Evolving Land–Atmosphere Preconditioning of Coastal Storm Surges: A Multi-Basin Analysis of Shifting Drivers

Coastal flood risk assessments traditionally treat storm surges as instantaneous responses to wind and pressure, assuming stationary physical drivers. However, under a warming climate, the thermodynamic processes preconditioning coastal catchments for extremes are evolving. This study quantifies how the influence of antecedent environmental precursors in modulating coastal surges has shifted over the last four decades. A physics-aware Convolutional LSTM framework analyzes 40 years (1984–2024) of ERA5 reanalysis data across six diverse global hotspots, including the U.S. Gulf Coast, Bay of Bengal, and East Asia. The model integrates lagged anomalies of soil moisture and integrated water vapor transport (IVT) to capture multi-day preconditioning. Explainable AI diagnostics—specifically temporal sensitivity analysis—are employed to assess changes in the relative importance of drivers between early (1984–2000) and late (2004–2024) epochs. Results indicate that while instantaneous wind stress remains the dominant control on surge peaks, the predictive weight of antecedent conditions has shifted. In the U.S. Gulf Coast and East Asia, the influence of 7–14-day soil moisture and IVT anomalies increased significantly in the recent period, particularly regarding flood duration and compound surge–precipitation likelihood. These findings reveal a strengthening coupling between terrestrial hydrologic memory and coastal extremes. Consequently, the results challenge "snapshot-based" hydrodynamic approaches, suggesting that effective early-warning horizons in a non-stationary climate must extend from days to weeks to account for these evolving precursor regimes.