Urban heat islands (UHIs), in which urbanized areas exhibit higher surface and air temperatures than surrounding non-urban areas, are often mitigated through greening. However, conventional buffer-based land surface temperature (LST) analyses collapse nearby conditions into one average and cannot identify the distance at which each factor is most influential. This study developed a distance-explicit multi-ring buffer framework for Seoul, South Korea, using Landsat-derived LST, Sentinel-2 vegetation indices, detailed land cover, topography, and tree-based machine learning. Concentric 50–500-m distance bands preserved radial information, and SHapley Additive exPlanations (SHAP) values were used for interpretation. The framework exhibited robust performance across XGBoost, LightGBM, Random Forest, and CatBoost; however, its main contribution was revealing non-monotonic distance effects hidden by single-buffer averages. The Normalized Difference Built-up Index (NDBI) peaked at the 100-m ring, vegetation and water indicators showed the strongest cooling in the 50–100-m band, and elevation effects were strongest at 200–350 m. Road-greening scenarios indicated detectable cooling within approximately 150 m of the restored corridors under early summer daytime conditions. The thresholds are specific to Seoul, but the framework can support locally calibrated UHI mitigation guidelines in other locations.