Optimizing the urban form is an effective way to alleviate the urban heat island (UHI) effect, and it has always been the focus of urban planning. Although previous studies have explored the effect of 2D/3D urban form on land surface temperature (LST), they have rarely revealed the nonlinear relationship from the perspective of land use and buildings. This paper, taking Wuhan as the case study, employed the random forest (RF) model to investigate the non-linear associations between features of urban form and LST. Results showed that the proportion of water area (PW) and building density (BD), proportion of green land (PG), floor area ratio (FAR), and distance to industrial land (DL) are the top five important variables that influence the LST. PW and BD are the domain factor, which together contribute more than 58.89% of the LST variances. The regulation amplitudes of these two dominant metrics for the average of LST based on regulatory planning unit are 1.5 ℃ and 2℃, respectively. Interestingly, we found the distance from the heat/cold source is more important than the attribute of land use type itself in influencing UHI. Moreover, the nonlinear relationships and marginal effect between urban form characteristics and LST are prevalent. The PW shows an obvious stepwise cooling effect on LST with three plateau phases of 10-25%, 25-45%, and more than 50%. BD has a significant heating effect on LST with slight steps. PG and DI present steady cooling effect up to 50% and 800m, respectively, after which they lose their effect. By comparison, the FAR performs heating effect up to 0.5, after which it fluctuated and keep stable at 2. These improved findings provide a scientific guidance for UHI mitigation by optimizing land use and buildings in urban planning.

urban form,land surface tempreture,urban heat island,nonlinear effect,random forest,urban planning