Simulating influences of land use/land cover composition and configuration on urban heat island using machine learning

Abstract

Urban heat island (UHI) has become a worldwide concern under global warming and urbanization waves, which pose significant threats to human health and urban sustainability. Despite Land Use/Land Cover (LULC) being regarded as a crucial factor influencing UHI, few studies simulated the nonlinear influences from LULC compositions and configurations, especially considering the difference between monocentric and polycentric cities. Therefore, we measured the patterns of LULC and UHI from 2006 to 2022 using cases of a monocentric city (Chengdu) and a polycentric city (Chongqing). Further, we simulated future and land surface temperature (LST) using Machine Learning (ML) models, and further calculated the simulated UHI in 2030 by urban-rural LST differences. The results showed that Chengdu had prominent UHI intensity in cores and the surrounding satellite towns. Chongqing's main center and subcenters had high UHI intensity under clustered patterns. ML results indicated that built-up percentage was the most crucial factor driving LST. Configuration factors dominated in impacting Chengdu's LST, while composition factors were more critical in Chongqing. Moreover, future UHI is estimated to decrease in Chongqing's core, compared with the remaining high UHI in Chengdu's core. These results might provide insights for understanding and mitigating future UHI effects.