In view of the characteristics of local weather prediction, the available numerical prediction products as well as the real-time radiosonde data, a practical scheme for predicting the summer precipitation is presented by introducing mesoscale data to the mesoscale numerical prediction model MM4 in order to increase the accuracy of prediction to the unexpected convective precipitation. The diagnostic field of the daily physical factors were calculated by using MM4 model. These calculation results are added into the numerical prediction production of the ECMWF and obtained 41 prediction fields, then the precipitation prediction equation is developed by using the canonical correlation analysis,successive regression as well as discriminatory analysis methods.