基于BO-LightGBM框架的城市功能区形态对地表温度的非线性关系及协同效应研究OACHSSCD

Against the backdrop of intensifying global climate change,enhancing urban thermal resilience through planning interventions has emerged as a crucial pathway toward sustainable urban development.However,evidence to support such interventions remains limited at the scale of urban functional zones(UFZs)—the practical units through which land-use regulation,renewal,and public-space governance are typically implemented.Existing studies often assess land surface temperature(LST)using administrative units or raster grids and emphasize linear effects,which may obscure threshold responses and synergistic interactions among morphology,natural conditions,building form,and human activity.This study therefore investigates UFZ-based thermal differentiation and quantifies the nonlinear and interaction mechanisms through which UFZ morphology shapes LST.We focus on the central urban area of Wuhan in China and integrate multi-source datasets,including Landsat-derived LST and NDVI,refined EULUC China 2.0 functional data corrected through AOI-based parcel segmentation and manual verification,road and water networks,building footprints and heights,points of interest(POIs),and SDGSAT-1 nighttime lights.UFZs are delineated via parcel segmentation and functional identification,and four representative types are analyzed:administrative(including major transport hubs),commercial,industrial,and residential.A set of 13 predictors is constructed across four dimensions:natural environment(e.g.,NDVI and distance to water),UFZ morphology(Area,an aggregation indicator PDI,fragmentation FRa,and landscape shape index LSI),building form(e.g.,building density,mean height,sky view factor,building-area variability,and mean nearest-neighbor distance),and human activity(facility density and nighttime light intensity).To capture complex nonlinearity and heterogeneity across UFZ types,we develop a Bayesian-optimized LightGBM framework(BO-LightGBM)and benchmark it against OLS,random forest,and XGBoost.BO-LightGBM delivers consistently superior predictive accuracy and generalization.Model behavior is interpreted using SHAP main effects and SHAP interaction values to quantify both marginal contributions and cross-factor synergies.LOWESS curves with confidence intervals are further used to visualize response shapes,while generalized additive models provide an additional robustness check for key nonlinear patterns.Results show that UFZs function as structural units of thermal differentiation:industrial and commercial zones form persistent hotspots,whereas residential and administrative zones are cooler on average but contain localized heat clusters.Morphological effects are strongly UFZ-dependent and threshold-governed.In residential zones,a moderate aggregation level(PDI ≈ 65.50-218.51 m)corresponds to a"small-block,dense-network"configuration and strengthens cooling,while excessive aggregation or over-dispersion weakens this benefit.Interaction analysis indicates that moderate fragmentation can amplify vegetation-related cooling via enhanced edge effects,whereas in highly aggregated areas higher facility density may partly offset warming by reducing travel demand and associated anthropogenic heat.These findings support UFZ-specific heat-mitigation strategies that combine morphology-oriented renewal with green-blue infrastructure networking and demand-management measures,thereby enabling more refined and evidence-based thermal governance.