Climatic and Geomorphic Controls on the Threshold Behavior of Upper Tail Floods

North China experienced some of the world’s most extreme floods, with flood peaks exhibiting pronounced heavy-tailed behavior that rivals global flood envelope curves. Unraveling the physical origins of these outliers is essential for reliable flood risk estimation and hydrologic design. Here, we analyze long-term rainfall and streamflow records (1950–2023) from 297 watersheds across North China to attribute the upper tail of flood peaks to storm structure and watershed behaviors. Although the region’s short-duration rainfall intensities are on the upper end of the rainfall spectrum, we find that the most extreme floods are not primarily driven by these high-intensity events. Instead, they are linked to long-duration events with exceptionally large accumulated totals, indicating that accumulated rainfall, rather than intensity alone, is the dominant meteorological control on catastrophic peak discharge. We further identify a threshold-type runoff response: once storm-total rainfall exceeds specific storage capacities, peak discharge increases disproportionately with additional rainfall, sharply widening the magnitude gap between extreme and ordinary floods. Our analyses suggest that this effective storage is governed largely by deep weathered-rock and aquifer systems rather than shallow soil layers. The susceptibility to this threshold behavior depends on the watershed's spatial organization. Integral geomorphic metrics reveal that watersheds in mountain-plain transition zones are structurally predisposed to rapid routing and nonlinear peak amplification once storage is surpassed. These watersheds are characterized by fan-shaped drainage networks and heterogeneous river longitudinal profiles. These findings highlight a latent catastrophic potential in watersheds where historical storms have rarely crossed these controlling thresholds. As climate change drives more persistent extreme rainfall, threshold exceedance may become more frequent, implying a heightened risk of unprecedented floods beyond historical experience.