Statistical Characteristics of a Real-Time Precipitation Forecasting Model

Abstract

At Colorado State University the Regional Atmospheric Modeling System (RAMS) has been used to produce real-time forecasts of precipitation for the Colorado mountain region since 1991. Originally a so-called dump-bucket scheme was used to generate precipitation, but starting in the fall of 1995 real-time forecasts used the bulk microphysics scheme available with RAMS. For the month of April 1995, a series of 24-h accumulated precipitation forecasts for the month were generated with both the dump-bucket and microphysics versions of the forecast model. Both sets of output were compared to a set of 167 community-based station reports and another set of 32 snow telemetry (SNOTEL) automatic pillow-sensor stations. The addition of microphysics improved the forecasting of the areal extent and maximum amount of precipitation, especially when compared to the SNOTEL observational set, which is found at locations more representative of the model topography. Climatological station precipitation forecasts were improved on the average by correcting for the difference between a station’s actual elevation and the cell-averaged topography used by the model. The model had more problems with the precise timing and geographical location of the precipitation features, probably due in part to the influence of other model physics, the failure of the model to resolve adequately wintertime convection events, and inadequate initializations.