Land use pattern impacts on urban flood resilience in Wuxi: a multi-scenario analysis integrating PLUS simulation and explainable machine learning

Rapid urbanization challenges urban flood resilience, yet the mechanistic links between land-use patterns and long-term resilience dynamics remain under-explored. This study proposes an integrated simulation-evaluation-interpretation framework to examine these impacts in Wuxi, China. Resilience for 2020 was assessed using an entropy-weighted Pressure–State–Response (PSR) model. Subsequently, the patch-generating land use simulation (PLUS) model simulated 2035 land use configurations under three scenarios: natural development (ND), urban priority (UP), and ecological protection (EP), with driving mechanisms attributed to eXtreme Gradient Boosting-SHapley Additive exPlanations (XGBoost-SHAP). Results show that land use patterns exert decisive, spatially differentiated influences. The UP scenario incurs a resilience debt, with extensive impervious expansion driving low-resilience clusters along urban fringes. Conversely, the EP scenario delivers the most substantial improvements by enhancing landscape connectivity. While land use/cover change (LUCC) remains the dominant driver, factor sensitivities exhibit a shift. EP strengthens natural-environmental regulation, whereas UP amplifies socio-economic dependence. These findings suggest that resilience-oriented planning must move beyond quantitative targets toward spatial configuration management, prioritizing ecological integrity and decoupling urban growth from risk accumulation. This framework offers a methodologically adaptable template for stress-testing resilience in flood-prone regions.