Abstract
A Model Output Statistics system for forecasting the conditional probability of precipitation type (PoPT) became operational within the National Weather Service in September 1978. Forecasts are provided for three precipitation type categories: snow or ice pellets, freezing rain, and rain. To develop the forecast equations, data are combined from different stations because of the limited amount of developmental data. To justify combining the data, the Limited-area Fine Mesh (LFM) model predictors are transformed from their original values through the use of the logit model. In one experiment, it is shown that probability of snow forecasts are made more accurate through an improved use of the logit model for predictor transformation.
The new transformation procedure is then used in the development of a set of experimental PoPT forecast equations. The experimental equations differ from the operational equations in other ways also. The developmental sample for the experimental equations included approximately three winter seasons more data than the sample used for the operational system. Also, improvements are made to the potential predictors used to develop the experimental equations. Finally, freezing rain mixed with any other precipitation type is defined as freezing rain in the experimental system; in the operational system, this mixture of precipitation is defined as rain.
A comparative verification between the experimental and operational systems on independent data indicates that, overall, the experimental PoPT forecasts are better than the operational forecasts, especially for 12–24 h freezing rain forecasts. Based on these results, new operational PoPT forecast equations are developed incorporating the features associated with the experimental equations. The new system was implemented in the fall of 1982.
