Abstract:From 21 May to 21 June 2022, the heaviest dragon-boat precipitation process in the last decade occurred in South China. This extreme precipitation process, featured with strong extremity, large accumulated rainfall and frequent occurrence of severe rainfall, caused significant economic losses. In this paper, the forecast products from two operational models, TRAMS and ECMWF, which are commonly used in South China, are selected to divide the torrential rain processes during dragon-boat precipitation into front-zone torrential rain and warm-sector torrential rain. The results are verified and evaluated, so as to understand the characteristics of the two models’ biases for the front-zone torrential rain and warm-sector torrential rain under the background of the extreme dragon-boat precipitation. Compared with the traditional point-to-point method, the MODE method can effectively avoid the phenomenon of high false alarm ratio caused by precipitation position deviation in the model. Further analysis of the number, position, precipitation area and intensity of torrential rain objects based on MODE method shows that the high-resolution model TRAMS has better ability to identify and match torrential rain objects than the global model ECMWF. The location of warm-sector torrential rain predicted by TRAMS is mostly biased to the east, while the front-zone torrential rain predicted by ECMWF is basically biased to the north. The deviations in precipitation position in the above two are closely related to the forecast errors of southerly airflow at low altitude by different models. The area prediction of the front-zone torrential rain by TRAMS is close to the observation, but the forecast area of warm-sector torrential rain is larger. The forecast areas by ECMWF for both front-zone torrential rain and warm-sector torrential rain are smaller. The prediction of torrential rain intensity and extreme value by TRAMS is closer to the observation than that by ECMWF, but it still underestimates the extreme precipitation. This study can provide new experience for forecasters to understand the prediction biases of different operational models for dragon-boat precipitation process. It also has some reference values for model developers to further carry out research on error source diagnosis and technical improvement of TRAMS model.