AVN-Guided MOS Weather Forecasts for Europe Based on Regression Screening and Cross Validation

A new Model Output System (MOS) on the basis of the AVN model output is going to be constructed for the European region, primarily for predicting maximum/minimum temperature and total cloud cover. It is generally recognized that the preparation, the preselection, and the maximum number of the predictors used in a prediction equation plays a prominent role. We enhanced the normally used MOS equations by combining the regression screening procedure with two approved expert systems and the cross validation method. Either an empirically build or a physically motivated expert system for predicting the cloud cover were optimally fitted to several stations. The spatial variation of the adapted parameters is small. Furthermore these calculations serve as additional predictors for maximum/minimum temperature foecasts. The cross validation method has the advantage that the prediction error can be derived on independent data while the whole data sample was used to estimate regression coefficients. So we can determine the optimum number of predictors in the regression equation in an objective manner.