GEFS Precipitation Forecasts and the Implications of Statistical Downscaling over the Western United States

Operational medium-range ensemble modeling systems produce quantitative precipitation forecasts (QPFs) that provide guidance for weather forecasters, yet these systems lack sufficient resolution to adequately resolve orographic influences on mountain precipitation. In this study, we verify cool-season (Oct-Mar) Global Ensemble Forecast System (GEFS) QPFs at upper-elevation sites in the western United States using Snow Telemetry (SNOTEL) observations. Results indicate that events of all sizes are routinely underforecasted, with all utilized performance metrics, such as equitable threat score (ETS) and Brier skill score, worsening toward the interior. The underdispersive ensemble spread captures only ~30% of upper-quartile events at Day 5, exhibits poor reliability, and is less skillful over the interior compared to forecasting with climatological probabilities.

In an effort to improve QPFs without exacerbating computing demands, we explore statistical downscaling based on high-resolution climatological precipitation analyses from the Parameter-elevation Relationships on Independent Slopes Model (PRISM), an approach frequently used by operational forecasters. Such downscaling improves model biases, ETSs, and hit rates. However, 50% of downscaled QPFs for upper-quartile events are false alarms at Day 1, and probabilistic QPFs still do not capture ~40% of such events at Day 5. These results should help forecasters and hydrologists understand the capabilities and limitations of GEFS forecasts and statistical downscaling over the western U.S. and other regions of complex terrain.