Characterizing the relationship between the spatial range of influence of urban land characteristics and surface temperature using geospatial explainable artificial intelligence models

The Urban Heat Island (UHI) effect leads to increased energy consumption, and a decline in urban residents' quality of life. Therefore, quantitatively analyzing this phenomenon and developing mitigation strategies is of critical importance. This study applies a Geo-Explainable Artificial Intelligence (GeoXAI) approach to quantify the influence of urban spatial configurations and land use characteristics on Land Surface Temperature (LST). LST was derived from Landsat 8 satellite imagery. Independent variables included vegetation indices such as the Normalized Difference Built-up Index (NDBI), and Green Normalized Difference Vegetation Index (GNDVI), as well as digital elevation models (DEM) and land cover data. Four tree-based machine learning models were compared. Among them, XGBoost demonstrated the highest prediction accuracy with an R² value of 0.9885 at the 150 m buffer distance. Additionally, the application of Shapley Additive Explanations (SHAP) identified NDBI, GNDVI, elevation (DEM), and roads as the most influential variables on LST. Furthermore, a scenario simulating the underground conversion of major arterial roads in Seoul and the restoration of the surface into urban parks revealed an LST reduction effect of approximately 0.45–1 °C, depending on vegetation density. These findings underscore the importance of green space restoration in mitigating the UHI effect.