Optimizing multidimensional land use for flood regulation supply-demand matching: Evidence from a GWRF-SHAP model

Enhancing Flood Regulation Ecosystem Services (FRES) offers an effective way to reduce urban flood risks. Land use influences both FRES supply and flood risk, thereby altering their supply-demand balance. However, current studies are mostly descriptive, and the few mechanism studies confined to either function or form provide only a partial view of land use impacts, obscuring complex nonlinear and spatially heterogeneous mechanisms and thus limiting targeted environmental management. To address this gap, this study constructs a framework integrating multidimensional "function-form-pattern-intensity" land use characteristics with the GWRF-SHAP model, an interpretable spatial machine learning approach. Validated in the flood-prone metropolis of Tianjin (China), this framework reveals the nonlinear and spatially heterogeneous mechanisms driving the FRES supply-demand ratio. Furthermore, K-means clustering was utilized to translate these mechanistic insights into zoning strategies for decision support. Results reveal severe mismatches, notably with 23.12% of units classified as "low supply-high demand". Our analysis identifies that land use intensity and pattern exert a stronger influence than form or function, and further uncovers distinct critical nonlinear thresholds for key indicators, particularly land use intensity (LUI) and built-up area cores (B_core). Spatially, driving mechanisms vary significantly: high-density areas are constrained by POI density, suburbs by LUI and B_core, while peripheral ecological zones are driven by landscape connectivity. Overall, we reveal pronounced FRES mismatch and highlight the need for targeted, threshold-aware land use interventions. This study advances FRES research from description to mechanism-based management, providing a replicable framework for optimizing land use toward flood-resilient and sustainable urban management.