Explainable Artificial Intelligence for Extreme Precipitation Simulation

Recent advances in artificial intelligence (AI) models have enabled effective simulation of extreme precipitation (EP), yet most models remain black-boxes with unclear contributions from individual meteorological variables. Here, we develop an Explainable Extreme Precipitation Simulation framework (XEP-Sim) to identify and quantify the key drivers of EP in the Asian monsoon region.

Six meteorological variables are found to be essential for accurately reproducing EP, with integrated water vapor (IWV) playing a dominant role. EP intensity increases significantly under extreme conditions of these variables, reaching up to 2.8 times higher when all six are extreme compared to none. In addition, interactions between extreme IWV and other variables further amplify EP intensity through dynamic pathways. These findings suggest the dominant role of IWV and its interactions with other key variables, offering physically interpretable insights that can improve EP prediction and inform variable selection in data-driven climate models.