Exploring the Nonlinear and Spatial Effects of Urban Activity Heterogeneity on the Nighttime Thermal Environment Using Machine Learning and GWR

Jingyi Lin, Wei Gan, Xiang Li, Xiaochi Xu,
Exploring the Nonlinear and Spatial Effects of Urban Activity Heterogeneity on the Nighttime Thermal Environment Using Machine Learning and GWR,
Building and Environment,
2025,
113928,
ISSN 0360-1323,
https://doi.org/10.1016/j.buildenv.2025.113928.
(https://www.sciencedirect.com/science/article/pii/S0360132325013988)
Abstract: The urban heat island (UHI) effect significantly influences both the quality of life and productivity of citizens, particularly at night, when elevated surface thermal radiation reduces the comfort of public spaces. While previous studies have predominantly emphasized the role of urban vegetation and surface albedo in mitigating the UHI effect, limited research has investigated the impact of urban activities on nighttime land surface temperature (NLST). This study aims to identify key drivers of NLST in the central urban area of Wuhan by integrating data on urban morphology, activity intensity, and green infrastructure. Several machine learning models including Random Forest (RF), XGBoost, Support Vector Machine (SVM), and LightGBM were compared based on predictive accuracy (R², RMSE, MAE), and an optimal approach combining XGBoost and Geographically Weighted Regression (GWR) was identified. This hybrid model effectively identifies activity drivers of NLST and employs SHapley Additive exPlanations (SHAP), Partial Dependence Plots (PDP), and Sankey diagrams to interpret nonlinear effects and visualize contribution pathways. Results reveal that, in addition to vegetation coverage (NDVI) and albedo, activity indicators such as the density of points of interest (POI_density) (importance score I = 0.09) and commercial activity intensity (I = 0.05) play a substantial role in shaping NLST. Furthermore, spatial heterogeneity of urban activity impacts is evident, with pronounced increases in NLST observed along both sides of the Yangtze River. These findings highlight the combined influence of urban activities and environmental factors on nighttime thermal conditions and provide insights for climate-responsive planning.
Keywords: Nighttime land surface temperature; urban heat island; urban activity; Geographically Weighted Regression; machine learning