Prediction and test of optimal integrated precipitation based on similar spatial distribution of precipitation

A precipitation integration method based on spatial distribution similarity was established by using ECMWF Global Model and Warms 2.0 precipitation Forecasts. The method basically contains 3 steps, i.e., step 1, spatial scale separation of real-time precipitation forecast field, step 2, retrieval of historical similar forecast cases, and step 3, determination of the optimal integration coefficient. In the first step of scale separation, Gaussian low-pass filtering is used to decompose the precipitation field into continuous and dispersive precipitation fields. The second step is similar case test, which uses image similarity technology to find similar cases in the historical period according to the continuity and dispersion of precipitation field. In the third step, the optimal integration coefficient is determined according to the historical similar cases and applied to the latest real-time forecast. By combining the methods of image recognition, weight optimization and establishment of historical sample database, the test results for flood season (June to August) in 2018-2019 are as follows: The accuracy of the weather forecast of multi-mode integrated products is significantly improved compared with the single mode, and the performance over time is relatively stable. In terms of precipitation forecast of magnitude above rainstorm, the overall TS score of the integrated product in the flood season is higher than that of the single model. When the scores of the ECMWF model and the East China model are similar, the integrated product tends to perform better, and the rainstorm range is more biased than the ECMWF model. Due to the small and large East China model, the range of integrated products is relatively moderate. Through CRA spatial inspection and analysis, it is found that the integrated product can not only compensate for the weakness of the global model in forecasting the mesoscale precipitation process to a certain extent, but also correct the deviation of the regional model in the prediction of the location of the precipitation area, and thus to improve the rainstorm TS score.