Analyzing the scale dependent effect of urban building morphology on land surface temperature using random forest algorithm

With continuous urban densification, revealing impacts of urban structures on thermal environment is necessary for climate adaptive design. In this study, random forest and partial difference plots were employed to depict the relative importance and interdependent effects of complex building morphology to land surface temperature (LST) variability. The six spatial factors of building density (BD), mean building height (MBH), building height difference (BHD), floor area ratio (FAR), building volume density (BVD) and mean compactness factor (MCF) were calculated at grids of 90, 300, 600 and 900 m. The results showed that BD, MCF and MBH exerted stable and significant impacts on LST with the highest prediction accuracy at 600 m neighborhood scale, and FAR and BVD were the least correlated to LST changes. Meanwhile, the influencing factors presented different correlation patterns with LST. Among them, the increase of BD had a positive linear effect on LST. MCF and MBH were nonlinearly correlated with the LST variation, and their threshold values of cooling effect were also identified. In addition to controlling BD, it also suggested that comprehensively arranging more small-volume buildings as well as increasing building height to enlarge shadow coverage were more conducive to ground heat mitigation.