The Use of Perfect Prog Forecasts to Improve Model Output Statistics Forecasts of Precipitation Probability

Abstract

A method of improving the accuracy of model output statistics (MOS) probability of precipitation (POP) forecasts was investigated. The method uses a perfect prog (PP) forecast as a potential predictor in a MOS equation. The PP method, with its larger developmental databases has the potential of incorporating additional information about local climatology, seasonality, and synoptic pattern type, which might be otherwise lacking in the MOS predictor dataset. Three PP models were developed: an analog model, a 1ogistic regression model and an analog/regression hybrid model. The POP forecasts were generated by the three PP models and the MOS model at four Pennsylvania stations by using 6 months of independent limited-area fine mesh (LFM) forecasts. Three MOS/PP combination models were derived by linearly combining MOS with each of the three PP models. The MOS/PP combination model forecasts were generated with the independent MOS and PP forecasts by using a cross-validation technique. The three MOS/PP combination models showed a small improvement over the MOS model. The probability that thew improvements were from random chance ranged from 6% to 33%.