Detection of East Asian Atmospheric Rivers Based on Physically Constrained Dual‐Threshold Methods and Deep Learning Algorithms

Atmospheric Rivers (ARs), as key carriers of global meridional water vapor transport, are crucial for the global hydrological cycle. Particularly in East Asia, jointly influenced by the Tibetan Plateau, complex coastlines, and monsoon circulation, ARs serve as critical drivers of extreme weather events including Meiyu floods and localized severe convection. However, current AR detection methods commonly face technical bottlenecks including lack of adaptive threshold setting, significant interference from Tropical Cyclones (TCs), and insufficient high‐quality, spatiotemporally consistent labeled data, severely constraining the in‐depth research on fine‐scale AR feature identification and precipitation effect mechanisms in East Asia. To address these issues, this study proposes a Physics‐Constrained Dual‐Threshold AR detection algorithm (PCDT‐AR) that integrates water vapor intensity, geometric morphology, and dynamical background information to construct a high‐resolution East Asian AR data set spanning 1996–2023 based on ERA5 data. The algorithm effectively mitigates tropical cyclone misclassification issues and enhances event boundary coherence. Experimental results demonstrate that compared to existing data sets (e.g., ClimateNet), labels generated by PCDT‐AR achieve an average Intersection over Union improvement from approximately 40% to over 80% across various deep learning semantic segmentation models, with overall accuracy exceeding 98%. Based on this data set, this study statistically analyzes the spatial distribution characteristics and interannual and seasonal variations of East Asian ARs, quantifying their contributions to regional total precipitation and extreme precipitation events. The research findings provide a reliable data benchmark for statistical characteristic analysis of East Asian water vapor transport pathways, extreme precipitation risk assessment, and validation of AR process simulations in climate models.