WSI realtime winter precipitation forecasting using WRF

Despite increases in NWP model resolution and improvements in model physics parameterizations, the accuracy of forecasts for some parameters directly output by NWP models continues to benefit from some type of post-processing. At WSI Corporation, 51 hr real-time forecasts are generated every 03 hours for the continental US with 12 km horizontal grid spacing using the latest version of the WRF. This past winter, forecasts of precipitation type and snowfall amounts were generated using several different algorithms developed recently at WSI. Performance of the algorithms was evaluated by comparison with observations, direct model output from the WRF, and with model output from the 40 km Eta.

The two precipitation type algorithms used for the forecasts are based on schemes developed by (1) Baldwin et al. (BTC) and (2) Bourgouin. The two snow density algorithms are based on (1) work by Dube and (2) a modified form of the National Weather Service (NWS) temperature-to-snow-density relationship. The two precipitation type algorithms are combined with the two snow density algorithms to create four snowfall algorithms. A comparison with observations from METARS for precipitation type and SNODAS-derived snowfall totals for two dozen or so case studies during Dec2004-Jan2005 shows good performance by the algorithms. Specifically, the BTC-NWS snowfall algorithm yields equitable threat scores for 06-30 hr snowfall forecasts that are comparable to or better than those from the Eta Model. Despite their good performance, the individual algorithms do exhibit particular biases. For example, the Bourgouin algorithm tends to generate too much mixed precipitation and the Dube algorithm tends to generate slightly lower snowfall totals in blizzard situations, possibly because it generates too much fragmentation due to the high winds, which results in overly high snow densities.

These preliminary results are encouraging and motivating. Future refinements to the WSI-implemented versions of these winter weather algorithms should yield additional forecasting skill.