Nonlinear and inequality-related heterogeneity in green cooling across Chinese urban agglomerations: a SHAP-based random forest analysis.

Land surface temperature (LST) is increasing with rapid urbanization and climate warming. However, the relative roles of green quantity, vegetation condition, and inequality in local green resource availability in shaping the urban thermal environment remain insufficiently understood across regions. This study examined built-up areas in five major Chinese urban agglomerations. A Random Forest model combined with SHAP analysis was used to quantify the relative importance of green-related, natural background, built-environment, and socio-economic variables and to identify their nonlinear associations with LST. Green-related variables contributed consistently across the five urban agglomerations, although their contribution structures differed markedly among regions. NDVI generally showed stronger explanatory power than green cover. The weighted Gini coefficient of local green resource availability exhibited a strongly nonlinear and region-dependent association with LST. Cross-regional comparisons indicated greater sensitivity to changes in the weighted Gini coefficient in the Yangtze River Delta and weaker responses in the Chengdu-Chongqing region. SHAP results further indicated that built-environment morphology was also relevant alongside green-related variables. Built-up density consistently showed positive SHAP values. These findings provide interpretable cross-regional evidence for heat-mitigation planning. Effective strategies should prioritize vegetation quality, the spatial continuity of green resources, and a more balanced distribution of green space rather than focusing solely on the total quantity of green space.