Decoding spatial patterns of urban thermal comfort: Explainable machine learning reveals drivers of thermal perception

Abstract

Thermal comfort (TC) is a pivotal indicator of urban quality of life and influences public health, productivity, and satisfaction. This study leverages remote sensing data from 2019 to 2023 to construct a national-scale TC framework using the Modified Temperature and Humidity Index (MTHI). This analysis reveals the spatial heterogeneity of TCs across China and their key driving mechanisms. The findings show a northwest–southeast gradient in the TC, with a decreasing contrast in this direction and a north–south disparity alongside a northward shift in heat discomfort centers. High-comfort zones are found in the western plateaus, northeastern regions, and southern mountains; conversely, low-comfort zones are concentrated in the northwest, North China Plain, and southern basins, particularly in the densely urbanized eastern coastal cities and central urban clusters. Coastal areas show high internal variability, whereas inland plateaus are more stable. Natural environmental factors have emerged as the primary drivers of TC. Shapley additive explanations (SHAP) values show a continuous upward trend, underscoring the crucial role of enhanced NDVI in improving TC. Although socioeconomic factors show increased SHAP values, their adverse impacts persist as urbanization and rising building density exacerbate TC deterioration. Landscape factors exert complex effects on TC, with water body landscapes displaying an optimal regulatory range. Interactions among driving factors, characterized by direct and complex trade-offs, further modulate and intensify the effects of TC. The proposed multiscale optimization framework provides strategic insights for managing China's urban thermal environment and offers guidance for other regions with similar climate challenges.