Ocean Moisture Source Evaporation Anomalies and Large-Scale Teleconnection Patterns: Attribution and Hazard Risk Assessment of Extreme Precipitation Events

Abstract

This study comprehensively investigates moisture source evaporation anomalies, large-scale teleconnection patterns, driver attribution, and hazard risk associated with the severe rainfall event of 2023 over the Hai River basin in North China. Utilizing high-resolution ERA5 reanalysis data, we performed Empirical Orthogonal Function (EOF) decomposition on a 12-station observational dataset. The analysis identified a dominant spatial mode explaining 37.4% of the total variance, characterized by a pronounced principal component (PC1) shift of +4.28 standard deviations at the event onset. This significant shift coincided with distinct warm atmospheric anomalies over the Ural Mountains, Lake Baikal, and the Western Pacific Subtropical High. To quantify the underlying mechanisms, an attention-based Long Short-Term Memory (LSTM) neural network model was developed for driver attribution. The model results indicated that large-scale teleconnection forcing served as the primary driver, accounting for 30.9% of the variance, followed by southern maritime evaporation at 23.7% and local thermodynamic dynamics at 23.3%. Furthermore, a composite Hazard Risk Index was constructed to evaluate regional vulnerability. The assessment revealed that Beijing and Baoding experienced critical hazard levels, with index values reaching 0.890 and 0.822, respectively. These findings provide crucial insights into the complex meteorological mechanisms driving intense precipitation in the region, offering a robust scientific foundation for future disaster mitigation, early warning systems, and climate adaptation strategies in highly urbanized areas.