Abstract
The Model Output Statistics (MOS) technique has been applied to the prediction of quantitative precipitation. Data at 233 stations for two winter seasons are pooled to develop generalized operator equations for prediction of the probability of precipitation amount (PoPA) in 5 categories for a 12–24 hr forecast projection. Predictors subjected to screening regression are obtained from the National Meteorological Center's (NMC) Primitive Equation Model and the Techniques Development Laboratory's Trajectory Model. The equations are developed by means of two approaches. The first, referred to as unconditional, uses both precipitation and no precipitation cases in the developmental sample. The second, referred to as conditional, uses only precipitation cases.
To test the system on independent data, forecasts at 58 cities in the United States are verified for October 1972. Probability forecasts are transformed to categorical forecasts by (1) maximizing the percent correct, (2) maximizing a quantitative precipitation forecast (QPF) score used to verify categorical forecasts at NMC, (3) maximizing a utility score, and (4) minimizing the categorical bias. The categorical forecasts are then compared to those obtained from (1) the Limited Area Fine Mesh Model, (2) subjective preparation at NMC, and (3) climatology. Verification scores include the percent correct, a QPF score used at NMC, and a utility score. In addition, the categorical bias is computed for each forecast method. The results indicate that the PoPA categorical forecasts obtained by minimizing the bias are, in general, slightly better than all the other forecasts. Also, “unconditional” forecasts are about as good as “conditional” forecasts.
