Abstract:Precipitation forecast of three typhoon rainfall processes affecting Henan Area in August 2018 from four numerical models, SHANGHAI_HR(SH), GRAPES_MESO(MESO), ECMWF_HR(EC), GRAPES_GFS(GFS) were evaluated using FSS (fractional skill score) and CRA (contiguous rain area) methods based on CMA radar-satellite-gauge merged precipitation (CMPA_Hourly V2.1) in this paper. The difference of two methods and the performance of each numerical model were discussed. The results show that FSS method can better distinguish the performance of different models through quantitive scores compared with the traditional TS method, and CRA method can reflect error sources of models more comprehensively. For local heavy rainfall or intense center of large-scale precipitation, regional models are more superior to global models. However, global models still perform well in 〖JP2〗predicting small-scale precipi-〖JP〗tation. For the two precipitation processes of typhoons ‘Yagi’ and ‘Rumbia’, the displacement errors of EC are more westward than the observation, and the same characteristics are also found in the prediction of ‘Rumbia’ precipitation by MESO and GFS. Precipitation scope and intensity tend to be underestimated by GFS model. The EC model can perform better a little, but still has some shortcomings in estimating precipitation extremes. Although regional models, especially SH, can forecast more intense precipitation centers, the scope and intensity can be easily overestimated. The displacement error for most models is main source of precipitation error, and intensity error and pattern error are roughly equivalent.