While NWP guidance on winter storms has significantly improved in recent years, forecasting the type of precipitation expected to fall remains a challenge. Even when model forecasts of vertical temperature and moisture profiles are skillful, a successful precipitation type forecast is inevitably dependent upon a skillful algorithm to diagnose whether rain, snow, sleet, or freezing rain is most likely.

A new mini-ensemble of precipitation type forecasts is run in real-time for the operational Eta model. These are not multiple model runs, but multiple solutions from a single run using different precipitation type algorithms. The starting point is the NCEP algorithm, a decision tree approach based on Baldwin and Contorno, 1993. This scheme contains an intentional bias towards prediction of sleet and freezing rain; this scheme generally predicts sleet or freezing rain in soundings marked by deep, isothermal layers with temperatures just below freezing. A second version of this scheme is run to eliminate that bias by using a different area check in the sounding. A third prediction comes from the Ramer precipitation type algorithm. Finally, the explicit percentage of frozen precipitation from the model's microphysics offers the fourth opinion.

An overview of the different schemes and results from selected cases in different regions of the country in the winter of 2002-2003 will be presented.