Investigating 2D/3D factors influencing surface urban heat islands in mountainous cities using explainable machine learning

Abstract

Surface urban heat islands (SUHI) pose significant risks to human health and urban sustainability. While the impact of urban features on SUHI has been widely studied, research on the influence of 2D and 3D indicators in mountainous cities remains limited. This study introduces a set of 2D/3D indicators and explores their effects on SUHI in Chongqing, a mountainous city, using explainable machine learning methods that integrate XGBoost and Shapley values. The findings show that SUHI intensity in Chongqing increased from 1.5 °C in 2009 to 2.5 °C in 2019, with high-intensity areas expanding from 1.8 % to 6.5 %, shifting from urban cores to the suburbs. Changes in both 2D and 3D urban factors significantly influenced SUHI, with 2D factors showing a greater impact than 3D factors. Notably, the percentage of industrial land contributed 25.0 % to SUHI in 2009 and 24.6 % in 2019. Among the 3D factors, building density accounted for over 15 % of the SUHI variance in 2009. Most 2D/3D factors demonstrated nonlinear effects on SUHI, emphasizing the complexity of mountainous urban systems. Specifically, 3D factors such as mean building height and terrain slope reduced SUHI in urban cores when they exceeded certain thresholds (20 m and 5°, respectively). Local SHAP analysis further revealed that the spread of industrial land exacerbated SUHI, while high-rise buildings mitigated its effects in older urban cores through shading. These insights contribute to a better understanding of SUHI dynamics in mountainous cities and offer potential strategies for its mitigation.