This evaluation assessed the content and accuracy of hydrologic messages issued by the Denver WSFO, and the sensitivity ofthe supporting hydrologic applications. "Content" was characterized by attributes measuring message detail on spatial, temporal, and magnitude scales, as well as completeness ofthe message. "Accuracy" was assessed categorically using radar-rainfall imagery coupled with the spatial characterization ofthe messages. Accuracy ofthe radar-rainfall estimates ofbasin average precipitation and sensitivity of the ADVIS flood forecasting procedure were also addressed. The message coding method produced a content "score" on various categorical scales of characterizing message detail (spatial, temporal, magnitude) and completeness. These scores were then reduced to various statistics characterizing the messages. The forecasters' editing increased the default message length from 9 to 12 lines to 15 to 19 lines. The message ratings seemed consistent with the message type and priority. For example, flash flood warnings, which would have a high priority, had the highest levels ofoverall detail and completeness. Overall message detail was rated at a less than intermediate level (2.0 in summer 1991, 2.4 in June 1992, on a 5- point scale). The statements and advisories had the highest ratings for (current and forecast) precipitation detail, but did not address flooding factors as often or in as much detail. A large difference in message issuance frequency and type was evident in comparing the summer 1991 and June 1992 data. More messages (229) were issued during June 1992 than during the four month summer season 1991 (166); the difference was attributed to the 1992 mesoscale forecasting experiment which emphasized a "nowcasting" approach to message issuance. Procedures developed for forecast verification were derived from data sets on radar-rainfall imagery and forecast message coding. For this study the radar-rainfall data were analyzed as storm cells characterized as ellipsoids having location, extent and rainfall intensity values. Message coding also involved spatial indexing by centroid and radius of coverage onto 10 km grids. Integration over a user-specified time window provided single images oftotal rainfall and message coverage. Logical overlays ofthe two data sets yielded a spatial mapping of verification types from which the categorical scores could be computed. Basic statistics on areas of concurrence and disparity between forecasts and rainfall occurrence were computed. Accuracy ofbasin average precipitation (BAP) estimates was assessed using gages only, radar Z-R, and radar calibrated with gages. Basin boundaries were selected to maximize rain gage densities and actual watershed boundaries played no part in the delineation process. The study ofradar-rainfall estimation accuracy s for the Colorado Front Range region has shown that accuracy’s on the order of 25 % (absolute) error are obtainable with calibration ofthe radar signal. Although only a small sample size was available, the results confirm those of other more comprehensive studies. The net benefit ofradar for precipitation estimation was a direct function of distance between gages and basin center and an inverse function ofthe number of gages used for calibration. Increasing the distance between gages and basin center increases the net benefit ofradar, while increasing the number of gages used for calibration decreases its net benefit. The Flash Flood Hydrologic Forecast Model, ADVIS, is the computer replacement of manual forecast procedures using tables. This interactive forecast model computes a crest forecast, plots a hydrograph, displays response plans, and displays past flood crests and their dates. Input data consists offour variables: basin average precipitation, moisture status, river stage and reservoir releases. Forecasts were computed as the time distribution offlow (expressed as cubic feet per second - cfs) using the unit hydrograph concept and the antecedent precipitation index. Although applied to only the Cherry Creek basin, the range of predictive uncertainty obtained from ADVIS were shown to be significant, particularly at high values of precipitation and low values ofRFC Guidance. The results suggest that the numerical predictions obtained from ADVIS are variable, and thus should only be used in a categorical manner.