Abstract:Based on the 2019-2021 rainfall observation and forecasts from multiple NWP models in Zhejiang Province, the performances of five operational models in forecasting torrential rains during the Meiyu period are analyzed. The 12 objective correction schemes are used to hindcast forecasts for the 2020 and 2021 Meiyu periods in Zhejiang, and are comparatively analyzed through the compressive evaluation. The results demonstrate that the skills of ECMWF, CMA-SH9 and CMA-MESO models are better than NCEP-GFS and CMA-GFS models in forecasting the torrential rains in the Meiyu period, and they have stable frequency bias relationships, therefore picked out for objective correction. Frequency matching fails to improve torrential rain quality because of the significant interannual variation in characteristics of torrential rains in the Meiyu period. The optimal score method (OTS) can improve the TS score of ECMWF rainfall forecast obviously, but its false alarm ratio is raised. The probability 〖JP2〗matching correction based on ECMWF〖JP〗 ensemble average and corrected forecast by OTS performs good skills in improving the forecast quality of the torrential rains in Meiyu period when steady torrential rain events with large rain bands are the dominant, but its correction is not effective for the convective-dominated torrential rain events. Various multi-model fusion schemes of preferred models, including multi-model averaging, adaptive integration and time-lagged ensemble forecast, can effectively improve the forecast quality of convective-dominated torrential rain events in the 2020 and 2021 Meiyu periods. Those schemes of multi-model fusion integrating OTS corrected models further improve the forecast skill for both steady and convective-dominated torrential rain events. Among them, the time-lagged ensemble algorithm integrating OTS corrected models has the best skill.