Urban Village Waterlogging in Shenzhen: Unveiling Mechanisms through Multi-Scale Buffer Spatial Analysis and Machine Learning during the “9·7” Rainstorm

Urban waterlogging has become increasingly severe under the combined pressures of extreme rainfall and rapid urbanization. Urban villages, as high-density and low-income settlements, are particularly vulnerable. Taking the “9·7” extreme rainstorm in Shenzhen as a case, this study investigates 623 urban-village study units across the Longgang River and Shenzhen River basins to identify the key drivers of waterlogging risk. Specifically, we examine the dominant role of the topographic low-lying effect and the amplifying role of street-canyon characteristics in urban-village vulnerability.We integrate multi-source data and construct a multi-ring buffer framework to quantify contrasts between urban villages and their surrounding environments. An XGBoost model, together with SHAP and partial dependence plots (PDP), is used to interpret factor contributions and interaction effects. The results show that: (1) the topographic low-lying effect dominates waterlogging risk and exhibits a clear threshold-switch pattern, with absolute elevation as the primary driver; (2) within a 0–400 m range, urban villages display pronounced morphological contrasts with their surroundings, forming a risk “vulnerability ring”; (3) street-canyon indicators are more important than conventional density metrics, highlighting the critical role of micro-scale morphology in regulating runoff pathways; and (4) maintaining the sky view factor (SVF) within 0.3–0.55 can effectively mitigate waterlogging risk. Overall, this study elucidates the synergistic driving mechanisms and key thresholds of waterlogging in urban villages, providing a scientific basis for targeted resilience enhancement.