Physical‐Dynamic‐Driven AI‐Synthetic Precipitation Nowcasting Using Task‐Segmented Generative Model

Precise and timely rainfall nowcasting plays a critical role in ensuring public safety amid disasters triggered by heavy precipitation. While deep‐learning models have exhibited superior performance over traditional nowcasting methods in recent years, their efficacy is still hampered by limited forecasting skill, insufficient training data, and escalating blurriness in forecasts. To address these challenges, we present the Synthetic‐data Task‐segmented Generative Model (STGM), an innovative physical‐dynamic‐driven heavy rainfall nowcasting model. The STGM encompasses three key components: the Long Video Generation (LVG) model generating synthetic radar data from observed radar images and data provided by the Weather Research and Forecasting (WRF) model, MaskPredNet predicting the spatial coverage of various rainfall intensities, and SPADE determining rainfall intensity based on the coverage provided by MaskPredNet. The STGM has demonstrated promising skill for precipitation forecasts for up to six hours, and significantly reduce the blurriness of predicted images, thus showcasing advances in rainfall nowcasting. Plain Language Summary Deep‐learning methods have proven superior to traditional techniques in rainfall nowcasting, but current models still face limitations such as low accuracy, insufficient training data, and a brief effective forecast period. To address these challenges, we've developed the STGM, a novel deep‐learning‐based architecture for predicting heavy rainfall. The STGM shows excellent performance by using a new design that breaks tasks into smaller parts and relies on AI‐generated data. Our results indicate that it excels in predicting heavy rainfall for up to 6 hr, accurately tracking the spatial and temporal evolution of storm cells over extended periods. Moreover, it significantly improves the clarity of the produced images compared to commonly used baseline models. Key Points Our novel STGM model outperforms common baseline models, showcasing high proficiency in predicting precipitation up to 6 hr ahead By integrating physical data from numerical weather prediction models, the performance of the STGM model is significantly enhanced The use of synthetic data provides a substantial boost to the predictive capabilities of the STGM model