Predicting Snow-to-Liquid Ratio in the Mountains of the Western United States

Details

• Journal Title:
  Weather and Forecasting
• Personal Author:
  Veals, Peter G. ; Pletcher, Michael ; Schwartz, Andrew J. ; Chase, Randy J. ; Harnos, Kirstin ; Correia, Jimmy ; Wessler, Michel E. ; Steenburgh, W. James
• NOAA Program & Office:
  NWS (National Weather Service) ; HPC (Weather Prediction Center) ; OAR (Oceanic and Atmospheric Research) ; WPO (Weather Program Office) ; CIRA (Cooperative Institute for Research in the Atmosphere)
• Description:
  The snow-to-liquid ratio (SLR) and its inverse, snow density, are crucial for forecasting snowfall in numerical weather prediction models and for estimating snow water equivalent (SWE) on the ground using remote sensing. SLR also varies widely in space and time, making it challenging to forecast accurately, particularly in the heterogenous terrain and climate of the mountains of the western United States. This study utilizes high-quality, manually collected measurements of new snowfall and new SWE from 14 mountainous sites across the region to build multiple linear regression (MLR) and random forest (RF) algorithms to predict SLR as a function of atmospheric variables. When an MLR algorithm is trained on a simple combination of wind speed and temperature from either the ERA5 reanalysis, the GFS, or the High-Resolution Rapid Refresh (HRRR), it predicts SLR with considerably more skill than existing SLR prediction methods. When a more extensive set of variables is considered, the skill improves further. The variables used to achieve the most skillful prediction of SLR are temperature, wind speed, relative humidity, specific humidity, maximum solar altitude angle during the observing period, convective available potential energy (CAPE), and HRRR quantitative precipitation forecast (QPF). When an RF algorithm is trained using these variables, it can predict SLR with R2 = 0.43 and mean absolute error (MAE) = 2.94. For the existing SLR prediction techniques currently used in operations, R2 ranges from 0.04 to 0.23 and MAE ranges from 4.01 to 9.45. Therefore, the algorithms built in this paper can drastically improve SLR prediction over the mountains of the western United States.
• Keywords:
  Machine Learning Numerical Weather Forecasting Snow Winter Machine Learning Numerical Weather Forecasting Cool Season Forecasting Techniques Numerical Weather Prediction Snowfall
• Source:
  Wea. Forecasting, 40, 1825–1838
• DOI:
  https://doi.org/10.1175/WAF-D-24-0233.1
• Format:
  PDF
• Document Type:
  Journal Article
• Funding:
  Grant no. NA19OAR4590137 ; Grant no. NA22OAR4590531 ; Grant no. NA23OAR4310278
• Rights Information:
  Other
• Compliance:
  Submitted
• Main Document Checksum:
  urn:sha-512:612155e628198651628cd690cc4f4681e93295bed98bef2c23ed1a24b53b38a38e2297bc8cf4554eab69c324d25f1258e42e3adfee113439539b88e01e1cb5ee
• File Type: