Abstract: In order to improve the precipitation prediction accuracy of numerical model during flood season in Shandong province and provide scientific references for selection and adjustment of physical parameter schemes of models, the mean absolute error, mean error, TS score of 24 hours and 48 hours precipitation simulated by the MM5, WRF-RUC and T639 models under sunny/rainy weather, general precipitation, grading precipitation conditions were analyzed during flood season (from June to September) of 2010. The results show that the accuracies of precipitation prediciton are similar for three models. TS scores of general precipitation and light rain are the lowest for the MM5 model and the highest for the T639 model. TS score of 24 hours precipitation prediction of middle rain and heavy rain etc. is the highest and that of heavy rain reaches 10.37% for the T639 model, while that of 48 hours precipitation forecast decreases significantly. Weather 24 hours or not 48 hours precipitation forecast, the mean absolute error is the lowest in September for the WRT-RUC model and from June to August for the T639 model, while the monthly mean absolute error is negative for the WRT-RUC model. Diffident precipitation verification schemes and weather processes of diffident types have certain impacts on the verifications results.

Key words: Multiple models, Precipitation forecasting, Verification