WRF Model Precipitation Type Algorithms and Verification

Winter weather events in North Carolina often consist of all precipitation types, including snow, ice pellets, freezing rain and rain. Visualizing where precipitation types will fall, as well as where a transition between two precipitation types will be located is useful to forecasters. Additionally, these events are generally not modeled well due to complex conditions during these storms. Therefore, knowing any biases of a model and understanding at what forecast hour the forecast becomes unacceptable is essential to meteorologists when forecasting these events. Gridded output from the Weather Research and Forecasting (WRF) model was used to evaluate different precipitation type algorithms. Two methods were employed: one using the temperature of warm nose and cold surface layer, and the other using the precipitation types from partial thicknesses on the precipitation type nomogram. The output was evaluated with surface point observations for each forecast hour for two different winter weather events. This study concluded that the warm nose temperature method verified better than the partial thicknesses, most likely due to some vagueness associated with precipitation types and partial thicknesses. Also, despite the WRF having a few degrees Celsius warm bias during the storm, the gridded WRF output had a high correlation with the observations (>.90) that gradually decreased as the forecast hour increased, dropping below .9 by forecast hour ~20. While the forecast precipitation type and variables were proven to correlate well with observations early on, it is important to keep in mind that winter storms are dynamic, hence using several precipitation type classification methods and model parameters are useful in creating a forecast.