Deep learning-constrained projection of global fluvial floods and their socioeconomic implications under global warming

Abstract Fluvial floods characterized by either large volume or prolonged duration severely impede the sustainable development of socio-ecological systems, which may intensify with global warming. However, the co-evolution dynamics of joint fluvial flood characteristics and their socioeconomic implications under different global warming targets are poorly understood. Here, we combine a deep learning (DL)-constrained hybrid model, a statistical approach and climate models to evaluate future flood occurrence within a bivariate framework (i.e. volume and duration) across 8735 catchments. After evaluating the performance of the DL-constrained hybrid model, we project a doubling of bivariate fluvial flood hazard under 3.0 °C warming. We assess the uncertainty of fluvial flood projections and reveal that the global climate models and coupled factors are the major uncertainty sources. Global warming from 1.5 °C to 3.0 °C tends to amplify flood exposure of gross domestic product, population and agricultural area, by ∼97%, ∼46%, and ∼67% under a medium emission scenario (SSP370), respectively.