This study evaluates the performance of the ECMWF global ensemble prediction system, the short-term-forecast mesoscale model, and the rapid-updated nowcasting model for South China based on GRAPES, by analyzing the spatio-temporal characteristics of precipitation forecast under the local climate background of Guangdong Province. Based on the evaluation results, a hybrid method for multi-model post-processing is developed to produce high-resolution gridded quantitative precipitation forecast (GQPF). The results are as follows: Multi-scale model integration is a promising technique for objective model post-processing. The frequency-matching and optimal-percentile methods show the advantages in ensemble forecast interpretation at synoptic scale, while the spatial correction and clear-rainy elimination based on localized stratified verification can help further optimize the space distribution and intensity for specific weather scenarios. Temporal downscaling based on the characteristics of the mesoscale model in diurnal variation is beneficial to improve the hourly precipitation forecasting. Considering the interdependence and complementary advantages of different methods, the GQPF method for Guangdong is established with a sequential flow of “frequency-matching, optimal-percentile, spatial correction, clear-rainy elimination, time-downscaling”, which improves the accuracy of precipitation objective forecast.