In the context of global climate change, this study aims to investigate the spatial heterogeneity and driving mechanisms of the urban heat island (UHI) effect within Xi’an’s second ring road area. We constructed a novel multi-source data fusion framework that integrates high-resolution remote sensing imagery, detailed building spatial data, and semantic indicators from street view imagery. Based on this framework, we extracted seven key environmental features and land surface temperature (LST) data. We employed Multi-scale Geographically Weighted Regression (MGWR) and machine learning models, including Random Forest, XGBoost, and Gradient Boosted Regression, to analyze both nonlinear interactions and spatially localized variations influencing UHI intensity. The results indicate that building density (BD), green view index (GVI), and road density (RD) are the dominant factors affecting LST, showing significant spatial heterogeneity. BD has the highest global importance with a SHAP value of 0.665 in the XGBoost model and shows positive effects on LST, especially in high-density areas. GVI exhibits stable negative correlations with LST, highlighting its cooling potential in medium- to high-density zones. MGWR regression coefficients for BD and GVI range from −0.66 to 1.38 and −0.53 to 0.33, respectively, revealing substantial local variation. Our analysis reveals the necessity of spatially differentiated climate adaptation strategies, and confirms the effectiveness of fine-grained environmental indicators in representing UHI formation mechanisms. The proposed multi-source data fusion and integrated MGWR-machine learning framework offers refined methodological tools and practical insights for enhancing urban thermal resilience and developing targeted microclimate regulation policies.
Meng et al. (2025) studied this question.