Understanding the spatial distribution and nonlinear drivers of the diurnal surface temperature range: Insights from ECOSTRESS and explainable machine learning

Urbanization and climate change significantly influence urban thermal environments, particularly the diurnal temperature range (DTR) and its limits, which have direct implications for thermal comfort and public health. Despite their importance, the spatial variability and underlying drivers of DTR across heterogeneous urban landscapes remain underexplored. In this study, we leveraged high-resolution thermal infrared imagery from ECOSTRESS to investigate spatial patterns in maximum (Tmax) and minimum (Tmin) surface temperatures, as well as DTR, in relation to natural, physical, and anthropogenic factors. Focusing on Seoul, South Korea, we applied a comparative modeling framework using generalized additive models (GAMs) and explainable artificial intelligence (XAI) to capture the nonlinear relationships between environmental variables and DTR. Our findings reveal that urban areas exhibit significantly higher DTR than natural landscapes, primarily driven by elevated Tmax relative to Tmin. Proximity to natural features such as forests and water bodies was found to mitigate DTR by reducing Tmax and stabilizing Tmin. Building morphology demonstrated strong nonlinear effects in terms of direction, magnitude, and threshold, with horizontal expansion increasing DTR variability and vertical densification dampening it. These insights highlight the need for targeted urban planning strategies to manage the intensifying DTR linked to rapid daytime surface heating.