Abstract:The total performance of short-range torrential rain forecasts during the 2020 Jiangsu main flood season (June-September) is verified based on the three global models (ECMWF, NCEP-GFS, CMA-GFS), three regional models (CMA-MESO, CMA-SH9, PWAFS), local objective forecast commonly used in Jiangsu operational forecast service and subjective forecast by forecasters. Verification and case analyses of stable and convective torrential rain forecast are also demonstrated respectively according to the types of precipitation. Results show that from the overall verification, the subjective torrential rain forecasts of forecasters only have positive TS skill compared to ECMWF within the 24 h forecast lead time. The torrential rain forecasts by local objective method and forecasters have TS>regional models>global models except ECMWF. It is not true that the higher model resolution, the better forecast performance for either global models or regional models. Most regional models have obviously larger torrential rain foreacst area and high false alarm ratio, while most global models except ECMWF have obviously smaller torrential rain forecast area and high miss rate. The CMA-MESO model initialized at 20:00 BT has obviously better forecast performance than that initialized at 08:00 BT. For stable torrential rain forecast, forecasters should take more references of the ECMWF forecasts, the local objective forecasts within 48 h lead time and the CMA-MESO forecasts initialized at 20:00 BT. As for convective torrential rain forecast, forecasters should refer to the region models CMA-SH9, PWAFS, and the 24 h local objective forecasts more.