Unveiling nonlinear dynamics in urban blue-green space cooling: Spatiotemporal supply-demand matching via integrated machine learning in Hangzhou

Urban blue-green spaces (BGS) are critical for mitigating heat islands, yet the nonlinear dynamics between cooling supply and demand remain underexplored. This study establishes a spatial identification-mechanistic diagnosis framework to investigate these dynamics in Hangzhou from 2016 to 2022. First, spatially explicit Urban Cooling Demand (CEDL) and Supply (CESL) indices were constructed using a Geographically Weighted Random Forest (GWRF) approach to capture local parameter sensitivity. Local Spatial Autocorrelation (LISA) was then applied to pinpoint supply-demand mismatched zones. Subsequently, LightGBM, SHAP, and Partial Dependence Plots (PDP) were integrated to diagnose the nonlinear drivers specifically within these mismatched regions. Results reveal a spatial polarization: high cooling demand areas expanded outward in dense urban cores, while high supply regions contracted inward near ecological barriers. The mismatch analysis identified dominant Low Supply-High Demand (LS-HD) zones in the city center, primarily driven by the ‘Distance to Nearest BGS’ and ‘Building Shape Index,’ with PDP revealing a critical cooling attenuation threshold at 200–300 m. Conversely, High Supply-Low Demand (HS-LD) zones in the periphery were sustained by landscape connectivity but faced increasing erosion from suburban sprawl. These findings move beyond global linear assumptions, providing data-driven, spatially targeted strategies—such as micro-park insertion in core zones and connectivity preservation in fringes—to alleviate urban heat inequity. • A “spatial identification-mechanistic diagnosis” framework integrates Geographically Weighted Random Forest (GWRF) and interpretable machine learning. • Spatiotemporal polarization intensifies: cooling demand expands outward in urban cores while supply contracts inward near ecological barriers. • ‘Distance to Nearest BGS’ and ‘Building Shape Index’ are identified as the dominant drivers of heat accumulation in Low Supply-High Demand zones. • Partial Dependence Plots reveal a critical nonlinear threshold where cooling benefits diminish rapidly beyond 200–300 m from blue-green spaces.