Unveiling nonlinear effects of built environment attributes on urban heat resilience using interpretable machine learning

Abstract

Built environment attributes (BEAs) play a significant role in influencing urban heat resilience (UHR). Previous research has examined the correlations and nonlinear relationships between BEAs and both land surface temperature (LST) and urban heat island (UHI) effects. Nevertheless, the investigation into the nonlinear effects of BEAs on UHR remains underexplored. Furthermore, the advantages of explainable machine learning in elucidating the mechanisms through which BEAs affect the urban thermal environment have been extensively validated. Consequently, taking Beijing, a highly urbanized city, as a case, we conducted an empirical study to investigate the nonlinear effects of BEAs on UHR. We first operationalized UHR as the differential in LST between extreme heat waves and normal heat days. Secondly, we constructed a set of influencing factors covering BEAs and control variables. Subsequently, by integrating the Gradient Boosting Decision Tree (GBDT) with SHapley Additive exPlanations (SHAP), the nonlinear relationships between BEAs and UHR are uncovered. The results demonstrate that: 1) Nonlinear relationships between BEAs and UHR are prevalent, as well as threshold effects. 2) Greening is the key BEA affecting UHR, accounting for 22.39% in contribution, with which increase, UHR increases at an accelerating rate. 3) From the city center outward, the growth of UHR exhibits a leapfrog effect, with the growth rate in the outer ring being 2.7 times that of the inner one. 4) Interactions between BEAs impact UHR. Our findings unveil the complex nonlinear effects of BEAs on UHR, clarifying the priority and optimal quantity thresholds of BEAs. We emphasize the importance of greening and urban scale, which could support decision-making for UHR planning and precise management.