Abstract

Floods remain among the most destructive natural hazards, and their frequency, intensity, and spatial distribution are being profoundly reshaped under the influence of climate change. Despite their global significance, comprehensive physical assessments of changing flood susceptibility at a planetary scale remain limited, with most studies constrained to regional or basin-level analyses. In this study, we present the first global comparative evaluation of flood vulnerability using multiple assessment methodologies designed to capture both the spatial complexity and the physical drivers of flood risk. By analyzing 5,105 historical flood events in conjunction with nine critical biophysical parameters—encompassing topography, terrain indices, precipitation regimes, vegetation cover, and soil characteristics—we reveal that approximately one-quarter of the Earth’s land surface currently exhibits high to very high flood susceptibility. Spatial heterogeneity is evident, with susceptibility concentrated in large river basins and densely populated alluvial plains.

Our multi-model comparison highlights the superior performance of machine learning approaches, particularly the Random Forest model (AUC = 0.85), which effectively captures nonlinear interactions between precipitation variability, terrain morphology, and land surface features. Key flood-prone regions identified include the Yangtze, Indus, Ganges-Brahmaputra, and Amazon basins, where compound drivers converge to intensify vulnerability. This research provides a high-precision, physically grounded assessment of global flood susceptibility, demonstrating how shifting precipitation regimes interact with terrestrial landscapes to create emergent patterns of flood risk. The findings offer valuable spatial intelligence for climate adaptation planning, water resource management, and disaster preparedness, especially in regions where traditional hydrological monitoring and infrastructure remain insufficient.