Quantifying Urban Heat Island Trends in the Appalachian Region Using Machine Learning and Geographically Weighted Regression

Urban Heat Islands (UHI)—the phenomenon where urban areas experience significantly higher temperatures than surrounding rural regions—represent a critical environmental and public health challenge, especially in rapidly urbanizing areas. The Appalachian region, despite its unique ecological, economic, and historical significance, remains notably understudied in large-scale UHI modeling efforts. To address this research gap, this study develops a novel, high-resolution, spatially enriched dataset spanning 13 Appalachian states, integrating satellite remote sensing data, topographic variables, infrastructural characteristics, and demographic attributes. Employing this comprehensive dataset, the study evaluates two prominent machine learning (ML) methods—Gradient Boosting (GB) and Random Forest (RF)—alongside Geographically Weighted Regression (GWR), a spatially explicit non-ML approach, to quantify and interpret UHI intensity. Comparative analyses reveal that RF slightly outperforms GB in predictive accuracy (RMSE: RF = 2.579, GB = 2.594; R²: RF = 0.5531, GB = 0.5475). While GB and RF effectively capture broad-scale trends, their global modeling frameworks obscure critical local spatial variations. Conversely, GWR, despite moderate predictive performance (RMSE = 2.29, R² = 0.65), significantly enhances interpretability by highlighting spatially heterogeneous relationships between UHII and predictors such as vegetation, population density, and land surface temperature. This research underscores the importance of combining predictive strength with spatial interpretability, demonstrating the value of hybrid modeling frameworks. Such approaches can better inform targeted mitigation and adaptation strategies, crucial for enhancing environmental resilience in complex, diverse regions like Appalachia.