Large-scale circulation shapes the spatial organization of anomalous precipitation over Northern Jiangsu

Anomalous precipitation events often exhibit coherent spatial organization rather than isolated intensity extremes, reshaping regional flood risk in subtropical monsoon zones. Using 1-km daily precipitation data from 1961 to 2019, we developed an unsupervised framework combining a convolutional autoencoder and a self-organizing map to identify event-scale anomalous precipitation patterns over Northern Jiangsu. A total of 543 anomalous days were classified into five distinct spatial types characterized by pronounced north-south contrasts and east-west differentiation. Interaction analysis reveals suppressive and co-occurring relationships among patterns, indicating structured atmospheric control. Phase variations of the Western North Pacific Subtropical High and the East Asian jet significantly modulate the probability of different patterns. Incorporating circulation-conditioned pattern probabilities into runoff forecasting improves peak discharge simulation. These findings highlight the importance of circulation-structured precipitation in linking atmospheric variability with hydrological response under nonstationary climate conditions.