Abstract
Based on NWP(Nmerical Weather Predictionmodel outputs and precipitation observation data , the SVM (Support Vector Machine) statistical method is used to establish precipitation forecast models at 14 meteorological stations in central and eastern China. The optimization parameters are chosen according to the cross-validation experiments with random samples. Then the global optimization could also be gotten by the cross- validation experiments. The comparison of support vector samples are performed to explain physical significances of the predictors and their roles, providing the guidance for predictors selection of precipitation forecast. Then, the best predictors are selected. Forecast experiments were conducted for the period of June to August 2006 and the results show that the forecast models with selected predictors have higher predictive accuracy and are superior to the forecast models with all predictors included.