IntroductionUnder global climate change and rapid urbanization, the continuous intensification of the surface urban heat island (SUHI) effect has become a critical issue threatening public health and urban sustainable development. However, the nonlinear coupling mechanisms among socioeconomic activities, blue–green infrastructure, and three-dimensional urban morphology remain insufficiently quantified, constraining the scientific formulation of fine-scale planning interventions.MethodsTherefore, taking Wuhan, a typical high-density “furnace city” in China, as a case study, this study integrates multisource data, develops and compares six regression algorithms, and couples the optimal model with the SHAP game‐theoretic interpretability framework to systematically identify the global importance of driving factors, nonlinear threshold responses, and bivariate interaction mechanisms.ResultsThe results show that: (1) the LightGBM model outperformed all comparison models (R2 = 0.631, RMSE = 2.781), confirming the superiority of interpretable ensemble learning methods in modeling the thermal environment of high-density cities; (2) socioeconomic intensity is the dominant driving force, with GDP and population density contributing most strongly to LST (>18%) and exhibiting significant nonlinear growth and thermal saturation effects; (3) blue–green infrastructure exhibits critical cooling threshold characteristics, with the water index (NDWI) and vegetation coverage ratio (VR) needing to reach specific thresholds to trigger significant cool‐island effects; and (4) interaction analysis reveals a severe spatial mismatch between “high heat sources” and “low cool sinks,” whereby areas of intensive economic agglomeration are often accompanied by a scarcity of blue–green spaces.DiscussionBased on these findings, this study constructs a spatial optimization framework for the urban thermal environment oriented toward “source–sink” balance, advocating stock renewal in high‐density built‐up areas, threshold‐anchored connectivity protection for blue–green networks, and the construction of ventilation corridor systems based on topographic characteristics. Methodologically, this study extends the application boundaries of explainable machine learning in urban climate research; practically, it translates complex nonlinear patterns into implementable and spatially differentiated planning strategies, thereby providing a scientific basis for mitigating the SUHI effect in high-density cities.