Deciphering the impacts of urban gray–green spaces on seasonal diurnal land surface temperature in a valley-basin city via explainable machine learning

The Urban Heat Island effect poses a major challenge to sustainable development, and understanding its driving factors is crucial for improving the urban thermal environment. The nonlinear effects of urban gray–green spaces on Land Surface Temperature (LST) under valley-basin conditions remain systematically underexplored. This study investigates Changsha, a typical valley-basin city, by integrating Morphological Spatial Pattern Analysis with landscape metrics to construct a comprehensive gray–green spaces framework. Using explainable machine learning (CatBoost-SHAP), we quantify the nonlinear and interactive effects of eight composite indicators on seasonal diurnal LST. Results show: (1) Gray space indicators contribute 51.6–59.2% to daytime LST across spring, summer, and winter, with BMSF as the dominant positive driver; green space indicators contribute 57.5–65.2% to nighttime LST, led by GACI and GMSF. (2) Threshold effects are evident: BMSF warming saturates beyond +0.5 SD, while GCI cooling diminishes after +0.5 SD, with marginal returns dropping from 0.25 °C per unit to near zero. GACI shows a unique positive effect (+0.3 °C) on winter nights. (3) Interactions reveal synergistic heating (BMSF×BCI > 0.25 SHAP), synergistic cooling (GMSF×GCI <−0.4 SHAP), and asymmetric buffering where high GMSF reduces BMSF thermal load by up to 0.6 SHAP. This study underscores gray–green interplay as a key driver of urban thermal dynamics in valley-basin cities, revealing quantifiable nonlinear balances. Based on these findings, we propose a regulatory strategy integrating spatiotemporal control, synergistic optimization, and topographic adaptation.