Based on the surface observation and ERA5 hourly reanalysis data from 2000 to 2014, 43 factors were selected to characterize the temperature, humidity and microphysical characteristics of four types of precipitation phases (rain, snow, sleet, freezing rain). The objective forecast method of precipitation type in the middle reaches of Yangtze River is studied by using CatBoost algorithm and the fine terrain height preprocessing. The results show that this method has good classification and forecast effect for rain, snow and freezing rain. The accuracy and spatial fineness of precipitation type discrimination can be improved by using the fine terrain height preprocessing. Compared with the ECMWF prediction, the prediction accuracy of rain, snow and freezing rain has increased by 9.9%, 39.1% and 11.1% respectively, but the improvement of sleet was not obvious.