Stereoscopic urban morphology metrics enhance the nonlinear scale heterogeneity modeling of UHI with explainable AI

Abstract

In urban heat island (UHI) effect modeling, traditional 3D morphological indexes often homogenize elements within units, leading to a loss of spatial detail. Here we introduce detailed stereoscopic urban morphology indices, including the Height Variation Coefficient, the Height Gravity Index, the Mixed Degree, and the Aggregation Degree, using urban building and vegetation height data to represent the 3D structures within urban units. We employed ensemble learning algorithms to explore nonlinear effects and statistical correlations across 20 buffer radii (100 m to 2000 m) related to the UHI effect. The integration of these inhomogeneous indexes substantially improved model accuracy, with XGBoost achieving an R2 of 0.95. The optimal scale for calculating such indexes was found to be between 300 m and 400 m. Our findings suggest optimizing UHI mitigation by lowering the Building Height Coefficient in high-density areas and increasing it in low-density regions. Furthermore, we propose tailoring the Vegetation Height Coefficient for green spaces, advocating higher values for larger parks and lower values for smaller ones. This approach offers practical insights for mitigating urban heat effects through customized morphological strategies, emphasizing the importance of precise urban planning and environmental management.