Abstract:Multitudinous precipitation predictions are verified and assessed during July an d August 2005. These products include the forecast of forecasters in Central Meteorologi cal Office and NWP models, such as T213l31, HLAFS25, MM5, MESO GRAPES 60km and 30km model，WRF20km, as well as Japanese and German global model outputs. Differ ent kinds of rainfall verification schemes and statistical verification variable s are used. The results show that the verification conclusions of different veri fication schemes on 8 kinds of NWP models and forecaster’s rainfall prediction from light rain to heavy rain are near each other, whereas there are obvious dif ferences on observation stations distribution between dense verification and sta ndard verification. TS of dense verification system is lower than that of standa rd verification system. But for torrential rain and severe rain, bias of two ver ification systems are obviously different. The reason is that most of the weathe r systems resulting in torrential rain are meso scale or micro scale weather s ystems, their frequency is similar with dense verification observation. For whole country, ETS is less than TS for the rainfall under heavy rain, and is near for the rainfall over torrential rain. However, the sort of verification r esults on ETS for kinds of models is similar with TS or SS score except for ligh t rain. Not only for dense verification, but also for standard verification, the best product of all verified forecasts is the same for all classes. If the area is the same, but the forecasts are different, the differences between dense ver ification and standard verification are different. For T213L31 model, the dense verification biases of some areas for torrential rain are obviously different fr om standard verification, but for forecaster’s prediction, the results of two v erification systems are similar.