Simulation of Subsurface Urban Heat Islands via the Random Forest Algorithm

The subsurface of urban areas is steadily warming, leading to subsurface urban heat islands. These phenomena confront cities with several issues. However, rigorous and expedient approaches to analyze subsurface heat islands are presently lacking. This paper presents a machine learning approach to effectively study ground temperature and strain anomalies caused by subsurface heat islands. Using the Chicago Loop as a case study, we identified physical features in support of the analysis of this phenomenon. These features were then incorporated into a random forest algorithm, which was trained by using experimentally validated finite element simulation results. The analyses reveal that high-resolution predictions across entire urban areas can be achieved based on a handful of data, with minimal calculation errors: temperature errors are typically below 0.5°C, and strain errors are under 7 με. The results also characterize the spatiotemporal distributions of ground temperatures and strains caused by subsurface heat islands.