Contrasting prediction skill for different precipitation patterns in Meiyu over eastern China using ECMWF subseasonal forecasts

The Meiyu season, characterized by the northward progression of the East Asian summer monsoon in the Yangtze River Valley, is a critical period of precipitation. This study assesses the prediction skills for various precipitation patterns during the Meiyu using S2S hindcast data from the European Centre for Medium-Range Weather Forecasts (ECMWF) from 2002 to 2021. Three primary spatial distribution patterns of precipitation have been identified: the Jiangnan, Yangtze River, and Huaihe River types. Significant difference exists in the model’s simulation capabilities across these patterns, with the Yangtze River Rainy type showing the highest accuracy. The forecasting skill differs among the precipitation types, primarily due to the model’s representation of key impact circulation systems and their impacts on precipitation. During Meiyu, the model effectively predicts low-latitude circulations but struggles with higher-latitude ones. Consequently, these differences in circulation simulation crucially affect the simulation outcomes for the various precipitation patterns. Meanwhile, the model’s performance largely depends on its ability to accurately represent the relationship between key circulations and corresponding precipitation patterns. The model captures the relationship between circulation and precipitation well for the Yangtze River Rainy type. However, it inaccurately simulates this relationship for the Jiangnan Rainy type, leading to significant errors. Additionally, different sea surface temperature (SST) states have varying external forcing impacts. Elevated SST in the central Pacific during the preceding spring favors the formation of a circulation pattern conducive to heavy rainfall in the Yangtze River Basin. Due to the good simulation relationship between key circulation configurations and precipitation in the model, the external forcing relationship of the corresponding SST distribution in the model is well reflected.