Comparative Analysis of Microscale Air Quality Response to Traffic and Built Environment from Machine Learning Perspective

With millions of people living and working in close proximity to busy roads, exposure to air pollution is a major health risk for residents in Hong Kong, a high-density city. As the city continues to grow and change, with new buildings under construction, old buildings under remodel, and increase in number of vehicles, assessing the microscale air quality response to these changes is essential to the development of planning strategies.

In this study, we conducted a one-month air quality monitoring campaign using a low-cost sensor network (LCS) along Tsuen Wan Highway, one of the busiest roads in Hong Kong. We applied proprietary pair differential filter technology to dynamically track the sensor baseline for correcting the effects of temperature and relative humidity in the gaseous modules of LCS. We introduced crowd-sourced Google real-time traffic status (TCI) as the indicator of traffic activities and frontal area index (FAI) to give temporal characteristics to built environment. Then we proposed a machine learning (ML)-based model to simulate air quality. The proposed model outperformed existing ML models and showed a high capability to reproduce hourly pollutant concentrations with cross-validation R2 > 0.71. The newly introduced indicator FAI proved to be key role in the model with an improvement of model performance by 8%.
Our monitoring results revealed three clusters of temporal air quality characteristics with similar trends. The model simulation results indicated that the air quality of these clusters responded differently to various traffic activities and built environments, although they were similarly impacted by meteorological factors. Using interpretable ML method, we found that traffic-related pollutants (TRAP) concentrations were not mainly affected by traffic activities, but rather, the impacts from the built environment played a more important role in the accumulation of TRAP in some clusters. For clusters with a higher density of built environment, the same increase in built complex resulted in a larger increase in NO2 and PM2.5 concentrations.
Our findings suggest that planning strategies should consider local features to balance urban development and air quality and provide insights to urban planners to develop strategies that mitigate the impact of urban development on air quality.