier than the typical El Niño as\nwell as the atypical El Niños is probably related to several factors including the nature of\nthe mean climate state at the outset of the event and the strength of the anomalies that\ndevelop in the atmospheric and oceanic circulation. The interplay of these and other factors\nin determining the duration of a warm event in the GCM may be investigated in further\ndiagnostic studies. The simulated atmospheric and oceanic anomalies are smaller than found\nin nature and suggest the possibility that the shorter duration in comparison with observations\nmay be related to the weaker strength of the signal.\n3. Mixed Layer Current Anomalies: Concomitant with the variation of SLP between the\neastern Pacific and Indonesia is a mixed layer current variation due to the coupling of the\nocean currents to the variation of the atmospheric forcing by windstress. During years of low\nSouthern Oscillation Index the SLP gradient across the equatorial Pacific decreases resulting\nin diminished Southeast trade winds. Firing et al. (1983), Wyrtki (1975, 1977) and Enfield\n(1981) have deduced from observations that a weakening of the South Equatorial Current\noccurs during El Niño resulting in an eastward transfer of warm water that is responsible for\n56","the anomalous increase in SST in the central and eastern Pacific. Rasmusson and Carpenter\n(1982) have noted the development of anomalous equatorward flow in both hemispheres in\nthe Pacific during a warm episode. The simulated anomalous ocean currents occur not only in\nthe tropical Pacific basin but also in extra-tropical regions. The current anomalies are shown\nusing velocity vectors of three different lengths; the smallest arrow is for speeds < 2cm s-1,\n,\na medium arrow is for speeds 2cm s-1but < 5cm s-1 and the largest arrow is for speeds\n5cm s-1. It is interesting to note that the anomalies in equatorial flow converge near\n115°W, 2°N, the grid-point noted earlier as having the strongest anti-correlation in SLP\nwith the Indonesian region. The South Equatorial Current has been observed to reverse in\ndirection, particularly in the western Pacific during El Niño years. The simulated anomalous\natmospheric forcing of the currents is weaker than observed, recalling that the amplitude of\nthe model Southern Oscillation Index is 1/2 3 to 1/2 2 that of the observed Southern Oscillation\nIndex (Sperber et al. (1987). Hence the currents do not reverse; however the anomalies are in\nthe observed direction. These eastward anomalies are supported by tropical flow converging\nto the equator from both hemispheres, as noted in the empirical observations of Rasmusson\nand Carpenter (1982).\nThe temporal evolution of the current anomalies during years of low Southern Oscillation\nIndex can be evaluated by examining successive seasonal anomalies. It will be seen that\nthe Pacific basin SST anomalies, given in Figure 1a-d evolve roughly in parallel with the\ndevelopment of the current anomalies.\nDuring spring, Figure 2a, eastward anomalies of 3-7.8cm s-1 are located in the equato-\nrial region west of the dateline. The corresponding distribution of SST anomalies, given in\nFigure 1a, shows the highest SST anomalies in the same region. By summer, Figure 2b, the\neastward equatorial Pacific anomalies west of 120°W and east of the dateline have increased\nin strength to 3-6cm s-1. Concomitant with the expansion of equatorial eastward current\nanomalies is the spread of positive SST anomalies into the central/eastern Pacific Ocean. The\ndistribution of SST anomalies in Figure 1b shows the central Pacific as the region of greatest\nwarming. Northeastward flow from 8°S and 4°S in the longitude range 160°W to 130°W\nconverges with the eastward equatorial current anomalies in the area of the largest positive\nSST anomalies. From the coast of South America to about 120°W westward anomalies of\n2-6cm s-1generally reinforce the climatological flow. These current anomalies are associated\nwith westward expansion of eastern Pacific positive SST anomalies (clearly seen in the in-\ndividual monthly composites). In the fall, Figure 2c, the eastward equatorial anomalies in\nthe Pacific are highly developed with velocities of 5-11.7cm s-1extending to nearly 120°W.\nThe SST anomalies, Figure 1c, span the Pacific basin in conjunction with the well developed\ncurrent anomalies. The equatorward flow in both hemispheres has increased relative to sum-\nmer together with greater eastward current anomalies at the equator. In winter, Figure 2d,\nthe current anomalies are highly organized. The eastward anomalies in the equatorial Pacific\nnow range from 6-26cm s-1. The strength of the equatorward current anomalies in both\nhemispheres has increased. Equatorward flow in the Pacific now comes from as far north as\n16°N and as far south as 8°S, particularly west of the dateline. The waning of the positive\nSST anomalies precedes the decrease of eastward current anomalies in the model.\n57","4. Discussion: Although we have concentrated our attention to the anomalous behavior\nin the Pacific basin the simulated ocean currents throughout the globe depart from the mean\nstate in a systematic fashion. The anomalous flow during low Southern Oscillation Index may\nbe characterized in general terms. Flow in the tropical and southern regions of the Pacific,\nIndian, and Atlantic Oceans opposes the normal circulation. In the regions north of the\ntropics in the Pacific and Atlantic Oceans the current anomalies tend to augment the usual\nflow.\nIt should be noted that high phase cold event simulated SO SST anomalies evolve in\nthe same fashion but with the sign of the anomalies reversed, a known characteristic of the\nempirical data (Meehl, 1987; Kiladis and van Loon, 1988). During this phase the trans-Pacific\npressure gradient is enhanced resulting in mixed layer current anomalies in the tropical Pacific\nOcean which augment the usual flow. Thus the cold event evolution is antithetical to that of\nthe warm event.\nReferences\nEnfield, D. B., 1981: El Niño-Pacific eastern boundary response to interannual forcing. in Re-\nsource management and environmental uncertainty, eds. M. H. Glantz and J. D. Thompson,\nWiley, New York, 213-254.\nFiring, E., R. Lukas, J. Sadler, and K. Wyrtki, 1983: Equatorial undercurrent disappears\nduring 1982-1983 El Niño. SCIENCE, 222, 1121-1123.\nKiladis, G. N., and H. van Loon, 1988: The southern oscillation. part VII: meteorological\nanomalies over the Indian and Pacific sectors associated with the oscillation. Mon. Wea.\nRev., 116, 120-136.\nMeehl, A., 1987: The annual cycle and interannual variability in the tropical Pacific and\nIndian Ocean regions. Mon. Wea. Rev., 119, 27-50.\nPhilander, S. G. H., 1983: El Niño Southern Oscillation phenomena. NATURE, 302, 295-301.\nRasmusson, E. M. and T. H. Carpenter, 1982: Variations in tropical sea surface temperature\nand surface wind fields associated with the Southern Oscillation/El Niño. Mon. Wea. Rev.,\n110, 354-384.\nSperber, K. R., 1989: Southern Oscillation simulation in coupled ocean/atmosphere global\nclimate models., Ph.D. Dissertation, State University of New York at Stony Brook, Stony\nBrook, NY, 157pp.\nSperber, K. R., S. Hameed, W. L. Gates, and G. L. Potter, 1987: Southern Oscillation\nsimulated in a global climate model. NATURE, 329, 140-142.\nWyrtki, K., 1975: El Niño-the dynamic of the equatorial Pacific Ocean to atmospheric forcing.\nJ. Phys. Oceanogr., 5, 572-584.\nWyrtki, K., 1977: Sea level during the 1972 El Niño. J. Phys. Oceanogr., 7, 779-7871.\nFigure Captions\nFigure 1. Evolution of SST anomalies during simulated composite warm events. The con-\ntour spacing is 0.2°C. Contours >0.2°C are shaded. a) early spring(0) b) late spring/early\nsummer(0) c) late summer/fall(0) d) winter(+1)\nFigure 2. Evolution of mixed layer current anomalies during simulated warm events. a)\nspring(0) b) summer(0) c) fall(0) d) winter(+1)\n58","30N-\n10N\n10S\nON\n30S\n150E\n180\n150W\n120W\n90W\nLONGITUDE\nb\n30N\n10N\n10S\n30S\n150E\n180\n150W\n120W\n90W\nLONGITUDE\n30N-\n10N-\n10S\n30S\n150E\n180\n150W\n120W\n90W\nLONGITUDE\nd\n30N-\n10N\n10S\n-.6\n30S\n150E\n180\n150W\n120W\n90W\nLONGITUDE\n59","28N\n16N\n5N\n6S\n17S\n28S\n130E\n150E\n170E\n170W\n150W\n130W\n110W\n90W\n70w\nLONGITUDE\nb\n28N\n16N\n5N\n6S\n17S\n28S\n110W\n90W\n70W\n170E\n170W\n150W\n130W\n130E\n150E\nLONGITUDE\nC\n28N\n16N\n5N\n6S\n17S\n28S\n130E\n150E\n170E\n170W\n150W\n130W\n110W\n90W\n70W\nLONGITUDE\nd\n28N\n16N\n5N\n6S\n17S\n28S\n130E\n150E\n170E\n170W\n150W\n130W\n110W\n90W\n70W\nLONGITUDE\n60","VECTOR INDEX OF THE NORTH ATLANTIC OSCILLATION AND ITS\nASSOCIATION WITH THE SURFACE AIR TEMPERATURE AND CIRCULATION\nAndrei V. Nemirovsky, Vladimir T. Radiuhin\nAll-Union Research Institute of Hydrometeorological\nInformation-World Data Centre, Obninsk, USSR\n1. Introduction.\nThe North Atlantic Oscillation detected by Walker,which is a\nlarge-scale time variation of the atmospheric circulation between\nthe subtropical high pressure belt in the Azores region and the\nsubpolar pressure low near Iceland, is an important factor in for-\nming the regional climate. A number of works (Lamb, Peppler, 1987;\nMoses et al, 1987; Rogers, 1984; Van Loon, Rogers, 1978; etc. ) stu-\ndying the phenomenon and its association with climate characte-\nristics, to obtain a quantitative NAO characteristic use an index\nwhich is a difference in normalized anomalies between the mean\nmonthly pressures at stations located near the long-term mean po-\nsitions of the Azores and the Iceland CAs. The works (Wallace, Gut-\nzler, 1981; Barnston, Livezey, 1987) have shown that the North At-\nlantic CAs are the centres of teleconnections in the geopotential\nfields associated with NAO, with the North Atlantic CA pair being\nthe only one in the Northern Hemisphere which can be noticed in\nall the months of the year (Barnston, Livezey, 1987). The present\nstudy to obtain a NAO characteristic considers a five-component\nvector which permits describing the temporal changes in the mutu-\nal positioning and intensity of the North Atlantic CA pair.\n2. Data and the indices computed.\nAs a vector index of NAO the vector INAO = {A,L,a,A, is\nconsidered, where A is the difference in pressures between the\nAzores high and the Iceland low centres [mb];L -the distance bet-\nween the centres obtained by the length of the sphere are connec-\nting the two CAs [deg]; a-the slope of L computed by moving anti-\nclockwise from the east[deg];and A, 4- -the Azores high coordinates.\nThe vector components have been computed from the centre mean\nmonthly coordinate and pressure archive for eight CAs in the Nor-\nthern Hemisphere for 1891-1989 (Baidal, Neushkin, 1988). To analy-\nze the INAO relationship with the surface temperature and circu-\nlation as well as to compare it with the traditional NAO index, we\nused the gridded data of mean monthly air temperatures for 1891-\n1986 (Gruza, Rankova, 1979), the circulation form frequency cata-\nlog for the Northern Hemisphere for 1900-1986 (Girs,1974) as well\nas the gridded data of mean monthly sea level pressures for 1891-\n1986 (Timopheeva, 1980). As a traditional NAO index we analysed the\ndifference in normalized anomalies between the mean monthly pres-\nsures:\nNAOym = (Pyma - (Pym -\nm\nwhere a and 1 are the Azores and Iceland; y and m - the year and\na\nmonth; Pym and Pym - the mean monthly sea level pressures ave-\nraged for four grid points (20 and 30 W, 40 N and 20 and 30 W,\n35 N) and for three grid points (10, 20, 30 W and 65 N); Pm -the\nmean long-term pressure; and o - the standard deviation of pres-\nm\nsure for the m-th month.\n61","3. The analysis method and results.\nTo study the statistical properties of INAO, decrease its\ndimension and pass to the non-correlated values, the eigenvectors\nand the corresponding eigenvalues of an INAO correlation matrix\nwere computed for each month and the principal component time se-\nries were obtained. All the components of the vector index have\nbeen previously normalized (by subtracting the long-term mean and\ndividing it by the standard deviation). Table 1 gives the eigen-\nvalues and the percentage of the variance of a vector under con-\nsideration explained by the principal components. As in our case\nthe sample size N, ranging from 86 to 96 for various months, is\nmuch greater than the vector dimension K, equal to 5, to assess\nthe statistical significance of the advantage of passing to the\nprincipal components (i.e. the differences in the eigenvalues) we\nused the asymptotic test by Wilks and Bartlett (Kendall, Stuart,\n1976). For nearly all the months of the year all the eigenvalues\ndiffer at a very high significance level (Table 2) Our further\nconsideration will be confined to the first two principal compo-\nnents corresponding to the eigenvalues >1 as their contribution\nto the total variance is greater than 60% for the majority of the\nmonths (excluding July and August). The analysis of the principal\ncomponent correlations with the initial INAO vector components\npermits interpreting the principal components. During the whole\nyear the first component accounts for the mutual spatial CA posi-\ntioning. As for the second one, it mainly accounts for the pres-\nsure differences between the CAs in winter and in March, June and\nJuly and - for their spatial positioning from April to November.\nTable 3 gives the correlation coefficients for three winter\nmonths, March and the central months of the rest of the seasons.\nOne can see that in January and February the first principal com-\nponent increase is characterized by the Azores high moving to the\nnorth-east, the North Atlantic pair centres coming closer to each\nother and the pair moving anticlockwise, while in other months the\nsame phenomena are associated with the first principal component\ndecrease. The time-series of the first two principal components\nfor January are given in Fig.1 As expected the second principal\ncomponent correlates rather well with NOAOym, especially in win-\nter (T = 0.73 for December, T = 0.75 for January and T = 0.61 for\nFebruary).\nTo obtain the INAO and NAOym relationships with the surface\nair temperature and circulation the coefficients of correlation\nof the principal components as well as NAOym with the mean month-\nly air temperature T and the monthly frequency of the circulation\nforms W, and E for the Atlantic Europian sector have been compu-\nted. The statistically significant correlations (the 1% - signi-\nficance level corresponds to /r/=0.3) have been revealed for all\nmonths excluding May and August (correlation with T) and May, Ju-\nne, August and September (correlations with W, C and E). The lar-\ngest values of T have been obtained for the winter T fields and\nthe second principal component of the INAO vector.Fig.2 shows the\ncorrelation isolines for the January values of T and the second\nprincipal INAO component which nearly coincides with the CA pres-\nsure difference. To the north-west of the long-term January posi-\ntion of INAO one can see a region of the negative T values, where\n62","at a large pressure difference there are mainly negative anomali-\nes of .To the east of the long-term position of INAO one can see\nan area of the positive T values, where at a large pressure dif-\nference there are positive anomalies of T. A similar picture of\nthe correlation isolines is obtained for the NAOym index and T,\nthe T value in this case being about 0.1 less in the positive va-\nlue area.\nTable 4 gives correlations of the circulation forms W, C and\nE with the NAOym index and the first two principal INAO compo-\nnents. Note that in winter the INAO correlation with the circula-\ntion forms is somewhat higher than the NAOym one and the value of\nT is maximal. From Table 3 it is clear that the correlations of\nNAOym and the principal INAO components with W are mainly accoun-\nted for by the intensity variation of the zonal flow over the At-\nlantics (see January and February) and their correlations with C\nand E-by the position of the upper-level ridge or trough over the\nWestern Europe (see February, April, July, October and December).\nThe analysis of the INAO components for the 1988-89 winter\nwhich was anomalously warm (Razuvayev, Bulygina, 1989) for the Eas-\ntern Europe has revealed substantial deviations of the components\nfrom the long-term means. The value of A for December and January\nwas greater than the standard deviation o and about 20 for Febru-\nary. The value of L was within the norm. The value of a for De-\ncember was more than O, about the mean for January and about - o\nfor February. The value of A for December and January was about\n- o and about the norm for February. The value of Q for December\nand\nJanuary was more than O and about the norm for February.\nFig.3 shows the deviations from the long-term January means\nof the cyclone frequency, the positions of the main trajectories\n(Kryzhanovskaya, 1977) as well as the positions of the North At-\nlantic CA pair. The above deviations are well seen in the values\nof the first two principal INAO components. In December the first\nprincipal component is less than - O, greater than O in January\nand about the norm in February. As for the second component its\nvalue is more than o in December and February and about 20 in\nJanuary.\n4. Conclusion\nThe five-component vector characterizing the North Atlantic\nCA pair in general is well represented (especially in winter) by\nits first two principal components. The first principal component\nexpaining some 50% of the total variance characterizes the chan-\nges in the mutual spatial CA positioning.As for the second compo-\nnent its behaviour is nearly the same as the one of the intensity\ncharacteristic of this pair - the pressure difference between the\npair's centres. The above characteristics are supposed to be the\nconvenient indices accounting for the NAO changes in time and can\nbe used to study the NAO influence on the regional climate.\nReferences\nBaidal M.Kh., A.I.Neushkin, 1988: The atmospheric circulation\nparameter catalog. Northern Hemisphere. Obninsk, VNIIGMI-WDC,\n452p (in Russian).\nBarnston, A.G., Livezey, 1987: Classification, seasonality and\npersistence of low-frequency atmospheric circulation pat-\nterns. Mon.Wea.Rev., 115,1083-1126.\n63","Girs A.A. 1974: The macrocirculation long-range meteorological\nforecasting.L.Gidromet. 488p. (in Russian)\nGruza, G.V., E. Ya.Rankova, 1979: Data on the climate structure\nand variability. Air temperature at sea level. Northern He-\nmisphere. Obninsk, VNIIGMI-WDC, 203p. (in Russian)\nKendall, M.G., A.Stuart, 1976: Desing and analysis, and time-\nseries. Nauka, Moscow, 736p. (in Russian).\nKryzhanovskaya,A.P., 1977: Long-term mean cyclone and anticyclone\ntrajectories and recurrence for the Northern Hemisphere.Maps\nand tables. Weather report Northern Hemisphere. Suppl 6,\nObninsk, 58p. (in Russian)\nLamb P.J., R.A.Peppler,1987: North Atlantic Oscillation: concept\nand an application. Bull.Amer.Met.Soc., 68, 1218-1225.\nMoses,T., G.N.Kiladis, H.F.Diaz and R.G.Barry, 1987: Characteris-\ntics and frequency reversals in mean sea level pressure in\nthe North Atlantic sector and their relationship to long-\n-term temperature trends. J.Clim. 7, 13-30.\nRazuvayev, V.N., O.N.Bulygina, 1989: Extremely warm winter over\nthe USSR during 1988-89. Proc. 14-th Annual Climate Diagnos-\ntios Workshop, La Jolla, California, USA.\nRogers, J.C., 1984: The association between the North Atlantic\nOscillation and the Southern Oscillation in the Northern He-\nmisphere. Mon.Wea.Rev. 112,1999-2015.\nTimopheeva, T.P., 1980: Data on the climate structure and varia-\nbility. Air pressure at sea level. Northern Hemisphere.\nObninsk, VNIIGMI-WDC, 208p. (in Russian).\nVan Loon, H., J.C.Rogers, 1978: The seesaw in winter temperatures\nbetween Greenland and Northern Europe. Part 1 : General Desc-\nription. Mon.Wea.Rev. 106, 296-310.\nWallace, J.M., D.S.Gutzler, 1981: Teleconnections in the geopo-\ntential height field during the Northern Hemisphere winter.\nMon.Wea.Rev. 109, 784-812.\nLegends of the Tables and Figures\nTable 1. Variance percentage (D%) explained by the principal com-\nponents (K).\nTable 2. Wilks-Bartlett statistic values for testing the signifi-\ncance of the difference between the eigenvalues (A).\nTable 3. Coefficients of the correlations of the first two prin-\ncipal components (K) with the initial INAO vector compo-\nnents.\nTable 4. Coefficients of the correlations of three circulation\nforms (w, C, E) with the NAO index and the first two INAO\nprincipal components.\nTable 5. Coefficients of the correlations of the INAO vector\nindex principal components with the traditional NAO index.\nFigure 1. Time-series and cubic trends of the first (a) and the\nsecond (b) principal components of the INAO vector index.\nFigure 2. Isolines for the coefficient of correlation between the\nsecond INAO principal component and T (January).\nFigure 3. Isolines for the deviations from the long-term means of\nthe cyclone frequency (%), main cyclone trajectories and the\nNorth Atlantic CA pair position for January 1989. The dashed\nline indicates the long-term mean January position.\n64","Table 1\nK\nJanuary\nFebruary\nApril\nJuly\nOctober\nDecember\nA\nD%\nED%\nA\nD%\nED%\nA\nD%\nED%\nA\nD%\nED%\nA\nD%\nED%\nA\nD%\nED%\n1\n2.52 50, 5 50, 5\n2.67 53.4 53,4\n2.13 42. 7 42. 7\n1.67 33.4 33.4\n2.17 43. 3 43, 3\n2.29 45. 9 45. 9\n2\n1.23 24.6 75. I\n1.18 23.6 77.1\n1.36 27.1 69. 8\n1.09 21. 9 55.3\n1.16 23. 66.4\n1.01 20. 3 66. 1\n0.87 17.4 83.9\n0,87 17.4 83. 5\n4\n0.36\n7.3\n96.\n8\n0,31\n6,3\n95.\n8\n0.53 10.5 93.7\n0.79 15.7 91.9\n0.57\n11.4\n95.2\n0.55 11.1 94. 6\n5\n0.16\n3,2\n100\n0.21\n4.2\n100\n0.32\n6.3\n100\n0.40\n8.0\n100\n0.24\n4,8\n100\n0,27\n5.4\n100\nTable 2\nDegrees of\nCritical region\nStatistic values for months\nE\nfreedom\nboundary\nJanuary\nFebruary\nApril\nJuly\nOctober\nDecember\n5%\n1%\n1\n10\n18.31\n23.21\n178.49\n168, 16\n100.2\n44.79\n104.26\n104.41\n2\n6\n12.59\n16.81\n91.07\n70.03\n48.96\n25.04\n52.04\n39.94\n3\n3\n7.82\n11.35\n46.41\n24.62\n11.84\n22.04\n33.15\n27.61\n4\n1\n3.84\n6.64\n11.37\n3.07\n5.57\n9.65\n15.23\n10.58\nTable 3\nJanuary\nFebruary\nMarch\nApril\nJuly\nOctober\nDecember\n1\n2\n1\n2\nI\n2\n1\n2\n1\n2\n1\n2\n1\n2\nA -0,07\n0.92\n-0,16\n0.93\n0.27\n0.74\n0,51\n0.48\n-0.20\n0.75\n0.45\n0.62\n-0.01\n0.99\nL\n-0.66\n0.48\n-0.78\n0.34\n0.73 -0.21\n0.74\n-0.27\n0.77\n0.38\n0.73\n0.19\n0.75\n0.07\n0.83\n-0.09\n0.90 -0.07\n-0,84 0.08\na\n-0,78\n0.04\n0.22\n0.56\n-0.83 -0,12\n-0,72 -0.09\nA\n-0.83\n-0.34\n-0.75 -0.32\n0,83 -0,43\n0.63 -0,70\n0.50 -0.23\n0.80 -0.38\n0.82 0.00\n0,84\n0.20\n0,81 0.29\n1\n-0.60\n0.69\n-0,47\n0.85\n-0.86\n0.17\n-0.29\n0.76\n-0,73\n0.15\nTable 4\nJanuary\nFebruary\nApril\nJuly\nOctober\nDecember\nW\nC\nE\nW\nC\nE\nW\nC\nE\nW\nC\nE\nW\nC\nE\nW\nC\nE\nNAO\n0,37\n-0,35 -0,09\n0.42\n-0,38\n-0,08\n0.13\n-0,08\n-0.06\n0.11\n-0.39\n0.24\n0.03\n-0.38\n0.28\n0.15\n-0.28\n0.04\n1\n0.17\n0.18\n-0.29\n-0.22\n0.51\n-0.24\n0.38\n0,13\n-0.44\n0.24\n0.29\n-0,41\n0.09\n-0,31\n0.18\n-0,21\n-0.19\n0.31\n2\n0,44\n-0,14 -0,29\n0.45\n0,03\n-0.43\n-0,02\n-0.33\n0.27\n0.04\n-0,33\n0.24\n0.03\n-0,08\n0.04\n0.19-0.13\n-0,10\nTable 5\nINAO\nJ\nF\nA\nJ\n0\nD\n1\n-0,04\n-0.45\n0,40\n-0,17\n0.30\n-0.04\n2\n0.75\n0.61\n0.42\n0.65\n0.40\n0.73\n3\n-0.28\n0.14\n0.38\n0.41\n0.49\n-0,15\n4\n0,00\n0.03\n0.13\n-0.22\n-0.33\n0.10\n5\n0.17 -0,07 0,07 -0,02 -0.23 0.24\n65","Fig. 1b\nFig.1a\na\n0\no\n1900 10 20 30 40 50 60 70 ao\n1900 10 20 30 40 50 60 70 80\nFig.2\n80\n0.3\n70\n0.4\n0.5\n0.6\nE\n60\n0.7\n-0.6\n0.6\n-0.4\n60\n-0.3\n40\n0.3\nso\n20\n40\n60\no\nso\n60\n40\n20\nFig.S\nao\n20\n70\nSo\nTO\n10\n60\n20\n50\n20\n-10\n10\n-10\nso.\nso\n00\n20\n40\nao\n60\n40\n20\no\n66","MOMENTUM EXCHANGE AMONG ATMOSPHERE, OCEAN, AND SOLID EARTH\nON 40-50 DAY TIME SCALES\nDavid S. Gutzler and Rui M. Ponte\nAtmospheric and Environmental Research, Inc.\n840 Memorial Dr.\nCambridge, MA 02139\nConsiderable research has been carried out to explain the presence of an\neastward-propagating near-equatorial oscillation of pressure, wind, and\nconvection with a 40-50 day time scale, first discovered by Madden and Julian\n(1971, JAS, p. 702; 1972, JAS, p. 1109) A spectral peak at that frequency\nhas also been found in time series of globally-integrated atmospheric angular\nmomentum M (Langley et al., 1981, Nature, p. 730). Rosen et al. (1989, JGR,\nin press) showed that the atmosphere-solid earth coupled system conserves\nangular momentum, as implied by the high 40-50 day coherence (with a phase\ndifference of 0+3 days) between fluctuations of M and the rotation rate of\nthe earth as measured from space. Madden (1987, JGR, p. 8391; 1988, JGR, p.\n5333) showed evidence for momentum transfer through surface wind stress\nacross the tropical Pacific, but his studies did not explain how momentum is\ncommunicated through the ocean to the solid earth. Enfield (1987, JPO, pg.\n1860) showed that variations in near-surface winds in the western and central\nequatorial Pacific were coherent with 40-50 day fluctuations in sea level at\nthe South American coast, which seemed to provide evidence for continental\ntorque due to sea level fluctuations, as suggested previously by Oort (1985,\nAdv. Geophys. 28A). Another possible mechanism for atmosphere-solid earth\nmomentum exchange, pressure torques across mountain ranges, was also dis-\ncussed by Oort (1985) but will not be considered explicitly in this study.\nWe address two related questions. (1) What is the phase relationship\nbetween the eastward-propagating, near-equatorial oscillation described by\nMadden and Julian, and globally averaged angular momentum? (2) What is the\ncoupling mechanism that effects momentum transfer between the atmosphere and\nsolid earth, and how (if at all) is the ocean involved in this transfer?\nThree data sets are used. (1) Daily rawinsonde time series of zonal wind\nat 850 mb and 200 mb at four near-equatorial stations in the western Pacific\n(Singapore, Koror, Truk, Majuro), where the 40-50 day spectral peak is\nprominent in zonal wind and proxy convection data (Madden and Julian, 1972).\n(2) Time series of atmospheric angular momentum, including the total zonally\nand vertically (1000 mb-100 mb) integrated angular momentum about the polar\naxis (M), and the zonal momentum [u] within individual latitudinal belts in\nvertical layers represented by the zonal wind u at 1000 mb, 850 mb and 200\nmb. (3) Sea level time series at four tide gauge stations (Rabaul, Malakal,\nNauru, and La Libertad) obtained from the TOGA sea level center, which have\nbeen filtered to remove tides with periods less than about 5 days. Nauru and\nLa Libertad are very close to the equator, within the oceanic equatorial\nwaveguide and therefore well-situated to detect sea level signals caused by\neastward-propagating baroclinic Kelvin waves. Rabaul and Malakal are located\noff the equator, where westward-propagating baroclinic Rossby waves show\nlarger sea level displacements. The rawinsonde and sea level stations and the\nboundaries of the latitudinal belts of momentum are mapped in Fig. 1.\nOur analysis is conceptually simple: we calculate coherence and phase\n67","relationships among the various time series and piece together those that are\nstatistically significant on the 40-50 day time scale. To address question\n(1) above we examine the phase relationship between local fluctuations of u\nat the rawinsonde stations and the zonal or global fluctuations of momentum.\nTo address question (2) we calculate the coherence and phase between M and\nsea level, between M and u 8 50 at the rawinsonde stations, and between M and\nthe near-surface belts of momentum.\nFigure 2 shows power spectra of u at the rawinsonde stations and the\ncoherence between 40-50 day fluctuations of the u winds and M. None of the\nU200 time series are coherent with M, which seems surprising in view of the\nsignificant coherence between zonally averaged near-equatorial upper tropo-\nspheric zonal wind [u] and M, shown in Table 1 (confirming earlier results of\nAnderson and Rosen, 1983, JAS, p. 1584). The dominant contribution to the\nzonally averaged momentum near the equator must therefore come from winds at\nlongitudes far from the convective activity that is thought to drive the 40-\n50 day oscillation. This conclusion is consistent with results of Gutzler and\nMadden (1989, JAS, p. 642), who showed that the largest near-equatorial 40-50\nday filtered 200 variances are found at stations east of Majuro and westward\nfrom Singapore.\nHowever, at least three of the U850 time series at these stations are\ncoherent with M, and the phases between U850 and M at all four stations are\nconsistent with Madden's (1987, 1988) results. These phases, and the phases\nof\n[u] 850 and [u] 1000 relative to M from Table 1, are consolidated to derive\na schematic depiction (Fig. 3) of the relationship between the eastward-\npropagating near-equatorial signal and 40-50 day fluctuations of M. Phases\nare lettered to correspond to the canonical phases of the near-equatorial 40-\n50 day oscillation defined by Madden and Julian (1972).\nA: Near-surface, near-equatorial westerly anomalies occur west\nof Singapore. Heavy anomalous convection takes place over\nthe far western Pacific.\nB: Maximum globally averaged total atmospheric angular momentum;\nnear-surface westerlies between Singapore and Koror; zonal-\naverage momentum at 200 mb maximized at +15 latitude.\no\nC: Near-surface westerly anomalies between Koror and Truk;\nmaximum near-equatorial zonal-average zonal wind.\nD: One-quarter cycle past maximum M, hence this is the phase\nof maximum torque on the solid earth by the atmosphere.\nNear-surface westerly anomalies near Majuro; maximum\nnear-surface zonal average zonal wind near 15-20°N.\nE: Diminishing near-surface westerly anomalies east of Majuro.\nCoherences between M and sea level (not shown) are insignificant on 40-\n50 day time scales at all four tide gauge stations. We conclude that exchange\nof momentum between ocean and solid earth does not come about via baroclinic\nocean waves. Thus the baroclinic Kelvin waves that couple western and central\nPacific stress with sea level at the South American coast (Enfield, 1987) do\nnot seem to transfer momentum between the atmosphere and continental walls.\n68","Nevertheless, the significant lower tropospheric coherences in Fig. 2\nand Table 1 support Madden's conclusion that Pacific surface stresses might\nprovide momentum transfer between atmosphere and ocean. Coherences between\ntime series of near-equatorial [u]850 and the local U850 at Singapore and\nKoror (which have the largest 850 40-50 day variance of these stations) are\nhighly significant with near-zero phase difference, confirming that the zonal\nwinds across the far western Pacific dominate [u] 850 However, these local\nu\nfluctuations, and the zonal average near-surface momentum in belts S2-N2, lag\nM by less than 60°. To account for the required torque, westerly stress anom-\nalies must lag M by one-quarter cycle (phase +90°). Such a phase lag occurs\nlater in the cycle, as westerly near-surface anomalies propagate eastward\nalong the equator with smaller amplitude (Fig. 2). Poleward propagation into\nthe Northern Hemisphere also occurs (Table 1). Although the coherence between\nM and the off-equatorial belts is slightly less than the 95% significance\nlevel, the phase lags between M and [u] 850 or [u] 1000 in belt N4 (15-20°N)\napproach +90°, leading us to suggest that a combination of near-equatorial\nstress east of the date line and off-equatorial stress in the Northern Hemi-\nsphere may provide the observed momentum transfer into and out of the atmos-\nphere. Otherwise, either our data do not properly represent the stress, or,\ndespite the high coherences between M and near-equatorial near-surface wind,\nsome other torque (e.g. mountain torque) is an important momentum transfer\nmechanism.\nMadden (1988) showed that the strongest stresses occurred some 20° off\nthe equator in the specific case he studied, though he did not emphasize that\nresult as we do here. If poleward propagation of surface stress anomalies is\nverified in subsequent studies to be a crucial element of ocean-atmosphere\ncoupling, then this feature will become an important component of any\ncomplete theory of the 40-50 day oscillation. Equally interesting is the\napparent unimportance for the global 40-50 day atmospheric momentum budget of\nwestern Pacific near-equatorial U200, for these are among the records that\ndisplay a 40-50 day spectral peak most strongly.\nIf a broad latitudinal patch of stress on the ocean is the principal\nmomentum transfer mechanism into and out of the atmosphere, then the nearly\ninstantaneous transfer of momentum between the ocean and the solid earth\nremains to be explained. We speculate that barotropic oceanic waves provide\nthe necessary transport. In comparison with the baroclinic waves that are\nmore commonly studied, barotropic waves are broader in latitudinal scale and\nfaster. These properties of barotropic waves, plus the fact that the verti-\ncally integrated currents of baroclinic waves tend to be nearly zero, make\nbarotropic modes potentially more effective for the rapid transport of\nmomentum to continental walls implied by the coherence and phase results.\nUnfortunately barotropic waves are difficult to detect in available oceanic\ndata because their sea level signals are swamped by those of baroclinic waves\n(Cane, 1984, JPO, pg. 1864), so testing this hypothesis will require modeling\nstudies. These are currently in progress at AER.\nAcknowledgments. This material is based upon work supported by the National\nScience Foundation under Grant ATM-8819825. Rick Rosen inspired us to inves-\ntigate this problem. Generous assistance with data sets was provided by Rol\nMadden (rawinsondes), David Salstein and Peter Nelson (angular momentum), and\nBernie Kilonsky (sea level)\n69","Rawinsonde (*) and Sea Level (a) stations\n20N\nBELT N4\nKoror\nTruk\nMajuro\nBELT N3\n10\nBELT N2\nSingapore\nMalakal\nBELT N1\no\nNauru\nLa Libertad\nBELT S1\nRabaul\nBELT S2\n10\nBELT S3\nBELT S4\n20S\n90E\n120\n150\n180\n150\n120\n90\n60W\nFIGURE 1: Map of rawinsonde stations (stars) and sea level stations (squares)\nand the latitudinal belts of zonally averaged momentum [u] at\n200 mb, 850 mb and 1000 mb (separated by dashed lines).\nTABLE 1\nCoherence (x100) and phase (degrees) between time series of M and equal\narea belts of zonal momentum at 200 mb, 850 mb, and 1000 mb (see Fig. 1)\nfor fluctuations with periods between 40 and 52 days. The 95% significance\nlevel for coherence is 53; coherences below 45 are denoted by X with no\nphase shown. Phase > 0 implies that M leads [u].\nBELT\nS4\nS3\nS2\nS1\nN1\nN2\nN3\nN4\nCoh\n74\n84\n85\n85\n81\n76\n79\n76\n[u] 200\nPhase\n+28\n-5\n-24\n-35\n-41\n-35\n-14\n+3\nCoh\n53\n61\n66\n78\n60\n(48)\nX\n[u] 850\nPhase\n+57\n+50\n+20\n+27\n+53\n+86\nCoh\n(51)\n56\n(49)\n[u] 1000\nX\nX\nX\nPhase\n+25\n+53\n+77\n70","1.000\n1,000\nMAJURO (850 MB) 1976-1985\nMAJURO (200 MB) 1976-1985\n0.100\n0.100\nintervals for spectral estimates in the upper right corner of each spectrum. Upper\nlocal wind and M with periods between 40 and 52 days. The 95% significance level\ncoherence is less than 45. Phase is in degrees; phase > 0 indicates that M leads\ncycles day\neach panel shows coherencex100 (top) and phase (bottom) between fluctuations of\nrow of panels is for 200; lower row is for u 850 . The pair of inset numbers in\nFIGURE 2 : Power spectra of zonal wind at four rawinsonde stations, with 95% confidence\n+92\n55\n0.010\nfor coherence is 53. An X is shown instead of coherence and phase where the\n0.010\nX\n0.001\n0.001\n1.0\n100.0\n10.0\n0.1\n10000\n1000\n100\n10\n1000.0\n1,000\n1.000\nTRUK (200 MB) 1976-1985\nTRUK (850 MB) 1976-1985\n0.100\n0.100\ncycles day\n63\n+52\n0.010\n0.010\nX\n0.001\n0.001\n10\n100\n10\n1000\n100\n10000\n1000\n10000\n1.000\n1.000\nKOROR (850 MB) 1976-1985\nKOROR (200 MB) 1976-1985\n0.100\n0.100\ncycles day\n60\n+15\nX\n0.010\n0.010\n0.001\n0.001\n10000\n1000\n100\n10\n10000\n1000\n100\n10\nthe wind.\n1.000\n1.000\nSINGAPORE (200 MB) 1976-1985\nSINGAPORE (850 MB) 1976-1985\n0.100\n0.100\ncycles day\n(47)\n-15\n0.010\nX\n0.010\n0.001\n0.001\n10000\n1000\n100\n10\n10\n10\n10\n10\n10","20N\nKoror Truk Majuro\nN3\n200\n10\nSinkapore\nN2\no\nmax. convection\n200\nS2\n10\nA\n20S\n90E\n120\n150\n180\n150W\n~4 days\n20N\n200\nN4\nKoror Truk Majuro\n10\nSingapore\n*\no\nmax. M\n10\nB\nS3 200\n20S\n90E\n120\n150\n180\n150W\n~4 days\n20N\nKoror Truk Majuro\n10\n*\nSinkapore\nN2\n850\nN1\no\nmax. stress on near requatorial Pacific\nC\n10\nS4 200\n20S\n90E\n120\n150\n180\n150W\n~ 7 days\n20N\n850\nN4\nKoror Truk Majuro\n10\nSinkapore\nO\nmin am/at\nD\n10\n20S\n90E\n120\n150\n180\n150W\n~ 5 days\n20N\nKoror Truk Majuro\n10\nSinzapore\no\ndiminishing near-surface westerlies\nE\n10\n20S\n90E\n120\n150\n180\n150W\nFIGURE 3: Schematic drawing of deduced phase relationships between 40-50 day\nfluctuations of near-surface, near-equatorial zonal wind (longitude\nof maximum westerlies shown by the solid arrow) , zonally averaged\nzonal wind at 200 mb or 850 mb in tropical latitudes ( latitude of\nmaximum westerlies shown by broad arrows along right edge of plots)\nand M. Phases are lettered according to the idealized phases defined\nby Madden and Julian (1972)\n72","HEAT AND MOISTURE FLUXES ON THE TIME SCALE OF\n20 TO 60 DAYS OVER THE INDIAN MONSOON AREA\nJean-Francois Gueremy\nDepartment of Meteorology, The Florida State University, Tallahassee, Florida\nOn leave from: Etablissement d'etudes et de recherches meteorologiques, Paris, France\nIn this report, we present the low frequency variability of the sensible and latent heat\nflux over the Indian monsoon area.\nFirst of all, we have used an atmospheric thermodynamic energy budget as well as the\nsimilarity theory in order to compute the surface fluxes on a daily basis. This method requires\nthe knowledge of the precipitation field and the OLR field, and notably the computation of the\nlarge scale apparent heating (Q1), and the large scale apparent moisture sink (Q2) (e.g. Yanai\net al., 1973), and the knowledge of the wind, the temperature and the specific humidity fields\nin the surface layer. This method is not always accurate on the daily time scale because of the\nlarge amount of data used, although we tried to be consistent in the choice of the datasets.\nHowever, we believe that this method is satisfactory for the studies of low frequency\noscillations. We have obtained monthly mean surface fluxes which are in agreement with\nthose computed by other authors. The other limitation of our study is the small number of\ndays available (122 days). This disadvantage has been reduced by the use of appropriate\nstatistical methods for short time series. Thus, we have obtained a good agreement with other\nauthors in the results concerning the low frequency variability of the fluxes (e.g. Krishnamurti,\n1988).\nOur major results may be summarized in the following three sections:\n1)\nVariability in the intraseasonal time scale (fig 1)\nWe have shown that the variability of the temperature and the specific humidity is\nmore important over land than over sea. This phenomenon may be mainly due to the soil\nmoisture which provides a low frequency variability to the land areas. The variance\npercentage of those two variables, for the considered time scale, is of the order of 30% over\nland, but only 20% over sea. Also, in the power spectrum, one peak has been found around\n20-30 days over land, and around 15 days over sea. The wind shows an important variability\nwhich geographical distribution is very well defined. In fact the summer 1979 has been\nrecognized as a year with pronounced amplitude for the 30-50 day oscillation. Thus, a large\narea with a variance percentage of 40% has been found from 5°N to 20°N (the northward\nmaximum extension stretching over India). In the power spectrum, the peak appears very\nclearly around 30-40 days. Finally, the surface fluxes show a variability which is a\ncombination of the previous ones. This gives an uneven variance percentage field with values\ngoing from 10% to 40%. In the same way, the variability peaks stretch from 10 to 40 days.\nWe have also noticed that in some places the wind strongly influences the surface fluxes,\nespecially the latent heat flux over sea.\n73","2)\nRelative influence of the different variables on the surface fluxes in the intra-seasonal\ntime scale (fig 2)\nThe results have been obtained by computing the surface fluxes generated by the low\nfrequency modes. In the computations, we have left one variable at a time constant (time\nmean value) in order to examine its contribution. The wind has a significant influence on the\nsurface fluxes over oceans; however, this is not quite evident over the land. Over the ocean,\nits influence seems to be more important for the latent heat than for the sensible heat flux. Of\nthe two other variables, the temperature appears to be the one which influences the fluxes the\nmost. This latter influence may be very strong over land. Thus, in some cases the sensible\nheat flux is entirely controlled by the temperature. The specific humidity appears to have an\nequal importance over land and ocean.\nThese results highlight the role played by the surface variables compared to the wind.\nThe surface variables have an important influence on the surface fluxes, which cannot be\nneglected especially over land for the intraseasonal time scale. This statement implies that a\nlow frequency wave over India (for example) may be modified by the surface variables; and,\nas a consequence, that the wave cannot be solely driven by the oscillations of the wind on this\ntime scale. On the other hand, we have pointed out the importance of the nonlinearity in the\nsurface fluxes, in the sense that different time scale phenomenon contribute to the fluxes on\nthe intraseasonal time scale.\n3)\nNorthward propagation of low frequency waves related to the surface fluxes and\nground variables (fig 3)\nWebster (1983) has used a model to expose a mechanism, which involves surface\nhydrological effects, to explain the northward propagation of low frequency waves over India.\nIn order to test his mechanism, we have shown the fields of precipitation, ground temperature,\nand sensible and latent heat flux (filtered in the 9-60 day interval) for a temporal sequence\nduring which low frequency wave propagation had been identified (Krishnamurti and\nSubrahmanyam, 1982). We have found relative patterns for the previous variables in very\ngood agreement with his mechanism: Thus, we had a temperature minimum zone coincident\nmore or less with the precipitation maximum zone. The drop in the temperature is caused by\nthe strong evaporation of the rain. In fact, we noted a latent heat flux maximum area on the\nsame location. On the other hand, a temperature maximum zone appears north of the\nprecipitation. The increase of the sensible heat flux destabilizes the atmosphere and induces a\nnorthward motion of the disturbance, according to Webster.\nFinally, these results and the concluding remarks of point 2 reinforce the idea that\nsurface processes can influence the propagation of low frequency waves over India.\nThis study might be continued to further validate its conclusions in the two following\nways. First, one could study much longer time series using observations coming from\nmeteorological stations at several locations. This kind of study could provide better accuracy\nin the surface fluxes estimation and also in the time series analysis, but with a loss in the space\ncoverage. Second, one could use a more realistic 3D model to test the influence of the surface\nprocesses on the evolution of the low frequency waves over the monsoon region.\n74","References\nKrishnamurti TN, Subrahmanyan D (1982) The 30-50 day mode at 850 mb during MONEX. J\nAtmos Sci 39, 2088-2095\nKrishnamurti TN, Oosterhof DK, Metha AV (1988) Air-sea interaction on the time scale of 30\nto 50 days J Atmos Sci 45, 1304-1322\nWebster PJ (1983) Mechanisms of monsoon low frequency variability: surface hydrological\neffects. J Atmos Sci 40, 2110-2124\nYanai M, Esbensen S, Jan-Hwa C (1973) Determination of bulk properties of tropical cloud\nclusters from large scale heat and moisture budgets. J Atmos Sci 30, 611-627\n35\nA\n30\n30\nA\n30\n38\nH\nto\nA\nD\n48\nH\n25\n0\n30\n30\n30\nA\nD\n30\nH\n20\no\n20\nH\nsv\n30\nL\n20\n15\n00\nA\n2\nlote\nF\n!\n28\nV\nyou\n2019\n10\n30\nH\n50\n80\n5\nH\nto\n130\n70\n80\n90\n100\n110\n120\n120\n130\n40\n50\n60\n40\n50\n60\n70\n80\n90\n100\n110\nb\na\nH\no\nD\n#\n-40\nH\nC\n800\nA\nH\nCity\nLIA\n4\n50\n30\n30\n40\n50\n60\n70\n80\n90\n100\n110\n120\n130\n35\nH\nL\n-\n0\n0\no\nA\n30\nB\nL\no\n4\nD\nU\n20\nA\nn\nH\nH\n25\nLi\na\na\n10\n<9\n20\n0\n1\nx\nC\nm\n15\n\"\n#\na\n:\nH\n32\nm\n10\nH\nH\n10\n5\n30\n0\n40\n50\n60\n70\n80\n90\n100\n110\n120\n130\n40\n50\n60\n70\n80\n90\n100\n110\n120\n130\nd\ne\nFig 1 Percentage of variance in the 20-60 day time interval. (a) ground temperature,\n(b) ground specific humidity, (c) wind, (d) sensible heat flux, (e) latent heat flux\n75","TIME SERIE OF S. W CST\nTIME SERIE OF S, T CST\n10\n10\n8\n8\n6\n6\n4\n4\n2\n2\n0\n0\n-8\n-2\ng\n1\n+\n-6\nul -8\n-8\n-8\n00\nDAY\nDAY\nTIME SERIE OF E, W CBT\nTIME SERIE OF E4 a OST\n120\n140\n100\n120\n80\n100\n60\n80\n60\n40\n40\n20\n20\n0\n0\n-20\n-20\n-40\nor -40\n-60\nof -80\n-80\n-80\n09\n00\n04\n80\nDAY\nDAY\nTIME SHIRID OF S. W CST\nTIME SURIE or S. T CST\n30\n30\n20\n20\n10\n10\n0\n0\n-10\n-10\n-20\n-20\n-30\n-30\n8\n00\nTO\n90\n09\n0\n08\nDAY\nDAY\nTIME SKRIE OF E, w CST\nTIME SERIE OF E. a CST\n30\n40\n20\n30\n20\n10\n10\n0\n0\n-10\n-10\n-80\n-00\n-30\n-30\n-40\n40\n-50\nDAY\nDAY\nUnits : W/m2\nb\nFig 2 Surface fluxes generated by the filtered time series of the temperature, specific\nhumidity, and the wind (20-60 day time interval) in (a) the Bay of Bengal (15° N, 84° E), and\n(b) the west coast of India (13° N, 75° E); the variation of one of the previous variables is\nsuppressed to see its effects on the fluxes. The dashed curve denotes the total fluxes.\n76","June\n35\n35\nH\n30\n30\nR\n30\n25\n25\nB\n20\n20\n10\n15\n15\nH\n0\n10\n10\nH\n5\n5\nL\nH\n0\n0\n70\n80\n90\n100\n70\n80\n90\n100\n35\n35\nH\nL\nA\n30\n30\nL\nH\nH\n25\n25\nR\n20\n20\n15\n15\n15\nH\nL1,\n10\n10\nH\n5\n5\n0.0\nH\n0\n0\n70\n80\n90\n100\n70\n80\n90\n100\n35\n35\nL\na\nD\n0\nH\nH\n30\n30\nH\n0\n25\n25\n20\n20\n20\n15\n15\n'H\nH\n10\n10\nH\n5\n5\n0\n0\n70\n80\n90\n100\n70\n80\n90\n100\n35\n35\nL\n30\n30\nH\nH\n9.0\n0.0\n25\n25\nG\nH\n20\n20\n25\n300\n15\n15\nL\nH\n10\n10\n5\n5\nH\n0\n0\n70\n80\n90\n100\n70\n80\n90\n100\nFig 3 Time evolution of the rain (left) and the ground temperature (right) filtered on\nthe 9-60 day time interval. (Units: W/m 2 and o C)\n77","ONSET OF THE AUSTRALIAN SUMMER MONSOON\nAND ASSOCIATIONS WITH THE 40-50 DAY OSCILLATION\nBrant Liebmann\nCooperative Institute for Research in Environmental Sciences\nUniversity of Colorado\nHarry Hendon\nCenter for Atmospheric Theory and Analysis\nUniversity of Colorado\nGreg J. Holland\nBureau of Meteorology Research Centre\nMelbourne, Victoria Australia\nThe original motivation for this project was to determine whether we could detect\na wavetrain propagating out of the tropics and dispersing into midlatitudes in association\nwith the onset of the Australian summer monsoon. The onset of the Australian summer\nmonsoon is the most suddenly occurring large-scale tropical convective event and, of\ncourse, it occurs in the same place every year. We use up to 31 years (1957-87) of data\nin our composites.\nOnset is defined as the first occurrence of 850 mb westerly winds at Darwin that\nare accompanied by rainfall of at least 7.5 mm/day (the climatological mean when winds\nbecome westerly) area averaged over some 30 stations that are located within Australia\nnorth of 20S. Figure 1 shows a time series of Darwin 850 mb wind and area averaged\nrainfall for 1968-69, smoothed with a 1-2-3-2-1 filter. Most of the years are similar in\nthat onset is well-defined and the monsoon fluctuates at intervals of a month or longer.\nDuring a few years the 850 mb winds at Darwin remained westerly for the duration of the\nmonsoon and in a few other years our objective definition produced a false onset.\nA composite of unfiltered zonal wind anomalies (defined as the departure from the\nDecember-March mean of 31 years) at Darwin (Fig. 2a) shows a deep baroclinic structure\nwith peak-to-peak variations of more than 11 m/s. Upper-level winds turn to easterly a\nfew days after lower-level winds become westerly. The overturning level of the\nanomalous zonal circulation occurs at about 500 mb prior to the occurrence of significant\nrainfall (Fig. 2b) and rises to about 300 mb with the onset of large-scale convection.\nOutgoing longwave radiation anomalies (Fig. 3) show a large-scale convective\nanomaly originating over the Indian Ocean 15-20 days prior to onset (Fig. 3a). A\nsequence of maps (not shown) shows this convection to propagate eastward. As the\nconvection approaches longitudes of Australia, it dips into the Southern Hemisphere. The\nconvection stalls and expands at onset (Fig. 3b). A few days later the convection resumes\neastward propagation and it retreats to the equator.\nThe signature of convection bears resemblance to that associated with the 40-50\nday oscillation. The impression that the monsoon is associated with (an approximate) 40-\n50 day oscillation is confirmed by a spectral analysis (not shown). There is a marginally\nsignificant spectral peak in Darwin 850 mb zonal wind at a period of around 35 days, but\nthe area averaged rainfall spectrum is quite red. The coherence between the two\nquantities, however, shows that while all periods longer than 20 days are statistically\nsignificant, there is a pronounced spectral peak near 35 days.\nSince the onset appears to be associated with a 40-50 day event and the monsoon\n78","exhibits 40-50 day variability, it seems appropriate to composite all 40-50 day events in\norder to increase the sample size. To isolate the 40-50 day events, we band-passed the\ndata using a Butterworth filter with half-power points at 30 and 60 days. Figure 4 is a\nfiltered version of Fig. 1, except that the rainfall curve has been shifted ahead by two\ndays to account for the weighted area average being east of Darwin.\nIn this particular year onset would be defined differently using the filtered series\nthan using the raw time series, although in many years onset occurs on nearly the same\ndate using filtered or unfiltered time series. Thus it is better to use (nearly) unfiltered\nthan bandpassed data to define onset.\nAnother feature evident in this particular time series and in every other year is that\nthe post-onset 40-50 day variability is much larger than pre-onset variability. We thus\nconclude that onset is the first 40-50 day event to affect Australia as the seasonal cycle of\nconvection shifts southward. The Darwin composite (not shown) using all 91 post-onset\n40-50 day events (based on filtered time series) is quite similar to the onset anomalies but\nexhibits slightly smaller amplitude.\nA stability analysis (not shown) reveals that only air below 900 mb can ever\nbecome unstable. Figure 5 is a composite time series of 950 mb moist static energy at\nDarwin. This figure shows that prior to day 0 (defined as the peak in 850 mb zonal wind\non this and the subsequent figure) there is no systematic rise in moist static energy.\nThere is a decrease in temperature at the time of maximum rainfall, presumably due to\nevaporation. There is a dramatic decrease in moist static energy, however, that occurs as\nrainfall decreases, and it is due to a decrease in moisture. We conclude that large-scale\ndynamics initiate rainfall, but a decrease in moisture is likely to contribute to its cessation.\nComposite Northern Hemisphere height anomalies are shown in Fig. 6. While the\npattern at day -2 is wavetrain-like in structure, there is no sense of emanation from the\ntropics, and it does not appear to disperse downstream. Furthermore, although the\nanomalies are statistically significant, they are quite small and would have a minimal\neffect on midlatitude 'weather.'\n15.0\n30.0\n22.5\n0.0\n15.0\n7.5\n-15.0\n0.0\n1 OCT\n1 NOV\n1 DEC\n1 JAN\n1 FEB\n1 MAR\n1 APR\nFIG. 1. Representative time series of Darwin 850 mb wind (solid) and northern Australia\nrainfall (dashed) for the 1968-69 season. The wind scale (in m/s) is at left and the\nrainfall scale (in mm/day) is at right. The series have been passed through a 1-2-3-2-1\nfilter. An 'X' denotes onset.\n79","25\n60\na\n20\nH\n100\n-6:41\n87\n15\n200\n300\n10\n0.0\n500\n5\nH\n850\n-4,23\n1000\n0\n20\n10\n0\n-10\n-20\nDAY\n5\nb\n0\n-5\nFIG. 2. 31 year composite onset of anomalous unfiltered a) zonal wind at Darwin and b)\nnorthern Australia rainfall. Day zero corresponds to the zero crossing of the 850 mb\nzonal wind. Note that time runs from right to left. Contour interval in a) is 1 m/s with\ncontours of easterly anomalies dashed.\n80","a\n-15\nH\n9.995\n7.30\n-10.9\nI\n-18.0\nis\n6\n07\n13.6\nH\n9.78\n2\n8\n-15.4\nH\n-13.0\n-10.5\n5.81\n&\n[\nb\n0\n9.81\n12\n7.12\nH\n15.00\n-24:51\n15.5\n-18.₫\nH\nT\n8.60\nt-\nH\n42\n3.4\n19.6\nL\nB\nL\n-9.09\n15.6\n-8.88\n-17.3\nFIG. 3. Composite unfiltered outgoing longwave radiation anomalies for a) 15 days prior\nto onset and b) onset. There are 12 years contained in this composite. Contour interval\nis 5 W/m2 starting at 10 W/m². Negative contours are dashed.\n81","5.0\n0.0\n-5.0\n1 OCT\n1 NOV\n1 DEC\n1 JAN\n1 FEB\n1 MAR\n1 APR\nFIG. 4. As in Fig. 1, except series have been filtered using a Butterworth filter with half-\npower points at 30 and 60 days and scale for both quantities is on left side of figure.\nQ. 14.6 - 15.7 G/KG\n5\nQ\nH. 3.398 - 3.428 X 10 J/KG\nT. 2% 298.1 K\nR. 5.- 25. MM/DAY\nR\n-20\n-10\n0\n10\n20\nDAY\nFIG. 5. Darwin unfiltered composite time series of 950 mb moist static energy, its\ncomponents, and rainfall based on peaks in 850 mb filtered wind at Darwin. Moist static\nenergy (solid line) is denoted by 'H', temperature (dot-dashed line) is denoted by 'T',\nmixing ratio (dotted line) is denoted by 'Q', and rainfall (solid line) is denoted by 'R.'\nScale ranges are noted in upper right corner.\n82","-4\n-8\n3\nb\nd\n+6\n+2\n8.43\ne\n6.10\nFig. 6. Composite unfiltered Northern Hemisphere height anomalies based on peak of 30-\n60 day filtered 850 mb wind at Darwin. There are 91 events included in composite.\nContour interval is 8 m starting at 4 m. Negative anomalies are dashed. Statistically\nsignificant anomalies are shaded.\n83","Biennial Variability in Tropical SST and Winds\nC. F. Ropelewski and M. S. Halpert\nClimate Analysis Center\nNMC/NWS/NOAA\nand\nE. M. Rasmusson\nDepartment of Meteorology\nUniversity of Maryland\nINTRODUCTION\nIn an earlier Climate Diagnostics Workshop paper, Ropelewski\net al. 1988, we described our preliminary examinations of\ntropical surface data, and presented a bibliography of studies by\nother investigators which suggested a link between the Southern\nOscillation (SO) and biennial variability. In this earlier work\nwe found a strong tendency for biennial temporal variability in\ntropical sea surface temperature (SST) and winds and several\nindices related to the SO. The analysis of time longitude cross\nsections along the equator, outlined briefly below, suggests that\nthis biennial variability plays a fundamental role in the\nevolution of the so. The work described here also supports the\nsuggestion that the SO can be analyzed and described in terms of\nthree temporal scales: 1) annual cycle, 2) biennial variations\nand 3) low-frequency variability.\nThe analysis briefly described here is taken from of an\nexpanded study of tropospheric biennial variability. A further\nexpansion of this work and a more detailed discussion of temporal\nscales through Singular Spectrum Analysis (SSA) of Southern\nOscillation (SO) - -related indices can be found in Rasmusson et\nal., , 1989.\nDATA\nThe Comprehensive Ocean Atmosphere Data Set (COADS) surface\nSST and wind data for 1950 to 1987 were analyzed to form monthly\nfields. An examination of the data distribution suggests that\nthese data are sufficient for analysis of equatorial cross\nsections from roughly 1957 onwards. The details of the field\nanalysis, data processing techniques and data distribution can be\nfound in Halpert and Ropelewski (1989)\nANALYSIS\nOur previous exploratory investigations have revealed that\nseveral so-related indices in the equatorial Pacific have a\nsignificant fraction of their total variance in the biennial\nrange. While the spectra of several of the SO indices, e.g.,\nDarwin sea level pressure, are very red, biennial variability in\n84","SST and 200 mb winds were demonstrated to carry more variance\nthan expected by chance. These earlier investigations prompted\nthe further analysis of tropospheric biennial variability in the\ntropics and an expansion of the analysis domain westward to\ninclude the Indian Ocean.\nSpectral analysis of the surface zonal wind anomalies,\nRasmusson et al., 1989, shows two distinct peaks above the red\nnoise level--one peak near 60 months and another centered near 24\nmonths (Fig. 1). While the period of the zonal wind component\nappears to be closer to exactly 24 months in the Indian Ocean\nbasin than in the western Pacific, the time series analysis of\nselected SO indices suggests that biennial variability is\npervasive throughout the equatorial regions. We note that the\nspectral analysis shows a clear separation between biennial\nvariability and the low frequency in the zonal winds.\nThe variance associated with the biennial and low-frequency\n(periods greater than 30 months) zonal wind components are\ncomparable across the equatorial Indian Ocean, with the low-\nfrequency component gaining in relative importance in the western\nand central equatorial Pacific (Fig. 2) The variance associated\nwith the annual cycle, not shown, tends to be largest near the\nocean-land boundaries. These three temporal scales, the annual\ncycle, biennial variability and low frequency variability account\nfor the bulk of the zonal wind variance in equatorial regions.\nTo obtain a more complete description of these various\ntemporal modes associated with the so, anomaly, biennially-\nfiltered, and low-pass filtered equatorial time-longitude cross\nsections of the SST were formed and compared. A description of\nthe filters can be found in Ropelewski et al., 1988. An example\nof the SST equatorial time-longitude cross sections for the\nperiod 1962 to 1980 appears in Fig. 3.\nThe annual cycle was discussed in detail in previous\nworkshop proceedings (Ropelewski et al., 1989) and will not be\ndiscussed further here. Biennial variability is evident in the\neastern and central Pacific unfiltered SST anomaly time-longitude\ncross section (left-hand panel of Fig. 3) In these regions, the\nsign of the SST anomaly tends to alternate each year for much of\nthe period, indicating that biennial variability is a fairly\nstrong temporal mode over much of the time sequence. A notable\nexception is the period from 1972 to 1976 when below normal SSTs\ndominate the cross section. In contrast to the eastern Pacific,\nbiennial SST variability is not obvious in the unfiltered\nanomalies across the western Pacific, nor is biennial variability\nevident in the Indian Ocean at any time during the period.\nThe biennially-filtered SST equatorial cross section (center\npanel of Fig. 3) emphasizes this component of SO variability.\nThe largest amplitude biennial SST variability extends from the\nSouth American Coast to the central Pacific. The long period\n85","modulation of biennial SST amplitude is one of the most striking\nfeatures of the cross section. The periods from 1962 to 1966 and\n1971 to 1978 show more biennial variability than the other years\ndepicted in these cross sections. These filtered cross sections\nalso reveal that the lower amplitude SST biennial variability in\nthe Indian Ocean tends to be in phase with the eastern Pacific.\nThe anomalies in the Indian Ocean and Western Pacific suggest\neastward propagation with time, while those in the eastern\nPacific suggest westward propagation of SST anomaly.\nThe warm ENSO episodes of 1963, 1965, 1972 and 1976/77 are\nassociated with positive extremes in the biennially-filtered SST,\nbut not all strong positive extremes in the biennial amplitude\nare associated with warm episodes, e.g., 1974/75. Conversely the\ncold SO episodes of 1964, 1970, 1973 and 1975 are associated with\nhigh negative amplitude excursions of the biennially-filtered\nSST. Thus, while each extreme of the biennial temporal mode\nisn't associated with an extreme in the so, the converse appears\nto be true over the analysis period; i.e., each of the extreme SO\nepisodes, warm or cold, is associated with a high amplitude\nexcursion in the biennially-filtered SST.\nLow-pass filtered SST equatorial time-longitude cross\nsections (right hand panel of Fig. 3) indicate no obvious link\nbetween the low frequency and biennial components of so\nvariability. We note, however, that each of the strong warm so\nepisodes over the period depicted here coincides with periods\nwhen both the low-frequency and biennial SST components show\npositive excursions. The weak 1963 warm episode was associated\nwith a strong positive phase in the biennial but not supported by\nwarming in the low-frequency component. We also note that the\nstrong positive biennial SST of 1974/75, which did not result in\nwarm SO episode, was opposed by a strong negative excursion in\na\nthe low frequency. This same low-frequency excursion appears to\nhave reinforced the strong negative biennial swings of 1973 and\n1975 to produce two cold SO episodes.\nDISCUSSION\nOur analysis of SO temporal variability, only briefly\noutlined here, supports a conceptual framework for the so. In\nthis framework so temporal variability is composed of three major\nmodes: 1) the annual cycle, which appears to fix the phase for\nmajor excursions in the so; 2) biennial variability; and 3) low-\nfrequency variability, which appears to set the amplitude for\nexcursions of the SO.\nA similar time-longitude cross section analysis of the\nsurface wind components, not shown here, indicates that the\nIndian Ocean and far western Pacific play a major, and perhaps,\nfundamental role in the evolution of the SO. While biennial\nvariability may have its origins in the temporal scales\n86","associated with ocean-atmosphere interactions, as suggested by\nsome numerical models, the sources of the low-frequency\nvariability are not evident. Nonetheless, it appears that the so\ncan be fruitfully investigated through analysis of its evolution\nin the three temporal modes discussed here.\nREFERENCES\nHalpert, M. S. and C. F. Ropelewski, 1989: Atlas of Tropical Sea\nSurface Temperature and Surface Winds. NOAA Atlas No. 8,\nU.S. Department of Commerce, Washington D.C. [Copies can\nbe obtained from the Climate Analysis Center W/NMC52,\nWashington D.C. 20233].\nRasmusson E. M. , X. Wang, and C. F. Ropelewski, 1989: The\nbiennial component of ENSO variability. Jour. of Marine\nSystems (submitted).\nRopelewski C. F. , M. S. Halpert, and E. M. Rasmusson, 1988:\nTropospheric biennial variability and its relationship to\nthe Southern Oscillation. Proceedings of the 12th Annual\nClimate Diagnostics Workshop, October 12-16, 1987, Salt\nLake City, Utah, 26-134.\n, and\n, 1989:\n,\nClimatological aspects of the tropical annual cycle.\nProceedings of the 13th Annual Climate Diagnostics\nWorkshop, October 30-November 3, 1988, Cambridge, MA, 57-\n60.\n87","1950-87 NO MEAN\n1950-87 NO MEAN\nU-150E.\nU- - 70E\n4\n4\n2\n2\n0\n0\n60\n24\n12\n6\n5\n60\n24\n12\n6\n5\nU-170E\nU- 90E\n6\n4\n3\n2\n0\n60\n24\n12\n6\n5\n0\n60\n24\n12\n6\n5\nFig. 1 Spectra of the zonal wind components averaged over 10°\nlatitude by 20° longitude boxes centered on the equator and 70°E,\n90°E, 150°E, and 170° E. The abscissa gives the period in months.\nRed noise 95% confidence levels are also shown. (Courtesy of X.\nWang [after the analysis of Rasmusson et al , 1989]).\n88","COADS EQUATORIAL U AMPLITUDE\n4\nBIENNIAL\nLOW FREQUENCY\n3\n2\nQ\n1\n0\n50E\n70E\n90E\n110E\n130E\n150E\n170E\n170W\n150W\n130W\n110W\n90W\nLONGTITUDE\nFig. 2 Equatorial variance profile in the low-frequency band (40\nto 80 months - - solid) and biennial band (22-33 months - dashed) . An\n\"X\" indicates no peak in the spectrum. . A dot indicates a\nstatistically significant peak. A circled \"X\" indicates spectral\npeaks that due not pass the 95% level of significance (from\nRasmusson et al . / 1989).\n89","1962\n1963\n1964\n1965\n1966\n1967\n1968\n1969\n1970\n1971\n1972\n1973\n1974\n1975\n1975\n1976\n1876\n1977\n1978\n1879\n1979\n1979\n1980\n1900\n1966\n100E\n160E\nBOW\n40E\n140W\nBOW\n1962\nSST\nfrom\nFig.\n3\n.\nleft\nThe\n1980\nto\npanel\nshows\nhand\npanel\nthe\nshows\nthe\nand\nlow-\nthe\nfiltered\npass\nSST.\n90","ENSO-related Variability as revealed by\nSingular Spectrum Analysis (SSA)\nXueliang Wang and Eugene M. Rasmusson\nCooperative Institute for Climate Studies\nDepartment of Meteorology\nUniversity of Maryland\nCollege Park, MD 20742 USA\n1. INTRODUCTION\nThe ENSO cycle is oscillatory, but not periodic. Consequently, the\nquestion of the fundamental time scale (s) of the ENSO cycle naturally arise.\nMajor ENSO \"episodes\" are said to have a recurrence interval generally between\ntwo-seven years, averaging three-four years. However, the episodes included\nin any particular compilation depend on a vaguely defined level of intensity\nof some Southern Oscillation Index (SOI). In fact, there is a broad spectrum\nof amplitudes in any SOI series. With the inclusion of smaller amplitude\nswings of the SOI, the average recurrence time shortens, and ultimately\napproaches a biennial time scale (Rasmusson et al. 1989) This time scale\ncorresponds with that of the \"ENSO episodes\" themselves, which typically span\na period of around two years and exhibit a preferred phasing with the annual\ncycle (Rasmusson and Carpenter, 1982)\nShown in Fig. 1 are the cross spectrum products for the Darwin and\nTahiti sea level pressure (SLP) time series which contribute equally to the\nvariability of the widely used Tahiti minus Darwin SOI. There is significant\ncoherence between these two series for periods ranging from about two to seven\nyears. However, the autospectra of Darwin and Tahiti SLP exhibit quite\ndifferent characteristics. Darwin exhibits a single significant spectrum peak\naround a five years, while the spectrum for the Tahiti series shows two\nsignificant peaks, one near five years, the other around 30 months. This\ndouble-peak feature is common in near equatorial zonal wind and SST data from\nthe Pacific- Indian Ocean sector, as shown by Rasmusson et al. (1989)\n2. SINGULAR SPECTRUM ANALYSIS\nSingular spectrum analysis (SSA) (Vautard and Ghil, 1989), has been used\nto objectively identify dominant time scales of ENSO time series. In this\nsection, we show the application of SSA to the Darwin and Tahiti SLP time\nseries.\nIn applying SSA, a regular sampled time series (n) n=1,\nN\nis\nembedded in an M-dimensional space by taking as state vectors the consecutive\nsequences (n) X (n+1)\nX (n+M-1) for n=1 to N-M+1 and m=1 to M.\nWithin this embedding phase space, the dynamics of the system Y (n,m), which is\nequivalent to single time series (n), can be described statistically in a\nlinear way by its principal axes. The time scales of the dynamics which can\nbe reconstructed are bounded from below by the sampling interval and from\nabove by the length of the embedding window M used in forming the two\ndimensional array Y (n,m) The SSA eigenvalues, eigenvectors, and PCs of the\ntime series X(n) are obtained from an empirical orthogonal function (EOF)\nanalysis of the two dimensional array (n,m) with state vectors forming the\nrows.\nThe singular spectra for Darwin SLP for window sizes ranging from 7 to\n121 months are shown in Fig. 2. Typically, there is an initial, strongly\nnegative slope of the spectrum. The associated eigenvectors describe the\ndynamically significant part of the time series, i.e., the signal. If the\nremainder of the variance is white noise, it will be reflected in the spectrum\nas a sharp transition from a negative slope to a flat spectrum \"floor\"\nThe\ndistinction between the slopes of signal and the slopes of noise part of\nspectra becomes less with increasing embedding window size (Fig. 2) A\nwell-defined oscillation above a red noise background whose time scale is\nshorter than the window length will appear as an even and odd eigenvector pair\nin quadrature with each other, in the shape of the oscillation. The\nassociated eigenvalues of this pair of eigenvectors will be equal. For\nexample, the first two eigenvalues for each window size in Fig. 2 are nearly\nequal, and their corresponding eigenvectors define the annual component of the\ntime series (figure not shown) The annual component retains less variance\nwith increasing window length as it approaches its mean state, i.e., the\naverage annual cycle.\nThe determination of a suitable embedding dimension is a crucial aspect\nof SSA. The embedding dimension must be small enough to provide the desired\nsignificance, but large enough to resolve the principal eigenmodes. Plotted\nin Fig. 3 are total variance as explained by the dynamically significant\n91","eigenvalues for each window size. For smaller windows, the first\nfew\neigenvalues absorb most of the variance of the time series. For the extreme\ncase of M=1, one eigenmode contains all of variance of the series. The first\ntwo eigenvalues of Darwin (Tahiti) SLP for window M=7 contain 92 (78) percent\nof the total variance. However, these eigenmodes have poor resolution, and\ninclude noise as well as signal. With increasing window size, the explained\nvariance of the signal eigenmodes decreases and approaches a saturation level,\nsuggesting the effective removal of noise variance. The choice of window size,\ntherefore, is a well-balanced compromise between the statistical significance\nthe smaller the window the better, and the dynamical resolution -- the\nlarger the window the better. Such an optimum choice of window size is\nrepresented by the initial part of spectrum shown in Fig. 3. For the analysis\nof the Darwin and Tahiti series, a window size around 60 months seems to be\nthe best compromise for describing the dynamics.\nUsing an embedding window M=61, SSA of the Darwin series shows annual\nand semi-annual components, and one pair of eigenvectors with nearly equal\neigenvalues in the interannual frequency band. The associated principal\ncomponents (PCs) describe a well-defined oscillation with a period of\napproximately 60 months (solid line in Fig. 4), consistent with the power\nspectrum analysis (Fig. 1). For Tahiti, after removal of the annual cycle,\nthere emerge two eigenvector pairs with nearly equal eigenvalues that have\ninterannual time scales. Their PCs capture the characteristics of\noscillations with periods of around 60 months and 30 months, again consistent\nwith the spectrum analysis (Fig. 1). The time series reconstructed from those\nfour eigenmodes (dashed line in Fig. 4) are compared with 30 month and 60\nmonth band-pass filtered time series of Tahiti SLP. The reconstructed series\nis exactly in phase with the 30 month filtered version of the original Tahiti\nSLP series, and the amplitudes are also nearly identical. The 60 month\nfiltered Tahiti SLP series has a smaller amplitude and appears to pick up long\nterm trend of the reconstructed series from SSA PCs. This means that the\ndominant oscillation described by those four eigenmodes of the Tahiti sea\nlevel pressure is of around 30 months period.\nIn summary, the dominant oscillation described by the Darwin SLP time\nseries has a period around 60 months. Tahiti SLP exhibits two dominant time\nscales, one near 60 months and the other around 30 months. However, the 30\nmonth oscillation appears to be the more fundamental. In other word, the\ninterannual variability of the Tahiti SLP time series is essentially captured\nby the oscillation with period around 30 months. Consequently, the Tahiti\nminus Darwin SOI is a combination of time series with different dynamic\ncharacteristics, and therefore could be misleading when investigating the time\nscales of ENSO.\n3. CONCLUSIONS\n1. SSA provides a powerful tool for identifying the principal modes of\nvariability of observed time series. The evolution, structure, and\nsignificance of each mode can be assessed by analysis of SSA eigenvalues,\neigenvectors, and associated principal components.\n2. The most important parameter of the SSA, the embedding dimension M,\ncan be estimated from the variance saturation level explained by dynamically\nsignificant part of eigenvalues.\n3. SSA indicates two dominant time scales of ENSO variability around 30\nand 60 months in the Darwin and Tahiti SLP time series. The widely accepted\nthree to four year ENSO time scale seems difficulty to support objectively.\n4. Darwin and Tahiti sea level pressure exhibit very different dynamic\ncharacteristics Therefore the use of this traditional SOI as the \"canonical\nSO index\" has its limitations, and could be misleading when investigating the\nmechanism of ENSO.\n92","CROSS SPECTRA: 1933-88 (NO MEAN)\nDARWIN SLP\nX 10\nTAHITI SLP\nX 10\n36\n36\n18\n18\n0\n0\n60\n24\n12\n6\n5\n60\n24\n12\n6\n5\nCO-AMPLITUDE\nX 10\nCOH. & PHASE\n36\n90\n270\n1\n90\n18\n0\n0\n60\n24\n12\n6\n5\n60\n24\n12\n6\n5\nPERIOD (MONTH)\nPERIOD (MONTH)\nFig. 1. Cross spectrum analysis of the sea level pressure (SLP)\nat Darwin and Tahiti. Upper left: autospectrum for Darwin.\nAbscissa is period (months) Also shown is red noise 95%\nconfidence level. Upper right: autospectrum for Tahiti.\nLower left: covariance spectrum. Lower right phase\ndifference (degrees) and coherence, including 95% confidence\nlevel.\n93","DARWIN SLP\n1882-1988 SSA\n-1.0\nM=7\nM=13\nM=19\nM=25\nM=31\nM=37\n-2.0\nM=43\nM=49\nM=61\nM=91\nM=121\n-3.0\n-4.0\n-5.0\n60\n70\n80\n110\n0\n10\n20\n30\n40\n50\n90\n100\n120\n130\nORDER IN MONTH\nFig. 2. Eigenvalues of the singular spectrum of the Darwin SLP\nseries 1882-1988) for window sizes from M=7 to M=121\nmonths. The dashed line is an approximate separation between\nthe signal part (above the line) and the noise part (below\nthe line) of eigenvalues.\n94","VARIANCE PERCENTAGE OF SIGNAL\n100%\n100%\n90%\n95%\n80%\n90%\n0\n6\n12\n18\n24\n30\n36\n42\n48\n54\n60\n66\n72\n78\n84\n90\n96\n102\n108\n114\n120\nEMBEDDING WINDOWS\nFig. 3. Total variance explained by the signal part of\neigenvalues for window sizes from 7 to 121 The solid line\nis for the Darwin SLP series scaled by the left ordinate\nwhile the dashed line for Tahiti SLP series (1933-1988)\nis\nscaled by the right ordinate. The short vertical line\nindicates the approximate saturation position (M=61) of\nwindow size.\nTAHITI PC3-6 SLP\nDARWIN PC3-4 &\n10\n5\n0\n-5\n-10\n26\n28\n30\n32\n34\n36\n38\n1940\n42\n44\n46\n48\n50\n52\n54\n56\n10\n5\n0\n-5\n-10\n58\n1960\n62\n64\n66\n68\n70\n72\n74\n76\n78\n1980\n82\n84\n86\n88\n(YEARS)\nFig. 4. The solid line is the reconstructed time series from the\nthird and fourth principal components, which represent the\n60 month period oscillation, revealed by the SSA for Darwin\nSLP. The dashed line is the reconstructed time series from\nthe third to the sixth PCs, which jointly describe an\noscillation combined 60 and 30 month periodicity with the\nlater dominating both phase and amplitude, of the SSA for\nTahiti SLP The major ENSO events may be characterized by\nthe coincidence of the positive peaks of the solid curve\nwith the negative extrema of the dashed curve (vertical\nshaded)\n95","On the Interaction of Multiple Time Scales in the Tropical Climate System\nT. P. Barnett\nScripps Institution of Oceanography, La Jolla, CA 92093\nAbstract\nGlobal SLP and SST fields are band passed filtered to isolate their quasi-biennial (QB,\n20-30 mo. band) and Southern Oscillation (SO, 36-80 month band) elements. The nonlinear\ninteraction between the QB and SO time scales is shown to be strong and highly significant\nstatistically. Future attempts to explain or model ENSO must account for this nonlinear\ninteraction.\n1.0 Introduction\nThere is a growing body of evidence that El Niño - Southern Oscillation (ENSO) events\nmay be a two-time scale process. Individual warm/cold events often appear to have a\ntwo year life cycle (e.g. Trenberth, 1976, 1980; Rasmusson and Carpenter (1982); Barnett\n(1988), among others) and hence a quasi-biennial (QB) appearance. The separation between\nindividual events is normally greater than two years; typically 3-8 years (Trenberth, 1976;\nQuinn, el al, 1978 among others). We shall refer to this longer time scale as the SO scale,\nalthough the implied reference to the Southern Oscillation may be a bit misleading as we\nshall see.\nThe present work investigates the interaction between the two time scales noted above.\nCan they be thought of as individual entities unrelated to each other, for instance as indepen-\ndent Fourier-Stieltjes components? Or do the time scales interact such as to produce ENSO\nevents? The possibility of interaction between the two time scales is nearly unexplored in\nthe literature. van Loon and Shea (1985) suggest a strong relation between the occurrence of\nwarm/cold SST anomalies off South America and an atmosphere QB measure. More to the\npoint, Lau and Sheu (1988) estimate the bispectrum (Hasselmann, el al 1963) for a global\nprecipitation index. These authors find a remarkably strong nonlinear interaction between\na QB time scale (~2 25 month period) and a longer (SO) time scale (~ 50 months) although\nno attempt is made to estimate the statistical significance of their results. Aside from these\npioneering efforts there seems little, if any, work that quantitatively attempts to investigate\nthe relation between the two time scales.\n2.0 Data and Methods\nTwo large data sets are used in this study. A near-global sea level pressure (SLP) set\ncovering the region 40°S - 90°N at monthly intervals between 1951-1987 was available for\nanalysis. The basic data come from the standard products of various weather services and,\nin the tropics, direct ship observations. These various data have been transposed to a 5°x\n50 grid. A description of the data set and methods used to derive it are given by Barnett\n(1985).\nThe second data set to be analyzed is the monthly global sea surface temperature (SST)\nfield for the period 1951-1985. This set was obtained from the COADS and extended COADS\nsets by the methods described by Barbour (1986). The original 4°x4° grid has been projected\nonto a 5°x5° grid to conform to the SLP set. Data below 40°S will not be analyzed here due\nto its sparse nature.\n96","The analysis will rely on various filtered versions of the data. Let the monthly time series\nof a variable (T) at position It' be denoted by T(x,t) where the mean, annual cycle and trend\nhave been removed from T so < T >t= 0. The Fourier expansion of T is just\nwhere dn are the (complex) Fourier coefficients and Wn = Not NT with St = one month and\nN - the total number of time terms in T. The first and last year of data were tapered with a\ncosine bell to reduce end effects prior to Fourier transformation.\nThe filtered series are obtained by reconstructing an estimate of T, say , retaining a\nlimited set of discrete Fourier components, e.g.\nwhere\nWn = 1\nfor\nN1<n<n2\nfor mio sn<n1\nWn =0 otherwise.\nThe properties of the filters are given in Table 1.\nTABLE\nName\nWnio\nWn,\nWn,\nWNri\nSO\n2T/88 mo\n2n/80\n2/36\n2/32\nQB\n2/35 mo\n2/30\n2/20\n2/18\nIn general, the SO filter captures up to 50% of the SST/SLP variance in the tropics while\nthe QB filter accounts for up to 30% of the variance.\n3.0 Interaction of Time Scales\n3.1 Nonlinear interaction coefficients\nThe purpose of this section is to quantify the interaction, if any, between two character-\nistic time scales: the SO (time scale T8 and frequency fs = 1/Ts), and the QB (time scale TQ\nand frequency fq). Since the Complex Empirical Orthogonal Functions (CEOFs) represent\nnatural modes of the data, their principal components, A will be used in the analysis (cf\nBarnett, 1983).\nNonlinear interaction between the time scales is given by what we shall term the nonlinear\ninteraction coefficient, i.e.,\n97","(1)\nWith this normalization, bijl = Bij is constrainted to lie in the range 0 to 1. We\nshall refer to B as the bicoherence. The side condition on the calculation of bij (a complex\nquantity) is that if the three members of the numerator have frequencies firfj, and fk\nrespectively, then fi+fj = fk. With this side condition (1) may be interpreted physically to\nrepresent two equal interactions between frequencies fi, fj and f-i-j and fii f-j and fitj\nThe center frequencies of the filters noted in Section 2 suggest we can explore the relations\nbetween SO and QB, i.e. (fsfsfq).\nThe physical meaning of (1) is made clear by realizing that under the restriction i = j\nthe first two terms of the numerator are the square of a particular signal, i.e. proportional to\nits energy. With that fact in mind, one can simply view (1) as a measure of the correlation\nbetween the energy of, say, the SO signal and the amplitude of the QB signal. Using this\napproach to studying nonlinear interactions between climate time scales is comparable to use\nof the conventional bispectrum (e.g. Hasselmann el al, 1963), but less fraught with numerical\nproblems given the extreme shortness of the climate time series.* The connection between the\ntime domain estimate of the nonlinear interaction coefficient (1) and the frequency domain\nbispectrum is provided by the side condition on the frequencies used in (1).\nAnother way to view the result of (1) is to consider Bij a measure of the degree to which\nthe signals at frequencies fi and fj are phased locked. Given the fact that the signals at fi\nand fj are produced by a finite band pass (Table 1) suggests that they could drift in and out\nof phase leading to low values of Bij if there is no phase locking between the signals. Thus,\nB provides a quantitative measure of the degree to which variations in the fi band can be\nthought of as harmonically coupled to the fj band.\nThe significance of Bij was determined via a Monte Carlo technique. The complex time\nseries, A(t) were modeled as second order autoregression processes whose relative phases are\nrandomly chosen. A thousand realizations of Bij from these randomly forced autoregressive\nmodels then provide a good reference probability distribution function on which to base\nsignificance claims. Although slightly dependent on variable types and CEOF mode number,\nvalues of Bij > 0.45 can conservatively be taken as significant in this study at the 0.05 level.\n3.2 Interaction of the SO and QB Time Scales\nThe BSQ for the first CEOF of the SST field was 0.54 and its argument, SQ =\nbij/Re bij) was 123°. This means, in effect, that in the tropical SST field there\nis a significant harmonic phase locking between the QB and SO frequency bands such that\nthe energy in the SO band, the magnitude of the SO signal, is closely associated with the\nQB variations. This value is characteristic of to the whole domain of analysis (40°S - 60°N).\nLocally in the tropical zone where the signal is strongest, the coupling is likely stronger.\nValues of BSQ for all other combinations of CEOF modes up to M = 3 were insignificant.\nApparently the principal SO-QB interaction in the SST field is associated with only the first\n*\nAs a rule of thumb, stable bispectral calculation in the frequency domain would require\nmonthly time series of length 500-800 years. Time series of climate data are rarely, if ever,\nthis long, e.g. a 100 year time series of monthly data generally would fall short of providing\nstable bispectral estimations unless transfrequency coupling were extraordinarily strong.\n98","CEOF. The fact that 0SQ ~ 123°means fluctuation in SST energy in the SO band leads those\nin the QB band by 1/6 of an SO cycle or roughly 8-10 months.\nA significant interaction in the SLP field also only occurred for the first CEOF (BSQ =\n0.57,0 = 96°). This interaction is of the same strength as that found in the SST field. The\nphase again suggests variation in the SO band that lead those in the QB band by 7-9 months.\nLike the SST result, no other SLP modes (up to M = 3) seem to interact with each other.\nOf considerable interest is the interaction between the SLP and SST field. This may be\ndetermined by using the CEOF series derived from the analysis of the two fields. Hence in\n(1) let (A, Aj) be As (t/SST) and Ak are AQ (t/SLP). In this example, bij estimates the\ndegree of interaction between the energy of the SO filtered SST field and the amplitude of\nthe QB filtered SLP field. It is critical here to ensure that the CEOF analyses have all been\nreferenced to a common base (the QB filtered SLP signal at Darwin in this case). Failure to\ndo this will not effect Bij, but will make the phase estimate meaningless.\nThe strongest cross interaction coefficient was associated with the triple (SSTs, SSTS,\nSLPQ) which had B = 0.62 and SQ = 107°. This means the energy in the low frequency\nSST field couples with higher frequency amplitude in the SLP field. The phase angle shows\nthat changes in the SO band of the SST field leads those in the SLP field by 7-8 months. The\ncomparable cross interaction between the SLP and SST field gave B = 0.55, SQ ~ 115°, i.e.\nthe cross interaction was nearly identical to that found for (SSTs, SSTS, SLPQ). As noted\nabove, the SO energy fluctuations in the SST field lead the QB fluctuations in the same field\nby 8-10 months; the difference between these estimates of lead being inconsequential. Thus,\nit appears the QB signal in the SST/SLP field can be thought of as harmonically coupled to\nthe SO time scale. The two time scales are not independent entities but interact with each\nother to modify their respective energies.\n4.0 Discussion and Summary\nThe above results clearly show a strong nonlinear interaction between the SO and QB\nbands. This important result dictates a new class of \"simple\" ENSO models, for none of\nthose currently available appear to allow for such an interaction. But the results also present\na dilemma - which is the more fundamental time scale? Is the QB just the first harmonic of\nthe SO band? Or, is the SO the first subharmonic of the QB frequency band? Physically,\nthe dilemma is stated as: (a) we can postulate that the energy of the SO signal is amplified\nby the strength of the QB; or, alternatively, (b) the QB signal is stronger at extremes of the\nSO. Which way does the energy flow? The bicoherence analysis, by itself, cannot directly\ndifferentiate between the two possibilities.\nThere are a series of facts that suggest the SO time scale may be the more fundamental\nof the two. These are listed below:\ni) The SO time scale generally contains more energy than the QB band over much of the\nIndo-Pacific.\nii) It is unlikely, though possible, that a subharmonic would be more energetic than the\nfundamental. More likely the first harmonic of a nonlinear process will be less energetic\nthan the fundamental (but again exceptions can be constructed). This favors the SO as the\ndominant scale.\niii) The phase associated with the cross-interaction coefficients suggest the SO variations in\nthe SST field lead all others. While discussions of phase of quasi-periodic signals are tenuous\nthe lead/lags inferred above seem physically more realistic than the alternatives.\n99","iv) The lack of a good phase locking of the QB energy with the annual cycle (shown elsewhere)\nsuggests we must look elsewhere for the origins of variations in this frequency band.\nv) Typical ENSO events are nonsymmetrical in time. They take longer to spin up than to\nspin down. Similarly, there may be a few years where \"nothing happens\" followed by an\nevent lasting 1-2 years, again an asymmetrical behavior in time. Such behavior will give rise\nto the harmonically peaked spectra shown by Rasmusson, el al (1989); i.e. the spectra are\njust the result of applying a linear analysis to the asymmetrical time series. The type of\ninteraction coefficients found here will also be obtained for such temporal behavior. Hence\nin this scenario the QB spectral peak is simply an artifact of the analysis.\nTaken together, i) - v) support the idea that the SO band represents the more funda-\nmental time scale in the Indo-Pacific. But this result must be taken as highly tentative.\nOther analyses suggest the QB and SO variations could be separate entities, i.e. generated\nby a different sets of physics. In fact, the above bicoherence results could be obtained in a\nscenario where the QB signal serves as a triggering mechanism to alter the state/phase of\nthe SO variations (Barnett, el al, 1990). This possibility is not precluded by significant val-\nues of Bij. Hopefully, this unsatisfactory state of understanding regarding the relationship\nbetween the QB and SO bands will be resolved shortly. For now, we can only conclude with\ncertainty that the two bands interact in an important, but nonlinear, fashion.\nAcknowledgement\nSupport for this work was provided by the NSF Climate Dynamics Program under Grant\nNSF ATM88-14571.\nReferences\nBarbour, P., 1986: Construction of a monthly sea surface temperature for the global ocean:\n1950-79, SIO Ref. 86-26, available from T. Barnett, A-024, Scripps Institution of\nOceanography, La Jolla, CA 92093.\nBarnett, T. P., 1983: Interaction of the monsoon and pacific trade wind system at interannual\ntime scales. Part I: The equatorial zone. Mon. Wea. Rev., 111, 756-773.\nBarnett, T. P., 1985: Variations in near global sea level pressure. J. Atmos. Sci., 42,\n478-501.\nBarnett, T. P., 1988: Variations in near global sea level pressure: another view. J. of Clim.,\n1, 225-230.\nHasselmann, K., W. Munk and G. MacDonald, 1963: Bispectra of ocean waves, in Time\nSeries Analysis, M. Rosenblatt (Ed.), John Wiley, New York, pp 125-139.\nLabitzke, K. and H. van Loon, 1988: Association between the 11 year solar cycle, the QBO\nand the atmosphere: Part I: The troposphere and stratosphere in winter, J. Atmos. and\nTerr. Phys., 50, 197-206.\nLau, K.M. and P. Sheu, 1988: Annual cycle QBO and Southern Oscillation in global precip-\nitation, J. Geophys. Res., 93 (D9), 10975-10988.\nQuinn, W., D. Zopf, K. Short, and R. Yang, 1978: Historical trends and statistics of the\nSouthern Oscillation, El Niño, and Indonesian drought, Fisheries Bull., 76(3), 663-678.\nRasmusson, E. M., and T. H. Carpenter, 1982: Variations in tropical sea surface temperature\nand surface wind fields associated with the southern oscillation/ El Niño. Mon. Wea.\nRev., 110, 354-384.\n100","Rasmusson, E., X. Wang, and C. Ropelewski, 1989: The biennial component of ENSO\nvariability. Proc. 21st Int. Liege Colloquium on Ocean Hydrodynamics, Liege, Belgium.\nTrenberth, K.E., 1976: Spatial and temporal variations of the Southern Oscillation, Quarerly\nJ. Royal Meteor. Soc., 102, 639-654.\nTrenberth, K. E., 1980: Atmospheric quasi-biennial oscillations. Mon. Wea. Rev., 108,\n1370-1377.\nvan Loon, H. and D. Shea, 1985: The southern oscillation. Part IV: The precursor south of\n15°S to the extremes of the oscillation. Mon. Wea. Rev., 113, 2063-2074.\n101","TROPICAL OCEAN-ATMOSPHERE INTERACTIONS AT THE\nINTRASEASONAL TIME SCALE\nDuane E. Stevens and Qi Hu\nDepartment of Atmospheric Science\nColorado State University\nFort Collins, CO. 80523\n1. Introduction\nAs first identified by Madden and Julian (1971, 1972) the intraseasonal oscillation\nin the tropical atmosphere is dominated by an eastward propagating disturbance in\nthe upper and lower level wind and pressure fields. This disturbance has the char-\nacteristics of one wavelet around the globe and a 10-15ms-1 traveling phase speed.\nFollowing MJ's discovery other observational studies have found the existence of sim-\nilar time scale oscillations in the atmospheric angular momentum; the convective ac-\ntivity and cloudiness; the outgoing long wave radiation; and the precipitation process\nin the tropical Indian/Western Pacific Oceans. Today, the 'intraseasonal oscillation'\nrepresents a complex of eastward propagating oscillations in the tropical atmosphere\nand ocean with a time scale of 30-60 days.\nUnderstanding the physical mechanism(s) of the intraseasonal oscillation is nec-\nessary because: (1) this oscillation is related to the El Nino/Southern Oscillation\nphenomenon (Lau and Shen, 1988), and (2) it affects the timing of the Asian and\nAustralian monsoons, as well as their strength and activity. This understanding may\nlead to greater knowledge and hence greater ability in prediction of global climate\nchange.\n2. The Hydrological Cycle in Tropical Atmosphere-Ocean System\nAn intensive review of the present studies, both observational and theoretical, on\nthe tropical intraseasonal oscillation has led us to focus on a consideration of the local\nthermodynamic processes in the tropical atmosphere-ocean system, i.e., the hydro-\nlogical cycle. In the tropics the intense solar heating at the ocean surface is consumed\nto drive the sensible and latent heat fluxes into the atmosphere. The latent energy\naccumulated in the lower tropical troposphere is eventually released in the upper tro-\nposphere and at the same time precipitation is produced. This release of latent energy,\nassociated with the phase change of water vapor in the atmosphere, represents the\nmajor heat transfer in the atmosphere and provides the needed energy to balance the\nenergy deficit in the upper troposphere due to infrared radiative cooling. During the\nrelease of latent energy, the cirrus clouds associated with convective activity interact\nwith solar radiation in the atmosphere and attenuate the amount of energy available\nat the ocean surface. As a consequence of the change in the energy budget, the sea\nsurface temperature and the water vapor flux from the ocean surface are reduced.\nThis negative feedback to the energy transfer process in the atmosphere causes the\nconvection to dissipate. This has been proposed as a hypothesis to interpret the\n102","intraseasonal time scale oscillation and has been tested in different one dimensional\natmosphere-ocean models (Stevens et. al., 1989).\n3. An Improved One-Dimensional Model\nIn the model presented in this article a detailed atmospheric temperature and\nmoisture structure is the model atmosphere. Some details about the model follow.\n(1) Model\nThe model is a 1-D atmosphere-ocean model. The model atmosphere is divided\ninto 17 levels and each level has its pressure, temperature and mixing ratio ; the\n'ocean' contains only a shallow mixed layer with fixed depth and temperature (and\ntherefore a fixed sea surface temperature (SST)).\n(2) Boundary Conditions\nAt the top of the atmosphere a balance between the net input solar radiation\nand the output terrestrial radiation is assumed. It is also assumed that there is\nno energy flux at the bottom of the ocean mixed layer. At the interface between\nthe atmosphere and the ocean, solar and infrared radiation absorbed by the ocean is\ncompletely converted into sensible and latent heat and transports into the atmosphere,\nconsistent with the assumption of fixed SST and the condition at the lower boundary.\nMonin-Obukhov similarity theory is applied to determine the amount of sensible heat\nflux and flux of water vapor above the atmospheric surface layer.\n(3) Dry and Moist Convective Adjustment in Model Atmosphere\nThe re-distribution of the sensible and latent heat added into the lower atmosphere\nis accomplished through dry and moist convective adjustments (Manabe et al., 1965).\nAs in the lower troposphere over the tropical oceans, vertical mixing due to buoy-\nant convection forced by surface evaporation is efficient (Riehl, 1979, and Lindzen,\n1981); in this model the water evaporated into the atmosphere is distributed in the\nlayer below 700mb (representing mixing by trade cumuli) before the moist convective\nadjustment. The atmospheric layers above the troposphere (here, from 1000mb to\n200mb) are assumed in radiative balance and not considered in the model.\n(4) Energy Balance in Model Atmosphere\nThe only energy sink in the atmosphere is the IR cooling. A Newtonian cooling\nformulation is applied in the model atmosphere to represent thermal dissipation and\nthe bulk effect of the neglected dynamic processes on the thermal balance of the model\natmosphere. This parameterization maintains the thermal state of the atmosphere\nnear the prescribed climatological mean state.\n4. Some Preliminary Results and Discussions\nAs shown in Fig. 1, the precipitation oscillates on an intraseasonal time scale with\nsignificant amplitude. (Note that rain is the only drying-out process in the model.)\nFig. 1 suggests that this oscillation, in nature, is determined by the moist ther-\nmodynamic characteristics and the radiation properties of the tropical atmosphere.\nAnother interesting feature from this result is that the oscillation asymptotes to an\nequilibrium situation. This, of course, is a product of the competition between con-\n103","vective heating and radiative cooling in the model atmosphere during the integration.\nThese competing processes eventually lead to a situation in which the convective fluc-\ntuations appear random. We speculate that the break up of this random fluctuation\nand reset of the oscillation may come from ocean feedbacks (not included here).\nRemoval of cloud-radiation influence on the surface energy budget causes two\ndifferent results. With a low SST value the oscillation feature essentially disappears\n(Fig. 2). With a high SST, a dominant period of 35 days is still significant (Fig. 3).\nThis suggests that over warm oceans, atmospheric thermal and radiation properties\nare fundamental for the existence of this long time scale oscillation. On the other\nhand, over cool oceans the evaporation fluctuation induced through cloud-radiation\ninteraction becomes a crucial pre-condition for the presence of this oscillation.\nOur thermodynamic modelling studies are continuing in order to further reveal the\nphysical processes responsible for the intraseasonal oscillation. The study of possible\nrelationships between the intraseasonal oscillation and interannual oscillation as well\nas the influence of the intraseasonal oscillation on Asian and Australian monsoon\nprocesses may contribute to a unified theory capable of explaining the observed large\ntime scale oscillation phenomena in the tropics.\n5. Acknowledgments\nThis research has been supported by NSF under grant ATM-8609731 and by\nNASA under grant NAG-5-1122 to Colorado State University.\n6. References\nLau, K.M. and S. Shen, 1988: On the dynamics of intraseasonal oscillations and ENSO. J.\nAtmos. Sci., 45, 1781-1797.\nLindzen, R.S., 1981: Some remarks on cumulus parameterization. Proc. of the NASA Cloud,\nClimate Conf., NASA Report.\nMadden, R., and P.R. Julian, 1971: Detection of a 40-50 day oscillation in the zonal wind\nin the tropical Pacific. J. Atmos. Sci., 28, 702-708.\nMadden, R., and P.R. Julian, 1972: Description of global-scale circulation cells in the\ntropics with a 40-50 day period. J. Atmos. Sci., 29, 1109-1123.\nManabe, S., J. Smagorinsky, and R.F. Strickler, 1965: Simulated climatology of a general\ncirculation model with a hydrological cycle. Mon. Wea. Rev., 93, 769-798.\nRiehl, H., 1979: The trade wind inversion. Climate and Weather in the Tropics. Academic\nPress, 611 pp.\nStevens, D.E., Q. Hu, G.L. Stephens, and D.A. Randall, 1989: The hydrological cycle of\nthe intraseasonal oscillation. Western Pacific International Meeting and Workshop\non TOGA-COARE. Noumea, New Caledonia, May 24-30,1989.\n104","12.0\n11.0\n10.0\n9.00\n8.00\n7.00\n6.00\n5.00\n4.00\n3.00\n2.00\n1.00\n0.\n0.\n100\n200\n300\n400\n500\n12.0\n11.0\n10.0\n9.00\n8.00\n7.00\n6.00\n5.00\n4.00\n3.00\n2.00\n1.00\n0.\n300\n400\n500\n600\n700\n800\n900\n1000\nTIME (DAYS)\n0.300\n0.250\n0.200\n0.150\n0.100\n0.0500\n0.\n200\n175\n150\n125 100 75.0 50.0 25.0 0.\nPERIOD (DAYS)\nFig. 1: The upper two panels show the model precipitation output during the time period of\n1000 days. The bump in the upper straight lines in these two panels indecates that the\nnumber of condensation levels in model atmosphere is 5. The bottom panel shows the\nresult of a power spectral analysis of daily precipitation amount from 300 to 1000 days.\n105","11.0\n10.0\n9.00\n8.00\n7.00\n6.00\n5.00\n4.00\n3.00\n2.00\n1.00\n300\n400\n500\n600\n700\n800\n900\n1000\nTIME (DAYS)\n0.300\n0.250\n0.200\n0.150\n0.100\n0.0500\n0.\n200\n175\n150\n125 100 75.0 50.0 25.0 0.\nPERIOD (DAYS)\nFig. 3: As in Fig. 2 except that SST=28.5°C.\n106","11.0\n10.0\n9.00\n8.00\n7.00\n6.00\n5.00\n4.00\n3.00\n2.00\n1.00\n300\n400\n500\n600\n700\n800\n900\n1000\nTIME (DAYS)\n0.300\n0.250\n0.200\n0.150\n0.100\n0.0500\n0.\n200 - 175\n150\n125 100 75.0 50.0 25.0 0.\nPERIOD (DAYS)\nFig. 2: Result of model precipitation and the dominant period of oscillations in the precipitation\noutput for a model condition of no consideration of influence of cloud attenuation of\nradiation on surface energy budget. SST = 25.0°C.\n107","RELATIONSHIP BETWEEN SUMMERTIME RIDGE POSITION\nAND CONTINENTAL RAINFALL REGIMES, 1929-1988\nArthur V. Douglas\nPhillip J. Englehart\nCreighton University\nMidwest Research Institute\nOmaha, Nebraska 68178\nKansas City, Missouri 64110\nRidge line position strongly effects moisture flux during the summer\nacross North America. Lower tropospheric ridge position across the\ncontinent can be calculated back to 1929 using pilot balloon data for\nthe 3,000m level (Figure 1). The 3,000m level is nearly equivalent to\nthe conventional 70kPa level, which during the summer is near 3,180m.\nThe pilot balloon data set is based on once daily observations, with\nmost stations reporting 26-30 observations per summer month (biased\ntowards non-cloudy days). Maps of ridge position for May through\nSeptember have been constructed for two observational data sets:\npilot balloon, 1929-1946; and radiosonde, 1947-1988. Ten year running\nmeans of ridge position during July and August at 95° W and 110 W are\ngiven in Figures 2 and 3.\nA statistical analysis of the data, 1929-1988, indicates that the\nridge position at 95 W shows greater fluctuation than that at 110 W\n(Table 1). One year lag correlations indicate statistically\nsignificant persistence from July to July and from summer to summer at\n95 W but not at 110 W. There is no indication of a QBO signal in\nridge position. The one year auto correlations clearly point towards\nthe tendency of drought years or wet years to persist in the Central\nPlains (see papers by Namias). This is borne out in the correlations\nbetween ridge position and Kansas statewide temperature and\nprecipitation frequency (days >.10\") The correlation between ridge\nposition and July-August temperature is 0.61 (significant at the 99%) .\nThe correlation between ridge position and July-August precipitation\nfrequency is -0.34 (significant at the 95% level).\nPositioning of the lower tropospheric ridge across North America is\nbelieved to be a key control in the summer monsoon of Mexico.\nWallen\n(1955) noted that rainfall in central Mexico tended to show large\ndecadel fluctuations, and these fluctuations were thought to be\nrelated to position, orientation and strength of the ridge. These\ndecadel fluctuations are clearly evident in Figure 4. The\nexceptionally wet period in the 1930s corresponded with persistent\nsevere drought in the Central Plains of North America.\nOverlaying the rainfall record of Tacubaya, D.F., onto the ridge\nposition at 95 indicates an in-phase relationship through the early\n1960s, but an out-of-phase relationship since then (Figure 5). Active\nmonsoon years prior to the early 1960s were associated with northward\nexcursions in the ridge (e.g. 1930s), while droughts occurred with a\nridge position farther south and close to the main monsoon region of\nMexico. This relationship is as expected based on the hypothesis\npresented by Wallen (1955).\n108","The exceptionally poor monsoons during the summers of the late 1940s\nand early 1950s were associated with positive height anomalies across\neastern North America and an anomalous southeast flow into Mexico\n(Figure 6). This unexpected relationship seems ideal for a wet\nregime, rather than the observed drought. Likewise, the wet years of\nthe 1960s and 1970s are associated with a weakened ridge and an\nanomalous northwest flow through Mexico. This anomalous pattern\nwould, at first glance, suggest drought should have occurred. An\nanalysis of the actual 70kPa data helps explain the climate dynamics\nassociated with drought in the 1940s and 1950s and enhanced monsoon\nrains in the 1960s and 1970s (Figure 7). The drought regime\nof\nthe\n1940s and 1950s was associated with very strong trades from the\nnormally dry Caribbean. Thus while the flow was enhanced, the air was\ndrier than normal in association with a strong trade wind inversion\nand descending motion. The wet years of the 1960s and 1970s were\nassociated with a weaker southeast flow that was presumably more moist\ndue to a weakened trade wind inversion.\nAnother factor influencing monsoon rainfall is tropical storm\nactivity. The four severe drought years of the late 1940s and 1950s\nwere typified by few Gulf of Mexico storms, with most Atlantic storms\nmoving due north. In contrast the four wet years of the 1930s had\ntwice as many storms, many of which moved due west into the Mexican\nplateau (Figure 8). The daily tracks of these storms indicated fairly\nlight easterlies.\nAn idealized representation of circulation during the 1930s is\npresented in Figure 9. As already shown in Figure 2, the ridge\nposition of the 1930s was very far north, favoring deep tropical\neasterlies and perhaps an intermittently established upper\ntropospheric easterly tropical jet across Mexico (as observed with the\nIndian monsoon) . Hurricane tracks and pilot balloon data suggest the\ndeep easterlies even penetrated into south Texas, where summer\nrainfall was above normal in the 1930s. At the same time, the Mexican\nmonsoon of the 1930s was additionally enhanced by strong\ncross-equatorial flowing during an extended period of cold SSTs\n(strong southeast trades) off South America (1931-1939). Enhanced\ntropical storm activity would have been favored during this cold\nperiod as a weakened TUTT in the Caribbean is generally associated\nwith anti-El Niño conditions.\nTABLE 1. RIDGE LINE STATISTICS AND AUTO CORRELATIONS.\nRIDGE LINE CORRELATIONS\nRIDGE LINE STATISTICS\n1929-1988\n1929-1988\n110°W\n95°W\nLAG PERIOD\n110°W\nMONTH/SEASON\n95°W\nARIZONA\nTEXAS\nARIZONA\nTEXAS\n1 YEAR LAGS\n.38*\n.11\nJULY-JULY\n32.8°N\nX\n32.3°N\nJULY\n-.05\n.12\nAUGUST-AUGUST\n32.7°\n32.8°\n.03\nAUGUST X\n.33*\nSUMMER-SUMMER\n32.8°\n32.4°\nJULY+AUGUST X\n2 YEAR LAGS\n1.4°\n2.2°\n.14\n.19\nJULY+AUGUST SD\nSUMMER-SUMMER\n*Significant .05 level.\n109","PILOT BALLOON\nWIND OBS\n3000m\nRIDGE LINE POSITION\nJULY 1936\nFigure 1. .\nExample of 3,000m pilot balloon wind data and\nridge position for July 1936.\nLatitude of 700 MB Ridge Line\nat 95° W (Texas)\n10 Year Means Ending at Year Plotted\n35.0\n34.0\n33.0\n32.0\n31.0\n5\n10\n15\n20\n25\n30\n35\n40\n45\n50\n55\n60\n65\n70\n75\n80\n85\n90\nYEAR\nFigure 2 .\nTen O. year running means of July-August ridge\nposition at 95 W. Period of analysis 1929-1988.\nLatitude of 700 MB Ridge Line\nat 110° W (Arizona)\n10 Year Means Ending at Year Plotted\n35.0\n34.0\n33.0\n32.0\n31.0\n5\n10\n15\n20\n25\n30\n35\n40\n45\n50\n55\n60\n65\n70\n75\n80\n85\n90\n95\nYEAR\nFigure 3 o\nTen year running means of July-August ridge\no\nposition at 110 W. Period of analysis 1929-1988.\n110","Total July and August Precipitation (mm)\nTacubaya, D.F., Mexico\n10 Year Means Ending at Year Plotted\n400\n375\n350\n325\n300\n275\n250\n225\n200\n80\n85\n90\n55\n60\n65\n70\n75\n25\n30\n35\n40\n45\n50\n5\n10\n15\n20\nYEAR\nTen year running means of Tacubaya July-August\nFigure 4 .\nrainfall (Mexico City National Observatory) .\nNote\nlarge\nterm trends with wet 1930s, 1960s and 1970s; and dry 1940s\nand 1950s.\nTotal July and August Precipitation (mm)\nTacubaya, D.F., Mexico\n10 Year Means Ending at Year Plotted\nLatitude of 700 MB Ridge Line\nat 95° W (Texas)\n10 Year Means Ending at Year Plotted\n400\n35.0\n375\n350\n34.0\n325\n300\n33.0\n275\n250\n32.0\n225\n31.0\n75\n80\n85\n90\n50\n55\n60\n65\n70\n5\n10\n15\n20\n25\n30\n35\n40\n45\nYEAR\nGraph of ridge position at 95°W and Tacubaya\nFigure 5. .\nreversal\nin\nrainfall\nfor\nJuly\nand\nAugust.\nNote\nrelationships early 1960s.\n111","60\n60.\nto\n15\nto\n10\n$\nB120\nASO\n16\n5140\n3160\n130\n18\n3170\n3180\n3190\n40.\n3120\n3140\n300\n03180\n20\n3170\n20\n20\n3160\n130\n100\n130\n100\n120-5\n110\n110 3150\n120\n10 for\n10\nSUMMER 70kPA ANOMALIES (m)\nSUMMER 70kPA (m)\n1947, 49, 52, 53, 54\n1947, 49, 52, 53, 54\n260\n3110\n3160\n3170\n5\n50\n50\n130\n3180\nSHO\no\n$190\n3/20\nI\n530\n31401\nCart\n3190\n30\n30\n3480\n3150\n5\n03170\n8\n3160\n20\n7\n20\n5\n130\n130\n100\n5\n120\n110\no\n120\n110\n10\n10\nSUMMER 70kPA ANOMALIES (m)\nSUMMER 70kPA (m)\n1963, 67, 69, 71, 73, 76\n1963, 67, 69, 71, 73, 76\nFigure 6. . The 70kPa height anomalies\nfor a dry period in the late 1940s and\n1950s and a wet period from the 1960s\nto 1970s.\nFigure 7 . The 70kPa heights for the\ndry and wet periods examined in Figure 6.\nNote large pressure gradient changes\nbetween the two periods from Mexico into\nthe western Caribbean.\n112","DRY\nWET\nYEARS\nYEARS\nTROPICAL CYCLONE TRACKS\nTROPICAL CYCLONE TRACKS\nJUNE - AUGUST\nJUNE - AUGUST\n1949, 52, 53,54\n1931, 33, 34,36\nFigure 8 . Hurricane tracks for the four wettest years and 4\ndriest years in Tacubaya (1930s vs. late 1940s and 1950s)\n113","1930'S REGIME\n30W\n110W\ngow\n70W\nshow\nFOR\nDROUGHT\nH\nRIDGE\n700MB\nRIDGE\n200MB\nH.\nWEAK\nEASTERLY\nWEAK\nJET\nTROF\nTRADES\n10N\n10S\nANTI-EL NINO\n30S\nH\n130W\n110W\n90W\now\n50W\n39W\n40S\nFigure 9. @ Idealized synoptic regime during the 1930s, based\non various proxy series discussed in the next. This regime\nresembles the much stronger Asian summer monsoon. Far north\nposition of the continental ridge favored enhanced monsoon\nconditions in Mexico in the 1930s.\n114","Do the Upper Air Data for the United States Indicate CO2 Induced Warming?\nJ. Gavin, G. Kukla and T. Karl*\nLamont-Doherty Geological Observatory\nPalisades, New York 10964\n*National Climate Center\nFederal Building\nAsheville, North Carolina 28801\nMost climate models predict that the doubling of CO2 will result in greater\nwarming of the middle troposphere with respect to the surface air temperature\n(Schlesinger and Mitchell, 1985). In order to determine whether current\nobservations are in qualitative agreement with the model predictions, we\ncompared trends of seasonal surface air temperature with those of the free\natmosphere between the 500 and 850 mb level for several geographic regions\nof the contiguous United States over the 1948-86 interval.\nThe surface data are based on approximately 500 stations from the\nmonthly Historical Climatology Network (HCN) which have been adjusted for\nmissing data, station moves, changes in instrumentation and observing\nschedules, as well as urban heat island effects. Height differences between\n500 and 850 mb have been calculated for 51 stations based on daily\nobservations taken at 00Z and 12Z. Both readings had to be present in order to\ncalculate the daily mean thickness and the monthly averages are based on a\nminimum of 26 days. Twenty meters was taken to represent a 1°C temperature\nchange. Departures of temperature and thickness in both data sets are from the\n1956-1985 means. The number of stations going into the average can vary over\ntime. The data were reduced to seasonal averages with winter represented by\nDecember, January and February.\nThe distribution of the upper air stations and the regional subdivisions\nreferred to in this preliminary study are shown in Figure 1. The HCN data are\nareally weighted for the East, Central and West divisions. The HCN data for the\nthree smaller regional blocks and for the upper air data for all study regions are\nrepresented by simple arithmetic averages. Trends are determined based on\nsimple linear regressions.\nFigure 2 shows the time series of differences between departures of\nupper air temperature and the surface air temperature for winter and summer\nover the United States. The data for winter show a slight warming of the mid\ntroposphere with respect to the surface in the West and Central regions and a\npronounced warming in the East. Tropospheric cooling relative to the surface is\nindicated in the West but warming is shown in the East. Little, if any difference is\nfound in the Central region of the country. No trends were found for the\ntransition seasons of spring and autumn.\nThe association between departures in upper air thickness and the\nsurface temperature is illustrated in Figure 3. A strong positive relationship is\nshown for all three regions for both summer and winter. The departures are well\nrelated throughout the remainder of the year.\nTime series for the winter season are also shown for several smaller\nregions separately for the upper air and surface air temperatures as well as the\ndifferences between the two. Over the Eastern Prairies the temperature trends\nat both the mid tropospheric and surface level show cooling with the decrease\n115","less pronounced in the upper air data. In the Northern Plains a very small\ndecrease is found in 850-500 mb level and practically no change at the surface.\nWarming is indicated in the upper air time series of the East Slope North\nCascades with little change at the surface.\nBased on this preliminary analysis we conclude that as yet no\nunambiguous signal of the CO2 differential warming of the middle troposphere\nrelative to the surface is found. We plan to further investigate taking into account\ndifferent circulation regimes and the potential influence of El Nino and La Nina\nevents, for example.\nReferences\nSchlesinger, M.E. and J.F.B. Mitchell (1985) Model projections of the time-\ndependent response to increased carbon dioxide. in: Projecting the\nClimatic Effects of Increasing Carbon Dioxide, (MacCracken M.C. &\nLuther, F.M., eds) U.S. Dept. of Energy, DOE/ER-0237, 81-147.\nFigure Captions\nFigure 1:\nA: The geographic regions for the West, Central and East\nUnited States delimited by solid heavy lines. Blank areas\nfrom West to East are the East Slope North Cascades,\nNorthern Plains and Eastern Prairies.\nB: The location of upper air stations marked by circles.\nSolid ones denote stations corresponding to blank areas\nshown in A.\nFigure 2:\nThe time series of differences between the 850-500 mb\nand surface air temperature departures from the 1956-\n1985 means for winter and summer over the East,\nCentral and West regions of the contiguous United States.\nFigure 3:\nDepartures from the mean thickness of the 850-500 mb\nlevel versus the departures in the surface air temperature\nfor winter and summer for the East, Central and West regions\nof the contiguous United States.\nFigure 4:\nTime series of temperature departures separately for the\n850-500 mb level and the surface as well as the\ndifferences between the two series. Shown for the\nEastern Prairies, the Northern Plains and the East Slope\nNorth Cascades for the winter season.\n116","FIGURE 1\nA\nB\n117","y = - 69.697 + 3.5317e-2x R^2 = 0.193\n1985\ny = - 16.990 + 8.6041e-3x R^2 = 0.148\n1985\nEastern U.S. Summer\nEastern U.S. Winter\n1975\n1975\n1965\n1965\n1955\n1955\n1945\n1945\n3\n2\n1\n0\n-2\n-3\n-1\n1.5\n1.0\n0.5\n0.0\n-0,5\n-1.0\n1985\nR^2 = 0.001\n60\nR^2 = 0.057\n1985\nCentral U.S. Summer\nCentral U.S. Winter\n1975\n1975\nFIGURE 2\ny = 2.1060 - 1.0637e-3x\ny = - 40.264 + 2.0433e-2x\n1965\n1965\n1955\n1955\n1945\n1945\n1.5\n1.0\n0.5\n0.0\n-0.5\n-1.0\n3\n2\n-2\n1\n0\n-1\n-3\n1985\nR^2 = 0.064\n1985\nR^2 = 0.103\nWestern U.S. Summer\nWestern U.S. Winter\n1975\n1975\n- 29.600 + 1.5032e-2x\ny = 14.598 - 7.4073e-3x\n1965\n1965\n1955\n1955\n=\n1945\ny\n1945\n1.5\n1.0\n0.5\n0.0\n-0.5\n-1,0\n3\n2\n0\n1\n-2\n-3\n-1","4\n2\ny = - 3.7626e-2 + 9.4758x R^2 = 0.833\n+\ny = - 0.44244 + 13.338x R^2 = 0.750\nSurface Temperature Departure\nSurface Temperature Departure\nEastern U.S. Summer\nEastern U.S. Winter\n2\n1\n0\n0\n-2\n- 1\n-4\n- 6\n-2\n0\n60\n40\n20\n-20\n-40\n-60\n20\n10\n0\n-10\n-20\n2\n4\nSurface Temperature Departure\ny = 0.65705 + 10.850x R^2 = 0.745\nd\nCentral U.S. Summer\nSurface Temperature Departure\ny = 0.92204 + 8.2586x R^2 = 0.679\n8°\n8\n1\nFIGURE 3\nCentral U.S. Winter\n2\n0\n0\n- 2\n-1\n- 4\n-2\n20\n10\n0\n-10\n-20\n6\n60\n40\n20\n0\n-20\n-40\n-60\n2\n4\nSurface Temperature Departure\ny = 0.13307 + 13.277x R^2 = 0.793\nSurface Temperature Departure\ny = - 1.2405 + 12.709x R^2 = 0.759\nWestern U.S. Summer\nWestern U.S. Winter\n2\n1\n0\n0\n2\n-1\n- 4\n2\n-6\n20\n10\n0\n-10\n-20\n60\n40\n20\n0\n-20\n-40\n-60","1985\n1985\n= 66.797 3.3836e-2x R^2 = 0.123\ny - 87.078 + 4.4144e-2x R^2 = 0.186\n1985\nEastern Prairies Winter\n= 137.33 6.9597e-2x R^2 = 0.173\nEastern Prairies Winter\nEastern Prairies Winter\n1975\n1975\n1975\n1965\n1965\n1965\n1955\n1955\n1955\ny\n1945\ny\n1945\n1945\n5\n3\n1\n-1\n-3\n-5\n3\n2\n1\n0\n5\n-1\n-2\n-3\n3\n1\n-1\n-3\n-5\n1985\ny = 29.769 1.5066e-2x R^2 = 0.032\n1985\ny - 4.8684 + 2.4871e-3x R^2 = 0.000\n1985\nto\n= 28.156 1.4269e-2x R^2 = 0.007\nNorthern Plains Winter\nNorthern Plains Winter\nNorthern Plains Winter\n1975\n1975\n1975\n1965\n1965\n1965\n1955\n1955\n1955\n1945\ny\n1945\n1945\n6\n4\n2\n0\n-2\n-4\n-6\n6\n4\n2\n0\n-2\n-4\n-6\n4\n3\n2\n1\n0\n-2\n-1\n-3\nFIGURE 4\n- 36.270 + 1.8340e-2x R^2 = 0.011\n1985\ny = - 104.74 + 5.3113e-2x R^2 = 0.223\n1985\nEast Slope North Cascades\nEast Slope North Cascades\n= - 65.324 + 3.3180e-2x R^2 = 0.073\n1985\nEast Slope North Cascades\n1975\n1975\n1975\nWinter\nWinter\nWinter\n1965\n1965\n1965\n1955\n1955\n1955\ny\n1945\n1945\ny\n1945\n4\n2\n0\n-2\n-4\n-6\n4\n2\n0\n-2\n-4\n-6\n4\n3\n2\n1\n0\n-2\n-1\n-3","Descriptive Statistics of Surface Winds in the\nCore of the Equatorial Western Pacific\nJames Frederick and Pao-Shin Chu\nDepartment of Meteorology\nUniversity of Hawaii\nHonolulu, HI 96822\nInterest in the relationship between westerly wind bursts in\nthe equatorial Western Pacific and El Nino is growing rapidly.\nThis interest has focused on the fact that wind bursts in the\nequatorial waveguide may act as a precursor to El Niño phenomena\nthrough oceanic dynamic processes. Since wind bursts are\nintermittent synoptic-scale disturbances, lasting for a few days,\nit is necessary to quantify the intensity, frequency, and the range\nof variability of daily wind data from the historical file prior\nto the detailed analysis investigating the nature of bursts. With\nthis in mind, daily surface winds from ship reports averaged over\nthe region 5°N-5°S, 140°E-160°E are analyzed by means of an\nexploratory data analysis. Data sources include COADS and interim\nCOADS, for the period 1958 through 1987 (i.e., 30 years).\nFig. la displays a schematic plot of daily wind data during the\nWinter (December to February). For convenience, wind direction\nis\ndivided into the 8 standard compass groupings of 45° each. In\nwinter, northeasterlies and northerlies are most common, since they\ntend to occur 26% and 21% of the time, respectively. Also,\nnorthwesterlies account for 17% of the total number of observations\nfor the winter (n=2524). Median wind speed is greatest from the\nnortheast (3.4 and the most extreme wind speed is from the\nnorthwest (17 ms-1). This diagram simply reflects the predominance\nof the Northern Hemisphere winter monsoon flow associated with the\ngigantic heat sink over the Asian continent. The monsoon trough\nis located around 10°S, resulting in northeasterly winds along the\nship track north of the equator, northerlies at the equator, and\nnorthwesterly winds south of the equator. The frequency wind rose\nin Fig. 1b shows that north to northeast winds between 2 and 4 ms-Superscript(2)\nare most frequent (9%) during winter months.\nFor Spring (March to May), easterlies are most common since\nthey occur 29% of the time (Fig.: 2a). The median wind speed is\ngreatest (3.5 ms-1) from the east. Easterlies are slightly skewed\ntowards higher speeds. The northeasterlies show a secondary\nmaximum of occurrence (19%) The extreme value wind speed of 23\nms-1\nis from the north. Westerly winds are most variable, as the\ninterquartile range (i.e., distance within the box) is the largest.\nThe pattern of the frequency wind rose in Spring has shifted toward\nthe east (Fig.2b) Northeast and east winds between 2 and 6 ms\noccur most often (7%).\n121","For Summer (June to August) , southeasterlies prevail (30%,\nFig. 3a), . Median wind speed is greatest from the southeast at\n3.1 ms 1. . In early Summer (June), the Northern Hemisphere monsoon\ntrough develops. This trough then strengthens and extends farther\nnortheastward by August. As a result, southeast trades,\neasterlies, and southerlies are all most common at this time of\nyear. The frequency distributions of wind speed for these\nparticular directions are approximately normal or slightly skewed\ntoward lower speeds. In Fig. 3b, the pattern of maximum contours\ncontinues its clockwise movement around the wind rose as nearly all\nthe contours are located in the southern half of the rose.\nFor Fall (September to November) , easterlies\nand\nsoutheasterlies are most common (Fig. 4a). Median wind speeds\nare\n-1\nabout 2.9 ms for winds from the east, southeast, south, southwest,\nand west. The extreme value is from the north at more than\n20 ms1 . The westerlies increase their frequency of occurrence from\n7% in the Summer to 13% in the Fall. This is mainly due to the\npresence of the double troughs on both sides of the equator in\nNovember. The frequency wind rose shows that light southeasterlies\nand easterlies, between 2 and 4 ms ¹, are most common (9%, Fig. 4b) .\nA secondary maximum is found in the westerlies and southwesterlies\n1\nbetween 2 and 4 ms . Northerly winds are rare in the Fall.\n122","20\nWinter\n18\n(a)\n*\n16\n14\n12\n10\n9\n8\n7\n6\n5\n4\n3\n26%\n21%\n2\n4%\n1\n0\nNE\nN\nE\nDirection\n(b)\nNW\nNE\n7\n9\n5\nE\n3\n-0\n2\n4\n6\n8\nSW\nSE\nTO\nS\nWinter\nFig.\ndata\nduring\nWestern\n1987\nin\nnumber\neach\nin\nthe\npercentage\nthat\nin\n(e.g.\nin\nwinter.\nwind\nspeed.\nspeeds\n(b).\n(ms)\ndashed\nlie\nby\ncontoured\ncircles\nbetween\nand\nSW.\nat 1, 3, 5, etc.\n123","22\nSpring\n(a)\n18\n14\n10\n9\n8\n7\n6\n5\n4\n3\n29%\n19%\n2\n9%\n10%\n11%\n6%\n1\n0\nN\nNE\nE\nSE\nS\nSW\nW\nNW\nDirection\nN\n(b)\nNW\nNE\n3\n5\n7\n3\nw\nE\n0\n2\n6\n8\nSW\nSE\nTO\n$\nSpring\nFig. 2. Same as Fig. 1, except for spring.\n124","20\nSummer\n18\n(a)\n16\n*\n14\n12\n10\n9\n*\n8\n7\n6\n5\n4\n3\n30%\n23%\n2\n18%\n10%\n7%\n3%\n1\n0\nN\nNE\nE\nSE\nS\nSW\nW\nNW\nDirection\n(b)\nN\nNE\nNW\n3\nE\n2\n7.\n5\n-6\n3\n8\nSE\nSW\nto\nS\nSummer\nFig. 3. Same as Fig.: 1, except for Summer.\n125","20\nFall\n18\n(a)\n16\n14\n*\n12\n10\n9\n8\n7\n6\n5\n4\n3\n20%\n2\n13%\n21%\n13%\n13%\n7%\n7%\n1\n0\nN\nNE\nE\nSE\nS\nSW\nW\nNW\nDirection\nN\n(b)\nNE\nNW\n3\n3\nE\nW\n9\n5\n5\n2\n6\n8\nSW\nSE\nto\nS\nFall\nFig. . 4. Same as Fig. 1, except for Fall.\n126","IMPACT OF THE DIURNAL CYCLE UPON INTRASEASONAL\nFLUCTUATIONS IN THE TROPICS\nA. N. Hahmann, J. D. Horel, and J. Paegle\nDept. of Meteorology - University of Utah\nObservational studies of outgoing longwave radiation (OLR) indicate that low-frequency\nvariability in convection tends to be suppressed in the core regions of active convection.\nThis reduction in variance is most noticeable over the islands of the maritime continent\nand the core of the Amazon Basin during austral summer. We investigate this issue further\nin order to determine the dynamical processes which may inhibit low-frequency variability.\nThe variance of intraseasonal fluctuations in OLR is shown in Fig. 1 for the three major\nconvective zones during the austral summer season, December-February (DJF). The vari-\nance of pentad averages of OLR about the mean OLR during each austral summer season is\ncomputed on a 2°: X 2° grid for the 15 year period 1974-1988. Fig. la shows extremely small\nintraseasonal variability in the core of the Amazon Basin, where the convection is most in-\ntense during DJF. The value of the intraseasonal variance in the core of the Amazon Basin\nis 2 times smaller than that around its perimeter. A similar reduction in intraseasonal\nvariability is evident over Central Africa (Fig. 1b). Over the large islands of New Guinea\nand Borneo, the intraseasonal variance is smaller than that over the adjacent oceanic areas\n(Fig. 1c).\nThe large diurnal cycle in convection over equatorial land masses may help to explain\nthe reduced low-frequency variance in those regions. Since the OLR data from polar\norbiting satellites can not resolve the diurnal cycle, fractional cloud coverage (FCC) of\ndeep convection derived from GOES infrared imagery is used to resolve both diurnal and\nlow-frequency variations. This data set has been developed by P. Arkin and collaborators\nin the Climate Analysis Center. We focus here upon the FCC of cloud tops colder than\n235K for the period December 1981 through February 1989. The FCC is summarized into\n3-hour totals for two periods per month. Although the diurnal cycle in deep convection is\nwell resolved with this data set, only low-frequency variability with periods greater than\n15 days can be investigated. Fig. 2 shows an example of the diurnal and intraseasonal\nvariability of FCC during the DJF season of 1982-83 for a 2.5° square box located over\nthe region of maximum convection in the Amazon Basin (11.25°S, 61.25°W). The diurnal\nvariations in convection have an amplitude of roughly 25% during this season and clearly\n127","dominate the intraseasonal fluctuations, which have an amplitude of roughly 5% in this\ncase.\nFigures 3 and 4 contrast the relative importance of low-frequency and diurnal variations in\nconvection over the Amazon Basin. There is a distinct minimum in low-frequency variance\nin the core of the Amazon Basin on the basis of the FCC data (Fig. 3), as is also evident in\nFig. 1a on the basis of the OLR data. Large values of diurnal variance are found over most\nof the Amazon Basin (Fig. 4). However, there is no clear one-to-one relationship between\nenhanced diurnal variability and diminished low-frequency variability.\nIn order to address this issue further, we have performed some numerical experiments\nusing the shallow water version of a global model developed by J. Paegle. Three 200 day\nsimulations were computed keeping all factors fixed except for the type of forcing used to\ndrive the numerical experiments. The forcing functions used for the three simulations are\nas follows: (1) an eastward travelling wavenumber 1 source with a period of 20 days; (2)\na stationary regional diurnal source; and (3) the combination of (1) and (2). The third\nsimulation exhibits less low-frequency variance compared to the first simulation. Further\nanalysis of these experiments reveals that the reduction of low-frequency variance is due,\nin part, to the modification of the basic state flow by the diurnal forcing.\n128","200.\n201\n0\ne\n0820\nNOV\n200\n400\n200\n100\n400\n400\n600\n400\nC\n600\nYEX\nago\n600\n,1000\n8\n1000\n00 400\nare\n800\n12\n800\n200\n2003\n8\n8\n400\n20\nFigure 1. Variance of the intraseasonal variability of OLR for the DJF\nseason over the three major convective zones: a) South America, b)\nAfrica, and c) the Maritime Continent and Northern Australia region.\nSolid contours are plotted in intervals of 200W2m-4 and the dashed line\ndenotes 300W2m-4\n129","","Figure 3. Low-frequency (interannual plus intraseasonal) variance of\nfractional cloud coverage for the DJF season. The diurnal cycle has\nbeen eliminated by first averaging the 8 values for each half-month.\nUnits multiplied by 100.\n60\nFigure 4. Diurnal variance of fractional cloud coverage for the DJF\nseason. The low-frequency variability has been removed by computing\nthe diurnal variance from the climatological means. Units multiplied by\n100.\n131","INTRASEASONAL OSCILLATIONS IN THE UPPER TROPOSPHERE\nOBSERVED WITH THE CHRISTMAS ISLAND WIND PROFILER\nJ.R. McAfee, K.S. Gage, B.B. Balsley, W.L. Ecklund,\nD.A. Carter, and G. C. Reid\nAeronomy Laboratory\nNOAA/ERL\nBoulder, Colorado\nThe Christmas Island wind profiler is a 50-MHz Doppler radar which\nmeasures line-of-sight wind components along two orthogonal beams directed at\n15 degrees off zenith. Provided the vertical velocity component is negligible\ncompared to the horizontal, these can be converted to the two horizontal wind\ncomponents, specifically zonal and meridional. The winds are sampled every\nfew minutes, averaged to yield a consensus hourly wind, and telemetered via\nthe GOES satellite for incorporation onto the GTS four times a day in a\nstandard pilot balloon format. Additional information about the Christmas\nIsland wind profiler may be found in Gage et al (1988).\nFor purposes of climate analysis, the four PIBAL results are averaged to\nproduce a daily average wind profile and reported on a monthly basis in both\nthe Climate Diagnostics Bulletin and the Surface Marine Bulletin. The results\npresented here are all based upon these one-day averages, which are available\ncontinuously from February, 1987 to the present.\nDuring the latter part of 1988, it became evident that there was a\ndistinct enhancement of the winds, both in terms of the average values and the\nvariability, particularly when compared to the same time period a year\npreviously. This change persisted through the early part of 1989. An example\nof this behavior is shown in Figure 1 in terms of the time series for both the\nzonal and meridional components at 11 km (about 250 mb). The annual cycle is\napparent in the zonal component, as well as much stronger westerlies in the\n1988-89 winter when compared to the 1987-88 winter. Also apparent is a good\ndeal of variability throughout the period. In contrast, the meridional\ncomponent shows the expected lack of both an annual cycle and any long-term\nnon-zero behavior. The 1988-89 winter versus 1987-88 winter difference shows\nup as a very large intraseasonal variability in the former case compared to\nthe latter.\nFourier analysis of these time series shows a strong signal in both the\nzonal and meridional components near a period of 60 days. The same series,\nafter being filtered through a simple Butterworth filter centered at 60 days\nand having half-power points at 45 and 75 days, is shown in Figure 2.\nAlthough there is a strong signal in the zonal component throughout this time\nperiod, with some annual cycle and an increase in magnitude with time, the\nmost striking feature is the presence of an even larger signal in the meridio-\nnal component during the 1988-89 winter, especially when compared to the same\ntime a year previously.\n132","More details concerning the comparative behavior for the particular time\nperiods beginning about the first of November and ending near the middle of\nMay are shown in Figure 3 for the filtered time series of Figure 2. Figures\n3a and 3b show the relationships between the two winters, 1988-89, and\n1987-88, for the two components. The zonal signal is reasonably consistent\nbetween the two years, not only in terms of magnitude, but also in phase. In\ncontrast, there is virtually no meridional signal during 1987-88. Figures 3c\nand 3d compare the zonal and meridional signals for 1987-88 and 1988-89,\nrespectively. The lack of a meridional signal in 1987-88 is again apparent,\nbut the comparison for 1988-89 shows both the greater strength in the meri-\ndional component and the approximately 180° phase difference between the two\ncomponents, the latter being consistent with Madden-Julian type waves.\nThe behavior and consistency in terms of altitude are shown in Figure 4.\nEach of the curves represents the filtered time series at a given altitude for\neight successive altitudes equally spaced between 7.53 and 11.59 km and, from\nlowest to highest, solid, dash, dot, dash-dot, solid, dash, dot, dash-dot.\nWhile the zonal component shows some consistency with altitude, the meridional\ncomponent is clearly so. Quite evident are the growth in amplitude with\naltitude and the regular but small phase change with altitude. The latter\nsuggests a vertical wavelength much larger than the 4 km altitude interval\ninvestigated here.\nREFERENCE\nGage, K.S. , J.R. McAfee, W.G. Collins, D. Söderman, H. Böttger,\nA. Radford, and B.B. Balsley, A comparison of winds observed at\nChristmas Island using a wind-profiling Doppler radar with NMC and\nECMWF analysis, Bull. Amer. Meteorol. Soc. , 69, 1041-1046, 1988.\n133","40\n1987\n1988\n1989\n30\nzonal,11.01 km\n20\n10\n0\n-10\n-20\n0\n100\n200\n300\n400\n500\n600\n700\n800\n900\n1000\nday\n30\n1987\n1988\n1989\n20\n10\n0\n-10\n-20\nmeridional, 11.01 km\n-30\n0\n100\n200\n300\n400\n500\n600\n700\n800\n900\n1000\nday\nFigure 1. .\nTime series for 11 km one-day average zonal and\nmeridional winds at Christmas Island, Feb. 1987\nthrough Aug. 1989.\n134","6\n1987\n1988\n1989\n4\n2\n0\n-2\n-4\n60 day filter zonal,11.01 km\n-6\n0\n100\n200\n300\n400\n500\n600\n700\n800\n900\n1000\nday\n10\n1987\n1988\n1989\n8\n6\n4\n2\n0\n-2\n4\n-6\n60 day filter meridion 11,11.01 km\n-8\n0\n100\n200\n300\n400\n500\n600\n700\n800\n900\n1000\nday\nFigure 2. Time series of 60-day filtered winds from Figure 1.\n135","200\n200\ncompar-\n180\n180\nison between 1987-88 and 1988-89. (c), (d) - comparison between zonal and meridional.\n60 day filter,87-88,11.01 km\n60 day filter,88-89,11.01 km\n-\n160\n160\n(b)\n140\n140\n,\nFigure 3. Filtered time series from Figure 2 for early November through mid-May. (a)\n120\n120\n(c)\n(d)\n100\n100\n80\n80\n60\n60\nmeridional\nmeridional\nzonal\nzonal\n40\n40\n20\n20\n0\n0\n10\n8\n6\n4\n2\n0\n-2\n-4\n-6\n-8\n-10\n10\n8\n6\n4\n2\n0\n-2\n-4\n-6\n-8\n-10\n200\n200\n180\n180\n60 day filter meridional, 11.01 km\n60 day filter zonal, 11.01 km\n160\n160\n140\n140\n120\n120\n100\n100\n(a)\n(b)\n80\n80\n60\n60\n88-89\n87-88\n88-89\n87-88\n40\n40\n20\n20\n0\n0\n10\n8\n6\n4\n2\n0\n-2\n4\n-6\n-8\n-10\n10\n8\n6\n4\n2\n0\n-2\n-4\n-6\n-8\n-10","10\n1988\n1989\n8\n6\n4\n2\n0\n-2\n-4\n-6\n60 day filter zonal,7.53-11.59 km\n-8\n-10\n0\n20\n40\n60\n80\n100\n120\n140\n160\n180\n200\n10\n1988\n1989\n8\n6\n4\n2\n0\n-2\n-4\n-6\n-8\n60 day filter mendional,7.53-11.59 km\n-10\n160\n180\n200\n0\n20\n40\n60\n80\n100\n120\n140\nFigure 4. Filtered time series for early November 1988\nthrough mid-May 1989 at 8 successive levels\nbetween 7.53 and 11.59 km.\n137","On the Annual Variation of Subtropical Jet Streams\nby\nTsing-Chang Chen and Ming-Cheng Yen\nDepartment of Geological and Atmospheric Sciences\nIowa State University, Ames, IA 50011\nIt is hypothesized that the annual variation of subtropical jet streams\nis induced by the adjustment process through the chain relationship between\nthe annual variations of tropical diabatic heating and planetary-scale\ndivergent and rotational circulations. To prove this hypothesis, a diagnostic\nscheme illustrating this chain relation is proposed. The 5-year (1978-1979,\n1984-1987) upper-air data generated by the Global Data Assimilation System\n(GDAS) of the National Meteorological Center (NMC) are used in the diagnostic\nanalysis.\n1. Diagnostic Scheme and Data\nApplying the inverse of Laplacian to the resultant equation by combining\nthe continuity equation and the vertical differentiation of the thermodynamic\nequation, we obtain\n(1)\n-\nXQ\nThe relationship between atmospheric divergent circulation, which is\ngenerally delineated with (VD,X), and diabatic heating is established by (1).\nThe notation used in (1) is conventional: X is velocity potential; Q,\n-T 20\ndiabatic heating; o = may be evaluated by the residual\nmethod\n(2)\nThe vertical differential heating may be inferred from\n(3)\nAtmospheric circulation is portrayed largely by rotational flow, which\ncan be well depicted by streamfunction The effect of divergent circulation\non rotational circulation can be revealed through the streamfunction tendency\ninduced by the vorticity source, Hx' shown in the following streamfunction\nbudget equation:\n+ f)] - -2 {V.(s + f)Vx|} + V-2 F.\n(4)\n4t\nYF\nAV\nX\nF includes vertical advection, twisting, and friction terms in the vorticity\nequation. The chain relationship between tropical diabatic heating, which\nmay be inferred from HVD' and divergent and rotational circulation can be\nobtained through the X maintenance equation, (1), and streamfunction (i)\n138","budget equation, (4).\nLet ) denote the annual variation component. Using (1), (3) and (4),\nwe propose a diagnostic scheme to illustrate the chain relationship between Q,\n~\nX, and 4, as follows:\n~\n(5)\nXQ = X X XHA' =\nTHAV\n&\n+\nX\nNote that - (5+f)]} where () = annual mean. This\napproximation was shown to be legitimate in our diagnostic analysis. It is\nclear from (5) that maintenance of the annual variation of\nplanetary-scale divergent circulation by diabatic heating can be realized\nit\nthrough XQ. The annual variation of rotational circulation represented by\nmay be caused by an imbalance between streamfunction tendencies induced by\nvorticity source (IX) and vorticity advection (TAV). The relationship\n4x\nbetween X and + is connected by V. D through\nThe upper-air daily data generated by the NMC GDAS for the period of\n1978-1979 and 1984-1987 were used in this study. The horizontal resolution\nof this data set is 2.5 x2.5 o The daily X, 4, & AV' and 4x fields are\nobtained over the entire globe by solving Poisson equation by the spectral\nmethod, without specifying boundary conditions at both poles. A regular\nempirical orthogonal function (EOF) analysis is applied to the monthly\naverages of the X, ' and 4x fields in order to extract their annual\nvariation components.\n2. Results\na. X Maintenance\nThe variances and fractional variances displayed in Table 1 show that\nthe annual variations of X, XQ, and are well represented by their first\nEOF eigenmodes. The maximum (minimum) annual variations of X and XO occur in\nmiddle northern summer (winter), but that of appears in northern spring\n(fall). The amplitudes of X and XQ are comparable to each other, while that\n-\nof is much smaller. Moreover, eigenvectors of the first X and xQ\n139","eigenmodes (not shown) well resemble each other in terms of spatial structure.\nThese features of the X, XQ, and XHA annual variation components suggest that\nthe annual variation of divergent circulation is supported primarily by that\nof diabatic heating. To substantiate this argument further, we construct\nseasonal mean charts of (VD,X) and\nBoth the northern winter- and summer-mean (VD,X) and H VD are exhibited\n-\nin Fig. 1. Note that the northern winter- and summer-mean charts are opposite\nin their spatial structure. The positive (negative) HVD centers are located\n-\nover three tropical continents of the summer (winter) hemisphere. These H\nVD\ncenters are consistent with those of cumulus convection observed by previous\nstudies. This consistency strongly indicates that the latent heat released by\ncumulus convection may be a major contributor to the annual variation\ncomponent of diabatic heating over the three tropical continents of the summer\nhemisphere.\nThe annual variation component of divergent circulation (VD,X) exhibits\na pronounced interhemispheric flow from the summer to winter hemisphere, with\nenhancement over the three tropical continents. Therefore, a pair of (VD,X)\ndivergent-convergent centers exists over each tropical continent. Of course,\nthe divergent (convergent) center is situated in the summer (winter)\nhemisphere. The contrast between (VD, X) and HVD VD clearly shows that the\n(VD,X) divergent (convergent) centers coincide with the positive (negative)\n~\nHVD centers. The similarity of X and Xn characteristics and the coincidence\nof (VD,X) and VD centers indicate that the annual variation of divergent\ncirculation (VD,X) is maintained primarily by that of tropical diabatic\n-\nheating, as inferred from HVD\nb. & budget\nThe variances and fractional variances of the first eigenmodes of t, 4\nLAV' AV' and TX demonstrate that the annual variation components of these fields\n~\nare well depicted by their first eigenmodes. The maximum phases of 4, +\nAV'\nTx appear in middle northern winter, when the northern subtropical jet\nand\nstreams reach their maximum intensity. On the other hand, I obtains its\n-\nmaximum phase during northern fall, when northern subtropical jet streams gain\ntheir maximum increase in wind speed.\n140","Moreover, the annual variation of atmospheric circulation is much more\npronounced in the Northern Hemisphere than in the Southern Hemisphere, and\nthus we will focus on the former hemisphere. To avoid redundancy, we will\ndiscuss only the case of northern winter. Shown in Fig. 2 is the\nnorthern-winter mean budget. The northern winter-mean VR(=kxVy), DAV' and\nTX\nconstructed with the first eigenmodes (Table 2) The VR direction\nin\nnorthern winter is opposite to that in northern summer (not shown). As\nindicated by VR' the annual variation of rotational flow is largely associated\nwith those of the northern subtropical jet streams, particularly with the East\nAsian jet. The significant It is associated with subtropical jet streams, as\nindicated by the kxVy vectors. The spatial structures and magnitudes of\nX\nand I AV\nare similar, but opposite in sign. These salient features of the\nbudget are consistent with Eq. 5, which suggests that the seasonal march of\nby which the annual variations of subtropical jet streams are determined, is\nattributable to it which is caused by the imbalance between & and & AV\nWe\nX\nsuperimpose Vy vectors on 4 in order to indicate both vorticity source\nX\n(divergence) and sink (convergence) regions. It is obvious that negative\n(positive) streamfunction tendencies are induced over vorticity source (sink)\nregions, i.e., upstream (downstream) sides of jet streams. These\nstreamfunction tendencies are counterbalanced by those induced by vorticity\nadvection.\nUp to this point, it is not yet clear how the annual variation of\ndivergent circulation VD influences that of jet streams (i).\nNote that can be approximated by As shown by our\nanalysis, 4x is more or less explained by TXI (not shown).\nIn other words,\n4x is decisively determined by the X field. Of course, the imbalance\nthe\nbetween AV and 4x' i.e., It's should be essentially decided by X. If we now\nbring together the analyses of the x-maintenance and & -budget equations, the\nannual variation of subtropical jet streams, which is caused by the chain\nprocess between Q, X, and &, becomes obvious.\nAcknowledgements\nThis study was supported by NSF grant ATM-8912451. We\nappreciate the help provided by Mrs. Reatha Diedrichs and by Ms. Anne Richards\nfor the preparation of this extended abstract. The computations of this study\n141","were performed on the XMP computer of the National Center for Atmospheric\nResearch, which is sponsored by the National Science Foundation.\nTable 1\nVariance (VE) explained by the first eigenmodes of the X, XQ' fields and\nfractional variance (FV) explained by the annual (12-month) harmonic of\neigencoefficient time series of these first eigenmodes. Both VE and FE are\nexpressed in terms of percentage (%). The maximum dates of the annual\nharmonic of eigencoefficient time series are denoted by (month/day), , i.e. ,\nthe parenthesized number following FV.\nAmplitude (m 2 S -1)\nVariable\nEigenmode\nVE(5)\nFV(%) (phase)\n1.27x107\nX\n1\n62.5\n83.2 (7/28)\n1.41x107\nXQ\n1\n67.9\n86.9 (7/9)\n6\n1\n46.0\n85.9 (4/22)\n5.98x10\nTable 2\nThe same as Table 1, except for the T, It's AV' and TX fields.\nVariable\nEigenmode\nVE(%)\nFV(%) (Phase)\nAmplitude\n4\n1.24x10 8 m 2 S -1\n1\n87.5\n97.2 (2/2)\nit\n17 m2s-2\n1\n40.8\n51.3 (11/24)\nt\ni\n2-2\n1\n68.1\n87.7 (1/8)\n462\nS\nAV\n&\n2-s-2\n1\n52.6\n78.9 (1/7)\n466\nm\nX\n142","1\n4 0ms\n-\nWINTER\n1979 + 84 + 85+86\nX and H\nVD\n45N\n15N\n15S\n45S\n90W\n301\n30W\n30E\n90E\n150E\n150W\n4.0ms\nSUMMER\n1979+84+85+86+ 87\nand H\nX\nVD\nURD\n45N\n15N\n15S\n45S\n150W\n90W\n301\n30W\n30E\n90E\n150E\nNorthern winter (December-February) and summer (June-August) mean\nFig. 1\nannual variations of divergent circulation, (VD,X), superimposed\nwith the annual variation of vertical differential heating, HVD\nThe positive (negative) HVD values are heavily (lightly) shaded.\nThe contour interval of X is 10 6 m 2 S -1\n.\n143","40ms\nVR\nWINTER\n1979 - 84 + 85 +86+87\n45N\n15N\n15S\nDI\n45S\n30W\n30E\n90E\n150E\n150W\n90W\n301\n6x10~ms-2\nVRt\nFALL\n1979+84+85+86+87\n45N\n15N\n15S\n101\n45S\n30W\n30E\n90E\n150E\n150W\n90W\n301\nWINTER\n1979 + 84 + 85 + 86 + 87\nAV\n45N\n&\nED\n320\n15N\nQ\n15S\n45S\n30W\n30E\n90E\n150E\n150W\n90W\n301\n1.5x10-4ms\nWINTER\n1979 + 84 + 85 + 86+87\nX\n45N\n15N\n15S\n45S\n30W\n30E\n90E\n150E\n150W\n90W\n301\nNorthern winter mean VR(=kxVy), DAV, and IX' and the northern\nFig. 2\n^\nfall\n(=kxV++)\nV\nIsotaches\nVR1\nmean\nof\nand\nVRt\nare included.\n=\nRt\nV\n-2\nI\n2x10\nare shaded. The contour intervals of\nAV\nms\nand\nRt\nI\n2 -2\nare 40 m S\nX\n144","POSSIBLE FACTORS CONTROLLING GLOBAL NIGHT MARINE AIR\nTEMPERATURE VARIATIONS OVER THE PAST CENTURY\nReginald E. Newell, Wu Zhongxiang and Jane Hsiung*\nDepartment of Earth, Atmospheric, and Planetary Sciences\n54-1522, M.I.T., Cambridge, Mass., 02139\nand Nicholas Newell, Somerville, Mass.\nFolland and Parker discussed time series of marine sea and air temperatures from\n1856 onwards at last year's workshop (pp. 160-165). The data set and correction\nprocedures are fully described in the Global Ocean Surface Temperature Atlas (GOSTA) by\nBottomley et al. (1989, in press). The present discussion draws on the night marine air\ntemperature records (henceforth abbreviated to MAT) in GOSTA for the Globe, the\nNorthern and Southern Hemisphere (NH and SH) and the tropical west and east Pacific\n(TWP and TEP), annual values of which are illustrated in Figure 1. The first point to note\nis that trend lines fitted by linear regression to these observations show a significant\nincreasing trend if only the last century is considered but the significance disappears for the\nNH when the record is extended back to 1856. There are at least three possible factors\ninvolved in the observed fluctuations: internal mechanisms, probably involving air-sea\ninteractions, which are characterized here by the Southern Oscillation Index (SOI); solar\nirradiance variations, a proxy for which is the sunspot number; and atmospheric\ntransmission for solar radiation for which we use the atmospheric turbidity calculated from\nCampbell-Stokes sunshine recorder data from Sonnblick, Austria (47°N, 13°E, 3106 m),\na high altitude station selected to minimize possible urban contamination. These three\nfactors (see Figure 2) form the independent variables in a multiple regression approach\ndesigned to account for the variability shown in Figure 1; the regression was only carried\nout for the past century due to the limit on turbidity data. The first factor corresponding to\neach MAT series is shown in Figure 3; it is turbidity except for TEP where it is SOI. The\nsecond factor, always a small contribution, is generally solar variability. For global MAT\nturbidity accounts for 50% of the variance, but solar only 5%. For TEP 47% of the\nvariance is accounted for by SOI, with turbidity second at 16%. When the contributions by\nthe factors which pass a significance test are removed from the original time series the\nresiduals show no significant trends (Figure 4).\nThree-month seasonal anomalies of global and SH MAT have been examined by\nharmonic analysis and power spectrum techniques (e.g. see Newell et al. GRL, 16, 311-\n314, 1989) and some of the results are summarized in Figures 5-12. Figures 5 and 9 are\nthe basic seasonal time series normalized to zero with a 41-season Gaussian smoothing\nsuperimposed which is designed to decrease the amplitude of periods shorter than 10 years\nby more than 50% (see \"Climatic Change\" WMO Technical Note No. 79.). For global\nMAT the sum of the first three harmonics is shown in Figure 6 which is subtracted from\nFigure 5 to yield Figure 8 and this in turn is also smoothed with the Gaussian filter. The\nlatter reveals a 22-year periodicity which also shows in the power spectrum analysis of\nFigure 7. A sum of the three harmonics near to 22 years shows that this variation is\npresent in both hemispheres with a close phase relationship (see Figure 10). This 22 year\nperiod is near to that of the solar magnetic cycle which is also known as the double sunspot\ncycle though this may be coincidence. A phase comparison between the double sunspot\ncycle and the near 22 year temperature variation is made in Figure 11 by superimposing the\nthree harmonics near 22 years from global MAT on to the sunspot record, which has been\ntreated with a frequency domain band-pass filter that has a response of unity at 11 years\ndropping to zero at 5.5 and 22 years (also from WMOTN No. 79). There is good\nNow at BBN Communications Corp., Cambridge, Mass.\n145","correspondence with the temperature record lagging that of sunspots. For the sake of\nargument here let us assume that a relationship exists. A sinusoidal solar forcing at the\nsurface would be expected to lead the surface temperature response by 1/8 of a cycle or\n2 3/4 years according to a simple thermal diffusivity model (e.g. see Sellers, Physical\nClimatology, 127-140, 1965). Moving the sunspot record forward by 2 3/4 years\nsubstantially improves the phase agreement of Figure 11 as seen from Figure 12. It is of\ninterest that Lean and Foukal have pointed out that the sun is brighter when more sunspots\noccur because the additional brightness due to the concomitant enchancement of faculae\noutweighs the darkening due to the spots and that from the recent records the brightest\ntimes would be expected to be 1958 and 1981 separated by twenty three years (Science,\n240, 906-908, 1988). Clearly future measurements of the solar irradiance over a full 22\nyear cycle and longer will be necessary to test these arguments.\nConclusions\n1) 50% of the variance of global MAT variations over the past century can be accounted for\nby atmospheric turbidity changes, which are thought to be controlled by volcanic\nactivity. The dip in temperature in the 1900-1910 period is associated with at least nine\nmajor eruptions having a volcanic explosivity index of 4 or more. 47% of the variance\nof TEP MAT is related to SOI changes, corresponding to internal adjustments within the\nocean-atmosphere climate system.\n2) Not more than two of three selected factors contribute significantly to the explanation of\nvariance of the five time series. After contributions from the two significant factors are\nsubtracted, trends in the residuals of each of the five selected time series are not\nsignificant.\n3) A twenty two year variation appears in the MAT data - is this evidence of a solar\nrelationship?\nTo explore item 1 further we appealed at the workshop for further information\nabout sunshine records prior to 1910. One participant located some early Japanese records\nin the Scripps Library and another two have provided further leads all of which are being\ninvestigated.\nTo explore item 3 further continuous irradiance measurements and much patience\nare required.\nAcknowledgement. Reginald E. Newell and Wu Zhongxiang received support from the\nNational Science Foundation Climate Dynamics Program (Grant No. ATM 8904957)\nduring preparation of this paper.\n146","Figure 2. Time series of independent variables and Lamb's dust veil\n1980\n1970\n1960\n1950\nDUST VEIL INDEX (LAMB)\nSONNBLICK TURBIDITY\nSUNSPOT NUMBER\n1940\nSOI_WRIGHT\nYEAR\n1930\n1920\n1910\n1900\nindex.\n1890\n2\n-2\n200\n100\n0\n2\n200\n100\n0\n1\n0\n-1\n7\n6\n5\n4\n3\nJAMA\n1980\nTREND(1887-1986)\nTREND(1856-1986)\n1970\nRAW DATA\nand their trends(dash line). Vertical lines are error bars of the\nFigure 1. Night marine air temperature time series(solid line)\n1960\n1950\nMAT TIME SERIES AND THEIR TRENDS\ntrends, corresponding to two standard deviations.\n1940\nTROPICAL EAST PACIFIC\nTROPICAL WEST PACIFIC\nSOUTHERN HEMISPHERE\nNORTHERN HEMISPHERE\n1930\nYEAR\nGLOBE\n1920\n1910\n1900\n1890\nH\n1880\n1870\n1860\n0.5\n0.0\n-0.5\n-1.0\n-0.5\n-1.0\n1.0\n-0.5\n-1.0\n1.0\n0.5\n0.0\n-0.5\n-1.0\n1.0\n0.5\n0.0\n0.0\n-0.5\n-1.0\n0.5\n0.0\n1.0\n1.0\n0.5","lines are the residuals of the regression. Dash lines are the final trends\nFigure 4. Night marine air temperature trends after regression. Solid\nafter regression. Vertical lines are the error bars of the trends, which\n1980\nRESIDUAL\nTREND\n1970\nMAT TRENDS AFTER REGRESSION(1887-1986)\n1960\n1950\nNORTHERN HEMISPHERE\nSOUTHERN HEMISPHERE\nTROPICAL WEST PACIFIC\nTROPICAL EAST PACIFIC\n1940\ncorrespond to two standard deviations.\nGLOBE\nYEAR\n1930\n1920\n1910\n1900\n1890\n1.0\n0.5\n0.0\n-0.5\n-1.0\n1.0\n0.5\n0.0\n-0.5\n-1.0\n1.0\n0.5\n0.0\n-0.5\n-1.0\n1.0\n0,5\n0.0\n-0.5\n-1.0\n1.0\n0.5\n0.0\n-0,5\n-1.0\n1980\nFigure 3. First factor of the regression. Solid lines are raw data, dash\nRECON. DATA\ncoefficient. TUR and SOI mean first factor is turbidity or SOI. The\nnumber in parentheses is the lag in years between the independent\nRAW DATA\nlines are reconstructed data using first factor. R is the correlation\n1970\n1960\nMAT - FIRST FACTOR (1887-1986)\n1950\nNORTHERN HEMISPHERE\nSOUTHERN HEMISPHERE\nTROPICAL WEST PACIFIC\nTROPICAL EAST PACIFIC\n1940\nGLOBE\nYEAR\n1930\n1920\nvariable and the MAT.\n1910\nR(4)_TUR. 0.669\nR(2)_TUR. 0.669\nR(4)_TUR. 0.712\nR(3)_TUR. 0.587\nR(0) SOI 0.682\n1900\n1890\n1.0\n0.5\n0.0\n-0.5\n-1.0\n1.0\n0.5\n0.0\n-0.5\n-1.0\n1.0\n0.5\n0.0\n-0.5\n-1.0\n1.0\n0.5\n0.0\n-0.5\n-1.0\n1.0\n0.5\n0.0\n-0,5\n-1.0","GLOBAL MAT, PERIODS > 27.8 YEARS\nGLOBAL MAT\n70\n70\n15\n60\n60\n15\n50\n50\n10\n10\n40\n40\n5\n5\n30\n30\n0\n20\n0\n20\n10\n-5\n10\n-5\n0\n0\n10\n-10\n10\n-10\n15\n20\n-15\n-20\n30\n20\n-30\n-20\n40\n-40\n25\n-25\n50\n-50\n30\n-30\n60\n-60\nYEAR\nYEAR\nFigure 6. Harmonics with periods greater than\nFigure 5. Global night marine air temperature\n27.8 years in global NMAT. Note the major dip\nanomalies for the period 1876-1986, with\ncentered around 1910.\nGaussian smoothing. The mean is set to zero.\nGLOBAL MAT, PERIODS < = 27.8 YRS\nGLOBAL MAT\n60\n60\n.50\n50\n50\n.45\n40\n40\n.40\n30\n30\n.35\n20\n20\n.30\n10\n10\n.25\n0\n0\n.20\n10\n-10\n.15\n20\n-20\n.10\n30\n-30\n.05\n40\n-40\n.00\n2\nYEAR\nPERIOD (YRS)\nFigure 7. Discrete power spectrum of global NMAT\nFigure 8. Harmonics with periods less than or equal\nto 27.8 years in Global NMAT, with Gaussian smoothing.\nwith mean set to zero. Superimposed are the 50\n95% confidence level of Markov \"red noise\" null\nspectrum derived based on the assumption that the\nfirst harmonic is a significant variation in the\nrecord and not just a manifestation of Markov-type\npersistence. 22.2-year harmonic falls above the 95%\nlevel of this null spectrum.\n149","S. HEMISPHERE MAT\nWORLD(solid), _HEM(++), S_HEM(--)\n80\n80\n20\n60\n20\n60\n15\n15\n40\n40\n10\n10\n5\n20\n20\n5\n0\n0\n0\n0\n-5\n-5\n-20\n-10\n20\n-10\n-15\n15\n-40\n40\n-20\n20\n-60\n60\n-25\n25\nYEAR\nYEAR\nFigure 10. A comparison of the periods in the range\nFigure 9. Southern hemisphere NMAT anomalies for the\n18.5 to 27.75 years(3 harmonics in the neighbourhood\nperiod 1876-1986, with Gaussian smoothing. The mean\nof 22 years) between the global, northern hemisphere,\nis set to zero. The 22 year variation can be detected\nand southern hemisphere NMAT records. The 22-year\nhere before filtering.\nvariation seems to be present in both hemispheres.\nFILT GLOB MAT, FILT SUNSPOTS\nFILT GLOB MAT, LAGGED FILT SPOTS\n20\n20\n20\n20\n15\n15\n15\n15\n10\n10\n10\n10\n5\n5\n5\n5\n0\n0\n0\n0\n-5\n-5\n-5\n-5\n-10\n-10\n10\n10\n-15\n-15\n15\n-15\n-20\n-20\n20\n-20\n-25\n-25\n25\n25\nYEAR\nYEAR\nFigure 11. A phase comparison between the double\nFigure 12. Same as figure 11, but the sunspot record\nsunspot cycle and the 22-year variation of global\nhas been shifted forward by 2.75 years, the lag\nNMAT. Periods between 18.5 and 27.75 years\nbetween a 22 year surface flux variation and\n(3 harmonics) in global NMAT are shown with\nthe its response in surface temperature that is\nthe sunspot record bandpass filtered to 11\npredicted by a simple thermal diffusivity model of\nyears. Note how alternating maxima and minima\noceanic vertical heat flux. Note the good phase\nin the temperature record lag successive peaks\nagreement that results from this shift.\nin the sunspot cycle.\n150","Atlantic Sea Level Pressure Teleconnections and\nAssociated Storm Track and Climate Variability\nJeffery C. Rogers\nDepartment of Geography, The Ohio State University\nColumbus, Ohio 43210-1361\nRotated principal component analysis (RPCA) is performed on monthly 11 X\n13 latitude/longitude grids of the sea level pressures (SLP), extending from\n1899-1986 and spatially from 20°: N to 70°N and 160 eastward to 40°E (excludes\nAsia). Generally, 8-11 unrotated components are retained per month for\nvarimax rotation, and 7 teleconnections are identified which are highly\nrecurrent on a month-to-month basis throughout the year. This paper focuses\nonly on the 4 January teleconnections (representative of the cold season\nmonths) identified in the Atlantic-European sector. These patterns (see Fig.\n1) include the North Atlantic Oscillation (NAO), the east Atlantic pattern\n(EATL), a pattern with centers over southern Europe and the northern Atlantic\n(SENA) and a pattern with a Scandanavian center (SCAN) Fig. 1 shows the\nteleconnection pressure differences (mb) obtained by subtracting the mean SLP\nin years with extreme (over one standard deviation) negative polarity from\nthat during positive polarity years. These differences are statistically\ntestable using a two-tailed t-test.\nFig. 2 shows the mean pressure fields in the polarity extremes used in\nmaking Fig. 1. The NAO is associated with large pressure differences in the\narea around the mean Icelandic Low. EATL is associated with latitudinal\nmovements of the Azores High in the eastern Atlantic around Great Britain.\nSENA positive mode resembles the NAO positive mode except that the Azores High\nlies over southern Europe, and SCAN is associated with longitudinal movements\nof the Icelandic Low and Siberian High.\nWave cyclone frequencies are obtained, using tracks published in Mariners\nWeather Log, for sets of nine winter months (DJF) with the highest positive\nand lowest negative teleconnection scores since 1957. The raw cyclone\nfrequencies, summed for 5° latitude by 5° longitude tessarae, are not adjusted\nfor latitude. During positive NAO months (Fig. 3a) the main storm track has a\nsharp northeastward orientation, while in negative mode (Fig. 3b) storms track\nalong latitude 45°N. A similar pattern to Fig. 3b occurs in positive polarity\ncases for EATL (Fig. 3c) except that the storm track has a more northeastward\ntrajectory and it crosses the east coast of the U.S. slightly south of its\nlocation in Fig. 3b. Note that both Figs. 3a and 3c are associated with a\ndeep Icelandic Low (Figs . 2a and 2c). SENA negative polarity cases (Fig. 3f)\nshow a secondary cyclone maxima over the northern Mediterranean Sea, in\naccordance with the mean trough of low pressure noted over this region in Fig.\n2f. The mean storm track extends farther east in the SCAN negative polarity\ncases (Fig. 3g), when the mean low extends northeastward toward the Barents\nSea (Fig. 2h).\nThe storm track characteristics in NAO and EATL are linked to variations\nin the mean zonal components of the 300 mb geostrophic wind (Figs. 4a-4d). In\nthe NAO, the axis of maximum zonal 300 mb winds exhibits latitudinal shifts\nbetween the polarity extremes (Figs. 2a and 2b) with an extensive\nnortheastward orientation in positive polarity. In the NAO, the mean\nIcelandic Low (Figs . 2a and 2b) is located 10 - 15° of latitude to the\n151","northeast of the 40 ms -1 mean zonal wind isotach, an area which lies to the\nnortheast of the mean jet exit region. In the EATL modes, the storm tracks\n(Figs. 3c, 3d) closely correspond to the orientation of the mean zonal wind\nisotachs (Figs. 4c-4d), especially the 20 ms 1 isotach and the mean Low (Figs.\n2c,2d) is displaced from the axis of fastest winds by over 15° of latitude.\nFig. 5 shows the areas where the mean 300 mb zonal wind differences,\nobtained by subtracting the mean zonal velocities in negative polarity cases\nfrom those in positive polarity cases (Fig. 4), are statistically significant\nwith at least 99.8% (99% in the SENA cases) confidence. The polarities shown\nfor each area indicate the teleconnection mode in which anomalously strong\nwesterlies occur, while areas with opposite teleconnection polarity experience\nanomalously weak westerlies. Use of a high statistical confidence interval\nrestricts the sizes of the areas and none of them overlap. The NAO (see also\nFigs. 4a and 4b) is associated with significant mean zonal wind variations\nover the subtropical and mid-latitude Atlantic as well as around Greenland.\nThe largest and most significant 300 mb zonal wind differences between the\nEATL polarity extremes (Fig. 5) occur south of the area of largest January\npressure variability (Fig. 1). Positive (negative) SENA polarity is\ncharacterized by a strong (weak) polar front jet over Scandanavia, especially\nalong 62. ON latitude, and an anomalous southward (northward) displacement of\nthe mean subtropical jet over northern Africa and the Mediterranean. Positive\n(negative) SCAN polarity is characterized by a displacement northward\n(southward) of the polar front jet into the Atlantic arctic (central Europe).\nWinter averaged (DJF) scores for the 4 teleconnections are correlated to\nwinter mean air temperatures at high latitude and European stations. Fig. 6\nshows the spatial distribution of significant correlations between these\nvariables. Note the similar locations of significant correlations in the NAO\nand SENA patterns (Figs. 6a, 6c). An opposition in the sign of the\ncorrelations exists in these modes between Europe, and eastern Canada,\nGreenland and the eastern Mediterranean. EATL (Fig. 6b) is associated with\nscattered significant correlations over southern Europe and parts of Norway\nand Sweden while the same occurs in SCAN (Fig. 6d) over north-central Europe\nand Iceland. Correlations between the scores and winter precipitation totals\nare low in the NAO and SENA, but higher ones occur over Great Britain in\nassociation with the EATL pattern, and in Scandanavia with the SCAN pattern.\n152","JAN\nJAN\no\nA\nA\nEATL\nLINAO\nJAN\nJAN\n+8\nSENA\nSCAN A\nA\n20\nThe 4 Atlantic-European sector January RPCA teleconnection\nFigure 1. e\npatterns, represented by SLP differences (mb) obtained\nbetween the polarity extremes in the eigenvector scores.\nNAO\nNAO\nH\nM6\n08\n16\n08\n24\n16\nA\nEATL\nEATL to\n08\n08\n16\n24\n16\n23\nD\nC\nMean pressure (mb) in the positive (left) and negative\nFigure 2.\n(right) polarity extremes of each of the 4 teleconnections .\n153","SENA +\nSENA -\n0\n00\n12\n16\n24\nSCAN +\nSCAN -\n24\nLed\n100\n00\n08\n08\n16\n16\n24\nFigure 2 continued.\nNAO\nNAO\n10\n40\n40\n30\n20\n20\n10\n0\nEATL\nEATL\n20\n30\n40\n20\n20\n10\n10\n0\nSENA\nSENA\n30\n30\n20\n10\n10\n0\nSCAN\nSCAN\n20\n40\n30\n40\n30\nG\nH\n0\nFigure 3. e\nWave cyclone frequencies for 9 months with positive and\nwith negative extremes in each of the 4 teleconnections.\n154","NAO -\nNAO +\n2\n20\n30\n40\n20\n30\nA\n10\n20\nEATL +\nEATL -\n20\n20\n10\nD\nC\nD\n20\nThe mean 300 mb zonal component of the geostrophic wind\nFigure 4. .\n(u s for the extreme positive and negative modes of NAO\nand EATL in Januaries.\nAreas where the 300 mb zonal winds (Fig. 4) are statistically\nFigure 5. .\nsignificant with 99.8% confidence (99% for SENA) . The reference\nto polarity indicates the mode with anomalously strong westerlies.\n300 mb Ug\nJAN\nEATL\nNAO +\nEATL +\nNAO -\n155","Low-frequency variability of subtropical jets\nChung-Kyu Park\nCentel Federal Services Corporation\nNASA/Goddard Space Flight Center\nGreenbelt, MD 20771\nSiegfried Schubert\nGoddard Laboratory for Atmospheres\nNASA/Goddard Space Flight Center\nGreenbelt, MD 20771\n1. Introduction\nA number of studies have examined possible links between tropical forcing and\nzonal wind fluctuations (e.g., Knutson and Weickmann, 1987; Paegle and Mo, 1988; Lau,\nKang, and Sheu, 1989). On the other hand, the work of Held (1983), and Nigam and\nLindzen (1989) suggests a potentially strong sensitivity of quasi-stationary waves to\nzonal wind fluctuations via changing wave propagation characteristics. This study\nexamines observational evidence of a link between the tropical forcing, low-\nfrequency zonal wind fluctuation, and quasi-stationary waves using ECMWF global\nanalyses from 1981 to 1987 for the Northern Hemisphere winter (DJF). Here,\n\"low-\nfrequency\" refers to time scales of 20 to 70 days. First, we describe the dominant zonal\nwind fluctuations by an empirical orthogonal function (EOF) analysis. The first two\nEOFs provide the baseline modes of zonal wind fluctuation. Next, we examine a possible\nlink between the tropical forcing and low-frequency zonal wind fluctuation by\ncompositing outgoing longwave radiation (OLR) for various time lags with regard to\nthe u-wind fluctuation. Finally, a link between the quasi-stationary waves and low-\nfrequency zonal wind fluctuation is examined in terms of the differing wave activity\nflux (Plumb, 1985) found during the extreme phases of the zonal wind fluctuations.\n2. Results\nThe first two EOF patterns of the filtered 200mb zonal wind in the Pacific sector\nand their coefficients are shown in Figure 1. The first EOF pattern is characterized by\na strong tropical signal in the western Pacific and a wave propagation feature in the\neastern Pacific which is associated with an emanating branch from the western\ntropical Pacific. The second EOF pattern shows the extratropical signals in the\ncentral North Pacific and the Australian region. The coefficients for the first EOF\nindicate a strong tropical oscillation during 1982/83 winter. Note that the time-series\nare only for DJF of ieach year. In the following these coefficients are used as an\nindex for the composites. In the compositing process, the high index refers to when\nthe coefficients fall one standard deviation (STD) above the mean, while the low index\nrefers to when the coefficients fall one STD below the mean.\nThe composite patterns of the 200mb total u-wind for the high and low index of\nthe first EOF are shown in Figure 2. It indicates that both the east Asian subtropical\njets and tropical easterlies are stronger in the high index composite and that there is\nno clear longitudinal displacement between the high and low cases. For the second\nEOF (Figure 3), however, there is a substantial eastward expansion of the east Asian\nsubtropical jets and tropical easterlies in the low index composite. The whole pattern\nin the western Pacific appears to be shifted eastward in the low index case. The\ncomposite patterns for the second EOF indicates that the zonal expansion and\ncontraction of the east Asian jets produce a significant variability in the central\nNorth Pacific. It is also noted that the strength of Australian jets appears to be out of\nphase with that of east Asian jets.\nIn order to investigate a possible connection between the tropical forcing and\nzonal wind fluctuation, we examined the composite patterns of OLR based on the zonal\nwind fluctuation. Figure 4 shows the difference patterns between the high and low\ncomposites for the first two EOFs with -10 day lag to +10 day lag. The composite\npatterns for the first EOF (in the left hand-panels) show eastward travelling tropical\nconvection in the Indian Ocean and central Pacific region. Here, the half cycle is\n156","about 20 days. It is consistent with the Madden-Julian oscillation period. There seems\nto be a connection between the eastward travelling tropical convection and the zonal\nwind fluctuation. As suggested by Paegle and Mo (1988), the tropical convection\nappears to be a driving force to the zonal wind fluctuation. The composite patterns\nfor the second EOF shows nearly standing oscillation with maximum convection\nfluctuations prior to the zonal wind extremes. The composite patterns of OLR indicate\nthat the convective activity associated with the second EOF is in the node of convective\nactivity associated with the first EOF.\nWe examined a possible link between the zonal wind fluctuation and quasi-\nstationary waves by using a horizontal wave activity flux (Plumb, 1985). The flux is a\nquasi-geostrophic quantity of stationary waves. The flux vectors indicate\nboth\nthe\nstrength of the wave activity and the direction of propagation while the divergence\n(convergence) gives the source (sink) of wave activity. Figure 5\nshows\nthe\nwave\nactivity flux patterns based on the 200mb eddy streamfunction composites for high\nand low index of the first EOF. For the high case, comparing to the climatological\nwave activity (not shown), there is a weaker source over east Asia and stronger\nsource near the Gulf of Alaska. For the low case, there is a substantial increase in\nwave propagation into the tropics associated with the stronger source in east Asia\n(northeast of the Tibetan Plateau). There is a prounced ridge pattern along the west\ncoast of the North America associated with an apparent wave reflection from the\nsubtropical Pacific into the middle latitudes. The wave reflection appears to occur\nnear the equatorward flank of east Asian jets. This agrees with the suggestion by\nBranstator (1983) that this may be a preferred region for wave reflection. Figure\n6\nshows the wave activity flux patterns for the second EOF. The pattern is close to the\nclimatological wave activity for the high case. For the low case, however, there are\nenhanced sources near the Gulf of Alaska and subtropical Pacific. These enhanced\nsources seem to be related to the eastward expansion of the east Asian jets. These\nwave activity flux patterns indicate that the zonal wind fluctuation is significantly\nrelated to the wave activity source and the propagation of extratropical quasi-\nstationary waves.\n3. Summary\nWe found observational evidence for a strong link between tropical convection\nand low-frequency zonal wind fluctuations in the tropical and subtropical Pacific\nregion. Furthermore, changes in wave activity flux patterns also suggest a close\nrelationship between quasi-stationary waves and low-frequency zonal wind\nfluctuation. In particular, we found that certain phases of the Madden-Julian\noscillation may be associated with significant reflection of wave activity emanating\nfrom the Tibetan plateau. These results support the notion that the low-frequency\nzonal wind fluctuations (which are in part influenced by tropical forcing) produce an\nindirect but important interaction between the tropical forcing and extratropical\nquasi-stationary waves\nReferences\nBranstator, G., 1983: Horizontal energy propagation in a barotropic atmosphere with\nmeridional and zonal structure. J. Atmos. Sci., 40, 1689-1708.\nHeld, I.M., 1983: Stationary and quasi-stationary eddies in the extratropical\ntroposphere: theory. Large-Scale Dynamical Processes in the\nAtmosphere. B.J. Hoskins and R.P. Pearce, Ed., Academic Press.\nKnutson,\nT.R. and K.M. Weickmann, 1987: 30-60 day oscillations: composite life cycles\nof convection and circulation anomalies. Mon. Wea. Rev., 115, 1407-1436.\nLau, K-M., I-S. Kang and P.J. Sheu, 1989: Principal modes of intraseasonal variations in\natmospheric angular momentum and tropical convection. JGR-Atmos.\nNigam, S. and R.S. Lindzen, 1989: The sensitivity of stationary waves to variations in\nthe basic state zonal flow. J. Atmos. Sci., 46, 1746-1768.\nPaegle, J.N. and K.C. Mo, 1988: Transient response of the southern hemisphere\nsubtropical jet to tropical forcing. J. Atmos. Sci., 45, 1493-1508.\nPlumb, R.A., 1985: On the three-dimensional propagation of stationary waves. J.\nAtmos. Sci., 42, 217-229.\n157","90W\nU200(90E 90W)\n0.2,\n0.21\n:0.2\n-0.4\n0.2\n1987\n-0.\n-074\n0.2\n120W\n0.2\n-0.2\n-0.2\n0.2\n1986\nPacific sector (90°E to 90°W and 60°S to 60°N) for the Northern Hemisphere winter\nContour interval is 0.1. Correlations greater than 0.2 or less than -0.2 are\nThe first (left) and the second (right) EOF patterns and the corresponding\ncoefficients of the band-pass filtered (for 20-70 day period) 200mb zonal wind in the\n0.4-0.2\n150W\n:0.2,\n1985\n-0.6\nEOF2 (8.3%)\nshaded. The coefficients are the time-series for the DJF of individual years.\nLONGITUDE\n-024\nD\nTIME\n180\n1984\nO.\n150E\n1983\n0.2\n-0.2\n£2\n0.2\n0.4\n0.64\n1982\n120E\nECMMF (DJF) COR\nO.\n0.2\n6:2\n1981\n-0.2\nb.4\n0.2\n90E\n60N\n30N\n30S\n60S\nEQ\n-2\n-3\n1\n3\n2\n0\n1\n-\n90W\nU200(90E -90W)\n0.2\nD.\n0.3\n0. 0.3\n0.3\n1987\n120W\nCO.S\n0.3\n0.5\n0.2\n0.2\n1986\nb:\n0.6\n0.3\n0.3\n150W\n0.3.0.4.\n6\n1985\n-0.3\n0.2\n0.\n-0.2\n0.3\nEOF1 (11.0%)\nLONGITUDE\nTIME\n0.2\n180\n1984\n0.2\n-0.3\n-0.2\n0.2\n-p.2\nFigure 1\n150E\n1983\n(DJF).\n1982\n120E\nECMWF (DJF) COR\n-0.54\n-0.3-0.2\n0.4,\n0.2\n-0.3\n0.3\n0.2\n0.3\n0.2\n1981\n90E\n60N\n30N\n30S\nEQ\n60S\n-2\n-3\n3\n2\n0\n-1\n1","HIGH(WO WGT)\nWP3 PC1\nU200\n90N\n5\nb\n5\n0\n0\n0\n10\n15\n5\n15\n20\n60N\n10\n5\n30.\n10\n00 35\n35\n.10\n20\n30N\n50\n65\n35\n50\n40\n5\n35\n10\n25\n5\n5\nEQ\n5\n10\n5\n5\n20\n10\n30S\n25\n10\n20\n35\n30\n35\n35\n15\n35\n25\n25\n60S\n10\n15\n-5\n5\n10\n5\n90S\n0\n0\n0\n0\n60E\n120E\n180\n120W\n60W\n0\nLONGITUDE\nWP3 PC1\nU200\nLOW (WO WGT)\n90N\n0\n0\n5\n15\n60N\n20\n03\n5\n25\n30\n35\n35\n30N\n15\n45\n50\n35\n35\n20\n30 35\n5\n5\n25\n5\n5\nEQ\n5\n5\n10\n6\n5\n25\n25\n30S\n20\n20\n35\n35\n40.\n20\n30\n35\n20\n10\n60S\n5\n90S\nH\n0\n60E\n120E\n180\n120W\n60W\n0\nLONGITUDE\nComposite patterns of the 200mb total u-wind for the high index (upper\nFigure 2\npanel) and the low index (lower panel) of the first EOF. Easterlies are shaded. Contour\ninterval is 5 m/sec.\n159","WP3 PC2\nU200\nHIGH(WO WGT)\n90N\nor\n0\n0\n0\n5\n10-\n10\n.15\n60N\n20\n15\n10\n25\n25\n35\n35\n65\n30N\n45\n50\n35\n40\n30\n35\n5\n25\n5\n10\n25\n5\nEQ\n-5\n10\n5\n15\n0\n0\n5\n10\n30S\n20\n25\n35\n15\n30\n35\n25\n35\n0\n35\n20\n25\n25\n60S\n5\n0\n0\n90S\n0\n60E\n120E\n180\n120W\n60W\n0\nLONGITUDE\nWP3 PC2\nU200\nLOW (WO WGT)\n90N\n0\n0\n5\n0\n10\n20\n60N\nG.\n20\ne\no\n25\n5\n5\n35\n35\n40\n15\n30N\n65\n50\n40\n35\n20\n35\n5\n10\n5\n25\nEQ\n5\nTO\n5.\n0\n5\n5\n20\n5\n25\n30S\n10\n25\n25\n15\n30\n35\n20\n30\n30\n35\n60S\n20\n5\n0\n90S\n0:\n0\n0\n60E\n120E\n180\n120W\n60W\n0\nLONGITUDE\nFigure 3\nSame as Figure 2, but for the second EOF.\n160","PC1\nOLR -10 DAY\nHIGH-LOW\nPC2\nOLR -10 DAY\nHIGH-LOW\nor\nor\nSON\nSON\nM\nISN\nINN\nCZY\n8\nED\n153\n159\n3\n308\n308\nU\n483\n254\n0\no\nBIE\nISDE\n100\n120M\nBOW\no\nSOE\n180E\n180\nLSDW\nBOM\n0\nLONGITUDE\nLONDITUDE\nOLR\n-5 DAY\nHIGH-LOW\nPC1\nOLR -5 DAY\nHIGH-LOW\nPC2\nor\nor\nSON\nSON\n0\nISN\nISN\n8\n8\nM\nA\n158\nLSS\na\n308\n303\nD.\nU\n458\n-\n180\n180W\nSDM\n0\n0\nBOE\n120E\n160\nLEOM\nBOM\n0\no\nSOE\n120E\nLONGITUDE\nLONDITUDE\nPC2\nOLR\n0 DAY\nHIGH-LOW\nPC1\nOLR 0 DAY\nHIGH-LOW\nor\nor\n()\nyou\nSON\nSUN\nISN\nISN\n8\nB\n1:\n153\n153\n.\n308\n308\n438\n138\n0\no\nODE\nISDE\n160\n120W\nBOM\n0\no\nSOE\nISDE\n180\n120W\nBOW\nLONDITUCE\nLONGITUDE\nOLR\n+5 DAY\nHIGH-LOW\nPCI\nOLR +5 DAY\nHIGH-LOW\nPC2\nby\nto\n(K.).\nSON\nSON\nISN\nISN\n8\n8\n198\n153\n300\n308\nD\n938\n8.\n0\n0\nSOE\n120E\n100\n120M\nBOM\n0\n0\nUDE\nISDE\n180\nLEON\nBDM\nLONDITUDE\nLONGITUDE\nOLR +10 DAY\nHIGH-LOW\nPC1\nOLR +10 DAY\nHIGH-LOW\nPC2\nto\nSON\nSON\nR\n15N\nISN\nED\n8\nthe\nor\n158\n159\n303\n303\n433\n100\n0\nGOE\nISDE\n180\n120M\nBOW\n0\n0\nBOE\n120E\n100\n120M\nBOM\n0\nLONDITUDE\nLONGITUDE\nThe composite patterns of OLR (high case - low case) for the first EOF (left)\nFigure 4\nand the second EOF (right) with -10 day lag to +10 day lag. The values greater than 10\nW/m2 2 or less than -10 W/m2 2 are shaded. Contour interval is 5 W/m ²².\n161","CMP HIGH\n200.1 STF PC1\n30\n90N\n60N\n30N\nEQ\n30S\n60S\n90S\n0\n60E\n120E\n180\n120W\n60W\n0\nLONGITUDE\nCMP LOW\n200.MB STF PC1\n30\n90N\n60N\n30N\nEQ\n30S\n60S\n90S\n0\n60E\n120E\n180\n120W\n60W\n0\nLONGITUDE\nFigure 5\nThe wave activity flux patterns based on the 200mb eddy streamfunction\ncomposites for the high and low index of the first EOF. Solid (dotted) lines represent\nthe divergence (convergence) of the wave flux.\n162","200.MB STF PC2\nCMP HIGH\n30\n90N\n60N\n30N\nEQ\n30S\n60S\n90S\n120E\n180\n120W\n60W\n0\n0\n60E\nLONGITUDE\n200.MB STF PC2\nCMP LOW\n30\n90N\n60N\n30N\n2\nEQ\n30S\n60S\n90S\n0\n0\n60E\n120E\n180\n120W\n60W\nLONGITUDE\nSame as Figure 5, but for the second EOF.\nFigure 6\n163","Climate Variability As Measured By Observations of Lightning Activity\nSteven J. Goodman\nNASA/Marshall Space Flight Center\nHuntsville, AL 35812\n(205)544-1683\nIntroduction\nGlobal lightning observations will be possible during the 1990s from geostationary (e.g.,\nNOAAs GOES-M) and polar orbiting platforms (NASAs Earth Observing System and Space Sta-\ntion Freedom). Previous studies have shown lightning is associated with heavy rainfall in many\nclimatic regions at the storm and meso-scales, is one of the major sources of naturally occurring\nNO and that thunderstorms are the generators of the Earth's global electric circuit. Ground-\nbased regional lightning networks offer an opportunity to examine the potential information\ncontent and uses of the space-based optical sensors currently under development to map global\nintracloud and cloud-to-ground lightning activity (Christian et al., 1989). Ground-based lightn-\ning networks in operation on a regional scale in the U.S. and elsewhere since the mid-1980s\nhave been used to explore the scientific use of regional, national, and global lightning observa-\ntions. Cloud-to-ground lightning activity can now be examined for its spatial, temporal, and\ninterannual variability and what this tells us about the variability of convection and the en-\nvironmental forcing. These issues will be addressed with data acquired since 1986 by a lightn-\ning network located in the Tennessee Valley (Goodman et al., 1988).\nResults and Discussion\nFigures 1 and 2 show the global spatial distribution of thunderstorm activity and its\ndiurnal variation (Whipple and Scrase, 1936). More than 60 thunderstorm days per year are\ncommon in low-latitudes, and extending as far north as central Alabama in the Southeastern\nU.S. The diurnal variation of global thunderstorm activity follows the passage of the sun and\nits heating of the major continents with the maxima occurring 6 h apart. However, much of\nthe warm season rainfall (up to 70% in the major crop growing regions), extensive flooding, and\nsevere weather is produced at the mesoscale. Fig. 3 shows the geographical distribution of\nthese organized regional scale convective weather systems, having lifetimes on the order of 12\nh\nand spatial extents of a few hundred kilometers. Note that the three major centers of mesoscale\nconvection in the Americas occur poleward of the thunderstorm day maxima. The orographic\nand environmental interactions that initiate and sustain these weather systems yields a different\npicture of thunderstorm significance that tends to get lost when looking only at the thun-\nderstorm days. For example, Goodman and MacGorman (1986) found that up to 25% of the\nannual lightning strikes at a given site can be produced by the passage of a single Mesoscale\nConvective Complex (MCC). Fig. 4 shows the typical life-cycle evolution of cloud-to-ground\nlightning activity and rainfall volume produced by MCCs in the Central U.S. There is a high\ncorrelation (r > .9) between the total lightning activity and the total rainfall, but the lightning\nactivity subsides as the total areal precipitation becomes increasingly dominated by stratiform\nprecipitation. The correlation between lightning and surface rainfall however, is greatest for\nthe convective rainfall areas dominated by the taller storms possessing strong updrafts and\nprecipitation sized ice particles. It follows that there should be a good relationship between\ndaily lightning totals and the instability of the atmosphere. We have computed a linear correla-\ntion between daily lightning flashes in the Tennessee Valley and the surface-50 kPa Lifted In-\ndex (not shown) from 12 UTC soundings taken near Huntsville during June and July 1986. The\ncorrelation coefficient is about -0.7 indicating, not surprisingly, that a greater number of\n164","lightning flashes occur in the more convectively unstable environments. These physical charac-\nteristics play an important role in the relationship between lightning, rainfall, and the environ-\nment at all space and time scales.\nThe June cloud-to-ground lightning and mean wind in the Tennessee Valley for the\nyears 1986-1989 shows the greatest differences in the years 1988 and 1989 (Fig. 5). During\nJune 1988 the Tennessee Valley was under the influence of a ridge of high pressure in the\ncentral U.S. attributable to the coldest ENSO episode in the past 15 years. Weak northwesterly\n85 kPa and 20 kPa flow with no vertical wind shear existed through the depth of the tropo-\nsphere. Fig. 6 shows minimal areal precipitation across north Alabama in June 1988 compared\nto 1986 and 1987 when the precipitation at Huntsville was above normal. June 1988 is of inter-\nest because it is the driest on record at Huntsville (HSV) with only 4.3 mm (0.17 in.) rainfall (17\nlightning flashes within 10 km of the airport). Interestingly, 3.8 mm of the total rainfall was\nrecorded on the last day of the month. Normal June rainfall at Huntsville (Fig. 7) is 100 mm\n(3.97 in.). The rainfall patterns for 1986-1988 show a general agreement between the lightning\nand rainfall patterns but there are some important differences as well. Some of the lightning\nmaxima do not correspond to maxima in the rainfall pattern. This is seldom the case for in-\ndividual mesoscale convective events containing tall thunderstorms (see Goodman and MacGor-\nman, 1986). The lack of one-to-one correspondence between the gaged rainfall and lightning is\npartially due to the undersampling of the Cooperative Observing Network (indicated by dots in\nFig. 6). An added complication to this picture is that intracloud lightning flashes are not\ndetected and located by the ground-strike networks. Perhaps more importantly is the contribu-\ntion of stratiform rainfall (when little or no lightning occurs) to the total rainfall amounts.\nStratiform rain events without lightning tend to diminish the contribution of a single intense\nthunderstorm.\nWhat a difference a year makes! June 1989 is also noteworthy because it sets the all\ntime June record with 380 mm (14.99 in.). A total of 416 lightning flashes within a radius\nof\n10 km of HSV were recorded during the month. In 1989 the moist low level flow from the\nGulf of Mexico returned and the upper level wind of nearly 20 m s-1 associated with the sub-\ntropical jet stream was the strongest in the 4 year period (Fig. 5). The 1989 mean 85 kPa wind\ndiffered from the earlier years only in direction, not in magnitude (5 m s-1). Deep convection\nis widespread in 1989 as depicted by the large number of counties having 150 or more lightning\nflashes per 100 kmi which represents about 20% of the estimated mean lightning flash density\n(8 flashes km-superscript(2) for this part of the country (MacGorman et al., 1984). Fig. 8 shows the daily\nJune 1989 rainfall and cloud-to-ground lightning time series at HSV. Except for the stratiform\nrain events on June 6 and 8, there is a good correspondence between lightning within 10 km of\nHSV and HSV rainfall. On June 6 a very electrically active storm moved toward HSV from the\nwest but the deep convection decayed and only stratiform precipitation was present as the storm\npassed by.\nConclusions\nMonitoring the changes in lightning activity provides a means to diagnose the nature and\nvariability of convection. Thunderstorms respond to regional and global scale changes in atmos-\npheric structure and dynamics, and new satellite observing systems will offer new opportunities\nfor observing and categorizing global scale convection.\n165","References\nChristian, H.J., R.J. Blakeslee, and S.J. Goodman, 1989. The detection of lightning from geos- -\ntationary orbit. J. Geophys. Res. (Special Issue), in press.\nGoodman, S.J., and D.R. MacGorman, 1986. Cloud-to-ground lightning activity in mesoscale\nconvective complexes. Mon. Wea. Rev., 114, 2320-2328.\n, D.E. Buechler, and P.J. Meyer, 1988. Convective tendency images derived from a com-\nbination of lightning and satellite data. Wea. and Forecasting, 3, 173-188.\nMacGorman, D.R., M.W. Maier and W.D. Rust, 1984. Lightning strike density for the con-\ntiguous United States from thunderstorm duration records. U.S. Nuc. Reg. Comm. Rept.,\nNUREG/CR-3759, 44 pp.\nMcAnelly, R.L., and W.R. Cotton, 1986. Meso-,--csale characteristics of an episode of meso-a-\nscale convective complexes. Mon. Wea. Rev., 114,1740-1770.\nVelasco, I., and J.M. Fritsch, 1987. Mesoscale convective complexes in the Americas. J.\nGeophys. Res.,92, 9591-9613.\nWhipple, F.J.W., and F.J. Scrase, 1936. Point discharge in the electric field of the Earth.\nGeophys. Memoirs (London) VIII, 68, 20.\n166","70\n60\n50\n20\n10\n20\n30\n40\n40\n30\n40\n40\n20\nto\nso\n10\n20\n20\no\n20\no\n20\n10\nso\n40\nn\n20\n30\n40\n20\n10\n10\n50\n60\n-\n70\n180\n165\n150\n135\n120\n105\n90\n75\n60\n45\n30\n15\no\n15\n30\n45\n60\n75\n90\n105\n120\n135\n150\n165\n180\nFig. 1. Annual number of thunderstorm days (shaded area for > 60 days).\n120\nThe World\n100\n80\nAmerica\nAfrica\n60\nand\nEurope\nAsia\nand\n40\nAustralia\n20\nNew Zealand\n0\n0\n4\n8\n12\n16\n20\n24\nHOUR\nFig. 2. Diurnal variation of global thunderstorm activity (annual mean).\nTime is UTC.\n167","40N\n20N\nso\n110W\n90W\n70W\n50W\n30W\n20S\nJan Feb\nMar Apr\nMay June\nJu'y Aug\nSept Oct\nNov - Dec\ns\n40S\nFig. 3. Geographic and monthly distribution of Mesoscale Convective Complexes (MCC) in and\naround the Americas (locations derived from the time of the maximum observed cold cloud\nshield extents. [From Velasco and Fritsch, 1987.]\n1.2\n450\n- Lightning\n- Precipatation\n1.0\n375\n0.8\n300\n0.6\n225\n0.4\n150\n0.2\n75\n0.0\n0\n-7\n-5\n-3\n-1\nM\n1\n3\n5\nMCC LIFE-CYCLE (IN HOURS)\nNote: The lightning rate is based on 10 MCC cases: rainfall rate is based on 3 MCC cases, which are centered on the maximum\ncloud extent.\nFig. 4. Hourly cloud-to-ground lightning rates and rainfall as a function of MCC life-\ncycle.[Adapted from Goodman and MacGorman, 1986; McAnelly and Cotton, 1986.]\n168","JUNE 1987\nJUNE 1986\n-\n0\nno\nJUNE 1989\nJUNE 1988\n10\n20\n50\n100\n150\n200\nFig. 5. Total lightning strikes for June 1986-1989 in 0.1 X 0.1° grids. The mean monthly 85\nkPa (long arrow) and 20 kPa (short arrow) wind directions for this region are shown above each\nmonthly summary (from the NMC final analysis depicted in the Climate Diagnostics Bulletin).\nA 10 km range ring is centered at Huntsville, AL.\n169","180\n?\n120\nFig. 6. Total monthly rainfall (mm) for June\n1986-1988 interpolated from the Alabama (only)\nCooperative Observing Network (stations indicated\n160\n100\n80\n100\nTO\nJUNE 1987\n120\nso\n140\naso\nIOTL (FIS) NO JUN\nby dots).\nMCIDAS\n40\nre\n80\n60\n100\n120\nJUNE 1986\nJUNE 1988\n140 120\n100\n10\n10\n180\n160\nTOTL (HB) MO JUN\nTOTL (HU) MO JUN","15\n0\n1927\n1989\nHUNTSVILLE\nFig. 7. Total June rainfall (in.) for Huntsville, AL 1927-1989. The 30 year mean (3.97 in.) is\nbased on the period of record 1941-1970. -\nCloud-to-ground Lightning and Rainfall\nJune 1989 at Huntsville, AL\n180\n80\nRainfall\nFlashes\n120\n40\n60\n0\n0\n25\n5\n15\nDay of Month\nFig. 8. Daily rainfall and cloud-to-ground lightning (within 10 km) for June 1989 at Huntsville,\nAL (HSV).\n171","Global Monitoring of Net Solar Irradiance\nat the Ocean Surface and Validation of General\nCirculation Model Simulations\nBeth Chertock\nNOAA/ERL/Wave Propagation Laboratory\nBoulder, CO 80303\nThe present study marks the first time a global, long-term,\nsatellite-based record of ocean surface solar irradiance has been\ncompared with the net surface solar irradiance simulated by\natmospheric general circulation models (GCMs) . The spatial and\ntemporal variability of the satellite data base was used to assess\nthe realism of the surface solar irradiance fields produced by two\nindependent GCMs.\nThis study only recently became possible when a new method\nwas developed and used to generate the first long-term (November\n1978 through October 1985) climatology of surface solar irradiance\nover the world oceans (Chertock, 1989)\nBased\nNimbus-7\non\nsatellite wide-field-of-view - planetary albedo data, the climatology\nhas ~1000 km horizontal resolution and monthly mean temporal\nresolution.\nThe new record was extensively validated by\ncomparisons with high-resolution satellite records from the MONEX,\nTOGA, and Tropic Heat field experiments, and was shown to be\naccurate to within 10 to 20 Wm 2\nThe two GCMs involved in this effort were the GFDL GCM\n(spectral model with rhomboidal truncation at wavenumber 15 and\nocean heat flux correction) from NOAA Geophysical Fluid Dynamics\nLaboratory and the GLA GCM (previously known as the GLAS fourth-\norder GCM) from NASA Goddard Space Flight Center.\nThe NOAA GCM was run for 10 model years, and the resulting\nlong-term monthly climatologies were computed. Difference fields\nwere obtained by subtracting the satellite-based climatology from\nthe corresponding model climatology for each calendar month. The\ndifference fields for December and June (Figures 1a and 2a,\nrespectively) exhibit a negative bias (i. too little surface\nsolar irradiance in GCM climatology) in the tropics of the summer\nhemisphere and a strong positive bias (i.e., too much surface solar\nirradiance in GCM climatology) in the extratropics of the summer\nhemisphere. March and September difference fields (Figures 1b and\n2b, respectively) show strong positive errors in the extratropics\nof both hemispheres. The GCM also maintains a positive bias in the\neastern Pacific.\n172","January and July climatologies (Figures 3a and 4a,\nrespectively) were obtained for the NASA GCM by averaging the\nmonthly values taken from four 47-day simulations in each case.\nAgain satellite-based climatologies (Figures 3b and 4b) were\nsubtracted from the corresponding model climatologies to produce\ndifference fields (Figures 3c and 4c). According to these results,\nthe NASA GCM has a positive bias in the extratropics and a negative\nbias in the tropics, especially in the summer hemisphere. The GCM\nalso fails to simulate the persistent stratus off western North\nAmerica.\nIn general, the positive biases suggest cloud fractions which\nare too small and/or clouds which are optically too thin, whereas\nnegative biases indicate cloud fractions which are too large and/or\nclouds which are optically too thick. Thus, the qualitative and\nespecially the quantitative information provided by this study\nallow for a more comprehensive evaluation of the cloud algorithms\nemployed by these GCMs than was previously possible.\nChertock, B , 1989: Global monitoring of net solar irradiance at\nthe ocean surface using Nimbus-7 satellite data. Doctoral\nDissertation, Scripps Institution of Oceanography, University\nof California, San Diego, 118 pp.\n173","90N\n(a)\n60N\n30N\n-20\nEQ\n20\n20\n-20\n0\n20\n20\n0\no\n30S\n0\n20\n40\n60\n60S\n40\n60\n90S\n0\n60E\n120E\n180\n120W\n60W\n0\n90N\n(b)\n60N\n40\n60\n40\n40\n60\n30N\n40\n20\n20\nEQ\n20\n0\n-20\n30S\n0\n-20\n20\n20\n40\n40\n60S\n40\n20\nGOOD\n90S\n0\n60E\n120E\n180\n120W\n60W\n0\nFigure 1 . Difference fields of net surface solar irradiance\nobtained by subtracting the satellite-based - monthly\n(Wm-2)\nclimatology from the corresponding GFDL GCM monthly climatology\nfor (a) December and (b) March. The contour interval is 20 Wm-2\nand negative contours are dashed.\n174","90N\n(a)\n-\n60N\n120\nS\no\n30N\n-60\n80\n-40\n20\nEQ\n-20\n-40\n30S\n-40\n-40\n-20\no\n0\n60S\n90S\n0\n60E\n120E\n60W\n180\n120W\n0\n90N\n(b)\n60N\n60\n40.\nS\n20\n0\n40\n-20\n30N\n20\n-40\n.\n40\n60\no\n20\nEQ\n0\n20\n40\n30S\n20\no\nC\n40\n20\n0\n20\n60S\n20\n90S\n60W\n0\n60E\n120E\n180\n120W\n0\nFigure 2. Difference fields of net surface solar irradiance\n(Wm-2 2 ) obtained by subtracting the satellite- - based monthly\nclimatology from the corresponding GFDL GCM monthly climatology\nfor (a) June and (b) September. The contour interval is 20 Wm-2\nand negative contours are dashed.\n175","90N\n(a)\n-\n60N\n40\n40\n100\n100\n30N\n160\n180\n180\n200\nEQ\n220\n220\n240\n260\n240\n30S\n280\n240\n280\n240\n260\n260\n240\n220\n220\n200\n180\n60S\n180\n180\n90S\n0\n60E\n120E\n180\n120W\n60W\n0\n90N\n(b)\n-\n60N\n20\n20\n100\n30N\n100\n180\n200\nEQ\n220\n240\n260\n240\n260\n280\n30S\n260\n260\n240\n240\n180\n180\n60S\n140\n160\n90S\n0\n60E\n120E\n180\n120W\n60W\n0\n90N\n(c)\n60N\n20\n30N\n40\n20\n20\no\n-20\n0\nEQ\na\n-60\n$\n-20\n20\n-40\n30S\n0\n20\no\na\n20\no\no\n=\n0\no\n60S\n20\n20\n90S\n0\n60E\n120E\n180\n120W\n60W\n0\nFigure 3 .\nJanuary climatology of net surface solar irradiance\n(Wm-2) for (a) the GLA GCM and (b) the satellite-based - record. The\ndifference field obtained by subtracting (b) from (a) is shown in\n(c), with negative contours dashed. In all figures the contour\ninterval is 20 Wm-2\n176","90N\n(a)\n-\n220\n60N\n<180\n200\n220\n30N\n220\n240\n220\n240\n200\n240\nEQ\n- 200\n200\n200\n30S\n100\n100\n100\n40\n40\n60S\n1\n90S\n60W\n0\n120W\n0\n60E\n120E\n180\n90N\n(b)\n120\n60N\n120\n200\n260\n30N\n280\n280\nV\n200\n220\nEQ\n200\n160\n100\n30S\n100\nto\n20\n0\n0\n60S\n90S\n60W\n0\n0\n60E\n120E\n180\n120W\n90N\n(c)\n100*\n60N\n60\n0\n-60\n30N\n-60\n-20\n-20\nEQ\n0\n20\no\n0\n20\n20\n30S\n40\n40\n60S\n698\n90S\n120W\n60W\n0\n180\n0\n60E\n120E\nFigure 4 o July climatology of net surface solar irradiance\n(Wm-2) for (a) the GLA GCM and (b) the satellite - based record. The\ndifference field obtained by subtracting (b) from (a) is shown in\n(c), with negative contours dashed. In all figures the contour\ninterval is 20 Wm\n.\n177","NATURAL VARIABILITY IN SUMMERTIME DROUGHTS\nOVER THE UNITED STATES GREAT PLAINS\nFong-Chiau Chang\nApplied Research Corporation\nGoddard Space Flight Center/NASA\nGreenbelt, MD 20771\nand\nK.-M. Lau\nGoddard Space Flight Center/NASA\nGreenbelt, MD 20771\n1.\nINTRODUCTION\nSummertime droughts over the central Great Plains have severe\nimpact on the socio-economic status of the United States. Climatological\nstatistics indicate the mean total rainfall during summer is between 15 cm\nand 30 cm everywhere in the United States east of Rockies, except along the\nsouth Atlantic and Gulf Coasts. Mean monthly total rainfall decreases very\nsharply west of the 100°W across the Great Plains. Summer rainfall\ntypically accounts for 40% of the annual average precipitation over the Great\nPlains. The variability of mean monthly total rainfall is large over the Great\nPlains.\n2.\nDOMINANT DROUGHT PATTERN OVER THE\nUNITED STATES GREAT PLAINS\nChang and Wallace (1987) have discussed summertime heat waves\nand droughts in the Kansas City area. In particular they focused on their\nassociated land-surface moisture conditions and large scale circulation\npatterns for the period from 1899 to 1980. They identified drought periods\nsimply by ranking the monthly mean surface air temperature anomalies at\nKansas City. Surface air temperature is a good indicator of drought\nbecause of the high spatial coherence of monthly mean temperature and total\nprecipitation anomalies, and the inverse relationship between these two\nclimate parameters over the central United States. The correlations (Fig. 1a)\nin summertime (June, July and August) monthly mean temperature\nanomalies between Kansas City and every climate divison during the period\n1931-1982, provides evidence of high spatial coherence in temperature over\nthe central United States. Precipitation also has high spatial coherence (see\nFig. 1b). The shape of the precipitation correlation pattern is similar to the\nprecipitation anomalies during April-May-June 1988; the latest severe\ndrought [see Fig. 4 of Trenberth et al. ( 1988)]. The local correlation\ncoefficients between monthly mean temperature and monthly total\nprecipitation anomalies over the United States during summer are negative\nalmost everywhere (Fig. 2), the relationships being strongest over the Great\nPlains. Correlations are stronger than -0.5 over much of the Southern Great\nPlains.\n178","The above relationships are supported by the results of Chang\n(1986), who examined the climate variations over the contiguous United\nStates using divisional monthly mean temperature and total precipitation.\nHer first orthogonal rotated eigenvector (Fig. 3) indicates that drought,\nsynonymous with a combination of high temperature and deficient rainfall,\nis a dominant type of climate fluctuation over the United States during\nsummer. The pattern indicates that much of the high plains are included\nwithin the region of large amplitude. Note the similarity of the shape of the\nEOF pattern and the relationship between temperature and precipitation in\nFig. 3 and those in Fig. 1 and 2. It is not surprise that the correlations\nbetween the principal component of this mode and temperature anomalies at\nKansas City for 156 months is as high as 0.85. Hence, it is reasonable to\nstudy the fluctuation of droughts in the United States Great Plains using the\ntemperature anomalies at Kansas City as a proxy indicator.\nPERIODICAL SIGNAL OF SUMMERTIME DROUGHTS\n3.\nOVER THE UNITED STATES\nFig. 4 shows the time series of temperature anomalies at Kansas\nCity. Most conspicuous in the 92-year time series is the clustering of\ndrought and non-drought years. Years of widespread severe drought\nclustered into epochs lasting 5-10 years. Drought years clustered most\nnoticeably in the 1930's and the 1950's. Stretches of several years without\ndrought occurred before and after the extreme drought (e.g., 1920-1929 and\n1940-1946).\nQuasi-periodic cycles (e.g., 22-year, 11-year and quasi-biennial) are\nknown to exist in climate data. It is difficulty to extract subtle periodic\nsignals from conventional spectrum analysis based on a short and noisy\nclimate record. Recently, Vautard and Ghil (1989) have shown that\nsingular spectrum analysis (SSA) can efficiently determine periodic signals\nfrom noisy and short climate time series. Essentially, the technique\nprovides objective time filters to separate dominant albeit sometimes\nlow-amplitude oscillations in a given time series.\nSSA is applied to the 92-year temperature anomalies at Kansas City.\nThe evidence for the sun's effect on drought appear in the first three\ncomponents. Fig. 5a presents the first mode of the 92 year temperature\nanomalies from SSA. Together with the second mode (which account for\n37 percent of the variance) is seen a period of about 22 years. That may be\nrelated to the perod of the sun's Hale sunspot cycle, during which the\nmagnetic poles of the sun and its sunspots flip and return to original\norientation. The third component (10.5%) shows an 11-year cycle (solid\nline in Fig. 5b) which compares well with the actual sunspot cycle (dotted\nline in Fig. 5b). The fifth mode of the drought (solid line in Fig. 5c)\nexhibits a significant quasi-biennial oscillation which compares well with\nthe quasi-biennial oscillation (QBO) index based on stratospheric wind for\nthe period after 1953 (dotted line in Fig. 5c).\n179","4.\nREMARKS\nIt has been shown in the preceding section that there are natural\nvariability associated with sunspot activities (11-year sunspot cycle and\n22-year Hale sunspot cycles), quasi-biennial oscillation present in the\nKansas City surface temperature data. A relatively weak 3-5 year irregular\ncycle associated with the ENSO is also appear in this 92-year time series\n(not shown). This may indicate that while tropical sea surface temperature\nforcing may contribute to regional United States droughts, other factors\nsuch as local soil moisture-insolation feedback may be more important in\nself-perpetuating consistent with the results of Chang and Wallace (1987).\nNo significant long term trend appear in the 92 year Kansas City\ntemperature anomalies. Whether this suggests a (non)relationship between\nthe United States drought and the greenhouse effect for global warming is\nnot clear. Further studies are needed.\n5.\nREFERENCES\nChang, F. C., 1986: On the seasonality of climate fluctuations over the\ncontiguous United States. Ph.D. Thesis, pp.161.\nand J. M. Wallace, 1987: Meteorological conditions during\n,\nheat waves and droughts in the United States Great Plains. Mon. Wea.\nRev., 115, 1253-1269.\nTrenberth, K. E., G. W. Branstator and P. A. Arkin, 1988: Origin of the\n1988 North America drought. Science, 242, 1640-1645.\nVautard, R. and M. Ghil, 1989: Singular Spectrum Analysis in Nonlinear\nDynamics, with applications to paleoclimatic time series, Physica, D.\n180","0.4\n0.6\n08\n8\n0.2\n6\n(a)\n2\n124%\n0.1\n2\n0.3\n0.2\n0.4\n0.6\n0.8\n.6\n(b)\n12.4%\nFig. 1. (a) Correlation coefficients in summertime\n3. The first rotated eigenvector derived from the\nFig.\nmonthly mean temperature anomalies between\ncombined temperature and precipitation fields over\nKansas City and every climatic division. Contour\nthe contiguous United States. Top panel:\ninterval 0.1. (b) As in (a), but for monthly total\ntemperature; lower panel: precipitation. Contour\nprecipitation. Based on temperature anomalies at\ninterval 0.1.\nKansas City and Historical Drought data for 150\nmonths (June, July, August of the summers\n1931-80).\nFig. 2. Local correlation between monthly mean\n54\ntemperature and precipitation over the United\n.5\n.2\nStates, based on Historical Drought Data for 156\nmonths (June, July, August of the summers\n1931-82). Contour interval 0.1.\n2\n181",".6\nJune\n4\nJuly\nAugust\n2\n0\n-2\n4\n-6\n2\n12\n13\n14\n15\n16\n17\n181\nYear\n6\n4\n2\n0\n-2\n4\n-6\n20 21 22 23 24 25 26 27 28 29 3031 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 484950\nYear\n6\n4\n2\n0\n-2\n.4\n-6\n51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 75 76 77 78 798081\nYear\nFig. 4. Time series of the temperatures anomalies for 276\nmonths (June, July, August of the summers\n1899-1980).\n182","4\n2\n0\n-2\n-4\n-6\n-20\n0\n20\n40\n60\n80\n3\n2\n1\n0\n-1\n-2\n-3\n80\n20\n40\n60\n-20\n0\n3\n2\n1\n0\n-1\n-2\n-3\n-4\n50\n60\n70\n80\nYear\n5. The first (a), third (b) and fifth (c) mode of SSA\nFig.\nof temperature anomalies of Kansas City.\n183","Seasonal Relations Between Tropical Pacific\nSSTs and Northern Hemisphere 700 mb Heights\nNicholas E. Graham\nScripps Institution of Oceanography\nLa Jolla, CA 92093\nLinear statistical models are used to examine the spatial and temporal characteristics of the links\nbetween tropical Pacific SSTs (equatorward of 18° latitude) and 700 mb heights in the Northern Hemisphere\n(poleward of 30° latitude). Three analyses are described.\nIn the first analysis, contemporaneous relationships are examined using canonical correlation analysis\n(e.g., Graham et al. 1987). As expected, the strongest associations were found in winter and the weakest\nin summer. During winter, spring, and fall two types of 700 mb patterns dominate, both of which tend\nto have highest amplitudes over the North Pacific. In winter and fall, one of these patterns can clearly be\nlinked El Niño activity In spring, such an association is suggested by both canonical modes. Particularly in\nwinter, the 700 mb pattern associated with extremes of the El Niño (La Niña) cycle is nearly identical to\nthat shown earlier by others (e.g., Horel and Wallace, 1981). Figure 1 shows the first canonical mode for\nwinter 700 mb heights and tropical Pacific SSTs. Note the large negative anomalies in eastern North Pacific\n700 mb heights when the tropical SST pattern is typical El Niño conditions. The canonical correlation (R2)\nof 0.60 indicates a close relation between these patterns. Figure 2 gives the cross-validated correlations for a\ncanonical correlation model for specifying winter 700 mb heights from tropical Pacific SSTs. High values in\nexcess of 0.4 are indicated over the sub-tropical western and eastern North Pacific. During summer, although\nthe statistical link is quite modest, the SST and 700 mb patterns are consistent with the results of Trenberth\net al. (1989) who suggest that remote forcing from the tropical Pacific may have contributed to the genesis\nof the central U.S. drought of 1988. The 700 mb pattern is similar to that noted by Namias (1982) in his\nanalysis of mid-American drought.\nThe second analysis concerned the possibility of using statistical models to forecast winter Northern\nHemisphere 700 mb height anomalies from tropical Pacific SSTs one to two seasons in advance. These\nmodels showed surprisingly good results suggesting that over parts of the North Pacific and U.S. as much as\n15% to 25% of the variance is predictable using the linear techniques applied here. In addition, this result is\nconsistent with the view that the lack of strong SST-700 mb teleconnections during summer and fall is due\nto poor communication between the fields, rather than inconsistent behavior of the tropical Pacific SST field\nduring those seasons. Figure 3 shows the first canonical mode linking summer tropical Pacific SSTs with\nNorthern Hemisphere 700 mb height anomalies during the following winter. Note that the 700 mb pattern\nis very similar to that seen in Fig. 1. Although less SO than in winter, the the SST pattern does show some\nfeatures typical of El Niño episodes. The canonical correlation (0.68) is suggests a strong association between\nthe patterns. Figure 4 shows the cross-validated correlations for predicting winter Northern Hemisphere 700\nmb heights on the basis of tropical Pacific SSTs during summer. The pattern is similar to that seen for the\nwinter specification model (Fig. 2) and shows the highest correlations of 0.4 to 0.5 over the central North\nPacific, the southeastern U.S., and over the sub-tropical eastern North Atlantic.\nThe third analysis examines the decrease in winter 700 mb heights which was observed over the North\nPacific in the years following the winter of 1975-76 (e.g., Venrick et al., 1987; Folland and Parker, 1988;\nTrenberth, 1989). The idea that this change represents a response to forcing from the tropical Pacific is\nexamined. Figures 6 shows the first canonical mode for winter outgoing longwave radiation over the tropical\nPacific and Northern Hemisphere 700 mb heights. The spatial patterns are quite similar to the change in\nwinter means for thosse fields observed before and after 1975-76. Although the period available for analysis is\nshort, the high canonical correlation and the agreement of the two time series at both high and low frequencies\nsuggests a consistent physical relationship between the fields. Similarly, the spatial patterns associated with\nthe second canonical mode for winter tropical Pacific OLR and SSTs are similar to that observed in the\nmean fields during the mid-1970's and the time series again suggest a strong relation between the fields.\nAcknowledgments. This work was supported by NOAA Grant NA86AA-D-CP104 for the Experimental\nClimate Forecast Center at the Scripps Institution of Oceanography.\n184","REFERENCES\nFolland, C. K. and D. Parker, 1988: Observed variations of sea surface temperature, Proc. NATO Ad-\nvanced Workshop on Climate Ocean interaction, Oxford, England; September 1988, D.\nReidel (in press).\nGraham, N. E., J. Michaelsen, and T. P. Barnett, 1987: An investigation of the ENSO cycle with statistical\nmodels. 1. Predictor field characteristics, J. Geophys. Res., 92, 14251-14270.\nHorel, J. D. and J. M. Wallace, 1981: Planetary-scale atmospheric phenomena associated with the Southern\nOscillation, Mon. Weather Rev., 109, 813-829.\nNamias, J., 1982: Anatomy of Great Plains protracted heat waves (especially the 1980 U.S. summer drought),\nMon. Weather Rev., 110, 824-838.\nTrenberth, K. E., G. W. Branstator, and P. A. Arkin, 1989: Origins of the 1988 North American drought,\nScience, 242, 1640-1645.\nTrenberth, K. E., 1989: Recent interdecadal Northern Hemisphere climate changes, in preparation.\nVenrick, E. L., J. A. McGowan, D. R. Cayan, and T. L. Hayward, 1987: Climate and Chlorophyll a: Long-\nterm trends in the central North Pacific Ocean, Science, 238, pp. 70-72.\n185","WINTER SPECIFICATION MODEL\nCANONICAL MODE I\nu=0.60\nSPATIAL PATTERNS\nA.\n0°\n60°W\n60°E\nBUTTE\n50°N\n700mb\n30°N\n70° NY\n120°W\n120°E\nDO\nO\n-80\n+80\n180°\n120°E\n140°\n160°\n180°\n160°\n140°\n120°\n100°\n80°W\n20°N\n-20\nFio\n-10\n40\no\n10\n20\n20\n-\n10\n0°\nTHE\n-\ne\nto\nIII\nSST\n40\n20\n20\nTO\n20\n20°S\nB.\nAMPLITUDES\n6\n4\nSST\n2\no\n-2\n700mb\n-4\n1950\n1955\n1960\n1965\n1970\n1975\n1980\n1985\nFigure 1. First canonical mode for boreal winter specification model for NH 700 mb heights and tropical\nPacific SSTs, 1950-86; A) Spatial patterns; zero line is heavy; large positive values are hatched;\nlarge negative values are stippled; units are relative; B) Temporal patterns (amplitudes); units are\nrelative.\n186","CROSS VALIDATED CORRELATION (x100)\nWINTER SPECIFICATION MODEL\n0°\n30\n10\nio\n-30\n60°W\n60°E\n50\nIO\n-\n8MI\nN\n120°E\n120°W\n30\n180°\nFigure 2. Cross-validated correlations (x100) for boreal winter specification model; zero contour is\nheavy; negative correlations are stippled.\n187","SUMMER TO WINTER CANONICAL PREDICTION MODEL\nMODE I u = 0.68\nSPATIAL PATTERNS\n0°\n60°E\n60°W\n40\n)\n700 mb\n20\nN. -20\n40\n-60\n120°W\n120°E\n80\n40\n180°\n120°E\n140°\n160°\n180°\n160°\n140°\n120\n100°\n80°W\n-10\n10\n20°N\n10-\n-40\n-10\n10\n0°\n-20\nSST\n6@\nCK\n-10\n20\no\n30\no\n20°S\n30-\n20\nAMPLITUDES\n700mb\nSST\n1\n1950\n1955\n1960\n1965\n1970\n1975\n1980\n1985\nFigure 3. As in (1), but for first canonical mode for boreal summer SSTs to winter 700 mb height\nprediction model.\n188","CROSS VALIDATED CORRELATIONS (x100)\nSUMMER TO WINTER\nCANONICAL PREDICTION MODEL\n0°\n40\n30\n60°W\n60°E\n-40\nF20\n0\n10\nO\n0\n10.0\n20.0\n120°W\n120°E\nO\n180°\nFigure 4. As in (2), but for boreal summer SSTs to winter 700 mb height prediction model.\n189","LEADING CANONICAL MODE FOR TROPICAL PACIFIC OLR\nAND NORTHERN HEMISPHERE 700 mb HEIGHTS\n(WINTER, 1975 - 1986)\nu=0.79\nA.\n0°\n60°E\n60°W\n2\n2\n70°N\n700 mb\n50\n-2\n120°E\n-4\n120°W\n6\n-8\n-10\n180°\n120°E\n140°\n160°\n180°\n160°\n140°\n120°\n100°W\n20°N\n-5\n15>20\n25S\n5\n10\n-5.\n0°\n60\no\n0\nOLR\n-10\n5\n15\n20\nC51\n5\n20°S\n10\n15\n15\nB.\nAMPLITUDES\n1.6\n700mb\n0.8\no\n-0.8\nOL R\n-1.6\n1972\n1974\n1976\n1978\n1980\n1982\n1984\n1986\nFigure 5. Leading canonical mode for boreal winter tropical Pacific OLR and NH 700 mb heights for\nwinters 1974-75 through 1985-86; A) Spatial patterns, 700 mb (upper), OLR (lower); zero con-\ntour is heavy; hatching (stippling) indicates large positive (negative) values; B) Temporal patterns\n(amplitudes).\n190","SECOND CANONICAL MODE FOR\nTROPICAL PACIFIC OLR AND SST (WINTER 1975-86)\nfl = 0.73\n20°N\n20125\n10\nOLR\n5\n-15\n0°\nOLR\nEQ\no\n10\n15\nA.\n120°E\n140°\n160°\n180°\n160°\n140°\n120°\n100°\n80°W\n20°N\n30\n10\n20\nSST\n0°\n80\no\n20\n10\n10\nO\n5\n20°S\n2.0\nB.\n1,6\n0,8\no\n-0.8\nOLR\nSST\n-1.6\n1972\n1974\n1976\n1978\n1980\n1982\n1984\n1986\nFigure 6. As in (5), but for second canonical mode for boreal winter tropical Pacific OLR (upper spatial\npattern) and SSTs (lower spatial pattern).\n191","Variability of Latent and Sensible Heat Fluxes\nover the Northern Oceans\nDaniel R. Cayan\nScripps Institution of Oceanography\nLa Jolla, California 92093\nThe spatial and temporal variability of latent and sensible heat flux anomalies over\nthe oceans is explored in relationship to variations in atmospheric circulation and in ocean\nsurface temperature over the last four decades. The anomalies are constructed by removing\nthe climatological mean fluxes, which are estimated using bulk formulae. The focus is on\ntheir behavior during fall and winter when these fluxes vent much of the ocean's accumulated\nheat to the atmosphere. The magnitude of the flux anomalies is often several 10's of W m-2,\nwith apparently significant values for the latent component occuring from the tropics through\nmiddle latitudes, and for the sensible component only occuring in middle-to-high latitudes.\nThere is considerable spatial structure in the anomaly variance, with greatest values along the\nwestern boundaries of the North Atlantic and North Pacific due to strong mean saturation-\nair humdidity differences (Aq's), sea surface-air temperature differences (AT's), and wind\nspeeds (w's, which interact with perturbations to produce large flux anomalies. In middle\nlatitudes, the latent and sensible flux components are strongly correlated, but there is still\na definite signal in the Bowen ratio (sensible/latent): if the wind is equatorward or off from\na continent, the Bowen ratio increases.\nDespite various sources of error which contaminates the monthly mean fluxes, their\nmonthly scale climate signal is strong enough to identify connections to the large scale at-\nmospheric circulation and to the sea surface temperature field over much of the oceans.\nEvidence that the flux anomalies are real is seen from: (1) their organization by the atmo-\nspheric circulation anomalies, a structure which is very consistent and apparent in both the\nNorth Atlantic and North Pacific basins, and (2) their connection to the upper ocean tem-\nperature, whereby averages over several months with similar atmospheric circulation reveal\nremarkably similar patterns of anomalies of flux and SST tendency.\nTo investigate climate variations of the fluxes, the flux anomalies are related to monthly\natmospheric circulation modes over the North Atlantic and North Pacific. Using sea level\npressure ( SLP ) empirical orthogonal functions (EOF's), the structure of anomalous w,\nAq, and AT reveals how the flux anomalies are organized on basin scales by the anomalous\ncirculation. For middle-latitude low pressure anomaly centers, positive flux anomalies are\nfound to the southwest of the low in the sector with large wind speeds and and large Aq's and\nAT's, while negative flux anomalies occur to the east of the low where small Aq's and AT's\nprevail. Due to teleconnections in the atmospheric patterns, remote flux anomaly centers\nare also found. Centered in the tropical Pacific, but extending into middle latitudes, there\n192","is also a distinct broad-scale pattern of these properties and the fluxes associated with the\nSouthern Oscillation.\nInterestingly, the marine data is good enough to provide some detail in addition to the\npatterns of the monthly mean fluxes. The intramonthly variability of the fluxes also exhibits\na well-organized spatial and temporal signal; the variance within a particular month is\nanomalously high where the monthly mean flux is high.\nTwo situations linking the flux and the SST anomalies are apparent. In the tropics, and\nin middle latitudes in summer and fall when SST is warm, latent flux anomalies tend to be\npositively correlated with local SST anomalies; this is the case where the ocean drives the\nanomalous surface heating (upper panels of Figure 1 show this for zonal averages across the\nPacific (left) and to a lessor extent for zonal averages across the Atlantic (right)). This case\nwhere the fluxes are driven by the SST anomalies is strongest in the eastern tropical Pacific.\nIn contrast, in the extratropics during most of the year the dominant relationship is that\nthe flux anomalies are correlated with the tendency of SST anomalies: positive ocean-to-\natmosphere flux anomalies produce negative SST anomaly changes (lower panels of Figure\n1 show this for zonal averages across the Pacific (left) and North Atlantic (right)). This flux-\nforcing-SST anomalies relationship is detected over most of the ocean, but appears strongest\nin anticyclonic ocean gyre regions in the lower mid-latitudes of both the North Atlantic and\nthe North Pacific. During months when a particular SLP EOF mode is strongly expressed,\nthe latent and sensible fluxes exhibit broad anomaly patches in excess of 50W m-2. and\nthese are accompanied by co-located tendencies in the SST anomaly of 0(0.2°C mo-1) (see\nFigure 2(a,b,c), showing flux and \\SST/At anomaly composites for \"strong\" and \"weak\"\nextremes of North Atlantic SLP EOF's 1 and 2 the North Atlantic Oscillation, and the\nEastern Atlantic pattern). For extratropical regions where the flux-SST anomaly tendency\nis most strongly linked, the inferred mixed layer depth varies over the year between about 50\nand 300 meters. While slightly high, the inferred mixed layer depths are roughly consistent\nwith observations, and also manage to capture the shoaling of the mixed layer in summer,\nas seen in Figure 3.\nAcknowledgements This report is a distillation of my PhD thesis, which has been sup-\nported by the National Science Foundation through a grant (ATM86-12853) to Dr. J.\nNamias, P.I., by the NOAA Experimental Climate Forecast Center at Scripps Institution\nof Oceanography, and by the University of California INCOR program.\n193","Pacific F(1+s) vs. SST\nAlantic F(1+s) vs. SST\nN lat\nN lat\n60\n-10 -24 -7 -11 13 10 19 8 -10 -24 5 - 11\n60\n55\n-11\n-22\n-2\n10\n26\n32\n54\n50\n35\n37\n55\n-4 -23 -17 -28\n41\n35\n9\n-10\n0\n3\n5\n-16\n3\n-2\n50\n4\n1\n5\n14\n5\n21\n43\n58\n50\n48\n35\n-19\n41\n-19\n-16\n-7\n3R\n18\n22\n17\nC30\n38\n-2\n29\n45\n-15 -10\nFASTES\n-5\n5\n15\n24\n66\n68\n45\n52\n40\n24\n7\n16\n-6\n16\n17\n4\n16\n34\n0\n26\n40\n-10 -20 -10\n2\n17\n21\n38\n53\n40\n-22-14\n53\n45\n-7\n26\n25\n-9\n4\n23\n21\n25\n20\n11\n4\n-6\n35\n-18\n17\n26\n35\n33\n43\n38\n-3-12\n26\n24\n4\n10\n-26\n25\n2\n-2\n13\n29\n21\n16\n21\n0\n30\n30\n-25 63\n25\n13\n7\n24\n30\n34\n15\n19\n-2\n14\n-11\n1\n-16\n10\n10\n18\n20\n18\n9\n-9\n25\n12\n32\n18\n23\n15\n25\n28\n20\n29\n-16\n22\n1\n4\n9\n-8\n0\n24\n-6\n3\n9\n7\n1\n-3\nI\n20\n38\n9\n29\n11\n18\n17\n26\n20\n22\n-2\n-12\n26\n32\n-10\n28\n-13\n-1\n20\n-3\n12\n28\n23\n12\n-7\n-1\n15\n42\n33\n19\n9\n-8\n4\n21\n15\n33\n13\n34\n37\n-10\n-11\n-6\n27\n27\n6\n8\n16\n20\n9\n2\n7\n8\n10\nB9\n82\n53\n25\n19\n22\nC5\n40\n10\n21\n41\n33\n9\n1\n29\n1\n5\n17\n16\n29\n13\n38\n33\n32\n23\n5\n56\n57\n60\n#7\n28\n5-17\n5\n3\n16\n-6\n10\n20\n-21\n32\n-1\n0\n9\n-6\n12\n-12\n-9\n3\n23\n27\n0\n34\n39\n28\n98\n55\n52\n51\n51\n0\n38\n4\n-1\n18\n6\n27\n25\n46\n27\n19\n4\n8\n13\n4\n-5\n-5\n78\n61\n42\n27\n54\n73\n78\n80\nE1\n-5\n39\n75\n14\n21\n24\n38\n17\n80\n47\n43\n36\n41\n21\n16\n-10\n18\n6\n-7\n-4\n24\n35\n31\n47\n-10\n61\n-12\n-3\n43\n10\nL8\n25\n25\n12\n12\n11\n1\n10\n8 -11 18\n-15\n17\n18\n10\n9\n19\n27\n-15\n52\n85\n37\n-8\n12\n14\n18\n34\n23\n6\n3\n-21-27-18\n19\n7\n7\n12\n-20\n1\n28\n37\n15\n12\n27\n27\n13\n-20\n13\n11\n26\n35\n-24\n17\n-28\n-13\n13\n8 -28\n-25\n8\n17\n29\n38\n3\n-3\n-25\n1\n17\n-19\n-5\n4\n-7\n15\n12\n-5\n-9\n0\n3\n-30\n3\n25\n11\n1\n-19\n-18\n-30\n-1\n8\n18\n-20\n0\n1\n15\n923\n4\n-6\nJ\nM\nA\nM\nJ\nJ\nA\nS\nF\n0\nM\nN\nA\nM\nD\nD\nPacific F(1+s) vs. dSST/dt\nAlantic F(1+s) vs. dSST/dt\nN lat\nN lat\n60\n60\n-21 -18 -13 -14\n55\n16\n15 -12 1 -6 -10\n55\n-15 -11\n3\n-7\n5\n-6\n50\n36 -9 23 -14\n50\n22 -21 -19\n88\n-12\n45\n12 -6 16 -17 122\n45\n0347-46\n-23\n40\n-26 -18\n3 -26 -26\n40\n10\n-16\n-27\n-19\n16\n35\n-16\n12\n10 -20\n35\n4\n3\n8\n5 -14\n30\n-9 -1 -20 -19\n30\n-6\n-7\n-8-16\n25\n25\n-9 -17 -22\n25\n20\n-2\n6\n-8\n39\n20\n27 -22\n15\n-28\n30\n80\n39\n16\n21\n15\n8\n13\n2\n10\n-24\n11\n19 -1-14\n10\n-18\n10\n2\n11\n1 -16\n5\n-15\n27432 13 -29\n5\n-27\n11\n-6\n12\n5 -23\n0\n11\n26\n8\n8 -10 12\n0\n-17\n-10\n10\n22\n10\n0\n8\n9 -22\n2\n4\n8\n-9\n0\n-5\n3\n30\n20\n40\n25\n-5\n-22\n-2\n3\n9\n00\nAC\n45\n28\n22\n25\n12\n-10\n25\n-10\n27\n4\n6\n-24-17\n17\n-10\n-5\n-25\n13\n-15\n23\n-15\n-27 -27\n-6\n19\n-20\n4\n-22\n28\n-20\n8 -14\n-25\n1\n20\n-25\n-10\n-29-15\n38-14\n40\n25\n-29\n-9\n-22\n-24\n-25\n-30\n-28\n-17\n-30\n-19\n-21\n-19\n-34\n-19\n-18\n-29\n-29\n12\n3\n4\nJ\nA\nD\nA\nS\n0\nN\nD\nFig. 1\nMIXED LAYER DEPTH FOR NORTH PACIFIC\nMIXED LAYER DEPTH FOR NORTH ATLANTIC\n400\n400\n350\n350\n300\n300\n250\n250\n200\n200\n150\n150\n100\n100\n50\n50\n0\no\nO\nF\nM\nA\nM\nJ\nA\nS\nO\nD\nJ\nJ\nF\nM\nA\nM\nJ\nA\nS\no\nN\nD\nSOLID - ESTIMATED. 4011-86N, 170E-130W\nROLID - ESTIMATED\n40N-6611.\n48W-61\nDOTTED - ESTIMATED. 20N-35N INOE-140W\nDOTTED - ESTIMATED. 20N-35N 65W-25W\nons. 40N-BRN 170E-130W FR. M. ROBINSON\nORS. IN-BRII, 46W-BW FR. P. LAMA. 1084\no ons, 20N-05N. 180E-140W FR. M. ROBINSON\nons. 20H-36N. RBW-25W FR. P. LAMB. 1984\nFig. 3\n194","FLUX DIFFERENCES [watts/m²]\nfor extreme N. Atl. SLP EOF 1 DJF 1950-'86\nB\nHATTS/METER\n-25\n-15\n-5\n0\n5\n10\n20\n30\n65\n60\n-\n90\n65\nstrong\n30\n60\n60\n30\n45\n40\n90\n30\nN LAT\n60\n35\nI\n30\n25\n20\no°\n30\nweak\n15\n30\n10\n5\nBOW\nLATENT\nHAITS/HETER==2\n-25\n-15\n-5 0 5 10\n20\n30\n85\n60\n30\n60\n90\n55\nstrong\nweak\n60\n45\n0\n30\n40\nN LAT\n35\n30\n25\no\no\n20\n0\n16\n10\n6\nSENSIBLE\nsaw\nFig. 2a\n195","FLUX DIFFERENCES [watts/m²]\nfor extreme N. Atl. SLP EOF 2 DJF 1950-'86\nWOTTS/HE IER==2\n-25\n-15\n-5 0 5 10\n20\n7\n-30\n65\n60\n65\n60\n90\n60\n-30\n30\n120\nweak\n60\n45\nstrong\n-30\n-30\n40\nN LAI\n0\n30\n35\n30\n30\n25\n20\n30\n30\n15\n10\nLATENT\n50W\n6\nWAITS/HETER==2\n30\n-25 -15 -5 0 5 10 20\n31\n65\n60\n65\nstrong\n60\n60\n30\n90\nweak\n45\n40\n0\nII LAT\n111111\n35\n30\n0\n25\n20\n15\n0\n10\nSENSIBLE\nsow\n6\nFig. 2b\n196","ASST\nDIFFERENCES (C/mo)\nAt\nfor extreme N. Atl. SLP EOF 1 DJF 1950 86\n.2\n6\n4\no\n4\n.2\n4\n0\n4\n4\n4\n-.2\n0\nsow\nfor extreme N. Atl. SLP EOF 2 DJF 1950 86\n4\n6\n.2\n2\n4\n2\n.2\n0\n-.2\n-.2\n-.2\n-.2\no\n50w\nFig. 2c\n1 97","Unstable modes in coupled ocean-atmosphere interactions in intraseasonal and interannual time scales\nK. - M. Lau\nLaboratory for Atmospheres\nNASA/Goddard Space Flight Center\nGreenbelt, MD 20771\nA. Hirst\nDivision of Atmospheric Research\nCSIRO, Aspendale, Victoria, Australia\n1. Introduction\nRecent studies with simple and intermediate coupled ocean-atmosphere models (e.g. Cane and Zebiak, 1986,\nSchopf and Suarez, 1988, Bettisti, 1988, Graham and White, 1988 and many others) indicate that these\nmodels are reasonably successful in simulating the recurring periods (3-4 years) of the El Nino. These\nmodels showed that the interplay between coupled air-sea interaction in the central and eastern tropical\nPacific and oceanic Rossby wave reflections from the western boundary are responsible for the interannual\ntime scale of the surface water warming in the eastern Pacific. However, none of these models were able to\nsimulate the eastward propagation of convection and sea surface temperature from the western Pacific such\nas those associated with the onset of the 1982-83 and the 1986-87 El Nino Southern Oscillation (ENSO).\nThe western Pacific in these models appears to play only a passive role. Moreover these models all seem to\npredict far too regular occurrence of ENSO than reality.\nOne of the key suspect for the above discrepancies is that all the above coupled models used essentially a\nsteady-state, Gill-type atmosphere whose state is completely determined by the model's sea surface\ntemperature (SST) anomaly field. Inherent in these models is the assumption that atmospheric time scales\nare much faster than that of the ocean and that the former will adjust rapidly to the relatively slow changes\nin the latter so that a steady state atmosphere is sufficient to capture all the coupled modes relevant to\nENSO. In essence, the steady state (dry) atmosphere used in most of the coupled models behaves as a\nslave to the ocean because in the absence of SST anomalies, it has no winds. However, this assumption\nmay be true only for a dry atmosphere. Yet, there are now an abundance of atmospheric observations\nand modeling results suggesting that a moist atmosphere is convectively active and possesses slow time\nscales (order of one to two months) even in the absence of SST anomalies (Lau and Chan, 1986, 1988,\nWeickmann, 1985 and many others). In fact, these slow time scales are ubiquitous in the moist tropical\natmosphere and there are significant overlaps between the slow time scales in the atmosphere and that of the\nequatorial ocean. Thus the basic assumption of separation of scales between ocean and atmosphere needs to\nbe re-examined.\nTwo issues are involved in considering the role of slow atmospheric motions (generally referred\nto\nas\nintraseasonal oscillations) in the coupled interaction between atmosphere and ocean. First, active\natmospheric components, in particular the higher frequency transients, may act as stochastic forcings which\ntend to obfuscate or even to alter the deterministic components of the coupled system largely governed by\noceanic dynamics (Lau, 1985). This will add a certain degree of uncertainty to the predictability of\nENSO. Although preliminary indication is that these stochastic forcings do not change the basic\ndeterministic characters in the long-term behavior of the Cane and Zebiak type coupled model (Zebiak,\n1989). Yet it is not clear that in a coupled model with a more realistic atmosphere, these effects are still\nsmall. Whether stochastic effects are important is still debatable. More important is the second issue which\nis of a more fundamental nature. Because of the overlapping or comparable times scales between\nintraseasonal oscillations and the gravest baroclinic modes of the equatorialocean (both of order of months),\nthe coupling of the two allows a non-equilibrium atmosphere and an ocean to mutually interact in the\ncoupling process and thus lead to a new branch of coupled modes. Consequently, the slave atmosphere-\nocean models only describe an incomplete set of coupled modes (cf. Lau, 1981, Hirst, 1986 and others).\nThis situation is illustrated schematically in Fig. 1 which shows the complete set of coupled ocean-\natmosphere modes in different time scales as a function of the coupling. The origin of the the atmosphere\nor ocean modes (at zero coupling) is denoted by the letters A and O. They represent the usual collection of\nuncoupled equatorial Rossby, Kelvin and mixed modes in the tropical atmosphere and ocean respectively.\n198","The excitation of unstable coupled modes at finite coupling are denoted by the coalesce of the atmospheric\nand oceanic branches. It is obvious that the slave-atmosphere coupled modes (referred to as Type II in Lau\nand Shen (1988)) are only a subset of unstable coupled modes coupled to a single atmospheric state. In\ncontrast, the non-steady atmosphere coupled modes (Type I) are those derived from a mutiplicity of\nintrinsic atmospheric modes associated with intraseasonal oscillations. In this paper, we aim at the\nidentification of unstable coupled modes in the two different branches shown in Fig. 1. A detailed\ndiscussion of the mechanism of the unstable modes will be presented in Lau and Shen (1988) and Hirst and\nLau (1989).\n3. Results\nAn instability analysis was carried out on a coupled system similar to that of Lau and Shen (1988) using\nthe procedure developed by Hirst (1986), with small ca* 3m/s) and Dm (=5-20 days damping) This\nis\nthe region of parameter space not explored by Hirst (1986) and is the region where the non-steady\natmosphere-ocean coupled modes shown in Fig. 1 are to be found. For clarity and maximum modal\nseparation, two limiting cases corresponding to different oceanic physics are discussed In the advective\nlimit, KT and KU are set to both zero and in the upwelling limit Tx is set to zero.\na. Advection limit\nFig. 2 shows the growth rate and frequency of the most unstable modes in Type I as a function of the\nevaporation parameter aE in the advective limit. The subscript A (O) indicates that the mode originates\nfrom the atmosphere (ocean) and therefore belong to Type I (Type II). At zero coupling, all waves have\nzero growth rate. As aT increases, the analytic continuation of the atmospheric Kelvin wave denoted by\nKA begins to develop a positive growth rate which reaches a value of (20 days)-1 at a typical value of aT.\nThis mode has a period of 43 days. The structure of the atmospheric and the oceanic component of KA\n(hereafter referred to as Mode A) is shown in Fig. 3a and b respectively. Both atmospheric and oceanic\ncomponents are dominated by Kelvin waves (strong zonal but weak meridional components). There is a\npositive (although weak) correlation between wind and current, thus energy is being transferred from\natmosphere to ocean via wind-stress work. Positive correlations are also evident between tropospheric\ntemperature and both the wind convergence field and SST anomaly. The atmospheric part of the mode\nshows a direct circulation with positive heating occurring in the region of warm air.\nAnother unstable mode is a Type II Rossby mode. However, the meridional structure of this mode\nbecomes increasing complicated as aE increases indicating the excitation of higher order oceanic\ncomponents. A mode with such complicated structure is likely be strongly modified by later diffusion and\nnonlinear advection in the real situation. Most likely this mode is not physical.\nb. Upwelling limit\nWith only upwelling and zero advection in the ocean, there are two unstable modes with simple meridional\nstructure in the small Ca* and low dissipation limit. The first is a Type I coupled mode which propagates\neastward with phase speed of about 3.0 m/s with a period of about 57 days. This structure of this mode\nresembles that of Mode A in the advection limit with Kelvin components dominating the atmospheric and\noceanic motion fields. The structure of the Type II unstable mode (Mode B) is shown in Figs 4 a and b.\nThis mode resembles the upwelling unstable mode of Lau (1981) and the large Ca-large Dm unstable mode\nof Hirst (1986). Mode B stems from an uncoupled oceanic thermal mode with low frequency in the range\n(1-3 years)-1\n4. Discussions\nAs a summary, the spatial structure of Mode A and Mode B in the equatorial X-Z plane are shown in Fig. 5.\nMode A, which is a Type I solution is a direct result of the destablization of an atmospheric mode by\ncoupling, represents a new class of unstable modes not previously known. This mode is favored by large\nnegative east-west SST gradient. The presence of this coupled mode suggests that there may be significant\nmodification of atmospheric intraseasonal oscillations by coupled air-sea interaction especially over region\nof negative SST gradient. On the other hand, Mode B can be identified with the slave atmosphere-ocean\n199","coupled unstable modes found in earlier studies (Lau, 1981, Philander et al 1984 and Hirst, 1986. They\nemerge as one of the two branches of unstable modes included in the present complete treatment of coupled\nmodes.\nAs found by earlier studies, Mode B is likely to be the dominant coupled modes responsible for the mature\nphase of ENSO when most of the action is taking place over the central and eastern Pacific and where\noceanic upwelling is large. It is possible that the relatively fast eastward movement of convection and SST\nanomaly (although weak) at the onset stage of ENSO may be related to Mode A or its hybrid modes. Mode\nA features enhanced convection over the western Pacific with strong surface easterlies over the central\nPacific: a favorable pre-condition for ENSO. One possible scenario is that as warm water keeps piling up\non the western Pacific as a result of some low frequency oceanic basin-wide adjustments to the strong\nsurface easterly phase (positive swing) of the Southern Oscillation, Mode A instability could be excited.\nAs a result, convection and and warm water rapidly migrates eastward, easing the large scale SST gradient\nA modal transition may then occur with Mode B becoming increasing important during the later stages of\nENSO.\nFigure Captions\nFig. 1 Schematic representation of the frequency interaction of atmospheric and oceanic modes with different\ntime scales as a function of the coupling.\nFig. 2 Growth rates and frequencies of several modes in the advection limit as a function of aT for DM=0\nand Ca*=2ms-1. Modes are analytical continuations of the atmospheric and oceanic Kelvin waves (KA and\nKo respectively) and the atmospheric and oceanic n=1 Rossby waves (R1A and R10, respectively).\nWavelength is 15,000 km. Growth rate and frequency have units of year\nFig. 3 Structure of destablized (Type I) ex-atmosphere Kelvin mode at standard at in the advection limit\nfor Dm=0 and Ca*=2 m/s. (a) Solid and dashed lines indicate positive and negative SST anomaly contours\nat 0.05 K intervals; arrows indicate wind vectors; T+ and T- and 0+ and 0- indicate locations of largest\npositive and negative perturbations of SST and atmospheric temperature (i.e. surface pressure), respectively.\n(b) Solid and dashed lines indicate positive and negative h anomaly contours at 4 m intervals; arrows\nindicate current vectors; T+ and T- are as in (a)\nFig.4 Same as in Fig. 3, except for Mode B in the upwelling limit.\nFig. 5 Zonal structure along the equator for Mode A and Mode B. Solid line indicates ocean-atmosphere\nboundary, dashed line indicates upper layer depth. T+ and T- indicate location of SST anomaly extrema, 0+\nand 0- for mid-tropospheric temperature. Horizontal arrows indicate zonal wind and current direction, and\nvertical arrows indicate location of maximum vertical motion.\n200","IA\nA\nSynoptic\nA\nand meso\nscales\nA\nA: Atmosphere\nO: Ocean\nIntraseasonal\nnon-steady atmosphere\noscillations\ncoupled modes\nBaroclinc oceanic\nslave-atmosphere\nmodes\ncoupled modes\nSteady state\natmosphere\nFig 1\nincreased coupling\nMODEL II\nD.: = 2 m s-1\nP==\n40\nR10\nKA\n20\n0\nR1 A\n-20\nK\n40\n10\nKA\n5\nK\n0\nR10\n-5\nR1\nA\n- 10\nFig. 2\n2\n3\n4\n0\n1\n(10-4 m² s-3 K-1)\naE\n201","MODEL II\nmode A\nCa = 2 m s' -1. - D = 0\nM\n2.5 m s-1\nSST, WIND, PRESSURE\n10°N\n0\n(a)\nT+\n0°\n-T\nO+\n10°S\n-6000 -4000 -2000\n0\n2000\n4000\n6000\n(km)\nX\n0.6 m s-1\nh, CURRENTS\n10°N\n0\n(b)\n0°\nT +\nT\no\n10°S\n- -6000 - -4000 -2000\n0\n2000\n4000\n6000\n(km)\nX\nFig. 3\n202","MODEL IV\nmode B\nC& = 2 m s-1 D = 0\nM\n3.0 m s-1\nSST, WIND, PRESSURE\n10°N\n(a)\n0°\ne\n+\n10°S\n-6000 -4000 -2000\n0\n2000\n4000\n6000\nX (km)\nh, CURRENTS\n1.0 m s-1\n10°N\n(b)\n0°\n10°S\n-6000 -4000 -2000\n0\n2000\n4000\n6000\nX (km)\nFig. 4\n203","Model IV\nZ\nMODE A\nx\nO+\n0\nT\nT+\nT-\nModel IV\nZ\nMODE B\nx\n0+\n0-\nT+\nT\nT-\nFeg.\n5\n204","Temporal Variability of the Thermohaline Structure of the North Atlantic Ocean\nSydney Levitus\nGeophysical Fluid Dynamics Laboratory/NOAA\nP.O. Box 308\nPrinceton, NJ 08542\nABSTRACT: Historical observations of temperature and salinity for the North\nAtlantic Ocean have been composited by one-degree squares and objectively\nanalyzed at standard depth levels for two pentads, 1955-59 and 1970-74.\nStatistically significant changes in the thermohaline structure of this ocean\n(of gyre and basin-wide length scales) have been found at depths extending from\nthe sea surface through 2000 m.\nThe permanent pycnocline of the subtropical gyre rose on the order of\n25-50 m from the earlier to the later pentad (Fig. 1). This resulted in a\ncooling of approximately 0.5°C and freshening of approximately 0.025 per mille\non constant depth surfaces in this region in the 500-1200 m depth range\n(Figs. 2-3). These results are in agreement with the variability observed from\ntime series data taken at the Panulirus Station which is located in this gyre\njust offshore of Bermuda (32°N, 64°W) (Roemmich, 1985). Hydrographic stations\nhave been taken at this location on a monthly or semi-monthly basis since 1954.\nThe shoaling of the pycnocline is responsible for a decrease in steric sea\nlevel of approximately 10 centimeters between the pentads. This is in\nagreement with a decrease in observed sea level measured from a tide gauge\nstation at Bermuda (Levitus, 1989d).\nThe subarctic gyre also exhibits statistically significant changes. In\nparticular, a fresh (0.5 per mille) anomaly occurred in the upper portion of\nthis gyre beginning in the late 1960's (Levitus, 1989b). Dickson et al. (1988)\nhave described this cold, fresh anomaly based on time series from ocean weather\nstations and coastal locations.\n205","The Upper North Atlantic Deep Water (UNADW) as characterized by difference\nfields at 1750 m depth exhibited a warming (approximately 0.1°C) and an\nincrease in salinity ( . 025 per mille) (Levitus, 1989c). This is consistent\nwith observations from Ocean Weather Station Bravo located in the Labrador Sea\nthat deep convection at this location ceased during the late 1960's and hence\nthe supply of relatively cold fresh water to deeper depths (e.g. 1500 m) was\ninterrupted (Lazier, 1980). This led to a warming and freshening of the UNADW.\nLabrador Sea Water is one of the two major components of UNADW.\nThe changes in temperature and salinity below 1300 E depth are nearly\ndensity compensating at all standard levels. Changes in in situ density are of\nthe order 0-2 X 10 5 g cm-3.\n2 06","References\nDickson, R.R., J. Meincke, S.A. Malmber, and A.J. Lee, 1988: The Great\nSalinity Anomaly in the Northern North Atlantic, 1968-82. Progress in\nOceanography, 20, 103-151.\nLazier, J.R.N., 1980: Oceanographic conditions at Ocean Weather Ship Bravo,\n1964-74. Atm.-Ocean, 18, 227-238.\nLevitus, S., , 1989a: Interpentadal Variability of Temperature and Salinity at\nIntermediate Depths of the North Atlantic Ocean, 1970-74 Versus 1955-59.\nJournal of Geophysical Research-Oceans, 94, 6091-6131.\nLevitus, S., 1989b: Interpentadal Variability of Salinity in the Upper 150 m\nof the North Atlantic Ocean, 1970-74 Versus 1955-59. Journal of\nGeophysical Research-Oceans, in press.\nLevitus, S., 1989c: Interpentadal Variability of Temperature and Salinity in\nthe Deep North Atlantic, 1970-74 Versus 1955-59. Journal of Geophysical\nResearch-Oceans, accepted.\nLevitus, S., 1989d: Interpentadal Variability of Steric Sea Level and\nGeopotential Thickness of the North Atlantic Ocean, 1970-74 Versus 1955-59.\nJournal of Geophysical Reearch-Oceans, submitted.\nRoemmich, D., 1985: Sea level and the variability of the oceans, 1985. In:\nGlaciers, ice sheets, and sea level: Effect of a CO2 induced climatic\nchange. DOE/ER/GO2 35-1, 104-115.\n207","26.0\n26.0\n-175\n26.5\n26.5\n50\nO'\nOB\n27 0\n0\n27.0\n25\n100\n25\n50\n25\n25\n27.5\n27.5\nEQ\n10N\n20\n30\n40\n50\n60\n70N\nFig. 1\nDepth difference (m) of potential density surface along 49.5°W as a\nfunction of potential density for 1970-74 minus 1955-59. Shading\nindicates negative values.\n80°N\n0.6\n70°\n60°\n-0.6\n0.4\n-0.4\n50°\n0.6\n1.2\n0.8\n1.0\n40°\n0.8\n-0.6\n30°\n-0.4\n-1.2\n0.0\n20°\n10°\nEQ\n100°W\n90°\n80°\n70°\n60°\n50°\n40°\n30°\n20°\n10°\n0°\nFig. 2\nTemperature difference (°C) for 1970-74 minus 1955-59 at 500 m depth.\nShading indicates negative values.\n208","80°N\n70°\n60°\n0.025\n-0.05\n-0.05\n50°\n0.00\n-0.025\n0.1\n0.075\n-0.15\n40°\n-0.025\n-0.1\n0.05\n30°\n0.00\n20°\n0.00\n0.05\n10°\nEQ\n30°\n20°\n10°\n0°\n80°\n70°\n60°\n50°\n40°\n100°W\n90°\nSalinity difference (psu) for 1970-74 minus 1955-59 at 500 m depth.\nFig. 3\nShading indicates negative values.\n80°N\n7\n0.3:\n70°\n-0.5\n0.2\n0.0\n60°\n-0.0\n-0.21\n-0.3:\n50°\n-0.5\n0.5\n40°\n0.3\n-0.1\n0.0\n0.1\n***\n30°\n0.2\n0.2\n20°\n(())\n10°\n0\n-0.5\nEQ\n40°\n30°\n20°\n10°\n0°\n100°W\n90°\n80°\n70°\n60°\n50°\nSalinity difference (psu) for 1970-74 minus 1955-59 at the sea surface.\nFig. 4\nShading indicates negative values.\n209","80°N\n70°\n60°\n0.15\n50°\n0.15\n0.00\n-0.20\n0.25\n40°\n-0.1\n0.10\na\n0.15\n0.00\nD\n0.05\n30°\n0.20\n0.10\n0.20\n20°\n0.05\n0.05\n10°\nEQ\n100°W\n90°\n80°\n70°\n60°\n50°\n40°\n30°\n20°\n10°\n0°\nFig. 5\nTemperature difference (°C) for 1970-74 minus 1955-59 at 1750 m depth.\nShading indicates negative values.\n80°N\n70°\n60°\n50°\n0.025\n0.00\n-0.025\n40°\n0.00\n0.025\n30°\n0.0\n20°\n10°\nEQ\n100°W\n70°\n90°\n80°\n60°\n50°\n40°\n30°\n20°\n10°\n0°\nFig. 6\nSalinity difference for 1970-74 minus 1955-59 at 1750 m depth.\nShading indicates negative values.\n210","The Development and Utilization of an AVHRR-Based\nVegetation Index for Climate Monitoring\nMichael Halpert, Phillip Arkin, and Chester Ropelewski\nClimate Analysis Center\nNMC/NWS/NOAA\nand\nRobin Tomlinson\nResearch and Data Systems Corporation\nA normalized difference vegetation index (NDVI) based on\nweekly digital data provided by the Advanced Very High Resolution\nRadiometer (AVHRR) on the NOAA operational polar-orbiting\nsatellites has been produced for the years 1985-1988. The\ndigital data are Global Area Coverage (GAC) at a resolution of 4\nkm, which is then sampled to 16 km. These data are combined into\na 1° latitude by 1° longitude grid. The vegetation index for any\n16 km pixel is computed by subtracting the reflectance in the\nvisible channel from the near infrared (IR) channel and then\nnormalizing by the sum of the channel reflectances. Since green\nvegetation is highly absorbent to incoming radiation in the\nvisible part of the spectrum and relatively reflecting in the\nnear IR, this process actually produces an index of the greenness\nof the vegetative surface. The CAC product consists of the mean,\nmedian, standard deviation, number of observations and a seven\nclass histogram for each 1 6 by 1° grid square. Data is archived\nbetween 55°S and 75°N.\nThere are several data problems associated with the NDVI.\nBecause the calibration coefficients are determined from linear\nregressions, low solar zenith angles or low illumination can\ncause channel radiances to appear negative. Therefore, the\nindividual channel data were examined and negative pixels were\neliminated. Pixels with negative vegetation index values were\nalso removed, since a negative NDVI would indicate greater\nreflectance in the visible than the near IR (which is not what is\nexpected from surface vegetation).\nThe largest source of data error is cloud contamination.\nAlthough the digital data are cloud screened by retaining the\nlargest NDVI value for a given week, cloud contaminated data\nstill exist in the data set. The AVHRR infrared window channel\nis used to further screen for clouds by eliminating pixels whose\ntemperature values are less than -5.0°c. Another source of data\ninconsistency results from the slow degradation of the optics of\nthe sensors. This degradation is usually less than 5% per year\nand is similar for both sensors. Thus, errors caused by this\nshould be small in the normalized difference.\nAnother source of data problems arises from the viewing\ngeometry associated with different scan angles. The AVHRR has\nscan angles which range from 55. o to -55.4° relative to nadir.\n211","NDVI values calculated from near nadir are generally lower than\nthose values from off-nadir positions. In addition, values from\nthe back-scatter direction (looking away from the sun) are lower\nthan values obtained from the forward direction. These\ninconsistencies can result in fairly large week-to-week\ndifferences. Due to the various problems mentioned above,\ntemporal filtering is often used to smooth the data. Figure 1\nshows three time series for 1988 over a 5° latitude by 5\nlongitude area of the midwestern United States. The top figure\nis unfiltered, the middle figure uses a three-week median filter,\nand the data in the bottom figure were smoothed with a three-week\nmaximum filter. The unfiltered time series shows the large week-\nto-week variations associated with scan angle differences. Both\nthe median and maximum filtered time series appear smoother, but\non closer inspection two large spikes in the maximum filtered\ndata set appear during weeks 31 and 35. These two large values\nare then maintained for the next two weeks. Since these weeks\ncorrespond to the U. S. drought of 1988, we suspect that these\ntwo points are inconsistent with the neighboring weeks.\nIn order to examine the spatial consistency of the NDVI, we\nproduced maps of the vegetation index over the United States for\n1987 and 1988. Difference maps for late June (Fig. 2) show large\nindex differences in the northern Great Plains and in Oklahoma,\nNebraska and Texas, but little difference between the two years\nin the corn growing regions of the Midwest. An examination of\nthe spring precipitation patterns for 1987 and 1988 reveals that\nprecipitation was deficient in this area during both years.\nHowever, comparisons of the June precipitation show normal to\nabove normal precipitation during 1987, but very dry conditions\ncontinuing in 1988. The difference map for late July (Fig. 3)\nshows that the largest vegetation differences between 1988 and\n1987 are now centered in Indiana and Illinois. This corresponds\nwell to the precipitation patterns, as well as to the known crop\nconditions during these two periods.\nA weekly vegetation index data set has been produced for the\nyears 1985-1988 on a 1° latitude by 1 o longitude grid. Despite\ncontinuing uncertainties with cloud contamination and scan angle\ndifferences, large-scale climate parameters have been identified.\nFurther work on eliminating these problems should produce an\nindex that can be used for global climate monitoring, as well as\nthe diagnostic study of climate variations.\n212","0.5\nPBP\n0.4\nPo\n0.3\n0.2\nOBQ\nto\n&\n0.1\n0\n-0.1\n23\n25\n27\n29\n31\n33\n35\n37\n39\n41\n43\n45\n47\n49\n51\n1\n3\n5\n7\n9\n11\n13\n15\n17\n19\n21\n0.5\n00g\n0.4\n0.3\n0.2\nDO\ns\nBed ge (000\n0.1\n0\n-0.1\n43\n45\n47\n49\n51\n1\n3\n5\n7\n9\n11\n13\n15\n17\n19\n21\n23\n25\n27\n29\n31\n33\n35\n37\n39\n41\n0.5\nFor e 86\n0.4\n0.3\n0.2\nso\ntoo\n0000\n0.1\n0\n-0.1\n1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45 47 49 51\nWEEK\nFig. 1. Time series of the 1985-1988 mean (0), 1988 total (+), and the 1988\ndifference (0) for the normalized difference vegetation index (NDVI). Data\nis averaged over an area from 38°N-43°N and from 90°W-95°W. The top figure\nis unfiltered, the middle uses a three week median filter, and the bottom\nuses a 3 week maximumn filter.\n213","122W\n10042 4 6\n80W\n2\n4\n0\n4\n2\n-4\n0\n50N\nSON\n2\n0\n0\n-2\n-2\n0\n40N\n40N\n30N\n30N\n-2024\n6\n-4\n'6\n-4\n0\n-d\n-8\n2\nL-8\n00\nI\nP\n20N\n20N\n120W\n-8\n100W\n-8 -8\n80W\n0\nFig. 2. NDVI difference (X 100) between late June 1988 and late June 1987.\nThree week median filters were applied to the total fields before filtering.\n214","BOW\nV\n22\n246 120W\n8642\n100W\n02\n4\n50N\n50N\n0\n2\nX-4\n4\n6\n4\n-6\n2\n10\n40N\n40N\n2\n2\n-2-2\n2\n2\n2\n30N\n30N\n-6\n-2\nL2\n0\n-4\n-2\n-2\n-2\n0\n4\n-0\nr2\n2\nDO 2\n4\n20N Fig. difference 120W (X 100) between late July 1988 fields and late before June 1987.\n20N\n2\n-2\nBOW\n4\n100W\nThree 3. week NDVI maximum filters were applied to the total\nfiltering.\n215","Interannual variability and trends indicated by OLR\nMuthuvel Chelliah and Phillip Arkin\nClimate Analysis Center, NMC, NWS, Washington D.C.\n1. Introduction\nRecently Nitta and Yamada (1989) (hereafter referred to as NY) exam-\nined annual to decadal-scale variations in sea surface temperature (SST)\nand concluded that tropical SST, especially in the central and eastern\nPacific and in the Indian ocean, has been increasing since the late 1970's.\nThey also claimed that, consistent with the warming in SST, the tropical\natmosphere has exhibited enhanced convective activity as detected by outgo-\ning longwave radiation (OLR) They used empirical orthogonal functions\n(EOFs) to analyze smoothed (3 month running mean) monthly anomalies of 38\nyears (1950-1987) of SST data and nearly 13 years (1974-1987) of OLR data.\nThe basis for their conclusion, that convective activity has increased, was\nthe appearance of a trend-like character in the time series associated with\nthe second unrotated eigenmode of tropical OLR. In this paper, we will\ndescribe the preliminary results of our study into the interannual varia-\ntions in tropical OLR, and try to identify any trends and their causes. In\nparticular, this study is an attempt to further explore the interannual\nvariability and trend indicated by OLR data beyond NY's study.\nThe methodology is as follows. We have referred to NY's EOF analysis\nof OLR anomalies as analysis A. We duplicated the results of NY, to the\nclosest extent possible, and called the resulting analysis B. We then\nanalyzed an extended data set and referred to this as analysis C. Then we\nhave shown that the results of analysis B is the same as that of A. The\nresults of analyses B and C are then subjected to further mathematical\nanalysis. Intercomparison of these extended analyses suggests that the\nperception of a trend towards enhanced tropical convection in the late\n1970's and early 1980's in the OLR data seems to have resulted from changes\nin satellite observation times. In section 2, we will describe analyses A\nand B and establish their similarity. In section 3, we will describe analy-\nsis C and compare it with analysis B both in their initial and extended\nform. Our preliminary conclusions are given in section 4.\n2.\nInitial analysis\nNY applied EOF analysis to the OLR data on a 5°X5° grid between 25°N\nand 25°: S over the period July 1974 through March 1987 (January 1978-March\n1979 is excluded) to detect annual to decadal-scale temporal and spatial\nvariations of tropical convection. There are a total of 138 months in their\nanalysis period. They performed EOF analysis on the 3-month running mean of\nthe monthly OLR anomalies, which are deviations from the 12- or 13-year\nmean OLR for the same month. Their EOFs are based on the correlation matrix\nand are non-rotated. Figure 1, taken from NY (their Fig. 4), shows the\nfirst two eigenmodes of OLR and their time coefficients. In analysis B we\nhave attempted to repeat NY's EOF analysis (called analysis A here), al-\nthough for a smaller region (20°N - 20°S) due to computer storage limita-\ntions. Examination of the spatial loading patterns for the first four\neigenmodes and their time coefficients for analysis B (not shown) reveal\nthat eigenpatterns 1 and 2 (and the time series) are, as expected, almost\nidentical to that of NY. The close correspondence between the two analysis\nassured us that we have successfully reproduced NY's results.\nAs NY point out, mode 1 exhibits a large east-west dipole pattern in\nthe equatorial Pacific, a characteristic of the ENSO phenomena, with large\namplitudes in the time series corresponding to the 1976, 1982 and 1986 warm\nEl\nNino episodes. Note that the 1982 ENSO event dominates the time series.\n216","Examination of the time series associated with mode 2 indicates, as noted\nby NY, a possible trend in the OLR data set with consistently negative\ncoefficients in the 1980s and positive coefficients in the 1970s. However,\na closer look at the mode 2 spatial pattern suggests that the confounding\nof two entirely different signals might be responsible for the appearance\nof a trend in the data. The first seems to be related to the Southern\nOscillation and, which we call the SO signal, has maximum positive loadings\nnear and east of the date line in the equatorial Pacific the Indian Ocean.\nThis pattern combined with large negative weights in the time series for\nthe northern winter of 1982/83 indicates to us that this mode is mostly\nassociated with the extraordinary warm ENSO event of 1982/83.\nOn the other hand, the second signal appears to be related to changes\nin satellite measurement times between the 1970s and 1980s. Because of the\nchanges in OLR observation time relative to the phase of the diurnal cycle\nin surface temperature, this signal will be maximized over the tropical\nland masses. We believe that the maximum loadings over Africa and south\nAmerica can be attributed to this signal. This mixing of patterns (signals)\namong various eigenmodes in unrotated eigenmode analysis is a result of\northogonality constraints and can lead to serious problems in interpreta-\ntion of the results (Richman, 1981).\n3.\nExtended analysis\nWe performed our analysis (C) over the region 30°S-30°N and covering\nthe period from June 1974 through March 1989 (excluding calendar year\n1978) A total of 166 months were included in the analysis. Principal\ncomponent analysis (PCA) was performed on monthly deviations of OLR, where\nthe deviations were computed from a 10 year (1979-1988) monthly OLR clima-\ntology. Note that our analysis period included the recent, major 1988/89\ncold ENSO event (Arkin, 1989). The PCA was performed on OLR anomalies on a\n10° lon X 5° lat grid. The explained variances of the leading 4 modes of\nanalysis C are uniformly lower than those of analysis B because analysis C\ncovers a bigger region and includes more months (166 versus 138) and hence\nhas to account for more degrees of freedom. Comparison of analyses B (or A)\nand C over the common domain 20°S-20°1 indicates that any reasonable corre-\nspondence between them exists only in the first mode, beyond which mode by\nmode comparison is very difficult. Most importantly, the mode 2 of analysis\nB (or A), associated with a trend in NY's analysis, does not have an exact\ncounterpart in our analysis C.\nBecause unrotated EOF (or PCA) analysis, by design, maximizes the\nvariance explained over the entire analysis domain for each eigenmode, we\napplied a varimax rotation (Richman, 1986; Barnston and Livezey, 1987) to\nthe principal components of analyses C and B. Estimation of sampling errors\nassociated with the eigenmodes (North et al., 1982) suggested that the\nleading four modes are reasonably well separated from each other and the\nremaining modes in both analyses B and C, leading us to rotate the first\nfour modes in each analysis. Figure 2 displays the rotated eigenpatterns\nand Fig. 3 shows the associated time series for the first four modes of\nanalysis C. After rotation, the leading 4 rotated modes of analysis C (Fig.\n2) compare very well to the top 4 rotated modes of analysis B (not shown)\nThis similarity of all four modes, between the two analyses, in spite of\nthe differences in the analysis period, analysis region and climatology,\nseems to confirm the validity of the signal contained in each of these\nmodes, and illustrates the effectiveness of the rotation in bringing them\nout, despite the differences noted earlier in the unrotated analyses. Also,\nwhen optimum number of modes are chosen, rotation does not disrupt or\nregionalize truly global scale phenomena, such as represented by mode 1.\n217","Due to the dominance of a single peak in the mode 1 time series, and the\nlocation of the maximum loadings in the equatorial Pacific east of the\ndateline and south of the equator, along with similar (in sign) loadings in\nthe southern Indian Ocean, we believe that the rotated mode 1 pattern is\nalmost exclusively associated with the major 1982/83 ENSO warm event. The\nspatial pattern of rotated eigenmode 2 describes the spatial characteris-\ntics of convection anomalies associated with a typical or canonical ENSO\nevent (Rasmussen and Carpenter, 1982) in the equatorial Pacific. Note that,\nunlike mode 1, maximum loadings in the Pacific are centered on the equator\nand on the dateline. This spatial pattern seems to represent the typical\nconditions during both warm and cold events. Indicated in the time series\nfor mode 2 are the major warm (W) and cold (C) ENSO events in the last 15\nyears.\nThe spatial patterns associated with rotated eigenmode 3 has maximum\nloadings over Africa, and with half the amplitude and opposite sign over\nthe equatorial Indian ocean, and even lower amplitudes elsewhere. Specifi-\ncally, no organized pattern with large loadings are found in the tropical\nPacific. This mode represents the second of the two signals discussed\nearlier in section 2. The time series associated with this third mode\nsuggests a systematic difference between the two decades (1970s versus\n1980s) However, changes in the time series appear to be associated most\nstrongly with changes in local observation times. That is very abrupt\nchanges in the unsmoothed time series accompany each satellite change. It\nshould be noted that the OLR data set that we (and NY) used, for the period\nfrom June 1974 through March 1989 has been adjusted (corrected) for the\nvarious instrument (radiometer) changes (Gruber and Krueger, 1984; Janowiak\net al. 1985), but not for changes in local measurement times.\nWhile the spatial loading patterns (and hence the time series) for\neigenmodes 1, 2 and 3 were quite robust (independent of the number of modes\nrotated), eigenmode 4 was less stable. i.e., the spatial loading pattern\nfor mode 4 changes continuously as higher and higher modes are included in\nthe rotation. Hence, even though the time series associated with mode 4, as\nshown (modes rotated = 4), does exhibit a trend that delineates the decade\nof 1980's from the 1970's, it is difficult to interpret the spatial loading\nThis situation leads to the following question. If there is a\npattern.\ntrend in the OLR data, what is it due to ? Is it due to a real climate\nsignal ? or Is it due to noise from the observation system ? The foregoing\nanalysis indicated that there is no trend associated with the first 3\nmodes. This is further confirmed by the absence of trend-like character in\nthe time series of OLR anomalies, reconstructed using the first 3 modes\nalone (not shown), separately over the tropical Pacific and over the global\ntropical belt. In other words, there is no decadal-scale trend in OLR data\n(1974-89) associated with ENSO type phenomena. Further research is needed\nto separate the signal from the noise in OLR data in the remainder of the\nmodes (modes > 3) and then investigate the real cause of the trend.\n4.\nSummary\nAn investigation has been made of monthly OLR anomalies from 1974\nthrough 1989 (includes 3 warm and 3 cold El niño episodes), using rotated\nprincipal component analysis (RCPA) While the presence of a trend in\nconvective activity in the tropics, as indicated by OLR, cannot be ruled\nout, our preliminary results indicate that there does not appear to be any\ntrend associated with ENSO type phenomena. Rotation of the leading 4 modes\nenables identification of one of the mode's time series with the changes in\nlocal observation times in various satellites. Further research is under-\nway.\n218","O.L.R. .\n(\n3-MONTH\n)\nFACTOR LOADING\nP.C.A. NO 1\n0\n20E\n40E\n60E\n80E\n100E\n120E\n140E\n-\n160E\n180\n160W\n40N\n140W\n120W\n100W\n80W\n60W\n40W\n20W\n0\n40N\n20N\n-19\n20N\n+\n4\nor\n8\na\nUA\nEQ\n09\nEQ\n+\nCO\n78\nor\n10\n+\n20S\n-\n+\n60\nr\n02\n1\no\n20\n+\n20S\nQ\n40S\n20E 40E 60E 80E 100E 120E 140E 160E 180 160W 140W 120W 100W 80W 60W 40W 20W\n0\n40S\n0\n0. L. . R.\n( 3-MONTH )\nFACTOR LOADING\nP.C.A. NO. 2\n0\n20E\n40E\n60E\n80E\n100E\n120E\n140E\n160E\n180\n160W\n140W\n120W\n100W\nBOW\n60W\n40N\n40W\n20W\n0\n2\n40N\n20N\n0\n20N\n(too\na\nin\nEQ\nco\n20.\nF.\nEQ\n+\n02\nCo\n10\n+\n+\nd\n20S\n<\n4\n-\n00\nON\n20S\n40S\n0\n20E\n40E\n60E\n40S\n80E\n100E\n120E\n140E\n160E\n180\n160W\n140W\n120W\n100W\n80W\n60W\n40W\n20W\n0\n(a)\nO.L.R. ( 3-MONTH )\nP.C.A. AMPLITUDE\n1974\n1975\n1976\n1977\n1978\n1979\n1980\n1981\n1982\n1983\n1984\n1985\n1986\n1987\n40\n20\nP.C.A.\n0\nNO. 1\nARA\n-20\n20\n-40\nP.C.A.\n10\nNO. 2\n0\n-10\n-20\n1974\n1975\n1976\n1977\n1978\n1979\n1980\n1981\n1982\n1983\n1984\n1985\n1986\n1987\n(b)\nList of Figures\nFig. 1\nThe spatial loading patterns and the associated time series for\nthe leading two unrotated EOF modes from Nitta and Yamada (1989)\n- Analysis A (here)\n219","ANALYSIS C - SPATIAL PATTERN - ROATATED MODE 1 - MAX. MODES ROTATED = 4\n30N\n0.2\naiy\n20N\n0.2\n10N\n10/03\n0.0\n0.0\n0.0\nEQ\n0.2\nQ.8\n10S\n0.2\nQ.6\n0.4\n20S\n0.4\n0.2\n30S\n0.2\nOE\n30E\n60E\n90E\n120E\n150E\n180E\n150W\n120W\n90W\n60H\n30W\nOE\nANALYSIS C - SPATIAL PATTERN - ROATATED MODE 2 - MAX. MODES ROTATED 0.10 = 4\nCONTOUR INTERVAL\n30N\n0.2\n0.2\n20N\n10N\n0.0\n0. u\nEQ\n0.6\n10S\n0,00\n20S\n0.2\n30S\nOE\n30E\n60E\n90E\n120E\n150E\n180E\n150W\n120W\n90W\n60W\n30W\nOE\nCONTOUR INTERVAL\nANALYSIS C - SPATIAL PATTERN - ROATATED MODE 3 - MAX. MODES ROTATED 0.10 = 4\n30N\n0.6\n20N\n0.0\n(0.2\n10N\n0.0\nEQ\n0.2\n0.0 0.0\n105\n0243\n0.2\n0.4\n0.2\n01\n20S\n0.2\n0+0\n30S\nOE\n30E\n60E\n90E\n120E\n150E\n180E\n150W\n120W\n90W\n60W\n30W\nOE\nCONTOUR INTERVAL\nANALYSIS C - SPATIAL PATTERN - ROATATED MODE 4 - MAX. MODES ROTATED 0.10 = 4\n30N\nONE\n20N\n0.00\n1\n0.\n0.0\n0.0\n10N\n0.2\n0.2\nEQ\n0.2\n(10\n0.2\n02\n10S\n0.0\n20S\n0.2.3.\n30S\nOE\n30E\n60E\n90E\n120E\n150E\n180E\n150W\n120W\n90W\n60W\n30W\nOE\nCONTOUR INTERVAL-\n0.10\nFig. 2 The spatial loading patterns for the leading four rotated principal\ncomponents of Analysis C.\n220","ANALYSIS C - TIME SERIES\nROTATED MODE NUMBER\n1\n200\n100\nmyham\n0\n-100\n-200\n1974\n1975\n1976\n1977\n1978\n1979\n1980\n1981\n1982\n1983\n1984\n1985\n1986\n1987\n1988\n1989\nYEAR\nANALYSIS C - TIME SERIES -- ROTATED MODE NUMBER 2\n200\n100\n0\nW\n-100\n-200\nYEAR\nANALYSIS C - TIME SERIES\nROTATED MODE NUMBER 3\n--\n200\n100\nMr\nMMM\nMy\n0\nTIROS\nNORA\nNOAA 2,3,4,5\nNO AA 7\n9\nNOAA\nNOAA 6\nN\n11\n-100\n4\n.\n4\n9:00 AM\n2:30\n3:30cM\nAM\n7:30\nM\n2:30\nAM\n2:30dm\n9:00P.M\n3:30'M\n7:30\nPM\nPM\n2:30\nPM\n2:30PM\n2:\n-200\n1974\n1975\n1976\n1977\n1978\n1979\n1980\n1981\n1982\n1983\n1984\n1985\n1986\n1987\n1988\n1989\nYEAR\nANALYSIS C - TIME SERIES -- ROTATED MODE NUMBER 4\n200\n100\nwas\nMango\n0\n-100\n-200\n1974\n1975\n1976\n1977\n1978\n1979\n1980\n1981\n1982\n1963\n1984\n1985\n1986\n1987\n1988\n1989\nYEAR\nThe time series associated with the leading four principal\nFig. 3\ncomponents of Analysis C.\n221","Verification of Satellite-Based Rainfall\nEstimates for Climate Models\nMark L. Morrissey\nDepartment of Meteorology\nand\nJ. Scott Greene\nGeography Department\nUniversity of Hawaii\nHonolulu, Hawaii 96822\nCurrent climate models require accurate area-averaged rainfall estimates in the\ntropics with resolutions of 5 days to 1 month and 75,000 km2. At present, rainfall estimates\nmade from satellite-based algorithms have not been validated over the various tropical\nclimatic regions. This is due to insufficient samples of area-averaged ground-truth rainfall,\nwhich are also used for the development, testing and calibration of these algorithms.\nAlthough area-averaged 'ground-truth' rainfall (defined as an estimate having minimal\nsampling error) has traditionally been acquired through the use of high-density raingauge\nnetworks, these are absent over the open ocean. However, widely separated raingauges are\npresent within many tropical oceanic climate regimes. A new and unconventional statistical\nprocedure using sparsely distributed raingauges is introduced that produces area-averaged\nground-truth rainfall estimates (AAR) for comparison with satellite information. The\ntechnique can be used to develop, test and calibrate satellite-based rainfall algorithms and\nverify rainfall estimates made from these algorithms.\nVarious satellite-measured parameters are thought to be related to rainfall indirectly\n(i.e. using infrared and/or visible wavelengths) or directly (i.e. microwave wavelengths).\nExamples of common infrared/visible satellite parameters are the number of highly\nreflective clouds and the fractional coverage of cold cloud within a given sized area.\nRecently, microwave iiradiances obtained from the Special Sensor Microwave/Imager\n(SSM/I) have been shown to be directly related to rainfall fluctuations.\nUntil now, AAR obtained from the few high-density raingauge networks or from\ngauge-adjusted radar provided ground-truth area-averaged rainfall. Each raingauge within\nthe network was assumed to accurately represent a small area surrounding the gauge (the\nsize of this area, known as the catch, depends on the rainfall's spatial scale). The gauges\nwere spaced close enough so that the small areas were contiguous. Ground-truth AAR was\nproduced by simply spatially averaging the raingauge measurements over a particular time\ninterval. The accuracy of AAR could be increased through the use of various weighting\nschemes. Satellite-based rainfall algorithms were developed, tested and calibrated by\nstatistically comparing ground-measured AAR values to satellite information corresponding\nto the location of the area represented by the raingauges.\nSince high-density raingauge networks (hereafter referred to as \"contiguous\"\nnetworks) are not available over the world's oceans, we propose the \"Non-Contiguous\n222","Raingauge\" procedure which utilizes widely scattered raingauges to compare ground-truth\nAAR with satellite information.\n1. The \"Non-Contiguous Raingauge\" (i.e. NCR) Procedure\nWithin a homogenous climate regime, AAR obtained by averaging a number of point\nmeasurements made at widely scattered raingauges is useful only if the resulting AAR\nprobability density function (p.d.f.) is the same as that determined from a contiguous\nraingauge network within the same regime. Fig. 1 illustrates how (for a spatially incoherent\nrainfield) the NCR method is used to construct an AAR estimate associated with a specific\nvalue of fractional cloud coverage. For the case shown, the AAR represents rainfall\naveraged over an accumulated area of 75,000 km2 (2.5° X 2.5° lat-long) and is compared with\na fractional cloud coverage of 0.4. This same comparison can be made using individual\ngauge measurements taken at different times (Fig. 2). Of course, the method can be applied\nfor other values of fractional cloud coverage and other satellite parameters as well.\nSince rainfall is spatially coherent, simple random averages of widely spaced point\nestimates leads to differences in the 2nd and higher ordered moments of the AAR p.d.f.'s.\nbetween a contiguous and a non-contiguous network. This, in turn, leads to incorrect\nassessments about the strength of the resulting statistical relationships. However, useful\nAAR values may be constructed from a non-contiguous network by selectively averaging a\nnumber of point estimates, such that the 2nd moment of the resulting AAR p.d.f. is the\nsame as that associated with a contiguous network. Since contiguous networks are absent,\nthe 2nd moment cannot be calculated directly. However, it can be estimated using the\nrainfall spatial correlation function1, the point variances and the sample size. A Monte\nCarlo technique has been developed which incorporates this information and results in a\nnumber of AAR values associated with each value of a satellite parameter. The technique\nhas been designed so that the higher ordered moments of the AAR p.d.f. will be similar\nbetween the two networks. The sampling error associated with these AAR estimates may\nthen be assessed and adjusted by increasing or decreasing the sample size (the number of\npoint measurements comprising each AAR estimate). A trade-off exists between the size\nof the sampling error and the number of AAR values available for comparison with the\nsatellite estimates.\n2. Evaluation of the \"Non-Contiguous Raingauge\" Method\nThe practicality of the proposed method is tested by selecting a contiguous and a\nnon-contiguous raingauge network within the same climate regime and assessing the\nstatistical relationships in each network among AAR and the fractional coverage of cold\ncloud (i.e. FCC; as obtained from satellite infrared data). If no significant differences are\nfound in the resulting statistics between the two networks then the proposed method can\nbe assumed valid. To reduce the effect of spatial inhomogeneity in the rainfall-FCC\nrelationships and varying amounts of sampling error within two separate networks, a non-\ncontiguous raingauge network was simulated by randomizing the individual raingauge\nOver open ocean regions, the spatial correlation structure of rainfall may be indirectly\nestimated using infrared satellite data, or directly measured using satellite microwave data.\n223","measurements within a contiguous network. This, in effect, discards the spatial information\non rainfall.\nDaily raingauge measurements were obtained from the Australian Meteorological\nBureau for December, 1983 through March, 1984 (122 days). A relatively homogenous\nregion encompassing 32 raingauges in the monsoon region of Northern, Queensland was\nselected as our study region. The region was enclosed by the boundaries of 140° - 142.5°E\nand 20.0° - 22.5°S (approximately 75,000 km ². The daily fractional coverage of cloud colder\nthan 265 K over this region was chosen as our cloud-top statistic.\nDaily AAR values were produced by simply averaging the 32 gauge measurements\nper day. The resulting 122 AAR values were then statistically compared with the\ncorresponding 122 FCC values using linear regression analysis.\nNext the 3904 (122 X 32) individual gauge measurements were randomized and 122\nAAR values were constructed using the Monte Carlo procedure (the spatial autocorrelation\nstructure of rainfall was assessed using the raingauges). Each AAR value was the average\nof 32 individual raingauge measurements, all of which were associated with the same value\nof FCC, but not necessarily taken on the same day. The resulting AAR values and the\ncorresponding FCC values were then plotted and a linear regression analysis was again\nperformed. A comparison of the results for both the contiguous and the simulated non-\ncontiguous network is shown in Fig. 3. The very small differences among the regression\nstatistics and parameters computed between the two networks are the result of errors\narising from the estimation of the rainfall spatial correlation function and the categorization\nof FCC. A comparison of the frequency distributions of AAR (not shown) indicates only\nminor differences in the p.d.f.'s of AAR between the two networks.\nThus, it appears the non-contiguous method may be useful for testing and calibrating\nsatellite-based rainfall algorithms. A suggestion that satellite infrared data may accurately\nestimate the rainfall spatial correlation structure comes from a comparison of spatial\ncorrelations between daily rainfall and infrared data ( < 265 W/m2 over 11 separate 2.52\nX 2.52 lat-long regions within Queensland (Fig. 4). The correlations were calculated using\ndata from the same 122 day period.\n3. Summary\nA technique has been illustrated that utilizes a network of widely scattered\nraingauges to investigate the relationships among area-averaged rainfall and cloud-top\ninformation. A Monte Carlo procedure has been developed to incorporate the rainfall\nspatial structure. The technique allows for the estimation of both linear and non-linear\nrelationships with a minimization of sampling error not obtainable with most contiguous\nraingauge networks.\n224","A SPARSE OR \"NON-CONTIGUOUS\" - 11\nRAINGAUGE NETWORK\n0.4\n0.4\n0.4\nx\nx\nx\n0.4\n0.4\nx\n0.7\nx\nx\n2.5 x 2.5 lat-long\n0.4\nx\nx\nx\nx\nx\nx\n0.4\nx\n0.8\n0.4\nx\n0.4\nx\nx\n1. An area averaged rainfall estimate associated with a fractional cloud coverage (FCC) of\n0.40 is constructed. Each raingauge is assumed to accurately represent a small amount of\narea surrounding the gauge (indicated by the small box surrounding each gauge). An\naveraged rainfall amount (to compare with FCC=0.40) representing 2.5° X 2.5° latitude-\nlongitude is produced by utilizing only the rain amounts from gauges where the FCC within\nthe 2.5° X 2.5° area surrounding each gauge (indicated by the larger box around each gauge)\nis 0.40.\n225","SPACE AND TIME CAN BE TRADED\nFCC\nFCC\n0.8\n0.3\n0.6\n0.4\n0.2\n0.8\n0.4\n0.2\nx\nx\ntime\ntime\nArea Average Rainfall\nFractional Coverage = 0.4\nx\nFCC\nFCC\n0.4\n0.2\n0.2\n0.9\n0.6\n0.4\n0.1\n0.7\nx\nX\ntime\ntime\n2. An area averaged rainfall estimate associated with a fractional cloud coverage of 0.40 is\nconstructed in a manner similar to that shown in figure 1, but the selection of 4 gauge\nmeasurements is made in time instead of space.\n226","Only the 10km X 10km pixels located over the individual raingauges\n2.5° X 2.5° lat-long areas located within Queensland, Australia.\nregression line, equation and the 95% confidence intervals are\nwere utilized in the spatial correlation calculations. The linear\nwith infrared irradiance ( <265 W/m2) within eleven separate\nRain Corr = 1.15 X IR Corr - 0.52\nIR < 265 W/m2\nRainfall VS IR Spatial Correlation\n4. A scatter plot of the spatial correlation of daily rainfall\n0.80\nIR Spatial Correlation\n0.75\n0.70\n:\n0.65\n0.60\nR = 0.82\n0.55\n0.4\n0.3\n0.2\n0.1\n0.0\nfor the conventional and the Non-Contiguous Raingauge method\nfor a 2.5° X 2.5° study region located in northern Queensland,\nScotter Plot: Rainfall vs Fractional Cloud Cover\nScotter Plot: Rainfall vs Fractional Cloud Cover\n1.0\n1.0\n+\n3. A Comparison of the results of the regression analysis\nfractional cloud coverage A 265 K\nfractional cloud coverage A 265 K\n0.8\n0.8\n0.6\n0.6\n0.4\n0.4\n0.2\n0.2\n0.0\n0.0\n50\n40\n30\n20\n10\n0\n50\n40\n30\n20\n10\no\nAustralia.\nStd Error (Intercept) = 0.69\nStd Error (Intercept) II 0.70\nRainfall Variance = 43.85\nRainfall Variance = 44.83\nStd Error (Slope) = 1.40\nNon-Contiguous Method\nStd Error (Slope) = 1.42\nConventional Method\nFc Variance = 0.14\nCorrelation = 0.45\nCorrelation = 0.45\nFc Variance = 0.14\nIntercept = 1.55\nIntercept = 1.59\nSlope = 7.79\nSlope = 7.80","Trimmed COADS - Problems and Solutions\nKlaus Wolter\nCooperative Institute for Research in Environmental Sciences, University of Colorado\nBoulder, CO 80309-0216\nTRIMMED COADS\nThe Comprehensive Ocean-Atmosphere Data Set (COADS Release 1; Slutz et al., 1985) has\nbeen compiled for use in global climate studies. It contains about 72 million ship reports for the\nnavigable globe during 1854 - 1979, covering best the last three decades and the North Atlantic\nand Pacific Oceans. Both monthly summaries for 2°x2° (also: 4°x4°) latitude-longitude boxes\nand individual marine reports are available for eight observed variables (SST, air temperature,\nwind speed, zonal and meridional wind component, SLP, total cloudiness, specific humidity).\nVarious wind stress and heat flux terms can also be obtained as COADS products.\nIn order to remove outliers, COADS observations were screened according to the following\nscheme (here simplified; ref. Slutz et al., 1985, Suppl. C, for more detail):\n*\ncreate 'trimming cube' based on adjacent 2°x2° boxes (<9) and months (3) for three distinct\nepochs: 1854 - 1909, 1910 - 49, and 1950 - 79;\n*\nwithin such an epoch and trimming cube, determine lower, median, and upper sextile for each\nmonth and variable, as long as at least 5 box-months with 3 or more observations are present;\n*\ncalculate trimming limits by adding (subtracting) 3.5o to the median, based on the difference\nbetween upper (lower) sextile and median, and remove all data outside these limits.\nInterim trimmed COADS releases for 1980-88 were prepared with 1950-79 trimming limits.\nPROBLEMS\nThis section will focus 'on problems that are unique to trimmed COADS. More general\nproblems, such as how to deal with inhomogeneities in COADS due to changes in instrumentation\nand observational practice will not be addressed here. Two types of statistical errors may have\noccurred in the process of trimming COADS: observations were excluded that should have been\nincluded (Type I error; Fig. 1a), and observations were included that should have been excluded\n(Type II error; Fig. 1b). While some COADS analyses have addressed Type II errors that may\nhave slipped through the trimming filter, this paper draws attention to Type I errors. Why ?\nClimate research is not confined any more to the study of long-term averages, but has focused its\nattention on climate anomalies. In fact, the bigger such anomalies are, the more interesting they\nbecome (e.g., the ENSO event of 1982/83), since the signal-to-noise ratio is increased. Extreme\nclimatic states, however, entail observational distributions close to and exceeding the trimming\nlimits (Fig. 1a). Trying to remove Type II errors through trimming can thus lead to Type I errors.\nIn addition, the introduction of drifting buoy data since FGGE should have improved the\ncoverage of high southern latitudes. However, interim COADS for the 1980's has been based on\n1950-79 trimming limits which do not extend to these latitudes, thus keeping such buoy data out\nof the trimmed 1980's set.\nEquatorial Pacific SST (10°N - 10°S, 180°W - 80°W), a region well known for persistent\nclimate anomalies, was chosen to examine this problem (Wolter et al., 1989). While that paper\nanalyzed the complete record of COADS Release 1 (1872 - 1979), the period 1950 - 88 is\ndiscussed here. If the trimming procedure removed only random statistical outliers (i.e., potential\nType II errors), roughly the same (low) number of observations would be taken out at the upper\nand lower limits, independent of the observed average SST. In our a priori chosen domain, this\nis not the case (Fig. 2), suggesting that a substantial fraction of the rejected data are Type I errors.\n228","Extreme El Niño and La Niña Januaries (e.g., 1983, 1976) even lose up to 10% of their data.\nThe difference between the upper and lower proportion of rejected observations thus correlates\nhighly with a trimmed SST index for the domain (Table 1a). An overall increase in trimmed\nobservations characterizes the 70's and 80's compared to the 50's and early 60's. This parallels\nthe substantial increase in telecommunicated (GTS) data in the last 20 years (with its potential for\ncoding and communication errors), combined with a decline in quality control at the data source.\nThe percentage of affected 2°x2° box summaries (Fig. 3a) is usually higher than the\npercentage of trimmed observations (Fig. 2), since one such observation is sufficient to change\nthe average SST within a box. Again, we find a strong association of the affected box rate with\nobserved SST (Table 1b). A complete loss of information due to the rejection of all data within a\n2°x2° box occurs on average for less than 1% from 1950 - 68, and above 2% (up to 11%) since\nthen (Fig. 3b), reflecting the increasing quality control problem of the last two decades (combined\nwith the outstanding climate anomaly of 1982/83).\nIn a case study, let us focus on February 83, the most extreme month within the ENSO event\nof the century (1982/83). Figure 4 shows histograms of observed SST with respect to the upper\ntrimming limits of all involved 2°x2° boxes. The SST distribution looks fairly \"normal\" prior\nto trimming (Fig. 4a), except for the tails at - 6°C and above +3°C. Enhanced tails indicate\npotential Type II errors. Observational anomalies appear realistic between - 5°C and +2°C with\nrespect to the upper limit. The trimmed SST distribution (Fig. 4b) is truncated on its right side,\nwhere the majority of screened observations occurs (12.4% versus 2.5% at lower limit). Quality\ncontrol measures appear to have removed most Type II errors before about 1965 in untrimmed\nCOADS (Wolter et al., 1989), while the present untrimmed COADS has a noticable percentage of\nsuch gross errors (1-2% of all observations in February 83) that need to be taken out.\nHow do these fractions of missing and changed data affect the spatial distribution of\nanomalies ? As can be seen in Fig. 5, the main losses of February 83 concentrate on the south\nequatorial Pacific (2°N - 6°S, 106 - 150°W), although reductions in the amplitude of anomalies\nare found almost everywhere. While the classic El Niño anomaly off the coast of Peru is mostly\npreserved (compare Fig. 5a versus b), the central Pacific anomaly is strongly reduced in size\n(~3°C) and extent in the trimmed map (Fig. 5b). Since most of the removed observations in this\nregion barely exceeded the upper trimming limits, the untrimmed SST anomaly of about 4°C (Fig.\n5a) is considered more realistic than the trimmed version. The implications for GCM case studies\nof 1982/83 based on COADS SST boundary conditions are obvious. On the other hand, extreme\nnegative outliers in the untrimmed version (Fig. 5a) are easily identifiable as coding errors. Since\ntrimmed COADS gets rid of them (Fig. 5b), the need for such quality control is reaffirmed.\nSOLUTIONS\nWhat can be done ? First of all, the old untrimmed data (say, prior to 1965) may be more\nrealistic (less Type I and only slightly more Type II errors) than the old trimmed data, since it\nappears quality-controlled. If in doubt, compare both sets against each other. The new data (post\n'65, especially post '79) needs some form of trimming. Three alternatives are currently being\nexplored and compared to the established procedure:\n1. We could raise the trimming limits from 3.5o to 5o. This approach would be very simple\ncomputationally, since it would involve only existing values for medians and O. However, it\nwould trade less Type I errors for more Type II errors, and perpetuate the questionable practice of\ntreating climate as an essentially invariant state with only sampling errors to worry about.\n2. We could apply a 'Winsorization' approach (Afifi and Azen, 1979), which uses monthly\nquartiles rather than decadal or trimming period limits, thereby embracing the idea of a variable\nclimate. Within each box, it sets all anomalies below the first and above the third quartile (25%,\n75%) equal to these quartile values. In other words, every observational value, whether an outlier\n229","or not (if within certain absolute limits about the long-term monthly mean), contributes to the\nmonthly box mean, but may be strongly reduced in its anomaly. Even though computationally\nsimple and applicable to all data, it suffers from the drawbacks of distrusting inherently 50% of\nall observations, and of largely removing the influence of the skewness on the monthly mean.\n3. We could trim each individual month about its median with limits of, say, 20 (based on the\naverage o computed from individual monthly sextiles). This approach has the advantage of\nleaving about 95% of the data intact, but it requires some particular rules for boxes with smaller\nnumber of observations. Both the second and the third approach would allow trimming limits to\nbe extended for the first time into areas now devoid of such limits and thus without any\ninformation in trimmed COADS.\nCONCLUDING REMARKS\nThis paper argues that the trimming of COADS based on several decades of monthly data\n(Slutz et al., 1985) is overly restrictive if the given month has extreme climatic conditions. Within\nmost of the modern record (1950-88), this translates only into small Type I errors in large-scale\narea averages since total losses of 2°x2° box summaries are rare. However, extreme events\n(be it the 1982/83 ENSO in the equatorial Pacific, or analogous extremes elsewhere) are\nsystematically underestimated and underrepresented within trimmed COADS. For instance, the\ntrimmed SST field of February 1983 features a large number of unnecessary gaps (where whole\nboxes were trimmed away), and is too cold by up to 1°C for much of the central Pacific (Fig.\n5).\nAlternative trimming procedures are currently being tested with the goal of minimizing both\nType I and Type II errors. The final approach will have to address the scatter of observations\nabout the individual monthly median rather than about a long-term average. We plan to trim the\nupdated COADS for the 1980's using the final revised procedure prior to its release. Further\nassessment may be needed of the extent of problems prior to 1980.\nAcknowledgments\nI thank Scott Woodruff at the Environmental Research Laboratory for keeping me straight on\nvarious aspects of trimmed COADS. CIRES Programmers Sandy Lubker and Mei Yuen Liu\nhelped with the programming. This research has been supported through EPOCS.\nREFERENCES\nAfifi, A. A., and S. P Azen, 1979: Statistical analysis: A computer oriented approach. 2nd edition, Academic\nPress, 442 pp.\nSlutz, R. J., Lubker, S. J., Hiscox, J. D., Woodruff, S. D., Jenne, R. L., Joseph, D. H., Steurer, P. M., and J. D.\nElms, 1985: Comprehensive Ocean-Atmosphere Data Set. Release 1. NOAA Environmental Research\nLaboratory, Boulder, Colorado, 268 pp.\nWolter, K., Lubker, S. J., and S. D. Woodruff, 1989: Trimming - a potential error source in COADS. Tropical\nOcean-Atmosphere Newsletter, 51 (Summer 1989), pp. 4 - 7.\nTABLE 1: Correlation coefficients for 1872 - 1909, 1910 - 49, and 1950 - 88, between trimmed January SST of\nthe equatorial Pacific (10°N - 10°S, 180°W - 80°W) and (a) the difference between the percentage of rejected\nobservations at the upper and lower trimming limits for the same area; (b) the difference between the percentage of\naffected 2°x2° box summaries due to the same trimming limits. Nonparametric Spearman Rank correlations are\nprovided for 1872-1988. Significance at the 99.9% (99.99%) level is indicated with a single (double) asterisk.\n1872-1909\n1910-49\n1950-88\nSpearman Rank Correlations for 1872-1988\n(a)\n+0.40\n+0.60**\n+0.70**\n+0.61**\n(b)\n+0.52*\n+0.64**\n+0.70**\n+0.55*\n230","TYPE I ERRORS\nTYPE II ERRORS\n4\n3\n2\n1\n0\nA\nB\n-1\n0\n5\n10\n15\n20\n25\n30\n0\n5\n10\n15\n20\n25\n30\nDAY\nDAY\nFig. 1. Schematic diagrams for statistical errors: (a) Type I, (b) Type II. Dots represent arbitrary daily\nobservations (normalized based on long-term statistics), solid lines stand for long-term medians, and dashed lines for\nupper trimming limits.\n10\nUPPER %\nLOWER %\n5\n0\n5\n10\n1950\n1955\n1960\n1965\n1970\n1975\n1980\n1985\n1990\nYEAR\nFig. 2. Percentage of rejected observations in Januaries 1950-88 for central and eastern equatorial Pacific SST\n(10°N-10°S, 180°W - 80°W) due to upper and lower trimming limits.\n231","30\nA AFFECT %\nB LOSS %\n25\n20\n15\n10\n5\n0\n1950\n1955\n1960\n1965\n1970\n1975\n1980\n1985\n1990\nYEAR\nFig. 3. Total percentage of 2°x2° box summaries (a) affected, and (b) removed by trimming procedure in Januaries\n1950-88 for equatorial Pacific SST (10°N - 10°S, 180°W - 80°W).\n225\n225\nA\nB\n200\n200\n175\n175\n150\n150\n125\n125\n100\n100\n75\n75\n50\n50\n25\n25\no\n0\n-6\n-4\n-2\n0\n2\n4\n-6\n-4\n-2\n0\n2\n4\nATEM\nATEM\nFig. 4. Histograms of equatorial Pacific SST observations (10°N - 10°S, 180°W - 80°W) for February 1983 with\nrespect to the upper trimming limits (based on 1950-79): (a) untrimmed, (b) trimmed. The left- and right most\ncolumns in the histograms refer to all values below - 5.5°C and above +5.0°C, respectively.\n232","FEBRUARY 1983 DEPARTURES FROM 1950-79 MEAN (G) FOR UNT SST\nA\n.75 GT ISST-G) LE\n..25\n25 CT (SST-G) LE\n.25\nB\n-1.25 GT (SST-G) LE\n-.75\n1\n.25 GT (SST-G) LE\n.75\nC\n-1.75 GT ISST-G) LE\n-1.25\n2\n.75 GT ISST-G) LE\n1.25\nD\n-2.25 GT ISST-C) LE\n-1.75\n3\n1.25 GT (SST-G) LE\n1.75\nE\n-2.75 GT (SST-G) LE\n-2.25\n4\n1.75 GT (SST-G) LE\n2.25\nF\n-3.25 GT ISST-G) LE\n-2.75\n5\n2.25 GT ISST-G) LE\n2.75\nG\n-3.75 GT ISST-G) LE\n-3.25\n6\n2.75 GT (SST-G) LE\n3.25\nH\n-4.25 GT (SST-G) LE\n-3.75\n7\n3.25 GT (SST-G) LE\n3.75\nI\n**** GT (SST-G) LE\n-4.25\n8\n3.75 GT (SST-G) LE\n4.25\n9\n4.25 GT (SST-G) LE\n****\n180W\n155W\n130W\n105W\nBOW\n10N\n10N\nA\n0\n0\n10S\n10S\n180E\n205E\n230E\n255E\n280E\n105W\nBOW\n130W\n180W\n155W\n10N\n10N\nB\n0\n0\n10S\n10S\n255E\n280E\n230E\n80E\n205E\nFig. 5. Spatial distribution of 2°x2° SST anomalies in equatorial Pacific SST (10°N - 10°S, 180°W - 80°W) in\nFebruary 1983 for (a) untrimmed, (b) trimmed observations. A legend is given on top of the figure. The area with\nthe highest incidence of trimmed 2°x2° boxes is outlined with a broken rectangle (2°N-6°S, - 106 - 150°W).","Global Tropospheric Temperature Monitoring with\nPassive Microwave Radiometers\nby\nRoy W. Spencer\nES43 Earth Science and Applications Division\nNASA Marshall Space Flight Center\nAlabama 35812\nJohn R. Christy\nJohnson Research Center\nUniversity of Alabama in Huntsville\nHuntsville, Alabama 35899\n1. Introduction\nIn contrast to conventional measurements, satellites can provide the global coverage that is\nneeded to monitor the earth's atmosphere, even on a daily basis. Unfortunately, the issue of\nsatellite instrument calibration has typically been a source of great concern and uncertainty in the\nremote sensing community. However, possibly the best calibrated instruments in earth orbit to\ndate have direct application to the global atmospheric temperature monitoring issue. These are\nthe Microwave Sounding Units (MSU's), built by the Jet Propulsion Laboratory for the National\nOceanographic and Atmospheric Administration's (NOAA) operational weather monitoring needs.\nBecause these instruments measure a vertically averaged atmospheric temperature, we feel that\nthey have the potential for significantly augmenting the surface-based thermometer record by\nproviding a measurement representing a significant depth of the troposphere, rather than just a\nthin near-surface layer sensitive to variable surface effects (urbanization, desertification, etc.).\n2. Theory\nThe type of radiometer currently represented by the MSU's is designed to measure the\nthermal emission by molecular oxygen in the atmosphere at different spectral intervals in the\noxygen absorption complex near 60 GHz. Oxygen abundance in the atmosphere is very stable in\nboth space and time, with all recorded measurements since 1910 falling within the range of\n0.20945 and 0.20952 by volume.\nAt microwave frequencies, radiance is directly proportional to the temperature of the\nemitting body. The radiometer output is usually expressed as a \"brightness temperature\" (TB).\nThe term \"brightness temperature\" acknowledges that the measurement is based upon radiative\nbrightness, which equals a thermometric temperature only when the measured object is radiating\nas a black body (unit emissivity, and thus zero reflectivity). Fig. 1 shows the weighting functions\nfor MSU channels 2 and 4, which reveal what levels in the atmosphere the radiation originates\nfrom. The measured (as well as theoretically calculated) channel 2 TB ranges from approximately\n200 K for a very cold polar atmosphere to about 260 K in tropical atmospheres. Comparisons\nwith rawinsonde data reveal it's TB variations are most highly correlated with mid-tropospheric\ntemperature changes (Fig. 2), although correlations with other pressure levels in the troposphere\nare also high. Thus, channel 2 measurements can be thought of as primarily a mid-tropospheric\nmeasurement, but practically speaking represent most of the troposphere.\n234","MSU-inferred temperature anomalies are closely related to those measured by\nradiosondes. Previous investigators have made comparisons between MSU measurements and TB\ncalculated from rawinsonde data. Considering both land and ocean surfaces, the standard error\nbetween the satellite measured and rawinsonde-calculated TB is 0.7°C. This error is found to be\nindependent of surface type. The fact that the error is larger than the single-footprint in-\nstrumental noise of 0.25°C can be attributed to sampling errors, i.e. spatially averaged satellite\nobservations compared to the point rawinsonde measurements used in computing TB, and random\nerrors in the baseline calibration of the rawinsondes before their release. When the MSU data\nare used for climate purposes, the large number of observations reduce the random error from\ninstrument noise to essentially zero.\nCirrus cloud effects are virtually non-existent at these frequencies, a significant ad-\nvantage over infrared temperature sounders. All other effects on the channel 2 TB (cloud water,\nsurface temperatures, land wetness, ocean roughening by wind) have been evaluated based on\ntheory, and on analysis of the data, and have been determined to be small for hemispheric com-\nputations, and probably the vast majority of regional computations.\n3. Method\na. calibration\nThe MSU radiometric measurements of atmospheric thermal emission are calibrated\nthrough their comparison to a view of cold space (2.7 K), and a warm target whose temperature\nis monitored with redundant platinum resistance thermometers. Interpolation between these two\nreference extremes to the earth observation value allows calculation of TB.\nb. MSU viewing geometry\nThe MSU's are mounted underneath the NOAA TIROS-N series of satellites, which have\nsun-synchronous near-polar orbits at nominal local crossing times of 0730 and 1930 LST for the\n\"morning\" satellites and 0230 and 1430 LST for the \"afternoon\" satellites. Each MSU scans\nacross the satellite sub-orbital track through nadir such that it measures eleven \"footprints\" on\neach scan, scans being spaced in the along-track direction by 170 km. Only the central five of\nthe eleven footprints are used in this study. An example of the geographical distribution of one\nday's total number of footprints falling within 2.5° grid squares (Fig. 3) shows the orbital swath\npattern and the inter-swath data void regions. The pattern in Fig. 3 moves eastward on a daily\nbasis due to precession of the orbit, at a rate which varies from satellite to satellite.\n4. Results\na. MSU noise estimates from satellite intercomparison\nOf special interest is a near-two year overlap period when the NOAA-6 and NOAA-7\nsatellites simultaneously observed the earth. An annual cycle of temperature was calculated for\neach satellite separately by averaging the two years' 2-day averages together, then low pass filter-\ning this cycle to provide a more representative (smoother) cycle. Then the original two-day\naverages were subtracted from these smoothed cycles to give the 2-day hemispheric anomalies\n(Fig. 4). The hemispheric and global anomalies for NOAA-6 and NOAA-7 averaged together are\nshown, and immediately below them the averaged difference in anomalies between the two satel-\nlites. Even though these separate satellites had different orbits, it can be seen that both satellites\nshow the same 2-day hemispheric temperature anomalies to within about 0.03°. The signal to noise\nrepresented in Fig. 4 is about 41 for the northern hemisphere, 31 for the southern hemisphere,\n235","and 46 for the entire globe. Thus, these two satellites measure essentially the same hemispheric\nanomalies. The ragged character of the time series of 2-day averages is intriguing, and shows\nthat jumps and dips in global temperatures of 0.5 o can occur in less than two weeks. That these\nabrupt changes are real is supported by the different orbits of the two satellites. Their differing\nprecession rates and daily geographical coverage exclude the possibility that they are both\ngeographically undersampling the earth in the same manner.\nAlso, during this overlap period, neither satellite drifts relative to the other at about the\n0.001° level. Assuming that both MSU's are not together drifting in exactly the same way, this is\nvery significant. If indeed we can expect stability on the order of 0.01 o to 0.001 o over a two\nyear period, then our atmospheric monitoring capability from satellites has achieved a very useful\nlevel of precision.\nb. comparisons with thermometer-derived anomalies\nMonthly averaged MSU data on a 2.5° grid covering the United States was compared to\nmonthly averaged surface air temperature data (Tom Karl, personal communication) for the\nperiod 1979 through 1984. A six-year average annual cycle was computed for each data set, and\nthen was differenced with the original monthly averages to obtain 72-month time series of tem-\nperature anomalies. The correlation between the MSU and thermometer anomaly time series was\n0.90 (81% explained variance), indicating good agreement. This is consistent with the results in\nFig. 2, which suggested that surface air temperatures should correlate well with a mid-\ntropospheric temperature measurement. The surface air temperature anomalies average about\ntwice the magnitude of the MSU channel 2 anomalies.\nTable 1. Monthly and annual explained variances between MSU and thermometer-measured -\ntemperature anomalies (Tom Karl, personal communication) for United States.\nMONTHLY\nANNUAL\nU.S.\nU.S.\nCorrelation (r)\n0.90\n0.96\nExplained\nvariance (r2)\n0.81\n0.93\nThe hemispheric level of agreement is not nearly as good, and will be addressed in the companion\npaper.\n236","The above results suggest that MSU channel 2 provides a very useful temperature\nmeasurement which, when monitored globally, can provide a powerful data set for analysis and\nmonitoring of climate anomalies. The companion paper will deal with resulting anomaly calcula-\ntions for the period 1979 through 1985.\nAcknowledgments\nWe are very grateful to Roy Jenne and Dennis Joseph at the National Center for Atmos-\npheric Research who provided the MSU data used in this study; to Ms. Robbie Hood who\nprovided programming assistance and library research support; to Pete Olsen, Fred Soltis, and Dr.\nPaul Swanson at the Jet Propulsion Laboratory, who assisted us in obtaining technical data on the\nMSU; to Dr. Gregory Wilson, who helped obtain the satellite data sets; and to Dr. James Dodge at\nNASA Headquarters, discussions with whom led to the present research.\n1\nNadir (0=0°)\nScan Limit (0 =56.6°)\n2\n3\n5\n7\nCH. 4\n(57.95 GHz)\n10\n20\n30\nCH. 2\n(53.74 GHz)\n50\nE = 0.5\nE = 1.0\n70\n100\n0.0\n0.2\n0.4\n0.6\n0.8\n1.0\nTemperature Weighting Function\nFig. I Temperature weighting functions for MSU channels 2 (53.74 GHz) and 4 (57.95 GHz), for\nthe nadir (solid) and extreme (dashed) scan positions. Also shown is the difference in the\nweighting functions for ocean (emissivity=0.5) and land (emissivity = 1.0) surfaces (figure\nadapted from Grody, 1983).\n237","Explained Variance (%)\n0\n50\n100\n10\nSpring\n20\n30\n40\n50\n60\nSummer\n70\nSpring\n80\n90\nsfc.\n-1.0\n-0.5\n0\n0.5\n1.0\nCorrelation Coefficient\nFig. 2 Correlation coefficient (r, solid lines) and explained variances (r2, dotted or dashed lines)\nbetween MSU channel 2 TB calculated from rawinsonde data and the rawinsonde measured tem-\nperatures at different atmospheric pressure levels.\nFig. 3 One day (10 January 1982) of earth coverage from a single MSU (NOAA-7), where 2.5°\ngrid squares containing one or more of the middle five footprints of a scan line are shaded from\nlight (1 print) to black (10 prints).\n238","NOAA-6 vs. NOAA-7\nGlobe\n+0.6°\nod = 0.0233°\n°s = 0.1584°\n0.0°\n-0.6°\nN. Hemisphere\n+0.6°\nOs=0.2004°\n=0.0315°\nMM,\n0.0°\n-0.6°\nS. Hemisphere\n+0.6°\nOs=0.1771°\nod = 0.0318°\n0.0°\n-0.6°\n16 April\n29 June\n1983\n1981\nFig. 4 Intercomparison between global NOAA-6 and NOAA-7 MSU channel 2 TB during a two\nyear period (29 June 1981 to 16 April 1983) when both satellites had MSU's operating. Time\nseries of global and hemispheric satellite averages (sum/2) and differences (difference/2) are\nshown by the large variation and small variation curves, respectively. Note the stability of the\ndifference between satellites during the period. The standard deviations of the two-satellite sums\nand differences are also shown, with signal-to-noise ratios (=[ sum/diff.12) of 46 for the globe,\n41 for the northern hemisphere, and 31 for the southern hemisphere.\n239","Global Tropospheric Temperature Variations since 1979 as measured\nfrom the Microwave Sounding Unit\nJohn R. Christy\nAtmospheric Science and Remote Sensing Laboratory\nUniversity of Alabama in Huntsville\nRoy W. Spencer\nEarth Science and Application Division\nNASA/Marshall Space Flight Center\nBecause the atmosphere is global in extent, one must examine it as a whole to gain an\nunderstanding of the variability of the climate system. Global properties such as mean surface\ntemperature have been generated from surface data, but inherent inhomogeneities limit the\nconfidence one may place in any results. Typical problems of surface measurements include\nchanges in instrumentation, exposure, observers, local environment, observation times as well as\ndifficulties related to instrument drift and poor spatial sampling.\nWith the Microwave Sounding Unit (MSU) scientists have the ability to monitor passive\nmicrowave emissions from space with a single instrument per satellite which is calibrated on every\nscan. Spatial sampling is excellent as these polar platforms orbit the earth 14 times per day.\nPossible uncertainties in temperature due to surface contamination, cloudiness and precipitation are\nextremely small for channel 2 (lower troposphere) and negligible for channel 4 (lower\nstratosphere). In this way, the brightness temperature is extremely close to physical temperature.\nSo robust are these measurements that separate MSUs aboard different satellites produce monthly\nglobal mean temperatures with an error standard deviation of less than 0.01°C.\nThe MSUs examined here flew on TIROS N, NOAA 6 and 7. Each instrument has its own\nset of calibration coefficients and equatorial crossing times, SO that absolute temperatures vary by a\nfew tenths of a degree between the units. We were able to eliminate instrument bias in NOAA 6\nand 7 by producing 2-day zonal mean values of the annual cycle from each satellite for the overlap\nperiod (21 months). Anomalies were then generated as departures from these annual cycles. The\neffective overlap between TIROS N and NOAA 6 was only 200 days, so the annual cycle\ncomputed from NOAA 7 (also an afternoon crossing) was substituted for the TIROS N annual\ncycle. Anomalies thus produced during the 200 day overlap were nearly identical between TIROS\nN and NOAA 6, with a signal-to-noise ratio of 2-day global mean values of 46. Another set of\nmonthly gridpoint anomalies was generated in the same manner over a 2.5° grid from 85°S-85°N.\nThese anomalies were then patched together into one time series beginning 13 November 1978 and\nending 18 February 1985.\nOf greatest interest is the record of temperature anomalies. Figure 1 displays the global,\nhemispheric and latitudinally weighted zonal anomalies from the patched data set for channel 2.\nThe low-pass filter allows for half of the power for 80-day periods to be retained, with greater\npower for longer periods. The most remarkable feature is the tropospheric warming of late 1982\nand 1983 in association with the major ENSO of that time. The bulk of the warming occurred\n240","within 20° of the Equator, as did the subsequent dramatic cooling of late 1983. The annual global\nanomalies (preliminary) from these data are (1979-1984, °C) -0.07, +0.11, +0.02, -0.06, +0.16\nand -0.16. Correlations between the MSU results and other surface estimates are given below.\nThe first value utilizes 7 footprints per scan while masking out all grids above 2500 m. The\nsecond correlation uses 5 footprints and no elevation masking.\nCorrelation with MSU channel 2 temperatures\nSize\nGlobal\nNH\nSH\n.61\nJones et al (1986) monthly\n72\n.58\n.55\n.52\n.37\n.45\n.87\n.84\n.79\n.80\nJones et al (1986) annual\n6\n.85\n.82\nHansen and Lebedeff annual\n6\n.71\n.68\n.78\n.74\n.56\n.79\nBecause the MSU has been verified as an instrument which measures surface temperature\nvariations (companion paper) we conclude that the \"global\" and \"hemispheric\" values of surface\ntemperature from ground-based data are of only marginal use since they do not sample the global\natmosphere in a systematic way. Of course, with only 6 years of data, we make no conclusions\nconcerning long term trends nor of the accuracy of these global surface temperature sets for\nspecific regions. It is clear that the data from Jones et al. is to be preferred over Hansen and\nLebedeff and that annual values are much more accurate than monthly.\nFigure 2 presents the monthly time series of the MSU channel 2 and of Jones et al (1986).\nNote that in the MSU case, the hemispheric values are more highly correlated with each other than\nin the surface observations. The reasoning is that the tropical belt behaves in a more-or-less\nconsistent manner, thereby affecting both hemispheres in the same way. Surface derived\nobservations have only a few data points in these regions.\nChannel 4 results (Fig. 3, lower stratosphere ~70 mb) show a single warm global anomaly\nbeginning in early 1982 and lasting for a year. The time series indicates a relatively slow variation\nin the global value with considerable interhemispheric compensation. The sudden warming near\n30°N in NH Spring 1982 may be related to El Chichon, however examination of the unfiltered\ndaily zonal time series shows that the bulk of the warming began on 20 February 1982 which was\nover a month prior to the major eruptions (28 March to 4 April).\nConsiderable work remains as these time series are being updated. Once this is\naccomplished, these data will be utilized for a wide range of research including studies on global\ntemperature variations, energy balances and the validation of the NCAR CCM1. Though\nforecasting in this business is fraught with danger, we feel the MSU data will provide the standard\nby which global temperature variations will be assessed in the coming years.\nAcknowledgement: These data would have been lost forever had not Roy Jenne (NCAR)\npossessed a determination to preseve them on a modern medium and Dennis Joseph (NCAR) had\nthe skill to transfer over 99% of the original information from the \"ancient\" TBM reels.\n241","SAT=124 CHAN= 2 DATES\n321\n2605\n0.5\nFILTER WIDTH\n41\n1979\n1980\n1981\n1982\n1983\n1984\n-0.5\n78321\n85049\n90.0\nFILTER WIDTH\n41\nailg\nEQ\nEQ\nEQ\nEQ\n1979\n1982\n1\n980\n1981\n1983\n1984\n- 90.0\nFig. 1 Filtered anomalies of tropospheric (Channel 2) temperatures from MSU's aboard TIROS\nN, NOAA 6 and NOAA 7. The top curves are global, NH and SH anomalies and the\nlower panel shows the latitudinally weighted zonal anomalies with contour interval of 0.2\n°C and the zero contour omitted.\n242","0\n0\n0\n1978\n1979\n1980\n1981\n0\n0\n0\n1982\n1983\n1984\n1985\n1\n0\n1\n1980\n1981\n1978\n1979\n1\n0\n- 1\n1985\n1982\n1983\n1984\nFig. 2 Anomalies of monthly temperatures from channel 2 MSU (upper) and Jones et al.\n(lower). The base period for both graphs was 1979-1984.\n243","SAT=124 CHAN= 4 DATES\n321\n2605\n1.0\nFILTER WIDTH\n41\n1979\n1980\n1981\n1982\n984\n1983\n1\n1.0\n-\n78321\n85049\n90.0\nFILTER WIDTH\n41\n800\n100\nEQ\nEQ\n1979\n1980\n1981\n1982\n1983\n1984\n90.0\nFig. 3 As in Fig. 1 but for lower stratospheric temperatures (Channel 4) with contour interval of\n0.4°C.\n244","A GCM Study On the 1988 U. S. Drought\nKingtse Mo and J. R. Zimmerman\nClimate Analysis Center\nand\nE. Kalnay and M. Kanamitsu\nDevelopment Division\nNMC/NWS/NOAA, Washington D. C. 20233\n1. Introduction\nThe United States drought of 1988 was the most severe in the\nMidwest since 1936. The driest and hottest periods in over 90\nyears were recorded in parts of the Plains, Midwest and lower\nMississippi Valley. The severity of the Midwest drought peaked\nin June, but the development of drought can be traced back one or\nmore years in different parts of the country. In the beginning\nof July, normal rainfall returned to some parts of the central\nUnited States and improved short term water deficiencies.\nFor the 1988 drought, Trenberth et al (1988) suggested that\nthe sea surface temperature anomalies (SSTA) in the tropical\nPacific were the main cause of the event. Their results were\nbased essentially on simulations of a linear model. Linear models\nmay not give a complete picture of the complicated relationship\nbetween tropical forcing and extratropical circulations. Palmer\nand Brankovic (1989) did GCM experiments using the ECMWF model to\nconfirm that the SSTA were linked to the US drought. But their\ncontrol experiment consisted of only a single 30 day forecast,\nwhich might not be stable. In this paper, we study the origin of\nthe June 1988 drought using the NMC Medium Range Forecast (MRF)\nmodel. By comparing experiments using the SSTA fixed at the\nstarting dates, and experiments using climatological SST but with\nthe same initial conditions, we are able to assess the influence\nof the SSTA. The importance of the particular initial conditions\nwill also be discussed.\n2. Circulation features during June 88\nIn this section, the anomalous atmospheric circulation and\nboundary conditions associated with the 1988 drought will be\nreviewed. All data presented here and the initial conditions for\nthe model, were derived from NMC analyses. Fig. la and 1b show\nmonthly mean 500 mb heights and standing waves for June 1988. In\nthe Northern Hemisphere (NH), they show a persistent ridge over\nthe north -central States and central Canada, while troughs can\nbe found just off the West Coast of the U. S. and in the Gulf of\nAlaska as well as along the East Coast of Canada.\nThe jet\nstream, as indicated by the 200 mb zonal wind map of June 1988\n(not shown) showed a northward displacement from the normal\nposition in the United States. Storms, which followed the path of\nthe jet, moved further to the north during that period, so areas\nin the Midwest were drier than normal. Standing waves in the NH\nduring June 1988 were similar to the climatology. Fig. 2a plots\nthe 500 mb standing waves from the June climatology archived at\n245","the Climate Analysis Center, which covers the period from 1978-\n1988. The similarity between the wavetrain pattern of June 88\n(Fig. 1b) in the NH and climatology is striking. The locations of\nhighs and lows are the same, but the magnitudes of the standing\nwaves for June 88 are much stronger. This can be shown by plot-\nting departures from climatology of the standing waves during\nJune 1988 (Fig.2b).\n3. Control Experiments\nAll numerical experiments were done using the operational\n1989 NMC spectral model (NMC Staff, 1988) except for the horizon-\ntal resolution. We used a T40 version of the model, while T80 is\nused for daily MRF operation. Since the wavetrain was stabi\nlized near May 25, 1988, we made three 30 day integrations as\ncontrol experiments starting with initial conditions of May 21,\n22 and 23 of 1988. The SST anomalies were fixed at the starting\ndate during the entire integration for each experiment. The soil\nmoisture field was initialized with climatological values but it\nwas updated during the integration.\nFig. 3a shows 500 mb standing waves for the ensemble mean of\nthe three control runs. The agreement between simulated and\nobserved heights is remarkably good in the NH. The wavetrain\nover North America is extremely well simulated both in location\nand strength. Even the features in Europe and Asia are also\nreasonably reproduced.\n4. Climatology SST (CSST) experiments\nTo examine the importance of SST anomalies to the drought in\nthe United States, we repeated three 30 day integrations with the\nsame initial conditions as the control experiments but with\nclimatological SST. Fig. shows the ensemble mean of heights\nand standing waves at the 500 mb level. The overall pattern is\nsimilar to the waves simulated by the control experiments\n(Fig. 3), but there are some differences. The trough, originally\nlocated at the west coast of the United States, now moves further\nout into the Pacific ocean. The ridge in the north-central part\nof the U.S is weaker.\nTwo questions exit: is the wavetrain simulated by the CSST\nexperiments stronger than the June climatology; and are the\nchanges between CSST experiments and control experiments statis-\ntically significant. We conducted a statistical significant test\nto answer these questions.\nFig. 4a gives the difference map between the simulated ensem-\nble mean of standing waves at 500 mb for the CSST experiments and\nJune climatology. The statistical significance test was done for\nfields at each grid point separately. Areas where features are\nstatistically significant at the 95% level are shaded. It shows\nthat the negative anomalies centered at the west coast of the\nUnited States are no longer stronger than climatology. However,\nthe most part of the ridge located in the North-central U. S.\nand central Canada is still reliable and stronger than climatolo-\ngy. Fig. 4b shows the difference map between the ensemble means\nof the control experiments (Fig. and the CSST experiments\n(Fig.4). The weakening of the trough in the Pacific is signifi-\n246","cant. But the part of the ridge located in the central U. S. and\nthe trough located in the eastern Canada are there for both sets\nof experiments. This suggests that the ridge in the central U.\nS. and Canada and trough in the eastern Canada are stable and\nsignificant features. They are still present even though the\nSSTA disappear.\nTwo additional 30 day integrations from May 22 1987 and May\n22 1989 and climatological SST have been performed. The charac-\nteristic wavetrain pattern of the 1988 integration is not appar-\nent in either integrations. The wavetrain pattern is not a part\nof the model's climatology.\n5. Barotropic instability\nFrom these experiments, we have seen that the SSTA in the\ntropics are not the main cause of the drought. We have also\nshowed the importance of the initial conditions to the mainte-\nnance of the wavetrains in the Pacific-north America region.\nFurthermore, we notice the unusual similarity between the stand-\ning waves for the 500 mb heights of June 1988 and the June clima-\ntology. Apparently, this pattern belongs to one of the most\nstable regimes in the atmosphere and can be easily excited by\ndisturbances, since a similar pattern has been obtained by forc -\ning a linear model with a tropical heating pattern (Trenberth et.\nal. 1988) . A possible explanation for this is that the wave -\ntrain pattern may be associated with the barotropic instability\nof the climatological base state. To confirm this hypothesis, we\nobtained the normal mode structures associated with barotropic\ninstability. The barotropic model used is described by Tribbia\nand Ghil (1989) The second mode has the e-folding time of 18\ndays, and a period of 45.6 days. Its structures in the Pacific\nand / North America region are similar to the wavetrain pattern of\n1988.\n6. Discussion\nWe have done GCM experiments to simulate the 1988 United\nStates drought and studied its origin. Control experiments were\nrun which include three 30 day integrations with the SSTA fixed\nat the initial date and initial conditions from May 21-23, 1988.\nForecasts in the NH especially in the Pacific - North America\nregion are very good. These experiments simulate the persistence\nof the pattern and the standing waves during that period.\nWe then did CSST experiments by repeating the control exper-\niments using the same initial conditions but the climatological\nSST. The wavetrain pattern is weaker, but there is still a ridge\nlocated in the north-central U. S. and Canada. The simulated jet\nstream in North America is also shifted north and differs very\nlittle from the control experiments. It seems that the SSTA of\n1988 are not the main cause of the wave pattern, and therefore,\nnot the mean cause of the 1988 drought over U. S.\n247","90N\n90N\n$400\n5520\nYOUR\n60\n60N\n60N\no\n12\n57602\n30N\n80\nSON\nSEED\no\n&\nEQ\nEQ\n5000\nsaso.\n305\n5780.\n30S\n60.\n53000\n60S\n120\n-00.\n60S\n-sb.\nsoud\n450\n905\n180W 150W 120H 90W 60W 30W OE 30E 60E 90E 120E 150E 180E\n90S\nso,\n180H\n150H\n120W\n90W\n60W\n30H\nCONTOUR INTERVAL. 60.00\nOE\n30E\n60E\n90E\n120E\n150E\n180E\nCONTOUR INTERVAL 30.00\na\nb\nFig. 1a: Monthly mean 500 mb geopotential height for June 1988\nfrom NMC analysis. Contour interval is 60 m.\n1b: Same as above, but for standing waves. Contour interval\nis 30m.\n90N\n90N\na\n.60\n60N\n60N\n60s\nsen\nso.\n(60\n12\n6d\n{\n30N\n30N\nbe\nB\n0\nEO\nEQ\n305\n30S\n60.\n60.\nso.\n-120\n60.\n605\nEGO\n60\nGO.\n605\n60\nsap\nis\nTOP\nCOO\n60.\na\n90S\no\n90S\n180W 150H 120H 90H 60H 30H OE 30E 60E 90E 120E 150E 180E\n160H 150H 120W 90H 60H 30H OE 30E 60E 90E 120E 150E 180E\nCONTOUR INTERVAL- 30.00\nCONTOUR INTERVAL 30.00\nb\na\nFig . 2 a : Standing waves for 500 mb heights for June climatology.\nContour interval is 30 m.\n2b: Difference between standing waves for 500 mb heights for\nJune 1988 and June climatology. Contour interval is 30 m.\n248","SON\n90N\nbe\n60\n60N\n60N\n60\n120\n60.\n30N\nSON\nI\nEQ\nEO\n1\n30S\n305\n-60.\n60.\n605\nso\nSOS\nto\n7.30\n600\n90S\n90S\n180W\n150W\n120W\n90W\n60H\n30W\nOE\n30E 60E 90E 120E 150E 180E\n180H\n150H 120W 90H 60W 30W OE 30E 60E 90E 120E 150E 180E\nCONTOUR INTERVAL 30.00\nCONTOUR INTERVAL 30.00\nb\na\nFig. 3a 30 day mean standing waves for the 500 mb heights averaged\nover three control experiments. Contour interval is 30 m.\nFig. 3b Same as Fig 3a, but for the CSST experiments\nor\nBON\n9\nSON\nD.\n0\nON\n10\n0.\nDr.\n0.\n20N\n0.\nED\n180H\n150M\n120N\n90H\n60M\n30M\nOE\n30E\n60E\n90E\n120E\n150E\n180E\nCONTOUR INTERVAL 20.00\nFig. 4a: Difference map between the 30 day mean standing waves for\n500 mb heights for the CSST experiments and June climatology.\nContour interval is 30 m. Areas where features are statistically\nsignificant at the 95% level are shaded.\nDIFFERENCE OF 500 MB STANDING WAVES\nBON\nao\nBON\n40\n20\nat\nee\nMON\n0\n29\n0.\n20N\n0.\nbe\nEO\n180E\n30M\nDE\nSOE\n60E\n90E\n12DE\n150E\n180M\n150M\n120N\n90M\n60M\nCONTOUR INTERVAL 10.00\nFig 4b. . Difference map between ensemble mean of standing waves\nfor 500mb heights for the control experiments and climatological\nSST experiments. Areas where features are statistically signif-\nicant at the 95% level are shaded.\n249","Comparisons of model-produced diabatic heating component fields\nDavid A. Salstein and Richard D. Rosen\nAtmospheric and Environmental Research, Inc.\n840 Memorial Drive, Cambridge, MA 02139\n1. Introduction\nThe means by which diabatic heating is distributed within the atmosphere is\nof fundamental importance to its climate and circulation. Thus, diabatic\nheating is responsible for generating the available potential energy (APE)\nthat is subsequently converted to kinetic energy. Despite this importance,\nthere is no direct means of observing heating in the atmosphere; rather, two\nindirect methods have been used to calculate it. In one, heating is deter-\nmined as a residual of the thermodynamic equation typically in either isobaric\n(e.g. Kasahara et al., 1987) or isentropic (e.g. Wei et al.) coordinates; this\nmethod requires atmospheric analyses of a number of variables. In the second\nmethod, each of the components of diabatic heating is calculated by the use of\nphysical parameterizations, in conjunction with a numerical model; these com-\nponents are then summed to form the total (see Schubert and Herman, 1981).\nUnlike the thermodynamic approach, the model approach allows us to study the\nheating components individually as well as in total. Here we use this method\nto examine and compare the heating component fields produced by several\nmodels. We study the dependence of latent heat on parameterization as well as\nthe heating variability on a number of time scales.\n2. Data and results\nWe have turned to the NASA Goddard Laboratory for Atmospheres (GLA) for a\nset of simulations from the FGGE first Special Observing Period (SOP). From a\ndata assimilation for the period 6 January - 4 February 1979, we averaged the\ndiabatic heating component fields that were produced by the GLA fourth-order\nglobal circulation model on a 4 X 5 degree latitude-longitude grid at nine\nsigma levels (Kalnay et al., 1983), separately at four synoptic hours. The\nfour heating components are due to:\n1. Shortwave radiation, whose calculation is based upon multiple scattering\nwith absorption due to ozone and water vapor, and uses ground and variable\ncloud albedo.\n2. Longwave radiation, whose calculation includes a detailed procedure for\nabsorption by water vapor, as well as precalculated O3 and CO2 transmission\nfunctions. The ground and variable clouds emit as blackbodies.\n3. Sensible processes, which include surface sensible fluxes, vertical diffu-\nsion, and dry convective adjustment, and\n4. Latent processes, including large-scale precipitation, middle-level, pene-\ntrative and low-level cumulus convection (Arakawa, 1972), as well as moist\nadiabatic adjustment.\nThe set of four heating components from the GLA run, averaged for 24\nhours, are plotted at the 850 mb level in Fig. 1. Shortwave processes have\nmaximum heating in low southern latitudes, and cooling occurs by longwave ra-\ndiation, with a maximum in the tropics. Maxima in sensible heating occur over\nthe southern continents and in the Northern Hemisphere over the western bound-\nary currents in the oceans where relatively warm water heats the lower atmo-\nsphere. Latent heat maximizes in a band across the tropics. Further separa-\ntion of these fields by synoptic hour (four times daily) reveals strong diurn-\nal variability in some of the fields. The most pronounced variation occurs in\nthe sensible component, values of which are high during the 6 hours ending 00\nUT over the western Americas, 06 UT over Australia, 12 UT over eastern Africa,\nand 18 UT over eastern South America and western Africa (not shown)\n250","On comparing the heating profiles with those derived using the thermody-\nnamic approach (from T. Schaack, U. Wisconsin), we discovered that insuffi-\ncient heating was being produced by the model approach in the mid-latitude\nmid-troposphere, especially over the oceanic storm tracks. This problem was\nreduced in another set of results in which latent heating values were produced\nusing a version of the parameterization by Arakawa and Schubert (1974) incor-\nporating fractional cloudiness and cloud formation at each of the 9 model\nlayers (Sud and Molod, 1988). A set of 8 independent \"climate\" integrations\nfrom the GLA fourth-order model representing four Januarys and four Julys,\ninitialized from ECMWF analyses 15 days prior to each month, were run with the\nArakawa-Schubert scheme. The problem with the older Arakawa parameterization\n(Fig. 2a) no longer occurs in the Arakawa-Schubert scheme (Fig. 2b), where\nheating is carried up much higher in mid-latitudes.\nThe availability of this eight-run set also allowed us to examine the vari-\nability of the heating components at a number of time scales. Variability on\nintramonthly, intraseasonal, and interannual scales are displayed for latent\nheating in Fig. 2c-e. The strong intramonthly variability represents the lo-\ncally transient nature of latent heating. Interseasonal variability is repre-\nsented by the difference between mean January and July values; in low lati-\ntudes larger values occur in the summer hemisphere, and smaller values in the\nmid-latitude lower troposphere in the winter hemisphere. The largest interan-\nnual variability occurs in the tropics, as well.\nWe have also investigated the initial adjustment of the GLA model by\nexamining heating at the first time step archived and its subsequent evolu-\ntion. An imbalance created in part by initializing the GLA model with ECMWF\nfields appears to have produced unreasonably strong latent heating in the\ntropics in the first time step (e.g. 15 December 1978) several days are re-\nquired to reduce these excessive values. However, no problem appears to occur\nin mid-latitudes where the latent heat value at the first time step is repre-\nsentative of the rest of the simulation. Because each archived time in the\ndata assimilation run involves new initial conditions, we may surmise that the\ndifference in mid-latitude heating between the new and old GLA versions is due\nto the change in model physics, and not to the difference between a data\nassimilation run and a climate integration.\nThe relative strength and sign of the integration-produced diabatic heating\ncomponents are displayed in Fig. 3, where they are vertically averaged and\nplotted by latitude. Latent heating dominates the shape of the total heating\nprofile, with sensible heating weaker at all latitudes. As expected, short-\nwave heating vanishes at the north (winter) pole, and longwave processes are\nnegative everywhere. It is interesting that latent processes provide a posi-\ntive generation of zonal APE because of their positive correlation with tem-\nperature, whereas longwave processes have just the opposite effect. (Calcula-\ntions of numbers for these generation terms will be reported elsewhere.) With\na suitable model, these statistics are potentially useful for climate\ndiagnostics.\nFGGE reanalyses that include archives of diabatic heating components have\nrecently been produced by other meteorological centers. At the Geophysical\nFluid Dynamics Laboratory (GFDL) a continuous insertion method was used in\nconjunction with a 42 wave, 18 level spectral model. The \"E4\" physics package\n(Gordon and Stern, 1982) was used in the reanalysis, and preliminary diagnos-\ntics were provided for part of the FGGE first SOP. Likewise, the National\nMeteorological Center (NMC) ran a reanalysis for part of this period, with the\nMRF 86 model, using 40 spectral waves and 18 levels as its basis. Although\nthe radiational heating parameters in the GFDL and NMC runs were affected by\nan unrealistic zonal cloud structure, the sensible and latent heating appear\nquite reliable. Latent heating from these centers is shown in Fig. 4. Like\nthe GLA Arakawa-Schubert simulations (Fig. 2b), latent heat is carried up into\nthe mid-latitude mid-troposphere in both GFDL and NMC fields. Somewhat weaker\n251","tropical latent heating at NMC results in part from three independent compu-\ntations, in which negative values from the shallow convective process in the\ntropical mid-troposphere reduce the strong positive values generated by deep\nconvection in that region, with little effect from large-scale condensation.\n3. Conclusions\nIt is clear that atmospheric diabatic heating fields are important climate\ndiagnostics; however, heating components can be calculated only within a model\napproach by using parameterization schemes. Fields that are produced several\ntimes daily are most valuable because of the strong diurnal cycle evident in\nsome of the components. We have found that latent heating appears to be most\ncritical for the heat budget. For this latent heating component, the GFDL,\nNMC and the GLA (Arakawa-Schubert) fields appear to be reasonable at mid-\ntropospheric levels in mid-latitudes. The variability of the heating terms,\nespecially on the intramonthly scale, as derived from the GLA climate runs, is\nlarge compared to the mean. Differences in heating from the different centers\ndo exist; nevertheless values of sensible and latent heating appear in general\nto be satisfactory. On the other hand, problems in radiational heating from\nthe GFDL and NMC runs are evident. Because the radiation schemes have already\nbeen updated at these centers, such problems can be eliminated in future\nanalyses.\n4. Acknowledgments\nY. Sud of the NASA Goddard Laboratory for Atmospheres kindly provided most\nof the GLA data used here; we benefited from useful discussions with him. We\nthank W. Stern of GFDL and B. Katz of NMC for providing their respective data\nsets. The research was sponsored by NASA under contract number NAS5-29377 and\nNSF under grant number ATM-8521788.\n5. References\nArakawa, A. 1972: Design of the UCLA General Circulation Model. Tech. Rep.\nNo. 7, Dept. Meteor., UCLA.\nArakawa, A. and W. H. Schubert, 1974: Interaction of cumulus cloud ensemble\nwith the large-scale environment, Part I. J. Atmos. Sci., 31, 674-701.\nGordon, C. T. and W. F. Stern, 1972: A description of the GFDL Global\nSpectral Model. Mon. Wea. Rev., 110, 625-644.\nKalnay et al., 1983: Documentation of the GLAS fourth-order general\ncirculation model. NASA Tech Memo. 86064, Vol. 1.\nKasahara, A., A. P. Mizzi, and U. C. Mohanty, 1987: Comparison of global\ndiabatic heating rates from FGGE Level IIIb analyses with satellite\nradiation imagery data. Mon. Wea. Rev., 115, 2904-2935.\nSchubert, S. D. and G. F. Herman, 1981: Heat balance statistics derived from\nfour-dimensional assimilation with a global circulation model. J. Atmos.\nSci., 38, 1891-1905.\nSud, Y. and A. Molod, 1988: The roles of dry convection, cloud-radiation\nfeedback processes and the influence of recent improvements in the para-\nmeterization of convection in the GLA GCM. Mon. Wea. Rev. 116, 2366-2387.\nWei, M.-Y., D. Johnson, and R. D. Townsend, 1983: Seasonal distributions of\ndiabatic heating during the First GARP Global Experiment. Tellus, 35a,\n241-255.\n252","80N\n60N\n4ON\n20N\n20S\n40S\n60S\n80S\n80N\n60N\n40N\n20N\n20S\n40S\n60S\n80S\nDiabatic heating components at 850 mb from the GLA assimilation run (daily means) for\no\nO\n180\n180\n850 MB\n850 MB\n>>\n90E\n90E\nSENSIBLE\nLATENT\no\no\nGLA/ASSIMILATION\n2\n6 January - 4 February 1979. Solid line contours are 2 °K day-1\n0\n90W\n90W\n-\n0\n180\n180\n180\n180\n850 MB\n850 MB\n90E\n90E\nSHORTWAVE\nLONGWAVE\no\no\n2\n90W\n90W\nFig. 1.\n*\n180\n180\n80N\n60N\n40N\n20N\n20S\n40S\n60S\n80S\n80N\n60N\n40N\n20N\n20S\n40S\n60S\n80S\no\nO","90N\n90N\n90N\ne\nINTRAMONTHLY VARIABILITY, 4 JANUARYS\nINTERANNUAL VARIABILITY, 4 JANUARYS\n60\n60\n60\nDIFFERENCE, 4 JULYS - 4 JANUARYS\n30\n30\n30\no\no\n0\n30\n30\n30\n6.5\n0.5\nGLA LATENT\n60\n60\n60\nP(MB)\n90S\nP(MB)\nP(MB)\n90S\n90S\n100\n150\n200\n250\n300\n400\n500\n700\n850\n925\n1000\n100\n150\n200\n250\n300\n400\n500\n700\n850\n925\n1000\n100\n150\n200\n250\n300\n400\n500\n700\n850\n925\n1000\n(a) Zonal mean values of latent heating from the\nGLA assimilation run (6 January - 4 February\n(c) Mean of the root-mean-square - (rms) values\n(d) Difference in latent heating between mean\n(e) Rms values of 4 January zonal means of latent\nSame as (a) but for a climate integration re-\naround a latitude band of rms daily values\nduring each of 4 Januarys for latent heating\npresenting January 1979 using the Arakawa-\n90N\n90N\n-1\nb\na\nAll are based on daily means; units are °K day\n1979) using the Arakawa parameterization.\n60\n60\nfrom GLA climate integrations.\nARAKAWA-SCHUBERT -\nJuly and mean January values.\n30\n30\n1.0\n2.01\nARAKAWA\n0\no\n2.0\nSchubert scheme.\n30\n0.5\n30\nheating.\n60\n60\nFig. 2.\nP(MB)\nP(MB)\n90S\n90S\n(b)\n500\n100\n300\n400\n700\n850\n925\n850\n200\n1000\n100\n200\n300\n400\n500\n700\n925\n1000","GLA/CLIMATE\nVERTICAL MEAN\n3\nLATENT\n2\n1\nSHORTWAVE\n0\nSENSIBLE\nTOTAL\n1\nLONGWAVE\n-2\n-3\n90\n60\n30\nO\n30\n60\n90\nS\nN\nLATITUDE\nFig. 3.\nZonal, vertical and daily mean values of 4 diabatic heating\ncomponents, and total heating, from the mean of 4 GLA January\nclimate integrations. Units are °K day-1\nLATENT\nGFDL\nP(MB)\n100\n200\n300\n400\n500\n700\n850\n0.5\n2\n925\n1000\n90S\n60\n30\no\n30\n60\n90N\nNMC\nP(MB)\n100\n200\n0.5\n300\n400\n500\n700\n0\n850\n925\nT\n1000\n90S\n60\n30\nO\n30\n60\n90N\nFig. 4.\nZonal mean of latent heating from the GFDL and NMC FGGE\nreanalyses, for the period 6 January - 4 February 1979, for\nUT. Units are °K day-1\n00\n255","Zero Potential Vorticity Gradient Basic States in the\nNeighborhood of the Equator\nRichard S. Lindzen & Volkmar Wirth\nCenter for Meteorology and Physical Oceanography, M.I.T.\nSumant Nigam\nDepartment of Meteorology, University of Maryland\nIn this talk we investigate zonally averaged basic states for linearized stationary wave models,\nwhere the basic state zonal winds have been modified in the neighborhood of the equator in such a\nmanner as to eliminate meridional gradients in the pseudo-potential vorticity near the equator. The\nunmodified basic states commonly used (i.e., Jacqmin and Lindzen, 1985, Nigam and Lindzen, 1989)\nin linear models typically have large potential vorticity gradients near the equator (~ B). There are\ntwo related aims to this study:\n1. To see whether the modified basic states differ from the unmodified basic states to a measurable\nextent; and\n2. To see whether the response to stationary forcing (especially to tropical heating) would be\naffected by the modification to the basic state.\nThe motivation for the study comes from a common shortcoming in linearized stationary wave\nmodels: namely, although the response to tropical forcing in geopotential height appears superficially\nreasonable, the associated wind fields are generally excessive. Large values of friction are used in\nthe neighborhood of the equator in order to bring the wind fields down to reasonable values. Now,\nto the extent that the forced waves are Rossby wave-like, their potential vorticity comes from the\npotential vorticity gradients in the basic state (Lindzen and Schoeberl, 1982). It therefore seemed\npossible that eliminating such gradients near the equator might reduce the response to stationary\nforcing. In addition, regions of zero potential vorticity gradient might also affect stationary waves by\ncausing the reflection of stationary waves (Lindzen and Tung, 1977). It has been noted by Nigam,\net al (1986), that nonlinear GCM's do not have this problem at the equator. A possible reason for\nthis might be that wave-mean flow interactions in the GCM might act to iron out the basic state\ngradients in potential vorticity. However, to the extent that basic states are based on observations,\nwe would hardly be justified in replacing them with states that differ pronouncedly from those that\nare observed; hence item 1 above.\nWe have, thus far, only completed item 1 above1. Our procedure is to replace zonal winds\nwithin\na distance, a, of the equator with winds satisfying the following equation (essentially the equation\nfor zero pseudo-potential vorticity gradient on an equatorial beta-plane):\n(1)\nwhere y = ao, =latitude, and where u(ta) is set equal to the unmodified values. We also\nsmooth the fields in the neighborhood of y = a in order to eliminate kinks in the velocity field.\n1 Since the meeting, work towards answering item 2, above, has been done; pronounced changes in the stationary\nwave response have been found - including both amplitude reductions and increased reflectivity.\n256","Figure 1a shows the unmodified basic state zonal wind and its associated distribution of potential\nvorticity gradients. Figure 1b shows the same for a modified basic state where a corresponds to\n5°; Figure 1c shows the same where a corresponds to 10°. A comparison of Figures la and 1c\nshows that the modified u-field is noticeably different from the original; however, a comparison of\nFigures la and 1b shows that the modified u-field is virtually undistinguishable from the original.\nThus, within about 5° of the equator, observed zonally averaged zonal winds are compatible with\nradically different distributions of potential vorticity gradient. The situation is even more vexing\nwith temperature. Figure 2 compares the zonally averaged temperature and zonal wind fields for\nmodified basic states with a corresponding to 10° and 20°. (The temperatures are in cyclostrophic\nbalance with the wind fields.) The zonal winds are very different, but the temperature field shows\nalmost no discernible differences. We are left with the uncomfortable situation where a variable\nwhich appears to be of fundamental importance to dynamics (namely, potential vorticity gradient)\ndoes not seem to be readily determined from existing data - at least in the immediate neighborhood\nof the equator. Nevertheless, if our modelling studies demonstrate that response is more accurate\nfor zero potential vorticity gradients at the equator, then this might provide compelling reason to\nuse this value regardless.\nAcknowledgements\nThis research was supported by NSF Grant ATM 84-14660, and NASA Grant NAGW-525 at M.I.T.,\nand by NSF Grant XXXX at the University of Maryland.\nFigure Legends\nFigure 1 (a) y - Z cross sections of u and potential vorticity gradient for unmodified basic state,\n(b) same for modified basic state with a = 10°, (c) same for modified basic state with a = 5°.\nFigure 2 Modified basic states for a = 10° and a = 20°. Wind is in m/s and temperature is in deg\nK.\nReferences\nJacqmin, D. and R.S. Lindzen, 1985: The causation and sensitivity of the northern winter planetary\nwaves. J. Atmos. Sci.,42, 724-745.\nLindzen, R.S. and M.R. Schoeberl, 1982: A note on the limits of Rossby wave amplitudes. J.\nAtmos. Sci., 39, 1171-1174.\nLindzen, R.S. and K.K. Tung, 1978: Wave overreflection and shear instability. J. Atmos. Sci.,\n35, 1626-1632.\nNigam, S., I.M. Held, and Lyons, S.W., 1986: Linear simulation of the stationary eddies in a general\ncirculation model. Part I: The no-mountain model. J. Atmos. Sci., 43, 2944-2961.\nNigam, S. and R.S. Lindzen, 1989: The Sensitivity of stationary waves to variations in the basic\nstate zonal flow. Accepted by J. Atmos. Sci.\n257","UNMODIFIED ZONAL WIND U\nPVGRAD OF U\n50\n50\n40\n40\nL\n646.\nH\nL\n1343\nH\n457\n30\n30\nL\n750.\n619.\nto\n9\n20\n20\n0\n229\n187629\nH\n10\n10\n363\n2\n577.\n597\n1220\n0\n0\n-30\n-20\n-10\n0\n+10\n+20\n+30\n-30\n-20\n-10\n0\n+10\n+20\n+30\n(a)\nMODIFIED ZONAL WIND U\nPVGRAD OF U MOD\n50\n50\n40\n40\n1828\nH\n134\n457.\nH\n30\n30\n178\nel\n$9.\n488\n20\n20\nH\n87626\n10\n10\n36:\n2\n835\nH\n220\n0\n0\n-30\n-20\n-10\n0\n+10\n+20\n+30\n-30\n-20\n-10\n0\n+10\n+20\n+30\n(b)\nMODIFIED ZONAL WIND U\nPVGRAD OF U MOD\n50\n50\n40\n40\nH\nL.\n15\n134\n457\n30\n30\nto\n9\n20\n20\n13\n87625\n10\n10\n36\n2\n220\n0\n0\n-30\n-20\n-10\n0\n+10\n+20\n+30\n-30\n-20\n-10\no\n+10\n+20\n+30\n(c)\nFigure 1\n258","MOD ZONAL MEAN WIND, Z = 0...50 KM, PHI = -90 +90 DEG\n50\nkm\n40.0\n20 0\n20.0\n0.0\nH\n0.0\n20.0\n0 km\nMOD ZONAL MEAN TEMP, Z = 0...50 KM, PHI = -90...+90 DEG\n50 km\n240.\n240.\n8\n240\n240\n280\n206.\n0 km\nSP\nNP\nEq\nb)\nMOD ZONAL MEAN WIND, Z = 0...50 KM, PHI = -90...+90 DEG\n50km\n78.5\n20.0\n0.0\nH\n20.0\n20.0\n0 km\n-49.4\nMOD ZONAL MEAN TEMP, Z - 0...50 KM, PHI - -90...+90 DEG\n50 hm\n240.\n240.\n198.\n8\n240\n240\n203\n280.\nSP\nEq.\nNP\nF igure 2\n259","Model Simulation of Atmospheric Circulation and Rainfall\nDuring June 1988 and the Low Frequency Vacillations in a\nPerpetual Run\nY. C. Sud, W. K.- M. Lau, G. K. Walker and W. E. Smith\nLaboratory for Atmospheres\nNASA/Goddard Space Flight Center, Greenbelt MD 20771\nUsing a 4° latitude X 50 longitude version of the nine sigma-layer Goddard Laboratory for\nAtmospheres (GLA) general circulation model (GCM) of Kalnay et al. (1983), we conducted an\ninvestigation to determine if the observed June 1988 sea-surface temperature (SST) anomalies\n(Fig. 1) were responsible for the atmospheric circulation and rainfall anomalies observed during\nJune 1988. Trenberth and Branstator (1989) have shown with a simple model that the\nparticular structure of the SST anomaly can force a stationary circulation that can in turn\nproduce the observed circulation anomalies over continental North America.\nTwo runs, both of which started from the analysed initial conditions based on the ECMWF\nanalysis of observations for May 15 1988 at 00Z, were differentially analysed; one used the\nclimatological SSTs and the other used the anomalous SSTs produced by superposing the 1988 SST\nanomaly observations on the climatological SSTs. The model was able to reproduce the observed\ndroughts over North America but could not reproduce the observed stationary wave pattern\n(Fig.2b). See Climate Diagnostics Bulletine for June and July 1988 for a comparison with\nanalysis of observations. Indeed, the excessive inital rainfall in the simulation virtually\ndestroyed the stationary wave pattern in the initial conditions and also gave rise to very wet soil\nmoisture conditions over North America. Presumably, this was responsible for the unrealistic\nsimulated patterns of June circulation. Because, in an extended run which allowed time for\nstationary circulation to redevelop and the land to dry out, the rainfall and 500mb geopotential\nheight and surface wind anomalies over North America and several other regions of the world\ndeveloped to resemble the observations for June 1988 (Fig. 3).\nIn an attempt to statistically assess the above results, it was necessary to determine the\nnatural variability of the model simulations. A perpetual July run was made in which the sun and\nthe SSTs were held constant for mid-July conditions. However, the soil moisture was allowed to\nevolve in response to surface water balance: rainfall, evapotranspiration, and gravitational\n260","drainage of water within a three layer soil moisture parameterization which is in the Biosphere\nmodel of Sellers et al. (1986) that is used in the GCM.\nIn a 365 day perpetual July simulation with the GLA GCM, it was found that even the\nconsecutive 30 day averages of most meteorological fields exhibit a large variability everywhere,\nincluding the tropics (Fig. 4a). The rainfall variability is large in all the convectively active\nregions even though the accompanying evaporation/evapotranspiration variability is much\nsmaller (Fig. 4b). Moreover, the regions of large evaporation variability do not necessarily\ncorrespond with the regions of large rainfall variability. This clearly indicates that the\nevaporation variability introduced as a cosequence of interactive soil moisture is not the primary\ncause for this vacillatory behavior of the atmosphere.\nA preliminary analysis of these results suggests that seasonal climate prediction as a\nboundary forced problem might suffer from some severe limitations unless models can\nsuccessfully forecast the low frequency behavior of the atmosphere. The mechanisms behind the\nunforced low frequency oscillation of the model simulations and its relationship with the\nobservations is currently being investigated. Future experiments to determine the roles of\ninteractive soil moisture and other physical processes such as cumulus convection on the\nmaintenance of these oscillations are planned.\nReferences\nKalnay, E., R. Balgovind, W. Chao, D. Edelmann, J. Pfaendtner, L. Takacs and K. Takano,\n1983: Documentation of the GLAS fourth order general circulation model.\nVolume I., NASA Tech. Memo. 86064, NASA Goddard Space Flight Center,\nGreenbelt, Maryland.\nSellers, P. J., Y. Mintz, Y. C. Sud, A. Dalcher, 1986: A simple biosphere model (SiB)\nfor use within general circulation models. J. Atmos. Sci., 43, 505-531.\nTrenberth, K. E. and G. W. Branstator, 1989: Global circulation anomalies during 1988. Third\nInternational Conference on Southern Hemisphere Meteorology and Oceeanography. Buenos Aires,\nArgetina, Nov. 13-17, 1989.\n261","or\n40N\n2DN\nEO\n205\n405\n11\n90E\n180\n90W\n0\n90€\n18J\nFig. 1. Analysed sea surface temperature anomaly for June 1988.\n7\nII\n2\n10N\n20N\nEQ\n2\n20S\n40S\no-\n6\n4ON\n20N\n2\nEC\n3\n2\n20S\nEi\n3\n40S\ncl\n18\n90M\n0\n90€\n90E\n180\nFig. 2 a) Top: Rainfall anomaly in mm/day and b) bottom: 500mb height anomaly in\ngpm simulated by the GCM for June 1988.\n0\n0\n4ON\n&\n3\n,\n2\n20N\nEO\n20S\n6\n40S\n02\nAON\n10\n(2)\n20N\nEQ\n20S\n2.\n40S\n90E\n90€\n180\n180\nSOW\n0\nFig. 3 a) Top: Rainfall anomaly in mm/day and b) bottom: 500mb height anomaly in\ngpm simulated by the GCM for the next 30 days in an extended run.\n262","40N\n1\n20N\nEQ\n40N\n2\n2ON\na\nEQ\n40N\nif\n3\na\n20N\nEQ\n40N\nTV\n4\n20N\nEQ\n40N\n12\n20N\na\n2\nEQ\n40N\n6\n20N\nEO\n40N\n7\n20N\nEQ\n4ON\n8\n2ON\nEQ\n40N\n9\n20N\nRD\n2\nEQ\n40N\n20N\ni\nEQ\n40N\n12\n20N\n17\nEQ\n4ON\n12\n0\n2ON\nEO\n45E\n90€\n135E\n180\n180\n135W\n90W\n45W\n0\nL ONGITUDI\nFig. 4a Twelve consecutive maps of 30 day average rainfall in mm/day produced by\nthe porpetual simulation for mid-July prescribed solar declination and sea-\nsurface temperature shown for 0-40N.\n263","40N\n20N\nEQ\n40N\nYV\nSD\n20N\nEQ\n40N\nM\n0\n2ON\nEQ\n40N\n2\n20N\nas\n3\nEQ\nFig. 4b Mean and standard deviations of evaporation(top panels) and rainfall (bottom\npanels) in mm/day, respectively. The standard deviation calculation is based on\n30 day consecutive means. Only the region from 0-40N is shown.\n264","A 3-D Dynamical/Chemical GCM for Simulating the Anthropogenical Effects on Ozone\nC. Y. Jim Kao, G. A. Glatzmaier, and R. C. Malone\nLos Alamos National Laboratory\nLos Alamos, NM 87545\nR. P. Turco\nUCLA\nLos Angeles, CA 90024\n1. Introduction\nIn spite of the significantly increased attention atmospheric ozone has received over the last\ntwo decades or so, a fully quantitative understanding of the three-dimensional ozone structure\nremains unavailable. In the past, one-dimensional (vertical column) models have provided most\nof the quantitative simulations of atmospheric chemical systems, because these models can treat\nlarge ensembles of chemical process economically. More recently, two-dimensional (meridional cross\nsection) models have become popular, in part because of increasing computer capabilities. There\nhas been little work using a sophisticated 3-D GCM coupled with a comprehensive photochemical\nmodel to study the chemical-dynamical interactions involving the ozone abundance. Namely, either\nthe dynamics or the chemistry is too simplified in most of the existing models (see a review in Kaye\nand Rood, 1989). Relatively more advanced photochemistry has been included in the stratospheric\nmodels (dealing the vertical domain from 20 to 70 km above the surface) at the expense of tro-\npospheric details (Kaye and Rood, 1989; Grose et al., 1987). An important motivation for this\nresearch has been the desire to estimate the magnitude of changes in the ozone layer induced by\nanthropogenic perturbations such as the injection of oxides of nitrogen and chlorine compounds.\nAccordingly, the attempt has been to develop a complete but feasible photochemical scheme and\nto combine it with the Los Alamos National Laboratory (LANL) 3-D GCM (Malone et al., 1986).\n2. The Model\nWe have utilized the NCAR CCM0A (Malone et al., 1983) modified to predict the effects\nof smoke on the global atmosphere. Particular modifications which benefit this current research\ninclude: a) stability-dependent vertical diffusion, b) subsurface storage and diffusion of heat and\n265","moisture, c) annual and diurnal cycles of sunlight, d) finer zoning in stratosphere, e) Langrangian\ntransport of aerosols, and f) improved 3-D transport scheme for chemical species.\nThe model results for the unperturbed atmosphere have also been validated against the long-\nterm climate analyzed by ECMWF (European Center for Medium-Range Weather Forecast). The\nmodel history tape for a summertime condition has provided meteorological conditions to the\nPNL's (Pacific Northwest Laboratory) Global Chemistry Model to evaluate the role of fair weather\ncumulus clouds in the long-range transport and chemistry of methane (Berkowitz et al., 1989).\nThe complete photochemical system in the stratosphere includes several dozen trace species\nand hundreds of photochemical processes. Our approach toward simplifying the photochemical cal-\nculations utilizes a \"family\" scheme (e.g., see Turco and Whitten, 1977) in which the key chemical\nconstituents are organized into groups that have relatively short inter-species equilibration times\n(compared to the computational time steps used in the GCM). Within a family, the constituent\nmembers can thus be assumed to reach a quasi-equilibrium state or to have simple time-dependent\nbehaviors that are amenable to analytical representation. On the other hand, the family itself is\nselected to have a photochemical time constant that is longer than (or equivalent to) the compu-\ntational time step. The families and species considered are summarized in Table 1. The essential\nphotochemical processes and rates are adopted from DeMore et al. (1987) and Turco (1975). The\npresent chemistry set constitutes an initial representation for the principal atmospheric chemical\nfamilies.\nThe fundamental equations that determine the time evolution of the family concentrations\ncan be derived from the complete set of photochemical continuity equations by making signifi-\ncant simplifications. For example, reactions of secondary importance in controlling the chemical\ntransformation (production and loss) rates of families are neglected; less important families in\nstratosphere (e.g., sulfur compounds) are omitted; and continuity equations for individual family\nmembers are summed over the family to eliminate reactions that simply cycle family constituents\n(i.e., intra-family conversion processes). Care must be taken in any simplifying approach to identify\nall of the key rate-limiting reactions (for example, those which control the odd-oxygen loss rate)\nand dominant cycling processes that fix the relationship between family members. in Table 2.\n266","3. Model Performance\nTo assure ourselves that the \"family\" photochemical scheme is numerically stable and compu-\ntationally feasible, the photochemical scheme described above has been implemented in the LANL\nGCM to test the effects of NO deposited by nuclear fireballs on ozone depletion under a post-\nwar condition by Kao et al., 1989. In essence, the large amount of instantaneous injection of NO\nduring the explosions rapidly converts into NOx which, in turn, produces a large amount of N2O5\nand HNO3. N2O5 is later involved with the transfer of NOx, and HNO3 can be photodissociated\ninto HOx. Both NO and HOX can react with odd-oxygen family members to deplete ozone. The\nnet results on ozone abundance are illustrated in Figs. 1 and 2. During this highly nonlinear\nprocess, the modified circulation due to the solar heating of smoke contributes to push the NOx\nrapidly upward into the levels where the maximum ozone concentration resides. The heating of\nthe stratosphere by rising smoke also speeds up several catalytic chemical reactions. Furthermore,\nthis chemical/dynamical model is run with an \"outline\" fusion SO that the diurnal cycle of the\nsunlight associated with photodissociation is fully implemented and the effects of ozone abundance\non the radiative transfer can be included. In contrast, most of the existing model are run in an\n\"off-line\" fashion in which the chemical module is driven by either the analyzed data of winds and\ntemperature or the data set generated from a GCM.\n4. Future Work\nWe will extend the top of the vertical domain from the present 33 km to about 70 km to cover\nthe whole stratosphere. The photo emical processes and their interactions with the transport\nmechanism will then be better resolved. In particular, the photodissociation of O2, which is the\ndominant source term for odd-oxygen family, will appropriately be included. Secondly, the numer-\nical difficulties due to the wave reflection from the rigid top boundary can be reduced. We will\nalso add families for chlorine, bromine, and hydrocarbon species, bringing the total number of con-\nstituents to roughly 65; i.e., a complete photochemical set and we will formulate a comprehensive\nset of photochemical processes for the GCM using the recommendation of the Panel of Chemical\nKinetics and Photochemical Data Evaluation (Demore et al., 1987).\nAcknowledgements\nThis research is jointly supported by the U.S. Department of Energy and Defense Nuclear\nAgency.\n267","TABLE 1. PHOTOCHEMICAL CONSTITUENTS\nActive\nTracer Fixed/Source\nFamily\nSpecies\nSpecies\nSpecies\nOdd-Oxygen\no, O3, O Superscript(1))\nO3\nO2\n(Ox)\nNitrogen\nNO, NO2, NO\nNO\nN2/N2O\nOxides\n(NOx)\nDinitrogen\nN205\nN205\n-\nPentoxide\nNitric Acid\nHNO3\nHNO3\n-\nWater Vapor\nH2O\nH2O\nH2,CH4\nHydrogen\nOH, HO2\nHOx\n-/H2O\nRadicals\n(HOx)\nHydrogen\nH2O2\nH2O2\n-\nPeroxide\nReferences\nBerkowitz, C. M., D. J. Luecken, D. C. Bader, G. A. Glatzmaier, and F. M. Vukovich, \"The role of\nfair weather cumulus clouds in long-range transport,\" Proceeding Abstract in 1989 Annual AGU\nMeeting, Baltimore, EOS, 70, 15, 287, 1989.\nDeMore, W. B., M. J. Molina, S. P. Sander, D. M. Golden, R. F. Hampson, M. J. Kurylo, C. J.\nHoward, and A. R. Ravishankara, \"Chemical kinetics and photochemical data for use in strato-\nspheric modeling: Evaluation number 8,\" JPL Publ. 87-41 (1987).\nGrose, W. 1, J. E. Nealy, R. E. Turner, and W. T. Blackshear, \"Modeling the transport of chemically\nactive constituents in the stratosphere,\" in Transport Processes in the Middle Atmosphere, edited\nby G. Visconti and R. Garcia, pp. 229-250, D. Reidel, Hingham, MA (1987).\nKao, C.-Y. J., G. A. Glatzmaier, R. C. Malone, and R. P. Turco, \"Three-dimensional simulations\nof ozone depletion under a post-war condition,\" submitted to J. Geophys. Res. (1989).\nKaye, J. A., and R. B. Rood, \"Chemistry and transport in a three-dimensional stratospheric model:\nchlorine species during a simulated stratospheric warming,\" J. Geophys. Res., 94, 1057-1083, 1989.\nMalone, R. C., L. H. Auer, G. A. Glatzmaier, M. C. Wood, and O. B. Toon, \"Nuclear winter:\nThree-dimensional simulations including interactive transport, scavenging, and solar heating of\nsmoke,\" J. Geophys. Res., 91, 1039-1054 (1986).\n268","Turco, R. P., and R. C. Whitten, \"Chlorofluoromethanes in the stratosphere and some possible\nconsequences for ozone,\" Atmos. Env., 9, 1045-1061 (1975).\nTurco, R. P., and R. C. Whitten, \"The NASA Ames Research Center one- and two-dimensional\nstratospheric models: Part I, The one-dimensional model,\" NASA Tech. Paper 1002, 29 pp.,\n(September 1977).\nFigure Captions\nFig. 1. (a) Initial zonal-averaged ozone concentrations in units of molecules cm-3. The contour lines\nare scaled by 10-10, (b) the same as (a) except at 20 days after the initial condition. The decreased\nconcentrations in the Northern Hemisphere are due to both the chemical reactions associated with\nNOx and the transport mechanism associated with the heating anomalies due to the absorption\nof sunlight by smoke. The increased concentrations in the Southern Hemisphere are due to the\ntransport.\nFig. 2. (a) Initial ozone column number density in units of molecules cm-2. The contour lines are\nscaled by 1 X 10-18 - , (b) the same as (a) except at 20 days after the initial condition. The readers\nare referred to the caption of Fig. 1.\n(a)\n(b)\n03\n03\n10\n20.\n60.\n30\n30\n200.\n20\n240.\n290\n20\n120\n99\n200.\n280\n25\n25\n30\n30\n298\n260.\n-50\ni50\n200\n280.\n20\n20\n160\nA\n,40\n70\n70\nM\nP\n309\n20.\n100\n60.\n320\n00\n00\n8\n280\nIS\nIS\n10.\ndi\n160\n160\n120\n-200\n700\n40.\n12.\n0\n10\n300\n000\n10.\n23.\n500\n500\nS\n5\n-700\n'00\n4850\nHSO\n1000\n0\nxx\n0\n.40\n-50\n60\n70\n80\n-90\n90\n80\n70\n50\n50\n40\n30\n20\n10\nJ\n0\n-10\n-20\n30\n10\n20\n30\n40\n50\n50\n70\n80\n70\n90\n80\n70\n60\n50\n40\n30\n20\n10\nLATITUDE\nCATITUDE\nFig. 1.\n269","(a)\nTOP VERT INTEGRAL\n03\nSIGMA = 1.0000\n90\n04\n96.\n60\n96\n96.\n88.\n68.\n88.\n80\n80.\nAn\n12.1\n32\n72\n64\n64\n64\n30\n56\n56:\n56:\n48\n48,\n48\n40\n40.\n40\n0\nL\n34\n10.\n40.\n40.\n98\n48.\n48.\n-30\nS6\n56.\n56.\n64\n64.\n64.\n72.\n72.\n72.\n-60\n80.\n80.\n80,\nH\nH\nyear\n84\n80.\n80.\n72\n-90\n-180\n-150\n-120\n-90\n-60\n-30\n0\n30\n60\n90\n120\n150\n180\n(b)\nTOP VERI INTEGRAL\n03\nSIGMA = 1.0000\n90\nGRU\n70.\nto\nH\n60\n79\nare\n60\n30\nSo.\n50\nH\n4F\n02\n0\n50\nso\n60\nSb\nZO\n60\n80\n80\n-30\ni\n90\no\n103.\nH\n-60\n90\nwe\n1061\n-90\n-180\n-150\n-120\n-90\n-60\n-30\n0\n30\n60\n90\n120\n150\n180\nLATITUDE\nFig. 2.\n270","Preliminary Coupling Experiments\nof an Atmospheric and an Oceanic GCM\nM. Fisher1, M. Ghil1,2 , D. Halpern3,1 , C.R. Mechoso¹, J. Spahr¹\n1 Climate Dynamics Center and Department of Atmospheric Sciences, University of\nCalifornia, Los Angeles.\n2 Institute of Geophysics and Planetary Physics, University of California, Los An-\ngeles.\n3 Jet Propulsion Laboratory, Caltech/NASA, Pasadena, California\nIntroduction\nCoupled atmosphere-ocean general circulation models (GCMs) are essential\ntools for the study of climate change on the time scales of months to millennia.\nHowever, systematic errors are invariably present both in atmospheric general cir-\nculation models (AGCMs) and in oceanic general circulation models (OGCMs);\nthese errors can be amplified and result in large errors and climate drift when such\nmodels are run in a coupled mode, free from the constraint of prescribed climato-\nlogical boundary conditions at their interface.\nWe are developing a coupled atmosphere-ocean GCM for use in data assim-\nilation, predictability studies, and climate change. In this paper, we compare\nsimulations of an OGCM in which surface wind stresses are prescribed either as\nclimatological values or from an AGCM.\nThe models\nThe OGCM is the primitive-equation tropical Pacific model developed at GFDL\n(Philander and Pacanowski, 1981). The model covers the region between 28°S and\n50°N and from 130°W to 70°E, with variable resolution in latitude - from 1/3°\nat the Equator to 3° at 50°N, and in the vertical - from 10m near the surface to\n650m at the uniform depth of 4000m, and with 1° resolution in longitude.\nThe AGCM is the UCLA atmospheric GCM developed by Prof. A. Arakawa\nand collaborators (Suarez et al., 1983). It incorporates an advanced finite-difference\nrepresentation of dynamical processes, and state-of-the-art parameterizations of\nconvection, the planetary boundary layer, and radiation. The data used for this\nstudy were taken from a three-year control integration of a version of the model\nwith 9 vertical layers and a horizontal resolution of 5° long. by 4° lat.\n271","Simulations\nFor the control simulation, we repeated the experiment performed by Philander\nand Siegel (1985). The OGCM was run for three years. Atmospheric forcing was\nprescribed using monthly-mean observed wind stress (Hellerman and Rosenstein,\n1983) and surface heat fluxes calculated from climatological surface temperatures.\nAs a first step towards full coupling of the OGCM and AGCM, we performed\nan OGCM simulation, which we call Experiment 1, in which surface wind stress\nwas prescribed as monthly-mean values taken from a multi-year simulation of the\nAGCM using climatological sea-surface temperature (SST). The SST distribution\nwas maintained at the climatological distribution used in the AGCM run by calcu-\nlating at each time step a surface heat flux which exactly compensated the tendency\nof the SST to depart from climatological values. The OGCM was run in this mode\nfor four years.\nResults\nFig. 1a shows monthly-mean wind stress for January from the AGCM. The ob-\nserved wind stress, used in the control simulation, is shown in Fig. 1b. The AGCM\nwind stresses are generally similar to in direction, but weaker than, the observed.\nThis discrepancy results in significantly weaker ocean currents in Experiment 1\nthan in the control simulation. Cross-sections along the Equator of zonal cur-\nrent for January from the OGCM simulations are shown in Fig.2. The Equatorial\nCounter-Current is approximately twice as strong in the control simulation as in\nExperiment 1. Surface currents are up to four times stronger along the Equator.\nThe discrepancy can be attributed, in part, to the fact that Hellerman and Rosen-\nstein (1983) used too high a value of the drag coefficient in deriving the wind stress\nfrom observed monthly-mean winds (Harrison et al., 1984).\nFig.3a shows heat flux across the ocean surface for January from the AGCM\nsimulation. The amplitude and pattern of heat flux are similar to observations (not\nshown). The heat flux required in Experiment 1 to maintain SST at climatological\nvalues is shown in Fig.3b Away from the equator, there is broad agreement in sign\nand magnitude. Both fields show strong downward heat flux along the equator; in\nExperiment 1, however, the heat flux is concentrated into a narrow band of values\nup to 300 Wm-2. The latitudinal scale of this band is smaller than the resolution\nof the AGCM.\n272","Conclusions and future work\nSurface heat and momentum fluxes from the AGCM have generally correct\npatterns. The amplitude of the heat flux is approximately correct, but that of\nthe wind stress is too weak. When the AGCM wind stresses are used as surface\nboundary conditions for the OGCM, the resulting ocean currents are too weak. The\nheat flux required to simulate AGCM SST in this case is similar to the observed,\nbut contains features of scale smaller than the resolution of the AGCM. The scale\nof these features is due to the narrow, boundary-layer type equatorial upwelling,\ncaptured by the increased resolution near the Equator of the OGCM.\nTo improve the compatibility of the AGCM and the OGCM for fully-coupled\nsimulations, we shall concentrate on matching the heat and momentum fluxes across\nthe air-sea interface. Variability of these fluxes on a daily, rather than monthly basis\nwill also be studied.\nReferences\nHarrison, D.E., V. Cordone, J.J. O'Brien, D. Reynolds, J. Sadler, D. Wylie, 1984:\nReport of the ad hoc committee on surface wind and wind stress fields. Univ. Corp.\nfor Atmos. Res. USTOGA 2, 40pp.\nHellerman, S. and M. Rosenstein, 1983: Normal monthly wind stress over the world\nocean with error estimates, J. Phys. Oceanogr., 13, 1093-1104.\nPhilander, S.G.H. and R.C. Pacanowski, 1981: Response of equatorial oceans to\nperiodic forcing. J. Geophys. Res.,86(C3), 1903-1916.\nPhilander, S.G.H. and A.D. Siegel, 1985: Simulation of El Niño of 1982-1983.\nCoupled ocean-atmosphere models. J.C.J. Nihoul, Ed., Elsevier Oceanogr. Ser.,\nVol. 40.\nSuarez, M.J., A. Arakawa and D.A. Randall, 1983: The parameterization of the\nplanetary boundary layerin the UCLA general circulation model: Formulation and\nresults. Mon. Wea. Rev.,111, 2224-2243.\n273","AGCM WIND STRESS JANUARY\n(a)\n50\n5.\n40\n30\n20\n10\n0\n-10\n-20\n-30\n130\n150\n170\n190\n210\n230\n250\n270\n290\nLONGITUDE\nOBSERVED WIND STRESS JANUARY\n(b)\n50\n5.\n40\n30\n20\n10\n0\n-10\n-20\n-30\n130\n150\n170\n190\n210\n230\n250\n270\n290\nLONGITUDE\nFig.1. Monthly-mean wind stress (dyne cm-2) for January: (a) from the AGCM;\n(b) observed.\n274","CONTROL ZONAL VELOCITY JANUARY\n(a)\no\n50\n50\n100\n80\n150\n200\n250\n300\n350\n400\n230\n250\n270\n290\n130\n150\n170\n190\n210\nLONGITUDE (DEGREES)\nEXPT 1 ZONAL VELOCITY JANUARY\n(b)\no\n50\nof\n100\n40\n150\n40\n200\n250\n0\n300\n350\n400\n250\n270\n290\n130\n150\n170\n190\n210\n230\nLONGITUDE (DEGREES)\nFig.2. Cross sections of zonal current (ms-1) for January: (a) from the control\n.simulation; (b) from Experiment 1.\n275","AGCM DOWNWARD HEAT FLUX JANUARY\n(a)\n50\n100\ne\n40\n30\n100\nL\n20\n-100\n10\n100\n0\n-10\n-20\n100\n-30\n130\n150\n170\n190\n210\n230\n250\n270\n290\nLONGITUDE\n(b)\nEXPT 1 DOWNWARD HEAT FLUX JANUARY\n50\n40\n30\n-100\n20\n100\n10\n0\n100\n0\n0\n0\n3003\n100\n-10\n0\n-20\n-30\n130\n150\n170\n190\n210\n230\n250\n270\n290\nLONGITUDE\nFig.3. (a) Downward heat flux (Wm-2) across the ocean surface. (b) Heat flux\nrequired in Experiment 1 to maintain SST at climatological values.\n276","Winter Simulation in a Simplified NMC 4-layer Global Spectral Model\nShyh-Chin Chen and John O. Roads\nScripps Experimental Climate Forecast Center\nUniversity of California, San Diego A-024\nLa Jolla, CA 92093\nA simplified primitive equations global GCM has been developed. The model is a lower\nresolution version of the operating NMC spectral model (See Sela, 1980. This model was\nkindly provided by J. Sela) with 4 sigma layers and a spectral truncation of rhomboidal\n20 (R20). A moisture cycle consisting of large-scale condensation and Kuo's convective\nparameterization is included. Through evaporation and sensible heating and surface friction,\nthe model interacts with the underlying surface which has the monthly climatological sea\nsurface temperature fields specified. Due to the relatively low vertical resolution and the\nuse of a \"Lorenz-type\" vertical grid, a computational stabilizing term suggested by Arakawa\nand Moorthi (1988) has been applied at the top and bottom layers of the model to suppress\nwhat Arakawa and Moorthi called baroclinic instability of the computational kind (BICK).\nA unique feature of the model is the use of a stationary residual forcing to empirically\nmimic processes, e.g. radiation cooling (or heating) and planetary boundary processes, that\nare excluded in the model but are responsible for maintaining the observed climatology. The\nresidual forcing can be found empirically, from a large ensemble of short term forecasts, by\ncalculating the ensemble averaged tendency (Chen and Roads, 1988). For example, if the\nuncorrected model is constantly drifting toward a warmer average state then a cooling must\nbe imposed to offset this warm drift. This cooling is equivalent to a crude but skillful pa-\nrameterization for radiational cooling. To calculate the forcing we have used the ECMWF 8\nJanuary (1979-1986) daily analyses as initial fields of short term weather forecasts. Although\na residual forcing could be applied to all prognostic equations, we have initially chosen to\nonly force the thermodynamic and vorticity equations.\nA 1600-day perpetual January simulation has been done. The model reaches an equilib-\nrium state after an initial adjustment period of 20 days. Thereafter it maintains a steady\nglobal mean temperature about .16°C higher than the observed climatology. The equilib-\nrium mean global precipitation rate is about 104 cm per year. The model simulated mean\ngeopotential heights capture most of the observed dominant features in both hemispheres\nin January (see Fig. 1 for 500mb). Both zonal mean subtropical jets are also reasonably\nwell reproduced, despite the fact that their strengths are somewhat stronger in the Northern\nHemisphere and weaker in the Southern Hemisphere compared to observation. The tran-\nsient behavior (RMS of 500mb heights) of the model also has a similar distribution to more\n277","sophisticated models and observations but with slightly weaker intensity. This simulation is\nbeing used for predictability experiments.\nFig. 2 shows the global mean normalized RMS error and pattern correlation (p) for\nthe simulated and observed streamfunctions at layer 3. The model demonstrates persistence\ncharacteristics similar to that observed in that the pattern correlation (error) drops (grows)\nrather quickly with time for the first few days. The average skill for 10 initial error ex-\nperiments (thick dashed curve) is also shown in Fig. 2. Each experiment is restarted from\na model day in the control run (30 days apart) but, at each initial day all amplitudes for\nwavenumbers higher than R15 are set to zero. The introduction of this small initial error\nrepresents (crudely) analyses errors in the current observations. The skill of the model can\nthen be evaluated by comparing the \"predicted\" and the \"observed\" streamfunctions. It is\nfound, with an estimated doubling time of 1.8 days, that the model error (skill) is as large\n(poor) as persistence forecast error (skill) after 20 days.\nWe are especially interested in examining the latitudinal distribution of the predictability\nerror. Fig. 3 shows the error growth as function of latitude and lapsed time for the model\npersistence and initial error experiment. The latitude-dependent skill can also be best shown\nby the Z score () in the bottom panel of Fig. 3. Note that the zonal mean of Z\nscore has been removed to highlight the distribution contrasts. It is found that the model is\nless persistent and less predictable (negative values) in the tropics. The NH mid-latitudes\nseem to be more predictable in the model atmosphere than the SH counterpart and the\ntropics.\nTo conclude, the empirically-forced GCM simulates well the time-mean and temporal\nJanuary circulation. We suggest that the time invariant forcing, which is derived empirically,\nis a skillful alternative to highly computational time consuming radiation parameterization\nand other excluded forcings in the model. Moreover, some preliminary experiments indicates\nthat the model demonstrates similar predictability characteristics as those in simpler models\n(e.g. Roads, 1987) and numerical weather prediction models (Roads, 1986).\nReferences\nArakawa, A., and Shrinivas Moorthi, 1988: Baroclinic instability in vertically discrete sys-\nitems. J. Atmos. Sci., 45, 1688-1707.\nSela, J. G., 1980: Spectral modeling at the National Meteorological Center. Mon. Wea. Rev.,\n108, 1279-1292.\nRoads, J. O., 1986: Forecasts of time averages with a numerical weather prediction model.\nJ. Atmos. Sci., 43, 871-892.\n, J. O., 1987: Predictability in the extended range. J. Atmos. Sci., 44 3495-3527.\n278","0\n0\nINTERVAL 25M\nINTERVAL 25M\nL\n0\nH\n35.5\n4\n1-1600\n58.8\nFig. 1. The 500 mb (a) geopotential heights and (b) the departure from zonal men for the 1600-day average from the\nempirically-forced NMC GCM. The corresponding observations compiled from January 1979-86 ECMWF analyses are\n00\n20\nH\n:\nECMWF 1979-86 JAN 500 MB Z*\n29 9\n90W\n90W\n32\nSGM CL 500 MB Z* MODEL CLIM DAY\n&\nO\nTO\nX\n0.0\n0.0\n0.0\n0\nH\n180\n180\nH\nO\n83\n34.2\nH\nD\n0\n72\n90E\n33.8\n90E\nfor\n0\n46\nA\n43\n30.8\nP\n(d)\n(b)\n0\n0\n30N\nEQ\n30S\n60S\nSP\nNP\n60N\nNP\n60N\n30N\nEQ\n30S\n60S\nSP\n0\n0\nINTERVAL 50M\nINTERVAL 50M\n587\nH\n-1600\n5230\n8600\n$440\n5800\nX200\n5000\n5400\n5920\n5800\n5800\n00\nH\n5116\nECMWF 1979-86 JAN 500 MB Z\n587\n90W\n90W\nSGM CL 500 MB Z MODEL CLIM DAY\n5867\nH\n5926\nH\n5200\n400\n5800\n5000\n5400\n5200\n5800\n5600\n5400\n180\n180\n5800\n5877\n5116\n5235\nH\ngiven in (c) and (d).\n*\nr\n90E\n90E\n400\n580001\nH\nOTHER\n5400\n5400\n5200\n5600\n5800\n5000\n5400\n2600\n500\n5800\n..\n5125\n(a)\n(c)\n0\n0\n60S\nSP\nNP\n60N\n30N\nEQ\n30S\nNP\n60N\n30N\nEQ\n30S\n60S\nSP","1.0\n(a)\n0.8\n0.6\n0.4\n0.2\n0.0\n0\n10\n20\n30\nDAY\n1.0\n(b)\n0.8\n0.6\n0.4\n0.2\n0.0\n0\n10\n20\n30\nDAY\nFig. 2. The 30-day forecasts of ensemble mean (a) normalized RMS error and (b) pattern\ncorrelation for ECMWF observed persistence (thin solid curves), model persistence (thick\nsolid curves) and initial error experiment (thick dashed curves).\n280","R15 TRUNC PREDICBILITY EXPERIMENT\nMODEL PERSISTENCE\n30\n30\n00\nG\n00\n808\n800\n10\n10\n600\n400\nE\n400\n200\n200\n800\n1\n1\n90\n60\n30\n0\n-30\n-60\n-90\n90\n60\n30\n0\n-30\n-60\n-90\nLAT\nLAT\n30\n30\n0.0\n0\n10\n10\n*\nZ\n1\n1\n90\n60\n30\n0\n-30\n-60\n-90\n90\n60\n30\n0\n-30\n-60\n-90\nLAT\nLAT\ndistribution of the normalized RMS error (top panel) and the are rescaled on the Z\nFig. 3 Latitudinal plotted against forecast days. Those for model persistence been removed,\nscore (bottom panel), experiment on the right. The zonal mean Z score has positive values.\nleft therefore and initial latitudes error with negative values are less predictable than other with\n281","A Comparison of General Circulation Model and Observed\nRegional Climates: Daily and Seasonal Variability\nDavid A. Portman and Wei-Chyung Wang*\nAtmospheric and Environmental Research, Inc.\n840 Memorial Drive\nCambridge, MA 02139\nThomas R. Karl\nNational Climatic Data Center\nNESDIS/NOAA\nAshville, NC 28801\nI. Introduction\nIncreasing concentrations of atmospheric greenhouse gases (CO2, CH4, N20\nand CFCs) are expected to cause changes in global and regional climates over\nthe next few decades. General circulation models (GCMs) are perhaps the best\ntools for assessing potential effects of climate change on regional scales.\nUnderstanding how GCMs simulate current climate is an important first step in\nusing GCMs to predict future climate changes. Many of the studies that use\nobservations to validate GCM output concentrate exclusively on annual or seasonal\naverages of global, hemispheric and zonal mean quantities. Although such long-\nterm and large-scale averages are useful for general global climate studies,\nthey reveal very little about how the model reproduces specific regional\nclimates.\nHere we develop a technique for relating the large-scale climate simulated\nin a GCM with the local climate observed at individual stations. We apply this\ntechnique by investigating the natural variability of atmospheric temperatures\n(1) between neighboring observational stations and (2) between GCM gridpoints\nand observational stations. We use daily as well as seasonally averaged data\nin order to take advantage of the best time resolution offered by the model.\nIn addition, we focus on the free-atmosphere away from the surface in order to\navoid complications introduced by surface-related differences between gridpoints\nand stations (e.g. elevations and boundary-layer conditions).\nII. Analysis\nDaily temperature data were obtained for 3 gridpoints in the NCAR Community\nClimate Model (CCM1) and for 6 U.S. upper air rawinsonde stations. Stations and\ngridpoints are listed in Table 1 and their relative positions are indicated in\nFig. 1. The CCM1 data are for 10 years of a control run that includes an annual\ncycle and interactive surface hydrology (Seasonal-Hyd: Simulation, Case 241).\nThe observed temperatures are for the 30 years 1957-1986. Surface climates\nat\nneighboring stations and gridpoints may or may not be similar, depending on\nlocal geography and topography. For each observational station, we used a Monte\nCarlo technique to generate one hundred 10-year long temperature time series\n*Currently at the Atmospheric Sciences Research Center, State University of\nNew York at Albany, Albany, New York.\n282","from 30 years of daily data. Each time series consititutes an independent sample\nof the current climate state. Ten-year mean and variance statistics were then\ncomputed interannually for 4 seasons--winter, spring, summer, autumn--and for\n365 days of the year. In order to compare gridpoint and station means (Hg) us)\nand variances (Og, os) directly, we adopted two statistical parameters proposed\nby Chervin (1980, 1981)\nthe\n(1)\n1/2\nand\nOg\n(2)\nOs\nParameter r1 is a difference in the mean values of two temperature time series\ndivided by a composite of their respective variances. If the mean temperatures\nare equal, as they would be in the case of two samples respresenting identical\nclimate states, r1 = 0. Larger and smaller r1 values would suggest that the mean\ntemperatures for the two samples may be different. However, r1 must exceed an\na priori two-tailed 95% level of statistical significance defined for a t-test\nin order for us to consider that such differences are important. Parameter r2\nis a ratio of the variances for two temperature time series. If the variances\nare equal, r2 = 1. Larger and smaller r2 values would suggest that the\ntemperature variances for the two samples may be different; r2 must exceed an a\npriori two tailed 90% level of statistical significance defined for an F-test\nfor us to consider that such differences are important. Of course, even if\nparameters r1 and r2 fall beyond their respective limits for statistical\nsignificance, there remains the possibility of determining that two sampled\nclimates are different when they are not and vice versa. Note that both r1 and\nr2 depend on the temperature variance (Og, os) We have imposed narrower limits\nfor r2 than for r1 in order to reduce the possibility of a conclusion of\ndifferent mean temperatures under an improper assumption of equal temperature\nvariances (c.f. Chervin, 1981).\nSeasonal distributions of parameters r1 and r2 are shown for all three\ngridboxes in Figs. 2-3. Boxplots are used to illustrate the variability of the\ndifference in the means (r1) and the ratio of the variances (r2) of seasonally\naveraged temperatures for the two observational stations in each gridbox.\nConsidering the median value of each boxplot (Fig. 2), we find that station-to-\nstation differences in temperature variance are generally not significant and\nthat interseasonal fluctuations are small. For some gridboxes, the largest\nvariability occurs near the surface probably due to differences in surface\nelevation and geography (see Table 1) Station-to-station differences in mean\ntemperature (Fig. 3) are somewhat more variable. Absolute values\nand\ninterseasonal fluctuations of r1 appear to be greatest near the surface.\nIn order to examine differences between one GCM gridpoint and 2 nearby\nobservational stations, seasonal distributions of r1 and r2 are shown in Figs.\n4-5 for the Southeast gridbox. Here we find that variability of the ratio of\n283","seasonally averaged temperature variance is small (Fig. 4). Furthermore,\ndifferences in mean temperature appear large near the surface and decrease with\nheight (Fig. 5). The r1 and r2 values shown in these figures suggest that the\nmodel's seasonal temperatures are generally cooler than the observed seasonal\ntemperatures and that gridpoint-to-station differences exhibit an annual cycle.\nWe can use interannual averages of daily temperature data to further examine this\nannual cycle. For example, gridpoint-to-station variability of parameters r1 and\nr2 is shown on a daily basis in Figs. 6-7 for the 500 mb level of the same\ngridbox. During winter, we find that there is little difference between modeled\nand observed temperature variances. However, during summer, model variance is\nsignificantly greater than observed variance (Fig. 6). Gridpoint-to-station\ndifferences in mean temperature, which approach zero during autumn, become\nsignificant during late spring and early summer (Fig. 7). By contrast, station-\nto-station differences in temperature mean and variance are smaller and remain\nrelatively constant throughout the year.\nIII. Summary\nWe have developed a technique for relating large-scale climate simulated\nin a GCM with local climate observed at individual stations and have applied it\nto atmospheric temperatures for one version of the NCAR CCM1. This technique\nmay also be applied to other atmospheric variables (i.e. geopotential height,\nrelative humidity) and for other GCMs. In general, intercomparison studies that\ncompare GCM output with observations should focus on upper air regions as well\nas the surface. Both gridpoint-to-station and station-to-station variability\nshould be considered. Finally, variability of important short-term\nperiodicities, such as the annual cycle, should not be overlooked.\nAcknowledgment\nThis work was supported by a grant from the Carbon Dioxide Research\nProgram, Office of Health and Environmental Research, U. S. Department of Energy\n(DE-FG02-86-ER60422).\nReferences\nChervin, R.M., 1980: On the simulation of climate and climate change with general\ncirculation models. J. Atmos. Sci. , 37 (9) 1903-1913.\n1981: On the comparison of observed and GCM simulated climate ensembles.\n,\nJ. Atmos. Sci , 38(5) 885-901.\n284","Table 1. NCAR CCM1 gridboxes and observational stations. Locations and\nelevations of the central gridpoint and included stations are indicated\nfor each gridbox. Winter (DJF), spring (MAM), summer (JJA), and autumn\n(SON) observed surface climatological averages for 1951-1980 are taken\nfrom Monthly Climatic Data for the World.\nTEMPERATURE (C)\nNORTH\nWEST\nHEIGHT\nNAME\nLATITUDE\nLONGITUDE\n(m)\nDJF\nMAM\nJJA\nSON\nNorthwest\nGridpoint\n46°40'\n120°00'\n941\nn/a\nn/a\nn/a\nn/a\n(A) Spokane WA\n47°40'\n117°25'\n721\n-1.7\n7.7\n19.1\n8.5\n(B) Porland/Salem OR\n44°57'\n123°01'\n61\n5.5\n11.5\n19.3\n13.2\nCentral\nGridpoint\n37°48'\n97°30'\n556\nn/a\nn/a\nn/a\nn/a\n(A) Topeka KS\n39°20'\n95°41'\n270\n-0.7\n12.1\n24.7\n13.4\n(B) Dodge City KS\n37°45'\n100°02'\n790\n0.7\n12.0\n25.3\n13.6\nSoutheast\nGridpoint\n33°20'\n82°30'\n116\nn/a\nn/a\nn/a\nn/a\n(A) Charleston SC\n32°48'\n79°58'\n13\n9.6\n18.0\n26.3\n18.9\n(B) Atlanta GA\n33°45'\n84°23'\n312\n6.0\n16.1\n24.9\n16.7\nB\nA\nB\nB\n125.0\n120.0\n115.0\n110.0\n105.0\n100.0\n95.0\n90.0\n85.0\n60.0\n75.0\n70.0\nLONGITUDE\nFig 1. U.S. map showing locations of CCM1 gridboxes and observational\nstations. Upper air rawinsonde stations (labeled A and B) lie east and\nwest of the central gridpoint in each gridbox. Latitudes, longitudes\nand\nsurface elevations of gridpoints and stations are summarized in Table 1.\n285","Fig. 3. . Test variate 11 standardized difference,\nin mean temperature between 2 observational\nstations in each gridbox. Boxplots are as in\nFig. 2. Dashed lines denote a priori statistical\nTopeka KS & Dodge City KS\nSpokane WA & Portland OR\nCharleston SC & Atlanta GA\n300 mb\n500 mb\nNorthwest\nSoutheast\nCentral\n850 mb\nsignificance at 95%\nSurface\nW\n4\n3\n2\n1\n0\n-2\n-1\n-3\n-4\n-5\n3\n2\n0\n-2\n-4\n1\n-1\n-3\n-5\n4\n3\n2\n0\n-2\n-3\n-4\n1\n-1\n2. Test variate r2 - ratio of temperature\nvariance between 2 observational stations in each\nBoxplots indicate 90th, 70th, 50th\n(median) , 30th and 10th percentile values at 4\nsummer (s) , and autumn (A) . Dashed lines denote\natmospheric levels for winter (W) , spring (s),\nTopeka KS & Dodge City KS\nSpokane WA & Portland OR\nCharleston SC & Atlanta GA\n300 mb\nS\nS\na priori statistical significance at 90%.\nW\nW\n500 mb\nNorthwest\nS\nSoutheast\nS\nCentral\nW\nW\nW\nA\n850 mb\nW\nW\nA\nA\nSurface\nS\nS\nS\nS\nS\nW\nW\nW\ngridbox.\n3.0\n2.5\n2.0\n1.5\n1.0\n0.5\n3.0\n2.5\n2.0\n1.5\n1.0\n0.5\n3.0\n2.5\n2.0\n1.5\n1.0\n0.5\n0.0\nFig.","Fig. 5. Test variate r1 - standardized difference\nin mean temperature between 2 observational\nstations in the Southeast gridbox (top) and\nbetween the Southeast gridpoint and each station\nBoxplots and\ndashed lines are as in Fig. 3. GCM gridpoint\nCharleston SC & Atlanta GA\n300 mb\nGridpoint & Charleston SC\ndata are for 3 atmospheric levels only.\nGridpoint & Atlanta GA\nW\nW\nseparately (center and bottom) .\n500 mb\nSoutheast\nSoutheast\nSoutheast\nW\nW\n850 mb\nW\nW\nW.\nSurface\n5\n4\n3\n2\n-2\n1\n0\n-1\n-3\n-4\n0\n-2\n-1\n-3\n-4\n-5\n-6\n-7\n-8\n-9\n0\n-2\n-4\n-6\n1\n-1\n-3\n-5\n-7\n4. Test variate r2 ratio of temperature\nvariance between 2 observational stations in the\nSoutheast gridbox (top) and between the Southeast\ngridpoint and each station separately (center and\nbottom) Boxplots and dashed lines are as in\nFig. 2. GCM gridpoint data are for 3 atmospheric\nCharleston SC & Atlanta GA\n300 mb\nGridpoint & Charleston SC\nS\nS\nGridpoint & Atlanta GA\nS\nW\nW\nA\nA\n500 mb\nS\nSoutheast\nSoutheast\nSoutheast\nW\nW\nA\nA\n850 mb\nS\nW\nW\nW\nA\nSurface\nS\nlevels only.\nW\n3.0\n2.5\n2.0\n1.5\n1.0\n0.5\n3.0\n2.5\n2.0\n1.5\n1.0\n0.5\n3.0\n2.5\n2.0\n1.5\n1.0\n0.5\n0.0\nFig.","Ratio of Temperature Variance\n500 mb\n3.0\n2.5\nGridpoint & Charleston SC\nGridpoint & Atlanta CA\n2.0\n1.5\n1.0\n0.5\n0.0\n0\n30\n60\n90\n120\n150\n180\n210\n240\n270\n300\n330\n360\nDay of Year\nFig. 6.\nTest variate -ratio of temperature\nvariance at 500 mb between the Southeast gridpoint\nand each of the 2 observational stations in the\nSoutheast gridbox. Each curve represents a series\nof 365 median (out of 100 computed for each day of\nthe year) r2 values which was filtered to remove\nperiodicities of less than 90 days. Dashed lines\ndenote a priori statistical significance at 90%.\nStandardized Difference in Mean Temperature\n500 mb\n1.5\n1.0\n0.5\nGridpoint & Charleston SC\nGridpoint & Attarita GA\n0.0\n-0.5\n-1.0\n-1.5\n0\n30\n60\n90\n120\n150\n180\n210\n240\n270\n300\n330\n360\nDay of Year\nFig. 7. Test variate r1 - standardized difference in\nmean temperature at 500 mb between the Southeast\ngridpoint and each of the 2 observational stations\nin the Southeast gridbox. Each curve represents a\nseries of 365 median (out of 100 computed for each\nday of the year) r1 values which was filtered to\nremove periodicities of less than 90 days. Dashed\nlines denote a priori statistical significance at\n95%.\n288","The Response of a GCM Atmosphere to Extratropical SST Anomalies.\nYochanan Kushnir\nAOS Program Princeton University, P.O. Box 308, Princeton, NJ 085421\nNgar-Cheung Lau\nGFDL, Princeton University, P.O. Box 308, Princeton, NJ 08542\nThe sea surface temperature (SST) field over the extratropical ocean and the\ntropospheric circulation, exhibit a strong mutual relationship in the form of large-scale\ncoherent patterns (e.g., Namias 1972, Wallace and Jiang 1986). Over the Pacific during\nwinter, for example, the prominent SST anomalies take the form of a large pool of warm\n(cold) water over the central ocean basin and cold (warm) water near the North American\nseaboard. The dominant 500 mb height anomalies associated with that SST pattern, exhibit\na north-south dipole-like structure over the central ocean basin, with high (low)\ngeopotential heights just north of the warm (cold) water and a low (high) anomaly to the\nsouth. This relationship is not consistent with results of linear, stationary wave theory\nwhen the effects of the SST anomaly are prescribed (e.g., Hoskins and Karoly 1981).\nPrevious attempts to address this problem with the aid of general circulation models\n(GCMs), did not reproduce all aspects of the observations. Recently however, Lau and\nNath (1989) showed that when an atmospheric GCM is forced with a time varying SST\ndistribution, based on the observations between 1950 and 1979, a relationship similar to\nthe observation emerges. The goal of this study is to better understand the results of\nprevious studies and to examine whether their differences can be reconciled.\nThis report presents preliminary results from a study in which a GCM with fixed\nJanuary insolation was used. The model was first integrated for 1350 days with the\nJanuary climatology of SSTs prescribed at its lower boundary, to form the so-called\n\"control\" integration. Next, the anomaly pattern, shown in Fig. 1 in its \"positive\" polarity,\nwas added to the climatological distribution of SSTs2. Nine integrations with\n\"anomalous\" SSTs were performed, each starting at a different initial atmospheric state,\ntaken from the control run. Eight of these integrations were limited to 180 days (the\n\"short\" runs), and one extended to 1350 days (the \"long\" run). The integrations were then\nrepeated with the polarity of the anomaly pattern reversed. Ensembles of 90-day averages\nwere calculated from the output of the control run and each of the experiments.\nTo examine the model's response to the SST anomaly we calculated the differences in\ngeopotential height between each of the experiment ensembles and the control ensemble.\nBoth 1000 and 500 mb height differences were calculated as well as the corresponding\nlocal values of a univariate t-variable. The t-statistics were calculated from the local height\ndifferences and the local rms value of the 90-day averaged geopotential height field.\n1\nPresent address: Lamont-Doherty Geological Observatory of Columbia University, Route 9W,\nPalisades, NY 10964.\n2\nThe anomaly was derived from a rotated principal component analysis of the seasonal SST field over\nthe North Pacific Ocean performed by Lau and Nath (1989).\n289","Hereafter we will regard as statistically significant the height anomalies which have a\ncorresponding t-values equal to or larger than 2 (assuming 9 independent \"realizations\" in\neach category, this is approximately the 95% confidence level).\nFigs. 2, 3 show the response to the SST anomaly in the long integrations. These\npatterns can be thought of as representing the equilibrium response of the atmosphere to the\nprescribed SST anomaly. Both the positive and the negative anomaly runs exhibit a\nstatistically significant response in the vicinity of the SST anomaly. In the positive\nanomaly experiment (Fig. 2) a negative, \"equivalent barotropic\" height anomaly is located\ndownstream from the region of anomalous warm water, centered around 130°W, between\n30 and 60°N. Over the subtropics, where the SST anomaly is negative, the response is\n\"baroclinic\", with a weak but significant high at the surface and a low aloft. In the long\nnegative anomaly experiment we again find an equivalent barotropic low around 130°W and\nbetween 30 and 60°N, this time downstream from the cold SSTs. North of the low\npressure, a statistically significant high pressure cell is found. Thus the model's\nmidlatitude response does not bear a simple linear relationship with respect to the polarity\nof the prescribed SST anomaly, a fact noted previously by Pitcher et al. (1988). In the\nSubtropics the response reverses its polarity with respect to the positive anomaly run, with\na low at 1000 mb over the warm part of the SST anomaly and a high at 500 mb.\nThe response during the first period of the short integrations (days 1-90) is shown in\nFig. 4, 5 for the positive and negative anomaly runs respectively. In the positive anomaly\nrun (Fig. 4) there is virtually no response over the midlatitudes and a weak baroclinic\nresponse, similar to the equilibrium response, over the subtropics. The negative anomaly\nrun (Fig. 5) however, exhibits a strong, equivalent barotropic low over the entire ocean\nbasin north of 30°N and a weak baroclinic response over the subtropics. Note that in the\nnegative anomaly run, the short-term response north of 40°N is different from the\nequilibrium one by the absence of the high latitude ridge. The center of the negative height\nanomaly is also shifted northwestward with respect to its counterpart in the long run.\nViewed with respect to the region of the SST anomaly, this response pattern is much more\nconsistent with the observations (and the results of Lau and Nath) than the equilibrium\nresponse. In the second 90-day period of the short runs the response tends to approach the\nequilibrium response as shown in Fig. 6 for the positive anomaly run and Fig. 7 for the\nnegative anomaly run. In particular, note the appearance of the equivalent barotropic low\nover the midlatitudes, downstream from the warm SSTs in the positive anomaly runs and\nthe change in scale, both zonal and meridional, of the mid- and high-latitude response in the\nnegative anomaly runs.\nIn summary, we find that the model's response to the prescribed SST anomalies is\nquite complex. It varies from one experiment to the other, is relatively weak (compared to\nobservations), and does not bear a linear relationship with respect to the sign of the SST\nanomaly. However, the t-values in the vicinity of the anomaly indicate that the response\npatterns are statistically significant. The model's response to the forcing seems to be\ngoverned by a relatively long, transient dynamical process with a time scale on the order of\nan entire \"season\" (~90 days). This raises the interesting possibility that in order to\n290","interpret the observed ocean-atmosphere relationship, we need to consider a transient rather\nthan an equilibrium response. If so, these results may explain why previous GCM\nexperiments which emphasize the transient behavior (e.g., Lau and Nath 1989) agree better\nwith observations than GCM experiments that emphasize the equilibrium response (e.g.,\nPitcher et al. 1988).\nREFERENCES:\nHoskins, B. J. and D. J. Karoly, 1981: The steady linear response of a spherical\natmosphere to thermal and orographic forcing. J. Atmos. Sci., 38, 1179-1196.\nLau, N.-C., and M. J. Nath, 1989: A general circulation model study of the atmospheric\nresponse to extratropical SST anomalies observed in 1950-1979. J. Climate, in press.\nNamias, J., 1972: Space scale of sea surface temperature patters and their cause. Fishery\nBull., 70, 611-617.\nPitcher, E. J., M. L. Blackmon, G. T. Bates, and S. Munoz, 1988: The effect of North\nPacific sea surface temperature anomalies on the January climate of a general circulation\nmodel. J. Atmos. Sci., 45, 173-188.\nWallace, J. M., and Q. Jiang, 1987: On the observed structure of interannual variability of\nthe atmosphere/ocean climate system. In Atmospheric and Oceanic Variability, Roy.\nMeteorol. Soc., H. Cattle ed., 17-43.\n0\nFigure 1: The anomaly added to the climatological distribution of SSTs in our study shown\nhere in its \"positive\" polarity. Contour interval is 2°C, negative values are\nstippled. Note that here and in following figures only the Northern Hemisphere\nbetween 120°E and the Greenwich meridian is shown.\n291","1000\n500\n20\n0\n10\n0\nD\n120W\n120W\nFigure 2: The \"equilibrium\" geopotential height response to the SST anomaly of Fig. 1\n(the \"positive\" anomaly). Height anomalies are shown in the upper two panels\nfor the 1000 and 500 mb levels. Contour interval is 10 m, negative contours are\ndashed. The distribution of the corresponding t-variable is shown in the lower\ntwo panels where the contour interval is 1.\n1000\n500\n10\n20\n10\n0\n2\n0\n120W\n120W\nFigure 3: As in Fig. 2 but for an SST anomaly with the reversed polarity (the \"negative\"\nanomaly).\n292","1000\n500\n20\n10\n0\n⑉\n120W\n120W\nFigure 4: Day 1 to day 90 geopotential height response to the SST anomaly of Fig. 1\n(the \"positive\" anomaly). Height anomalies are shown in the upper two panels\nfor the 1000 and 500 mb levels. Contour interval is 10 m, negative contours are\ndashed. The distribution of the corresponding t-variable is shown in the lower\ntwo panels where the contour interval is 1.\n1000\n500\nTO\n0\n10\nTO\n0\n0\n120W\n120W\nFigure 5: As in Fig. 4 but for an SST anomaly with the reversed polarity (the \"negative\"\nanomaly).\n293","500\n1000\n20\n0\n10\n10\n0\n10\n0\n0\n0\n0\n0\n0\n0\no\n120W\n120W\nFigure 6: Day 91 to day 180 geopotential height response to the SST anomaly of Fig. 1\n(the \"positive\" anomaly). Height anomalies are shown in the upper two panels\nfor the 1000 and 500 mb levels. Contour interval is 10 m, negative contours are\ndashed. The distribution of the corresponding t-variable is shown in the lower\ntwo panels where the contour interval is 1.\n1000\n500\n10\nNO\n-10\n0\n2\n120W\n120W\nFigure 7: As in Fig. 6 but for an SST anomaly with the reversed polarity (the \"negative\"\nanomaly).\n294","MODELLING NORTH PACIFIC SST RESPONSE TO ATMOSPHERIC ANOMALIES\nUte Luksch and Hans von Storch\nMax-Planck-Institut fuer Meteorologie\nBundesstrasse 55, D2000 Hamburg 13, FRG\nLarge-scale sea surface temperature anomalies (SSTA) in the North Pacific\nocean often persistent for several months during wintertime (Namias et al.,\n1988). There is observational evidence that these patterns are forced by\nanomalous atmospheric circulation (Fig. 1, schematic diagram): The SST can be\nmodified by\n*\nanomalous latent and sensible heat fluxes which are initiated by the\nadvection of cold or warm air by anomalous winds.\n*\nchanged turbulent mixing and upwelling due to wind field anomalies\nThree experiments with a multi-level primitive equation model of the North\nPacific have been conducted to study the influence of such anomalous\natmospheric circulation on the SST.\nIn the standard experiment anomalous wind stress and anomalous latent and\nsensible heat fluxes are specified as anomalous forcing. The anomalous\nwind field data are derived from the 1950-1979 COADS subset, heat fluxes\nare parameterized with the anomalous surface wind, an air temperature\nderived from a simple advective prognostic model, and SST.\n*\nIn two sensitivity experiments the influence of anomalous wind stress and\nanomalous heat fluxes on the SST are examined separately.\nIn the standard experiment the GCM is capable to reproduce the main aspects of\nthe observed evolution of SST, in particular in winter (Fig. 2). The high\nvariability of the SST in the West Wind Drift Area is good reproduced by the\nmodel (Fig. 3). The low variability in the eastern subtropical Pacific,\nhowever, is overestimated in the model. This problem might be related to the\nuse of constant bulk transfer coefficients in the parameterization of the\nsensible and latent heat fluxes.\nIn sensitivity experiment 1 (only wind stress anomalies) the temporal\nvariability of SST is systematically too low; nevertheless, the spatial\n295","patterns of variability are reproduced and the time evolution of SST compares\nwell with the observed.\nThe SST response to anomalous heat fluxes only (sensitivity experiment 2) is\nvery similar to the results of the standard experiment; overall in the North\nPacific the correlation of SST in the two experiments is large (with more than\n99.9% significance) and the temporal variances are comparable.\nThe SST responses to anomalous heat fluxes and wind stress separately\n(sensitivity experiment 1 and 2) are positive correlated in midlatitudinal\nregions (Fig. 4), especially west of the dateline and east of 150°W. That\nmeans in these areas the response to wind stress and heat flux anomalies are\nadditive. In the central North Pacific and in the subtropics the correlations\nare weakly positive or negative.\nSince the atmospheric circulation is in part related to the tropical El\nNiño/Southern Oscillation (ENSO) phenomenon it is hypothesized that part of\nthe North Pacific SSTAs may be interpreted as remote oceanic response to\nanomalous equatorial Pacific SSTs. Observed and simulated SSTAs are correlated\nwith ENSO (Fig. 5). To filter out the SO related North Pacific SSTA\ncharacteristics, we plan to perform experiments in which the anomalous forcing\nis restricted to the remote response to the SO.\nReferences:\nLuksch, U., Storch, H.v. and Maier-Reimer, E., 1989: Modeling North Pacific\nSST anomalies as a response to anomalous atmospheric circulation. Submitted to\nthe J. Marine Systems. (also available as MPI Report 37)\nNamias, J., Yuan, X. and Cayan, D.R. , 1988. Persistence of North Pacific sea\nsurface temperature and atmospheric flow patterns. J. Climate, 1: 682-703.\nvan Loon,H. and Rogers, J.C., 1981. The Southern Oscillation. Part II:\nAssociations with changes in the middle Troposphere in the northern winter.\nMon. Wea.Rew., 109: 1163-1168.\n296","Figure 1:\nSchematic of the anomalous wintertime flow during a warm SO event (after a 700\nmb height composite given by van Loon and Rogers, 1981). The map indicates\ncooling of the central North Pacific and warming in the eastern North Pacific.\nThe anomalous wind advects cold air from the northwest to the central North\nPacific and cools the central North Pacific; at the same time the warm air\nblowing from the tropical Pacific warms the eastern North Pacific. The\ncyclonic shear in the central part of the ocean causes upwelling of cold water\nin that area. Along the american coast the anomalous northward flow is\nassociated with coastal downwelling.\n60°N\nCOLD\nDRY\n40°N\nWARM\nWET\n180°\n150°W\nFigure 2:\nCorrelation coefficients between observed and simulated SSTs (standard\nexperiment) in January. Areas with positive correlation coefficients are\nstippled and coefficients greater than 0.25 (0.50) are hatched (cross\nhatched).\nSince 30 independent samples (Januaries) are available the null-hypothesis of\nno local relationship between the observed and the simulated SST is rejected\nwith 95% (99%) confidence if the estimated correlation coefficient is greater\nthan 0.31 (0.42).\n-0\n60°N\n0\n0.25\n0.50\n0.50\n0.75\n40°N\n0.25\n0.50\n0.25\n0.50\n7\n180°\n150°W\n297","Figure 3:\nMonth-to-month variability of the observed and simulated SSTAs\nin\nJanuary (standard experiment).\nTop: Observations from COADS 1950-1979. Areas with standard deviations greater\nthan 0.5K are stippled. Contour interval 0.1K.\nBottom: Ratio of the standard deviations of SST in the standard experiment and\nin the observations. Areas with ratios of 1+25% are stippled. Contour\ninterval 25%.\n60°N\n0.4\n0.5\n0.6\n0.7\n0.7\n40°N\n0.4\n0.6\n0.5\n0.5\n0.4\n180°\n150°W\n60°N\n1.0\n1.5\n1.0\n1.0\n40°N\n2.0\n2.0\n1.5\n> 2.5\n180\no\n150°W\n298","Figure 4:\nCorrelation coefficients of SST simulated in sensitivity experiment 1 (only\nwind stress anomalies) and in experiment 2 (only heat flux anomalies).\n60°N\n0.25\n0.50\n0.50\n40°N\n0\n0.25\n180°\n150°W\n299","Figure 5:\nCorrelation coefficients between the observed SOI (air pressure difference\nDarwin and Tahiti) and North Pacific SST in January. Areas with negative\ncorrelation coefficients are stippled and areas with coefficients AI -0.25 and\n+0.25 are hatched.\nSince 30 independent samples (Januaries) are available the null-hypothesis of\nno local relationship is rejected with 90% (95%) confidence if the estimated\ncorrelation coefficient is greater than 0.24 (0.31).\nTop: observed SST\nBottom: simulated SSTAs (standard experiment)\n60°N\n0.25\n0\n40°N\n0.25\n-\n180°\n150°W\n60°N\n0\n0\n0.25\n0.25\n40°N\n0\n10.25.\n180°\n150°W\n300","Variability of Skill of Long Range Forecasts\nand Implications for Their Use and Value\nRobert E. Livezey\nClimate Analysis Center, NMC/NWS/NOAA, Washington DC 20233\n1. INTRODUCTION\nThe status and utility of long-range predictions (e.g., of monthly and\nseasonal averages of temperature and precipitation) has often been the subject of\ndebate in the meteorological literature. Few would argue that such forecasts\naccurately made would not be enormously useful. However, the consensus view has\nbeen that the skill, albeit real, is in fact quite small. This opinion has been\nshared by both producers and users of long-range forecasts alike as well as the\nbroader meteorological community. The fact that prospects do exist for improve-\nments in the skill of long-range forecasts provide considerable justification for\nthe continued support of research and development on prediction of climate fluctua-\ntions. Additionally, recent research documents real forecast skill over that of\nrandom forecasts for a variety of schemes, both objective and subjective, leaving\nlittle doubt that such forecasts are practical. Thus, one might conclude that\nskillful long-range forecasts can be made but only with modest accuracy, but that\nsmall gains in this accuracy are possible with potentially large increases in\neconomic value for certain applications.\nIn much of the remainder of this essay I intend to argue that the characteri-\nzation of current long-range forecasting skill as slight is an inappropriate\ngeneralization, and that as a consequence it should be possible right now to match\nforecasts to applications with substantial economic payoffs. The crux of the\nargument is that many observed variations in skill by predictand, season, loca-\ntion, and situation reflect real variability in predictability that will be\nreasonably insensitive to the forecast sample. Thus, many of the large excursions\nof skill above rather modest overall levels represent reliable and considerable\nlocal (in space, time, etc.) skill which can be matched beneficially to certain\nusers.\nUntil quite recently, skill variations had to be conservatively viewed as\nsampling fluctuations; there was too small a fund of data, insufficient forecast\nexperience, and complete absence of a theory for predictability. Much of this has\nchanged: (1) Data sets have slowly but steadily grown while more powerful evalua-\ntion techniques have become possible with bigger computers, (2) experience with\nexisting and new forecast methods has expanded, and (3) rudimentary predictability\ntheories have been developed. While there is a growing awareness of the possi-\nbilities inherent in exploitation of skill variations at least among producers of\nofficial forecasts in the United States this is generally not the case among users\nand most atmospheric scientists. One purpose here is to alert a wider audience to\nan untapped resource for decision makers and the associated research required to\nrealize this benefit.\n2. EXAMPLES OF RELIABLE AND POTENTIALLY EXPLOITABLE SKILL VARIABILITY\nAt the National Weather Service's Climate Analysis Center (CAC) forecasts of\none of three mean temperature and precipitation classes (above, near, and below\n301","normal) and of the probabilities of each of the classes at 100 U. S. stations are\nmade every month for the following 90-day period. The latter forecasts have been\navailable only since the early 1980's but the categorical forecasts for the four\nnon-overlapping 90-day periods corresponding to winter (December-January-February),\nspring (March-April-May), etc. have been routinely made since 1959. Thus, it is\npossible to characterize the overall skill at least for categorical forecasts in\nthe four seasons. This turns out to be 6% for temperature and precipitation com-\nbined for all stations and seasons using the skill measure employed herein. This\nscore varies between -50 and 100%, the former representing no correct categorical\nforecasts and the latter all correct. Since a skill of zero is the expected\nskill of categorical forecasts assigned randomly from a climatological distribu-\ntion, the skill is often referred to as the \"percent improvement over climatology.\"\nBecause the 6% score summarizes a very large number of forecasts it is highly\nsignificant statistically even though it is small. Consequently, official fore-\ncasts confirm the broader conclusions stated in the introduction. However, while\nthis score unambiguously establishes that long-range forecasts have real skill, it\nis likely to have no particular relevance to any potential user of the forecasts.\nSuppose, for example, the user is the gas industry and the concern is demand and\nstorage of natural gas for wintertime heating. As Knox et al. (1985) point out,\npeak demands for gas space heating are sensitively dependent on temperature and\nits extremes. Low temperatures imply heavy use of gas, with sizable amounts drawn\nfrom pipelines or storage with attendant drop in pressure and perhaps interference\nwith delivery. On the other hand, an unusually warm winter may lead to a glut and\nhigh storage costs due to carryover. The gas manager then is specifically con-\ncerned about temperature and its extremes in the wintertime in the particular\nregion his company supplies. Thus, in the context of the official three class\nU.S. 90-day forecasts, the most (some would argue only) relevant skill for his\nproblem is that for forecasts of only above and below normal temperatures (in the\nthree-class system) in the wintertime in a particular region. Let us examine the\nimplications on skill of stratification of forecasts first by type, then by season,\nnext by location, and finally by class. For each forecast stratification the\nobserved variability in skill scores will be examined in terms of its reproduci-\nbility in an independent sample.\nAt each stage the sample of forecasts will be considerably reduced, so, except\nfor the first step (skill stratified for temperature and precipitation forecasts\nseparately), the discussion will be based on a 38 year sample of independent fore-\ncasts produced objectively by the operational analog prediction system that is one\nof the forecast tools used by seasonal forecasters at CAC. Before we begin, it\nshould be pointed out that it is unlikely that the gas manager could effectively\nutilize forecasts with a skill level as low as 6%.\nFirst, when official forecasts of temperature and precipitation for all sea-\nsons are scored separately, their skills are 8.1 and 4.2% respectively. This is\nconsistent with the relationship between temperature and precipitation forecast\nskills for other time scales (e.g., 3 to 5 days). It is well known that the pre-\ncipitation prediction problem is much more complex than the temperature problem\nand involves a broader spectrum of time and space scales, so the fact that the\nformer's skill is less than the latter's is expected.\nBecause the analog system's overall temperature skill of 8.3 is compatible\nwith that for the official forecasts, the remainder of this discussion will focus\non its skill score. Table 1 consists of a list of skill scores for all stations/\ncategories for each of four seasons. Here the score of interest (for winter) to\n302","the hypothetical gas manager happens to be the largest in the table and is twice\nas large as the original unstratified skill. Its relationship with other table\nentries is consistent with what might be expected from the characteristic time and\nspace scales of the atmospheric circulation and corresponding signal-to-noise\nratio estimates in the different seasons. It is likewise consistent with results\nfrom verification of official seasonal forecasts and other objective systems.\nConsequently, the observation that skill is highest in the winter can also be\nconsidered a reliable result.\nTo more closely pinpoint the information most needed by users in this\nextended example, the geographical distribution of skill of the analog predictions\nin the wintertime is presented in Fig. 1. The skills are for predictions of three\nequally-distributed temperature classes. The spatial variability of skill is\nlarge; scores are high over the Southwest, north and south central states, and the\nEast, and particularly over the Southeast, but are low over large parts of the\nNorthwest, in the lee of the Rocky Mountains, and over northern New England. If\nthe broad features of this skill variability are reliable, for example the huge\ncontrast in forecast skill between the southeastern United States and the high\nplains states, the implications for potential gas industry users of the predic-\ntions in the two areas are dramatically different. Obviously, the manager in\nNebraska should rely on climatological information rather than the forecasts, but\na manager in North Carolina might profit considerably from intelligent exploita-\ntion of the forecasts.\nThere are at least three different ways to explore whether the differences in\nskill from place to place in Fig. 1 are more than just sampling fluctuations.\nFirst, most of the larger scale features shown on the map are essentially dupli-\ncated in maps for wintertime temperature skill in three classes by other forecast\nmethods. These approaches include official forecasts and other experimental\nobjective systems substantially different from that used here.\nSecond, the pattern of skill in Fig. 1 is predicted with considerable vera-\ncity in calculations by Barnett (1981) and Madden (1981) using potential predicta-\nbility theory. The only important difference between their maps (Barnett's, 1981,\nis shown in Fig. 2) and Fig. 1 is the absence of forecast skill for New England in\nthe latter. An explanation for this apparent inconsistency has been offered by\nGilman (1986), and is related to the sensitivity of the winter climate in New\nEngland to snow cover and the inability to forecast the presence or absence of\nthis cover. Nothwithstanding this, there is substantial reason to regard the\ngross features of the skill pattern in Fig. 1 as fundamental characteristics of\nthe forecast system and to utilize this information.\nA third way to check this idea is to test the sensitivity of features in Fig.\nto removal of different sets of fixed numbers of forecast years from the sample.\n1\nThis approach is the least desirable because of limitations in total sample size,\nand was not used here because it wasn't needed. In many other instances, however,\nit may be the only test available.\nThe final stratification of forecast skill made here focuses as closely as\npossible on the climate information of most value to the gas industry in the\nwintertime, namely the prediction of local temperature extremes. The skill of\nprediction of the two-thirds of all forecasts that are for either above or below\nnormal (excluding near normal) is shown by area in Fig. 3. Note that overall\nskill for the two extreme classes is considerably greater than that for all three\nclasses and that this is true virtually everywhere in the United States. By\n303","inference, then, skill in predicting near normal conditions must be close to zero,\nboth overall and locally. This is indeed the case here and in fact seems to be an\nalmost pervasive characteristic of monthly and seasonal forecasts, whether subjec-\ntive or objective. It is important to note here that again there are considerable\ngrounds to expect the differences in skill implied by this additional stratifica-\ntion to be reliable and therefore exploitable. Even with very conservative (i.e.\nlarge) estimates of sampling error, forecasts of above and below normal mean\ntemperature classes in much of the wintertime eastern United States should be\nreliably several times as skillful as the overall forecast skill of about 6% cited\nat the start of this section, and perhaps an order of magnitude greater in some\nlocations. This, along with the low skill in the high plains, should be important\ninformation for a number of potential users.\nOther useful skill stratifications than those discussed SO far are also\npossible. One commonly employed is to separate cases for a particular season by\nthe antecedent climate or atmospheric circulation regime, the latter gaining\nincreasing attention for extended range (6-10 days) to monthly prediction. A good\nexample of the former is the separation of forecasts for years in which an index\nof the E1 Niño/ Southern Oscillation (ENSO) indicates anomalously warm waters in\nthe equatorial Pacific (i.e. a warm event; sometimes accompanied by the so-called\nE1 Niño), anomalously cold waters (a cold event), or near normal waters.\nRopelewski and Halpert (1986, 1987, 1989), among others, have documented a number\nof instances in which the anomalous surface temperature, precipitation, or both\nfor a particular location/season can be reliably predicted given the knowledge\nthat a cold or warm event is taking place.\n3. SUMMARY AND RECOMMENDATIONS\nA case has been made that forecasts of mean conditions for up to a season can\nbe made with appreciable reliability for certain parameters, places, seasons, and\nsituations, thereby making them potentially of definite benefit to certain users.\nTwo conditions are necessary before users will take full advantage of skill-\nful long-range prediction, namely (1) the creation of a perception in the user\ncommunity that forecasts are skillful enough to be of some value, and (2) the\nmatching of reliable forecast situations with users who can exploit them, and the\ndevelopment of analytic frameworks to facilitate this exploitation.\nNeither of these problems has received the research attention accorded other\nresearch thrusts. The first requirement implies a new direction for producers of\nlong-range predictions, specifically the analysis of skill variability, including\nits description and reproducibility in independent forecast samples. The second\nrequirement, however, is a task for meteorological statisticians and economists,\nparticularly through quality/value studies like that of Katz et al. (1987).\n304","Table 1. Skill of Seasonal Temperature Forecasts for the Operational Analog\nPrediction System\nSkill (percent)\nSeason\n12.6\nWinter\n8.6\nSpring\n9.3\nSummer\n2.8\nFall\n8.3\nAll\nAll Forecasts , Winter\n10\nyo\n10\no\n20\n20\no 10\n10\no\n25\n10\n20\n10\n20\n10\n20\no\n10\n20\n29\n20\n20\n10\n40\n40\n57\n46\n10\n40\n20\n20\no\n10\n30\n10\n29\n21\n10\n20\n20\n20\n25\n120°\n20\n90°\nS = 0.126\na\n+\n10\n$10\nFig. 1. Skill (S) of U.S. winter mean temperature forecasts of three\nequally-distributed temperature categories (above, near, and below normal)\nmade by the operational analog prediction system (1950-51 to 1987-88)\n305","100\n30\n20\n13\n40\n20\n50\n50\nB\n40\n35\n<20\n30\n30\n40\n25\nWINTER\n0.\n115\n110\n105\n100\n95\nas\nao\nUSCOHM WB-DC\n75\nFig. 2. Distribution of winter potential predictability as percent of\nvariance (from Fig. 13, Barnett, 1981).\no\nAbove and Below Forecasts , Winter\nis\n0\n20\n20\n40\n20\n0\nis\n40\n52\n20\n40\n20\n48\n39\n60\n20\nx\n80\n82\n20\n20\n:\n60\n31\n40\n20\n300\n40\n20\n20\n(2\n120°\n20\n40\n90°\nSa=0.213\nb\n+\n20\n20\nFig. 3. Same as Fig. 1 except for above and below normal forecasts only.\n306","Prediction of ENSO episodes using canonical correlation analysis\nAnthony G. Barnston and Chester F. Ropelewski\nClimate Analysis Center, NMC/NWS/NOAA\nWashington, DC 20233\n1. Introduction. In this study the skill potential of canonical correlation\nanalysis (CCA) in predicting fluctuations of the cycle of the E1 Niño/ Southern\nOscillation (ENSO) is examined. The prediction of the phase of ENSO is worth\npursuing because of the strong impact on the Pacific Coast of South America as well\nas on numerous related remote climate fluctuations such as the Indian monsoon or\nthe North American winter temperature and precipitation patterns (Rasmusson and\nCarpenter 1982, Ropelewski and Halpert 1986). This study follows the lead of\nBarnett and Preisendorfer (1987) and Graham and Michaelsen (1987a, 1987b) in\napplying CCA to the prediction of short-term climate fluctuations, both for exami-\nnation of the features of the optimized linear relationships and for determination\nof the levels of the accompanying skill. Real-time ENSO prediction is commencing\nboth using empirical approaches such as CCA and using numerical models (Cane and\nZebiak 1986, Inoue and O'Brien 1984). CCA has previously been described in the\nnon-meteorological literature (Hotelling 1936, Anderson 1984) as well as in empiri-\ncal meteorological or geophysical contexts (Glahn 1963, Davis 1977, Barnett 1981,\nNicholls 1987).\nCCA is a multivariate statistical technique that calculates linear combina-\ntions of a set of predictors that maximizes relationships, in a least-squares\nsense, to similarly calculated linear combinations of a set of predictands. Be-\ncause the predictor and predictand are each a set of variables, the method can\ndepict relationships between corresponding favored spatial patterns of either\n(e.g., if the sets of variables are gridded fields). Furthermore, if either set is\na temporal sequence of such fields (referred to as \"stacked\" fields), CCA can high-\nlight the evolution of the predictor or predictand spatial patterns. For example,\nthe first 100 out of 200 predictor variables might be a 10 by 10 two-dimensional\ngridded field of 1-month mean geopotential height centered 1.5 months before fore-\ncast time, while variables 101-200 are the same field centered a half month before\nforecast time. The analysis could find predictor patterns at both prior times\n(implying a temporal evolution) that are associated with a specified predictand\npattern which itself might be \"stacked\" in time.\n2. Data. In this study the predictor data are gridded near-global 3-month mean\nsea-level pressure (SLP) starting 1, 2, 3 and 4 seasons prior to forecast time, and\nthe predictands are 3-month mean area average sea surface temperature (SST) in\nseveral rectangular areas in the tropical Pacific. The SLP data are obtained from\nthe National Meteorological Center (NMC) and the SST data from the COADS (Slutz et\nal. 1985) through 1987 and blended ship/ satellite data thereafter (Reynolds 1988).\nThe period of study is 1956-1989, constrained by the onset of relatively good trop-\nical Pacific data in the middle 1950s.\nEach seasonal field of predictor SLP consists of 828 grid point values for 10\ndegrees longitude spacing and 5 degree latitude spacing from 40S to 70N. There\nare\n3312 predictor variables within the four temporally \"stacked\" SLP fields.\nSeven predictand area average SSTs were used (Table 1). Note that there is\noverlap between the BP87 and CAC areas. A recommended pre-CCA EOF analysis will\neliminate these redundancies as it will do for the spatially as well as temporally\ncorrelated \"stacked\" predictor data (Sec. 3).\n3 07","Before beginning the CCA analysis, the predictor and predictand data are\ndetrended and then standardized, so that all variables possess temporal station-\narity and are weighted equally to produce a correlation-based CCA. The detrending\nis done to filter out artifactual intrusions such as temporally differential data-\nset biases, but might remove some authentic climatic nonstationarities also.\nTable 1. The seven SST areas used as predictors. BP87 refers to Barnett and\nPreisendorfer (1987) ; CAC is Climate Analysis Center.\nLocation\nBP87 Name\nLocation\nCAC Name\n10N-10S, 140-180E\nP2\n0-10S, 80-90W\nNiñol+2\n5N-5S, 120-170W\nP3\n5N-5S, 90-150W\nNiño3\n0-10N, 80-100W\nP4\n5N-5S, 150W-160E\nNiño4\n0-10S, 80-100W\nP5\n3.\nOverview and Procedure of CCA. As a linear modeling technique, CCA is maxi-\nmally advanced in that it fits two or more predictors with two or more predictands.\nThis is more complex than multiple linear regression, which handles only one pre-\ndictand at a time. CCA uses eigenvalues and eigenvectors in a specialized way,\nsuch that the structure of the covariance between predictor and predictand vari-\nables is defined under the constraint of maximization of cross-dataset covariance\nexplained with each successive mode. It is more complex than standard empirical\northogonal function (EOF) analysis which calculates an orthogonal coordinate system\nthat maximizes variance explained in a single dataset with each successive mode.\nThe CCA procedure is now very briefly described; for more detail the above-cited\nreferences, including the appendix in Barnett and Preisendorfer (1987), may be\nconsulted.\nFollowing any preprocessing of the predictor and predictand datasets (such as\nfiltering, detrending or standardization), it is advisable to pre-orthogonalize\nthese datasets individually using EOF analysis, and then inputting the resulting\ntime series to the CCA. This eliminates redundancies in the data, which safeguards\nagainst degeneracies in estimated inverse matrices in the CCA. It also reduces\nnoise; this helps reduce the degree of overfitting in the CCA.\nNext, a covariance matrix between each predictor (predictand) and all predict-\nands (predictors) is formed; note that only cross-dataset pairings are used. This\nis generally not a square matrix, but becomes square when it is post- or premulti-\nplied by its transpose, each of which operation is carried out for the predictor\nand pre-dictand CCA solutions, respectively. The predictor and predictand product\nmatrices are then used to obtain CCA eigenvectors and eigenvalues for predictors\nand predictands, with equal eigenvalues for corresponding mode numbers. The square\nroots of the eigenvalues are the correlations between the predictor and predictand\ncanonical component time series associated with the canonical loading patterns for\nthe mode in question. The predictor and predictand canonical loading patterns are\nlinear combinations of the respective input variables that maximize the cross-data-\nset covariance accounted for by the mode.\nTo perform CCA for varying predictand lead times, the process is repeated for\neach case. In this study the number of seasons between the final predictor season\nand the predictand season is varied from 0 to 4.\n4. Cross-validation tests to determine skill. To estimate the forecast skill of\nCCA for each of the seven predictand SSTs for each of the five lead times for each\nof the twelve three-month seasons, a cross-validation technique was used as follows:\nEach of the 34 years from 1956 to 1989 was held out in turn and CCA was used to\ndevelop a prediction model from the remaining 33 years (for the 4-season lead con-\n308","dition, one less year was available). The predictor data for the withheld year was\nthen projected onto the predictor CCA loading patterns and predictand values were\ngenerated and verified against observed data. The predicted anomaly values were\nuniformly inflated such that their standard deviations equaled those of the\nobserved data. Over the 34 iterations of this process, three statistics were com-\nputed: (1) the temporal correlation between forecast and observed anomalies, (2)\nroot mean square error (RMSE), and (3) mean absolute error (MAE).\nAs a competitor for the first statistic, the autocorrelation of the SST over\nthe lead time in question (i.e., persistence) was computed. Note that the forecast\nversus observed correlation is unaffected by the inflation of the forecast anomaly.\nThis first statistic will be used predominantly in the evaluation of skill to follow.\nStrictly speaking, the autocorrelation version of the persistence skill has an\nunfair slight advantage relative to the cross-validation CCA skill because the\nformer is computed directly rather than in a cross-validation mode. Using the\nlatter approach would give differentially damped persistence forecasts among target\nyears; this will be done in a forthcoming journal paper on this subject.\n5. Results. In general, area P3 in BP87 (5N-5S, 120-170W) is the best predicted\nof the seven SST areas, followed by Niño3 and Niño4. The eastern areas (P4 and P5\nin BP87, and Niñol+2) are predicted less well still, and P2 in BP87 is predicted\nwith virtually no skill. There is a tendency for the predicted fluctuations in the\neastern Pacific to be phased slightly earlier than those in the central Pacific.\nHowever, with the exception of area P2, there is a high correlation among the pred-\nictions of all the areas, even at the longest lead periods.\nThe temporal correlation between the forecast and the observed SST for area P3\nfor 0, 1 and 2 season predictand lead periods is shown in Fig. 1 for all 12 3-month\nseasons (defined by middle month of predictand period). When the CCA forecast\nskill exceeds that of persistence skill (as depicted -- and slightly overestimated\n-- by the autocorrelation from final predictor to predictand period), an X indicates\nthe persistence skill and a vertical line connects the corresponding skill levels.\nTwo features of the skills are apparent. First, there is a strong seasonal\nskill variation. While a winter-to-summer skill difference is quite obvious at\nfirst glance, closer study of Fig. 1 reveals that the SST forecast is poor when the\ninterval between the final predictor period and the predictand period includes\nspring. Apparently spring introduces a quasirandom discontinuity in the evolution\nof the anomalies; the SST climate system is in effect \"reset\" each year. Any lead\ntime greater than two seasons has unfavorable expectations, because at least one\nyear separates final predictor and predictand period midpoints. In fact, the three\nseason lead skills (not shown) never attain the 0.4 skill level. The skill-maxi-\nmizing season in which to make a forecast for the first three lead times (0, 1 and\n2) is after the completion of Jun.-Jul.-Aug., which captures the peak in 0 season\nlead skill in Sep.-Oct. Nov. in one season lead skill in Dec.-Jan.- Feb., and the\nsecond highest value of two-season lead skill in Mar.-Apr.-May. If the two-season\nlead forecast is most desired, the best forecast would need to be made one month\nearlier.\nA second obvious feature is the standing of CCA skill relative to persistence.\nPersistency skill (which is slightly overestimated here) is frequently superior to\nCCA skill, and much of the time when CCA skill is higher both are at low (<0.6)\nlevels. However, there are several instances in which CCA nontrivially outperforms\npersistence at high skill levels: for one season lead forecasts of Nov.-Dec.-Jan.\nand Dec.-Jan.-Feb. and for two season lead forecasts of Mar -Apr. and (with\nmarginal CCA skill) Mar.-Apr.-May. These four cases occur with the two optimal\ntimes to make forecasts (see the preceding paragraph) and feature 10 to 15 percent\nincreases in variance explained over persistence. The one season lead forecast for\nDec.-Jan.-Feb. is particularly useful in view of established associations between\n309","ENSO and extratropical North American winter atmospheric anomalies (Ropelewski and\nHalpert 1986).\nPlots comparable to Fig. 1 for the other SST areas (except for poorly pre-\ndicted area P2) are generally similar but have shorter lasting like-phased seasons\nof somewhat lower peak skills. Persistence likewise loses some accuracy.\n6. Interpretation of canonical loading patterns. One would hope that the pre-\ndictor canonical loading patterns, which hold potential to depict systematic near-\nglobal SLP pattern evolution, might suggest established or hypothetical scenerios\nrelated to subsequent tropical Pacific SST behavior. For example, a Southern\nOscillation signature would be expected in modes containing strong predictand load-\nings for the central Pacific SST areas. We now explore this for the two-season\nlead model forecasting winter.\nThe leading canonical mode, explaining 94 percent of total variance, is loaded\nupon heavily by the central Pacific SST areas and somewhat heavily by the eastern\nPacific areas. Examination of the set of four predictor loading patterns reveals\nfairly weak and physically questionable patterns in the earliest two maps, and the\nevolution of a strong Southern Oscillation dipole in the final two. Other \"centers\nof action\" are found in the extratropics in all four maps, particularly over moun-\ntainous continents. The effect of removal of such areas, which might test the\nreality of their association with the SSTs, will be discussed below.\nHigher order modes do not exhibit clear Southern Oscillation signatures.\nTheir contribution to skill will be discussed below.\nThe one season lead model forecasting winter is generally similar to the two\nseason lead model. Models for longer leads have weaker loading patterns and the\nSouthern Oscillation dipole loses prominence.\n7. Skill sensitivity tests. Several tests were performed to determine whether\nforecast skills could be increased further, particularly for central Pacific SST\nareas for two and three season lead times. The tests might also provide informa-\ntion about the CCA model characteristics.\nFirst, SST area B2 was removed because it was poorly predicted. Resulting\nskills for the remaining areas were virtually identical, confirming that irrelevant\ninformation does not distort the appearance of the meaningful associations.\nBecause of the presence in the predictor loading patterns of continental\nextra-tropical centers of questionable relevance to tropical Pacific SST, the CCA\nwas rerun (1) without the mountainous portions of the predictor fields and (2)\nwithout any land areas. In both cases, results were slightly different but there\nwas no overall skill improvement or degradation. The original full field was thus\nretained in the absence of understanding of the physical meanings of many of the\nextratropical centers.\nElimination of the earliest two of the four predictor fields lowered skill\nslightly, and made more difference in long than short lead forecasts. Apparently\ndevelopments in the early half of the four-season-long predictor period contribute\nto the forecast skill despite the lack of obvious physical understanding.\nWhen higher order canonical modes were omitted from the analysis the skills\nwere slightly lowered, indicating again that the less obvious features in the second\nand third modes should not be discarded. Because the cross-validation procedure is\nused on time series already prefiltered with an EOF analysis, increases in skill\nfrom these poorly understood sources cannot be assumed to be due principally to\nsampling variability or noise fitting.\nA planned further test is lag averaging, in which forecasts for a given target\nseason made from two or more adjacent months will be averaged.\n310","8. Outlook for 1990. The CCA forecasts made on completion of Jul.-Aug.-Sep. 1989\nindicate a continuation of slightly to moderately below-normal tropical Pacific\nSSTs through the 1989-90 winter, followed by a switch to above-normal in the\neastern areas in early spring of 1990 and the central areas in middle to late spring.\nThe above normal SST is predicted to continue through the four season lead time of\nOct.-Nov.-Dec. of 1990. The skill of the forecasts from mid-spring forward is too\nlow to trust. This fact is consistent with some rather large alternations (cold-\nwarm-cold-warm) for consecutive overlapping seasons within spring 1990 for most of\nthe SST areas.\nReferences\nAnderson, C. W., 1984: An Introduction to Multivariate Statistical Analysis, 2nd\ned., Wiley and Sons, 675 pp.\nBarnett, T. P., 1981: Statistical prediction of North American air temperature\nfrom Pacific predictors. Mon. Wea. Rev., 109, 1021-1041.\nBarnett, T. P. and Preisendorfer, R., 1987: Origins and levels of monthly and\nseasonal forecast skill for United States surface air temperatures determined\nby canonical correlation analysis. Mon. Wea. Rev. 115, 1825-1850.\nCane, M. A., S. E. Zebiak and S. C. Dolan, 1986: Experimental forecasts of E1\nNiño. Nature, 321, 827-832.\nDavis, R., 1977: Techniques for statistical analysis and prediction of geophysical\nfluid systems. Geophys. Astrophys. Fluid Dyn., 8, 245-277.\nGlahn, H., 1963: Canonical correlation and its relationship to discriminant\nanalysis and multiple regression. J. Atmos. Sci., 25, 23-31.\nGraham, N. E. and J. Michaelsen, 1987a: An investigation of the E1 Niño-Southern\nOscillation cycle with statistical models. 1. Predictor field characteris-\ntics. J. Geophys. Res., 92, 14251-14270.\nGraham, N. E. and J. Michaelsen, 1987b: An investigation of the E1 Niño Southern\nOscillation cycle with statistical models. 2. Model results. J. Geophys.\nRes., 92, 14271-14289.\nHotelling, H., 1936: Relations between two sets of variates. Biometrika, 28, 321-\n377.\nInoue, M. and J. J. O'Brien, 1984: A forecasting model for the onset of a major E1\nNiño. Mon. Wea. Rev., 112, 2326-2337.\nNicholls, N., 1987: The use of canonical correlation to study teleconnections.\nMon. Wea. Rev., 115, 393-399.\nRasmusson, E. M. and T. H. Carpenter, 1982: Variations in tropical sea surface\ntemperature and surface wind fields associated with the Southern Oscillation/\nE1 Niño. Mon. Wea. Rev., 111, 517-528.\nReynolds, R. W., 1988: A real-time global sea surface temperature analysis. J.\nClimate, 1, 75-86.\nRopelewski, C. F. and M. S. Halpert, 1986: North American precipitation and tem-\nperature patterns associated with the E1 Niño/Southern Oscillation (ENSO) .\nMon. Wea. Rev., 114, 2352-2362.\nSlutz, R., S. J. Lubler, J. D. Hiscox, S. D. Woodruff, R. J. Jenne, D. H. Joseph,\nP. M. Steurer, and J. D. Elius, 1985: Comprehensive Ocean Atmosphere Data\nSet, National Oceanic and Atmospheric Administration, Boulder, CO. (Available\nfrom Climate Research Program, ERL, R/E/AR6, 325 Broadway, Boulder, CO 80303.)\n311","Figure 1. Temporal correlation over 34 years between CCA-forecast and the observed\nSST in area P3 (5N-5S, 120-170W) for 0, 1 and 2 seasons lead. The X's show skill\nof persistence (which has a slight unfair advantage; see text) when it is outper-\nNDJ\n12\nOND\n11\nSON\n10\nASO\n9\nJAS\n8\nJJA\nPREDICTED SEASON\n7\nMJJ\n6\nAMJ\n5\n0-season lead\nMAM\n2-season lead\n4\nFMA\n3\nformed by CCA.\n1 -season lead\nJFM\n2\n0\n2\nDJF\n1\n1.0\n.9\n.8\n.7\n.6\n.4\n.5\n.3\n.2\n.1\n0","Numerical Model Precipitation Forecasts\nJ. O. Roads\nN. Maisel*\nScripps Experimental Climate Forecast Center\n*Climate Analysis Center\nObservations and Forecasts\nThere are a number of observational data sets that could be used to verify the NMC MRF\nmodel forecasts. One data set consists of the U.S. cooperative station network consisting of\na few thousand stations that report weeks or months later. These data are archived by the\nNational Climatic Data Center (NCDC). From these cooperative stations a subset of very\nlong term stations have been selected by NCDC to develop a U.S. historical climatology\nnetwork. It is this data set that we use here to verify the NMC model. Only stations that\nare not believed to be influenced to any substantial degree by strictly local changes are\nincluded in this network, a requirement that eliminates most of the first order stations. An\nadditional requirement made here was that the stations be active as of 1989 and have serial\nrecords that cover a substantial number of years. For example, starting in 1950, 832 stations\nhad relatively serially complete records for 37 years. Continuing on to March 30 of 1989,\nonly one station dropped from the list.\nThe NMC MRF model used to make these forecasts is a state-of-the-art general circula-\ntion model first developed by Sela. The MRF model forecasts are archived on the NMC 65 X\n65 grid, which is an octagonal square grid of resolution 300 km overlying a polar stereographic\nprojection (true at 60° N) of the northern hemisphere. Individual stations collocated within\nthe NMC gridpoints are averaged and 'gridded' to provide suitable large scale observations\nthat could be directly compared to the NMC forecast.\nConsider first the time series of areally averaged precipitation over the U.S. Figure 1a\nshows the seasonal variations beginning in the spring (March 1) of 1950 and ending in the\nwinter of 1988-89 (March 1, 1989). The dashed line shows the normal seasonal cycle which\nis a maximum during the summer and a minimum during the winter with the spring season\nbeing slightly more rainy than the fall. The solid lines show the actual amounts which\ncorrespond to the annual amounts shown in Figure 1b. Quite a lot has happened since\n1950. During the middle 1950's the country was in wide spread drought and during the\n1980's the country was experiencing higher than normal precipitation. During the winter\nof 1987-88 the drought returned for four more seasons until the fall of 1988 when the rains\nreturned once more to normal. Finally, the last winter evaluated was also below normal.\nThe annual averages shown in Figure 3b indicate that there has been an almost insignificant\ntrend since 1950 toward higher precipitation. The major purpose in evaluating a trend for\nthese evaluations is to determine whether it might positively influence perceived forecast\nskill. For the forecast models to be discussed subsequently it had a negligible influence.\nSkill Score\nSince precipitation distributions are more difficult to deal with than, say, the 500 mb\nheight which has a more Gaussian-like distribution, we must accordingly evaluate precip-\nitation forecasts somewhat differently than we would forecasts of the 500 mb height. For\n313","simplicity, we are going to arbitrarily assume a forecast is good if it has the appropriate sign.\nThat is, if the observations are anomalously wet then a good forecast is one that predicts\nanomalously wet conditions. Obviously, for the predominantly dry regions, a forecast of dry\nweather would be successful most of the time. To remove the influence of this dry bias on\nthe skill evaluation, we use the standard Heidtke skill score, S.\nMRF Precipitation Forecasts\nForecasts of 2-hour accumulations by the MRF model were archived for the period\nMarch 1, 1987 - March 1, 1989. This time period is broken up into two discrete years. 87\nrefers to the period March 1, 1987 - February 29, 1988, and 88 refers to the period March 1,\n1988 - February 28, 1989. Each of these periods contains a winter and summer season, which\nis evaluated separately. Also, since a bias correction must be made, the separate evaluation\neach year provides an independent method to evaluate the numerical predictions.\nThe archived precipitation was available at -day intervals for forecast days 1 to 3. In\norder to compare this data to observed 24-hour accumulations the 12 hour forecasts were\nsummed. After forecast day 3, only the last 12 hour accumulation of each day was archived.\nA simple interpolation of these last 12 hour accumulations was then used to obtain the\none-day accumulation.\nThe major problem with this interpolation is that, as may be seen in Figure 2, there\nappears to be an unresolved diurnal precipitation cycle. The second 12 hour accumulation\nof each day is larger than the first 12 hour accumulation. Now the first 12 hour accumu-\nlation refers to the accumulation from 0000 UTC to 1200 UTC. At 90° west longitude this\ncorresponds to 6:00pm local time of the previous day to 6:00am local time of the same day.\nSimilarly, the second 12 hour accumulation refers to 6:00am local time to 6:00pm local time\naccumulation of the same day.\nThe dominance of the second 12 hour accumulation suggests that over the country, on\naverage, there are greater amounts of forecast precipitation during the day than during the\nnight. This diurnal cycle has substantial influence on the perceived forecast precipitation\namounts. For example, during 1988 the second 12 hour accumulation on the third day was\nabout .11mm/hour, whereas the first 12 hour accumulation was about .085mm/hour. The\ninterpolation of the last 12 hour forecast after day 3 thus yields precipitation amounts that\nmight be unfairly perceived as being larger than they actually are. Presumably 24 hour\naccumulations or, even better, archival of the missing 12 hour intervals would be beneficial\nto showing a lower bias.\nStill, in comparison to 1987, the model bias has improved considerably. The wet bias has\nbeen reduced by about .02 mm/hour even though the observed values had only a slight re-\nduction of .005 mm/hour. During 1987 the bias was about 50% higher than the observations\nand during 1988 the bias was about 30% higher than the observations. Again, part of this\nperceived bias could be due to the previously mentioned archival problem of only having the\nsecond 2 1 day forecast (6:00am to 6:00pm). However, it is also clear that there is a tendency\nfor the model to become wetter than the observations at increasing forecast time.\nConsider now the forecast skill of time averages in comparison to observed averages,\nshown in Figure 3. For comparison, the daily forecasts are also shown. Although only MRF\nforecasts of maximum 10 days are available, MRF model forecasts beyond this point are\nmade by simply persisting the 10-day forecast. As shown previously this may be better than\neven attempting to extend the numerical forecast, which may provide more noise than useful\n314","information at this range.\nOnly the dependent case is shown here. That is, precipitation breakpoints (precipitation\namounts delineating wet and dry conditions) and distributions (distributions determining\nchance) are calculated from the cases being evaluated. The intersection of F with the\nenvelope curve of persistence forecasts (i.e. at about 20 days for forecasts considered over\nthe entire year) provides a difference between the numerical forecast and the long term mean\npersistence forecasts of one standard deviation. If we consider the difference between the\npersistence forecasts and the numerical forecasts made for the specific year, then the standard\ndeviation of the difference is 20 and a 20 difference only has an 8% chance of occurring\nby chance. In fact, only for the daily forecasts does persistence ever become better than the\nnumerical forecasts. All of the time averaged numerical forecasts considered here (at least\nthose out to 30 days which are all that are shown here) are better (probably significantly)\nthan the time averaged persistence forecasts.\nAs might be expected, the skill is greatest when using the dependently determined break-\npoints and distributions. However, the evaluation in which the observed distribution was\nalso used to evaluate the skill shows that the numerical forecasts could be applied directly,\nalthough presumably some massaging might be useful to present a more skillful official fore-\ncast. The skills are lowest here, especially for the daily forecasts, since the forecast model\nhas a wetter distribution than is observed. Still, the skills are not a lot lower.\nThere appears to have been a marginal improvement in the skill of the numerical precip-\nitation forecasts for at least the first few days (after the first day) during the second forecast\nyear. During 1988, the forecasts were even better during the second day, especially during\nthe wintertime. In the medium to monthly range the skill increased somewhat during the\nsecond summer, although persistence also increased here.\nSkill levels are always greatest during the wintertime at short range and during the sum-\nmertime at the extended range. The highest skills tend to be reached during the wintertime,\nespecially for the second or last winter evaluated (December 1988, January, February 1989)\non the second day. This is especially noticeable for the forecast of a two-day average. The\nforecast skill here is about .5 (depending upon details of the skill evaluation), which indicates\na correct forecast 75% of the time.\nAcknowledgements\nThis research was supported by USGS grant (14-08-0001-G1743), CAC and NSF grant\n(ATM 85-20540), NOAA grant (NA86-AA-D-CP104), and Calspace. The figures were pre-\npared by M. Schultz and the manuscript was typed by P. Whan. Numerous people provided\ndata for this study. NMC observations were provided by J. Janowiak of NMC/CAC and the\nfinal observational data set was provided by S. Doty and T. Karl of NCDC as part of the\nExperimental Climate Forecast Data Center program. We especially thank NMC (especially\nJ. Ward) for providing the forecast data, despite their initial reservations about the utility\nof such data. We also thank E. Bainto, D. Nghiem, and B. Ronberg for their help in the\ndata analysis. Useful comments were provided by D. Cayan.\n315","SEASONAL VARIATIONS\n1.3\n1.2\n1.1\n1.0\np .9\n8\n.7\n.6\n1955 1960 1965 1970 1975 1980 1985 1990\n.5\nANNUAL VARIATIONS\n1.3\n1.2\n1.1\n1.0\nTREND\nAVERAGE\n.9\np\n.8\n.7\n.6\n1960 1965 1970 1975 1980 1985 1990\n.5\n1955\nSeasonal and annual variations in the line U.S. indicates averaged the precipitation, average seasonal p (in\nFigure 1. In the seasonal variations the and dashed end in the winter of 1988-89 (DJF); give the the\nmean. .1 mm/hour). The curves start in the spring of and 1950 the minimum to the winter. Solid give lines the annual\nmaximum corresponds amounts. to the In summer the annual variations estimated plot, standard the dashed deviation lines of the 37 year\nactual observed with the envelope corresponding to the trend line with an envelope corresponding considered to\naverage mean. estimated The dotted + one lines standard show deviation a least squares variation trend. in the slope. Over the 39 years\nhere an there is an almost significant upward\n316","U.S. AVERAGE\n1.4\n87 NMC\n1.3\n1.2\n1 day\n88 NMC\n1.1\n1.0\n50-86 NCDC\n.9\n87\n.8\n88\n1 day\n.7\n.5\n1\n5\n10\nt\nFigure 2. U.S. average forecast amounts by the NMC MRF model for 1987 and 1988 in\ncomparison to the 50-86 average and + one annual standard deviation. The NMC forecast\naverage is shown separately for for the 12 hour archives and interpolated 24 hour forecasts.\n317","D\n87 A\n88 A\n.8\nFL\nB\n3\n3\nto\nS\nS\nva\nto\n0\n0\n-3\n-3\n,\n10\n30\n1\n10\n30\nt\n87 Su\n88 Su\n.6\n.0\na\n3\nS\nS\n0\n0\n-3\n-3\n1\n10\n30\n1\n10\n30\nt\nt\n87 W\n88 W\n.6\n.6:\nF\nF\na\n3\nvo\nS\nS\n0\n0\n-3\n-3\n1\n10\n30\n1\n10\n30\n1\n1\nFigure 3. MRF model and persistence skill for daily and time averaged amounts.\nSkill, S, is shown as a function of lag time in days for daily forecasts and as a function of\naveraging time in days for the time averaged forecasts. The long term mean persistence skill\nalong with an envelope of skill appropriate to one year's persistence forecasts is shown for the\nannual values. In this figure, the precipitation breakpoints and distribution determining the\nchance of making a good forecast are determined from the dependent data. Skills for 1987\nand 1988 are shown separately. Also shown are the skills for summertime and wintertime\nforecasts.\n318","Dynamic forecast of the next month's flow.\nH. M. van den Dool\nCooperative Institute for Climate Studies\nDepartment of Meteorology\nUniversity of Maryland\nCollege Park, MD 20742\n1. Introduction\nWhat would happen to anomalies in low frequency atmospheric\nflow if the seasonal cycle evolved and everything else stayed the\nsame? Or, how do quasi-steady anomalies adjust to a change in the\nbackground flow, i.e. a change in the refractive index for Rossby\nwaves. Since the change in climatological background flow is known\nin advance there may a small amount of skill in predicting the\nchanges in anomalies. That is the issue we pursue here.\n2. A model and its forcing\nConsider the following approxomate vorticity equation\nB , DOY\n0 =\n= -Un\n-Vn\n+F\n(1)\nwhere (Un, Vn ) is a windfield prescribed as a function of x and\ny\n(\nthe monthly climatology in section 3), and is the vorticity\n(the model variable to be solved for). The dissipation consists of\nfourth order diffusion and linear drag ( k = 5.10-- and D is\nchosen such that diffusion and drag are equal at total wavenumber\n30). If F is determined from climatology, i.e.\nd&n\nDVS\nF = Un\n+\nVn\n+\nk\n+\n+\n(2)\nthen F is just the forcing needed to keep the climatological waves\nin the In field still. A global T30 model based on (1) was run\nforward in time in search for a steady state. Using F from (2) the\nsolution is obviously n. In this and other experiments (below)\nsatisfactory convergence is obtained in about 200 1-hr time steps\nfor the strength of the dissipation constants chosen. The solution\nis found starting from an arbitrary intial S-field.\n3. Forecasting the climatology\nSuppose and are the climatological vorticities of\nsuccessive months. Considering (1) and (2) the task of forecasting\nwould formally consist of changing a) the basic state wind\n(\nUni, Vn1 ) to ( Un2, Vn2 ), and b) changing the forcing F1 to\nF2. In this section we will do only a), i.e. change the background\nflow while persisting the previous months' forcing. For example we\ntry to forecast in in April from F in March and ( Un / Vn ) in\nApril. This is not a problem of great practical appeal, but it\ndoes serve as a test for the proposed methodology.\nChanging the prescribed winds in a previously balanced flow\nwill cause the vorticity advection to change which alters the\nreponse to nearby as well as remote forcing. In all 12 months we\nfound convergence to a new steady state. Defining observed and\nforecast tendencies as Ynz - Griz and s Fest - Sna respectively we\n319","calculated the tendency correlation (TC) in order to verify the\nforecasts. TC has to be larger than 0.5. We applied the model at\n250 mb because we anticipated the residual calculation of F to be\nmost meaningful at a level where the true F ( roughly the\ndivergence term) is large.\nFigures 1 and 2 show TC for the 12 climatology forecasts for\ndifferent verification areas - USA (Fig. 1), and extratropics and\ntropics (NH, SH and TR) in Fig. 2. For the USA, NH and TR domains\nwe find TC to be larger than 0.5 in 10, 11 and 12 out of 12 months\nrespectively. So there is a good deal of profitable change in\ngenerated by the change in background windfield. The exception is\nthe SH where the forecasts are little better than random. We\nverified that over USA, NH and TR the residual F is indeed\nreasonably correlated to ( f + *Divergence. This is not the\ncase\nin SH. Hence the 10 year climatology used here is dynamically\n(in) consistent in the (SH) NH, USA and TR. For this reason alone\nour attempts must fail in SH.\nThe observed fields in August and September are shown in\nFig. 3 while the observed and forecast changes in G. are given in\nFig. 4 for the USA domain. This is the most successful of the 12\nforecasts. Note the non-trivial changes caused by changing the\nprescribed windfield. The magnitude of the forecast changes is\ngenerally about half the observed, leaving the remainder to\nchanges in the forcing. From March to April the change in forcing\nis apparently dominant in the seasonal changes in the Sn-field\nover the USA.\n4. Forecast of actual monthly flow\nWe have engaged in a sofar very limited experiment on\nforecasting actual monthly flow. For each of 10 Augusts (79-88)\nthe forcing was evaluated as\nas\nDV.S2\nF2 = U2\n(3)\n+\n+\nwhere U1, V1 and are actual August means. The model is run to\nnew steady state (considered to be a September forecast) after\na\nadjusting for the seasonal cycle by\nF1\nto\n(F1 - Fni)\nand\n+\n(U2 ,V1) to (U1 - Uni, V1 - Vn1) + (Un2, Vn2)\n.\nWe have chosen W = cos (latitude) to express greater confidence in\nthe persistence of anomalies in tropical flow and forcing. The\nmean of the 10 forecasts for September is indeed very close to\nSnz, i.e. a near-perfect forecast of the climatology. The more\nchallenging question concerns the anomalies. The amount of change\nin the vorticity anomalies caused by the seasonal adjustment is,\nin variance units, only 10-15% of the observed change (for USA-\narea and August to September, 250 mb). Note also that the observed\nmonth-to-month changes in anomalies are much larger than those in\nthe climatology. Although (1) may be a good approximation for the\nclimatology, it may be less accurate for anomalies. We found\nanomalies in F to be only very weakly correlated to anomalies in\ndivergence. Hence the residual may not be an accurate\nrepresentation of the forcing. One option not tried yet is to\ninclude transient eddies on the right-hand-side of (2) This will\ncertainly improve the estimate of F.\n320","VERIFICATION USA-AREA\n100.00\n60.00\n0.5\n20.00\n0,00\n4100\n8.00\n12.00\n-20.00\nMONTH 1=JAN ETC\nFig.: . 1 The tendency correlation (in %) of forecasts of\nclimatological vorticity at 250 mb as a function\nof target month using a model based on (1) #\nThe verification are is the United States, , see Fig.3.\nCLIMATOLOGY\n80.00\nNH\nTR\nSH\n0.5\n40.00\n0.00\n12.00\n0.00\n4.00\n8.00\nMONTH 1=JAN ETC\nFig 2 As Fig.1 but now for different verification areas.\nTR is tropics, NH and SH are the midlatitudes of the\ntwo hemispheres.\n321","120W\n100W\nBOW\n2,5\n2,57\n0.0\n-19.0\n-5.0\n0,0\n-1010\nSON\nSON\n-10.0\n-2.50\n10.0\n12.\n-7.5\n15.0\n17.5\n-5.0\n2.5\n-2.5\n2010\n40M\n40N\n0.0\n2.5\n2.5\n2.5\non\n5.0-30N\n30N\n0.0\n-2.5\n-5.0\n-12.5\n-7.5\n-2.5\n-10.\n10.0-7.5 -5.0\n-10.0\n-10.0\n-1050\n-2.5\n20N\n20N\n0\n120W\n100W\nBOW\nOBSERVED TENDENCY OF RELATIVE VORTICITY\n120W\n100W\nBOW\n-7.5\n-5.0\n-2.5\n-2.5\n-5.0\n50N\n50N\n-7.5\n10.0\n12.5\n-5.0\n15L0\n-5.0-\n17.5\n-2.5-\nX\n40N\n40N\n0.\n20,0\n-2.5\n2.5\n5.0\n7.5\n7.5\n5.0\n-2.5\n30N-\nV\n2.5\n30N\n0.\n-2.5\ns\n-2.5\n-5.0-\nI\n,\n-5.0-2.5\n-2.5 -5.0 -7.5-7.5 -5.0\n0.0\n0.0\n20N-\n-20N\n120W\n100W\nBOW\nFORECAST TENDENCY OF RELATIVE VORTICITY\nFig.3 The observed climatological vorticity fields in August and\nSeptember over the USA. Units are 10 - -.\n322","120W\n100W\nBOW\n0\n7.5\n5.0\n2,5\n0,0\n2,5\n5.0\n7.5\n10.0\n12,5\n15.0\n17.5\n50N\n50N\n-17.5\n-15.0\n12.5\ntrue\n77.5\n-15\n-2.5\n0.0\n-2.5\n-20.0\n5.0\n10.0\n40N\n7.5\n40N\n12.5\n15.0\n5.\n-15.0\n2.5\n-12.5\n0.0\n-10.0\n-7.5\n30N\n30N\n-5.0\n-2.5\n-2.5\n0.0\n2.5\n>\n-2.5\n-2.5\n-5.0 -5.0 -2.5 0.0\n0200.0\n2.5\n20N\n20N\n0\n120m\n100W\nBOW\nZ1-AUGUST - CLIMATOLOGY OF RELATIVE VORTICITY\n120W\n100W\nBOW\n0\n.\n15.0\n17.5\n0.0\n2,5\n150\n7.5\n10.0\n12.5\n50N\nSON\nPN.5\n-15.0\n12.5\n10.0\n-7,5%\n0.0\n0.0\n-2.5\n-5.0\n7.5\n2.5\n3\n12.5\n40N\n40N\n5.\n15.0\n7.5\n17.5\n17.5\n7.5\n-15.0\n2.5\n-12.5\n30N\n30M\n0.0\n-10.0\n-2.5\n-7.5\n-5.0\n-5.0\n-7.5\n-10.0\n0.0\n-12.5\nD\n-2.5\n0.0\n-12.5\n-12.5\n$5.0 -12.5 $1.5-5.0\n-20N\n20N-\n0\n120W\n100W\nBOW\nZ2-SEPTEMBER CLIMATOLOGY OF RELATIVE VORTICITY\nFig. . 4 The observed and calculated tendencies in climatological\nvorticity from August to September over the USA. Units are\n10-'s-1. Calculation is based on (1) .\n323","Dynamical Prediction in the Extended Range\nEmploying low-frequency PNA Mode\nW. Y. Chen\nClimate Analysis Center\nNOAA/NWS/NMC\nWashington, D. C. 20233\nIn a recent investigation, Palmer (1988) reports that the skill of the\nmedium/extended range forecasts over the Pacific/North America region is\nstrongly correlated with the fluctuations of the Pacific/North American (PNA)\nmode of low frequency variability (Wallace and Gutzler 1981; Barnston and\nLivezey 1987). Using a simple barotropic model, Palmer finds that the growth\nof forecast error is weaker when the basic state has a strong positive PNA\nindex (negative height anomaly over the Aleutians), while the growth is\nstronger when the basic state has a large negative PNA index. It should be\ninteresting to see whether the National Meteorological Center's operational\nmedium-range forecasts (MRF) also show this PNA mode/skill relationship.\nIn association with strong E1 Nino/Southern Oscillation (ENSO) (e.g., see\nRasmusson and Wallace, 1983) condition in the tropics, the northern winters of\n1982/83 and 1986/87 had a very prominent positive PNA circulation pattern. Are\nthe MRF skills of these two winters substantially higher than those of other\nwinters? Furthermore, due to the possible influence of the prevailing PNA\nmode, are skills over the Pacific sector substantially higher than those over\nthe Atlantic sector during ENSO winters?\nAlthough there have been substantial advances in dynamical weather fore-\ncasting, the current limit of useful skill of MRF model is still only about\nsix days on average (Chen, 1989a; Saha and Van den Dool, 1988; Roads, 1988)\nTo obtain medium-range forecasts with useful skill, it becomes essential to\nidentify the geographical regions and the flow regimes that produce better\nthan average forecasts. Being able to identify a priori those more predict-\nable states will in effect raise the limit of useful forecasts.\nSix hundred cases of wintertime 1- to 10-day operational forecasts made\nby the NMC's MRF System are examined for their variabilities in performance.\nIn addition to hemispheric-wide assessment, the North Pacific/North America\n(PAC) and North Atlantic/Eurasia (ATL) sectors are also evaluated separately\nand compared. Tests of statistical significance of results are performed.\nDuring E1 Niño/Southern Oscillation (ENSO) winters, the forecast skill over\nthe PAC sector is found significantly higher than over the ATL sector. As\nexamples, Figures 1 and 2 show the difference between the two sectors. For\nwintertime as a whole, the average skill over the PAC sector is also found\nsignificantly higher during ENSO winters than during non-ENSO winters.\nWe also investigate whether the PNA-like flow regime within each ENSO\nwinter does indeed contribute most to the higher performance. For this pur-\npose, we denote the PNA anomaly pattern (taken from the rotated EOF analysis\nof Barnston and Livezey 1987) as PNAP and it is shown in Figure 3 for illus-\ntration. Then we obtain the anomaly pattern correlation, over the PAC sector\n324","only, between this PNAP and each of the winter D10F, where D10F with an over-\nbar here denotes an 11-day-mean forecast centered on day 10. The forecasts we\nare dealing now are those of the NMC DERF (dynamical extended range forecast-\ning) 30-day integrations. For detail of the DERF Experiments, the readers are\nreferred to the recent article of Tracton et al. 1989. The operation of this\nanomaly pattern correlation reduces those two maps into one value and its\nmagnitude represents the extent of PNAP-like behavior of D10F. Let us there-\nfore denote this value as FPE (forecast pattern extent). A unity implies a\nperfect match of D10F to PNAP. All 108 D10Fs were thus reduced to FPE 108-\npoint time series. The dashed curve of Fig. 4 shows the variabilities of FPE\nof D10Fs for the DERF Experiments, which happens to fall into an ENSO winter.\nComparison with observational pattern extent (OPE), the solid curve, is also\nshown in the figure. We see that most of the time the D10F shows strong PNAP-\nlike circulation, as expected of the ENSO winters. However, there are times\nthat the 11-day-mean forecast deviates substantially from PNAP. Those with\nhigh PNAP-extent are expected to yield better forecast skill than those with\nlow PNA-extent (Palmer, 1988). Indeed, this is the case, as seen in Figure 5,\nwhen using FPE as criterion to discriminate, among 108 cases of DERF experi-\nments, those with better forecast skill from those with worse forecast skill.\nThe capability to forecast the skill is remarkable.\nTherefore, within DERF ENSO winter, the contribution to their better\nperformance in the PAC sector comes mainly when the large-scale circulation is\nPNA-like, consistent with the recent results of Palmer (1988) that the skill\nof forecast over the PAC sector is strongly correlated to the fluctuation in\nthe PNA mode of low frequency variability.\nMore skillful forecasts can also be associated with greater than average\natmospheric persistence. Their relationships have been explored previously\n(e.g., Palmer and Tibaldi, 1988; Chen, 1989b). In a recent NMC dynamical\nextended range forecasting investigation, Tracton, et al., (1989) note that on\naverage the scores of the DERF predictions are much more skillful than persis-\ntence. Reporting on 18 DERF experiments, Mansfield (1986) also notes that\nwhile more persistent flow tends to yield better forecasts, persistence alone\nis not sufficient to account for the predictability in the extended range.\nUsing the 600-case dataset, the 5-day persistence forecasts have also been\ncalculated for the PAC sector. The winter mean results, for both daily and\npentad data, indicate that the mean score for the 1986/87 ENSO winter is lower\nthan both the preceding and the following winters (Chen 1990). Palmer (1987)\nalso finds lower average persistence for the 1986/87 winter than the 1985/86\nwinter. The 1987/88 persistence was not available when he published that\npaper. The results obtained so far seem to strengthen the conjecture that\npersistence alone is not sufficient to explain the interannual variability in\nskill observed over the PAC sector.\nThe relationship between low frequency flow regimes in the initial condi-\ntions of the NMC MRF model and subsequent forecast systematic error signatures\nhas recently been investigated by O'Lenic and Livezey (1989) They find\nunusually strong error signatures associated with the PNA mode among other low\nfrequency variabilities. The dependence of model skill upon initial condi-\ntions is also assessed by Horel and Roads (1987) using a two-level spherical\nquasi-geostrophic model. They find no more successful forecasts that begun\nduring qusai-stationary periods than those begun during transient flow\nregimes. However, they also find dependence of model skill on the strength of\nzonal flow across the North Pacific Ocean. In this regard, our results from\n325","the NMC MRF operational forecasts appear to be consistent with Horel and\nRoads' assessments as well as Palmer's investigations on model forecast\nskills.\n1.0\n0.9\n0.8\n0.7\n0.6\n0.5\n0.4\nFigure 1. DERF II period, 108-\ncase mean anomaly correlation\n0.3\nscores for 11-day time-mean\n0.2\nforecasts: Solid curve for PAC\nsector and dashed curve for ATL\n0.1\nsector.\n0.0\n-.I\n-.2\n30\n0\n5\n10\n15\n20\n25\nFORECAST RANGE\n1.0\n0.8\n0.6\n0.4\n0.2\nFigure 2. Performance of 5-to-\n0.0\n15 day time-mean forecasts for\nall 108 cases of DERF 30-day\n-.2\nintegrations. Upper panel is\nfor PAC sector and lower panel\n0.8\nfor ATL sector.\n0.6\n0.4\n0.2\n0.0\n-.2\n-.4\n-.6\n-.8\n0\n10\n20\n30\n40\n50\n60\n70\n80\n90\n100\n110\nCASE NUMBER\n326","110\n30\n5. Using FPE of each F10D as criterion (larger than mean + one standard deviation or those smaller with\nFigure 4. PNA pattern extent in term of\nanomaly correlation between PNAP and\nAnalysis/forecast: Solid curve for OPE\n100\n90\n25\n80\nFORECAST RANGE\nthan Figure mean-one standard deviation) to discriminate, among 108 cases of DERF II experiments,\nCASE NUMBER\n20\n70\nbetter forecast skill (left panel) from those with worse forecast skill (right panel) .\n60\nand dashed curve for FPE.\n15\n50\n40\n10\n30\n20\n5\n10\n0\n0.8\n0.6\n0.4\n0.2\n0.0\n-.2\n-.4\n-.6\nC\n1.0\n0.8\n0.6\n0.4\n0.2\n:\n-.2\n4\n-.6\nS\n0\nL\nS\nB\n30\nRotated EOF mode that represents\nPNA pattern, taken from Barnston and Livezey\n10\n25\n10\nFORECAST RANGE\n20\n10\n-10\nA\n12-20\n15\n-30\n4Q\ngo\n10\n60\nUs\n30'\n-10\n125\n300\n20\n5\nFigure 3.\n10\n10\n20\n1987.\n00\n20\n0\nBO\n30\n1.0\n0.8\n0.6\n0.4\n0.2\n0.0\n-.2\n4\n- .6","SEA-SURFACE TEMPERATURE HINDCASTING\nEXPERIMENTS USING AN ISOPYCNIC COORDINATE\nMODEL OF THE TROPICAL PACIFIC\nArthur J. Miller®, Josef M. Oberhuber#, Nicholas E. Graham® and Tim P. Barnett\n@Scripps Institution of Oceanography, La Jolla, CA 92093\n#Max Planck Institute for Meteorology, 2000 Hamburg 13, West Germany\nA tropical Pacific version of a sophisticated ocean general circulation model (Oberhuber,\n1989, in preparation), developed at the Max Planck Institute for Meteorology in Hamburg,\nhas been implemented on the Lawrence Livermore National Laboratory (LLNL) CRAY-2\ncomputers for experiments on the diagnosis, hindcasting and prediction of EL Niño events.\nSince the model was tuned to reproduce climatological ocean fields, we seek to determine if\nthe model can also reproduce anomalous observations of equatorial sea-surface temperature.\nWe first determine how well the observed monthly-mean winds, with all other atmospheric\nvariables held at the long-term climatotlies, are sufficient forcing for hindcasting El Niño\nevents in the tropical Pacific. As a second step, we plan to include anomalous atmospheric\ntemperature, relative humidity and cloudiness, as well as the anomalous winds, as forcing\nagents to determine the extent model hindcasts are improved. Through analysis of the\noceanic structure and response in these experiments, we can help to determine what processes\nare most important in generating El Niños.\nThe Oberhuber ocean model integrates the primitive equations of motion in isopycnic\n(constant density) coordinates. The surface boundary conditions for the interior flow are\ndetermined via a mixed-layer model which is fully coupled to the interior. The mixed-layer\nmodel which was used here was chosen based on the results of numerous sensitivity exper-\niments (Schroter, Gering and Oberhuber, in preparation), in which the best simulation of\nthe seasonal cycle for various mixed-layer models was optimized by tuning the model pa-\nrameters. Oceanic variables (except surface salinity) are dynamically predicted with forcing\nby the atmospheric fields of wind stress, wind speed, air temperature, relative humidity and\ncloudiness, which are specified from long-term monthly-mean observations (1950-1979) from\nthe COADS, except for the wind-stress and wind fields which are taken from the Florida State\nUniversity observations from 1961-1987 (Goldenberg and O'Brien, 1981, MWR, 1190-1207).\nBecause the evaporation/precipitation observations are unreliable, the surface salinity field\nis presently forced by a Newtonian method towards the climatological annual mean values.\nThe model is integrated on LLNL's CRAY-2 computers for a +40° latitude region of\nthe equatorial Pacific, with otherwise realistic basin shape and topography. The latitudinal\nresolution is enhanced at the equator (0.57 degrees) and gradually decreases within 10 degrees\nof the equator to approximately 2.4 degree resolution in the mid-latitudes. The longitudinal\nresolution is approximately 2.7 degrees. An implicit time-stepping procedure is used which\nallows a time step of two days. Typical costs of computer time per step are 15 seconds on\nthe LLNL f-machine for a 7-layer, 64 by 54 point horizontal grid.\nThe model was first spun-up for 12 years from initial conditions corresponding to the\nLevitus annual-mean temperature-salinity observations using the monthly-mean atmospheric\nobservations as forcing. The model satisfactorily reproduces the observed circulation patterns\n328","(Figure 1) when forced by the mean forcing functions. Typical model velocities in the South\nEquatorial Current are 30-50 cm/s, while in observations they are 40-60 cm/s. Typical\nmodel velocities in the North Equatorial Current are 10 cm/s, which is of the order of the\nobservations. Typical model velocities in the Equatorial Countercurrent are 10 cm/s, while\nin observations they are about 20 cm/s. The Peru Current is much weaker in the model\nthan in observations. The core of the Equatorial Undercurrent occurs with peak velocity\nnear 30 cm/s, which is weaker than the observed values (near 80 cm/s) probably because\nthe core's width is approximately 4 degrees of latitude wide, rather than +2 degrees as\ntypically observed. The model SST pattern has an equatorial cold tongue comparable to\nobservations and differs from the observed mean SST usually by less than a degree (Figure\n1).\nWe next forced the model with observed (Florida State University) monthly-mean wind\nstress (with air temperature, relative humidity and cloudiness held at the climatological\nmonthly-mean values) from the years 1961 through 1973. The SST anomalies (relative\nto the 61-73 monthly mean values) which occur in this run (Figure 2) are weaker than\nobservations and decay too quickly, since the climatological atmospheric fields result in\nfluxes which damp the anomalies too strongly (e.g., Schopf, 1983, JPO, 1878-1893). The\ntiming of the SST events in the Niño-4 region (Figure 2, left) is reasonably consistent with\nthe observations, apparently since the local wind anomalies are most energetic there. The\nnature of the excitation of SST anaomalies can be seen in Figure 3 which shows the anomalous\nintegrated effects of advection, surface heat fluxes, entrainment and horizontal diffusion on\nthe equation for SST evolution. In the Niño-4 region wind-induced anomalous advection\n(thick line) initially excites the anomalies, followed by strong damping by the anomalous\nfluxes. In the Niño-2 region, anomalous entrainment (thick dashed line) also influences the\nSST anomalies, though they are weak and short-lived there.\nThe sea-level displacements, on the other hand, occur with amplitude and structure\nfairly similar to the observations (Figure 4). This suggests that the oceanic dynamics are\nresponding properly to the variable wind forcing.\nIn order to improve the thermodynamic response of this model, we are considering several\nadditional experiments. One will be to allow atmospheric air temperature (which affects\nsensible, latent and radiative fluxes of heat) to be some function of the model SST, SO as\nto reduce the size of the heat-flux damping imposed by the climatological atmospheric air\ntemperature. Such a relation between air temperature and SST may be phenomenologically,\nobservationally or statistically based. Another experiment will be to use \"observed\" (actually\natmospheric model-derived) fields of air temperature, relative humidity and cloudiness, along\nwith the FSU winds, as forcing agents for these hindcasts. By these experiments, we hope\nto be able to diagnose the relative importance of the latent heat fluxes, radiative heat fluxes\nand the various internal ocean flux processes of different regions of the tropical Pacific.\nAcknowledgments. Support was provided by NOAA Grant NA86-AA-D-CP104 and the\nUniversity of California INCOR program for Global Climate Change. Supercomputing re-\nsources were provided by the Atmospheric and Climate Research Division of the Department\nof Energy through Lawrence Livermore National Laboratories.\n329","SEA SURFACE VELOCITY\nSCRIPPS-PACIFIC\n120 °E\n150 E\n180 °\n150 ow\n128 °V\n90°W\n60°W\nLONGITUDE\nCOMMENTS.\n250.\nMM/S\n=\n-\n250.\nMM/S\n=\nLABEL. ISOPYCNAL OGCM / OBERHUBER / MPI-HH-FRC\nDATE, 11-OCT-89 TIME 09.20.13\nCURRENT SST ERROR\nSCRIPPS-PACIFIC\n7\n2]\nD\n0\nF\nN\nX\n120 E\n150 E\n180 *\n150\n120 ow\n90 °W\n60°W\nLONGITUDE\nREFERENCE LINE\n0\nUNIT\nKELVIN\nMIN= -2\nMEAN= 0\nCONTOUR INTERVAL\n1\n< 0\nMAX= 6\nSTDV= 1\nLABEL\nISOPYCNAL OGCH / OBERHUBER / MPI-HH-FRG\nDATE 11-OCT-89\nTIME. 09.33.50\nFigure 1. (Top) Surface velocity field for December 1 when model is forced by monthly-\nmean FSU winds and monthly-mean COADS fields of air temperature, cloudiness and hu-\nmidity. Length of velocity vector is proportional to the square root of the velocity. (Bottom)\nError in SST field (°K) for December 1 for the same experiment as above.\n330","0.7\n3.0\n0.6\n0.5\n2.0\n0.4\n0.3\n1.0\n0.2\n0.1\n0.0\n0.0\n-0.1\n-0.2\n-1.0\n-0.3\n-0.4\n-2.0\n-0.5\n-0.6\n-0.7\n-3.0\n68\n69\n70\n71\n72\n73\n74\n60\n62\n63\n64\n65\n66\n57\n68\n69\n70\n71\n72\n73\n74\n61\n60\n61\n62\n63\n64\n65\n66\n67\nYEAR\nYEAR\n2. (Left) Time series of SST anomalies in the Nino-4 region filter (0.25/0.5/0.25). for the model\nFigure and observations (thick line), both smoothed with a 3-point ON-5S; dashed), Nino-\n(thin line) series of model SST anomalies in the Nino-2 (90W-coast, not smoothed.\n(Right) Time 5N-5S; thin) and Nino-4 region (160E-150W, 5N-5S; thick), is reasonable\n3 (150W-90W, the timing of the model SST anomaly events in the model Nino-4 region is too small\nAlthough with the observed events, the amplitude of the response humidity and cloudi-\nwhen (especially compared in the Nino-3 and Nino-2 regions), because air temperature,\nness are fixed at the long-term monthly mean state.\n18\n18\n16.\n16.\n14.\n14.\n12.\n12.\n10.\n10.\n8.\n8\n6.\n6\n4.\n4.\n2.\n2.\n0.\n0\n-2\n-2\n-4\n-4.\n-6\n-6\n-8\n-8.\n-10\n-10\n-12\n-12\n-14\n-14.\n-16\n-16.\n-18.\n-18\n67\n68\n69\n70\n71\n72\n73\n65\n66\n60\n61\n62\n63\n64\n65\n66\n67\n68\n69\n70\n71\n72\n73\n60\n61\n62\n63\n64\nYEAR\nYEAR\n3. Time series of the anomalous model SST budget (instantaneous and 2 monthly-\nFigure contribution minus long-term monthly-mean value) for the Nino-4, 3 regions.\nmean Units in 108 o K/s for total surface heat flux (sensible, latent, and radiative; entrainment thin line),\nhorizontal are advection (thick line), horizontal diffusion (thin dashed line) and then\n(thick dashed line). In the Nino-4 region, advection initially excites the SST anomalies,\nthe surface fluxes rapidly extinguish them.\n331","01\n91\nDI\nB\n21\n01\n9\n8\n9\n2-\n9-\n8-\n9-\n01\n21.\n8-\nD1-\n01\n91-\n09\n19\n29\nE9\n99\n99\n99\nL9\n89\n69\n01\n12\n2L\nCL\n09\n19\n29\nE9\n99\n59\n99\n19\n89\n69\nOL\n12\n2L\nEL\nREAL\n02\n91\n81\nD1\n91\n21\nDL\n01\n21\n01\nB\n8\n9\nlike\n9-\nB-\n01-\n21-\n01-\nDI\n21-\n91.\n81-\n02-\n91.\n09\n19\n29\n£9\n99\nS9\n99\n19\n89\n69\nOL\nIL\n21\nCL\n09\n19\n29\nE9\n99\n59\n99\n19\n89\n69\n0L\n12\n21\nEL\n'(MLL 'SSI) JO SOLIES\n-qo pue pour JOJ 'NL) pue (M18 'NG) \"I '(MLSI'NG)\n7d-E e sey SOLIDS The\nLiou SI pur 07 esuads LL\nresuodse.ru the the\nZEE","Statistical prediction of January-February mean North American climate\nfrom the 11-year solar cycle and ENSO, for west QBO phase years.\nAnthony E. Barnston and Robert E. Livezey\nClimate Analysis Center, NMC/NWS/NOAA\nConsiderable attention has been focused recently on an apparent association\nbetween 10.7 cm solar flux and the tropospheric Northern Hemisphere winter climate\nfor QBO phase-stratified (west or east phase) samples of years (van Loon and Labitzke\n1988). Global (or field) significance tests on the field of correlation between\nJanuary-February mean 700 mb height and solar flux, accounting for both temporal\nand spatial autocorrelational features, produced highly significant results for the\n1951-1988 period, especially for the west QBO phase (Barnston and Livezey 1989;\nhenceforth called BL89).\nIn January-February of 1989, however, the 700 mb height anomalies were strong-\nly opposed to those expected for the west QBO phase with a highly positive solar\nflux anomaly. The correlation field of 700 mb height versus flux for the west QBO\nphase was thus considerably weakened (Fig. 1) relative to that for the 1951-88\nperiod (Fig. 2 in BL89). The field significance of the Fig. 1 map is 0.14, in\ncontrast to 0.003 for the 1951-88 period.\nA clue to the 1989 exception is found by considering the correlation between a\nSouthern Oscillation Index (SOI; low values correspond to E1 Niños or warm ENSO\nepisodes) and 700 mb height (Fig. 2). This SOI is based 50 percent on the Tahiti\nminus Darwin sea level pressure and 50 percent on the SST in the maximally associ-\nated central tropical Pacific area 10S-10N, 120-160W. Note that over North America\nthis pattern is quite similar to that of the flux-height correlation field for west\nQBO phase (Fig. 1), although over the Pacific they differ. The cold ENSO event of\n1988-89 apparently superseded the solar/QBO situation in forcing the January-Febru-\nary climate pattern.\nInspection of the historical data reveals that for west QBO phase years from\n1951-88 there was a weak negative correlation between the SOI and solar flux; i.e.,\nwarm ENSO events tended to occur slightly more frequently during high flux than low\nflux winters, perhaps bolstering the apparent relationship between flux and North\nAmerican climate. After 1989 this relationship is nearly neutralized, presumably\nhaving been due to chance in the small (21-year) 1951-88 sample.\nThe seeming influence of both quasi-independent factors, ENSO and solar flux,\non North American January-February mean climate for west QB0 phase years lends\nitself to a two-predictor multiple linear regression model predicting 700 mb height\n(or surface temperature) at a given location or over a relevant area.\nThe result of such a multiple regression for 1951-89 is shown in Fig. 3 for\nNorthern Hemisphere January-February 700 mb height, with significant (p<0.05) flux\ncoefficients underlined at the individual predictand grid points in question and\nsignificant SOI coefficients not underlined. Because the data were standardized\nprior to the analysis, the coefficients are the partial correlations between pre-\ndictor and predictand. Locations having an R2 value (unadjusted) of 0.5 or higher\nare circled, and 0.4 to 0.5 broken-circled. The asterisk denotes insignificant\nindividual predictor coefficients but a significant regression equation, implying\nsome participation on the parts of both predictors. The SOI appears to have a more\npervasive effect on the climate, but the solar effect is strong in several loca-\ntions and thus worthy of consideration. Areas that appear to be affected by both\nfactors are (1) the western Atlantic near 30°N, (2) Canada and (3) North Africa.\nThe first two apply to North American climate and will be studied further. A simi-\nlar analysis applied to mean January-February U.S. temperature (not shown) indi-\ncates significant effects from both the SOI and flux in the southeast U.S., as\n333","opposed to a relative domination there by SOI in the heights. Regarding the latter\nfeature, it is noted that for the 22-year west QBO phase sample the correlation\nfield of SOI with 700 mb height (not shown) is similar to Fig. 2 but has a stronger\nsoutheast U.S. positive center and a weaker Canadian negative center.\nOn the basis of the regression results, a general ENSO/flux-related dipole\nregion over North America was defined for the west QBO phase years. One pole of\nthe dipole is over Canada, comprising the area 35-120°W, 55-65° and the other\npole is in the southeast U.S./western Atlantic, with area 45-100°L 25-35°N. With-\nin the area of either pole each of the two predictors acts in the same direction,\nalthough at some grid points one predictor may overshadow the other. The Aleutian\nIslands area is not used as a third component of the index because the two predict-\nors combine differently there (the flux affects the heights in the same sense as in\nCanada, whereas the SOI has the opposite sign of effect as in Canada). An appro-\npriate index for the Atlantic-Canada dipole, based on the corresponding seesaw\nfeature in the relationships of both the SOI and the flux with the heights, is the\n700 mb height difference between the Atlantic and the Canada areas. Either a high\nflux or a low SOI (warm ENSO event) would tend to produce negative difference anom-\nalies (i.e., positive height anomalies in Canada and negative anomalies in the\nAtlantic), and vise versa.\nWhen the multiple regression model is applied to the Atlantic-Canada differ-\nence index itself, the relationship is of approximately the same strength as near\nthe center of one of the dipole poles. Specifically, R2=0.43, the SOI coefficient\nis 0.46, the flux coefficient is -0.42, and the significance p-values for the SOI,\nthe flux, and the regression equation as a whole are 0.008, 0.026 and 0.005, re-\nspectively. The strength of the relationship between the dipole difference index\nand each of the two predictors is shown in Fig. 4, where the abscissa is the stan-\ndardized SOI, the ordinate the standardized solar flux, and in the body of the\nfigure are individual year values of the standardized (X100) dipole height differ-\nence index. Values greater than +1 standard deviation are encircled (implying\nsubnormal Canadian heights and/or above-normal southeast U.S./Atlantic heights),\nand those more negative than -1 standard deviations are enclosed in squares.\nDotted enclosures denote greater than one-half standard deviation departures. Some\nbroad clustering is quite apparent, with years having high height differences tend-\ning toward the lower right portion of the domain. The 1989 value defies this,\nbeing in the upper right portion. While the 1989 predictand value might be expected\nto be much weaker (being forced oppositely by the flux versus the SOI), its high\npositive value can at least be explained by the high SOI value, leaving the sug-\ngested relationship with the flux much less damaged than it would be in a univari-\nate regression model based on the flux alone. Note that 1989 is the only year in\nthe sample having both a high flux and a high SOI, underscoring the serious samp-\nling problem arising from the stratification by QBO phase and from the use of two\npredictors instead of one. Note also the weak negative correlation between flux\nand SOI when the 1989 point is removed, and the associated lack of points with low\nvalues for both predictors (which still remains after including 1989).\nThe 700 mb height difference index might be regarded as an a posteriori pre-\ndictand in the sense that it was defined on the basis of an exhaustive series of\nmultiple regression experiments using each of 358 grid points of 700 mb height as a\npredictand. A more logically formulated predictand might be a principal component\ntime series associated with a selected January-February mean 700 mb height loading\npattern that emphasizes the Canadian-southeast U.S./Atlantic seesaw noted in Figs.\n1 and 2. Candidate patterns are (1) the (rotated) Tropical/Northern Hemisphere\n(TNH) pattern (approximated by Figs. 5c and 5d in Barnston and Livezey 1987), (2)\nthe (rotated) Pacific/North American (PNA) pattern (Fig. 3d in same paper), (3) the\n(rotated) North Atlantic Oscillation (NAO) pattern (Fig. 2a in same paper), and (4)\nthe (unrotated) mode 1 pattern (Fig. 19 in BL89).\n334","The resulting regression equation performance, including predictor coeffi-\ncients (partial correlations in this case), is shown for each candidate predictand\nin Table 1. (Coefficient signs are keyed for trough over Canada.)\nCoefficient (followed by p-value)\nEquation\nR 2 (p-value)\nFlux\nPredictand\nSOI\n.43 (.005)\n700 mb height dipole\n.46 (.008)\n-.42 (.026)\ndifference\n-.42 (.045)\n.28 (.042)\n.28 (.10)\nTNH amplitude\n.48 (.002)\n.69 (.001)\n.18 (.28)\nPNA amplitude\n.12 (.52)\n-.21 (.37)\n.06 (.54)\nNAO amplitude\n.49 (.013)\n-.22 (.26)\n.32 (.026)\nunrotated mode 1\namplitude\nTable 1. Normalized multiple regression coefficients and R2 with significance\nvalues for 22 cases of west QB0 phase, using several January-February North\nAmerican climate indices as predictand and SOI and solar flux as predictors.\nThe coefficients indicate that while the 700 mb height difference index may\noptimize the combined effects of the SOI and the flux, the levels and significances\nof those effects are also attainable using amplitudes of previously established\nmodes of 700 mb circulation. These modes tend to be associated with one predictor\nat a time: the PNA and unrotated mode 1 with the SOI, and the TNH more with the\nflux than the SOI. Some sensitivity of these indices to future inclusion of new\ndata points is anticipated. It may take 20 to 40 years (resulting in 11 to 22 new\nwest QBO phase data points) for the relationships (and, to a lesser degree, the\nmodes) to stabilize; learning more about the physical mechanisms underlying the\nsolar and ENSO effects may also progress during these years. The possibility of\nnonstationarity in the climate system further complicates our task. In the mean-\ntime, the apparent effects may be cautiously accepted and used in the presently\nillustrated statistical mode to help develop operational winter long-range forecast\ntools at the Climate Analysis Center. Such efforts are already in progress, both\nin an analog system (Livezey and Barnston 1988, Barnston and Livezey 1989) and in a\nregression scheme as shown here. A similar study for east QB0 phase years is\nunderway.\nReferences\nBarnston, A. G. and R. E. Livezey, 1987: Classification, seasonality and per-\nsistence of low-frequency atmospheric circulation patterns. Mon. Wea. Rev.\n,\n115, 1083-1126.\nBarnston, A. G. and R. E. Livezey, 1989: An operational multifield analog predic-\ntion system for United States seasonal temperatures. Part II: Spring,\nsummer, fall and intermediate three-month period experiments. J. Climate, 2,\n513-541.\nBarnston, A. G. and R. E. Livezey, 1989: A closer look at the effect of the 11-\nyear solar cycle and the quasi-biennial oscillation on Northern Hemisphere 700\nmb height and extratropical North American surface temperature. J. Climate,\n2, 1295-1313.\nLivezey, R. E. and A. G. Barnston, 1988: An operational multifield analog predic-\ntion system for Unites States seasonal temperatures. Part I : System design\nand winter experiments. J. Geophys. Res., 93, 10,953-10,974.\nvan Loon, H., and K. Labitzke, 1988: Association between the 11-year solar cycle,\nthe QBO, and the atmosphere. J. Climate, 1, 905-920.\n335","..\ng\nL20Hg\n55\nL-13\n4-14\n56\n38\nL\n42\n42\n####\n100\nFigure 1. o Correlation (X100) between 10.7 cm solar flux and 700 mb height for\nthe west QBO phase for the January-February period (22 years, 1951-1989).\n1246\n1006\n005\n604\n-65\n-\n1-36\n60\n44\nH\n57\n1120\n1000\n-\n000\nFigure 2. . Correlation (X100) for January-February - between a Southern Oscil-\nlation index (Tahiti minus Darwin SLP and SST in area 120-160W, , 10N-10S) and\n700 mb height (39 years, 1951-1989).\n336","58\n33\n*\n*\n40\nBOTTH\n20\n337","2\n-192\n147\n68\n-135\n16\n32\n974\n36\n93\n2\n-3\n3\n2\nFigure\nheight\n100.\n338","Prediction of Forecast Skill using Monte Carlo Ensembles of 30 Day Forecasts\nDavid P. Baumhefner\nNational Center for Atmospheric Research, 2 Boulder, Colorado 80307-3000\nThe concept of estimating forecast reliability, or in other words predicting forecast\nskill, was first proposed in the theory of stochastic-dynamic prediction (Epstein, 1969).\nIt was suggested (Leith, 1974) that an approximation to the full set of equations for\nstochastic-dynamic forecasts might be achieved using a Monte Carlo sample of forecasts.\nThese basic ideas have been applied to the monthly mean deterministic forecast problem.\nPreliminary forecasts results have shown a large variation in skill of 30 day forecasts.\nTherefore, it will be necessary to determine in some way the accuracy of the operational\nforecasts beforehand.\nIn the past, the Monte Carlo technique was difficult to implement because the mod-\neling error was much larger than errors arising from initial uncertainty. These error\ncharacteristics would overwhelm any predictability signal that might be present in the\nMonte Carlo ensemble. The present-day models (GCM's) that are used to simulate climate\nscenarios have reduced many of the modeling error sources such that the two sources of\nerror are now nearly equal in many cases. If the error sources are of the same magnitude,\nthen the signal from the ensemble can be more easily determined.\nThe relationship of the characteristics of a Monte Carlo ensemble to actual forecast\nskill was examined by testing an extensive set of integrations. Thirty-six initial winter\nstates throughout the decade of the 80's were used to construct forecast ensembles. Each\nensemble contained 10 member forecasts which were integrated from slightly altered initial\nanalyses to simulate the spread in forecast skill from initial state uncertainty. These\nensembles were then analyzed for the actual forecast skill of the ensemble mean, the spread\nof the individual members' skill, and the dispersion of the forecasts compared to each other.\nThe skill of each ensemble is illustrated in Figure 1 for the 30 day time-mean of the\n500 mb geopotential. The anomaly correlation between forecast and observed conditions\nis calculated as the skill measure and a value of greater than .5 is considered skillful. The\nopen circle on each vertical bar locates the skill of the ensemble mean forecast (all forecasts\naveraged together), while the solid vertical line outlines the range of skill of the individual\nmembers. The average skill of the ensemble mean for the entire 36 initial states is .42 with\nskill ranging from a high of .73 to a low of -.18 for each date. Fifteen ensembles have useful\nskill (.5), and if the model were further improved only slightly, another 10 or so cases\nwould also become skillful.\n1 Proceedings of the 14th Climate Diagnostics Workshop, Climate Analysis Center,\nNMC/NWS/NOAA, WWB Room 606, 5200 Auth Road, Washington, D.C. 20233\nThe National Center for Atmospheric Research is sponsored by the National Science\n2\nFoundation.\n339","Clearly, it would be very useful to determine which forecasts would be accurate in\nsome a priori manner. Examination of individual ensembles for 15DEC88 and 1JAN89\n(shown in Figure 2) reveals a possible relationship. The anomaly correlation score from a\ncomparison of daily values of each member forecast in the ensemble to observed conditions\nis shown on the left: the 15DEC88 case on top and the 1JAN89 case on the bottom. These\nplots measure the forecast skill of the ensemble. The anomaly correlation of each member\nforecast compared to the other nine members is presented on the right-hand side. These\nplots show the dispersion, or error growth, due to uncertainties in the initial condition.\nWith 10 members in each ensemble, there are 45 pairs that can be evaluated for dispersion.\nAt first glance it appears that less skillful forecasts (15DEC88) have a more rapid dispersion\nof individual forecast comparisons in the ensemble than do more accurate forecasts. This\nwould make sense if the actual forecast error was strongly affected by the initial state\ninstability in some manner. In other words, the predictability error growth characteristics\nof each ensemble may contain information regarding the actual skill of the ensemble.\nA first attempt to establish a quantitative relation between skill and prediction is\nplotted in Figure 3. The forecast skill, as measured in this case by the standard deviation\n(SD) of the error field, is correlated in a scatter-plot with the mean spread (also measured\nas a SD) of the ensemble forecasts. As can be seen, the relationship between parameters\nis virtually non-existent for the 30 day mean (left). Examination of Figure 2 suggests\nthat the different rates of dispersion maximize at the middle period of the 30 day forecast.\nTherefore, the 10 day mean centered at days 10-19 was also analyzed, with similar results.\nThe reasons for the poor relationship so far are undoubtably quite complex. It appears\nthat some of the ensembles contain a signal but they are clearly in the minority. Since\nthe number of samples is reasonably large, it may be possible to sub-categorize ensembles\nthat may better fit the condition of equal error sources mentioned in the introduction.\nProblems with the modeled variance and the mechanisms that produce the variability at\nthe one month time scale are also possibly playing a role by distorting the estimates of\ndispersion within the ensemble. The current method of sampling to produce the Monte\nCarlo ensemble is probably not sufficient to estimate the correct dispersion as well. Finally,\nthe measures currently being used to estimate the predictability need to be examined,\nrefined and expanded to better represent the dispersive characteristics of the ensemble.\nReferences\nEpstein, E. S., 1969: Stochastic dynamic prediction. Tellus, 21, 739-759.\nLeith, E., 1974: Theoretical skill of Monte Carlo forecasts. Mon. Weather Rev., 102,\n409-418.\n340","FORECAST SKILL (A.C.)\n30 DAY TIME AVERAGE\n20-90N\n500 MB HEIGHT\n.5\no\n86-87\n87-88\n88-89\n8I-82\n82-83\n83-84\n84-85\n85-86\n78-79\n79-80\n80-81\nFigure 1\n341","HT1\nFIELD ON PRESSURE SURFACE 500MB\nHT1\nFIELD ON PRESSURE SURFACE 500MB\nAREA, 1-180.0 180.0 20.0 90.0)\nAREA. 1-180.0 180.0 20.0 90.0)\nMODES. M=1 0.901 N=1 0.90)\n15DEC88 12Z\nMODES. M=1 0.90) N=1 0.90)\n15DEC88 12Z\n11\n11\n1.0\n1.0\n..\n09\n0.0\n0.0\n0.7\n0.7\n0.6\n6.6\n0.5\n0.5\n0.4\n04\n8.3\n# 3\n8.2\n02\na\ne\n6.0\nB.C\n-0.\n-\n-0.2\n-0.2\n-0.3\n\"\n-0.4\n-0.4\n0\n18.11.12.13.14\n18\n11.\nDAYS\nDAYS\nHT1\nFIELD ON PRESSURE SURFACE 500MB\nHT1\nFIELD ON PRESSURE SURFACE 500MB\nAREA, (-180.0 180.0 20.0 90.0)\nAREA, (-180.0 180.0 20.0 90.0)\nMODES. M=( 0.90) N=1 0.90)\nMODES. M=( 0.90) N=1 0.991\n1JANB9 00Z\n1JAN89 00Z\n11\n1.1\n1.0\n1.8\n..\n04\n0.0\n0.0\n0.7\n0.7\n8.6\n0.0\n0.5\n9.5\n9.4\n8.4\n0.3\n0.3\n0.2\n0.2\n8.1\n0.1\n0.0\n0.8\n-e\n-\n-0.2\n-4.2\n-03\n-0.3\n-6.4\n-4.4\n0.\n1.\n8.\n32\nDAYS\nDAYS\nFigure 2\n342","150-60W;25-70N\n500 MB HEIGHT 10 DAY AVE (10-19)\n150-60W,25-70N\n500 MB HEIGHT\n30 DAY AVE\n160\n90\n140\n80\n120\n70\n100\n60\n80\n50\n60\n40\n40\n60\n70\n80\n90\n35\n40\n45\n50\n55\n50\nMEAN SPREAD (S.D.)\nMEAN SPREAD (S.D.)\nFigure 3\n343","Potential for Long-Range Predictability of\nMonthly Mean Surface Temperatures\nover North America\nDennis J. Shea and R. A. Madden\nNational Center for Atmospheric Research\n1\nBoulder, Colorado 80307\n1. Introduction\nA certain amount of variability of time-averaged meteorological data is to be expected\ndue to statistical sampling variations alone. That is, one monthly average of temperature\nat a given location will differ from another because each is a finite time average of daily\nvalues which result from a continuous process of passing weather systems. It represents\nvariability that cannot be deterministically predicted at long range (thought to be about\ntwo weeks). This sampling variability of time-averaged weather which is strictly due to\nthe atmosphere's internal dynamics will be referred to as natural variability or \"climate\nnoise\". Here, estimates of the climate noise for monthly averages over North America\nare made and compared with the actual interannual variability of monthly means. The\ninterannual variability of monthly means is assumed to consist of a \"climate signal\" and\nthe climate noise. The climate signal is assumed to arise from changing external conditions\nsuch as slowly varying sea surface temperatures, the greenhouse effect, etc. Generally, any\nprocess whose time scale is long relative to daily weather fluctuations is considered to be\nexternal. Where the interannual variability of monthly means exceeds that suggested by\nclimate noise it is assumed that the signal is large enough to potentially allow long-range\npredictability.\nThis work extends an earlier paper on the subject (e.g., Madden and Shea, MWR,\n1978, hereafter MS) by using 25 to 80 years of daily temperature data from 350 U.S. and\nCanadian stations versus 22 to 24 years of daily data from 107 stations over the U.S. only,\nthat were used in MS. In addition, comparisons with data from a 6000-day, perpetual\nJanuary general circulation model (GCM) help to demonstrate the credibility of the \"low-\nfrequency white-noise\" (LFWN) statistical model that forms the basis for estimating the\nclimate noise.\n2. Estimating climate noise\nEstimating the potential for long-range predictability is approached as an analysis of\nvariance problem. The variance within seasons, which is presumed to consist of only\nvariability due to weather systems that are unpredictable at long range (climate noise), is\ncompared to that between seasons which is presumed to be made up of both the climate\nnoise and any potentially predictable climate signals. To estimate the climate noise, spectra\nwithin seasons are computed using daily data and from them the variance of monthly means\n1\nThe National Center for Atmospheric Research is sponsored by the National Science\nFoundation.\n344","is determined. That is\n(1)\ndf\nwhere ON is the variance of the climate noise of a time-average of length N. For monthly\nmeans N is 30 or 31. S(f) is the spectrum of daily data within seasons, f is the frequency\n(day-1) and H2 (f) is the power transfer function of the process of averaging N values.\nThe spectrum of daily data at each station S(f) was estimated by the following procedure\noutlined in MS.\nThe actual variance of monthly means is computed in the standard way, that is\n(2)\nwhere n is the number of months, Ti is the mean temperature of the ith month, and\nT is the grand mean temperature over all available years. The ratio 02/02 then gives\na measure of the potential for long-range predictability of monthly mean temperatures.\nWhere 02 > o2 we assume that the excess variance is, at least, potentially predictable.\nIn principle, the separation of variability to that within and that between seasons is\nstraightforward. In practice, we really want to separate variance caused by internal dy-\nnamics (natural variability) as manifest by the daily weather from that possibly caused\nby changing external conditions which might constitute predictable signals. Many of the\nproblems involved have been underscored by Shukla (MWR, 1983), Madden (MWR, 1983),\nand Trenberth (MWR, 1984). An especially troublesome one is the fact that slowly chang-\ning external conditions probably have some influence on the daily data within a season.\nThis could make the autocorrelation in the data larger than it would be under the influ-\nence of only weather variations. As a consequence the low frequency, 1/96 day-1, spectral\nestimate and thus o2 will be overestimated. Trenberth (MWR, 1984) used simulated\ndata to study this problem. He added a reasonable quasibiennial oscillation to climate\nnoise modeled by a red-noise process, and found that the methods used here only slightly\noverestimated the true climate noise.\n3. Results\nEstimates of the standard deviation of the climate noise, ON, of monthly mean of surface\nair temperatures for winter, spring, summer and fall based on eqn. (1) are shown in Fig 1.\nGenerally, minimum values of ON tend to be found along the coasts of the continent, ON\nestimates are less in the south than in the north and maxima are found in the interior and in\nwinter. During winter, the largest ON values lie over the Rocky Mountains. As an example,\nwe interpret these ON estimates to mean that 31-day monthly average temperatures during\nwinter over parts of Montana and Alberta would be approximately normally distributed\nwith a standard deviation of >4°C simply as a result of sampling variations due to daily\nweather variability.\nThe ratios 02/02 (Fig. 2) are similar in most respects to corresponding monthly values\npublished in MS over the United States. Some differences are found over the southeastern\n345","U.S. where the previous ratios exceeded 2.0 for January and are less than 2.0 here in Fig. 2.\nThe relative minimum in the ratio over the central U.S. during January is a feature that is\ncommon to other studies. Not surprisingly potential predictability tends to be largest near\nthe coasts in proximity to the long-memory oceans. There are, however, relative maxima\nover the interior of Canada during all seasons but winter.\nThe \"signal-to-noise\" ratios are F-ratios with known degrees of freedom in both the\nnumerator and denominator. Thus significance levels may be calculated. For example,\nareas over the Pacific Northwest during winter are significantly larger than one at the 5%\nlevel. However, we prefer to interpret the potential predictability patterns of Fig. 2 in a\nrelative sense. That is, the potential for long-range predictability is greater over the west\ncoast of North America than over the center of the continent in January, and greater in\nJanuary than July (not shown) over the Canadien Rockies, etc.\nA ratio of 1.5 means that 33% of the total variability exceeds climate noise and is there-\nfore potentially predictable. For a ratio of 2.0, 50% of the total variability is potentially\npredictable. If we were to be able to predict 50% of the variance, we could predict shifts in\ntemperature distributions from 50/50 (above/below average) to 25/75 more than 50% of\nthe time (see Madden, JC, 1989). These would likely prove to be very useful predictions.\nOn the other hand, over the central U.S. where ratios tends to be relatively small the\nability to accurately predict such large shifts in temperature would be rare.\nSince 02 which is assumed to contain any potentially predictable climate signal is deter-\nmined by monthly values separated by one year, these estimates of potential predictability\nstrictly refer to predictions made a year or more in advance. However, we suspect that, at\nleast qualitatively, they apply to predictions with shorter lead times.\n4. Sensitivity of the LFWN assumption\nOf the several assumptions used to determine the estimates of potential predicability,\nthe most crucial is the LFWN extension. After multiplying the power transfer function\nH2 (f) with the spectrum S(f) the noise estimates are primarily determined by the spectral\nestimates at zero cycles/day (i.e., the LFWN estimate) and the lowest resolved estimate,\n1/96 day-1. We have tested this using a long GCM integration. The sensitivity of the\ncalculated estimates (Fig. 2) to other assumptions is also briefly described.\na) Model\nData generated by long runs of GCM's with constant external conditions provide an\nopportunity to test the credibility of the LFWN model in principle. That is, they can\nprovide insights into the low frequency behavior of idealized weather variations.\nData from a 6000-day perpetual January run from the NCAR Community Climate\nModel (\"CCM0B\"), was examined to specifically test the LFWN assumption. The model\ngrid is Gaussian with a resolution of about 4.4° of latitude and 7.5° of longitude giving 1920\ngrid points. The model was run with fixed sea surface temperatures. At each grid point\nthe 6000 day time series of temperatures nearest to but not at the surface (991 mb) was\nbroken into a variable number of 32, 48, 72, 96, 128 and 192 day \"seasons\" or segments\nafter the 6000-day model linear trend was removed to minimize model bias. This leads to\n346","the LFWN assumption spanning 1/32 day to 1/192 day-1 Fifty seasons with lengths\nof 32, 48, 72 and 96 days were used while for the 128 and 192 day seasons 40 and 27\nseasons were used, respectively. For example, at each of the 1920 grid points a time series\nwas divided into 50 seasons of 96 days. Each 96 day season was separated by 24 days to\nassure independence. The spectra S(f) were computed as described in MS. The LFWN\nextension was used to complete the spectra. The middle 31 days of each season were used\nto compute the 50 \"monthly\" means. The variance about the overall monthly mean was\ncomputed using eq. 2. Since the ratio 02/02 is an F-ratio, the distribution of ratios from\nall grid points should approximate that of the F-distribution if 02A = O2. The distribution\nof ratios from the 1920 grid points of climate model data, the theoretical F-ratios and\ntheir 5 and 95% confidence limits are presented in Fig. 3. Because the model results are\nsystematically displaced to the right of the theoretical values for season lengths of less than\n96 days, the potential predictability estimates are larger than would be expected, while\nuse of season lengths of 96, 128 and 192 days yield ratios which agree reasonably well with\nthat expected by theory. Given sampling variations, these results indicate that the LFWN\nmodel is a good assumption when season lengths are about 96-days or so.\nb) Observations\nThe CCM comparison suggests a seasonal segment should be at least 96 days or SO long\nto avoid underestimating the climate noise. However, in practice long seasonal segments\nincrease the possibility that the resulting spectra will be affected by changing external\nconditions. To assess the magnitude the calculations were repeated for 48- and 32-day\nseasons during winter. The ratios for the 48-day segment are very similar to those shown\nin Fig. 2. The only exceptions are the West Coast and the Southeast United States, where\nthe ratios increased slightly. This is to be expected from the model comparisons of Fig. 3.\nIt is an indication that the spectra at these stations are still \"red\" while for the majority of\nstations the spectra have essentially become \"white\" by 48 days. The ratios for the 32-day\nsegments are reduced everywhere. The increase in ratios is geographically dependent with\nsome large increases west of the Rockies while in the center of the U.S. and Canada the\nratios increased only 1 to 10%. In neither case do the basic patterns change. That is,\nthe potential for long-range predictability over the central portions of North America is\nconsiderably less than that over coastal oceans.\n5. Conclusions\nEarlier work by MS has been extended to include Canada. The updated estimates of\nclimate noise and potential predictability have been determined using more stations with\nlonger time series and are generally consistent with those published earlier. The crucial\nLFWN assumption used to obtain these estimates has been shown to be a good model by\nusing a 6000-day run of the NCAR CCM in perpetual January mode. In winter, the noise\nestimates are somewhat sensitive to the length of the season used to calculate the seasonal\nspectra from which the noise estimates are obtained. Using a seasonal length of 96 days\nfor observed data seems to be a prudent choise.\n347","3.5\n30\n30\n4,0\n3.5\n3.0\n3.0\n2.5\n2.0\n4.0\n2.0\n1.5\n2.0\nJ....A\n2.0\n2.0\n1.5\nFigure 1. Standard deviation in °C of the estimated climate noise (ON) for 31-day monthly\nmean temperatures during winter. Dots represent location of stations.\n1.5\n2.\n2.0\n2.5\nI.5\nno\n3.0\n2.5\n3.0\n2.0\nI.5\nI.5.\n2.0\n2.0\n1.5\n2.0\nFigure 2. Ratio of actual interannual variance of January monthly mean temperatures to\nthe variance of the climate noise for winter.\n348",".24\n.24\n.22\n32\n.22\n48\n.20\n.20\n.18\n.18\n.16\n.16\n.14\n.14\n.12\n.12\n.10\n.10\n.08\n.08\n.06\n.06\n.04\n.04\n.02\n.02\n0\n0\n0\n.2\n.4\n.6\n.8\n1.0\n1.2\n1.4\n1.6\n1.8\n2.0\n2.2\n2.4\n2.6\n0\n.2\n.4\n.6\n.8\n1.0\n1.2\n1.4\n1.6\n1.8\n2.0\n2.2\n2.4\n2.6\n.24\n.24\n.22\n.22\n72\n96\n.20\n.20\n.18\n.18\n.16\n.16\n.14\n.14\n.12\n.12\n.10\n.10\n.08\n.08\n.06\n.06\n.04\n.04\n.02\n.02\n0\n0\n0\n.2\n.4\n.6\n.8\n1.0\n1.2\n1.4\n1.6\n1.8\n2.0\n2.2\n2.\n4\n2.6\n0\n.2\n.4\n.6\n.8\n1.0\n1.2\n1.4\n1.6\n1.8\n2.0\n2.2\n2.4\n2.6\n.24\n.24\n.22\n.22\n128\n192\n.20\n.20\n.18\n.18\n.16\n.16\n.14\n.14\n.12\n.12\n.10\n.10\n.08\n.06\n.06\n.06\n.04\n.04\n.02\n.02\n0\n0\n0\n.2\n.4\n.6\n.8\n1.0\n1.2\n1.4\n1.6\n1.8\n2.0\n2.2\n2.4\n2.6\n0\n.2\n.4\n.6\n.8\n1.0\n1.2\n1.4\n1.6\n1.8\n2.0\n2.2\n2.4\n2.6\nF-RATIO\nF-RATIO\nFigure 3. Model (light) and theoretical (dark) F-distributions for 32, 48, 72, 96, 128 and\n192 day season lengths\n349","6-10 DAY PROBABILITY FORECASTS OF DAILY PRECIPITATION FREQUENCY\nWilliam H. Klein\nCooperative Institute for Climate Studies, University of Maryland\nCollege Park, MD 20742\nEdward S. Epstein\nClimate Analysis Center, NMC, NWS, NOAA\nCamp Springs, MD 20233\nThree times a week, the Climate Analysis Center (CAC) of the National Weather\nService routinely forecasts the class of precipitation amount across the contiguous United\nStates during the 5-day period from 6 to 10 days in advance of forecast day. Because of the\ndifficulty of predicting quantitative precipitation that far in advance, CAC is planning to\nchange the format of its 6-10 day forecast from class of precipitation amount to probability of\n0, 1, and 2 or more (2+) days of measurable precipitation (at least .01 inches in 24 hours)\nduring the 5-day period. The climatological background for these predictions consists of\nharmonically smoothed frequencies of 36 years of precipitation data for the period 1948-1983\nat a network of 146 surface stations (fig. 1). The results provide \"normal\" probabilities of 0, 1,\nand 2+ days for each station and each 5-day period. In order to provide guidance for the new\n6-10 day forecasts, we are now developing two sets of multiple regression (specification)\nequations for each station and each month by applying a forward selection procedure to screen\nthe observed probability anomalies (obtained by taking the difference between the normals and\neither 0 or 1, depending on the observed event) as a function of the field of concurrent 5-day\nmean 700 mb height anomalies. The equations give the anomalies of the (unconditional)\nprobabilities, the first for 0 wet days in 5, the second for 2+ wet days in 5. Since the\nanomalies must add to 0, the probability anomaly of exactly 1 wet day in 5 then equals the\nnegative of the sum of the two derived anomalies. The forecasts can be expressed as actual\nprobabilities by adding the normal probabilities to the specified anomalies.\nThe potential predictors consist of 5-day mean 700 mb height anomalies (H) analyzed\nat a network of 79 gridpoints over North America and surrounding oceans (fig. 2) during the\n30-year period 1951-1980. As with precipitation, normal values of 700 mb height were\ncomputed for each day of the year by harmonically smoothing running 5-day means. Other\nvariables that are more directly related to precipitation, such as humidity, vertical motion, and\ndaily synoptic features, were not included because we doubted they could be forecast skillfully\nbeyond six days. Four candidate predictands, or dependent variables, were defined as:\nPAO = anomaly of the unconditional probability of 0 wet days in 5\nPA1 = anomaly of the conditional probability of 2+ wet days, given that the number of\ndays with precipitation (NDP)>0\nPA2 = anomaly of the unconditional probability of 2+ wet days in 5\nPA3 = anomaly of the conditional probability of 0 wet days, given that NDP<2.\nIn terms of the climatological probability of 0 wet days in 5 (CPO) and the climatological\nprobability of 2+ wet days (CP2), these predictands become:\nif NDP = 0\nPAO =\n1\n- CPO\nif NDP > 0\n0\n- CPO\n=\nif NDP = 1\nPA1 = - CP2/(1-CP0)\nif NDP > 1\n1 - CP2/(1 - CPO)\n=\nif NDP = 0\nnot defined\nif NDP < 2\nPA2 =\n0 - CP2\nif NDP > 1\n1\nCP2\n=\n-\nif NDP = 0\nPA3\n1 - CPO/(1 - CP2)\n=\nif NDP = 1\n- CPO/(1 - CP2)\n=\nif NDP > 1\nnot defined\n350","In the forecast mode, three pairings of these probability anomalies will allow estimates to be\nmade of the probabilities (p.) of each category (0, 1, and 2+):\npo = CPO + PÃO\nSystem A.\nP1 = [1-po] [1-CP2/(1-CP0) - PÂ1]\nP2 = [1-po] [PA1+CP2/(1-CP0)]\nPo = [1-P2 ] [PA3+CP0/(1-CP2)]\nSystem B.\nP1 = [1-p2] [1-CP0/(1-CP2) - PÂ3]\nP2 = CP2 + PÃ2\nPo = CPO + PÃO\nSystem C.\nP1 = 1 - Po P2\nP2 = CP2 + PA2\nwhere the carat () indicates a predicted value.\nWe combined each month with its two adjacent months and used overlapping five-day\nperiods, centered on each day of the year, to enhance the sample size for the screening\nregression. Since these procedures made it difficult to estimate the true number of degrees of\nfreedom, we could not apply classical tests of significance. Also, Monte Carlo simulation\nwould be too time-consuming. Therefore, we arbitrarily terminated our screening procedure\nby using the first three variables selected in each multiple regression equation. This cut-off\ncriterion was adopted because it is completely objective, leads to simple equations that are\neasy to understand, approximates equations obtaining by stopping when a new variable adds\nless than 1% to the explained variance, and sacrifices little predictive power while maintaining\nstatistical stability. Before deciding on a final system for operational use, we experimented\nwith the four candidate predictands and three specification modes described above. We\nselected three stations for this experiment; San Francisco (SFO) because of its pronounced\nannual cycle, Buffalo because it is wet throughout the year, and Yuma because it is very dry.\nFor each month and station, we derived 3-term regression equations for each of our four\npredictands. The specification equations for SFO in December for System C were:\nPAO = .0158 + .0066H(45N,125W) - .0030H(50N,120W) - .0018H(25N,125W) and\nPA2 = .0166 - .0068H(45N,125W) + .0030H(50N,120W) + .0017H(25N,125W)\nwhere the latitude and longitude of the predictive heights are given in parentheses, and the\nreduction of variance (RV) was 47.4% for PAO and 52.7% for PA2. Note that the two\nindependently derived equations tend to be mirror images, with the same (or adjacent)\ngridpoints selected but with coefficients of opposite sign. The second and last terms indicate\nthat dry (wet) weather is favored by anticyclonic (cyclonic) easterly (westerly) flow of\ncontinental (maritime) air around a high (low) near the coast of Oregon.\nThe specification equations were applied to the dependent sample to give computed\nprobabilities of 0, 1, and 2+ wet days for each of systems A, B, and C. The ranked probability\nscore (RPS) was used to evaluate the probability specifications. The RPS is given by\nRPS\nwhere p = (po, P1, ....Pk-1) and d = (do, d1, ....dk-1) are the forecast probabilities and the\nobservations in an ordered sequence. dn equals 1 if class n occurs and 0 otherwise. k is the\nnumber of classes (here 3). The RPS is a quadratic scoring rule that is sensitive to the distance\n351","between the forecast and observed categories. It has a range between 0 (for a forecast that\nassigns 100% to one of the extreme categories when the other extreme occurs) and 1 (for a\nforecast that assigns 100% probability to whichever category is observed). Thus higher scores\ncorrespond to more accurate forecasts.\nValues of the RPS obtained by hindcasting for SFO separately by systems A, B, and C\nare plotted in fig. 3. Results for the three systems are nearly identical, and they are superior to\nthe RPS yielded by climatology (zero anomalies). All the curves exhibit a strong annual cycle,\nwith a sharp maximum in summer when rain is a rare event and precipitation is very hard to\nspecify (RV=6.5% for PA2 in July). As a result, the improvement of regression over\nclimatology is very small in summer but much larger in winter. An opposite type of annual\nRPS cycle is exhibited by Buffalo (fig. 4); all systems have highest scores in winter and\nminima during summer. Here dry periods rarely occur in winter (CP0=0.01 in January) when\nthe RV for specification is very low (3.9% for PAO in February). Consequently, the difference\nin RPS between forecasts made by regression and by climatology is greater in summer than in\nwinter. In all months, however, the curves for systems A, B, and C are nearly identical.\nFigures 3 and 4 show that the difference in RPS between systems A, B, and C is extremely\nsmall at SFO and Buffalo. Since the same finding also applied to Yuma (not shown), we\nselected for operational use the system (C) which is based on the two unconditional\npredictands. This system emphasizes the two extreme categories (0 and 2+) and is easiest to\nunderstand.\nFigure 5 shows that, in terms of RPS, it is difficult for a forecast to beat climatology.\nIn fact, the maximum possible improvement is only 0.25. We therefore decided to evaluate\nthe forecasts in terms of the maximum possible improvement over climatology (1-C), rather\nthan just C itself. We did this by converting the RPS to an RPS-based skill score (analogous\nto the Heidke skill score) defined as: RS = 100 (F-C)/(1-C), where C and F are RPS scores\nyielded by climatology and by any other forecast system. Figure 6 shows the annual cycle of\nRS for hindcasts made by system C at SFO. Our regression specifications now have maximum\nRS values in the winter and minima in summer, opposite to the pattern plotted in fig. 3.\nHowever, the annual cycle of RS is similar to that for RV, not only for SFO but also for other\nstations. The geographical distribution of RS scores is plotted in fig. 7 for January and fig. 8\nfor July. In Jan., values range from 50 at SFO to 4 at Burlington, VT. In July, the highest\nscore is 34 at Portland, OR, and the lowest is 6 at Los Angeles. Scores are generally lower in\nJuly because of its smaller scale of precipitation.\nThe specification equations have not yet been used extensively in a predictive mode.\nNaturally, we expect large loss of skill when they are applied to 6-10 mean prognostic 700 mb\nheights. Every effort will be made to remove any systematic bias from the heights before\nentering them into the regression equations, and the forecast height anomalies will be adjusted\n(deflated) to account for the loss of predictability at this extended range. We also plan to\ncombine our statistical equations with precipitation forecasts obtained directly from the\nnumerical model. Details of these efforts will be reported at a later date.\nACKNOWLEDGMENTS\nThis research was funded in part by the Climate Analysis Center, National Weather\nService, NOAA, under grant no. ATM-84AA-H00026 to the Cooperative Institute for Climate\nStudies, Department of Meteorology, University of Maryland. We thank Don Gilman,\nTom\nPerrone, Fran Hughes and Bruce Whistler for valuable assistance in this project.\n352","siw\nIN\nto\n80\nTO\n65\n60\nU.S. DEPARTMENT OF COMMERCE\nWEATHER BURCAU\nTRUE KCALE $ 32,803,000 AT\nPOLAR STEREOCAUPHIC PROJECTION\n40\n.\n35\n30\n0\n25\n25\n0.\n115\n110\n105\n100\n95\n90\nas\n80 USCOMM-WB-DC\n75\nLocations of the 146 stations used in this study.\nFig. 1.\n00\nO.\n70\n,\nS\nG\nit\n&\n&\n24\nGO\n50\n9.\nE\n$\nD\n&\nS\n08\nin\n8\nY\nR\n50\n2\nEO\n70\nx\n80\na\n6\n1)\nx\na\n%\nG\nR\n3\nF\n.\n3\n:\n30\n0.\n(2\nM\n2\nof\nFINAL 79 POINT GRID\nD\nwi\n20\no\n00\n0\n700 MB. HEIGHT.\n3907\non\nNetwork of 79 gridpoints used to delineate the field of 700 mb heights.\nFig. 2.\n353","COMPARISON OF RANKED PROBABILITY SCORE\nFOR SAN FPANCISCO\n0.97\n0.96\n0.95\n0.94\n0.93\n0.92\n0.91\n0.9\n0.89\n0.83\n0.87\n0.86\n0.85\n0.84\n0.83\n0.82\n0.81\n0.8\n0.79\n0.78\n0.77\n0.76\nJan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nev. Doc.\nc\n\"A\"\n\"B\"\n+\n\"0\"\nCilmo.\nC\n-\nFig. 3.\nAnnual cycle of ranked probability scores for probability specifications of\nprecipitation frequency at SFO by climatology and by three alternate regression\nsystems.\nCOMPARISON OF RANKED PROBABILITY SCORE\nFOR BUFFALO\n0.96\n0,95\n0.94\n0.93\n0.92\n0.91\n0.9\n0.89\n0.88\n0.87\n0.86\n0.85\n0.84\n0.83\n0.82\n0.81\n0.8\nJan. Feb. Mar. Apr. May Jun. Jul. Aug. Sep. Oct. Nev. Doc.\n0\n\"A\"\n\"B\"\n\"C\"\n+\nCilmo.\na\nFig. 4.\nSame as Fig. 3, but for Buffalo.\n354",":\n.98\n90\n96\no\n94\n57\n0.92\na\n0.90\nee\n0.96\n0.80\n0 Bill\n0.70\n90\n92\n76\n90\n=\n98\n00\n0.10\n0.80\nC.\n1.00\n0.00\n0.10\n0.20\n0.30\n0.40\n0.50\n0.00\nPROBABICITY OF 0 BAINY DAYS IN\n5\nGraph giving the expected ranked probability score (dashed curves) for forecasts\nFig. 5.\nof climatology as a function of the climatological probability of 0 and 1 rainy day\nin 5.\n100\n90\n80\n70\n60\nN°\nR\n50|-R\nR\nR\nR\nR\n40\nR\nR\n30\nR\nR\n20\nR\nR\n10\n0\n1\n2\n3\n4\n5\n6\n7\n8\n9\n10\n11\n12\nMONTH\nFig. 6.\nAnnual cycle of RPS skill scores produced by regression system C at SFO.\n355","-\na\n65\nw\nU.S. DEPARTMENT or COMMERCE\nWEATHER BURCAU\nTOUR SCALL 123.101.000 AT us wa\nPOLIC PROCECTION\n10\n30\n20\nH\n0\n10\n5\n-15\n10\n40\n15\n50\nH\nH2\n10\n.35\n35\n25\n20\n35,\nH\nH\n40\nH\n50\nso\n30\n25\n20\n20\n25\n15\n15\nHI\nRPS SKILL SCORE\nJANUARY\n15\nGeographical distribution of RPS skill scores produced by operational\nFig. 7.\nspecification equations in dependent sample in January. Isopleths are analyzed\nevery 10% with intermediate lines dashed.\n10\n\"\n110\n100\nUS DEPARTMENT or COMMERCE\nWEATHER BURCAU\n15\nTRUE SCALL 122 100 000 AT LAI\n30\nPOLAR STERIOGRAPHE PROCECTION\nH\n33\n10\n20\n15\n30\nH\n20\n10\n15\n10\nM\nH\n15\n10\n15\nH\n10\n15\nD\nH\nRPS SKILL SCORE\n10\nJULY\n0.\na USCOMM.WB-\n75\nFig. 8.\nSame as fig. 7 but for July.\n356","STRUCTURE OF RECENT SURFACE AIR TEMPERATURE (SAT) TRENDS.\nG.V.Gruza, E.Ya.Ran'kova, E.V.Rocheva\nHydrometeorological Center of the U.S.S.R.\nThe estimates of recent climate changes based on historical records\nhave been given in a number of papers (Budyko, Vinnikov, 1976; Gruza,\nRan'kova, I980; Jones, Raper, Wigley, I986; Jones, Wigley, Kelly et al,\nI986; Vinnikov, I986; Hansen, Lebedeff, 1987).\nThese estimates concerned regionally averaged air temperature (global,\nhemispherical, for latitudinal zones etc.). They had been produced\nusing different data sets and processing methods (Hansen, Lebedeff,\nI987; Jones et al, 1986; Gruza, Ran'kova, 1980).\nThese series agree with each other well. Their main features (for the\nNorthern Hemisphere) are reflected in fig.I, plotted on the U.S.S.R.\ndata. Hemispherical and zonal air temperature anomalies are presented\nfor the annual and seasonal (winter, summer) means.\nIt is apparent from the figure that:\n- in I940's warming was replaced by cooling which lasted till 1960's.\nWarming periods in I940's and I980's were the most significant ones.\nBrief cooling in 1985 and I986 was once more replaced by warming;\n- Northern Hemisphere air temperature in 1987 and I988 was 0.2 and\n0.37cc, respectively, over the norm. Anomaly value of I988 was the\ngreatest for the period I89I-I988.\n- Latitude versus time section of the anomalies of temperature averaged\nover the latitudinal circles for winter (fig.Ia) and summer (fig.Ib)\nseasons showed, that winter temperatures, especially in high latitudes,\nwere the best illustration of the changes: many a warmest winter\ncoincided with the global warming, the coldest winters occurred during\nthe cooling period.\nSecular course of the zonal temperature anomalies can easily be followed.\nWarming in the middle of the century is distinct from that in the end:\nin the middle of the century it showed in the 35-80°N and 5-10°1 N zones,\nin the end - in all latitudes.\nThe last 3 years are analysed in more detail (with a I-month step) in\nfig.2. In the left of the figure are the norms for I95I-I980; in the\nright - zonal temperature values as deviations from the norms. During\npractically the whole of I986 and in almost all latitudes negative ano-\nmalies, and in I988 - the positive ones, prevailed. In 1987, though, po-\nsitive anomalies steadily kept only to the south of 35°N.\nFig.3 presents annual mean air temperature anomaly fields in the Northern\nHemisphere grid points for the last 3 years.\nAreas of anomalies exceeding I°C in absolute value are dashed.\nIn I986 negative anomalies prevailed over the oceans (and were most\n357","significant over Atlantic) and over Eurasia positive anomalies were ob-\nserved.\nIn 1987 negative anomalies were still there over the Atlantic ocean and\nthe Arctic Sea; over the continents positive anomalies became stronger.\nWe should note the increase of the positive anomalies' area, especially\nin lower latitudes and the reduction of the area occupied by negative\nanomalies, observed for the greater part in high latitudes.\nLatitude versus time sections for the 85-5 o N latitudinal belt (averaging\nis done along the meridional section for all of the Northern Hemisphere)\nare presented in fig. 4. Such presentation of data enables us to follow\nthe main features of the temperature regime over the continents and the\noceans for the considered period (continent-ocean boundaries are conven-\ntionally chosen along the IO°W, 140°E, 120°W, 60° W meridians).\nIt attracts one's attention that:\n- Positive anomalies prevail (in 1987 and, especially in 1988) practically\nover the whole Hemisphere;\n- Over the oceanic sectors anomaly sign was the same for a longer period\nof time than over the continents. So, over the Eastern Pacific positive\nanomalies lasted for nearly 2 years;\n- Anomalies greater than 2°c were observed over the continents and in\nwinter mostly.\nUntil now we have been analysing specific features of the last 2 years\ntemperature regime.\nBut it seems of special importance to estimate them in comparison with\nthe secular global temperature series.\nThe results of the detailed analysis of the structure of secular fluctua-\ntions are given in fig.5. Coefficients of linear trend A are presented\nhere as function of the length of the trend estimation period L (vertical\naxis) and the ending-year of the period T (horizontal axis). The value in\nthe given point (e.g., in the point with L=20, T=I940) is the linear trend\ncoefficient, estimated for the L-year period (in our case, 20-year),\nending with the year T (in our case,1940), that is, for I92I-I940. On the\nwhole, the figure gives estimates for all moving (with a I-year step)\nperiods from 5 to 60 years. It was but natural to expect, with small L's\nfrequent change of the sign is characteristic of all the series under\nconsideration. With L=30, during I89I-I988 global cooling in the Northern\nHemisphere was twice replaced by warming.\nMain temperature trends in the Northern Hemisphere and the globe were\nrather similar.\nWhat goes for comparison of the Northern and the Southern Hemispheres, is\nthat\ncooling, which began in I940's, in the Northern Hemisphere kept for a\nlonger period of time than in the Southern, with L 20.\nUsing the discussed above approach, we can obtain new data on the structure\n358","of long-term trends of global air temperature; it gives a vivid illustra-\ntion of the complex nature of air temperature changes in this century and\nthe vulnerability of rigid assessments of the tendency of global climate\nchanges without clear indication of the considered time scales and periods.\nThe pattern of the trends is extended with estimates of cross-correlations\nof air temperature values for different regions, latitudinal belts and\nzones.\nCorrelation coefficients between regionally averaged air temperatures for\nthe Northern Hemisphere are listed in fig. 6. In January the hemispheric\ntemperature correlates most significantly with that of Eurasia (0.68)\nand the North America; in June - of the North America (0.75), Eurasia\n(0.62) and the Pacific Sector (0.59).\nIt appears from fig.6 that hemispheric temperature has stronger relation-\nship with moderate latitude temperature (50-65 N), especially in cold seasons\nmainly because of the continental regions. Correlation with the air tem-\nperature over the oceans was less significant (fig.7).\nRelationship between regionally averaged temperatures seems very weak (fig.\n8).\nThe above-mentioned estimates enable us to choose the \"active\" regions,\nmaking most significant contributions to the changes of temperature over\nthe U.S.S.R. and the Northern Hemisphere on the whole.\nREFERENCES.\nI. Budyko M.I., Vinnikov K.Ya. I976: Global warming. - Meteorol.and Hydrol.,\nNo.7, p.p. I6-26.\n2. Vinnikov K.Ya. I986: Climate sensitivity. - Leningrad, Hydrometeoizdat,\n223 p.\n3. Gruza G.V., Ran'kova E.Ya. I980: Structure and variability of observed\nclimate. - Leningrad, Hydrometeoizdat, 72 p.\n4. Hansen J. and Lebedeff S. I987: Global trends of measured surface air\ntemperature. - Journal of Geophysical Research, vol.92, No.DII,\np.I3.345-I3.372.\n5. Jones P.D. et al. 1986: Northern Hemisphere surface air temperature\nvariations: I85I-I984 /P.D.Jones, S.C.B.Raper, R.S.Bradley, H.F.Diaz,\nP.M.Kelly and T.M.L. Wigley/. - J.Clim.Appl.Meteorol., vol.25, p.I6I-\nI79.\n6. Jones P.D. et al., I986: Southern Hemisphere air temperature variations:\nI85I-I984. /P.D.Jones, S.C.B.Raper, T.M.L.Wigley/. - J.Clim.Appl.\nMeteorol., vol.25, p. I2I3-I230.\n359","°C\nannual, N.H.\n0.5\na)\n0,0\nb)\n0,5\n0,0\n-0.5\nwinter\n-1.0\n80\n60 40 DROP and\n20\n0\n0,5\nd)\n0,0\nsummer\n0.5\nU:U\nYUU\n80\n60\nN\n40\n20\n0\nsail\n1891\n1900\n1910\n1920\n1930\n1940\n1950\n1960\n1970\n1980 1988\nFig. I. Time series of annual mean (December-November, the upper\npart of fig . Ia), winter December-February fig.Ib) and summer (June-\nAugust, fig.Ic) surface air temperature anomalies for the Northern\nHemisphere (85-5°N) and latitude versus time diagram of the temperature\nQ\nanomalies averaged over the latitudinal circles for winter (fig.Ic)\nand summer (fig.Ie) seasons. Base period: I95I-I980.\nE&VT\nNo\nan\nso\n40\n20\n0\n,\n,\nw\nII\nv\nVM\nas\nKi\n/\n,\nMI\nv\nUN\nIX\nKI\n,\n⑉\nv\nVII\nIX\nXI\n⑉\nv\nVN\nIX\nXI\n1987\n1988\n1989\n1986\n=\n2\n2-0;\n8:8\n0-2;\nIIII\n>2\nFig.2. Latitude versus time diagram of monthly mean air temperature\nover the Northern Hemisphere (85-5 o N) . Left: sea level normals (I95I-\n1980). Right: anomalies relative to the I95I-I980 period for June I986 -\nMarch 1988.\n360","60 E\n60 W\no\n120 E\n240 W\n180 E\nN\nN\n1986\n90\n90\nI986\n80\n80\nTO\n70\n-\n60\n60\n50\n50\n40\n40\n30\n30\n20\n20\n10\n10\no\no\n60 W\n60 E\n120 E\n180 E\n240 W\no\n60 W\n60 E\n240 W\no\n120 E\n180 E\nN\nN\n90\n1987\n90\n0\n80\n1987\n80\nTO\n70\n60\n60\n50\n50\n40\n40\n30\n30\n20\n20\n+\n10\n10\no\no\n60 W\n60 E\n120 E\n180 E\no\n240 W\n60 W\n60 E\no\n120 E\n180 E\n240 W\nN\nN\n90\nI988\n90\n80\n80\nI988\nG\nTO\n70\n60\n60\n50\n50\nas\n40\n40\n<- 1\n30\n30\n1-0\n0-1\n20\n20\n>1\n10\n10\no\no\n60 W\n60 E\n120 E\n180 E\no\n240 W\nFig 3.\nAnnual mean (December-November) air temperature\nanomalies over the Northern Hemisphere for I986, I987 and\n1988. Base period: I95I-I980.\n361","Eevt,c\nT.\nUSSR\n60\nNH\nLongitude\ndate\n180°E\n240°E\n50'E\n120°E\n60'\n50\n1986,VI\n40\nVII\nVIII\nso\nX\nx\n20\nP.\nXI\nXII\n10\n1987\n11\nill\nTNH\nJWK\n60\nNH,\nIV\nv\n50\nVII\n40\nVIN\n30\nIX\nx\n20\nFARIDATE\nXI\nXII\n10\n1988,\nIII\nTSH, SH, JWK\n60\nIV\nv\n50\nVI\n40\nVII\nVIII\n30\nIX\nx\n20\nO\nXI\nXII\n10\n1989\ncare\nnB\nIII\n60 TGL. JWK\nPO\nNA\nAO\nEA\n2-0\n6-2\n0-2\nlatte\n>2\nso\n40\nFig.4. Time-longitude\n30\nsection of monthly mean\n20\nsurface air temperature\nPARAB\n10\nanomalies, Northern He-\n1891\n1900\n1910\n1920\n1930\n1940\n1950\n1960\n1970\n1980\nmisphere (85-5° N), 1986,\nending-year\n405\n0.5-0\n0-05\nHIB\n0.5\nJune - 1988, March. Base\nFig.5. Regionally averaged surfa-\nperiod: I95I-I980.\nce air temperature (SAT) linear trend\ncoefficients (C/IOyears) as function\nof the period length (vertical axis)\nand ending-year (horizontal axis).\n362","latitude\nnl\nVIII\nIX\nX\nXI\nXII\nI\nIf\nIII\nIV\nV\nVI\nN.\nX\n30\n30\n30\n10\nas\n30\n50\n79\n60\n50\n65\n70\n70\n60\n55\n70\no\n50\n50\n35\n60\nC\n70\n25\n30\n50\n10\nis\n30\n30\n30\nFig. 6. Correlation coefficients between air temperature means for the\nNorthern Hemisphere and the temperature of separate latitudinal circles.\nR\n100\n90\n60\nEA\n70\nNA\n60\n/\nx\n40\nPO\n30\nAO\n20\nApp\n10\n0\n-W\nI\n2/\nIII\nVII\nVIII\n1X\nIV\nV\nVI\n,\nH\nmonths\nFig.7. Seasonal course of correlation between the mean hemispheric tem-\nperature and that of different regions. I - Eurasia, 2 - North America, 3 -\nAfrica, 4 - the Pacific Ocean, 5 - the North Atlantic.\nb)\nPO\nPO\na)\n08\n23\n-39\n24\nI4\nOI\nOI\nI8\n45\nI5\nEA\nNA\nEA\nNA\n-02\n08\n05\n-OI\n05\n28\n08\n25\nAO\nAf\nAL\nAO\n-26\n-03\nFig.8. Correlation between the regions in January (fig.8a) and July\n(fig.8b): EA - Eurasia, NA - North America, PO - the Pacific Ocean,\nAO - the Atlantic Ocean, Af - Africa.\n363","EXPERIENCE ON PROBABILISTIC LONG-RANGE FORECASTING IN THE U.S.S.R.\nG.V.Gruza, E.Ya.Ran'kova\nHydrometeorological Center of the USSR, Moscow\nA long-range forecasting scheme, known as \"a group of analogues\", is\nused in the U.S.S.R. since 1975. We call the method, realized in this scheme\na \"dynamic climatology method\", implying conditional climatology, which\ndescribes the expected perspective of the current (prognosed) process.\nThe materials, concerning all the stages of the scheme's development\nand use, were published in the U.S.S.R. (beginning with 1976) and presented\nat international symposiums (Gruza, Ran'kova, 1980; Gruza, I985; Gruza,\nRan'kova, 1986) and in monograph by Gruza, Ran'kova, 1983.\nThe idea of \"group analogs\" was used by Murphy, 1987; Livezey, I987.\nIn this report, without describing the scheme and method in detail, we\nwill give some results, generalizing the experience of the model's use since\n1984.\nWe will present 2 versions used in the U.S.S.R. operative practice,\ngiving the Northern Hemisphere surface air temperature (SAT) forecasts:\n- GRAN-84, corresponds to Gruza, Ran'kova, 1980;\n- GRAN-87, corresponds to Gruza, Ran'kova, 1986.\nThe forecast is given in the probabilistic form for 3 equiprobable\ncategories. On this basis categorical forecast is produced by means of 4\nmethods: preevaluated distribution expectation and median, maximum probabi-\nlity and utility category medians. The matrix\nI.O\n0.50\no\nwhich\n0.25\nI.O\n0.25\no\n0.50\nI.O\ndoes not encourage middle category (norm) forecasting was used as utility\nmatrix.\nThe examples of probabilistic forecast are given in figures I and 2 for\nthe Northern Hemisphere and the U.S.S.R. territory, correspondingly.\nThe difference between the GRAN-87 and GRAN-84 schemes lies in the\nfollowing:\nI. A priori estimates of expected analogue skill and optimization of\npredictors' subspaces to be used are introduced: out of all possible predic-\ntors the scheme automatically forms the optimal \"composite factor\", maximi-\nzing analogue sample compactness for the given (prognosed) case. That is, for\neach forecast their own predictors are used.\n2. Forecast is not based on a single analogue group for all the territory\n(hemisphere), but is actually a compilation of independent forecasts for sub-\nregions.\nWe should make our subregions small enough and take into account the\nphysical and geographical features and statistical structure of the predictant.\nPredictor' optimization and selection of analogues are made individually for\neach one of such subregions.\nAt present we were forced to restrict our investigation to optimization\nof only 3 sectors (fig.3) because the computer capacities availiable were\nunsufficient.\n364","3. Instead of one smoothed SAT field, potential predictors' set\nembraces more than a hundred SAT- 9 SLP-, and SST-field presentations\n(differing in the spacial segmentation and seasonal course elimination\nmethods and metrics for estimation of their analogousness).\nGRAN-87 version was tested monthly for 2 years (1986--998 ix ), whereas\nGRAN-84 - since December of 1983.\nAll estimates (fig.4-7) are given for SAT forecasts with time over-\nhead of I month (using the data for October-December, etc.), though the\ntests have also been carried out for 2-months seasons and for precipita-\ntion as well.\nIt should be noted in fig.4, that, on the whole, performance of pro-\nbabilistic and categorical forecasts in both versions is above the climatic\none ( APSce =0.333, Pce =0).\nComplicating the scheme was effective when \"predictability\" was above\nclimatic; when forecasts are bad, they are bad in all versions.\nTwo points are of interest in figures 5 and 6:\n- GRAN-87 forecasts present probabilities for all intervals, reaching\nsometimes 90-I00% (there were none of the kind in GRAN-84 forecasts; and\nin Outlook a probability over 60% simply cannot be given - none exceeding\n50% ever was).\n- extreme categories (above norm, below norm) have better forecast\nperformance than middle ones (norm).\nFig.7 compares features of decision-making, using the 4 above-mention-\nned criteria.\nForecast optimization procedure lowered frequency overestimates of\nclass\" categories, especially in decisions on mathematical expecta-\n\"norm\nYM\nDecision-making on the maximum expected utili-\ntion\nand median\nbasis is seen as most effective.\nty\nCONCLUSIONS\nI. Both versions of \"group analog\" scheme showed positive skill over\nprobabilistic climate forecast.\n2. The latest GRAN-87 version has marked advantages over the previous\none. The most important is it's ability to automatically adapt predictors'\nsubspace in every single case, taking into account current meteorological\ndata.\n3. GRAN-87 version makes use of only a part of the advanced \"group\nanalog\" model capabilities, and has considerable potential for further\nimprovement:\n- optimization is done for unjustifyably small number of subregions\n(II out of 45), and not for separate subregions, but within each of the\n3 large sectors on the average (fig.3).\n365","- proper version of the scheme is required for forecasting on the\nU.S.S.R. stations, with it's own (independent) optimization for each sta-\ntion. Some stations can be considered as a single unit, but this question\nshould be considered for each particular case;\n- We should include in predictor set daily meteorological station data\nand series of the geopotential fields' (H 1000! H 500' 700' HIOO' etc) and\ntheir indexes, and among them the values, derived from hydrodynamical models.\n4. In general, the recent version of \"group analogs\" scheme is especial-\nly interesting in that it allows using operative data, to carry out detailed\nanalysis of current (prognosed) process, repeating in this particular case\nthe procedure of forecast development with the given time overhead.\nREFERENCES\nI. Gruza G.V. Ran'kova E. Ya., 1980: Long-range forecasting using a group of\nanalogs and evaluation of meteorological predictability. @ -WMO Symposium on\nProbability and Statistical Methods in Weather Forecasting (Nice, 8-I2 Sept.,\n1980), p.269-276.\n2. Gruza G.V., Ran 'kova E.Ya.,I983: Probability meteorological forecasts. Le-\nningrad,Hydrometeoizdat, 20 a.l.\n3. Gruza G.V.,I985: Monitoring and probability forecast of monthly and seasonal\nvariations in hemisphere atmospheric processes.-Long-range Forecasting Research\nSeries, No. 6, vol.II,WMO/TD.No87,p.682-688.\n4. Gruza G.V. Ran'kova E.Ya., I986:Experience on objective probability long-\nrange forecasting of air temperature in the Northern Hemisphere.-Long-rang\nForecasting Research Report Series, No 18,WMO/TD.No.I47, p.3I3-320.\n5. Kruizinga S. Murphy A.H,, 1983: Use of an analogue procedure to formulate\nobjective probabilistic temperature forecasts in the Netherlands.-Mon.Wea.Rev..\nvol. . III, II,P. 2244-2254.\n6. Livezey R.E., Barnston A.G.,I988: An operational multiplied analog/antianalog\nprediction system for United States seasonal temperatures.I, System design and\nwinter experiments.-J.Geophys.Res.,vol.93,No.D9,p.I0.953-I0.9758\nExpected probabilities of monthly\ntotal precipitation anomalies\nin January 1987\nIssue\nKORECK\nY\nCaseraph\nKOTCH\nwas\nCarpanack\nTORAFC\nTyaras\nPa\n0.333 (\"above norm class\nprobability)\nAV\nPy\n0.333 (\"below norm\" class\nprobability)\n?B\n0.333 (\"near norm\" class\nSalipares\nprobability)\nPb\n0.333(uncertainty)\n(climate forecast)\nFig.2.\n366","Mc.\n10-8-42\nIf\nNo\nall\nIII\n10-08-62\n22-14-66\n23-38-39\nin\na\n09-38-53\nml\nlia\nXIII\nus\n33-30-57\n23-07-24\nh\n12-41-27\nan\nDE\nil\nun\n27-32-\n43-33-24\nMy\nU.\nIis\nhl\nNit\nin\nlb\n11a\n30\n50-33-\n40-35-25\n-18\n38-19-\nU\n3-48-19\nH\n1-19-32\nan\nur\n28-35-37/1\nin\nat\n32-36-32\nIN\nEL\n39-22-39\n30-40-30\nso\n39-31-30\n27-42-31\n43\n36-21\n47-07-06\n,\nor\nLegood\n5C\n24-28-28\nin\n2\nthe\nalla 0.323\nFig.3.GRAN-87 experimental forecast\nregions and their subregions (dashed\nFig. 1. Mean monthly air temperature\nareas) used for predictors set\nprobability forecast over the Northern\noptimization.\nHemisphere, October I989.\nAPS\n0,5\n0/4\n03\nS\n0,4\nAway\n02\n0,0\n-02\n-04\n1988\n1987\n1989\n1990\n1984\n1985\n1986\n:\nFig.4. Skill score estimates of operational (GRAN-84) and\nexperimental (GRAN - -87) probability forecasts for mean\nmonthly air temperature over the Northern Hemisphere as\ncompared with climatic forecasts.\n367","A\nAssociation of predicted probabilities Poredwith observed frequencies\nFig. 5.\nfobs by the results of GRAN-87M and GRAN-87S testing\nSAT, Northern Hemisphere; I986, X - 1988, IX\nForecast for the month\nForecast for the\nseason\nGRAN-84\nGRAN-87M\nGRAN-87S\nfobs\n34\nn = 3216\nn = 3225\nn = 3237\n80\n28\n54\n57\n162\n60\n246\n48\n68\n283\n86\n262\n40\n650\n683\n581\n10\n152\n486\n362\n458\n492\n63\n640\n672\n1015\n606\n496\n20\n360\n414\no\nIOO\n80\n60\n40\n20\no\nGRAN-87M for T '40' U.S.S.R.\nOutlook, U.S.A.\n(see Gilman, 1986)\nI00\nn = 2805\nn = 3600\n80\n4\n60\n55\n23\n389\n600\n261\n40\n115\n588\n506\n20\n264\n0\n20\n40\n60\nIOO\n0\n80\n20\n40\n60\n80\nPpred\n0\n20\n40\n60\n80\nPpred\nwarm\nnorm\n- cold\n368","GRAN-84\nGRAN-87\nfobs.\nn=3225\nn=3216\n80\n54 28\n246\n60\n63\n283\n48\n10\n.\n40\n650\n152\n883\n581\n492\n362\n63\n640\n1015\n496\n20\n366\n0\nf(P)\n40\n30\n20\n10\nPpred\n0\n80\n100\n20\n40\n60\n40\n80\n100\n0\n20\n60\nFig.6. Skill scores of the monthly probabilistic forecast, SAT,\nNorthern Hemisphere, 1986-1988 IX Top: Correspondence of predicted\nprobabilities p pred to observed frequences. Bottom: Distribution\nof probabilities f(P) over the IO%-intervals.\nP%\n100\n90\nGRAN-87\n^pred'\n80\nGRAN-84\npred'\n70\nobs\n60\n50\n40\n30\n20\n10\n3\nYou\n3\nPM\n2\n3\nYp\nI\n2\n1\n2\n3\n1\nYE\n1\n2\nFig.7. Distribution of observations F(Y) and categorical\nforecasts F(Y) in a time series, 1986 -1988\nIX\n369","The Distribution of the Correlation between Large Scale Averages and\nComponent Members\nStanley L. Grotch\nLawrence Livermore National Laboratory\nIntroduction\nAlthough it should be self-evident from the wording alone, the \"global\naverage temperature anomaly\" is, in fact, the average of many individual grid\npoint values, that fact is easily overlooked by both the general public and by\npolicy makers. As a result, there is a temptation to use the average curve as a\nsurrogate for the temperature behavior of let us say; western Europe, the\ncentral United States, the state of California, or Washington, D.C.\nWhen presenting data graphically before smoothing or averaging, the\noriginal data values should also be shown. In Fig. 1 (right panel) the area\nweighted average NH annually averaged temperature anomaly curve of\nJones et. al. (CRU) is presented. In the left panel of Fig. 1, using the same\nscaling, are presented the individual grid point values from which these\naverages were estimated. The enormous scatter of these data relative to the\nchanges seen over the century is obvious. These plots should immediately\nalert us to the potential danger of using the average curve as \"representative\"\nof any individual gridpoint values.\nThe present study addresses the question: \"How representative is the\naverage of a number of time series of any of the individual components?\"\nMore specifically, what is the resultant distribution of the correlation\ncoefficient for a set of K time series when each individual member is\ncorrelated with the average of the K series?\nDATA USED\nThe CRU data for the NH of monthly temperature anomalies over the\ncentury 1887-1986 were used in this study. To avoid problems with missing\ndata, only those 134 gridpoints in the NH were used which contained at least\n90% of the monthly data during this 100 year period. When cross correlated\nwith their average, the average cross correlation between any gridpoint and\nthe composite average of the 134 is small, 0.34. Thus, relatively little\npredictive information is gained beyond using the climatological average at a\ngridpoint by incorporating information derived from the larger scale average\n370","behavior. The distribution of these correlation coefficients is displayed in Fig.\nIs it possible to predict the mean and the variance of this distribution in\n2.\nterms of a simple model?\nA Simple Model for Describing the Cross correlation coefficients\nAssume that we have K time series: yk(t), k=1,2,3,,,,..K which are\narithmetically weighted to yield a composite series: y (t). . Assume that for k\n= 1,2,3,,,.K.\n(1)\nWhere N(0, o2x) is a normal distribution of mean = 0 and variance =\nAssume that oxis a constant independent of both: (1) the individual\ngridpoint k, and (2) time. Monte Carlo simulations were performed using as\nthe underlying series y (t), the average of 134 CRU NH annual temperature\n2\nseries for 1887-1986 and a Gaussian random number generator varying 'ox\nand generating sets of 134 composite series using Eqn 1. From these\nnumerical experiments it was found that the mean cross correlation, u(p),\nwas accurately predicted as:\n-1\n2\nExpected value (p)\n(2)\nu(p)\n= =\n2\nIn Eqn 2 the variance of the base average series is o Eqn. 2 can be\nderived analytically if one assumes that: (1) the mean value of each\ncomponent series is zero, and (2) the noise added to the base series is\nindependently correlated with each series. The simple noise model permits\none to accurately predict the mean of the component cross correlations with\ntheir average. However, the standard deviation of this distribution from\nthese simulated experiments is about 2.4 times smaller than that observed.\n371","MORE COMPLEX MODEL\nAssume now that the noise variance is different for each series k, but\nstill is independent of time.\n(3)\nA second set if Monte Carlo experiments was performed in which the\nnoise added to each series k had the same variance as that found for the 134\nmembers:\n100\n(4)\nfor k = 1,2,..,134\nThese simulations again accurately predict the mean of the actual cross\ncorrelations observed, but the standard deviation is still too low, but now by\nabout a smaller factor of 1.6 Fig. 2 displays the three correlation distributions\nfor: (1) the 134 series, (2) simulated noise model of Eqn 1 and, (3) simulated\nnoise model of Eqn. 3.\nAnother result which may at first seem surprising is important to\nnote, particularly with regard to regional correlations. The composite average\nseries for the 29 North American gridpoints yields a cross correlation with\nthe CRU NH average of 0.75. One might intuitively suspect that the 29\ncomponent gridpoints comprising the average would show correlations\nwhich would scatter about this average correlation, some above and some\nbelow 0.75. In fact, however, every individual gridpoint in the region has a\ncorrelation with the hemispheric average which is lower than does their\naverage. This is also found to be the case for most regional areas (with one\nexception at lower latitudes). The implications of this result are important for\nregional scale impact studies.\n372","©\n9\nO\no\no\no\nO\no\nD\nB\nO\nO\nI\n1840\n1860\n1880\n1900\n1920\n1940\n1960\n1980\n2000\n1840\n1860\n1880\n1900\n1920\n1940\n1960\n1980\n2000\nYEAR\nYEAR\nFig. 1 CRU NH annually averaged temperature anomaly for 1851-1986. In\nthe right panel the area-weighted average curve is given and in the left panel\nthe individual gridpoint anomalies are shown with the same scaling.\n373","18\n18\n20\n18\n16\n16\n16\n14\n14\n14\n12\n12\n12\n10\n10\n10\n8\n8\n8\n6\n6\n6\n4\n4\n4\n2\n2\n2\n0\n0\n0\n.8\n-.4\n-.2\n0\n.2\n.4\n.6\n1\n2 .25 .3 35 .4 45 .5 .55 .6\n.4\n.6\n.8\n1\n-.4\n--2\n0\n.2\nConstant noise variance\nVariable noise variance\nActual 134 series\nSdev(corrin)= 0.09\nSdev(corrin) = 0.13\nSdev(corrin) = 0.19\nFig. 2 Three distributions of the correlation coefficient for 134 individual\nannual temperature anomaly series when correlated with their composite\naverage over the century 1887-1986: (1) actually observed, (2) simulated using\nnoise model of Eqn 1 and , (3) simulated using noise model of Eqn 3.\n374","Medium- and Long-Range Forecasting\nat the National Meteorological Center*\nA. James Wagner\nPrediction Branch, CAC\nNMC, NWS, NOAA\nWashington, DC 20233\nAt the medium- and long-range time periods for which forecasts are made\nwithin the Climate Analysis Center (CAC), a mix of dynamical tools based on\nnumerical weather prediction (NWP) models and statistical tools based on past\nclimatological records is used. As is generally well known, as one progresses\nfrom short-range forecasts to seasonal outlooks, the methodology employed\nchanges from almost exclusive reliance on NWP products with a minimum of sta-\ntistical modification at the shortest ranges to an array of entirely statisti-\ncal tools for seasonal time intervals (Fig. 1). The accuracy and reliability\nof both the tools and the final forecast products decline steadily from the\nshortest to the longest time intervals.\nThe basis for the 6-10 day, medium-range forecast is a modified 500mb\nprog chart consisting of a blend of 500mb heights and anomalies from the\nEuropean Center for Medium Range Forecasts (ECMWF) and NMC's own Medium Range\nForecast (MRF) models. Regression corrected versions of the MRF output are\navailable for days 1 through 5 and 3 through 7 as well as for the 6-10 day\nperiod. The regression correction is based on systematic errors in the MRF\nfor the past 60 days, and thus at least partially corrects for inherent model\nbias as well as for any regime - dependent errors related to the past two\nmonths' weather patterns. The forecaster may also choose one or two observed\n5-day mean patterns from the past 60 days if they are judged (both objectively\nand subjectively) to be reasonably good analogs. Any proportional weights\nadding to 100 percent may be assigned to whatever height fields the forecaster\nwishes to put into the blend, based on his judgment of recent model\nperformance and other factors (Fig. 2, top).\nTwo different objective methods are used to derive temperature anomalies.\nOne is applied to the subjectively modified 500mb 6-10 day mean height\nforecast and the other is computed directly from the regression corrected\nversion of the MRF model output for days 6 to 10. The forecaster considers\nboth predictors and modifies them for better agreement with prevailing anomaly\npatterns observed with past observed cases having key height anomaly centers\nin the same locations as his forecast pattern. The precipitation forecast\ndirectly by the MRF model is used as basic guidance for the precipitation\nprog. It is modified to correct for known regionally and seasonally varying\nmodel biases and for better agreement with similar past observed cases\noccurring with key circulation anomalies similar to that forecast; in\naddition, .03\" is subtracted from the model output everywhere to adjust for\nthe MRF's tendency to be too wet (Fig. 2, bottom).\n* This paper is based on \"Medium and Long Range Forecasting,\" by the same\nauthor, Weather and Forecasting, 4, No. 3, Sept. 1989, pp. 413-426.\n375","Forecasts at the monthly range are made using a mix of both numerical\n(dynamic) and statistical methods. The consistency of these indicators is\nchecked with the aid of teleconnection patterns, which also are often helpful\nin filling in portions of the circulation pattern not clearly indicated by\neither the dynamical or statistical tools (Fig. 3, top).\nTemperature and precipitation categories over the conterminous United\nStates are assigned primarily on the basis of Klein objective specifications\nfrom the predicted mean 700mb height pattern. Objective temperature specifi-\ncations are available for the remainder of North America and extratropical\nportions of Eurasia, although precipitation still must be subjectively esti-\nmated from the forecast height patterns outside the United States. Surface\ntemperature and precipitation auto- and cross-correlations both simultaneous\nand at one-month lag are used to refine the temperature and precipitation\ncategorical forecasts, as are past observed mean temperature and precipitation\nanomaly categories during months at roughly the same time of year with mean\nheight anomaly patterns similar to that being forecast. Finally, probabili-\nties are assigned to the categorical forecasts on the basis of the degree of\nagreement among the forecast tools, the past record of skill at that time of\nyear, the geographic variability of potential predictability, and the strength\n(standardized values) of the objective anomalies based on the predicted height\nanomaly pattern (Fig. 3, bottom)\nSeasonal forecasts are prepared entirely from statistical predictors. A\nqualitative sketch of the predicted principal mean 700mb height anomaly\ncenters is prepared using height anomaly autocorrelations at several different\nlags from one month to ten seasons in the past, based on a sample of about 40\nyears of data. Usually only a few areas are clearly indicated with good sta-\ntistical significance and other parts of the pattern have to be indirectly\ninferred with the help of teleconnections. A number of special tools that are\nstatistically significant only during certain seasons are used at the appro-\npriate times when there is a useful input signal (Fig. 4, top). Additionally,\nthere is now available for all seasonal forecasts an analog technique that\nuses both analogs and anti-analogs in a true predictive sense.\nAnother difference between the seasonal forecast procedure and the\nmonthly and 6-10 day operations is that due to the large amount of subjective\njudgment needed for the seasonal forecast, 3 or 4 forecasters independently\nmake their own forecasts, which are then discussed before a consensus forecast\nis made. As an additional aid in preparing the seasonal precipitation, a\nsearch is made through about 75 years of data for the five best matches to the\nseasonal temperature forecast, and the average of the accompanying precipita-\ntion patterns is used as a specification analog.\nProbabilities are assigned on the basis of the degree of agreement among\nthe forecasters' independent progs, as well as in the monthly forecasts accor-\nding to the agreement among the forecast tools, the strength of the indica-\ntions, potential predictability and the geographical and seasonal variation of\nskill in the past. Seasonal forecasts are currently made for the lower 48\nUnited States only.\n376","Methods of Predicting Mid-tropospheric Circulation\nFigure 1.\nat Different Time Ranges\nAccuracy/Skill/Reliability\nTime Frame\nInput(s) Used\nDaily iterations of\nHighly Accurate\nShort Range\n(1-2 days)\nNWP Models\nDaily iterations of\nGood but rapidly\nExtended Range\n(3-5 days)\nNWP Models\ndeclining accuracy from\nday 3 to 5\n5-day mean for days 1-5\nMean is usually highly\nmay also be constructed\naccurate\nMean has highly variable\nMedium Range\n5-day mean for days 6-10\n(6-10 days)\nquality from good to\nStatistical correction to\nvery poor\n5-day mean reduces model\nerrors and climate drift\nDaily progs available;\nLittle or no skill in\nsometimes useful to see\ntiming of short waves\nwhere precip comes from\nfrom daily progs\nHighly variable, depends\nMean for first 10 days\nMonthly\non model skill, area of\nor so from models\nLong Range\nglobe time of year,\n(30 days)\nand even on regime\n(sensitive to phase of\nPNA during cold season)\nNumerical input beyond\nRemainder of month from\n10-14 days usually\npersistence from first 10\ndegrades quality of\ndays in areas where good\nmonthly mean estimate due\nsignal (anomaly) and\nto increasing errors and\nwhere relationship is\n\"climate drift.\" \"\nsignificant\nCorrelations significant only\nWhole month point-by-point\nin limited areas, often where\nheight anomaly auto-\nnot too useful (subtropics)\ncorrelation with previous\nmonth.\nSmall but useful skill\nStatistical point-by-\nSeasonal\non balance in limited\npoint autocorrelations\nLong Range\nareas but several statis-\nwith previous month\n(90 days)\ntical inputs available\nand several preceding\nseasons back as far as\n2 1/2 years\nSignificant only during\nSpecial statistical\ncertain seasons and can\nrelationships with ENSO,\nbe used only when there\nmid-latitude Pacific\nis good input signal\nSSTA's, QBO and solar\nflux are sometimes useful\n377","Figure 2. Construction of \"best\" numerical model-based\n6-10 day mean height/anomaly prog\nECMWF\nMRF\nREGRESSION\nMODEL\nMODEL\nCORRECTED\nMRF\nX (%\nY%\nZ%\nRecent 5-day\nMODIFIED\n- P% - >\nobserved mean\n500MB PROG\nP+Q+X+Y+Z=100%\n(Analog #1)\n\"The Blend\"\nRecent 5-day\nQ%\nobserved mean\n(Analog #2)\nForecaster may use any or all of these inputs in any proportions\nadding to 100% depending on his judgment of quality of the various\nmodels as determined by:\nRECENT MODEL PERFORMANCE\nCONSISTENCY BETWEEN SUCCESSIVE RUNS OF EACH MODEL\nAGREEMENT BETWEEN MODELS\nREASONABLENESS OF EVOLUTION FROM FIRST 5 DAYS\nSUBJECTIVE EVALUATION OF PATTERN AS TO AGREEMENT WITH WELL- -\nKNOWN TELECONNECTIONS, CLIMATOLOGICALLY PREFERRED POSITIONS OF\nTROUGHS AND RIDGES, HIGHS AND LOWS, ETC.\nOBJECTIVE ANOMALY CORRELATION AND RMS DIFFERENCE SCORES OF\nCANDIDATE ANALOGS FROM PAST TWO MONTHS\nConstruction of Temperature Anomaly and Precipitation Category\nProgs for the 6-10 day forecast period\nREGRESSION\n\"BLEND\" MODIFIED 500MB MEAN\nMRF\nCORRECTED\nHEIGHT AND ANOMALY PROG\nMODEL\nMRF\nFOR DAYS 6-10\nDAYS 6-10\nMODEL\nREGRESSION\nKLEIN OBJECTIVE\nDIRECT*\nCORRECTED\nTEMPERATURE ANOMALIES\nPRECIP\nSURFACE\nBASED ON BLEND PROG\nOUTPUT\nTEMPERATURE\nFROM\nANOMALIES\nMRF\n* .03\"\nOFFICIAL\nANALOGS\nOFFICIAL\nTEMPERATURE\nPAST 5-DAY MEANS OBSERVED\nPRECIPITATION\nANOMALY\nWEATHER PATTERNS THAT OCCURRED\nCATEGORY\nPROG\nWITH MEAN HEIGHT ANOMALIES IN\nPROG\n(CATEGORIES)\nTHE SAME LOCATIONS AS ON THE\n(CLASSES)\nMODIFIED 500MB PROG\n378","Construction of \"best\" 30-day mean height/anomaly prog\nFigure 3.\nBlend 500 mb 6-10\nStatistical persistence\nStatistical point\nautocorrelation of\nday prog (can include\nof first 10 days to\nmonthly mean height\nfirst 5 days also)\nwhole month\nanomalies with\nprevious month\nConsistency checked by Teleconnections and estimated\nreliability of each method for that particular forecast\n(depends on time of year, etc.)\nSubjective estimate of predicted 700mb mean height\nand anomaly pattern for the coming 30-day period\nConstruction of 30-Day Temperature and Precipitation Forecasts\nAnalagous past 30-day mean\nSubjective 30-day mean\nobserved height anomaly\n700mb height/anomaly prog\npatterns selected within\n+/- 1/2 month\nKlein objective temperature\nanomalies and categories for\nAccompanying observed\ntemperature anomaly\nUS, rest of North America\nand Eurasia. Precipitation\nand precipitation\nobjective specification\ncategories\ncurrently for US only (lower\n48). Special equations for\nHawaiian precipitation\nSurface temperature and\nPreliminary temperature\nprecipitation auto- and\nand precipitation\ncross-correlations simul-\ncategorical forecasts\ntaneous and 1 month long\nDegree of agreement among\nall forecast tools\nPast observed skill\nFinal temperature and\nat same time of year\nprecipitation patterns\nexpressed in probability\nGeographic variability of\nformat\npotential predictability\nStrength (standardized\nvalues) of Klein objective\nanomaly patterns based on\nheight prog\n379","Figure 4. CONSTRUCTION OF \"BEST\" SCHEMATIC 700MB HEIGHT ANOMALY PROG\nFOR 90 DAY PERIOD\n8, 9, OR 10-\n4-SEASON\n3-SEASON\n-SEASON\n1-SEASON\n1-MONTH\nSEASON LAG\n(1-YEAR)\nAUTOCORRELATIONS\nOF 700MB HEIGHT\nERICKSON\nENSO\nSUBJECTIVE SKETCH\nINDEX\nSTATISTICAL\nOF SEASONAL MEAN\nRELATIONSHIPS\nRELATIONSHIPS\n700MB HEIGHT\n(Chen)\nANOMALIES EXPECTED\nFOR PREDICTAND SEASON\nOTHER\nNEW TOOLS\nSOLAR/QBO\nGULF OF ALASKA\n(Barnston-Livezey)\nSST-SLP\nRELATIONSHIPS\n(Davis)\nAs with monthly forecasts, the various indications (predictors) are\nchecked against teleconnection patterns, and conflicts between different\npredictors are resolved with their help and by comparing the statistical\nsignificance of the various relationships.\nOne or more (usually more!) of these inputs may be too weak (either due\nto weak signal input or weak/insignificant correlation) to be useful in any\nCONSTRUCTION OF SEASONAL TEMPERATURE AND PRECIPITATION\nPROBABILITY FORECASTS\nSubjectively chosen past\nObjectively chosen analogs based\nobserved 30- and 90-day\non height, thickness, SST anomalies\nheight anomaly patterns\nin preceding season and/or months\nwith key features resembling\n(Multiple statistical analog/anti-\nanomalies in the circulation\nanalog method) (Barnston & Livezey)\nsketch judged to be significant\n(Specifications Analogs)\nAccompanying observed\nRanked set of temperature analogs\ntemperature and\nand anti-analogs following 10 best\nprecipitation patterns\nmatches to preceding season's\npredictors. Also composite of\nthe 10\nSpecial forcing factors\n(ENSO, Solar, etc., with\nsignificant signals at\nthat season\nComposite final consensus\nAll the above processes\ntemperature and preliminary\nare gone through by 3\nprecipitation categorical\nor 4 forecasters working\nforecasts\nindependently\n5 best analogs to consensus\ntemperature category forecast\nFinal Precipitation Categorical\nselected from US statewide-\nForecast (quality of 5 analogs\naveraged data going back to 1899\nsubjectively filtered or\nevaluated)\nPast scores of Analog method\nPast skill of official temperature\nFinal Temperature and\nand precipitation forecasts\nPrecipitation forecasts\nseasonal persistence at 1 season\nfor U.S. lower 48 expressed\nand 1-year lag (SU) for Temperature.\nin probability format\nGeographical variation of potential\npredictability\n380","Modernization of Long-Range Prediction Operations at NMC\nEdward A. O'Lenic\nClimate Analysis Center, NMC, NWS, NOAA\nWashington, DC 20233\n1. Introduction.\nThe Prediction Branch (PB) of the National Meteorological Center (NMC)\nhas a venerable tradition in its production of the Monthly and Seasonal\nWeather Outlook (Gilman 1985, Namias 1968) of average temperature and precipi-\ntation. During the nearly twenty years since the inception of this product\nnumerous improvements have been made to the basic method of constructing the\nOutlook. These include improved statistical tools, longer data sets and the\nuse of medium range forecast (MRF) models (Gilman 1985, Wagner 1989). Yet\ndespite these noteworthy improvements, throughout its history, the actual\nprocess of constructing the Outlook has remained essentially the same.\nRecently, however, the National Weather Service has begun to modernize\nmuch of its operations. Among the changes which have ensued are a transition\nto more automated activities and away from activities performed manually by\nmeteorological technicians or by forecasters themselves. This wave has not\npassed by the Prediction Branch. Changes are also taking place in our opera-\ntions. The purpose of this paper is to outline some of those changes and the\nactivities attending them, and to briefly lay out our plans for the ongoing\nmodernization of our prediction operation. Although the modernization project\ncurrently involves both the 30- and 90-day operations, to only the 30-day\noperation will be discussed in detail here.\n2. The Forecast Process.\nUntil very recently, the method of constructing the 30-day Outlook was as\nshown in the schematic illustrated in Fig. 1. The two-day process, normally\nrequiring the full-time attention of one forecaster and two meteorological\ntechnicians (MTs), begin an MT initiates a computer program on the NAS 9000,\nNMC's mainframe computer, to place certain data into files for subsequent use.\nAnother MT initiates a program to produce a monthly mean observed upper air\nmaps from the previous month and a linear regression forecast, based upon the\nobserved data, valid for the upcoming 30 days. These maps are hand colored\nand made available to the forecaster. While this is going on, the other MT\nquality controls his data by visual inspection, then plots the data via line\nprinter and hand analyzes and colors the resulting maps.\nIn addition to the observed data and linear regression forecast maps, the\nforecaster also has access to the MRF 6-10 day forecast made at 00 UTC and the\nECMWF 6-10 day forecast made at 12 UTC the previous day, a catalog of maps of\nobserved monthly mean upper air height maps and the corresponding maps of\nobserved surface temperature and precipitation, and certain other resources\nsuch as knowledge of the current status of ENSO, a catalog of teleconnection\nmaps, knowledge of typical month-to-month persistence as a function of month,\netc\n381","The forecaster uses all of this information to formulate a prognostic 700\nmb height departure from normal (DN) maps which he sketches by hand. This map\nis used as a first guess for Day 2 activities, when updated MRF and ECMWF 6-10\nday forecasts are available (Fig. 1). The forecaster finalizes the DN prog-\nnostic map and gives it to an MT who would then manually digitize it, use that\ndata to compute prognostic surface temperature and precipitation maps via\nscreening regression (Klein progs), and finally hand analyze and color the\nmaps and present them to the forecaster.\nUsing an ensemble of tools including knowledge of the seasonal variation\nof forecast skill, analogs, and other tools, the forecaster formulates hemis-\npheric probability prognostic maps of temperature and precipitation and\nsketches them by hand. He also formulates a written narrative of the Outlook\nfor temperature and precipitation over the U.S.\nThe final set of procedures shown in Fig. 1 may be called the publication\nprocess. This involves several highly laborious procedures including manual\ndrafting of the observed and prognostic maps, doing the layout of the Outlook\ndocument itself, digitizing the prognostic maps in order to place them into\nthe AFOS system, and typing and distribution of the Outlook narrative and maps\n(not included in the Outlook, but prepared for the press and public). In all,\nthe Outlook publication alone requires eleven drafted maps and three tables.\n3. A Plan for Modernization.\nAmong the motivations for modernizing the above forecast system are,\nfirst, the impending loss of MT support (they will not be replaced by other\nMTs when they retire or leave). This factor requires that we streamline the\nprocess and reduce its labor intensity. Other motivating factors include a\ndesire to explore such options as increasing the flexibility of the forecast\nsystem, expanding the options for analysis and computation available to the\nforecaster and development of a more convenient system for developing, testing\nand implementing new forecast tools and procedures.\nThe second major motivating factor is that NMC's NAS 9000 mainframe is\nheavily committed to operational requirements related to short and medium\nrange forecasting, i.e., hourly data acquisition, storage and transmission,\nrunning numerical analysis and forecast systems, and the formulation, plotting\nand transmission of myriad operational graphical and alphanumeric products.\nWhile this dedication of the mainframe is wholly appropriate to NMC's medium\nand short-range operations, it results in both scheduled and unscheduled\nlosses of service which are inconvenient for our operation. A one or two-hour\ndelay on Day 2 of the forecast process often means a like delay in getting the\nfinished Outlook materials to the printer, resulting in additional cost to us\ndue to overtime and other costs.\nClearly, one of the keys to enabling PB to achieve our modernization\ngoals is to develop the capability to perform our computing and graphics\nindependently from the mainframe. The computing system chosen for this task\nis an Apollo 3500 workstation equipped with a digitizer. This powerful device\npermits development of sophisticated graphics and has the capability to run\nlarge software programs and to communicate with a variety of other machines\nincluding the NAS 9000, and other Apollos, via an Ethernet line. Because our\n382","machine is one of several Apollos which are part of a local area network\ncalled a ring, information and CPU can be shared among the individual\nmachines. Nearly all of the graphics software developed for our Apollo system\nwas developed by Mr. Keith R. Ward of Ellsworth Associates, Inc.\nThe activities of the modernization process are:\n1. Automate all manual operations such as computer job initiation,\nquality control, data plotting, analysis and coloring, digitizing,\ndrafting and layout of documents. Also, transfer all computing\nactivities from the mainframe to the workstation.\n2. Develop an interactive forecast/computation system including:\na. Rapid access and display of forecast material including model\noutput and long-term data bases.\nb.\nCapability to perform statistical computations tailored to a\nparticular forecast.\nC. Multivariate statistics.\n3. Expand automation to desktop publication of the Outlook, updating of\ndata bases and dissemination of documents to users via Ethernet.\nAs of this writing, the first stage of the modernization process is about\n70 percent complete. Virtually all of the manual activities marked with an\nasterisk in Fig. 1 have been automated to some degree, and the process will\ncontinue. Ultimately, we hope to expand the options available for exploring\nsolutions to both forecasting and research problems. Obviously, any appli -\ncation of forecasting tools developed using the new system must be carefully\ntested to ensure consistency. The new flexibility of the modernized system\nwill make such tests much easier and efficient to conduct, and we will then be\nable to not only keep abreast of new developments in the technology of long-\nrange prediction, but also forge some of those developments ourselves.\nReferences\nGilman, D. L. 1985: Long-range forecasting: The present and future. Bull.\nAmer. Met. Soc., 66, 159-164.\nNamias, J. , 1968: Long-range weather forecasting - History, current status\nand outlook. Bull. Amer. Met. Soc. , 49, 438-470.\nWagner, A. J. , 1989: Medium and Long-Range Forecasting. Weather and\nForecasting, 4, 413-426.\n383","Day 1\nN.H. observed\ndata available\nAutomated\nAcquire obs.\ncolor\n700mb mean\nT. precip. data\nZ maps\nAutomated\nQuality Control\nlinear regr.\ncolor\nforecast\nPlot observed\nmaps\nAnalyze/color maps\nMRF prog\nAnalogs\nForecaster\nENSO, RPCA,\nTeleconnections\n30-day mean\n700mb Z DN\nprog map\nDay 2\nMRF\nAnalogs\nForecaster\nENSO.\nteleconnections,\netc\nFinal DN\nmap\nDigitize\nKlein\nregression\nAnalyze.\ncolor maps\nMRF\nENSO, etc.\nSkill Variability\nForecaster\nWritten\n*\nT. precipitation\nnarrative\nprobability maps\nTyping\n*\nDrafting\nDigitizing\nDistribution\nPublisher\nAFOS\n*\nFigure 1. . Schematic of activities required to produce the monthly Outlook\nof average temperature and precipitation. Processes marked with\nan asterisk indicate activities which have been automated.\n384","The Status of Long-Range Forecasting at the UK Meteorological\nOffice\nMSJ\nHarrison, A Dickinson, SF Milton, DS Richardson, AW Colman\nand A Woodcock\nUnited Kingdom Meteorological Office\nLondon Road\nBracknell, Berks RG12 2SZ\nLong-range forecasts for the United Kingdom are prepared at the\nMeteorological Office every second Monday to cover three periods,\n1 -5, 6-15 and 16-30 days ahead. Model predictions from either\nthe UK Met. Office or ECMWF form the basis for the 1-5 day\nforecasts, in which there is a high level of skill. At longer\nranges ensemble forecasts, run using a reduced horizontal\nresolution version of the UK Met. Office NWP model, and\nstatistical techniques are employed. In all cases predicted\nsurface pressure fields and, at days 1-5 , 1000-500 mb thickness\nfields are converted to temperature and rainfall prognoses for 10\nregions across the UK through linear specification equations.\nReal-time ensemble forecasts were introduced in December 1988,\nconsisting of nine integrations run from analysis times separated\nby six hours. Additional hindcast ensembles have been run to\nprovide a time series of 47 ensembles from September 1987 to June\n1989. Average anomaly correlations (ACC) for the North Atlantic\nand West European area for the ensemble mean are 0.79 at days\n1-5, 0.31 at days 6-15 and 0.15 at days 16-30. However there is a\nmarked case-by-case and seasonal variation in ensemble skill\n(Fig. 1), with highest skill in winter (ACC = 0.52 at days 6-15\nand ACC = 0.41 at days 16-30) and lowest skill in the\ntransitional seasons (ACC < 0.1 The ensemble mean is\nconsistently more skilful than persistence, with higher anomaly\ncorrelations in 74% (53%) of cases at days 6-15 (days 16-30) (Fig\n1. - lower panel). The impact of ensemble averaging is greatest\nwhen there is some residual skill (ACC > 0.2) in the individual\nensemble members. In these cases the ensemble mean is more\nskilful than the latest member beyond days 1-5 (Table 1). Spread\n(average anomaly correlation between each member and the ensemble\nmean) across the ensemble may provide a useable guide to forecast\nskill in some seasons. Strong spread-skill relationships occur\nat days 1 to 5, but correlations for other periods vary both\nseasonally and regionally (Fig. 2). In general, highest spread-\nskill correlations occur in winter and spring (Fig. 2a, 26), with\nparticularly poor correlations at days 6 to 15 in summer (Fig.\n2c, 2d).\nStatistical forecasts have been made throughout the 1980's using\ntechniques based on eigenvectors of surface pressure over regions\nof the Northern Hemisphere. The earliest technique, SPEVR,\n(Surface Pressure EigenVector Regression) uses eigenvector time\n385","coefficients representing the observed northern hemisphere\ncirculation over recent months to predict time coefficients for\neigenvectors across the North Atlantic/west Europe region.\nA\nmore recent development, MVA, uses discriminant analysis to\nselect between six clusters formed from eigenvector time\ncoefficients for the North Atlantic/west Europe region.\nPredictors for MVA include eigenvectors of the hemispheric\nsurface pressure and 1000 to 500 mb thickness over toroids of the\nNorthern Hemisphere and SST's in selected regions. Based on\nhindcasts over 18 years it appears that MVA skill is marginal\nwith highest skill in winter (Jan-Feb) and pre-winter (Nov-Dec),\nMost of the winter skill originates with a single surface\npressure predictor eigenvector (Fig. 3) with a strong weighting\nover north-west Siberia. A positive weighting (i.e. positive\npressure anomaly over Siberia) often precedes a blocked\ncirculation over the North Atlantic - i.e. a positive weighting\nof the eigenvector pattern of Figure 4 (which resembles the North\nAtlantic 0scillation) Equally, low pressure over Siberia is\noften a precursor of low pressure around Iceland.\nThe\ncorrelation of 0.38 between the predictor pattern for January 1\nto 15 and the predictand pattern for January 16 to 31 (Fig. 5) is\ntypical of half-month lags between these two patterns throughout\nwinter and is significant at the 99% level. This is an unusually\nhigh correlation for lagged relationships in the northern\nextratropics. Although overall MVA skill is low the technique is\nbetter (for unknown reasons) at forecasting certain clusters than\nothers (Table 2). On the basis of present evidence for the years\n1987-89, ensemble predictions in general provide highest skill\nout to 15 days and in winter/pre-winter (Nov-Feb) and pre-summer\n(May-June) out to 30 days but statistical techniques are\ngenerally more skilful than the ensembles over 16 to 30 days in\nthe transitional seasons, although skill levels are very low.\n(Table 3).\nMonthly temperature prediction skill, based on the Folland-\nPainting skill score which uses information theory to allocate\nmarks dependent on the accuracy with which temperature is\nforecast in quints for each of the 10 UK regions, has recently\nbeen at the highest level ever on the basis of a running one-year\nmean assessment (Fig. 6) In part this high level of skill may\nbe attributable to the persistence of relatively warm\ntemperatures over the UK since November 1988. Skill was not\nhigh, however, in past persistent periods and the introduction of\nthe ensembles is thought to have made a major contribution to\nthis skill level. Rainfall prediction skill (based on terces)\nhas not reached the levels of temperature prediction skill,\npartly due to systematic under-prediction in the south and over-\nprediction in the north during the 1988/89 winter.\n386","SKILL BINS (AC x 100)\n1\n2\n3\n4\n5\n84\n89\n46\n68\n03\n04\nD1-5\n-\n-\n89%\n9%\n2%\n80\n75\n44\n38\n01\n09\n-55 -40\nD6-15\n-\n17%\n43%\n34%\n6%\n81\n70\n45\n41\n06\n15\n-57 -29\nD16-30\n-\n9%\n36%\n45%\n11%\nTable 1. Skill of ensemble mean and latest ensemble member for\nthe N. Atlantic and Europe, stratified according to the mean\nskill of the ensemble members. Skill bins are 1 (1. >AC> 6) ,\n2 (.6>AC>. 2) , 3 (.2>AC>-.2), 4 (-.2>AC>-.6) - 5 (- .6>AC>-1.).\nEach box shows the percentage of forecasts where the mean skill\nof the members falls in the specified bin and the average over\nthose cases of the skill of the ensemble mean (left-hand\nfigure) and the latest member (right-hand figure).\nCLUSTERS\n1\n2\n3\n4\n5\n6\nAVG\nWINTER\nACC\n0.089\n-0.013\n0.131\n-0.020\n0.089\n0.303\n0.124\nFREQ\n20\n10\n9\n10\n10\n13\n72\nSPRING\nACC\n0.088\n0.037\n0.028\n0.558\n0.233\n0.096\n0.115\nFREQ\n20\n5\n18\n3\n8\n18\n72\nPRE-SUMMER ACC\n0.250\n0.228\n0.010\n0.302\n0.149\n-0.193\n0.064\nFREQ\n8\n3\n31\n8\n7\n15\n72\nSUMMER\nACC\n-0.265\n0.132\n0.057\n0.062\n-0.030\n0.053\n-0.013\nFREQ\n1\n13\n22\n1\n28\n7\n72\nAUTUMN\nACC\n0.018\n0.111\n-0.123\n-0.050\n0.132\n-0.079\n0.012\nFREQ\n20\n18\n8\n4\n8\n14\n72\nPRE-WINTER ACC\n0.140\n0.337\n0.120\n0.120\n0.117\n0.075\n0.158\nFREQ\n14\n6\n8\n13\n21\n10\n72\nTable 2 - Skill of predictand clusters. Anomaly correlations for\nmean sea level pressure over a UK 30 point grid.\nFIRST HALF MONTH\nSECOND HALF MONTH\nMVA\nE. MEAN\nMVA\nE. MEAN\nWINTER\n0.105\n0.506\n0.034\n0.404\nSPRING\n0.029\n0.304\n-0.086\n-0.130\nPRE- SUMMER\n0.055\n0.247\n0.130\n0.287\nSUMMER\n0.210\n0.173\n0.038\n0.008\nAUTUMN\n0.001\n-0.016\n0.125\n-0.290\nRE-WINTER\n0.252\n0.504\n-0.018\n0.243\nTable 3. - Anomaly correlations of the top cluster from\nstatistical (MVA) forecasts 1986 - 89 versus the ensemble mean\nfor a 30 point grid surrounding the UK. The first half month\nMVA forecasts are compared with days 6-15 of the ensemble mean\nforecasts and the second half months with days 16-30. (Each\nseason is composed of four half months, e.g Winter is Janl,\nJan2, Febl, Feb2) .\n387","1.0\nW\nW W WW\nS\nWS\nA\nW\nA\nT\nT\n0.5\nA\nS\nT\nT\nA\nW\nT\nT\nA\nS\nW\nW\nS\nA\nW\nW\nS\nS\nT\nA\nA\n0.0\nTT\nA\nSs\nW\nS\nS\n-0.5\nA\nA\nA\n-1.0\nO\n10\n20\n30\n40\n50\n1\no\n-1\n-2\no\n10\n20\n30\n40\n50\nENSEMBLE NUMBER\nFigure 1 - (Top panel) Time series of ensemble mean anomaly\ncorrelations for the ensembles Sept 87 - June 89. Mean sea level\npressure for N.Atlantic and Europe (60W-40E, 30N-70N) at days 6-15.\n(Lower panel) Difference in anomaly correlation between the ensemble\nmean and persistence forecasts for MSLP at days 6-15.\na)\nNHEM\nb) NATL/EUR\n12\n1.0\nR=0.67\n0.7\n0.5\n02\n00\nR= 0.51\n-03\n-05\n0.75\n080\n085\n090\n095\n100\n06\n0.7\n08\n0.9\n10\nSPREAD ( ACC )\nSPREAD ( ACC )\nd) NATL/EUR\nc) NHEM\n1.0\n1.0\nR=0.12\n05\n0.5\n00\n00\n-05\nR=0.00\n-05\n-10\n06\nU.7\n08\n09\n:0\n07\n08\n09\n10\nSPREAD ( ACC)\nSPREAD ( ACC )\nFigure 2 - a) Scatterplot of skill (anomaly correlation) against\nspread (see text) for days 6-15 of Winter and Spring ensembles for\nthe Northern Hemisphere (15N-90N) Closed circles = Winter, open\ncircles=Spring. b) as (a) for N. Atlantic and Europe. c) as (a) for\nSummer/Autumn closed circles=summer open circles=autumn. d) as (c)\nfor N.atlantic and Europe.\n388","60\nFigure 3 - Predictor eigenvector pattern (EV3) for Jan 1-15\n40\n10\n90\nso\n30\nby\nSb\n40\nFigure 4 - Predictand eigenvector pattern (EV1) for Jan 16-31\n389","PMSL EOF 3 i JAN1) y PMSL EOF\n(JAN2\nJAN PMSL EV\nJAN PMSL EV I\nWINTER PMSL EIGENVECTOR TIME COEFFICIENTS FOR 1921-08\nJAN PASL EV 1\nIS\nJAN PHSL EV 3\n:\nR=0.38\n10\n10\nS\n,\n0\n-S\n-10\n10\n-15\nIS\n1921\n1931\n1941\n1951\n1961\n1971\n1981\nDATE\nFigure 5 - Time series of Winter MSLP eigenvector time coefficients\nfor the predictor pattern (EV3) and predictand pattern (EV1) 1921-88\nONE YEAR RUNNING MEAN SKILL\nTEMPERATURE QUINTS PLOTTED EVERY TWO MONTHS (ONE NATURAL SEASON)\n4.00\n63.00\n7.00\n78.00\n2.00\n00.00\n80.00\n!YEAR\nISSUED FORECAST SCORES\nPERSISTENCE FORECAST SCORES\nCLIMATOLOGY FORECAST SCORES\nFigure 6 - Time series of 1 year running mean of skill of issued\ntemperature forecasts 1964 - 89. The skill measure is the Folland -\nPainting score (see text) .\n390","The Relationship between Global Sea-Surface Temperatures and Sahel Rainfall - Investigations using a GCM\nD.F.Rowell, K.Maskell and J.A.Owen\nMeteorological Office, Bracknell, Berks, U.K.\nIntroduction: Interannual variability of Sahel rainfall is known to depend on sea-surface temperature\n(SST) anomalies on 2 near-global scale (Folland et al., 1986, 1989 and Wolter, 1989). This paper provides\nfurther evidence for such a relationship based on integrations of a general circulation model (GCM). We\nillustrate the simulated large-scale circulation changes between wet and dry Sahel years, and also briefly\ndiscuss the forecasting capability of the model.\nThe Model: The GCM used is similar to that of Cunnington and Rowntree (1986), having 11 levels in the\nvertical, a horizontal resolution of 2.5 latitude by 3.75° longitude, and detailed physical\nparameterizations. It has a good rainfall climatology in the Sahel, both temporally and spatially, during\nthe main part of the wet season (July to September).\nSimulations: A number of experiments have been carried out, each consisting of a 210-day integration\nstarting from identical initial atmospheric conditions (26 March 1984). Observed monthly global SST data\n(from April to October) were taken from the MOHSST dataset (Parker and Folland, 1988), interpolated\nspatially to the model grid, and interpolated temporally such that they were updated every 5 days.\nTo date six years have been simulated: two very wet years (1950 and 1958), two very dry years (1983 and\n1984), one dry year (1976), and one average year (1988). Results are shown in Fig.1, and it is clear that\nthe model is skilful in its ability to simulate interannual variability of Sahel rainfall based on SST data\nalone. An interactive soil moisture scheme is normally used, whereby the moisture content varies according\nto precipitation and evaporation. Replacing this with non-interactive climatological soil moisture on all\nland surfaces has a relatively minor impact (Fig.1), demonstrating, that in the model, forcing from the SST\nanomalies is dominant. The impact of using different sets of initial conditions is also being investigated.\nFig.2 illustrates that the model is also well able to simulate the change in spatial pattern of rainfall\nbetween a wet year and a dry year.\nCirculation Changes: Using a range of diagnostics from four of the experiments (1950, 1958, 1983 and\n1984), we have studied the differences in large-scale circulation which directly influence the rainfall of\nNorth Africa. These are all illustrated as July to September means, and it is important to note that the\nresults are computed from data extracted from the model every 6 hours. It is essential that the diurnal\ncycle is well sampled in this way to avoid misleading results. The following paragraphs summarize our\nfindings.\nMoisture flux: The primary source of moisture for much of the Sahel is a low-level flow originating from\nthe tropical Atlantic. Fig.3 shows cross-sections of the zonal and meridional components of moisture flux\nin the lower troposphere. The strength of the moisture flux into the Sahel is seen to be substantially\nreduced in drought years, and the ITCZ (which may be defined as qu=0 or qv=0) is seen to be further south.\nBoth factors will clearly have an impact on the rainfall.\nHadley Circulation: This is illustrated using plots of the divergent meridional component of vertically\nintegrated mass flux (Fig.4). Contours may be interpreted as streamlines of the meridional divergent flow\nand its contribution to the vertical flow. The major difference is that the southern cell is significantly\nweaker in drought years, contributing to the reduced supply of moisture to the Sahel and decreased large-\nscale ascent.\nWalker Circulation: This is illustrated using graphs of 200 and 850mb zonal divergent wind. These\nindicate the magnitude and direction of the divergent flow, and the divergence may be inferred from the\ngradient, hence enabling areas of large-scale ascent/descent to be located. The mean for the four years is\nshown in Fig.5, and drawn from this is a sketch of the model's mean Walker circulation. The anomalous\ncirculation for each year relative to the four-year mean is shown in Fig.6. Wet years show anomalous ascent\nover Africa, and dry years anomalous descent. In 1950 and 1983 (an E1 Nino year) there appears to be no\nstrong coherence between the anomalous vertical motion over Africa and other cells of the Walker\n391","circulation. However, in 1984 the anomalies are associated with a strengthening of the entire circulation,\nas well as a slight eastward shift. 1958 is also characterized by large global anomalies, this time\ncorresponding to a westward shift of all cells apart from the western part of the Indian cell. Clearly\nmodelled global circulation anomalies may be of quite differing character, irrespective of Sahel rainfall\nanomalies.\nForecasting Capability: 'Hindcast' experiments have been carried out for five of the years simulated.\nSST anomalies are computed for either April, May or June, and persisted through to October by adding then\nto an SST climatology which is updated every 5 days. Initial conditions are taken from the appropriate\npoint in the corresponding simulation, i.e. day 30, day 60, or day 90. The results are shown in Table 1.\nThere is clearly potential to provide skilful forecasts from June SST anomalies, though as the lead time\nincreases this skill declines, 50 that only the sign of the rainfall anomaly may be predictable. The\nhindcasts for 1976 are particularly poor, and the reason for this is not yet clear,\nFurther Work: The GCM used in this study seems well able to simulate interannual variability of Sahel\nrainfall, when forced by observed SSTs, Model diagnostics are of great value in aiding our understanding of\nthe links between SST anomalies and tropical drought, but detailed comparison of the modelled circulation\nchanges with available observations is also necessary. Furthermore the effects of individual ocean basins\nand their (non-linear) combination must also be investigated. With this end in mind a further series of\nexperiments is under way, and initial results are reported in Folland et al. (1989).\nIn order to extend the lead time at which skilful forecasts may be provided, prediction of SST anomalies\nis required. One aspect of this which is currently being studied, is the 'typical' seasonal evolution of\nSST anomaly patterns during wet and dry Sahel years.\nReferences:\nCunnington, W.M. and Rowntree, P.R. (1986) Simulations of the Saharan atmosphere - dependence on moisture\nand albedo. Quart. J. R. Met. Soc., 112, 971-999.\nFolland, C.K. Palmer, T.N. and Parker, D.E. (1986) Sahel rainfall and worldwide sea temperatures, 1901-85.\nNature, 320, 602-607.\nFolland, C.K., Owen,J.A. and Maskell,K (1989) Physical causes and predictability of variations in seasonal\nrainfall over sub-Saharan Africa. Proc. of the IAHS Third Int. Assembly, Baltimore, MD, May 1989, 87-95.\nParker, D.E. & Folland, C.K. (1988) The Meteorological Office Historical Sea Surface Temperature data set.\nRecent Climatic Change - a regional approach (ed. S Gregory), Belhaven Press, 41-50.\nWolter, K. (1989) Modes of tropical circulation, southern oscillation and Sahel rainfall anomalies, J.\nClimate, 2, 149-172.\nYear\nJuly to Sept\nrain (mm/day)\n1950\n1958\n1976\n1983\n1984\nObserved\n5.5\n4.2\n2.7\n2.0\n1.4\nSimulated\n5.9\n4.7\n2.4\n1.3\n2.0\nJune Hindcast\n6.6\n4.1\n2.9\n1.4\n1.8\nMay Hindcast\n4.0\n4.2\n4.3\n3.1\n1.4\nApril Hindcast\n3.4\n3.3\n3.8\n2.4\n2.4\nTable 1: Comparison of hindcast results with observed and simulated Sahel rainfall. The observed 1951-80\nmean for July to September is 3.2mm/day.\n392","Modelled (interactive soil moisture)\nObserved\nModelled (climatological soil moisture)\n8\n6\n4\n2\n1976\n1983\n1984\n1988\n1950\n1958\n0\nJ A S o\nS\no\nJ\nJ\nA\nS\no\nJ\nJ\nJ\nA\nS\no\nJ\nA\nS\no\nJ\nJ\nA\nJ\nJ\nA\nS\no\nJ\nMonths\nMonths\nMonths\nMonths\nMonths\nMonths\nFig.1: Simulated and observed Sahel rainfall in six past years.\n15°\n30°\n15°\n0°\n45°E\n30°W\n30°N\n0\n1\n-1\n0\n-2\n-4\n15°\n-4\n8\nI\n+4\n+4\n+2\n+1\n+8\\\n+16\n0\n+1\n1\n+4\n+4 +2 +1\n0°\nFiq.2a: Observed rainfall differences between 1984 and 1950, for July to September, Contours are at 0, +1,\n42, +4, +8, +16 day-1, -1 with negative contours dashed. The area we define as the Sahel is also marked.\n0°\n15°\n30°\n45°E\n30°W\n15°\n30°N\n0\n1\n-\n-2\n4\n15°\n-8\n-2\n0\n-1\n0\n+1\n+2\n+2\n+2\n2\n0\n+1\n+1\n-1\n+1\n+1\n0°\nF10.2b: As Fig.23, but modelled rainfall differences between 1984 and 1950.\n393","500\n500\n(a)\n(b)\n550\n550\n0\n600\n600\n650\n650\n700\n700\n-2\n750\n750\n800\n800\n850\n850\n900\n900\n950\n950\n3\n3\n3\n,0.\n1000\n1000\n25\n20\n15\n10\n5\n25\n20\n15\n10\n5\nLATITUDE (N)\nLATITUDE (N)\nFio.3a.b: Cross-sections of the zonal component of moisture flux (qu), averaged from 153W to 37.5°E. for\nJuly to September (a) mean of 1950 and 1958 experiments, (b) mean of 1983 and 1984, Units: x100 gkg-1as-1\n500\n500\n(c)\n(d)\n550\n550\n600\n600\n650\n650\n700\n700\n750\n750\n0\n800\n800\n0\n850\n850\n900\n900\n2\n950\n950\n-2\nS\n3\n3\n1000\n5\n1000\n25\n20\n15\n10\nS\n25\n20\n15\n10\n5\nLATITUDE (N)\nLATITUDE (N)\nFio.3c.d: As Fig.3a,b, but for the meridional component (qv).\n394","100\n100\n200\n(a)\n(b)\n200\n2\n2\n3\n3\n300\n300\n4\n4\n3\n3\n400\n400\n500\n500\n600\n600\n700\n700\n5\n800\n800\n900\n900\n3\n-1\n2\n2\n2\n1\n1000\n1000\n0\n0\n8\n0\n0\n0\n0\n25\n20\nis\n10\n5\n-5\n- Yo\n25\n20\n15\n10\n5\n0\n-5\n-10\nLATITUDE IN)\nLATITUDE (N)\nFiq.4 Cross-sections of the divergent meridional component of vertically integrated 0855 flux, averaged\nfrom 15°W to 37.5°E, for July to September (a) mean of 1950 and 1958 experiments, (b) mean of 1983 and\n1984. Units: kgs -1\n2.5\n2.0\n1.5\n(a)\n1.0\n0.5\n0.0\n-0.5\n-1.0\n-1.5\n-2.0\n-2.5\n-3.0\n-3.5\n-4.0\n-4.5\n-5.0\n0\n30\n60\n90\n120\n150\n180\n210\n240\n270\n300\n330\n360\nLONG4TUDE (E)\n(b)\nINDIAN\nPACIFIC\nMEXICAN\nATLANTIC\nCELL\nCELL\nCELL\nCELL\nFig.5: (a) Zonal component of divergent wind, averaged from 5°N to 25°N, for July to September Mean of\n1950, 1958, 1983, and 1994 experiments. Solid line is 850mb, dashed line is 200mb. Longitude is degrees\neast of Greenwich, and the Sahel lies between 345°E and 37.5°E.\n(b) Sketch of BEER Walker circulation between 5°N and 25°N, for July to September\n395","0.6\n0.4\n(a)\n0.2\n0.0\n-0.2\n-0.4\n-0.6\n-0.8\n0\n30\n60\n90\n120\n150\n180\n210\n240\n270\n300\n330\n360\nLONGITUDE (E)\n1.4\n1.2\n(b)\n1.0\n0.8\n0.6\n0.4\n0.2\n0.0\n-0.2\n-0.4\n-0.6\n-0.8\n-1.0\n0\n30\n60\n90\n120\n150\n180\n210\n240\n270\n300\n330\n360\nLONGITUDE (E)\n0.4\n(c)\n0.2\n0.0\n-0.2\n-0.4\n-0.6\n-0.8\n0\n30\n60\n90\n120\n150\n180\n210\n240\n270\n300\n330\n360\nLONGITUDE (E)\n1.2\n1.0\n(d)\n0.8\n0.6\n0.4\n0.2\n0.0\n-0.2\n-0.4\n-0.6\n-0.8\n-1.0\n-1.2\n-1.4\n-1.6\n-1.8\n-2.0\nC\n30\n60\n90\n120\n150\n180\n210\n240\n270\n300\n330\n360\nLONGITUDE (E)\nFig.6: As Fig.5, but anomalies from the four-year mean shown in Fig.5a. (a) 1950 experiment, (b) 1958, (c)\n1983, and (d) 1984.\n396","The Relationship of Rainfall Variability in East Africa\nto ENSO and the Atlantic and Indian Oceans\nS.E. Nicholson\nDepartment of Meteorology\nFlorida State University\nTallahassee, FL 32306\nB.S. Nyenzi\nDirectorate of Meteorology\nP.O. Box 3056\nDar es Salaam, TANZANIA\nEast Africa is an area of complex climates and highly variable rainfall.\nPast studies have shown a relationship between rainfall and both the Southern\nOscillation and sea-surface temperatures (SSTs) far to the west along the\nBenguela Coast. This paper further examines the association with SSTs in a\nlarger marine sector, encompassing parts of both the Atlantic and Indian\nOceans.\nSeveral characteristics of rainfall variability in East Africa have been\ndemonstrated. Interannual variability is strongly quasi-periodic, with fluctu-\nations occurring predominantly with time scales of 2.3, 3.5 and 5 years (Fig.\n1). The last is by far the strongest of the three periodicities. Despite a\nhigh diversity of mean climatic conditions, East Africa is markedly homo-\ngeneous with respect to interannual variability of rainfall and a single time\nseries suffices to describe the temporal fluctuations over the entire region\n(Fig. 2). Finally, the \"short rains\" of October-November-December (Fig. 3)\nproduce most of the interannual variability and account for the strong spatial\ncoherence of the rainfall fluctuations.\nThe time scale of 5 to 6 years is also the dominant one in the spectra of\nthe Southern Oscillation Index and of SSTs in the Atlantic and Indian Oceans\n(Fig. 4) Rainfall tends to be enhanced during ENSO years. However, in the\nfrequency domain the only coherent association between rainfall and ENSO is on\nthe Quasi-Biennial time scale of \"2.2-2.4 years.\nThe fluctuations with periods of 5 to 6 years appear to be associated\nwith SSTs in both oceans. This time scale is the dominant one through a\nsector encompassing the equatorial Indian Ocean and approximately 5°N to 30°S\nin the Atlantic Ocean (Fig. 5). Rainfall in East Africa varies coherently\nwith both oceans on this time scale. The strongest association is with the\nsoutheastern Atlantic (especially the Benguela coast) and with the Somali\ncoast in the Indian Ocean (Fig. 6)\nWe have also examined the overall pattern of SSTs for the five wettest\nrainy seasons and five driest rainy seasons (Fig. 7) Two analyses were\ncarried out: one for the \"long\" rainy season of April-May-June and one for\nthe \"short\" rainy season of October-November-December The antecedent pat-\nterns of SSTs during the three previous seasons were also derived for both\ncases. These are depicted in Fig. 8 as the SST difference between the five\nwet cases and five dry cases.\nFor both the long and short rains, significant SST anomalies are evident\nduring the season of abnormal rainfall. However, the pattern for the long\nrains is nearly opposite that for the short rains. For the April-to-June\n397","season, the wet minus dry composite (Fig. 7) shows negative SST differences in\nthe south Atlantic to -30°N, positive differences from there to 40° S. This\npattern is reversed in the Indian Ocean from 10°S to 40 S. In the areas of\ncoastal upwelling (Benguela and Somali coasts) SST differences are negative,\nimplying greater upwelling during wetter seasons. The wet-minus-dry compo-\nsites for the October-to-December season show instead positive SST differences\nthroughout most of the South Atlantic (to 35°S), in areas of coastal\nupwelling, and in the equatorial Indian Ocean. Large negative differences are\nconfined to 35-40°S in the Atlantic and 20-35°S in the Indian Oceans.\nboth the long-rain and short-rain cases, these patterns were evident to some For\ndegree one season in advance.\nPerhaps the most interesting results of the wet-minus-dry composites are\nthe SST patterns which were established two and three seasons prior to\nrainy season. Unlike the simultaneous SST patterns, the antecedent the\nare the same for both the long rains and short rains and anomalies are patterns\nsimilar 15°S in the Atlantic and Indian Oceans. SSTs are generally lower quite from\nlow to 35°S prior to seasons of abnormally high rainfall than prior to\nward, of rainfall. Positive differences are apparent both poleward and those\nforecasts especially two seasons prior to the rainy season. Thus, long-range equator-\nmight be realized on the basis of these antecedent patterns.\n200\nPERIOD IN YEARS\n200\n50\n10\n7\n5\n3\n2.5\nZONE 1\n2\n1.5\nZONE 2\nSPECTRUM\n100\n100\nEast Africa\n.04\n0\n0\n5.6 yrs\n-100\n-100\n.03\n2.2 yrs\n99% CL\n-200\nn\n-200\n3.6\n1931\n1941\n1951\n1961\n1971\n1981\n1931\n1941\n1951\n1961\n1971\n1981\nyrs\n200\n200\n95% CL\nZONE 3\nZONE 4\n.02\n100\n100\n0\n.01\n-100\n-100\n-200\n-200\n1931\n1941\n1951\n1961\n1971\n1981\n.05\n1931\n1941\n.10\n.15\n1951\n.20\n1961\n1971\n1981\n.25\nFREQUENCY\nYEARS\nYEARS\n2. Time series of normalized rainfall\ndepartures 1901-1984 for East Africa\n1. Normalized spectrum of rainfall\na whole (Kenya, Uganda, Tanzania; as\nvariability in East Africa, 1901-1984.\ndashed line), and for four climatically\nhomogeneous sectors within East Africa\n(units: 100 = one standard deviation).\n398","OND\nJAS\nAMJ\nJFM\n25\n25\n35\n50\n35\n25\n35\n50\n25\n25\n35\n25\n35\n25\n50.\n25\n50\n35\n35\nOND\nJAS\nJFM\nAMJ\n4.70\n50\n50\n50\n50\n50\n.70\n70\n50\n50\n70\n50\n.50\n50\n50:\nis\nhis\n70\n70\n50\n50\n70\n70\n50\n50\n70\n50\n50\n70\n70\n50\n50\n3. Top: Seasonal contribution (%) to the mean annual rainfall (stipling, 35\nto 50%; shading, >50%). Bottom: Correlation between seasonal and annual\nrainfall departure series, multiplied by 100 (approximately equal to the\nsquare root of the annual variance explained by each season; stipling, 50 to\n70%; shading, >70%).\nPERIOD (YEARS)\nPERIOD (YEARS)\n1\n50107543252\n32.5\n1.5\n1.5\n0.5\n10\n5\n.15\n12%\nSSTA EDF1\nSOI\n(ATLANTIC)\n5-6yrs\n.12\n99%\n3.5yrs\n.09\n95%\n6%\n2.3yrs\n.06\n.03\n.00\n.20\n.25\n.00\n.05\n.10\n.15\no\n4. Spectra of the Tahiti-Darwin - Southern Oscillation Index and of the first\nprincipal components of SST fluctuations in the Atlantic and western Indian\nOceans (analysis domain is shown in Fig. 5) .\n399","0\n20\n3.5\n3.5\nSomali coast\n10\n0\n(north)\n5\n5\n5\n(south)\no\n0\n5\n5\n10°\n-\n5\n3.5\n-20\nBenguela\n5\n5\ncoast\n3.5\n30\n3.5\n-\n5\nSE Atlantic\n5. Location of Atlantic and Indian Ocean sectors for which co-spectra and\ncoherence-squared have been derived between SSTs and East African rainfall\n(numbers denote the period in years of the dominant spectral peak in each\nsector).\nSE Atlantic\nBenguela coast\nSomali coast\nSomali coast\n(north)\n(south)\n:.0\nDEGREES OF FREEDOM 10.0\n.9\nRECREES OF FREEDEM 10.0\nDEGREES OF FREEDOM 10.0\n96- PERCENTLY\nDECREES OF FREEDOM 10.0\nPERCENT\n96- PERCENT\n95-PERCENT\n.8\n.7\n.6\n.5\n.08\n.07\n.06\n.05\n.04\n.03\n.02\n.01\n.00\n20\n10\n7\n5\n4\n32.5\n2\n20 10 7 5 4 3 2.5 2\n20 10 7 5 4 3 2.5\n2\n20\n10\n7\n5\n4\n3\n2.5\n2\nYEARS\nYEARS\nYEARS\nYEARS\n6. Coherence-squared between East African rainfall and SSTs for the period\n1946-1979 and SST spectra for two Atlantic sectors and two Indian Ocean\nsectors (see Fig. 5 for location).\n400","LONG RAINS\nAMJ\n+4:\n4\n-4\n4:\n8\n4\nSHORT RAINS\nOND\n4\n4\n4\n4\n7. Composites of the SST difference between the five wettest rainy seasons\n.\nand the five driest. Top diagram is based on the \"long rains\" and is wet-\nminus-dry for the April-May-June season. Bottom diagram is based on the\n\"short rains\" and is wet-minus-dry for the October-November-Decembe - period.\nUnits: tenths of degrees centigrade.\n401","(2 SEASONS PRIOR)\n(1 SEASON PRIOR)\n(3 SEASONS PRIOR)\nAMJ\nJAS\nJFM\na. The \"long-rain\" case. b. The \"short-rain\" case. Units: tenths of degrees centigrade.\n8. Wet-minus-dry SST composites for the three seasons prior to the rainy season.\n80\nb.\n(3 SEASONS PRIOR)\n(2 SEASONS PRIOR)\n4\n(1 SEASON PRIOR)\nWHITE\nOND\nJAS\nJFM\nC4\n4\n::\n80\na.","SPECIFYING MONTHLY MEAN ANOMALIES OF FIRE-WEATHER\nELEMENTS IN THE UNITED STATES\nWilliam H. Klein and Bruce T. Whistler\nDepartment of Meteorology\nUniversity of Maryland\nCollege Park, MD 20742\nThe U. S. Forest Service needs to know the potential for wildland fires several weeks in\nadvance in order to deploy personnel and equipment in the most effective manner. They\nwould therefore like to have monthly fire-weather outlooks for the entire nation during each\nmonth of the year. The weather elements of greatest importance to fire control are\nprecipitation, temperature, humidity, and wind speed. These elements can be combined into\nvarious measures of fire behavior such as the Burning Index and Fire Danger Rating Index.\nHere we discuss the derivation and properties of multiple regression equations for specifying\nmonthly mean anomalies of precipitation frequency and midday temperature (T), dew point\n(D) and wind speed (S) in the contiguous United States. These specification equations are\nbeing applied to prognostic monthly mean 700 mb height anomalies routinely produced by the\nClimate Analysis Center of the National Weather Service to yield monthly outlooks for each\nfire weather element in map form. These outlooks can then be combined to compute values of\nvarious fire weather indexes.\nMonthly mean anomalies of T, D and S observed about 1 p.m. local standard time at\n116 surface stations (Fig. 1) from 1964 to 1983 were computed from daily data compiled by\nthe Forest Service in East Lansing, MI. These data served as the predictands (dependent\nvariables) for this study. The potential predictors (independent variables) were concurrent\nmonthly mean 700 mb height anomalies observed at 121 grid points in North America and\nsurrounding oceans (H) (Fig. 2). The data were processed by a forward selection (screening)\nprogram, with statistical significance determined by a Monte Carlo technique. In order to\nincrease sample size, each month's data was pooled with its two adjacent months during the\nscreening process. This procedure yielded multiple regression equations for specifying T, D,\nand S, respectively, at each city from the field of simultaneous 700 mb H, plus the previous\nmonth's local value of that weather element. In addition, specification equations were derived\nfor the number of rainy days per month in anomaly form (PF) in each of 60 climate divisions\n(CDs) across the nation as a function of 79 values of H (a subset of Fig. 2).\nThe statistics involved in selecting a sample equation are illustrated in Table 1 for\ndewpoint during winter months at Charlotte, NC (35°N, 81°W). The first predictor selected is\nH at 35°N, 75°W, just east of Charlotte, with an RV of 71% and an F of 142.4. The second\nvariable is H at 20°N, 110°W, well southwest of Charlotte, which added 5% to the total RV\nwith an F of 12.0. The third predictor is H at 55°N, 115°W, northwest of the station, which\nadded 3% to the RV. The positive sign of the regression coefficient for the first predictor, in\nconjunction with negative signs of the second and third predictors, indicates that high (low)\ndewpoints are favored by anomalous flow of moist (dry) air from the south (north). The\nselection process was stopped at this point, with a standard error of 2.4°F and 79% of the\nvariance explained by only three terms, because no other predictor made good synoptic sense\nor produced an F-value in excess of the critical F.\n403","Nationwide means were computed by month for both climatological and specification\nstatistics. Fig. 3 shows the annual cycle of the reduction of variance (RV) for T, D, and S.\nAlthough T exhibits higher values than D and S each month, the difference in RV between T\nand D is small in January, when both elements are well specified by the large scale 700 mb\ncirculation. However, the difference is large in July, when D is sensitive to boundary layer\nphenomena and small scale factors (e.g., soil moisture and evaporation) which are not well\nrepresented by the large scale 700 mb pattern. Generally, S is poorly specified for all months.\nFig. 4 shows the annual cycle of the average RV for PF (dashed) and precipitation amount\n(PA) (solid) for the same 60 climate divisions and period of record. PF outperforms PA each\nmonth, with greatest difference in the summer when precipitation is difficult to specify\nbecause of localized convective activity. Fig. 5 shows the annual cycle of the SE for T, D, S,\nand PF. T, D and S follow an approximately sinusoidal curve, with maximum values in winter\nand minimum values in summer, while the SE of PF is highest in spring and lowest in fall.\nMaps were analyzed for the geographical distributions of each statistic for each month.\nFigs. 6 and 7 show that the values of RV for T range from a maximum of 80% over central\nCalifornia in July to a minimum only 30% over the same area in Jan. This marked diminution\nof RV during winter can be attributed to frequent fog and stratus clouds that reduce surface\ntemperatures in inland valleys under warm ridges aloft. Also notable is a marked decrease in\nRV from 80% in winter to 40% in summer in Florida. This decrease can be explained by the\nhigh frequency of cooling sea breezes and cumulus clouds in July without any effect at\n700 mb. Figs. 8 and 9 show the geographical distribution of RV for D during Jan. and July.\nThe RV ranges from 80% in winter in Florida and the coasts of South Carolina and Georgia to\n20% in summer in a band extending southeastward from Utah through New Mexico, Texas,\nLouisiana, and Mississippi to the coast of Alabama. A marked decrease in RV is apparent\nfrom Jan. to July in the Atlantic and Gulf states, with greatest change in Georgia (80% to\n20%). As with temperature in Florida, this decrease is probably caused by the high frequency\nof sea breezes and convection in summer.\nOn an overall basis, our specification equations explain approximately 68% of the\ntemperature variance, 57% of the dewpoint variance, 41% of the precipitation variance, and\n37% of the wind variance by means of 2-4 variables. However, the equations show marked\nmonth-to-month and area-to-area variations in their properties. In general, specification\nstatistics for D exhibit similar broad scale features as those for T and PF. However, PF and D\nare smaller in scale and more localized than T, especially in summer. Also in summer, the RV\nof both T and D generally increases with increasing latitude.\nACKNOWLEDGMENTS\nThis research was supported by the U. S. Department of Agriculture, Forest Service,\nunder Cooperative Agreement No. PSW-87-0023CA. We thank Francis Fujioka, Morris\nMcCutchan and Bill Main of the Forest Service, Don Gilman of CAC, and Ferd Baer, Hal\nBloom and Runhua Yang of the University of Maryland for help in various aspects of this\nproject.\n404","us DEPARTMENT or COMMENCE\nMCATHER DUREAU\nsee was wa\nnew\n&\nGEO\nOLM\nMSO\nVKM.\nGTF\nPOX\n@\nANL\nPW\nPOT\nBIL\nMFR\nDIS\nDLH\nSS\nSM\nCO\no\nFAR\n.\nBOI\nn\nALT\n40\nORM\nRBL\nSIIR.\nSYA\nPIH.\nMSP\ne\nGRO/\nBDL\n.\nLND\nRAP'\nOUF\nRGM\nLSE\nFNT\nCPR\nFSD\nSAC\nADCREWA\no\ns\nRNO\n2\nOFF\nsux\nCAI\nACY\nARO\nCYS\nSAN\n.\nFAT\nDSM\n3\nDWI\nELY\n.\nLBF\nMLI\nBLC\nCMH\n.\n.\nGAII\nIND\nCRWEKII\nGJT'\nWEN\n\"\nCVG\nSPI\nINC\n.\nOLD\ns\nDFL\nTOP\nLAS\nSTL\nLEX\nROA\n.\nEVV\nDDC\nLAX\nICT\nROU\nTAI\n.\nCLT\nTYS\nDNA\nSAN\nADQ\nTUL\nGSM\nFSM\no\nPHX\nOKC\nGMA\nPEN\nCAF\nc\nLIT\nUSV\nATL\nCH:\n3\nSP\nTUS\nBHM\n100\nMCN\nCAY\nMGM\nELP\nJAN\nME!\nSHV\na\nAC\nSAX\nSJ:\n1100\nTW\na\nSTR\nSAT\nPA\n25\n116 Surface Stations\n0\nfor Fire Weather\nBRO\n0.\na\n93\nis\n037742 ****\n115\n110\n105\n1W\n-\nFig. 1.\nLocations and call letters of the 116 surface stations for which temperature,\ndewpoint and wind speed were available during the 20-year period: 1964-1983.\n50\nor\n12.5? 51\nX\n0 :\n50\nOF\n3\n6\n110\n120\na\na\n70-\na\nB\n11\n⑇\nG\n7\na\n.\na\n2\ne\na\nd\n23\nI\nto\ne\nC\n3\nS\ne)\n16\ns\n19\nSo\nB\na\n0g\na\n&\na\n0\n3\no\n&\n3\na\n3\n.50\n500\n@\nQ\n140\n130\nB\n0\nID\n0\nR\n0\n&\n40\nx\n(\n3\n30\n50\n3\n20\n140\n120\n130\nNetwork of 121 grid points used to delineate the field of 700 mb heights for\nFig. 2.\nspecification of monthly mean temperature, dewpoint and wind speed.\n405","Table 1.\nSummary of the regression statistics obtained by screening monthly mean\nsurface dewpoint at Charlotte, NC (35 N. 81 W) during Dec. Jan. and Feb\n1964-1983, as a function of the concurrent field of 700 mb height anomalies\nThe critical F-values were determined by a Monte Carlo procedure The final\nequation selected for use in specification is starred.\nGrid. point\nReg. coeff.\nAdded\nTotal\nClassical\nProb\nCritical\nSE\nStep\nfor height\non entry\nRV (3)\nRV(8)\nF-value\nlevel (8)\nE-value\n(°F)\n1\n35°N,75°W\n+0.123\n71.1\n71.1\n142.4\n100\n12.6\n2.9\n2\n20°N,110°W\n-0.095\n5.1\n76.2\n12.0\n99.9\n8.3\n2.6\n3*\n55°N,115°W\n-0.029\n2.9\n79.1\n7.7\n99.2\n6.7\n2.4\n4\n20°N,130°W\n+0.034\n1.1\n80.2\n3.0\n91.0\n6.2\n2.4\n5\n30°N,90°W\n+0.051\n1.0\n81.2\n2.7\n89.4\n5.6\n2.3\nL TEMPERATURE\nDEWPOINT\n75\nO\nWIND SPEED\n4\n70\n0\n6 6\n60\n65\n50\n45\n40\no\nc\nO\no\no\n36\n30\nJAN FEB MAR APR MAY JUN JUL AUG SEP OCT NOV DEC JAN\nMONTH\nFig. 3.\nReduction of variance (%) of specification equations, based on 700 mb heights\nplus local persistence, for midday mean temperature, dewpoint and wind speed,\naveraged over 116 surface stations by month.\n406","52\n50\n48\n46\n44\n42\n40\n38\nPF\n36\n34\n32\nP A\n30\n_...\nD\nJ\nF\nM.\nis\nM\nJ\nJ\nA\nS\no\nN\nD\nJ\nF\nMonth\nReduction of variance (%) of monthly precipitation specification equations,\nFig. 4.\naveraged over 60 climate divisions, as a function of concurrent 700 mb height\nanomalies for precipitation frequency (PF) and amounts (PA) during 1951-1980.\n2.0\n2.0\n2.7\n2.4\n1.6\n1.4\n2.0\n1.2\n1.1\n3.0\n1.0\n1.4\n1.7\n1.0\n,\nWIND SPEED\n1.2\n..\n1.0\n0.9\n..\n1.00\n1.40\nPRECIP FREQUENCY\n1.40\n1.43\n1.40\n1.20\n1.36\n1.34\n1.32\nJAN FEB AUG NOV DEC JAN\nFig. 5.\nStandard error of estimate for specification equations for anomalies of monthly\nmean temperature, dewpoint, wind speed and precipitation frequency.\n407","100\n90\nIN\n110\n80\n10\n65\nw\nU.S. DEPARTMENT OF COMMERCE\nWEATHER BUREAU\nTRUE SCALL 122,500,000 AT ur.com\n70\nPOLICE STEALOGRAPHC PROJECTION\n60\n80\n0\n80\n50\n50\n50\n.\n.\nH\n70\n60\n70\n60\nw\nw\no\nL\nHh\n60\n7,5\n60\n60\nL\n70\n60\n40\n(30\n70\n80\nL\n80\n25\n40\nH\nH°\n0\n60\n80\n60\nx\nL\n.\nH\n60\n.\nL\n,\nb\no\n80\n25\nH\n.\n70\n80\nReduction of Variance\n70\n60\n60\nTemperature (Jan)\n80\n0.\n,\n50\n50\n115\n1:0\n105\n100\n95\n90\n85\n&\nUSCOUN.WB.DC 75\nFig. 6.\nSpatial distribution of reduction of variance (%) during winter months for midday\nmonthly mean temperature at 116 surface stations (solid circles).\n110\n100\nso\n80\n70\n65\nw\nU.S. DEPARTMENT or COMMERCE\nWEATHER BUREAU\nnot sell 122 500 000 41 LAT wk\nnew\n60\n70\n80\n70\nA\n50\na\n70\n@\nH\nL\nH\n50\n60\ne\n.\no\nH\n80\nH\n50\n60\n.\n50\n.\n60\nH\nL\n.\n@70\n70\nH\n50\n60\n50\n70\n50\nL.\n.\n50\n60\nH\n.L\n70\n60\nd\n.\nH\nH\n60\nC\n60\nL\nis\n40\n40\nReduction of Variance\n50\n50\nTemperature (July)\nC.\n60\n60\n113\n:\nis\n100\nvi\nes\n-\nFig. 7.\nSame as Fig. 6 but for summer.\n408","110\n100\nso\nw\n10\nw\nw\nU.S. DEPARTMENT OF C JUCHCE\n150\nWEATHER BUREAU\n70\n60\nTADE SCALL LAT wa\nNUR POLICITION\n45\nis\n50\nL\n70\n60\nL\n60\n60\n60\n60\n70\nw\nH\nw\n70\n70\nH\n60\n75\n0\nH\n@\n6.1\n/9\nH\n50\nL\nN\ne\n60\n50\n80\n.\nH\nH\n35\n35\n40\no\n75.\n70\n50\nH\n50\nH\n2\n60\n80\n0\nH\n60\n25\n50\n60\n25\n70\nReduction of Variance\n50\nDewpoint (Jan)\n40\n40\n0.\n:\n:10\n105\n100\n95\n90\nas\nw\nUSCOMM.WB:5C\n75\nFig. 8.\nSame as Fig. 6 but for dewpoint.\nN\nis\nas\nw\nU.S. DEPARTMENT or COMMERCE\n19\nWEATHER BUMCAU\n60\nH\nTOUL SCALL wa\n50\nPULL SECURITY PROCECTION\n60\n40\n40 50\n50\nH\n50\n40\n40\n30\n60\n50\nH\nL\n40\n30\n50\nL\nL\n50\n30\nH\n30\nL\nH\n20\n50\n35\n50\n301\n0\nL\n20\nH\n50\nH\n60\n40\nH\n40\n3\n60\nL\n50\n30\n20\n40\n20\nL\n30\n40\n30\nH\nLn\n20\n25\n30\n30\nReduction of Variance\nH\n20\nDewpoint (July)\n0.\n30\n1:3\n110\n105\n100\n95\n90\nas\nto USCOMM.WS.DC\nis\nSame as Fig. 7 but for dewpoint.\nFig. 9.\n409","LONG-RANGE FIRE WEATHER FORECASTING\nFrancis M. Fujioka\nMorris H. McCutchan\nPacific Southwest Forest and Range Experiment Station, Forest Service\nU.S. Department of Agriculture, Riverside, CA 92507\n1. Introduction\nForest managers have a keen sense of appreciation for weather, particularly in fire\nmanagement. A sound fire management program anticipates changes in fire potential\ndue to weather, usually through a detailed 24-hour forecast for an area on the order\nof 103 km². With ever increasing suppression costs, managers seek innovative ways of\ntackling persistent fire problems. One solution is to share firefighting resources. But federal\nagencies require lead times of 30 days or more, to identify national firefighting resources,\nrequest supplemental firefighting funds, and boost national fire prevention efforts. The\nconventional short-range fire weather forecast does not provide information for this level\nof planning. Instead, a long-range fire weather forecast system is needed, to support\ncoordinated national planning. Forest Service, National Weather Service, and university\nscientists are jointly developing a system to make 30-day forecasts of fire potential.\nThe objective of this forecast system is to describe mean monthly afternoon weather\nconditions for the United States. The fire emphasis requires consideration of local af-\nternoon dry-bulb temperature, relative humidity, windspeed, precipitation amount and\nfrequency (defined as the number of days in a month that at least 2.54 mm of precipita-\ntion occurred). Subsets of these basic variables can be used to compute a variety of fire\nweather indices. A study showed that correlations between fire weather indices and fire\nactivity were time-dependent, and no single index was uniformly highly correlated with\nfire activity (McCutchan and Main, 1989). Indices that are most highly correlated with\nfire activity at any given time are candidates for the long-range fire weather forecasts.\n2. Forecast Method\nSpecification equations were developed to make extended forecasts of the anomalies\n(departures from the monthly mean) of the following fire weather variables observed at\napproximately 1300 Local Standard Time (Klein and Whistler, 1988; 1989):\nDry-bulb temperature\nDewpoint temperature\nWindspeed\nEquations were derived for each fire weather variable, for each month, at 116 Na-\ntional Weather Service (NWS) stations (fig. 1). Using U.S. climate divisions delineated by\nEnglehart and Douglas (1985), equations were also developed for precipitation forecasts:\nNumber of days that daily precipitation exceeded 2.53 mm (precipitation frequency)\nPrecipitation amount\nThe forecast equations were derived by the \"perfect prog\" method, a regression pro-\ncedure in which observed values of predictors and predictand for a long period of record\n410","(20 years, in this study) are used in the development phase, but forecast values of most of\nthe predictors are used in application:\n(1)\nin which f(x,t+1) is the forecast surface weather anomaly for location x and month\n1, f(x,t) is the observed surface weather anomaly for month t at location x, and\n{Api} are forecast 700 mb height anomalies at designated gridpoints for month t + 1.\nThe {ai} are parametric constants determined by regression analysis. The constant ao is\nsometimes 0, which then implies that persistence is not a strong component. A screening\nprocedure is used to select the kx gridpoints (usually two to four), from a variable spacing\nNorthern Hemisphere grid which contained 180 gridpoints. The equations are applied to\nmonthly mean 700 mb height anomalies forecast by the National Oceanic and Atmospheric\nAdministration (NOAA) Climate Analysis Center. To obtain the predicted monthly field,\nthe climatological means are added to the forecast anomalies (fig. 2).\nThe forecast fire indices are calculated from the forecast weather variables. This\nunfortunately may introduce a bias, when the index is a convex or concave function of the\nfire weather variable(s). With a convex (concave) function, the index may underestimate\n(overestimate) the actual monthly mean, when the function is evaluated with the actual\nmean of the independent variable(s). For example, if a daily index is a convex function g\nof a weather variable S, the mean of the daily values over a month will be at least as great\nas the value obtained by evaluating g for the monthly mean of S:\nin which E is the expectation operator. These systematic deviations are explained by\nthe Jensen inequality (Fujioka, 1985). Equality obtains in the unlikely event that S stays\nconstant over the month. The system is sensitive to predictions biased toward low estimates\nof fire potential.\n3. Forecast Normalization\nThe forecasts are normalized, because fire managers are as much interested in the\nrelative severity of the forecast conditions, as in the forecast itself. Moreover, without\nnormalization, indices of varying scale and location parameters cannot be compared. The\nforecast is normalized by climatology, as follows. The empirical distribution function (Hogg\nand Tanis, 1977) for each variable, station, and month is determined for each of the 116\nstations in the fire weather network. This function, call it F, is the nondecreasing step\nfunction in the interval (0,1), obtained by sorting the n data points in the period of record\nin ascending order, and adding the increment 1 to F at each data point (fig. 3). For\nn\nthis study, the empirical distribution functions were determined for each month from 20\nmonthly values obtained in the period 1964-1983, except for the precipitation elements,\nwhich were derived from 30 monthly values in the period 1951-1980.\n411","GEG\nDWM\nINL\nOLM\nGTF\nSSM\nVKM\nCON\nDLH\nMSO\nBIS\nALB\nPDX\nPDT\nFAR\nORH\nBIL\nSYR\nGRB\nMSP\nBUF\nPVD\nBOI\nRAP\nBGM\nSHR\n.\nMFR\nBDL\nLSE\nFNT\nABE\nFSD\nCPR\nEWR\nPIH\nSUX\nERI\nCAK\nLND\nRBL\nBFF\nBWI\n.\nACY\nCMH\nSBN\nLBF\nRFD\nCYS\n.\nGRI\nDSM\nCRW\nMLI\nSAC\nCVG\nRIC\nRNO\nSLC\nIND\nDEN\nGJT\nEKN\nELY\nSPI\nEVV\ne\nGLD\nROA\nTOP\n@\nFAT\nSTL\nTRI\nLEX\nRDU\nDDC\n.\nICT\nLAS\nCLT\nTYS\nILM\nBNA\nBFL\n.\nGSP\nTUL\nCAE\nABQ\nFSM\nMEM\nLAX\nAMA\nHSV\nATL\n.\nOKC\nCHS\nLIT\nSAN\nPHX\nMCN\nLBB\nBHM\nMEI\nSPS\nSAV\nTUS\nSHV\nMGM\nJAN\nJAX\nELP\nSJT\nTLH\nACT\nBTR\nMOB\nTPA\n116 SURFACE STATIONS\nSAT\nMIA\nFOR FIRE WEATHER\nBRO\nFigure 1: Locations and call letters of the 116 National Weather Service stations\nfor which 30-day fire weather forecasts are prepared.\n40\n30\n40\n50\n25\n60\n50\n60\n73\n70\n70\n30-Day Forecast\n1300LST Temperature °F\nNovember 1989\nFigure 2: Thirty-day forecast of mean November 1989 afternoon temperatures\nin the contiguous U.S.\n412","BAKERSFIELD, CA - NOVEMBER\n100\n80\n60\n40\n20\n0\n55\n60\n65\n70\nTEMPERATURE °F\nFigure 3: Modified empirical distribution function of mean November 1300 LST\ntemperature (°F) for Bakersfield, California, based on 1964-83 data. The mod-\nification was made by connecting the data points with straight line segments,\nwhich yields a piecewise linear function called an ogive.\n45\n45\n30 - Day Forecast\n1300LST Temperature Percentiles\nNovember 1989\nXXXXXXXXXX\n>97%\nVIII 90 - 96\n******\n45-89%\n<45%\nFigure 4: Normalized 30-day forecast of mean November 1989 afternoon temper-\natures in the contiguous U.S. The contoured values are percentiles p, obtained\nfrom the ogive at each of the 116 NWS stations.\n413","Having determined F, the forecast values may be normalized by the scale- and\nlocation-invariant transformation:\nP = 100F(f),\n(2)\n0    100\nwhere f is a forecast value obtained from the process described above, and p expresses the\nproportion of the climatological data which the value f equals or exceeds.\nThe values of p (percentile) may be plotted on a map and contoured. For the fire\npotential graphics, it is necessary to have increasing values of p signify increasing fire\npotential. This is clearly not the case for precipitation, for which increasing values are\nassociated with decreasing fire potential. In such cases, we plot 100 - p. Moreover, pre-\ndesignated values of p may be used to delineate fire potential classes. Under one such\nclassification, p 97 is considered extreme fire danger, p 90 is very high, and p 45 is\nhigh (fig. 4 is an example of a normalized temperature forecast). The normalization pro-\ncedure hence rates the forecast values relative to climatology, and in this sense, different\nindices {fi} may be compared, after the transformation is applied.\nAs of September 1989, the NOAA 700 mb monthly height anomaly forecast and pre-\nvious monthly means for the 116 stations are imported twice monthly to a Macintosh II\nmicrocomputer, which in turn computes monthly forecasts of the fire weather elements\nlisted above.\n4. Further Research\nThe complete fire potential forecast system that we envision will integrate weather\ninformation and the fire management parameters relevant to agency decision processes.\nIn order to make judicious use of the forecast information, the fire manager requires an\nassessment of forecast reliability. With some probabilistic measure of forecast accuracy,\nwe intend to provide the user with information needed to evaluate the forecast condition\nand the forecast itself in a decision analytic context (see, for example, Brown and Murphy,\n1988). Work in this area is in the incipient stages. At present, we can provide the user an\nindicator of forecast quality, with maps of the reduction of variance statistic obtained in\nthe perfect prog process.\nThe extended-range forecast equations were developed from first-order NWS station\ndata. The locations therefore represent largely airports and major cities. While this\nmethod may sufficiently capture the large-scale features of the expected weather patterns,\nadditional research is needed to improve the spatial resolution of the forecasts. This could\nbe done by correlating spatial variations over the NWS network with variations in the\nfire weather network, which the current system does not use (e.g., McCutchan and Tsou,\n1989). Studies are underway to enhance the spatial detail of the fire potential forecasts in\nforests and wildlands.\nWork is also being done to develop corresponding systems for medium-range and sea-\nsonal fire weather forecasts, to expand the spectrum of planning horizons for fire managers.\nWith medium-range fire weather forecasts (Ueyoshi and Roads, 1989), we hope to have\ndaily resolution, which is necessary to describe the dynamic variations in fire potential.\n414","REFERENCES\nBrown, B.G. and A.H. Murphy, 1988: On the economic value of weather forecasts in\nwildfire suppression mobilization decisions. Can. J. Forest Res., 18:1641-1649.\nEnglehart, P.J. and A.V. Douglas, 1985: A statistical analysis of precipitation frequency\nin the conterminous United States, including comparison with precipitation totals.\nJ. Clim. Appl. Meteor., 24:350-361.\nFujioka, F. M., 1985: Estimating wildland fire rate of spread in a spatially nonuniform\nenvironment. Forest Sci., 31:21-29.\nHogg, R.V. and E.A. Tanis, 1977: Probability and Statistical Inference. Macmillan Pub-\nlishing Co., Inc. New York. 83-97.\nKlein, W.H. and B.T. Whistler, 1988: Specifying surface temperature, dew point, wind\nspeed and precipitation frequency for monthly fire-weather outlooks. Proc. 13th An-\nnual Climate Diagnostics Workshop, Cambridge, MA, Oct. 31 - Nov. 4, 1988, NOAA,\n204-209.\nKlein, W.H. and B.T. Whistler, 1989: Specifying monthly mean anomalies of fire weather\nelements in the U.S. Proc. 14th Annual Climate Diagnostics Workshop, La Jolla, CA,\nOct. 16-20, 1989, NOAA, in press.\nMcCutchan, M.H. and W.A. Main, 1989: The relationship between mean monthly fire\npotential indices and monthly fire severity. Proc. 10th Conf. on Fire and Forest\nMeteor., Ottawa, Ontario, Apr. 17-21, 1989, in press.\nMcCutchan, M.H. and T. Tsou, 1989: Spatial interpolation of fire potential forecasts in\ncomplex terrain of California. Proc. 10th Conf. on Fire and Forest Meteor., Ottawa,\nOntario, Apr. 17-21, 1989, in press.\nUeyoshi, K. and J.O. Roads, 1989: Use of numerical weather prediction models for\nmedium-extended-range fire weather forecasting. Proc. 14th Annual Climate Diag-\nnostics Workshop, La Jolla, CA, Oct. 16-20, 1989, NOAA, in press.\n415","Observed Fire Weather Variability\nKyozo Ueyoshi and John O. Roads\nScripps Institution of Oceanography\nUniversity of California, San Diego\nLa Jolla, CA 92093\nSuccessful management of forest resources often requires anticipating weather conditions\nweeks to months in advance. In a recent article Fosberg and Fujioka (1987) emphasized\nthe potential importance of accurate long-range weather forecasts for developing appropri-\nate responses to wild fires and described a national research plan to integrate present-day\nmedium/long-range numerical weather predictions in national and regional planning of forest\nfire management. As part of this effort, the goal of the work described herein is to assess\nthe capability of present-day operational numerical weather prediction models to forecast\nfire potential on a U.S.-wide scale.\nOur initial efforts were devoted to describing climatological and temporal variations in\nthe Fosberg Fire Weather Index (FWI) on a nation-wide scale (Ueyoshi and Roads, 1989).\nFor this purpose we utilized the National Meteorological Center (NMC) global objective\nanalysis of twice-daily temperature, relative humidity, and wind speed at 1000mb level,\nwhich is available on a 2.5° X 2.5° grid mesh for the period from 1 July 1976 to 31 December\n1988. Several major episodes of forest fires in California during the same period were found\nto be closely associated with positive anomalies of the standardized FWI from the NMC\n1000mb analysis. For example, the forest fire activity in Southern California during the latter\npart of November 1980 is clearly seen in the positive anomalies of the twice-daily patterns\nof the standardized FWI. However, since the FWI is calculated using meteorological data\nextrapolated to the 1000mb level, the analysis at this stage may not be appropriate in many\nregions of the U.S. where the ground surface is far from the 1000mb height.\nIn the second and current phase, we therefore are concerned with comparing the FWI\ncalculated from the NMC grid data with individual station observations on a nation-wide\nscale. This evaluation has been carried out using daily-mean surface observations from\nover 300 U.S. stations in the Summary of the Day First Order data file (TD-3210) from\nthe National Climatic Data Center for the period beginning 1 January 1984 and ending 31\nDecember 1987. Although unsmoothed and irregularily distributed surface data results in\nrelatively noisy patterns, the U.S.-wide spatial distributions of the overall (4-year) means of\nthe basic variables and the FWI are relatively similar to corresponding patterns of individual\nquantities obtained from the NMC grid data. A local maximum of the FWI is located over\nKansas, Oklahoma and Texas, and the FWI shows a strong to moderate dependence on wind\nspeed and relative humidity while its temperature dependence is weak. However, in order to\nattain an approximately normalized view across the U.S., it seems more appropriate to use a\nstandardized index for evaluating the climatological and temporal variations in the FWI. A\nstandardized monthly FWI at a station, for example, is obtained by dividing the departure\nof each monthly station average from the overall mean by the overall standard deviation\nappropriate to that station.\nAs was the case for the NMC 1000mb grid data, the standardized seasonal FWI also ex-\nhibits a clear dichotomy between the western and eastern parts of the conterminous U.S. (see\nFigs. 1 and 2). During the summer the FWI is high in the West and low in the East. A\nreversal occurs during the winter. However, relatively small local extremes appear in the\nspatial distributions of the standardized monthly-mean FWI from the TD-3210 data, and\nthe details of the two analyses are considerably different from each other.\n416","Daily variations of the nation-wide FWI during a recent major forest fire activity are\nexamined next. Shown in Figs. 3a and 3b are the standardized anomalies of the FWI from\nthe NMC 1000mb analysis at 00UTC of 6 and 7 September 1987, respectively. The positive\nanomalies of the FWI over Southern California correspond favorably to a major outbreak\nof forest fires in California during the first part of that month. On the other hand, the\ncorresponding standardized daily surface FWI anomalies, which are illustrated in Figs. 3c\nand 3d for 5 and 6 September 1987, respectively, do not seem to exhibit noticeable positive\ncorrelations to fire activity in California. One of the possible reasons for the difference is that\nthe FWI from the NMC analysis uses instantaneous values of the meteorological elements in\nthe afternoon, while the FWI from the TD-3210 data set is calculated using the daily means.\nA previous study has shown that correlations between fire weather indices and fire activity are\ntime-dependent, and no single index derived from the monthly-mean meteorological elements\nis uniformly correlated with fire activity (Fujioka and McCutchan, 1989). Our examination\nseems to indicate that the short-term FWI anomalies often depend primarily on time scales\nmuch shorter than a month, and even the use of daily-mean variables for calculations of the\nFWI may filter out significant positive correlations between the FWI anomalies and actual\nfire activity that might be apparent if instantaneous variables were used.\nTo illustrate long-term temporal variations of the FWI at individual locations, time series\nof the 30-day moving-averaged FWI at two NMC grid points are plotted in Figs. 4a and 4b\nover the entire data period. Without a rigorous mathematical analysis we detect relatively\nregular seasonal cycles in the FWI for a grid point located in the northeastern region of the\nU.S. (Fig. 4a). However, a sudden change in amplitude beginning in the middle of 1986 may\nbe related to the changes in objective analysis method adopted at NMC in May 1986. On the\nother hand, seasonal cycles in time series of the FWI at a grid point in Southern California\nare less obvious. We have found, however, a Santa Ana event in California during which\nthere is higher potential of major forest fires is often represented as a peak in amplitude of\nthe 30-day moving-averaged FWI (Fig. 4b).\nCorresponding time series of the 30-day moving-averaged FWI at two surface stations\nare shown in Figs. 4c and 4d. As it is evident in these figures, the absolute amplitudes of\nthe surface FWI are, in general, considerably smaller than those calculated from the NMC\n1000mb analysis. Moreover, unlike the above NMC case, the Santa Ana event in January\n1986 does not show up in the 30-day moving averaged time series shown in Fig. 4d; however,\nthere is some indication of the event that started on 29 March 1987. These differences may\nagain be due to the same reasons described earlier and also because the behavior of the\nsurface FWI time series varies significantly among the stations even within a relatively small\nregion such as a NMC grid mesh. In other words the time series shown in Figs. 4c and 4d\ndoes not necessarily represent typical behavior of the FWI in surrounding regions.\nIn summary, the surface FWI obtained from the TD-3210 station observations in present\nform seems to show no better skill to express spatial or temporal fire potential than the NMC\n1000mb FWI.\nThe third phase of this work is to assess the capability of present-day forecast models to\nforecast fire potential by utilizing the 108 30-day numerical forecasts from NOAA (DERF II\ndata set). This stage will be initiated as soon as the preliminary preparation of the data set\nis completed. We are also looking into the possible use of the NMC medium range forecast\n(MRF) model for fire weather forecasting.\n417","300\nNMC 2.5-DEG GRID DATA OBSERVATIONS\nNMC 2.5 DEG GRID DATA ORSERVATIONS\nX)\n100\nSTD FIRE WEATHER INDEX AT 1000MB\nSTD FIRE WEATHER INDEX AT 1000MB\nMONTH OF JAN\nMONTH OF APR\nLA\nTIME AVERAGED 1978/7 - 1987\nTIME -AVERAGED 1976/7 - 1987\nNO. OF OBSERVATIONS - 591\nNO. OF OBSERVATIONS 576\nCONTOUR INTERVAL - 0.1\nCONTOUR INTERVAL - 0.10\nMAX - 0.6 MIN - -0.4\nMAX - 0.462 MIN - -0.159\n100\nAPR\nJAN\nb\na\n10g\n201\n1001\nx\n⑆2\n0\n0\nis\n100\nA\n100\na\n300\n200\n200\n100\nas\nNMC 2.5-DEG GRID DATA OBSERVATIONS\nNMC 2.5 -DEG GRID DATA OBSERVATIONS\nN\nK\nSTD FIRE WEATHER INDEX AT 1000MB\nSTD FIRE WEATHER INDEX AT 1000MB\nMONTH OF OCT\nMONTH OF JUL\n200\nTIME AVERAGED 1976/7 1987\nTIME -AVERAGED 1976/7 1987\n200\nNO. OF OBSERVATIONS 692\nNO. OF OBSERVATIONS 669\n102\nCONTOUR INTERVAL - 0.10\nCONTOUR INTERVAL 0.10\nMAX - 0.280 MIN -0.367\nMAX - 0.337 MIN - -0.613\nJUL\nOCT\nd\nC\n100\n2\nand\n00\n85\n100\n@\n900\nboo\nuna\nSRO\nFig. 1. Seasonal variations of the standardized monthly-mean FWI obtained from the NMC\ntwice-daily 1000mb grid data: (a) for January; (b) for April; (c) for July; and (d)\nfor October. Solid-line contours indicate positive anomalies, while broken-line contours\nrepresent negative anomalies.\n418","TD-3210 SUMMARY OF DAY\nTD 3210 SUMMARY OF DAY\nFIRST ORDER\nFIRST ORDER\n1 JAN 1984 - 31 DEC 1987\n1 JAN 1984 - 31 DEC 1987\nSTD MEAN MONTH OF JAN\nSTD MEAN MONTH OF APR\nSTD FIRE WEATHER INDEX\nSTD FIRE WEATHER INDEX\nCONTOUR INTERVAL . 0.2\nCONTOUR INTERVAL - 02\n0\nMAX VAL - 0.81 MIN VAL -\n-1.11\nMAX VAL = 1.46 MIN VAL -\n-0.34\nb\nJAN\nAPR\na\na\nO\n0\no\n0\n4-OCT-89 17.15.4\n4-OCT-89 17.20\nTD 3210 SUMMARY OF DAY\nTD-3210 SUMMARY OF DAY\nFIRST ORDER\nFIRST ORDER\n18\n1 JAN 1984 - 31 DEC 1987\n1 JAN 1984 - 31 DEC 1987\nSTD MEAN MONTH OF JUL\nSTD MEAN MONTH OF OCT\nSTD FIRE WEATHER INDEX\nSTD FIRE WEATHER INDEX\nCONTOUR INTERVAL 0.2\nCONTOUR INTERVAL 0.2\nMAX VAL - 1.08 MIN VAL - -0.80\nMAX VAL = 0.98 MIN VAL = -0.73\nOCT\nd\nJUL\nC\n+\n6\n0\nC\n0\nD\n0\n4-0C1-89 17.37.48\nFig. 2. As for Fig. 1 except that the standardized monthly-mean FWI was calculated using\nthe daily-mean surface values of the TD-3210 station data.\n419","NMC 2.5 DEG GRID DATA ANALYSIS\nNMC 2.5 - DEG GRID DATA ANALYSIS\nMONSTD FIRE WEATHER INDEX\nMONSTD FIRE WEATHER INDEX\n1847500\n07-SEP 1987 00Z\n08-SEP-1 1987 00Z\nCONTOUR INTERVAL as 0.75\nCONTOUR INTERVAL as 0.75\nMAX VAL - 2.92 MIN VAL - -1.74\nMAX VAL - 4.82 MIN VAL - -1.57\n750,\nb\na\n750\n75a\n10\n1.50\n750\nCO\nTD-3210 SUMMARY OF DAY\nTD 3210 SUMMARY OF DAY\nFIRST ORDER\nFIRST ORDER\nMONSTD FIRE WEATHER INDEX\nMONSTD FIRE WEATHER INDEX\n6 SEP 1987\n7 SEP 1987\nCONTOUR INTERVAL 0.50\nCONTOUR INTERVAL = 0.50\nMAX VAL - 2.64 MIN VAL - -1.94\nMAX VAL - 2.15 MIN VAL - -2.66\nd\nC\ne\nO\n0\nC\n6\n0\n0\nof\n1-MOV-89 10.38/57\nFig. 3. The standardized anomalies of the FWI during a recent forest fire activity in Cal-\nifornia: (a) and (b) illustrate the results obtained from the NMC 1000mb analysis for\n00UTC of 6 and 7 September 1987, respectively, while (c) and (d) show the anomalies\ncalculated from the daily-mean surface values of the TD-3210 data for 5 and 6 September\n1987, respectively. The standardized anomaly was obtained by dividing the departure of\neach instantaneous or daily FWI from the overall monthly mean by the overall monthly\nstandard deviation appropriate to each grid point or station.\n420","80\n80\nFWI AT 1000 MB\nFWI AT 1000 MB\na\nb\nAT NHC 2.5 DEG GRID POINT (44.12) NE\nAT NHC 2.5 DEG GRID POINT (26. 91 CA\n30-DAY HOVING AVERAGE\n70\n30-DAY MOVING AVERAGE\n70\nJUL 1976 DEC 1987\nJUL 1976 DEC 1987\nMEAN\n18.4\nMEAN\n15.5\nSTD DEV\n14.6\n60\nSTD DEV\n9.7\n60\n50\n50\n40\n40\n30\n30\n20\n20\n10\n10\n0\n0\n76\n77\n76\n79\n80\n81\n82\n83\n84\n85\n86\n87\n76\n77\n78\n79\n80\n81\n82\n83\n84\n85\n86\n87\nYEAR\nYEAR\n80\n80\nTO3210 SURFACE FWI\nTD3210 SURFACE FWI\nSTATION NR 14739 BOSTON/LOGAN INT L ARPT\nMA AT 14222N 07102W)\nSTATION NR 23174 LOS ANGELES/INT L ARPT\nCA AT 13356N 11824W)\n30-DAY MOVING AVEGAGE\n70\n30-DAY MOVING AVEGAGE\n70\nI\nJAN 1984 31 DEC 1987\nd\nC\n1 JAN 1984 31 DEC 1987\nMEAN\n14.7\nMEAN\n10.1\nSTD DEV\n6.9\n60\nSTO DEV\n4.2\n60\n50\n50\n40\n40\n30\n30\n20\n20\n10\n10\n0\n0\n76\n77\n78\n79\n80\n81\n82\n83\n84\n85\n86\n87\n76\n77\n78\n79\n80\n81\n82\n83\n84\n85\n86\n87\nYEAR\nYEAR\nFig. 4. Time series of the 30-day moving-averaged FWI for individual locations: (a) and (b)\nare, respectively, the NMC grid points located at 42°30'N and 72°30'W in Massachusetts\nand at 35°30'N and 117°30' W in Southern California, while (c) and (d) are the TD-\n3210 surface stations located at Boston Logan International Airport and at Los Angles\nInternational Airport, respectively. The horizontal solid line indicates the overall mean\nFWI, while the dotted lines illustrate to limits. The arrows in (b) and (d) indicate three\nSanta Ana events started approximately on: (1) 26 November 1980; (2) 7 January 1986;\nand (3) 29 March 1987.\n421","Large-Scale Influences Vs. Land Surface-Atmosphere\nInteraction as Factors in Sahel Rainfall Variability\nS.E. Nicholson, A.R. Lare and H. Zheng\nDepartment of Meteorology\nFlorida State University\nTallahassee, FL 32306\nSevere drought has often plagued the West African Sahel. So far\nits\nmeteorological cause have remained elusive. In this region droughts (as well\nas wetter conditions) demonstrate an unusually high degree of persistence, on\nthe order of one to two decades. The present dry episode has lasted over\ntwenty years (Fig. 1). Numerous theories have been proposed to explain rain-\nfall variability in the Sahel. These can be broadly assigned to two classes:\nlocal land-surface forcing and large-scale atmospheric-oceanic forcing.\nBoth the long-term persistence and shorter year-to-year variability must\nbe explained; different causal mechanisms may be involved. Likewise, the\nfactors which initiate a dry or wet episode may not be the same ones which act\nto maintain the episode over several years.\nWe hypothesize that large-scale atmospheric and oceanic factors initiate\nsuch episodes and produce high frequency fluctuations (e.g., 2 to 5 years)\nwhile land-surface/atmosphere feedback acts to positively reinforce the condi-\ntions of anomalous rainfall, thereby creating the long-term (>10 years)\npersistence so apparent in the Sahel We are testing this by examining\nseparately the high and low frequency components of rainfall variability.\nThis paper presents some of our preliminary analyses.\nThree means are used to isolate the high frequency components. One is\nto\nconsider only the period 1920-49, during which no persistence is evident in\nSahel rainfall. Another is the use of high-pass-filtered time series. The\nthird is to examine the change in rainfall from one year to the next. These\nhave variously been related to the Southern Oscillation Index (SOI), ENSO\nevents and sea-surface temperatures in various sectors of the Atlantic Ocean.\nThe potential influence of land-surface processes is evaluated using the\nsurface energy balance \"climatonomy\" model of Lettau to simulate dry and wet\nconditions.\nThe spectra of rainfall in the Sahel and Soudan (just to the south) are\nshown in Fig. 2. Most of the variance is clearly at low frequencies, i.e.\nperiods of ten years or greater. Discrete high frequency peaks are also\nevident. Those at ~2.6, 3.6 and 5.6 years are noteworthy because they are\nstrongly apparent in rainfall elsewhere in Africa and in SSTs and the SOI.\nRainfall in the Soudan varies coherently with the SOI on all three time\nscales, but for the more northern Sahel, significant coherence is seen only at\n5.6 years (Fig. 3). In the high-pass filtered series (Fig. 4), and for the\nperiod 1920-49, these same three frequencies are clearly the dominant ones and\ncoherence with the SOI is also apparent.\nNevertheless no association between the SOI and Sahel-Soudan rainfall is\napparent in the time domain (Fig. 1). ENSO years show no preference for above\nor below normal rainfall. It is interesting to note, however, that there is a\nstrong tendency for ENSO years to be drier than the previous year (Fig. 5).\nThis might be interpreted to mean that ENSO acts to reduce rainfall from the\nprevailing mean and that the ENSO-induced fluctuations are masked in the time\ndomain by the dominant low frequency fluctuations.\n422","The high-pass-filtered series were compared with SST fluctuations in four\nregions of the Atlantic (Fig. 6) In the three sectors farthest south, the 5. -\n6 year time scale is the dominant one (Fig. 6), but in the Cape Verde sector,\nthe variance is instead concentrated at 2.3 and 3.5 years. The only strong\nassociation with Soudan rainfall is at 5-6 years with the Benguela coast (Fig.\n7). The Sahel shows a somewhat stronger association with SST and it is\napparent at both 5-6 and 2.5-3 years and, to a lesser extent, at \"14-16 months\n(Fig. 8).\nThe absence of associations between rainfall and ENSO or SSTs in the time\ndomain, but presence in the frequency domain, suggest that the large-scale\ninfluences are not the dominant factors in rainfall variability. The question\nremains as to the causes of the lower frequency fluctuations and the decadal\npersistence. The rest of the paper addresses the latter issue. In order to\ndetermine whether land surface - -atmosphere feedback plays a role in the persis-\ntence of drought, we have used the climatonomy model of Lettau to assess the\nchanges of surface energy and hydrologic balance parameters which accompany\nextreme changes in rainfall. We have used the model to simulate an idealized\ncase of the rainfall fluctuations which occurred in the Sahel between 1950 and\n1979.\nThe climatonomy model consists of three parts: shortwave radiation,\nevapoclimatonomy and thermoclimatonomy. The first computes absorbed solar\nradiation from insolation at the top of the atmosphere. The ground-absorbed\nradiation plus precipitation force the second submodel, which computes\nhydrologic parameters including soil moisture. The third part predicts the\nlatent and sensible heat fluxes.\nThe results are shown in Figs. 9 and 10. Soil moisture change by nearly\na factor of two between wet episodes and dry ones. The latent and sensible\nfluxes also exhibit a large change. When rainfall is reduced from 800 mm/year\n(1950-59) to 400 mm/year (1970-79), sensible heat flux increases from \"3.1 to\n5.3 MJ/m 2 day, while latent is reduced from 4.9 to 2.7 MJ/m day. This does\n2\nnot yet prove the role of land surface processes, but it does show the tremen-\ndous changes of surface energy balance which accompany anomalously wet or dry\nconditions.\n150\n100\n50\no\n-50\n-100\n- -150\n1981\n1931\n1941\n1951\n1961\n1971\n1901\n1911\n1921\n1. Normalized rainfall departures for the West African Sahel, 1901-1985;\nENSO years are shaded (units: 100 = one standard deviation).\n423","PERIOD IN YEARS\nPERIOD IN YEARS\n50\n10\n7\n5\n4\n3\n2.5\n2\n1.5\n50\n10 7\n5\n4\n3\n2.5\n2\n1.5\nT\nSPECTRUM\nSPECTRUM\nSahel\nSoudan\n.03\n99% CL\n.04\n95% CL\n.02\n.03\n99% CL\n-\n2.6-3.8 yrs\n95% CL -\n.02\n5.6 yrs\n.01\n2.6-3.6 yrs\n.01\n.05\n.10\n.15\n.20\n.05\n.25\n.10\n.15\n.20\n.25\nFREQUENCY\nFREQUENCY\n2 Normalized spectra of rainfall in the Sahel and Soudan zones, 1901-1973\n(see Fig. 6 for location).\nPERIOD IN YEARS\nPERIOD IN YEARS\n20\n10 7\n5\n4\n3 2.5\n2\n20 107\n5\n3 2.5\n2\nCOHERENCE-SQUARE\n.9\nCOHERENCE-SQUARE\n.9\nSOI/Sahel\nSOI/Soudan\n1.8\n.8\n.8\nBW\n.7\n.7\n.6\n.6\n5.0\n5.0 3.7 2.6\n.5\n.5\n3.6\n1.8\n2.2\n$1.4\n.4\nC2.6A\n.4\n.3\n.3\n.2\n.2\n.I\n.1\n.0\n.0\n.05\n.10\n.15\n.20\n.25\n.05\n.10\n.15\n.20\n.25\nFREQUENCY\nFREQUENCY\nCoherence-squared between the Tahiti-Darwin Southern Oscillation Index\n3\n( SOI) and Sahel and Soudan rainfall, 1935-73.\n424",".07\n.13\nSAHEL\n14 mo\n.12\nSOUDAN\n.06\n.11\n.10\n.05\n.09\n.08\n.04\n.07\n2.5 3 yr\n> 10yn 5\nV\n.06\n2.5-3 yr\nyr\n.03\n.05\n14 me\n5\n.04\nyou\n.02\n.03\n.02\n.01\n.01\n.00\n.00\n20 10 7 5 4 2.5 2\n20 10 7 5 4 32.5 2\nYEARS\nYEARS\n4. Normalized spectra of rainfall in the Sahel and Soudan zones; time series\nhave been high-pass - filtered to remove periodicities longer than 7 years.\n200\n150\n100\n50\no\n-50\n-100\n-150\n-200\n1901 1911 1921 1931 1941 1951 1961 1971 1981\n5. Time series of the difference in rainfall between each year, 1901-1985 and\nthe previous year. Units: 100 = one standard deviation. ENSO years\nare\nshaded.\n20'\n0\nSAHEL\nC.VERDE\nSOUDAN\n00\nGUINEA\nCOAST\nNE\nBRAZIL\nBENGUELA\n20\nCOAST\n6. Location of the Sahel and Soudan zones and four sectors of the Atlantic\nfor which SST time series are derived and compared with rainfall.\n425","SOUDAN/C.VERDE\nS QUDAN BENGUELA\nCOHERENCE SPECTRUM\nREGION 3/SST 2\nCOHERENCE SPECTRUM\n3/SST 8\nYEARS\nCoherence-squared between Soudan\nYEARS\n20 10 7 S 4 32.5 2\nrainfall and SSTs in the four Atlan-\n20 10 7 5 4 32.5 2\nREGION\n5yr\nPREECE 10.0\n10.0\nPERCE\ntic sectors shown in Fig. 6.\nCEORES\n1.0\n.9\n.8\n.7\n.6\n.5\n.4\n.3\n.2\n.1\n.0\n1.0\n.9\n.8\n.7\n.5\n.5\n.4\n.3\n.2\n.1\n.0\nSOUDAN/GUINEA COAST\nSOUDAN/NE BRAZIL\nCOHERENCE SPECTRUM\nREGION 3/SST 1\nCOHERENCE SPECTRUM\nREGION 3 /SST 5\nYEARS\nYEARS\n20 10 7 54 32.5 2\n2\n32.5\nDECALES FREEDOM 12.0\nCLERED or FREEDOM 10.0\n4\n8.\n5\nPERCENT\nPERCENT\n20 10 7\n1.0\n.9\n.8\n.7\n.6\n.5\n.4\n.3\n.2\n.1\n.0\n1.0\n.9\n.8\n.7\n.6\n.5\n.4\n.3\n.2\n.1\n.0\nBENGUELA\nSAHEL/C.VERDE\nCOHERENCE SPECTRUM\nCOHERENCE SPECTRUM\nREGION 1/SST 2\nREGION 1/SST 8\nSAHEL\nYEARS\n7. Coherence-squared - between Sahel\nrainfall and SSTs in the four Atlan-\n2.5-3\nYEARS\n20 10 7 5 4 32.5 2\n20 10 7 5 4 32.5 2\nSCORES . FREEDOM 10.0\nDECREES OF FREEDOM 10.0\n5\ntic sectors shown in Fig. 6.\n- %\nPERCEN\n1.0\n.9\n.8\n1.0\n.7\n.6\n.5\n.4\n.3\n.2\n0\n.9\n.8\n.7\n.6\n.5\n.4\n.3\n.2\n0\n.1\n.1\nSAHEL/GUINEA COAST\nSAHEL/NE BRAZIL\n14 mo\nCOHERENCE SPECTRUM\nREGION 1/SST S\nCOHERENCE SPECTRUM\nREGION 1 /SST 1\n2.5-3\n2.5-3\nYEARS\nYEARS\n20 10 7 S 4 32.5 2\n2010 7 54 32.5 2\n10.0\nOCCUEES If register 10.0\n5\nPERCEA\n0\n1.0\n3\n3\n7\n5\n5\n4\n3\n2\n1\n1.0\n.5\n0","EVAPOCLIMATONOMY (1950-1979)\n70\n60\nEVAPOTRANSPIRATION\n50\nSOIL MOISTURE\n40\n30\n20\nRUNOFF\n10\n0\n1950 1952 1954 1956 1958 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978\nYEAR\n9. Soil moisture, evapotranspiration and runoff as calculated by the clima-\ntonomy model for a 30-year period during which Sahel rainfall declines from\n800mm/an to 400mm/an.\nSURFACE FLUXES (1950-1979)\n5.4\n5.2\n5\n4.8\n4.6\n4.4\n4.2\n4\n3.8\n3.6\n3.4\n3.2\n3\n2.8\n2.6\n1950 1952 1954 1956 1958 1960 1962 1964 1966 1968 1970 1972 1974 1976 1978\nYEAR\nLATENT HEAT [LE]\nSENSIBLE HEAT [Q]\n+\no\n10. Sensible and latent surface heat fluxes as calculated by the climatonomy\nmodel for a 30-year period during which Sahel rainfall declines from 800mm/an\nto 400mm/an.\n427","Southward displacement of the tropical Atlantic wind field\nassociated with the Sahel rainfall trend 1948-83\nStefan Hastenrath\nDepartment of Meteorology\nUniversity of Wisconsin, Madison\nMadison, Wisconsin 53706\nParticularly remarkable about the current drought regime in the Sahel zone of\nWest Africa (Figs. 1, 2) is its persistence and progressive aggravation. Earlier work on\ninterannual climate variability (Lamb 1978a, b; review in Hastenrath, 1988, p. 303-309) has\nshown that individual drought years are characterized by an anomalously far southerly\nposition of the near-equatorial low-pressure trough and associated confluence between\nnorthern and southern hemispheric airstreams, as well as a zone of anomalously cold\nsurface waters across the tropical North Atlantic. Whether corresponding shifts of\ncirculation systems over the Atlantic have paralleled the downward trend of Sahel\nrainfall also on the decadel time scale is the central theme of the present study. While\nsome recent papers (Folland et al., 1986; Oort et al., 1987) have called attention to sea\nsurface temperature trends in various portions of the World Ocean, the investigation is\nhere concentrated on the nearby tropical Atlantic sector. Various elements at sea and\nover the continent and the entire annual cycle have been examined, but the present\nsummary is limited to the pressure and wind fields in July-August, the height of the\nSahel rainy season.\nThe data base consists of ship observations in the tropical Atlantic and rainfall\nrecords in the Sahel during 1948-83. The ship observations were obtained from the\nNational Climatic Center at Asheville, North Carolina, and served as a source for\nclimatic atlases (Hastenrath and Lamb 1977). They were obtained with a one degree\nsquare spatial resolution and then compiled into five degree square areas. Ship data were\nprocessed for the boxes of 5 degrees latitude by 25 degrees longitude shown in Fig. 2. As a\nmeasure of rainfall anomalies in Subsaharan West Africa, an index of April-October\nprecipitation, SAHEL (Lamb, 1985) was used (Fig. 1). Individual monthly totals for the 20\nstations of this set were kindly provided by P. Lamb.\nFig. 3 illustrates the trends of the atmospheric circulation setting over the tropical\nNorth Atlantic in the course of 1948-83. Note that the trends depicted in panels a and C\nare significant at the one percent level, and those in panels b and d at the five percent\nlevel (see Oort et al., 1987, for testing method). Fig. 3a shows a pressure rise over the\nnorthern portion of the tropical North Atlantic, and Figs. 3c and d indicate an\nacceleration of the Northeast trades. Note that the speed increments exceed values\nplausibly attributable to changes in observation practices (Wolter and Hastenrath, 1989).\nFig. 3b exhibits the latitude of zero meridional wind component, as evaluated over the\nsector 15-40 °W, with a one degree latitude meridional resolution. This is a measure of\nthe latitude position of the near-equatorial wind confluence. In context, the panels of Fig.\n3 indicate a southward extension of the North Atlantic high, acceleration of the\nNortheast trades, and a southward shift of the near-equatorial wind confluence.\n428","The southward displacement of the near-equatorial wind confluence depicted in\nFig. 3b should be placed in perspective with the magnitude of the downward trend in\nSahel rainfall. Evaluation of the 20 station ensemble forming the basis for the index\nplotted in Fig. 1 yields for July-August a meridional rainfall gradient of 87 mm per\n100 km, and a trend of 156 mm over the 36 years. This translates into a southward\ndisplacement of the (essentially zonally oriented) isohyets by 179 km over the 1948-83\nperiod. By comparison, the southward shift of the near-equatorial wind confluence\nshown in Fig. 3b corresponds to 200 km over that time span, and thus accounts for the\nobserved downward trend in Sahel rainfall.\nAcknowledgements:\nThis study was supported through through NSF Grant ATM-8722410 and NOAA Grant\nNA86AA-D-AC064.\nReferences\nFolland, C., Palmer, T.N., Parker, D.E., 1986: Sahel rainfall and worldwide sea\ntemperatures, 1901-85.\nNature, 320, 602-607.\nHastenrath, S., 1988: Climate and circulation of the tropics.\nReidel, Dordrecht, Boston, Lancaster, Tokyo, second printing, 455 pp.\nHastenrath, S., Lamb, P.J., 1977: Climatic atlas of the tropical Atlantic and eastern\nPacific Oceans.\nUniversity of Wisconsin Press, Madison, 113 pp.\nLamb, P.J., 1978a: Case studies of tropical Atlantic surface circulation patterns during\nrecent sub-Saharan weather anomalies: 1967 and 1968.\nMon. Wea. Rev., 106, 482-491.\nLamb, P.J., 1978b: Large-scale tropical Atlantic surface circulation patterns associated with\nsub-Saharan weather anomalies.\nTellus, 30, 240-251.\nLamb, P.J., 1985: Rainfall in Subsaharan West Africa during 1941-83.\nZeitschrift fur Gletscherkunde und Glazialgeologie, 21, 131-139.\nOort, A.H., Pan, Y.H., Reynolds, R.W., Ropelewski, C.F., 1987: Historical trends in the\nsurface temperature over the oceans based on the COADS.\nClimate Dynamics, 2, 29-38.\n429","1950\n60\n70\n80\n+1\n+1\n0\n0\nSAHEL\n-1\n-1\n1950\n60\n70\n80\nFig. 1. Time series plot of Sahel rainfall index SAHEL (Fig. 2; source: Lamb, 1985).\n30 W\n0\n10 E\nP\n20 N\n20 N\nSAHEL\nU, V\n10 N\n10 N\n0\n0\n10 S\n10 S\n30 W\n0\n10 E\nFig. 2. Orientation map showing location of five degree latitude bands in the eastern\nAtlantic and domain of Sahel rainfall index (Fig. 1). SLP, u and V, indicate strips for\nwhich values of pressure and zonal and individual wind components are plotted in\nFig. 3a, c, and d, respectively.\n430","80\n80\n1950\n60\n70\n1950\n60\n70\n12\n1022\no\nN\nmb\n-10\n1020\nB ITC\nA SLP 30-25 N\n8\n1018\n80\n80\n1950\n60\n70\n1950\n60\n70\n+1\n+1\n0\n0\n-1\nms\n1\nms\n1\n-1\n2\n-2\nC\nU-WIND\n15-10 N\nD\nV - WIND 15-10 N\n3\n-3\n80\n1950\n60\n70\n1950\n60\n70\n80\nFig. 3. Scatter plots of circulation parameters in the Atlantic for July-August 1948-83\n(ref. Fig. 2).\n(a) sea level pressure 30-25 °N, mb;\n(b) latitude of wind confluence (zero meridonal wind component), °N;\n(c) zonal wind component, 15-10 °N, ms-1;\n(d) meridonal wind component, 15-10 °N, ms-1;\n431","THE ANOMALOUSLY WARM WINTER OF 1988-89\nIN THE USSR\nV.N. Razuvayev and O.N. Bulygina\nAll - Union Research Institute of Hydrometeorological Information -\nWorld Data Center, USSR\nINTRODUCTION\nAn analysis of the cold season temperature patterns over the USSR is extremely important\nfor a better understanding of regional climate variability.\nThis paper considers the winter of 1988-89. Significant positive anomalies of air\ntemperature were observed over the greater portion of the USSR. The results of daily\nmeteorological observations were used to investigate the development of the anomalously\nlong period of warmth in February-March 1989. This climatic anomaly is described by\nusing factor analysis.\n1. Climatological Analysis of the Temperature Pattern for the 1988-89 Winter\nIn November-March, large positive and negative air temperature anomalies were observed\nin various parts of the USSR (Fig. 1). Most notable were the anomalies of the central\nUSSR where temperature was above normal for each month from November to March. The\nmonthly anomalies were as great as +12°C. In January-March, large positive anomalies\nwere observed in the European USSR (T=+8°C). Significant temperature variations were\nobserved in the eastern USSR. During November-January, a zone of negative air\ntemperature anomalies was found and in February some positive anomalies for individual\nstations were the largest ever observed. To estimate the magnitude of the temperature\npattern anomaly, the coefficient K (Bagrov, 1966) was used:\n2\n1\nK=-\nN\ni=1\n(1)\nwhere (T - Ti)/o is the normalized anomaly of the mean monthly air temperature for the\n1st station.\nThe statistical characteristics of the coefficient K have been considered in various\npapers (e.g., Vysotsky, 1986). The value of K has an x2-distribution with N degrees of\nfreedom. However, the empirical function of the distribution of K is usually compared\nwith the x--distribution with fewer degrees of freedom, as the anomalies of temperature\nat closely located stations are not independent even in case of a very sparse station\nnetwork. To compute K, the values of T for 33 stations were used. It has been shown\nby analyzing the empirical functions of the distribution of K for individual months that\nthe upper boundary of the interval corresponding to the accumulated probability 66.7%\ndoes not exceed 1.25. Table 1 lists values of K for individual months of the 1988-89\nwinter as well as the rank of a given K from 1881 to 1989. The largest value corresponds\nto a rank of 1.\n432","46\n0\nNovember\nJanuary\nTura\nTo Tura\nDecember\nFebruary\n8\n0.1\nas\n&\n6\nTura\nMarch\nMean monthly air temperature anomalies (°C) (see text).\nFig. 1\n.\n433","Table 1. The magnitude and rank of the anomaly coefficient, K during the winter 1988-89\nover the USSR.\nNovember\nDecember\nJanuary\nFebruary\nMarch\n1988\n1988\n1989\n1989\n1989\nValue of K\n1.41\n1.60\n1.20\n2.14\n2.53\nOrdinal number of K\nin the ranked series\nfrom 1881 to 1989\n16\n11\n28\n6\n2\nMost K values are greater than 1.25 and can be looked upon as anomalous. March 1989 was\nthe second warmest for the period from 1881 to 1989.\nAnalysis of mean monthly air temperatures allows assessment for the magnitude of\ntemperature pattern anomaly, but to perform a detailed analysis of the evolution of\ntemperature variation in particular regions as well as its\nAT/G\nVologda\na)\n3.0\n2.0\n1.0\n0.0\n-1.0\n-2.0\nTura\nB)\n-\n3.0\n2.0\n1.0\n0.0\n-10\n-2.0\nMarkovo\n3.0\nI\n2.0\n10\n0.0\n-1.0\n-2.0\n10 15 20 25 day\n5\n10\n15\n20\n25\n30\n5\n10\n15\n20\n25\n30\n5\n10\n15\n20\n25\n30\n5\n10\n15\n20\n25\n5\nNovember\nDecember\nMarch\nJanuary\nFebruary\nFig. 2 Normalized anomalies of the mean daily air temperatures (see text)\n434","association with the atmospheric processes, requires the use of mean daily and three-\nhourly observations data. Fig. 2 gives values of the normalized anomalies of mean daily\nair temperature ( T/o) for the stations Vologda, Tura, and Markovo, located in a zone\nof the largest air temperature anomalies in the western, central, and eastern USSR,\nrespectively (Fig. 1). On the basis of the data in Fig. 2, one can conclude that large\npositive air temperature anomalies for the 1988-89 winter differ in their intramonthly\nstructures. The mean daily air temperatures in the western USSR were above normal during\nthe entire period from January-March. In the central and eastern USSR, only short\nperiods of positive anomalies occur, but the mean daily air temperature anomalies were\nquite high, 20-25°C (on February 6, the value for the station Markovo was as great as\n+26.4°C, corresponding to 36). It is interesting to determine the synoptic conditions\nwhich occurred during these short warm episodes since these episodes were associated with\nthe large-scale positive air temperature anomalies during the 1988-89 winter.\n2. Analysis of the Process of Significant Temperature Anomaly Formation\nThree-hourly meteorological observations data for 222 stations were available for\nanalysis. Daily fields of sea level pressure P, air temperature T, water vapor pressure\nE, and zonal and meridional components of wind speed V. and V y were used at 12GMT for\neach day of the winter period. To jointly analyze the fields of meteorological data,\nthe R-technique of factor analysis was used (Iberla, 1980). The factor analysis is\ncarried out for each day. The factor loadings 1 indicate the degree of the factor\nassociation with each of the above meteorological elements. The factor scores indicate\nthe association of each factor with the observation site. Considering the temporal\nchanges in loadings for the factor most closely associated with temperature allows\nassessing the peculiarities of the atmospheric processes affecting the air temperature\nvariability. As an example, consider the loadings from V and Vv. Their magnitude\nindicates the role of advection and their signs - the direction of the advective\nprocesses. Additional information can be obtained by comparing the factor score fields\nat various times. This allows an assessment of the stability of the factor solution in\ntime and also gives an idea of the evolution of atmospheric processes. Fields of factor\nscores however, are not comparable in terms of the R-technique as they correspond to the\nfactor solutions differing in the structures of their factor loadings. The paper\n(Bulygina, 1988) considers a technique for obtaining the factor loadings common for\ndifferent times. A matrix of initial data is thus defined as Z(t)\nZ(t2), and a matrix of\nfactor scores assumes the form F(t)\nF(t2). Considering the coefficient of correlation between\ntwo parts of the vector of factor scores (e.g., of the first factor, i.e.,\nN\nFp1,2),\ninformation on the similarity of factor solutions and hence, the atmospheric processes\nat various times can be obtained. It is evident that a correlation coefficient close\nto unity reflects a persistent atmospheric process.\nLet us consider an application of the above technique. Consider the warm period in the\neast of the USSR from 1 to 18 February which is characterized by large values of T.\nFig. 3 gives the absolute values of loadings for the first factor explaining the largest\nportion of variance of the meteorological fields. The correlation coefficient of this\nfactor for each pair of subsequent days is given as Fig. 3d. The fields of air\ntemperatures and factor scores for the first factor for February 6 when the mean daily\nair temperature anomaly was largest, given as Figs. 3a and 3b, respectively. Judging\nby the loading values and the signs, the air temperature increase resulted from a deep\ncyclone (a loading from P is negative) and an advection from the Pacific. The region\nof the first factor score maxima coincides with the area of maximum values of temperature\n435","Air temperature\na)\nFACTOR 1\n45\nmop.\nFebruary 6\n0\nd)\nc)\nR\n10\n0.9\n08\n0.8\n06\n0.7\n0.4\n0.6\n0.2\nday\n5\n15\n17\n5 7 9 11 13 15 17 day\nair temperature (T)\nsea level air pressure (P)\nwater vapour pressure (E)\nwind speed zonal component (Vx)\nwilld speed meridional component(V.)\nAnalysis of the temperature anomaly formation processes (see text).\nFig. 3\n436","anomalies. The atmospheric processes were persistent for a long period of time (r >\n0.7). On February 10 and 11, the pressure field changed. The cyclone increased and\nthe Siberia anticyclone decreased. The disturbance in the atmospheric persistence is\nmanifested in the lower values of correlation coefficients. The smaller contribution\nof the loading from the zonal component and the reversal of sign of the loading from the\nmeridional component of wind speed on February 15 can be accounted for by the cyclone\nfurther increasing and moving to the west which resulted in advection to the North,\nincreasing over the entire area under consideration. Similarly, the atmospheric\nprocesses in the rest of the warming periods shown in Fig. 2 were analyzed. Within the\nEuropean USSR, the warm periods considered are mostly associated with the short-lived\ncyclones. Increase in the number of cyclone trajectories in Eastern Europe during the\n1988-89 winter was noted in Nemirovsky (1989). It would be interesting to continue the\nstudy of warm periods on the basis of the above techniques using daily data. This might\nprove useful in studying regional climate variation.\n3. References\nBagrov, N.A., , N.N. Myakisheva, 1966: Some characteristics of the monthly mean air\ntemperature anomalies. Proceedings of the World Met. Centre, 9, 3-17 (in Russian)\nBulygina, O.N., , V.N. Razuvayev, 1989: Using factor analysis for describing anomalous\nclimatic situations. Proceedings of VNIIGMI-WDC, 150, 138-147 (in Russian).\nIberla, K., 1980: Factor Analysis, Finance and Statistics. Moscow, 398 (Russian\ntranslation from German).\nNemirovsky, a.V., V.T. Radjuhin, 1989: Vector index of the North Atlantic Oscillation\nand its association with the surface air temperature and circulation. Proceedings 14th\nAnnual Climate Diagnostic Workshop, La Jolla, CA, USA.\nVysotsky, A.M., M.K. Zdonova, V.N. Razuvayev, N.N. Shevchenko, 1986: Anomalies in the\ntemperature patterns of cold periods on the USSR area. Proceedings of the Third Intl.\nConf. on Stat. Climat., Vienna, 265-270 (in Russian).\n437","REASONS FOR THE REVERSAL OF CIRCULATION\nAND U.S. ANOMALY PATTERNS\nBETWEEN FALL AND WINTER, 1988-89\nby\nJerome Namias, Research Meteorologist\nClimate Research Division\nScripps Institution of Oceanography\nUCSD, A-024\nLa Jolla, CA 92093\nA major reversal in pattern of temperature anomalies occurred between autumn of 1988\nand winter 1988-89, a climatic break which reflected a change in phase of the prevailing\nmid-tropospheric wind pattern over much of the western hemisphere. This transformation is\ndescribed and attributed to an attempt by the atmosphere to adapt to forcing by remote and\npersistent circulations acting in concert with seasonal forcing. Practically all conventional\nforecast tools, especially those based largely on climatic contingencies, failed to predict the\nbreak. However, instability of the flow pattern by November was clearly indicated by: (1)\nconserving the standardized November height anomalies into winter, (2) testing the stability\nof this new hypothetical pattern with the help of statistically derived teleconnections, (3)\napplying a modified mean barotropic model to the just named chart, and (4) computing the\nlong period trends of circulation from the autumn months. Making use of this material, a\nforecast made for winter at the end of November enjoyed very high skill in spite of the failure\nof persistence between seasons.\nThe dramatic reversal in temperature anomaly pattern between fall and winter, 1988-89\nis shown in Figure 1 (a and b). The 700 mb anomalies for the two seasons (not shown) illus-\ntrated that Klein's specification equations showed that temperature patterns were consistent\nwith circulation - low heights in the west relative to high heights in the east. This pattern\nwas essentially the reverse of that in fall. Some model experiments including the resulting in\nFigure 2 offered an alert by the end of November for the changed winter circulation pattern.\nVarious temperature anomaly specifications from objective 700 mb circulation predictions\nare shown in Figure 3. In 3a, the standardized 700 mb height anomalies were conserved from\nfall to winter (a type of persistence forecast), while in 3b a modified barotropic forecast was\n438","made from 3a. 3c employs a trend method computed from months in the fall season, and\n3d used the barotropic prediction made from 3c. The final temperature anomaly prediction\nprepared in late November for winter resembled 3d and scored 60 out of 99 points correct in\nthe U.S. with no two class errors (compare Figure 3d with Figure 1b).\nLegends to Figures\nFigure 1. Temperature anomalies for (a) autumn 1988 (Sep. Oct. Nov.) and (b) winter\n1988-89 (Dec. Jan. Feb.), expressed in terciles: A, above normal; B, below normal; and N,\nnear normal where the base period is 1951-80.\nFigure 2. Barotropic forecast of 700 mb height (broken, in m) and anomalies (solid, in\nm) made from initial trend chart. Temperature anomaly specifications from this chart are\nshown in Figure 3d.\nFigure 3. Specifications of temperature anomalies (terciles) from 700 mb height made for\nwinter 1988-89; (a) assuming conservation of standardized height anomalies from autumn;\n(b) from barotropic model forecast using (a) as initial; (c) from long period trend chart; and\n(d) from barotropic model applied to trend chart. Note that charts b, c, and d all indicated\nthe break between fall and winter while persistence of standardized height anomaly (a) failed.\n(Compare with Figure 1.)\n439","U.S. TEMPERATURE ANOMALIES\n130\n120\n110\n100\n90\n80\n70\nA\nN\nNO\nN\n30\nN\nA\n25\nAUTUMN 1988\n(a)\nOBSERVED\n110\n100\n90\n80\n130\n120\n110\n100\n90\n80\n70\n60\nN\nN\nN\nA\n35\nN\nA\nN\n30\nA:\n-A\nB\nA\nWINTER 1988-'89\nOBSERVED\n110\n100\n90\n80\n440","BAROTROPIC FORECAST\nFROM WINTER 1988-'89 TREND FORECAST\n0.0\n2800\n0.0\n2800\n0008\n441","(b)\n(d)\nWINTER TEMPERATURE SPECIFICATIONS\nBAROTROPIC FROM TREND\nBAROTROPIC FROM DM/o\nN\nN\n2:\n(a)\n(c)\nN\nN\nFROM AUTUMN\nDM/o CONSERVED\nTENDENCY METHOD\nN","CLIMATE DATA SETS\nKenneth D. Hadeen\nDirector\nNational Climatic Data Center, NOAA/NESDIS\nAsheville, N.C.\nHigh quality, homogeneous data sets are\nrequired before credible research results\nare accepted by the scientific community.\n84,000\nUSER CONTACTS HANDLED\nThis paper discusses several aspects of\n4,848\nBULLETINS PREPARED\ndata management, levels of quality\n1,483,000\nCOPIES PRINTED\nassurance, and data availability of\n1,069,000\nCOPIES SUPPLIED TO 40,830\nselected data sets at the National\nPAID SUBSCRIBERS\nClimatic Data Center. The NCDC has\n2,177,182\nORIGINAL FORMS ARCHIVED\nrestructured its organization to increase\n39,378\nSATELLITE HARDCOPY IMAGES\ninteraction and provide selected Climate\nARCHIVED\nQuality or baseline data sets to the\n120,500\nTAPES/CARTRIDGES ARCHIVED\nResearch Community.\n(2.169 X 1013 BYTES)\n6,543,622\nRECORDS KEYED\nThe National Climatic Data Center was\ngranted official status in 1951 as a\nFederal Center for weather records under\nFigure 1.\nFY-88 Production\nauthority of the Federal Records Act of\n1950. The head of GSA at that time\nauthorized a joint Weather Bureau, Air\nForce, Navy (WBAN) Weather Records Center\nto be established in Asheville, N.C. The\nWBAN relationship exists to this day and\nNCDC is an operational data center that\nremains one of the strengths of the\ncollects, sorts, digitizes, runs quality\nFederal Climate Data Complex. Operating\nassurance programs, processes, prepares\nLocation-A of the Air Weather Service\nmonthly publications, updates digital\n(AWS)\nEnvironmental\nTechnical\narchives, and disseminates massive data\nApplications Center (ETAC) operates a CO-\nholdings each month. A measure of the\nlocated computer and tape library with\nmagnitude of the normal operational\nNCDC. The Navy Oceanography Command\nworkload is given in Figure 1.\nIn\nDetachment (NOCD) has also been a\naddition to the monthly production of new\nmainstay of the triumvirate through the\ndata, the total archive must be\nyears. The co-location of these agencies\nmaintained and updated as necessary.\nprovides access to a wealth of climate\nFigure 2 is an estimate of NCDC current\ndata and products unavailable any place\nholdings. It does not show the\nelse in the world. Several Global and\napproximately 100 Terabytes of\nRegional Data Sets are becoming available\nGeostationary Satellite Data maintained\nto the research community because of this\nat the University of Wisconsin under\nrelationship.\ncontract to the NCDC.\n443","To better illustrate what I am talking\nabout when I refer to Climate Data Bases,\n*\n280 DATA SETS\nwe can consider the Historical\n*\n210 MILLION PAPER RECORDS\nClimatological Network (HCN) data bases\n*\n1 MILLION MICROFICHE\nproduced in cooperation with the Carbon\n*\n120,500 MAGNETIC TAPES/CARTRIDGES\nDioxide Laboratory of the Department of\nEnergy (DOE). Starting with the basic\n(1 X 10 13 BYTES OR 20 TERABYTES)\nnetwork of more than 20,000\nclimatological stations shown in Figure\nFigure 2. HOLDINGS OF THE NATIONAL\n3, we determined that in order to qualify\nCLIMATIC DATA CENTER\nfor the study a station must have had at\nleast 80 years of continuous record, be\nlocated in a rural area, and must not\nhave been affected by station moves or\nAlthough some of these data sets could\nurban buildup. After candidate stations\nmeet the quality and completeness\nwere selected based on period of record,\ncriteria to be considered Climate Data\nstation history (Metadata) information\nSets, most require additional work.\nI\nwas examined in detail to ensure that the\ndon't consider data collected from the\nobservation conditions were basically\nGlobal Telecommunication System (GTS) or\nstable throughout the period. The result\nour data that have gone through\nwas a compilation of monthly values of\npreliminary quality assurance programs as\naverage maximum/minimum temperatures and\nadequate to be considered as Climate Data\nprecipitation for 1,219 stations. Thus,\nSets. The reasons are primarily because\nfrom a massive data base a climate data\nof the different observing practices,\nset was extracted that meets certain\ninstrumentation, missing data, and the\npredefined criteria. The criteria used\ntendency to throw out observations that\nfor a specific study or program should be\nare extremes. Also, in data sparse\npredefined and may vary with data type.\nregions numerical analyses performed in\nData for the HCN are now available in\nan operational mode tend to produce\ndigital form, are well documented, and\nclimatology of the model and not of the\ncan be used with a high degree of\nreal world.\nconfidence.\nNCDC is constantly looking for better\nways to disseminate data and information.\nFigure 4 shows some of the methods and\nmedia used by the Center. The Compact\nDisks Read Only Memory (CD-ROM) is\nproving to be very popular and provides\nlarge quantities of data rapidly and\ninexpensively to the user.\nThe\nproliferation of discs has forced more\nstandardization on the industry. Figure\n5 shows four CD-ROMs NCDC has been\ninvolved with through joint activities\nwith the private sector. The CD-ROM\ndeveloped with the University of\nWashington, called WORLD WEATHERDISC,\nFigure 3. U.S. COOPERATIVE STATIONS\nCLIMATE DATA FOR THE PLANET EARTH, has 18\n(JULY 1985)\ndata sets on it as shown in Figure 6.\n444","PAPER\nFILM\nThis is the first attempt to place a\nPUBLICATIONS\nMICROFICHE\ncomprehensive description of the Earth's\nCOPIES OF ORIGINAL\nMICROFILM\nclimate on a single disc. The contents\nRECORDS/CHARTS\nMOVIE LOOPS\nare given below:\nATLASES\n35MM FILM\nSPECIAL SUMMARIES\nPHOTOGRAPHS\n1. WORLD MONTHLY SURFACE CLIMATOLOGY\nCOMPUTER OUTPUT\n2. CLIMATIC DIVISION DATA\nMICROPRINTS\n3. WORLDWIDE AIRFIELD SUMMARIES\n4. U.S. STATION CLIMATOLOGICAL SUMMARIES\nDIGITAL\nOTHER\n5. U.S. MONTHLY NORMALS\nMAGNETIC TAPE\nTELEPHONE\n6. COADS SHIP DATA\nFLOPPY DISKS\nFAX\n7. SADLER TROPICAL CLIMATOLOGY\nCD-ROM\nVISIT\n8. U.S. FREEZE DATA\nELECTRONIC MAIL\n9. DAILY WEATHER STATISTICS\nON-LINE DIAL-UP\n10. JONES ET AL GRIDDED DATA\n11. NMC MONTHLY-MEAN GRIDS\nFigure 4. MANY FORMS OF DATA AND INFORMA-\n12. NCAR WORLD MONTHLY WEATHER RECORDS\nTION DISSEMINATION SUPPORTED BY\n(TD-9648)\nNCDC\n13. HISTORICAL SUNSHINE DATA (TD-9788)\n14. STATION HISTORICAL FILE (TD-9767)\n15. TORNADO DATA (TD-9714)\n16. THUNDERSTORM BEGINNING AND ENDING\nTIMES (TD-9945)\n17. LIGHTNING STATISTICS (TD-9617)\n18. CYCLONES (1871-1986)\nNORTH AMERICAN\nOBSERVATIONAL DATA\nFigure 6. CONTENTS OF WORLD WEATHER DISC,\nFOR JULT-DECEMBER 1987\n-\nopen are\nCLIMATE DATA FOR THE PLANET EARTH\nWorld\nWeatherDisc\nClimate Date for\nthe Planet Earth\nNCDC has several ongoing activities in\nCLIMATE\nthe development of Reference Data Sets.\nTEDATA\n1151:\nThese are listed below and indicate if\nMISC\nthey are joint projects like with the\nCarbon Dioxide Information and Analysis\nCenter (CDIAC) or others. Also, the\nSUMMART FTEB-WESTERN\nPERIOD OF RECORD\nperiod of record available and data set\nLICENSE\nOFTICAL\navailability are indicated.\nFigure 5. CLIMATE DATA ON CD-ROM\n1. USA Max/Min temperature and precipitation\n(mostly Cooperative)\na. Monthly - 1,219 stations - B.P.O.R.-\nA Climate Research Data Catalog-Global\n1987. Available now.\nClimate Change Studies was produced for\nb. Daily - 138 stations - B.P.O.R.-1988.\nthis workshop and 50 copies were\nAvailable Spring 1990.\ndistributed at the workshop. This\nC. Monthly Extremes - 950 stations -\ncatalog was produced by using the\nB.P.O.R.-1988. Available late 1990.\nautomated Climate Inventory and Catalog\nd.\nDigital Station Histories to 1987.\nSystem recently implemented at the NCDC.\nAvailable now.\nIt is our plan that access to this system\nwill be made available to outside users\nat some unspecified future date.\n445","2. USA Clouds - Monthly (197-240 principal\n(3) Thickness of Surface-Based\nstations) (NCDC/CDIAC)\nInversion\nb. Surface Max/Min/Precip/Snow Depth\na. SR-SS Total Sky Cover - B.P.O.R.-1987.\n24-Hour\nAvailable December 1989.\nC. AM/PM Medians\nb. Total Sunshine - B.P.O.R.-1987.\n(1) Temp\nAvailable now.\n(2) Humidity\n(3) Winds\n3. USA Pressure - Monthly (197 principal\n(4) Total Cloud (Including Ceiling)\nstations) (NCDC/CDIAC)\na. Water Vapor Pressure - B.P.O.R.-1987.\nOne problem in interpreting historical\nAvailable December 1989.\nrecords is illustrated in Figure 7, which\nb. Station Pressure - B.P.O.R.-1988.\nis a plot of mean cloud amount for\nAvailable Spring 1990.\nAlbany, N.Y , for the period 1900 to\nC. Sea Level Pressure - B.P.O.R.-1988.\n1988. What appears to be a clear trend\nAvailable Spring 1990.\nin increased cloudiness becomes very\nsuspect when the Metadata is evaluated.\n4. USA First Order Data (175-240 principal\nFigure 8 shows biases due to observing\nstations) (NCDC/CDIAC)\npractices.\na. Daily Snow Depth - B.P.O.R.-1988.\nAvailable Summer 1990.\nALBANY, N.Y. (UNEDITED)\nb. Monthly Snowfall - B.P.O.R.-1989.\n- CLAND INCLUD\nAvailable now.\nSUPPER\nX\nMINTER\nC. Monthly Max/Min/Precip - B.P.O.R.-\n1989. Available now.\nd. Monthly Heating & Cooling Degree Days.\nAvailable now.\n5. USA Daily Snow/Max/Min/Precip (319 Coop-\nerative stations) (NSF - Prof. Dave\nAIRPORT\nRobinson). Available 1990.\nCITY\na. Snowfall\nb. Snowdepth\n1920\nISCED\n1940\nLEED\n1990\n1970\nC. Total Precip\nYEAR\nd. Max/Min Temperatures\nFigure 7. MEAN CLOUD AMOUNT, ALBANY, N.Y.\n6. Mexican Monthly (350 stations) (Creighton\nUniversity - Prof. Art Douglas).\nAvailable 1990.\nSUNSHINE INSTRUMENTS\na. Long-Term Temperature (100)\n1888-1907 Photographic Recorder\nb. Long-Term Precipitation (250)\n1907-1952(Circa) Thermometric Recorder\nC. Formation of Climate Divisions\n1952(Circa)-Present - Photoelectric Sunshine Switch\nd. Drought Indices\nCLOUD OBSERVATIONS\n7. USA Concomitant Surface/Upper Air (60\nstations) (NCDC/CDIAC) Available now -\nUP TO = 1940 Frequent personal observations,\nCanadian 1990. B.P.O.R.-1988.\nbut only two observations 0800 and 2000 LST\nentered on 1001 form.\na. Twice daily 100, 85, 70, 50, 30 KPA\n(1) Temp, Height, Wind, Moisture\n1940 - PRESENT Hourly Observations.\n(2) Precipitable Water\nFigure 8. BIASES IN CLOUD AND/OR SUNSHINE\n446","Satellite data can be a very valuable\nsource of information in monitoring large\nareas that have little or no in situ\nAVHRR\nMSU\ndata. NCDC is participating in a\nUpper Air\nSurface\ncalibration/validation program as part of\nthe International Global Precipitation\nClimatology Project. We have established\nNational Climatie\na Surface Reference Data Center for the\nData Center\nproject with the goal of collecting all\nDate\nAnalysis\nthe in situ rainfall, radar, and\nSotallite\nN.C.\nResearch\nState\nsatellite data for approximately 20 test\nLOD\nareas around the world. A detailed\nanalysis of the data in these areas will\nimage\nProcessing\nbe produced and used as a\nDate\nAnalysia\ncalibration/evaluation/validation of\nsatellite precipitation estimates and\nFigure 10.\nNASA/NOAA/UNIVERSITY\ndevelopment of new algorithms.\nA\nPROGRAMMATIC RESEARCH;\ndepiction of this activity is given in\nURBAN HEAT ISLANDS AND\nFigure 9.\nTHE GREENHOUSE EFFECT.\nNCDC (SRDC)\nDATA\nREFERENCE\nDATA SETS\ngages\ndata\nareal\nCARDS\nradar\nDaily\nmgmt\nIntegration\nStation\nClimatological Aerological Reference\nMistaries\nexp sensors\nECMWF\nData Set\nU.S.\nsample\nSort &\nRonoom\nmethods\nAirforce\ndata\nHomogeneity\nApprogate\nErrors\nNCDC\nInterogeney\nUniversity\nUSSR\ninteregency\nRESEARCH\nDistribute\nINSTITUTIONS\nDate\nUpeates\nDate\nAnalyais\ndevelop Integration\nmethods\nNCDC\nNCDC\nUniversity\nInteregency\nFigure 9.\nGLOBAL PRECIPITATION\nHomogeneity\nCLIMATOLOGY PROJECT.\nUpdates\nProduce Data\nAnalysis\nRancom Errors\nFigure 11. CLIMATOLOGICAL AEROLOGICAL\nA program looking at the possibility of\nREFERENCE DATA SET (CARDS).\ndetermining Urban Heat Island effects\nwith NASA, the University of North\nCarolina, and the NESDIS Office of\nResearch is depicted in Figure 10.\nThe NCDC operates the World Data Center A\nWe are in the initial stages of\n(WDC-A) for Meteorology under the auspices\ndeveloping a Climatological Aerological\nof the International Council of Scientific\nReference Data Sets with several other\nUnions and the World Meteorological\nnational and international organizations.\nOrganization. A part of the\nThis is depicted in Figure 11.\nresponsibilities of the World Data Centers\nis to archive data products, etc., from the\nWorld Meteorological Centers (WMCs).\n447","The National Weather Service/National\nMeteorological Center is considered one\nof the WMCs and NCDC archives many of the\nproducts of NMC.\nNCDC actively participates in the\nBilateral Exchange of data with the\nSoviet Union. We have exchanged several\nlong-term data sets that may or may not\nhave been transmitted as part of the WMO\nexchange.\nFigure 12 depicts this\nactivity.\nMeteorogical\nMeteorological\nQuality\nQuality\nDate e.g.\nDate e.g.\nUpper Air\n6-houny date\n1000 stations\nUniversity\nParticipation\nUniv. of\nMossocnusetts\nUniv. ct\nNCDC\nAURIHI\ninstitute\nOkionome\nAsneville\nColumbio Univ.\nObninsk\nRutgers univ.\nExamples:\n_eagates Global Precio\nRobinson Daily Snow\nBrooley GIODO Precio.\nSlimate\nClimate\nQuality\nQuality\nDate e.g.\nDate e.g.\n-CN\nDolly\nMax-Min,\nPrecio,\nSnow\nFigure 12. BILATERAL EXCHANGE OF CLIMATE-\nQUALITY AND METEOROLOGICAL-QUALITY\nDATA SETS\nIn conclusion, NCDC's emphasis for the\n1990's falls in the development of Global\nand Regional Reference Data Sets,\nproviding better and more efficient\nservice to the Research Community as well\nas our many other users and constituents.\nWe have a major task in developing\nsystems to archive a massive increase in\ndata from new satellites, from land\nsurface systems, and in the definition of\nstandards to provide compatible data\nsets. We must have strong interaction\nwith the Research Community in the\ndevelopment of climate-quality data sets\nand in setting priorities of which data\nsets to be enhanced.\n448","WET/DRY SPELLS VIEWED FROM SAI TIME SERIES\nIN THE CANADIAN PRAIRIES\nby\nLeo O. Mapanao\nCanadian Climate Centre\nAtmospheric Environment Service\nABSTRACT\nPreliminary results of our analysis of the SAI (Standardized Anomaly Index) series, which\nrepresents the imbalance of heat and moisture in the atmosphere are discussed with a view on\nsome aspects of climate variability, climate change, and predictability of the index. Focusing at the\ndrought-prone Canadian prairies with at least 100 years record of precipitation and temperature,\nthe analysis was carried out for Regina, Saskatoon, Medicine Hat and Calgary. Monthly cumulative\ndistribution functions are presented to help in the estimation of drought probabilities and return pe-\nriods at varying intensities.\n1.\nBackground\nFuture climate is certain to be variable and changing, but unlike the past, this time man's\ninfluence on the normal transparency of the atmosphere to solar radiation can inadvertently upset\nnatural climate fluctuations. As society is already tuned to average weather, it is the spell of dry or\nwet, hot or cold phenomena that often bring the greatest problems to our society and economy.\nMonitoring and anticipating the emergence of these events against planned vulnerabilities inferred\nfrom statistical studies, is our immediate goal at the Hydrometeorology division/Canadian Climate\nCenter.\nSo, to supplement the weekly monitoring of hydrologic components and to support the\nWorld Climate Program - Water Project A.2 (see WMO/TD-No.224,1988), design for 88 analyzing\nlong time series of hydrological data and indices with respect to climate variability\", we have un-\ndertaken to study the SAI series at four representative locations in the Canadian prairies, with\nhopes of acquiring some capability to anticipate the atmosphere's relative wetness or dryness at\nlong lead times.\n2.\nThe \"SAI\" Index\nThe SAI index is simply the difference between normal deviates of mean monthly temper-\nature from monthly total precipitation, expressed as :\nSAI = P'/8p' - T'/st'\n(1)\nwhere P' and T' are anomalies (based on 1951-80 normals) of the monthly total precipita-\ntion and mean monthly temperature respectively; and sp'st' are the corresponding mean square\ndeviations.\nSAI is a measure of relative atmospheric wetness/dryness as it only refers to the mean\nclimate values of precipitation and temperature for specific months and regions. Above normal\nprecipitation combined with below normal temperature translates to positive SAI values, thus in-\ndicating moisture surpluses, while negative values denote deficits. For our purposes, SAI => 2\n449","would signify a potential flooding situation while SAI <=-2 is indicative of drought. These arbitrary\nthresholds may be interpreted as the number of standard errors for which the odds against the OC-\ncurrence of a deviation as great as or greater, is better than 20 to 1. These criteria also help delin-\neate wet/dry years in the SAI series for validation against reality. For comparison Italy uses -0.85\nfor drought, while Russian experience favors a range of -1.3 to -2.0 in their agricultural districts.\nTwenty one basic statistical parameters were calculated, but we now present only six fig-\nures that would graphically summarize and depict typical interannual features and structures of the\nSAI series at Medicine Hat and Regina for the winters, the summers and the annuals (mean values\nfor each year).\n3.\nResults\nFeatures of the SAI Annual Series\n3.1\nIn the set of Fig. 1, the variability of the index for the last 105 years is clearly portrayed with\nindications of their relative intensities. For a SAI = < -2, 18 drought years can be picked out\nat\nMedicine Hat but only 8 at Regina. This threshold matches the -2 Palmer drought index. In par-\nticular, we point out the very dry winters of the 30's which aggravated the droughts of the ensuing\nsummers. Farmers are very much aware what lack of snow can mean to subsequent crop yields.\nOn the other hand index near 2 was found to correspond to at least 170% of normal annual\nprecipitation.\nThe third panels depict the cumulative distribution functions for a typical year. A trend to-\nward dryness is clearly noticeable during winters and springs and some skewness can be seen\nmarkedly in the tails. These empirical near-normal distributions may be used to calculate probabil-\nities of droughts and return periods at various intensities. If climatic change is supposed to be re-\nflected in changes of distribution parameters (e.g. means, variances, etc.), certainly they are man-\nifested in the winter and spring values.\nLong-term Fluctuations\n3.2\nIn Fig. 2 plots, a 22-year Gaussian filter captures the longer time scale changes more\nclearly. While short-term fluctuations are more evident in the winter plots, the long term drying\ntrend from the start of the century is seen vividly to persist through the thirties in the summer and\nannual curves. Most likely the dry spell of the depression years was a segment of this longer term\nclimate signal.\nAutocorrelation Functions and Periodograms\n3.3\nThe autocorrelation plots of Fig. 3, would identify any systematic dependence of the ob-\nservations on past history. Here, we note higher amplitudes (.284) for the summers, yet they\nwould only explain about 8% of the variance. We also infer the possible uniqueness of generating\nmechanisms for each seasonal series, and the year overall.\nThe marginal autocorrelation coefficients observed are complemented by the cumulative\nperiodograms of Fig. 4. We especially note that possible signals stay well within the \"white noise\"\nband. However, in the annual plot, the presence of a \"red\" spectrum below 6 years is an encour-\naging sign for the development of a more tractable and fruitful SAI prediction method. Initial\nexamination of the feasibility of an autoregressive scheme suggests that a fifth-order Markov pro-\ncess should account for at least 17% of the variance. It is emphasized that drought prediction on\nan annual bias, even with marginal skills, is of great practical and economic value.\n450","3.4\nNormalized Variance Spectrum\nIn Fig. 5, we get an idea of the strength of the relationship between the power of the latent\nharmonics tuned in against its frequency. When compared to a purely random process whose\nspectrum would appear as a horizontal line, these figures show the decomposition of total variance\ninto possible periodicities which may then be link with geophysical and anthropogenic causes.\nHowever, in analyzing the structure of the spectrum from the data, there is always the problem of\ninterpretation because it is sometimes difficult to identify \"side band\" and \"echo\" effects. Further,\nalthough we can always fit a Fourier series to the data, in general there's no reason to suppose\nthat it corresponds to any reality. Nonetheless, the \"sun spot\" cycle (Van Loon and Labitzke, 1988)\nmay be a reasonable explanation for the 11-year peak observed at Medicine Hat with considera-\ntions of a 2-year margin of error. The 6/7-year peak seen in the annual diagrams may be atmo-\nspheric memory link to ocean surface forcings. With a sampling interval of one year, these dia-\ngrams should detect at least half of a cycle.\n3.5\nTrend Statistics for Means and Variances\nThe curves in Fig. 6 are intended for visual detection of major change points in the means\n(crossed curve) and variances (boxed) of the observations. Short-term fluctuations and long-term\nchanges can be identified, with the drought years of the 30's unmistakably a part of the longer\nscale trend. General agreement in the trend in the mean SAI is noted with reversal during 1931 as\nfar as the annual plot could detect.\n4.0\nConcluding Remarks\nThis paper concludes the following:\nThe SAI index should be a suitable tool for the monitoring of the relative wetness/dryness\nof the atmosphere then possibly used as input to other systems/models which are largely depen-\ndent on climate information.\nThe index was able to detect short and long-term fluctuations of heat/moisture imbalance\nfor the last century, for which the drought phenomena of the 30's are clearly depicted as part of the\nlong-term climate signal.\nSince evaporation is essentially a function of temperature, the index can rightfully be a sur-\nrogate to the runoff (Precipitation - Evaporation) index customary employed in hydrological calcu-\nlations and monitoring.\nFinally, the empirical cumulative distribution functions obtained can be used to estimate\nprobabilities of dry/wet phenomena and return periods.\n451","Fig. 1. Standardized Anomaly Index (SAI) series for Medicine Hat (left)\nand Regina (right)\nThe Indes Tari PXR\nm\nJanuary\nJanuary\na\n2\na\n-3\n-3\n0080\n1000\n1923\n1040\n2000\n1000\n1000\n1020\n1040\n1000 1000 8000\n1000\n1980\nYEAR\nYEAR\nYXR\nJuly\nJuly\n-3\n1000\n1920\n1000\n1000\n1020\n1040\n1960\n1980\n2000\n1960\n1080\n2000\nYEAR\nYEAR\nCumulative Distribution functions\nBASE\nYSR\nS\n90%\n3\n90%\n75 %\n75%\n50\n25\n50%\n-25%\n5%\n5%\nFEB APR MAY JUN JUL AUG SEP OCT NOV DEG\nJAN MAR APR MAY JUN JUL AUG SEP OCT NOV DEC\nMONTHS\nMONTHS\nFig. 2. Long-term fluctuations detected with low-pass filter.\nJanuary\nJanuary\n-9\n1800 1000 1010 1020 1030 1040 1080 1000 1970 1880\n1600 1900 1010 1020 1030 1940 1030 1060 1870\nYEAR\nYEAR\n0\nJuly\nJuly\n-3\n1000 1920 1030 1040 1030 1970 1000\n1800 1900 1010 1920 1030 1840 1930 1000 1970\nYEAR\nYEAR\na\na\nAnnual\nAnnual\no\n1880 1050 1610 1020 1020 1090 1070\n.... 1020 1030 1030 1060 1000\nYEAR\n,YEAR\n452","Fig. 3. Autocorrelation functions for Medicine Hat (left) and Regina (right)\nJanuary\nJanuary\n12\n13\nLAO\nLAC\n0.30\nJuly\nJuly\n0.20\n0.21\n0.00\n0.01\n-0.04\n-0.00\n-0.14\nLAG\nLAO\nAnnual\nAnnual\n7\n10\n11\n12\n13\n14\n15\nLAC\nLAG\nFig. 4. Cumulative Periodograms for Medicine Hat (left) and Regina (right)\n1.0\nJanuary\n0.0\nJanuary\n0.0\n0.7\n0.0\n0.5\n0.0\n0.3\n0.3\n0.2\n0.1\n0.0\n0.0\n0.3\n0.0\n0.2\n0.3\n0.4\n1.0\nJuly\n0.0\n0.0\n0.7\n0.0\n0.3\n0.4\n0.3\n0.2\n0.8\n0.1\n0.0\n1.0\nAnnual\nAnnual\n0.0\n0.0\n0.7\n0.0\nO.D\n0.4\n0.3\n0.2\n...\n0.0\n0.4\no\n453","Fig. 5. Variance spectra for Medicine Hat (left) and Regina (right)\nPRINT\nJanuary\nJanuary\n0.10\n10.00\n40.00\n0.00\n>0.00\n14\n14\n0.89\n0.81\nJuly\nJuly\n0.00\n0.17\n0.15\n0.13\n0.01\n0.00\n0.08\n0.00\n0.03\n0.01\n16\nov\n. 10 \" 12 13 94\nLAD\nLAG\nAnnual\nAnnual\n0.00\n40.00\n0.04\n0.02\n0.00\nis\n10\n19\n13 is 18\nLAG\n640\nFig. 6. Trends in mean SAI (cross) or variance (box) at Medicine Hat (left)\nand Regina (right) .\nFis\nU\na\npl\nJanuary\nJanuary\nS\no\nS\n2\n1940 8000\n1080\n1000\n1980\n0080\n1840\n1060\n1000\n1880\n1000\n1035\n1840\nYEAR\nYEAR\nD\na\nJuly\nJuly\ne\nacce\n7 1000\n1000\n1920\n1840\n1008\n1605\n1000\n1800\n9080\n1840\n9000\n1000\nVEAN\nYEAR\nE\nTRA\nM\nAnnual\nAnnual\na\nCLs\n1680\n0080\n1000\n1000\n1000\n1000\n1000\n1080\n1040\n1000\n1000\n2000\n1000\nVEAN\nYEAR\nI\n454","INTER-ANNUAL VARIABILITY IN NORTH AMERICAN MOISTURE FLUXES:\nSTREAMFLOW RECORDS MEET ATMOSPHERIC ANALYSES\nNeil E. Sargent\nClimate Diagnostics Research Group\nCanadian Climate Centre\nDownsview, Ontario, M3H 5T4, Canada\nIntroduction: Operational atmospheric analyses prepared by the National Meteorological Center (NMC)\nare an attractive source of data regarding energy and water budgets of the atmosphere on a wide range\nof time scales. If the data they contain faithfully represent actual conditions at the times of observation\nthey would form a good starting point for studies of the dynamics behind, among other things,\nintermediate to long term moisture supply anomalies. As tempting as it is to launch into such studies\nthere are reasons to believe the analyses may not contain useful information about moisture fluxes and\ntheir divergence. The moisture field is considered to be among the least reliable of all in the analyses\n(Lambert, 1988, Trenberth and Olson, 1988) particularly in the tropics (Rosen and Salstein, 1985). In\naddition to problems with the analysis of moisture, horizontal wind divergence fields are notoriously\nnoisy and difficult to analyse. Rasmusson (1967) has suggested twice daily observations of the wind\nfield may not be frequent enough to adequately represent the true moisture flux. In view of these\ndoubts about the usefulness of moisture fluxes derived from operational atmospheric analyses, this\npaper attempts to evaluate the quality of such flux values by comparing the variance in moisture\nsupply values derived from analyses with variances derived from independent data on river streamflow\nrates over North America.\nData and Technique: Twice daily atmospheric analyses of global data from NMC for the period 1979-\n1987 are used. These analyses were interpolated from 2.5° X 2.5° grid squares to a Gaussian grid and\nthence to sets of T30 spectral coefficients. Wind and moisture fields were available at 12 levels in the\natmosphere. Monthly mean flux divergences were computed as in Boer and Sargent (1985). The wind\nfield was adjusted to remove spurious divergence. Moisture fluxes m were derived by multiplying\nadjusted horizontal winds V by specific humidity q: m = Vq. These moisture fluxes were averaged over\nthe month ( m then vertically integrated from the top of the atmosphere Pt to the earth's surface\nPs (using T30 topography) to give a total monthly average horizontal moisture flux M:\n=\n(1)\nconvergence of M at a point gives the sum of the rate of increase Q in atmospheric storage of\nThe\nmoisture and the difference between the rates of precipitation P and evaporation E over that point:\n-V.M=Q+P-E.\n(2)\nRasmusson (1967), Chen (1985) and others have computed, displayed and discussed the quality of\n-V.M and M as computed from analyses for various time periods. This paper inquires into the quality\nvariability of - D.M For this purpose an independent estimate of the right hand side\nof\nestimates\nof\nof equation (2) for moisture flux convergence is obtained using records of streamflow across North\nAmerica obtained from the U.S. Geological Survey and the Canadian Water Resources Branch. The use\nof streamflow records for this purpose raises several points. Data were used only from streams and\nrivers which are not regulated or controlled in any way and for which complete records for the period\n1979-1987 were available. Over a sufficiently long time streamflow rate R expressed as a depth or\nvolume of water per unit area per unit time can be represented as the difference between P - E and\nthe rate of increase S in soil (or other) moisture storage over that area:\n= P - E S.\n(3)\nThis representation of the relationship among these variables ignores lags between the supply of\natmospheric moisture to the surface and its appearance at the streamflow gauge and is valid only for\n455","long time periods. For this reason only inter-annual variances (comparing total moisture supply over\nan October-September water year) are considered in this paper. Thus atmospheric moisture flux\nconvergence which occurs in the winter as snow falls is correctly matched with runoff which occurs the\nnext summer as that snow melts. The right hand sides of (2) for atmospheric convergence and (3) for\nstreamflow runoff are comparable also only if Q and S can be ignored. This, too, is more plausible on\nan annual time scale, but future studies should look more closely at lags and storage terms. Streamflow\nstations can only be treated as a sampling of values for the right hand side of (3) for surface runoff. In\nan attempt to avoid contamination of that sample with unusual local conditions affecting runoff and at\nthe same time obtain values which did not include runoff from distant regions, rivers draining basins\nsmaller than 300 km 2 and larger than 3000 km 2 were eliminated. Annual averages of moisture flux\nconvergence computed from atmospheric analyses were interpolated onto the locations of the remaining\nstreamflow guages across North America (676 in all) allowing comparisons to be made with\ncorresponding annual average runoff rates for comparable locations. All variances and correlations at\neach point were among the eight (water) yearly averages in the period October 1979 - September 1987.\nResults: Inter-annual streamflow and moisture flux convergence variances (Fig. 1) have similar\nmagnitudes and both show a tendency for maxima on the coasts but atmospheric flux convergence\nvariance estimates are larger overall and a maximum in streamflow variance over the Gulf coast is\nreplaced in the flux convergence variance map with a maximum over the central plains where\nstreamflow variance has a minimum (Fig. 2a). Temporal correlation of annual anomalies in the two\nfields (Fig. 2b) was generally positive, especially over the northwest and northeast coasts but negative\nin some areas. With the thought that the excess of flux convergence variance might be concentrated in\nhigh (short) spatial wave numbers spectra of flux convergence and streamflow variances were computed\nfrom spatial correlation functions based upon correlating all possible station pairs over the eight annual\nvalues (cf. Julian and Thiébaux, 1975). The spectra (not shown) display higher variance in flux\nconvergence at all wavenumbers but particularly so at scales shorter than global wavenumber 20. The\ncorrelation of the two fields is weakly positive at all wavenumbers with no one band of scales\nresponsible for the negative correlations. Smoothing the two fields with an influence function\nappropriate to a T20 truncation halves the maximum difference in variance of the two fields (Fig. 3a)\nand expands the area over which they are positively correlated (Fig. 3b). Further smoothing\ncorresponding to T15 halves the difference in variance of the two fields once again but does not result\nin any significant additional improvement in their correlation.\nConclusions: Streamflow records provide a plausible corroboration of atmospheric moisture flux\nconvergnce computations. Inter-annual variance of moisture flux convergence anomalies over North\nAmerica as computed from NMC analyses is found to be larger than that of streamflows and weakly,\nbut generally positively correlated with them. Spatially smoothing the data appears to at least partially\nimprove the correspondence, reducing the difference in variance of the two fields and producing values\nwhich are more positively correlated.\nReferences:\nBoer, G.J. and Neil E. Sargent, 1985: 'Vertically integrated budgets of mass and energy for the\nglobe', J. Atmos. Sci, 42, 1592-1613.\nChen, Tsing-Chang, 1985: 'Global water vapor flux and maintenance during FGGE', Mon. Wea. Rev.,\n113, 1801-1819.\nJulian, Paul R. and H.J. Thiébaux, 1975: 'On some properties of correlation functions used\nin\noptimum interpolation schemes', Mon. Wea. Rev, 103, 605-616.\nLambert, Steven J., 1988: 'A comparison of operational global analyses from the European Centre for\nMedium Range Weather Forecasts (ECMWF) and the National Meteorological Center (NMC)',\nTellus, 40A, 272-284.\nRasmusson, Eugene M., 1967: 'Atmospheric water vapor transport and the water balance of North\nAmerica', Mon. Wea. Rev, 95, 403-426.\nRosen, Richard D. and D. A. Salstein, 1985: 'Effect of initialization on diagnoses of NMC large-scale\ncirculation statistics', Mon. Wea. Rev., 113, 1321-1337.\nTrenberth, Kevin E. and J.G. Olson, 1988: Evaluation of NMC global analyses: 1979-1987, NCAR\nTechnical Note, NCAR/TN-299+STR, 82pp.\n456","N.09\nNoOZ\n04\n.OEL\nM.OZL\nOIL\nM\nM.OOL\nM.OG\nNo.\nNo.\nOE'O\n08'9\n1.0E\nNoOZ\nOTL\n09\nM\nM.OEL\nM.09\nM o 021\nM o OLL\nM.OOL\n08\nM.OB\nSIT\n(e\nI\nLSt","60° N\n50°\n30° N\n2a\no W\n140\n60° W\n70° W\n120\nW\n110° W\n80°\nW\n100 W\n90° W\n20°\n60\nW\n140\n60° W\n120° W\n70° W\n110° W\n80°\nW\n100° W\n90° W\nFig. 2 a) Excess of inter-annual rms variance of moisture flux convergence over that of streamflows.\nUnits are mm/day.\nb) Temporal correlation of annual moisture flux convergence anomaly with annual streamflow\nanomaly based on the water years 1980-1987. Correlations greater than 0.4 are shaded.\n458","52\nN .09\nN .09\nNot\nN 1.08\nDO\nDE\nN.02\nM.09\n04\nM\nM.09\nM ,OEL\nM.OL\nM .001\no\nM .08 o\nM o OIL\nM o 001\nM .08\nN 1.09\nI\nN .09\nQ\nNot\n9°0\n9\nN .0€\n9E\nNo.\nM . 09\n01\nM\n9 i\nM 09\nM o OEL\nM.04\no\nM or\nM o 08\nM o ou\nM 0 ,00L\nM .08\n07 [elieds e pue inq 7 The OZ.L E d\ne\nJO\n657","A CLIMATIC CLOUD ATLAS FOR NORTH AMERICA*\nJohn E. Hanzis\nArthur E. O'Brien\nThe MITRE Corporation\nBurlington Road\nBeford, MA 01730\nThe objective of this study was to develop a long range planning tool to assess the impact of\ncloud cover on remote sensing from a high altitude platform about the North American continent\nand adjacent waters. Long-term records of cloud cover observations have been processed to\nprovide climatic representations of the frequencies of occurrence of three levels of cloudiness:\nclear (0 tenths cloud cover), clouded (1 to 7 tenths cloud cover), and overcast (8 to 10 tenths cloud\ncover). The cloud data used in this study was for altitudes up to 30,000 feet, and the time scales\nwere annual, seasonal, and monthly.\nThe data for this atlas was obtained from the United States Air Force Environmental\nTechnical Applications Center (USAFETAC) through the Staff Meteorologist Office at Hanscom\nAFB. The United States Air Force 3D-nephanalysis and RT-nephanalysis cloud archives were\nused as the source of the cloud data. The data, which is a compilation of cloud information from\nconventional surface and upper air reports and satellite data, has an averaging period of 1980 to\n1987.\nFrequency of occurrence contours of constant cloudiness were generated from the data with\nthe use of a graphics software package called CA-DISSPLA TM** Generation of contours from\nthe data was a three step process. The first step involved plotting the frequency of occurrence\ndata\nas demonstrated in figure 1. The second step smoothed the raw data to soften surface\nirregularities; this is illustrated in figure 2. Finally, the two-dimensional contours of constant\npercent frequency of occurrence were generated from the smoothed three-dimensional surfaces.\nMap 1, 2 and 3 are the annual contours for the three levels of cloudiness: clear, clouded, and\novercast, respectively. The solid lines represent 10 percent contours and the dashed lines represent\n5 percent contours. The monthly and seasonal contours are available from The MITRE\nCorporation in MITRE document M89-44.\n*This work was sponsored by USAF Electronics Systems Division (AFSC), Hanscom AFB, MA.\n** CA-DISSPLA- TM is a registered trademark of Computer Associates, Inc.,©1988, Computer\nAssociates International, Inc.\n460","Figure 1. Raw Data Plot\nFigure 2. Smoothed Data\n461","-40\n45","E9\"","50\n3.\nMap","GLOBAL TROPICAL FORCING OF THE EXTRATROPICAL PLANETARY WAVE FIELD\nTom Murphree\nDepartment of Meteorology\nNaval Postgraduate School\nMonterey, CA 93943\nAbstract: The sensitivity of the extratropical planetary wave response to the global spatial structure\nof anomalous tropical heating is studied in a simplified model of tropical-extratropical\nteleconnections.\nINTRODUCTION\nIn the past decade our picture of how anomalous tropical atmospheric heating might force\nperturbations in the extratropical planetary wave field has become much more complex. A number\nof modeling studies have shown that in order to explain the process by which such teleconnections\noccur it is necessary to understand much more than just simple planetary wave propagation on a\nsphere. Many of these studies assign major propagation and amplification roles to various features\nof both the tropical and extratropical background flow. These might be, for example, extratropical\nbarotropic instabilities (Simmons et al, 1983), highly responsive tropical regions (Branstator, 1985),\nvertical wind shear (Lim and Chang, 1986), the sign and zonal gradient of the tropical zonal wind\n(Webster and Chang, 1988), vorticity advection by the divergent wind (Sardeshmukh and Hoskins,\n1988), and subtropical westerly barriers and avenues for meridional energy propagation (Murphree\nand Weare, 1989).\nMany of the conclusions from these, and from a number of other simplified modeling studies,\nare based on the use of idealized tropical forcing structures, usually a localized or monopolar\nsource compensated by a broadly distributed sink and / or dissipation elsewhere. Presumably, such\nmonopole forcings are motivated by a desire for simplicity and the basically monopole appearance\nof the tropical Pacific sea surface temperature anomalies (SSTA) associated with El Niño/Southern\nOscillation (ENSO) events. However, observations and general circulation model (GCM) results\nshow that the tropical tropospheric heating anomalies associated with ENSO events are far from\nmonopolar (e.g., Boer, 1985; Mureau et al, 1987). Furthermore, since the nature of the spatially\nvarying background flow appears to be important to the teleconnection process, it would seem to be\nimportant in simplified model or GCM studies of tropical forcing of the extratropics to specify, or\ncalculate, tropical tropospheric heating anomalies with a realistic global spatial structure.\nIn this paper, the preliminary results of a study of the sensitivity of a simplified model to\nthe global, or overall, tropical forcing spatial structure are presented. The emphasis is on the\nsensitivity of the extratropical response to the method of compensating a localized heating region\n(e.g., compensating with a zonally distributed or localized cooling).\nAs with most simplified model studies of teleconnections, the primary goal of this work is\nto clarify the mechanisms operating in the real atmosphere and in GCMs. An additional goal is to\nunderstand possible weaknesses in GCM simulations. For example, the conversion within a GCM of\na specified SSTA into a global tropical convective heating anomaly, which is an especially\nproblematic process. This process is presumably more difficult to simulate far from the SSTA. If\nthe global, and not just local, structure of the heating anomaly is significant in teleconnection\nmechanisms, then it may be important for GCMs to correctly simulate this overall structure before\nthey are applied to the teleconnection issue.\nMODEL\nThe model is based on the nonlinear shallow water equations, in spectral formulation with R25\nresolution, on a global spherical domain. The equivalent depth is 2000m, corresponding to a deep\ninternal mode. The basic state fields are European Centre for Medium-Range Weather Forecasting\n(ECMWF) 1979-87 mean global analyses of northern winter (DJF) 200mb streamfunction, velocity\npotential, and geopotential. A basic state forcing is calculated and applied to the equations to\nmaintain the basic state flow in the absence of any perturbation forcing. Dissipation is relaxation\nA-1","damping back to the initial conditions with a time scale of 15d, plus biharmonic diffusion with a\nnumerical constant of The equations are integrated in time and give steady state solutions\nafter about three to eight model weeks. To represent the response to persistent tropical heating\nperturbations, a steady forcing term is applied to the geopotential equation. Regions of positive\nperturbation forcing are localized and are compensated either zonally or locally. This means that\neither:\n(i) the negative forcing is broadly distributed zonally, so that the zonal mean forcing is zero and\nthe overall forcing is approximately monopolar; or\n(ii) the negative forcing regions are also localized, so that the overall forcing pattern is dipolar or\nmultipolar.\nIn all cases, the global mean forcing is zero. All localized forcing regions are centered near the\nequator and are oval in shape, spanning a maximum of 250 of latitude and 50° of longitude. Within\nthese regions, the forcing ranges from zero at the edges to the equivalent of 5K/d heating or\ncooling at the centers. The response to the total perturbation forcing is calculated as the solution\nwith this forcing minus that without.\nRESULTS\nThe geopotential height response to monopolar forcing located at three different Pacific\nlongitudes is shown in Fig. 1. For the case of west or central Pacific forcing, the response contains\na low over the north Pacific, a high over North America, and a relative low over the north Atlantic.\nFor the east Pacific forcing, this pattern is reversed. In the southern hemisphere, the pattern of lows\nand highs tends to shift eastward with the monopole. The much more zonally asymmetric basic state\nin the northern hemisphere than in the southern, especially in the subtropical jets (not shown),\nimplies that the basic state flow strongly controls the northern extratropical response. Additional\nexperiments with the monopole shifted around the entire tropical belt (not shown) show that the\nnorth Pacific and North American areas are prominent response nodes for almost all monopole\nlongitudes. Further, this north Pacific-North American (NPNA) response pattern is relatively fixed\nin space and of one phase (negative over the north Pacific, positive over North America) when the\nmonopole is located within the 40E - 180E sector, and is relatively fixed and of opposite phase\nwhen the forcing is within the 120W - 40W sector. Thus the zonal asymmetry of the basic state\nappears to constrain the response so that, within these two broad sectors, the NPNA response is only\nmoderately sensitive to the tropical monopole forcing longitude. These results are broadly consistent\nwith several previous studies (e.g., Simmons et al, 1983; Palmer and Mansfield, 1984; Yamazaki,\n1988; Branstator, 1989)\nObservations show that anomalous heating in the tropical west Pacific is often accompanied\nby anomalous cooling in the east Pacific, and vice versa, corresponding to amplification, or\ndeamplification, of the Walker circulation (e.g., during ENSO events, Mureau et al, 1987). Thus it\nis interesting to consider the model response to dipolar forcing. The responses to a positive monopole\nin the west Pacific, a negative monopole in the east Pacific, and a positive-negative west-east dipole\nare shown in Fig. 2a, b, and c, respectively. Both of the monopole responses show NPNA patterns\nof the same phase, while the dipole response indicates a constructive interference of the responses\nto the two centers of the dipole. The low-high-low pattern between the north Pacific and north\nAtlantic is especially pronounced in the dipole solution.\nThe marked impact of the dipole forcing on the amplitude of the response in this and other\nmodel experiments indicates that it is important to account for both localized sources and sinks\nwhen analyzing tropically induced teleconnections. Further, these results imply that one should be\ncautious in associating a given extratropical response with a particular localized tropical forcing\n(cf. Trenberth et al, 1988). This is because the response may be similar for: (i) a number of\ndifferent forcings located within a large longitudinal sector; and (ii) even very widely separated\nforcings, if the signs and relative locations of the forcings are suitable.\nTo get a sense of how realistic these dipole responses are, the model results are compared in\nFigs. 3 and 4 with observed anomalies. Fig. 3 compares the response to a west Pacific - east\nPacific heating-cooling dipole with the observed 300mb streamfunction anomaly during DJF 1983-\n84. During this time, the outgoing longwave radiation anomaly (OLRA) was negative in the west\nPacific and positive in the central and east Pacific. The two panels of Fig. 3 show a number of\nA-2","qualitative similarities, including anticyclones in the north African - south Asian region, paired\ncyclones in the tropical east Pacific, cyclones east of Japan, and anticyclones near the west coast\nof North America.\nFig. 4 is like Fig. 3 but for a case representing anomalous west Pacific cooling and east\nPacific heating. The streamfunction anomaly is for DJF 1982-83, when the OLRA was positive in\nthe west Pacific and negative in the central and east Pacific. Again, there are qualitative similarities\nbetween the model response and observed anomalies; in particular, anticyclone and cyclone pairs\nin the tropical Pacific and Atlantic, respectively, and cyclonic regions in the north Pacific. Note also\nin comparing Figs. 3 and 4 that there is an approximate antisymmetry in both the response and\nobservations.\nIn many ENSO or anti-ENSO events, the observed tropical heating anomalies are distinctly\nmultipolar. A dipolar heating anomaly structure across the Pacific is commonly supplemented by\nsmaller localized anomalies in the Indian, South American, or African sectors (cf. Mureau et al,\n1987). Some studies have suggested that the extratropical response may be especially sensitive to\nforcings located in certain regions (e.g., the Indian Ocean - Indonesian sector, Branstator, 1985).\nTo test this possibility, several experiments with multipolar forcing have been done. One example\nis shown in Fig. 5. A comparison of the responses to dipole and multipole forcings (Fig. 5a,b)\nshows that the smaller Indian and Atlantic forcings are of secondary importance, but have some\nimpact on the NPNA and Eurasian responses.\nOther experiments with smaller forcings outside the Pacific also show generally small affects\non the overall extratropical response, especially the NPNA response. This may be explained by the\nlow sensitivity of the response to forcing location when the forcing is located within the 40E-180E\nsector or the 120W-40W sector. That is, different forcings located within one of these two sectors\ntend to produce responses with similarly situated nodes, at least across much of the NPNA region.\nThus, if the major dipoles occupy these sectors, then small additional forcings within these sectors\nwill tend to have, for the NPNA region, only small quantitative impacts.\nCONCLUSIONS\nA simplified model with varying tropical forcing spatial structure gives plausible extratropical\nplanetary wave responses (cf. Wallace and Gutzler, 1981; Simmons et al, 1983; Sardeshmukh and\nHoskins, 1988). The northern hemisphere winter extratropical response is relatively insensitive,\nespecially in the NPNA area, to the forcing longitude when the forcing is located within one of\ntwo longitudinal sectors (40E-180E or 120W-40W). This is apparently due to the controlling effects\nof the strong zonally asymmetric basic flow on wave amplification and propagation. Since the\nstrength and zonal asymmetry of the basic flow varies through the year, one may expect this\ninsensitivity to be most apparent in winter.\nThe method of compensating localized tropical forcing is important to the amplitude and\nphase of the extratropical response. In particular, balancing localized heating with realistically\ndistributed localized cooling may significantly impact the response. This is in part because, on a\nglobal basis, the longitudes of the localized forcings are important, with the responses to very widely\nseparated forcings having the potential to significantly interfere with each other in certain\nextratropical areas (e.g., the NPNA area).\nThe possibility of such interference implies that a large variety of tropical forcing structures\nmay contribute to a given extratropical response. Thus it is important to account for the overall or\nglobal perturbations of tropical heating when simulating or interpreting observed extratropical\nresponses to such disturbances. In simple models of tropically induced extratropical wave anomalies,\nthe tropical forcing patterns should be more realistic than those which have been used in most studies\nso far. In GCM studies, the model should calculate reasonable tropospheric heating throughout the\ntropical band (and not just in the area of the prescribed SSTA, for example).\nREFERENCES\nBoer, G., 1985: Modelling the atmospheric response to the 1982/83 El Niño. Coupled Ocean -\nA-3","Atmosphere Models, J. C. J. Nihoul, Ed., Elsevier.\nBranstator, G., 1985: Analysis of general circulation model sea-surface temperature anomaly\nsimulations using a linear model. Part I: Forced solutions. J. Atmos. Sci., 42, 2225-2241.\n1989: Low-frequency patterns induced by stationary waves. Submitted to J. Atmos. Sci.\nHorel, J. D., and J. . M. Wallace, 1981: Planetary scale atmospheric phenomena associated with\nthe southern oscillation. Mon. Wea. Rev., 109, 813-829.\nLim, H., and C.-P. Chang, 1986: Generation of internal and external mode motions from\ninternal heating. Effects of vertical wind shear and damping. J. Atmos. Sci., 43, 948-957.\nMureau, R., J. D. Opsteegh, and J. S. Winston, 1987: Simulation of the effects of tropical\nheat sources on the atmospheric circulation. Mon. Wea. Rev., 115, 856-870.\nMurphree, T., and B. C. Weare, 1989: Large scale tropical - extratropical interactions in a\nmodel atmosphere. Proceedings of the Thirteenth Annual Climate Diagnostics Workshop.\nPalmer, T. N., and D. A. Mansfield, 1984: Response of two atmospheric general circulation\nmodels to sea-surface temperature anomalies in the tropical East and West Pacific. Nature, 310,\n483-485.\nSardeshmukh, P. D., and B. J. Hoskins, 1988: The generation of global rotational flow by\nsteady idealized tropical divergence. J. Atmos Sci., 45, 1228-1251.\nSimmons, A. J., J. M. Wallace, and G. Branstator, 1983: Barotropic propagation and\ninstability, and atmospheric teleconnection patterns. J. Atmos. Sci., 40, 1363-1392.\nTrenberth, K. E., G. W. Branstator, and P. A. Arkin, 1988: Origins of the North American\ndrought. Science, 242, 1640-1645.\nWallace, J. M., and D. S. Gutzler, 1981: Teleconnections in the geopotential height field\nduring the Northern Hemisphere winter. Mon. Wea. Rev. 109, 785-812.\nWebster, P. J., and H. R. Chang, 1988: Energy accumulation and emanation regions at low\nlatitudes: Impacts of a zonally varying basic state. J. Atmos. Sci., 45, 803-829.\nYamazaki, K., 1988: Influence of sea surface temperature anomalies over the Indian Ocean\nand Pacific Ocean on the tropical atmospheric circulation. J. Met Soc. Japan, Ser. II, 66, 797-806.\n10.0\n0.0\n0.0\n0.0\na\nb\nFig. 1. Geopotential height responses to\nmonopole forcings. Contour interval = 10m.\nForcing signs and centers marked by +. For\nforcing center: (a) in west Pacific, (b) at\ndateline, (c) in east Pacific.\n20.0\nC\nA-4","Q\ne\n0.0\n0.0\nb\na\n20.0\nFig. 2. Geopotential height responses to\n20.0\nmonopole and dipole forcings. Contour\n10\ninterval = 10m. Forcing signs and centers\nmarked by +, - For forcing center(s): (a) in\nwest Pacific, (b) east Pacific, (c) west and\neast Pacific.\n20.0\nC\n200\nTE.\nFig. 3. (a) Streamfunction response to dipole\nforcing. Contour interval = 100m2/s. Forcing\n200\nsigns and centers marked by +, -\n.\n(b) Observed 300mb streamfunction\nanomaly for DJF 1983-84 when anomalous\nheating was positive in west Pacific and\nnegative in central and east Pacific. Contour\ninterval = 5x10 0 m /s. From Mureau et al\n(1987).\n5\n0\nb\nA-5","00\nFig. 4. (a) Streamfunction response to dipole\n0\nforcing. Contour interval = 10 0 m²/s. Forcing\na\nsigns and centers marked by +, -\n(b) Observed 300mb streamfunction\nanomaly for DJF 1982-83 when anomalous\nheating was negative in west Pacific and\npositive in central and east Pacific. Contour\ninterval = 5x10 0 m /s. From Mureau et al\n20\n(1987).\n45\n(210\nGo\nb\nFig. 5. Geopotential height responses to (a)\na\nmultipole and (b) dipole forcings. Contour\ninterval = 10m. Forcing signs and centers\nmarked by +, - . Indian and Atlantic sector\nforcings in (a) are one fourth the magnitude\n40.0\nof those in Pacific sector.\nLas\n7500\n0.0\non\nb\nA-6","The sensitivity of the high-latitude response to tropical\nheating anomalies in real-data extended range forecasts\nL. E. Buja, J. Paegle,\nUniversity of Utah\nD. P. Baumhefner\nNCAR\nGeisler et al. (1985) have studied the response to tropical sea-surface\ntemperature anomalies in the Community Climate Model (CCM) of the National Center\nfor Atmospheric Research (NCAR). They found that a \"Pacific North American\"\n(PNA) type of geopotential height response occurs in perpetual season experiments\nwherein solar and other geophysical parameters are maintained at values\nappropriate for January over the entire 1200 day duration of their forecasts.\nThis PNA response pattern is characterized by relatively low geopotential heights\ncentered in the Gulf of Alaska and near the east coast of North America, with\nan anticyclonic, elevated height response over Central and Western North America.\nOne goal of the present research is to ascertain whether similar responses\nare perceived in more recent versions of this model (the \"CCM1\") that are\ninitialized with real data, and integrated for approximately one month. Closely\nrelated goals are to determine when the transient response begins to approach\nthe perpetual season response and how sensitive it is to temporal oscillations\nand longitudinal displacements of the tropical heating.\nAll present integrations are performed with the NCAR CCM1, a spectral model\nrun at R15 truncation, using 12 vertical sigma levels, convective adjustment\nparameterization of cumulus convection, and relatively complete treatments of\nradiation. The runs extend over 30 to 36 days duration, and are initialized from\neight different initial states selected from the 00:00 UTC observations on 1\nJanuary of each year from 1977 through 1984. This number of different cases is\nconsidered in order to minimize sensitivity of the response to a particular\ninitial state, and to enable estimates of statistical reliability.\nEach case is run for two separate forecasts. The first of these constitutes\nthe control, wherein the model is integrated in its standard form for 36 days,\nand the second is the experiment in which a heating term is added to the\nthermodynamic equation over the eastern tropical Pacific Ocean. This term is\nintended to simulate the effect of a strong latent heating anomaly supposed to\ndevelop suddenly at the start of the forecast, and persist for its duration.\nThe peak, column averaged value of this anomaly is approximately 5°C/day,\ncorresponding to the latent heat released by approximately 2.7 cm rain per day.\nThe anomaly decreases radially outward from the center of this peak in Gaussian\nprofiles in latitude and longitude with half width on the order of 1000 km in\nlatitude and 2000 km in longitude.\nThe response of the heated experimental forecast with respect to the control\nis described as the case averaged difference between them. Figure 1 presents\nthe time evolution of this measure of the 200 mb height response. It shows\nrising heights in the region of experimental heating for the day 1-5 average and\nNorth American anticyclonic response by the day 11-20 average. Weak cyclonic\nA-7","responses are also evident in the latter period over the Gulf of Alaska and\nnorthern portions of the North Atlantic. This incipient PNA response is even\nmore evident in the day 21-36 average, and also characterizes the full 36 day\naverage. We conclude that a PNA type of response to eastern tropical Pacific\nheating anomalies is evident by about two weeks, and becomes fully established\nwithin the time frame of an extended range forecast.\nIn order to investigate the effect of transiency of the tropical heating,\nthe previous set of experiments were repeated, imposing a time variation of the\nheating as depicted in Fig. 2. The time average of this heating equals the\nconstant value imposed in the previous experiments, it has maxima on days 6 and\n30, and vanishes between days 12 and 24. Fig. 3 displays the 1-36 day averaged\nresponse in this transiently forced experiment (top), the 1-36 day averaged\nresponse of the previously steadily forced experiment (middle) and the difference\nof these responses (bottom). Both the transiently forced and steadily forced\nexperiments have PNA signals over North America, and the difference between the\ntwo responses is less than either one. We conclude that the presence of a high-\nlatitude, 36 day averaged PNA forecast response to Eastern Pacific tropical\nheating is not very sensitive to time variations of the heating.\nThe final set of experiments is designed to investigate the sensitivity\nwith which the remote PNA response depends upon the longitudinal placement of\nthe tropical heating anomaly. For this purpose we have analyzed the 200 mb\nheight response produced by the experiments of Buchmann et al. (1990) The\nlatter consisted of 30 day integrations of the CCM1 wherein the heat source was\nplaced over the tropical Atlantic Ocean, rather than the Eastern Pacific, as in\nthe case corresponding to Fig. 1. The same set of cases was integrated as in\nthe previous experiments, and the case averaged response over the 30 day duration\nof the forecasts is depicted in the top panel of Fig. 4.\nThis displays\nrelatively low heights over the Gulf of Alaska and the northern areas of the\nNorth Atlantic Ocean, separated by an anticyclonic response centered at high\nlatitudes of North America. This response pattern could also be characterized\nas a PNA pattern. We conclude that the high-latitude PNA forecast response is\nnot very sensitive to the longitudinal position of the tropical heating anomaly.\nThis conclusion is consistent with Geisler et al (1985) and Buja (1989), both\nof whom found little sensitivity of the PNA response to longitudinal\ndisplacements of tropical heating anomalies within the Pacific Ocean.\nIn summary, we conclude that a PNA response to anomalous tropical heating\nis relevant to time averaged forecasts of monthly duration with the NCAR CCM1,\nthat this response appears in approximately two weeks, and that its time averaged\nstructure is relatively insensitive to both temporal oscillations and\nlongitudinal shifts of the heating. It should be emphasized that these\nconclusions have been demonstrated only for high latitude (poleward of\napproximately 50°) height response. More detailed examination of the wind field\nresponse by Buja (1989) shows that the momentum response is strongest in tropical\nand subtropical latitudes, and this is much more sensitive to the structure of\nthe heating than is the high latitude geopotential height response.\nAcknowledgements: This research was supported by NSF grants ATM 8611952 and\nATM 8905369 to the University of Utah. The model forecasts were made at NCAR\nwhich is supported by the National Science Foundation.\nA-8","References\nBuja, L. E. , The evolution and time scales of the atmospheric response\nto\ntropical diabatic heating events. Doctoral dissertation, the University\nof Utah. 210 pp.\nBuchmann, J. , J. Paegle, L. E. Buja, and R. E. Dickinson, 1990: Further FGGE\nforecast experiments for Amazon Basin rainfall. Mon. Wea. Rev., , in press.\nGeisler, J. E. M. L. Blackmon, G. T. Bates, and S. Munoz 1985: Sensitivity of\nJanuary climate response to the magnitude and position of equatorial\nPacific sea surface temperature anomalies. J. Atmos. Sci., 42, 1037-1049.\nA-9","(A) Days 1-5 -\nSSEP Z 200mb\n(C) Days 11-20\n14.6\n3\n⑇\n24.0\nL\n7.48\nH\nH\nt\n36.6\n32.4\n667\n-95\n57.8\n22.4\n23.9\n(B) Days 5-10\n(D) Days 21-36 -\nillin\n79.0\n9.02\nID\n209E\nHH\n25.\n82.4 35.0\n-20\n2\n92.4\nilli\n53\n(E) Days 1-36\na\n49.17\n9.8\n18\n50th\nFig. 1 Case averaged 200 mb geopotential height response for tropical Eastern\nPacific heating. Contour interval is 50 m, and regions displaying more than 99%\nstatistical significance according to T test are shaded.\n1.0\n.9\n.5\n.7\n.6\n80\n.5\n.4\n.3\n.3\n. &\no\n0\n5\n10\nIS\n20\n25\n30\n35\n10\nDAYS\nFig. 2 The temporal structure of the transient heating.\nA-10","(A) Days 1-36\nSTEP ZR 200mb\n-54\n11.8\nIOD.\n00\n00\nH\nb\n23.9\n5.30\n98\n2\nH\n24,3\n-\n70.7\n68.3\n34.8\n(B) Days 1 -36 -\nSSEP\n100\n58\n7\n44\nH\n49.7\n49.8\n18\n50.2\n18.800\nH\nDO\nH\n2.7\nSTEP-SSEP\n(C) Days 1 - -36\n24\nH\n1.93\n84.2\n6.37\n-59.7\n68.-S\nH\nH\n.974\n2.31\n1.33\n8.59\n71.8\nH\nH\n-58.0\n13,5\n24.8\n70.6\nFig. 3 Time, case averaged 200 mb geopotential height response A) transient\nheating, B) steady heating, C) difference A - B. . Contour interval is 50 m, and\nshading indicates statistical significance in excess of 99%\nA-11","SSNA ZR 200mb\n(A) Days 1-30 -\n100,\n-12.8\n12\nH\nPPQ.\n16\nHI\n44\n8\n28%\nBE\n0\nOR\nH\nAA\n40:\nL.\nH\n209\nH\n105\n39.4\n16.\n50\n30RO\n121.\nSSEP\n(B) Days 1 -36 -\n100\n58.51\nSeTD\n7\nSD\nH\n44\n0\n100:\n49.7\n49.8 118 50.2\n18.00\nH\nSSNA-SSEP\n(C) Difference\n35\n-46. ₽\n6\n50.5\n188\n-48\n8\n183\n152\n97.7\nA\nH\nH\n27\n80.4\n[05-\n41.6\n27.6\n-22.\nFig. 4 Time, case averaged 200 mb geopotential height response A) tropical\nAtlantic heating, B) tropical Pacific heating\nC) difference A - B. Contour interval is 50 m, and shading indicates statistical\nsignificance in excess of 99%\nA-12","MONTE CARLO FORECASTING WITH DYNAMIC STRATIFICATION\nSiegfried Schubert, Max Suarez and Jae-Kyung E. Schemm\nGlobal Modeling and Simulation Branch\nNASA/GSFC Code 611\nGreenbelt, MD 20771\nOne of the problems in implementing the Monte Carlo forecasts (MCFs) using\natmospheric general circulation models (GCMs) is the expense involved in generating a large\nnumber of forecasts. Here, we propose a method to make the MCF more efficient by\nemploying an optimal sampling strategy to reduce the number of ensemble forecast members.\nThe stratified sampling is done with a large number of the full GCM forecasts of a very short\nperiod preferably less than 1 day and the forecasts are extended to longer periods on a limited\nnumber of samples. The method provides potentially large reductions in the variance of the\nmean and variance estimates compared to a more traditional random sampling approach.\n1.\nStratified Sampling\nFor a simple stratified sampling from a sample of size n with L strata, unbiased\nestimates of the mean and variance of a variable y are (Yates, 1981)\n(1)\nand\n(2)\nwhere\nand\nHere, nh refers to the sample size in stratum h, Nh to the number of members in the population\nbelonging to stratum h, and\nN=ENn N\nh=1\nn=Enn\nThe population size N is assumed to be large but finite.\nA-13","The sample nh is chosen to minimize the variance of the sample mean by adopting the\nNeyman allocation such that (Cochran, 1963),\n(3)\nwhere Sh is the population variance of stratum h. (From here on, the capital letters refer to\npopulation values and the lower case letters refer to sample values.) Eq. (3) states that the\nsample size nh should be larger for those strata with more members and larger variance. The\nimprovement in the estimate of the mean for stratified sampling using Eq. (3) compared to\nsimple random sampling is (Chochran, 1963)\n(4)\nwhere S is the population standard deviation, the subscript r refers to an estimate from simple\nrandom sampling and\nIn practice, most of the contribution in making the ratio R small comes from the ratio\nof the variance between strata to that within strata. This suggests that in order to optimize the\nsampling, the samples should be grouped in a way to minimize the average within strata\nvariance. One of the clustering algorithms with such criteria is the KMEANS algorithm and it\nis used in this analysis.\nEstimates of Error Growth\n2.\nIn this study, two methods of estimating future error distribution are examined. Both\nmethods are based on the initial error growth distributions obtained from an ensemble of short\nterm forecasts with a GCM. The baseline method is to simply stratify based on the error\ndistribution from an ensemble of the short term forecasts and no attempt was made to\nextrapolate this distribution further into the future. The second approach is to extrapolate the\nerror distribution into the future based on Monte Carlo estimates of linear dynamic operator\ndeveloped by Leith (1980).\n3.\nTest with GCM Forecasts\nIn testing the stratified sampling strategy, a two-level GCM with primitive equation\nrepresentation of atmospheric flow was employed. The model was run in perpetual Northern\nwinter mode. A detailed analysis of the model error growth behavior is given in Schubert and\nSuarez (1989). In this study, we superimposed 100 perturbations onto a randomly chosen\ninitial state from a 9000 day control run. The population values of the mean and variance were\nobtained from the ensemble of 100 runs (it is assumed that N=100 is large enough to provide\nthe population values). The 500 mb height field was chosen as the test variable for the analysis\nA-14","and it was decomposed into empirical orthogonal functions (EOFs). The results reported here\nare based on the anlysis of the first 25 EOF modes, which explain more than 99% of the total\nvariance.\nIn Fig. 1, the reduction of variance R at forecast day 1 is shown as a function of the\nnumber of clusters. It is evident that most of the reduction in variance (approximately 60%)\noccurs in the first EOF mode with just two clusters. Additional reduction is provided by the\nsubsequent EOF modes as the number of clusters increases. However, the reduction in\nvariance levels off after 4 or 5 clusters. Based on these results and after examining the\nstructure of the clusters, the number of clusters is limited to three in the following analysis.\nFig. 2 shows the reduction of variance as a function of forecast days with three\nclusters when the clustering is based on actual forecasts from day 0 to day 3. Stratified\nsampling is performed with clusters fixed throughout the rest of the forecast period up to day\n6. They are compared to the potential improvement which could be achieved when the clusters\nare determined independently at each forecast period. In general, clustering based on forecasts\nat a later period gives more reduction of variance as one might expect and the persistence of the\nday 1 clusters gives approximately 15 % reduction up to day 5.\nExtrapolation of forecast error distributions based on a large number of short term\nforecasts was attempted by utilizing Monte Carlo estimates of the linear dynamic operator\n(Leith, 1980). The linear operator was estimated based on forecast error growth distributions\nbetween 6 hour and 12 hour forecasts. This forecast period was chosen to avoid the rapid error\ngrowth during the initial adjustment period and yet retain some of the error growth pattern of\nshort time scale features. With the linear operator, forecasts into the future can be estimated in\n6 hour increments. Cluster analysis based on the forecast estimates was performed in the same\nmanner as with the actual forecasts. Fig. 3 illustrates the reduction of variance with clusters\nobtained from the estimated forecasts. In this case, the persistence of clusters was applied\nbackward in time up to 12 hours in addition to the forward application because the linear\noperator estimates could be obtained once the 12 hour forecasts were made. Clusters based on\nforecast estimates up to day 3 provide the reduction of variance comparable to that obtained\nwith the clusters based on the actual forecasts. It is apparent, at least in this case, that the linear\noperator estimates provide reasonable basis for sampling strategy for forecasts further into the\nfuture\n4.\nConclusions\nThe test case presented here indicates that significant improvements in the variance of\nthe sample mean and variance are possible with a stratified sampling approach compared to the\nconventional random sampling. The simple persistence approach along with the linear operator\nextrapolation into forecast period longer than 1 day achieves approximately half of the potential\nimprovement of the independent sampling approach. Several other cases have been tested\nusing different intitial states and initial error distributions. The results show that the potential\nimprovements are similar in all cases. However, the degree of improvement of the persistence\napproach depends on the initial conditions and initial error distributions. The linear operator\nextrapolation method provides some additional improvements.\nA-15","5.\nReferences\nCochran, W. G., 1963: Sampling Techniques. John Wiley & Sons, New York.\nLeith, C.E., 1980: Monte Carlo Estimation of Local Initial Error Growth.\nWorkshop on stochastic dynamic forecasting. ECMWF, 9 - 30.\nSchubert, S. D. and M. Suarez, 1989: Dynamical Predictability in a Simple\nGeneral Circulation Model. J. of Atmos. Sci., 46, 353 - 370.\nYates, F., 1981: Sampling Methods for Censuses and Surveys. Macmillan\nPublishing Co., New York.\n1.2\n1.0\n3\n0.8\n0.6\n2\n2\n3\n0.4\n1\n0.2\n0.0\n1\n2\n3\n4\n5\n6\n7\n8\n9\n10\nNUMBER OF CLUSTERS\nFig. 1 The reduction in variance (R) as a function of the number of clusters at forecast day 1.\nThe heavy solid line is for the total variance of 25 EOF modes. The numbered lines refer to the\nfirst 3 EOFs.\nA-16","1.0\n0.8\n0.6\no\n1\n2\n3\n4\n5\n6\nFORECAST DAY\nFig 2. The reduction in variance (R) as a function of the forecast day for the total variance. The\nheavy solid line at the bottom is for the potential reduction with the clusters determined at each\nforecast day. The other lines are for the clusters of day 0, 12 hours, and days 1 to 3 persisting\nup to day 6. The starting point of each curve indicates the forecast period of the clusters.\n1.0\n0.8\n0.6\n0\n1\n2\n3\n4\n5\n6\nFORECAST DAY\nFig 3. Same as Fig. 2 except that the clusters are deterimined with the forecast estimates of the\nlinear operator extrapolation. The labels * and 1 to 4 indicate the forecast period of clusters at\n12 hours and days 1 to 4.\nA-17\nU.S. GOVERNMENT PRINTING OFFICE: 1990 - 261 - 915 20503","Let Us Be Your Guide\nCommerce\nPublications\nUpdate\nthe latest publications and trade, press\n0\nto on business, the economy, technical\nreleases trends, scientific and from the\ncensus developments, and related topics\nU.S. Department of Commerce.\n2 weeks, Commerce Publications from the\nEvery lists dozens of new titles Census\nUpdate International Trade Administration, Standards,\nBureau, National Bureau of\nNational Oceanic and Atmospheric of\nAdministration, and other areas the\nEach listing includes number\nof Department. price, and complete ordering major\ninformation pages, and-in the case of the\nreports-a brief description of\ncontents.\nnow to take advantage of this easy-to-\nBegin use guide to a broad range of Government\ninformation.\nsubscription to the Commerce and\nPublications Enter your Update by completing\nmailing the order form below, today.\nSuperintendent of Documents Subscriptions Order Form\nCharge your order.\nOrder Processing Code\n* 6370\nVISA\nMasterCard\nIt's easy!\nYES, Subscriptions of please COMMERCE PUBLICATIONS All prices include UPDATE, regular domestic for $21.00 postage each and per handling year. and are subject to change.\nsend me the following indicated subscriptions:\nYou can fax your order to (202) 275-0019.\n1. The total cost of my order is $\nInternational customers please add 25%\nPlease Type or Print\n3. Please choose method of payment:\nCheck payable to the Superintendent of Documents -\n2.\n(Company or personal name)\n(Additional address/attention line)\nGPO Deposit Account\nVISA or MasterCard Account\n(Street address)\nThank you for your order!\n(City. State. ZIP Code)\n(Credit card expiration date)\n( phone including area code)\n)\nMail (Daytime To: Superintendent of Documents, Government Printing Office. Washington, D.C.\n(Signature)\n20402-9371\n4.\n12/89"]}