【摘要】 四川盆地地形地貌复杂，强降水预报难度大，对模式降水预报产品进行订正，是提升强降水预报质量的重要手段。本文选取2018—2019年发生在四川盆地的35次强降水过程，对ECMWF、CMA＿MESO和SWC＿WARMS三种模式的24 h强降水预报采用常规评分和空间平移两个方法进行检验，并利用最优评分、多模式集成和位移订正三种方法进行订正试验。结果表明:最优评分订正方法可以有效改善各模式降水预报的强度，而多模式集成订正法可以改进降水落区和极值预报，在此基础上计算位移偏差，根据最优的位移偏差值对降水预报进行位移订正，可以进一步改进强降水落区预报效果。然后利用2020—2021年强降水过程进行订正效果验证，结果显示：经订正后的降水极值预报更接近实况，各量级降水预报评分明显优于单一模式，暴雨和大暴雨预报的TS评分提高率较最优单模式分别可达24.3%和42.8%，订正后空报率基本维持，漏报率显著下降，订正效果良好。更多还原

【Abstract】 The terrain and landform of the Sichuan Basin are complex, which also make the forecast of heavy rainfall challenging,therefore the correction of the model precipitation forecast products is the crucial method to improve the quality of heavy rainfall forecasting. In this paper, a total of 35 heavy rainfall processes that occurred in the Sichuan Basin from 2018 to 2019 were examined to verify the 24-hour heavy rainfall prediction of ECMWF, CMA＿MESO, and SWC＿WARMS by using conventional score and spatial translation methods. Correction experiments are then conducted on these models using three methods, including optimal score, multi-model integration, and displacement correction. The results show that the optimal score correction method can effectively improve the intensity of precipitation prediction, while the multi-model integration correction method performs better in both the precipitation fall area and extremum prediction. On this basis, the displacement deviation is calculated, which is used to correct the displacement of precipitation prediction according to the optimal value, and thus further improve the forecast of heavy rainfall area. Finally, the performance of the correction is evaluated by using the heavy rainfall processes from 2020 to 2021. The results show that the corrected precipitation extremum prediction is closer to the actual observations, with the precipitation prediction score of each magnitude being obviously better than that of the single model. Compared to the optimal single model, the TS score improvement rate of the rainstorm and heavy rainstorm prediction can reach 24.3 % and 42.8 %, respectively. After correction, the false alarm rate(FAR) basically remains unchanged, while the missing rate(MR) significantly decreases, suggesting good correction effects.更多还原