Lingjie Li

Study region: An independent drainage system in Hangzhou, China, a typical coastal and estuarine city.Study focus: Compound floods driven by concurrent rainfall and boundary water levels pose severe threats to coastal cities, yet systematic understanding of how underlying surface factors dynamically govern flood evolution across periods remains limited. This study establishes a bidirectionally coupled 1D drainage network and 2D surface hydrodynamic model (SWMM–ANUGA) and designs 16 compound scenarios.The Compound Flood Amplification Index (CFAI) is introduced to quantify risk amplification, and the flood process is divided into rising, peak, and recession periods using a 90% maximum-inundation-area threshold. The XGBoost-SHAP framework is then applied to disentangle the contribution intensity and switching patterns of underlying surface factors across flooding modes and evolution periods.New hydrological insights for the region: This work reveals a flood amplification effect under the extreme scenario, where inundation area and flood duration increase substantially. In the rainfall-dominated mode, terrain-driven runoff concentration prevails (DEM: 39%), with newly inundated areas concentrated in low-lying zones. In the water-level-backwater mode, drainage obstruction dominates, with the drainage system contribution surging from 21.6% to 33.7% and inundated areas extending toward downstream drainage ends. During the rising period, topography is the primary control (DEM: 32.3%); at peak, drainage proximity dominates (distance to rainwater wells: 39.4%); during recession, topography and the drainage network jointly regulate the process.

Compound floods driven by the concurrent action of rainfall and boundary water levels pose a severe threat to coastal and estuarine cities, yet systematic understanding of how underlying surface factors dynamically govern flood evolution across different periods remains limited. Taking an independent drainage system in Hangzhou, China, as a case study, this study establishes a bidirectionally coupled 1D drainage network and 2D surface hydrodynamic model (SWMM–ANUGA) and designs 16 compound scenarios.The Compound Flood Amplification Index (CFAI) is introduced to quantify nonlinear risk amplification, and the flood process is divided into rising, peak, and recession periods using a 90% maximum-inundation-area threshold. The XGBoost-SHAP framework is then applied to disentangle the contribution intensity and switching patterns of underlying surface factors across flooding modes and evolution periods. Results demonstrate a nonlinear flood amplification effect under the extreme scenario, where inundation area and flood duration increase substantially. In the rainfalldominated mode, terraindriven runoff concentration prevails (with a DEM contribution of 39%), and newly inundated areas are concentrated in lowlying zones. In contrast, in the waterlevelbackwater mode, drainage obstruction becomes the dominant factor, with the contribution of the drainage system surging from 21.6% to 33.7%, and inundated areas extending toward the downstream ends of the drainage network. During the rising period, topography is the primary control (DEM: 32.3%). At the peak, drainage proximity dominates (distance to rainwater wells: 39.4%). During recession, topography and the drainage network jointly regulate the process. This study provides a scientific basis for whole-process flood prevention and control.

Multi-source fusion techniques have emerged as cutting-edge approaches for spatial precipitation estimation, yet they face persistent accuracy limitations, particularly under extreme conditions. Machine learning offers new opportunities to improve the precision of these estimates. To bridge this gap, we propose a hybrid artificial neural network–geographically weighted regression (ANN–GWR) model that synergizes event recognition and quantitative estimation. The ANN module dynamically identifies precipitation events through nonlinear pattern learning, while the GWR module captures location-specific relationships between multi-source data for calibrated rainfall quantification. Validated against 60-year historical data (1960–2020) from China’s Yongding River Basin, the model demonstrates superior performance through multi-criteria evaluation. Key results reveal the following: (1) the ANN-driven event detection achieves 10% higher accuracy than GWR, with a 15% enhancement for heavy precipitation events (>50 mm/day) during summer monsoons; (2) the integrated framework improves overall fusion accuracy by more than 10% compared to conventional GWR. This study advances precipitation estimation by introducing an artificial neural network into the event recognition period.