Dependence between urban morphology and outdoor air temperature: A tropical campus study using random forests algorithm

The study proposes a nonparametric approach that uses machine learning to predict outdoor temperatures based on field measurements at the National University of Singapore (NUS) Kent Ridge campus from February 2019 to July 2019. Six urban morphology variables (e.g. BDG, PAVE, WALL, HBDG, SVF, GnPR) were extracted from geographical information system (GIS) maps, three dimensional (3D) model and field surveys. This study compares the predictive power between ordinary least squares linear regression (LR) and machine learning (e.g. random forest (RF)). By using RF as a regression model, the air temperature has a greater RMSE reduction than LR, ranging from 10 % to 33 %. The relationship between outdoor air temperature and urban morphology variables based on non-parametric regression is presented. On average, lower SVF space can reduce the heat on campus. As the WALL increases, the temperature rises, but the changes in day and night are different. Greenery reduces daytime temperature due to the shading of solar radiation, although the shading from trees blocks the release of long-wave radiation from artificial materials at night. The observation indicates nighttime temperature can be reduced by planting low height greenery while keeping SVF unchanged.