Beyond CFD: explainable machine learning for efficient assessment of urban morphology impacts on pedestrian level wind and thermal environment

Abstract

Urban morphology plays a crucial role in affecting pedestrian level wind velocity and air temperature. Most studies use CFD to simulate these parameters, which requires substantial computational resources, detailed inputs, and modeller expertise. To assess the impact of urban morphology on these parameters, comparative analyses are often used but are highly dependent on pre-designed scenarios. The study established two XGBoost-based models to predict wind velocity and air temperature efficiently while applying SHAP to investigate the effect of urban morphology. SHAP results revealed that the horizontal and vertical distances from the array centre showed a greater influence on microclimate parameters than conventional building design variables. This study introduces efficient machine learning models predicting pedestrian-level wind and thermal environment with high temporal granularity while providing quantifiable insights into the relationships between urban morphology variables and microclimate conditions through SHAP-based analysis, offering practical design optimization support for creating more comfortable urban environments.

Article Content