Characterizing the Thermal Effects of Urban Morphology Through Unsupervised Clustering and Explainable AI

Abstract

Highlights: What are the main findings? A data-driven classification using the K-Means unsupervised clustering algorithm reveals that urban morphology is a primary determinant of the thermal environment. Among the classified types, ‘Compact Mid-rise’ exhibits the highest temperatures, whereas ‘Open High-rise’ is the coolest. The Normalized Difference Built-up Index (NDBI) is identified as the most significant warming factor, while the Sky View Factor (SVF) emerges as the most crucial cooling factor. However, the precise influence of these factors is highly contingent upon the specific urban morphology type. The influence of three-dimensional (3D) urban morphology on Land Surface Temperature (LST) is both nonlinear and dichotomous. For instance, within compact built-up areas, increasing Building Height (BH) and density presents a double-edged effect. On one hand, it can impede heat dissipation, leading to higher temperatures through a ‘heat trapping’ effect. On the other hand, it can provide a cooling benefit by blocking solar radiation via a ‘shading’ effect. What is the implication of the main finding? These findings advocate for a shift in urban cooling strategies, moving away from a ‘one-size-fits-all’ approach towards precisely targeted policies tailored to different local urban morphologies. For instance, urban planning should prioritize the optimization of spatial building layouts in compact zones, whereas in open, low-density areas, the strategic deployment of green infrastructure should be the primary focus. Urban morphology significantly mediates the thermal effects of different land use functional zones. For functional zones with high anthropogenic heat emissions, such as industrial districts, planning interventions should favor sparse or open layouts to mitigate thermal stress on adjacent areas. The urban thermal environment poses a significant challenge to public health and sustainable urban development. Conventional pre-defined classification schemes, such as the Local Climate Zone (LCZ) system, often fail to capture the highly heterogeneous structure of complex urban areas, thus limiting their applicability. This study introduces a novel framework for urban thermal environment analysis, leveraging multi-source data and eXplainable Artificial Intelligence to investigate the driving mechanisms of Land Surface Temperature (LST) across various urban form types. Focusing on the area within Beijing’s 5th Ring Road, this study employs a K-Means clustering algorithm to classify urban blocks into nine distinct types based on their building morphology. Subsequently, an eXtreme Gradient Boosting (XGBoost) model, coupled with the SHapley Additive exPlanations (SHAP) method, is utilized to analyze the non-linear impacts of ten selected driving factors on LST. The findings reveal that: (1) The Compact Mid-rise type exhibits the highest annual average LST at 296.59 K, with a substantial difference of 11.29 K observed between the hottest and coldest block types. (2) SHAP analysis identifies the Normalized Difference Built-up Index (NDBI) as the most significant warming factor across all types, while the Sky View Factor (SVF) plays a crucial cooling role in high-rise areas. Conversely, road density (RD) shows a negative correlation with LST in Open Low-rise areas. (3) The influence of urban form is twofold: increased building height (BH) can induce warming by trapping heat while simultaneously providing a cooling effect through shading. (4) The impact of land use functional zones on LST is significantly modulated by urban form, with temperature differences of up to 2 K observed between different functional zones within compact block types. The analytical framework proposed herein holds significant theoretical and practical implications for achieving fine-grained thermal environment governance and fostering sustainable development in the context of global urbanization.