Quantitative assessment of factors that influence heat vulnerability in residential areas using machine learning and unmanned aerial vehicle

Abstract

Climate change and urbanization have intensified the urban heat island (UHI) effect, significantly impacting urban living environments. While existing studies have yielded valuable insights into macro-scale thermal environments, this study shifts the focus toward microscale residential contexts, where localized urban form and land use patterns critically shape thermal conditions.

In this study, we analyzed the temporal variations in LST in a residential neighborhood of Okgye-dong, Jung-gu, Daejeon, South Korea. High-resolution thermal imagery captured by unmanned aerial vehicles (UAVs) and interpretable machine learning (ML) techniques were used to model and analyze thermal patterns at the microscale. The study site, adjacent to a river and designated as an Urban Regeneration Area, is particularly vulnerable to summer heat.

Exploratory data analysis (EDA) was conducted to examine statistical characteristics and spatial patterns, followed by confirmatory data analysis (CDA) using nonlinear regression models such as CatBoost, Random Forest, and XGBoost. The results showed that the importance of variables influencing LST varied by time of day. However, meteorological variables such as solar radiation, wind, and humidity were not included due to data limitations.

Among the key findings, alley width, shadow ratio, and distance from the river emerged as dominant variables affecting thermal conditions in residential areas. This study contributes to identifying time-sensitive drivers of urban thermal vulnerability by leveraging UAV-based imagery and ML. Based on these findings, we propose specific policy-oriented strategies for heat mitigation in urban regeneration areas, including improving airflow in narrow alleys by removing obstructions or illegal parking, expanding riverside green spaces to enhance cooling effects, and installing vertical shading elements to reduce localized heat stress and improve thermal comfort.

These results are particularly valuable for urban regeneration projects, where thermal vulnerability is often intensified by high building density and limited green infrastructure. The proposed strategies—such as optimizing alley width, increasing shade coverage, and enhancing riverside green spaces—can be effectively incorporated into localized urban redevelopment plans to improve thermal comfort and resilience.