A multi-model blending precipitation prediction method based on the dynamic weight of multi-level TS score

Quantitative precipitation grid forecast is the focus and challenge in forecasting services. In this study, a muti-mode blending method was conducted which can fully consider the prediction performance of different orders. Based on the global scale model forecast ECMWF, the mesoscale model forecast CMA-MESO, and the CMA-SH9 forecast 24 h cumulative precipitation forecast. The optimal TS score algorithm was used to revise the numerical model, and then the multi-level dynamic weight was used to fuse the revised forecast products. the probability forecast was used to eliminate the weak precipitation. The blending forecast was verified and case applicated for Heilongjiang Province from November 2022 to October 2023. The results are as follows. (1) The forecast corrected by OTS can effectively improve the small precipitation false alarms, reduced quantitative errors, provide higher quality members. (2) Compared with ECMWF, CMA-MESO and CMA-SH9, the accuracy rate of the blending forecast was improved by 12.71%, 9.36% and 6.4%. And the TS score of the rainstorm was improved by 27.14%, 45.6% and 79.27%. (3) Two typical case analyses found that the blending forecast can improve the deviation of the model in the heavy rainfall area, effectively adjust the location and magnitude of rain areas, effectively improve the problem of precipitation false alarms.