Yansu Qi, Han Li, Xiuhe Yuan, Dongmiao Zhao, Chao Liu, Journal of Information Processing Systems Vol. 21, No. 4, pp. 355-370, Aug. 2025  

https://doi.org/10.3745/JIPS.04.0354
Keywords: Deep Learning, Remote sensing, Shapley additive explanations, Urban Thermal Environment
Fulltext: PDF Full Text PubReader

Abstract

Recent studies on the thermal environment in cities have concentrated on the macro-scale level, with limited attention given to specific built-up areas. Additionally, there is scarce research on the comprehensive effects of the physical features of urban spaces on their thermal environment, especially on the regional scale. We explored how urban built environment factors influence the thermal environment using Qingdao as a case study. A neural network model with a feed-forward architecture was employed to map the complex interactions between land surface temperatures and built environment characteristics. The model’s performance was compared with traditional methods. Various built environment factors were analyzed, considering both spatial and morphological features, with data sourced from multiple channels and pre-processed for quality. The Shapley Additive exPlanations method was applied to interpret the impact mechanism, quantifying the contribution of each factor. The results indicate that an impervious area percentage significantly increased the land surface temperature in the summer, while vegetation coverage and building density helped maintain surface temperatures during the winter. This study presents quantitative insights into the importance, direction, and critical impacts of variable urban factors on the thermal environment. The findings provide useful directions for urban design and rehabilitation, especially in reducing the environmental impacts of urban heat islands.

Statistics
Cite this article