Improving Cool-Season Snowfall Forecasts from Operational Modeling Systems over the Western Continental United States

Winter precipitation poses a major challenge for numerical weather prediction and operational weather forecasting and frequently leads to air- and surface-transportation disruptions, vehicle accidents, avalanches, and other hazards. In the western continental United States (CONUS), complex terrain strongly modulates precipitation amount and type, necessitating kilometer- or sub-kilometer-scale prediction for applications from winter-road maintenance to avalanche mitigation activities.

Over the past few years, the University of Utah has produced high-resolution gridded snowfall forecasts over the western CONUS using operational modeling systems run by national and intergovernmental modeling groups, including global ensembles and the US Global Forecast System (GFS). Quantitative precipitation forecasts (QPF) from these modeling systems are downscaled using climatological precipitation analyses to 800-m grid spacing. High-resolution snowfall forecasts above the estimated snow level are produced by machine learning techniques based on training to high quality manual snow-to-liquid ratio observations at multiple high mountain sites. Application to the ensemble members enables the derivation of probabilistic quantitative snowfall forecasts.

The approach is straightforward and does not require a long record of high-resolution precipitation analyses or forecasts. The latter is often difficult to obtain from operational forecast systems. For shorter forecast lead times, these techniques have also been incorporated into the US High Resolution Rapid Refresh and next-generation Rapid-Refresh Forecast System. Precipitation type algorithms and additional observations are being employed to increase the skill of these snowfall forecasts for a full-CONUS gridded product. Verification relative to observations indicates that these forecasts improve upon model output and existing SLR techniques.