Cooling varies with green space characteristics: Unraveling nonlinear spatial heterogeneity in cooling effects of urban green spaces with geographic explainable AI

Existing research on the relationship between urban green spaces (UGS) and land surface temperature (LST) has overlooked the nonlinear mechanisms associated with multiple UGS characteristics, diverse climatic backgrounds, and particularly spatial heterogeneity. To address these gaps, we established a systematic UGS assessment framework integrating three core dimensions: UGS composition, configuration, and morphological structure. Leveraging geographic explainable artificial intelligence (i.e., GeoShapley), we then quantified the nonlinear cooling effects of various UGS metrics and their spatial heterogeneity across five cities representing distinct climatic zones in China. Results demonstrated that (1) UGS composition, configuration, and morphological structure collectively regulated cooling effects, but their contribution magnitudes, cooling thresholds, and nonlinear responses displayed significant climate-specific differences. (2) GeoShapley quantified the role of location features, which substantially reshaped UGS metric importance for cooling, and uncovered pronounced nonlinear spatial heterogeneity in UGS cooling effects across urban areas. (3) UGS with high tree cover, strong connectivity, and dominance by large patches generally exhibited robust cooling effects, yet these benefits were constrained in dense built-up areas; in contrast, in regions with strong thermal backgrounds and limited ventilation, moderately dispersed small- and medium-sized UGS patches, together with their edges adjacent to grassland and water, could enhance cooling efficiency. These findings provide climate-specific, spatially explicit guidance for targeted urban greening interventions and advance sustainable urban heat mitigation strategies. Methodologically, this study has pioneered the integration of geographic explainable AI into urban thermal environment research, offering a novel approach to disentangle complex nonlinear and spatially dependent relationships in UGS cooling effects. • UGS cooling effects are assessed considering UGS composition, configuration, and morphological structure. • Geographic explainable AI reveals nonlinear thresholds and spatially heterogeneous cooling mechanisms. • Location features (GEO) contributed substantially and reshaped the importance ranking of UGS metrics. • UGS composition exerts cooling effects at lower thresholds in arid/plateau regions than in humid/hot zones. • Moderately dispersed small–medium UGS patches enhance cooling efficiency under strong thermal backgrounds.