Nonlinear Drivers and Interaction Patterns of Land Surface Temperature in a Coastal Mountainous City: A Mountain–Sea–City Perspective

Against the backdrop of global warming and rapid urban development, urban thermal conditions have become an increasingly important challenge for sustainable planning, particularly in coastal mountainous cities with complex geographic settings. Taking the main urban area of Yantai, China, as a case study, this study integrates multi-source remote sensing and geographic data and applies an XGBoost–SHAP framework to examine the relative importance, nonlinear responses, spatial heterogeneity, and pairwise interactions of 11 explanatory variables associated with daytime land surface temperature (LST). The random test set achieved an R2 of 0.834, while five-fold spatial block cross-validation yielded a mean R2 of 0.786 ± 0.055, indicating relatively stable predictive performance under spatially independent validation. LST exhibited a “coastal low-temperature belt–inland thermal core–mountainous cold spots” pattern. Building density was the dominant predictor, followed by NDVI, green space ratio, distance to the sea, and elevation. Their SHAP responses were markedly nonlinear and characterized by model-derived sign transitions rather than universal physical thresholds. TreeSHAP analysis identified NDVI × building density and distance to the sea × elevation as the strongest pairwise interactions. These findings highlight the joint and context-dependent contributions of urban morphology, ecological conditions, and the mountain–sea setting, providing an integrated basis for climate-adaptive planning in coastal mountainous cities.