Abstract: Taking the central urban area of Kunming as a case study, this research retrieved land surface radiant temperature based on Landsat 5 and Landsat 8 imagery using the single-window algorithm. The spatiotemporal evolution of blue-green spaces and their mitigating effects on the urban thermal environment were analyzed by integrating methods such as the standard deviation ellipse and landscape metrics. Furthermore, local spatial autocorrelation was employed to investigate interactions between the spatial distribution of blue-green spaces and land surface temperature (LST) gradients, thereby revealing underlying mechanisms. To further quantify these effects, we compared the performance of Ordinary Least Squares (OLS), Geographically Weighted Regression (GWR), and Multiscale Geographically Weighted Regression (MGWR) models in assessing the impact of blue-green spatial pattern indices on LST variations. The results indicate a southeastward shift in the spatial centroid of blue-green spaces, accompanied by a declining trend in their overall area—decreasing by 138.23 km² over the past 20 years. The annual dynamics were −0.49% for the first nine years and −0.56% for the subsequent decade, suggesting a relatively uniform rate of change. Among all landscape metrics, the landscape edge index, shape index, and patch density of green spaces exerted the most significant influence on LST variations. Thus, proper configuration of landscape morphology can enhance urban ecological resilience. Overall, the MGWR model demonstrated the highest explanatory power for LST level variations (R² = 0.66), providing more detailed spatial insights and providing valuable guidance for refined urban planning and climate adaptation strategies.