Vertical variations of low-altitude thermal and wind environment across different 3D urban morphologies in Singapore
Authors:
Ruohan Xu, Nyuk Hien Wong, and Shisheng Chen
Abstract:
Rapid urbanization intensifies urban heat and weakens natural ventilation, and this has become a critical concern in high-rise tropical cities such as Singapore. Although previous studies have related urban morphological (UM) parameters to thermal or wind conditions based on surface or pedestrian observations, systematic and citywide assessments of low-altitude vertical profiles of both air temperature and wind speed across diverse three-dimensional urban morphologies remain limited. This study integrates a large ensemble of ENVI-met simulations with machine learning models and Accumulated Local Effects (ALE) interpretation to quantify citywide height-specific relationships between morphology parameters and both temperature and wind speed. Built-up areas were classified into six morphology clusters using GIS-derived UM parameters, and 650 representative 200 m × 200 m sites were selected. Each site was simulated under a common clear, hot-day boundary condition to obtain air temperature and wind speed from 1.5 m to 61.5 m. For each height, separate regression models were trained using the corresponding temperature or wind speed and UM parameters to identify the most influential predictors and their effects. Non-linear machine learning models, particularly Extreme Gradient Boosting, substantially outperformed multiple linear regression, achieving RMSEs of about 0.22 °C for temperature and 0.34 m/s for wind speed. ALE further quantified the height-specific impacts of key UM parameters on temperature and wind, revealing systematic shifts in their dominant controls with height. The resulting height-specific response patterns were translated into design-readable ranges of morphology indicators that can support urban cooling and ventilation strategies in dense tropical environments.
Keywords:
Urban morphology; Low-altitude microclimate; Thermal and wind environment; Urban cooling and ventilation; Machine Learning; ENVI-met simulation