Evaluating the seasonal effects of building form and street view indicators on street-level land surface temperature using random forest regression

Current studies of the influence of urban morphology indicators on land surface temperature (LST) usually focus on administrative or grid-based research units, and the limited inclusion of similar indicators easily occurs due to multicollinearity. This study implements Random Forest (RF) models with multi-source data, to study the relative importance and marginal effects of eight building form indicators as well as six street view indicators on street-level LST across all four seasons for Shenzhen, China. Our results show that the RF models explained 79.56%, 79.07%, 76.42%, and 64.74% of the LST variations in the spring, summer, autumn and winter, respectively. The building view factor (BVF) and green view index (GVI) were identified as the two most important indicators across all seasons. However, BVF was the dominant indicator in the spring and summer, and GVI played more significant roles in the autumn and winter. The relative importance of building density (BD), average building height (BH), standard deviation of building height (BH_SD) and sky view factor (SVF) showed noticeable variations with the seasons as well. The trends of marginal effects remained stable for each indicator across the four seasons. BVF, BD and SVF had warming effects in each season, while GVI, BH and BH_SD had cooling effects in each season. These findings contribute to our understanding of the relationship between urban morphology indicators and LST and provide valuable design suggestions for improving urban thermal environment, especially in high-density cities.