A multi-source data-driven framework for probabilistic flood risk assessment using cascade machine learning models: case study in the Sichuan Basin.

Along with global climate change, more frequent extreme climate phenomena have led to an increasing number of and increasingly severe flood disasters. Extreme rainfall events are capable of generating substantial amounts of surface runoff. Concurrently, the progression of urbanization has given rise to the expansion of impervious surfaces, thereby augmenting the likelihood of flood disasters. As China's most flood-vulnerable region, the Sichuan Basin has sustained recurrent catastrophic flooding throughout history. This study establishes three policy-relevant hotspots across the basin as critical testbeds to quantify climate-driven changes in flood recurrence intervals. Utilizing CMIP6 projections under different SSPs scenarios, we developed a physics-informed process-based modeling framework integrating statistical downscaling, adjustments for extreme values and flood frequency analysis, specifically addressing how anthropogenically modified hydrological regimes amplify extreme event probabilities in this monsoon-dominated basin. The findings suggest that by the conclusion of the current century, the study area is likely to experience a notable increase in temperature, with an anticipated rise of approximately 1.7 ℃, and an intensification in precipitation, with an increase of 9.4% percent. Furthermore, the likelihood of extreme flood disaster events is projected to double, underscoring the imperative for robust climate adaptation and disaster mitigation strategies.