Nonlinear relationship between urban form and street-level PM2.5 and CO based on mobile measurements and gradient boosting decision tree models

Abstract Exposure to PM2.5 and CO has been proven to be closely related to physical health. Since 2012, they have been added to the pollutant list for national monitoring in Wuhan. However, the fine-scale variation in pollution, especially at the street level, is complex and requires further exploration. In this study, the influence of urban form on the street-level air pollution distribution was comprehensively assessed according to the urban factors, high-resolution meteorological data, PM2.5 and CO concentration data collected via mobile monitoring along roads in Wuhan. Furthermore, potential urban factors, including the land-use and urban form characteristics, were obtained from geographic information system. Both linear regression and gradient boosting decision tree (GBDT) models were developed to explore the relationship between the observed concentrations and the predictor variables. The modeling results demonstrate that the GBDT model, which captured the non-linear relationship, helps to better explain more of the variations in the pollutant concentrations than the linear model. This study provides insights into machine learning models for pollution prediction and demonstrate the important relationship between urban form and street-level pollutants. The results suggest that the urban form of the podium-level porosity can be set higher than 0.7 to promote the ventilation of street-level PM2.5 and CO, building density can be less than 0.35, and the standard deviation of height can be set to ~10 m in central Wuhan to mitigate street-level PM2.5. Thus, quantitatively demonstrating the impact of urban form on the PM2.5 and CO concentrations can help decision-makers with urban planning and management.