Use of Model Output Statistics for Predicting Ceiling Height

Abstract

The model output statistics (MOS) technique consists of determining a statistical relationship between the forecast output of numerical prediction models and a predictand. This paper presents some results obtained in applying the MOS technique to the prediction of ceiling height by means of screening regression. Data from 3 winter seasons and 95 eastern U.S. stations are combined in a generalized operator approach to develop multiple regression equations. The potential predictors subjected to screening include surface variables observed at 0700 GMT and forecast output from both the National Meteorological Center's primitive-equation model and the Techniques Development Laboratory's subsynoptic advection model. Prediction equations are developed for 5-, 11-, and 17-hr forecast projections representing ceiling height forecasts valid at 1200, 1800, and 2400 GMT, respectively. Ceiling height is treated both as a categorized and as a continuous predictand. Where ceiling height is categorized, the regression estimation of event probabilities (REEP) screening technique is used to develop probability forecast equations. Where ceiling height is treated as a continuous variable, specific ceiling height forecast equations are developed by ordinary screening regression. The independent sample used for testing consists of data for 20 stations in the eastern United States from the winter of 1970–71. Several verification scores, including the Brier P-score, Allen utility score, Heidke skill score, and percent correct, are presented. The verification results indicate that forecasts from the REEP equations are generally better than those from the equations that produce specific heights. Also, the REEP forecasts are better than persistence and climatology.