Abstract: In order to measure the forecasting capability of different numerical models for heavy rainfall processes in the southern Sichuan Basin, SAL and refined forecast verification were compared in 23 precipitation processes for a 2-year (2020—2021) period based on the CMA Multi-source Precipitation Analysis System (CMPAS) data. The results show that: (1) the SCMOC, ECthin and CMA-MESO models all overestimated the precipitation area (S>0), with ECthin and CMA-MESO showing the best performance in forecasting the precipitation structure starting from 08:00. ECthin underestimated the precipitation intensity (A<0), while SCMOC and CMA-MESO overestimated it (A>0), with ECthin having the smallest deviation in precipitation intensity forecasting starting from 08:00. (2) For precipitation forecasts above the large rainfall threshold, the Threat Score (TS) for each model were unimodal, with the highest scores occurring between 23:00 and 02:00, and the lowest scores occurring between 14:00 and 20:00. (3) All models performed well in forecasting the precipitation frequency in Luzhou, but their predictions for the central regions of precipitation frequency in Zigong and Yibin are significantly southward. (4) ECthin and SCMOC had biases within 3 hours (1 time interval) for the most likely start and end times of precipitation, with SCMOC at 08:00 showing the best performance in forecasting the start and end times of precipitation in the southern Sichuan Basin, but tends to forecast the end of precipitation too early in the more southern regions. The research provides an in-depth analysis of the forecasting characteristics of different models for heavy rainfall processes in the southern Sichuan Basin, offering guidance for improving the refined forecasting capability of intense precipitation.