Dynamic Traffic Data in Machine-Learning Air Quality Mapping Improves Environmental Justice Assessment

Area of Science:

• Environmental Health
• Air Quality Monitoring
• Machine Learning Applications

Background:

• Air pollution is a significant public health issue in megacities, with uneven exposure patterns.
• Existing models struggle to capture the fine-scale, dynamic nature of air pollutant concentrations.
• Community-level air quality assessment is crucial for understanding health impacts.

Purpose of the Study:

• To develop a machine learning approach for continuous, community-scale air pollutant estimation.
• To integrate dynamic traffic profiles for improved spatial and temporal accuracy.
• To identify and map traffic-induced air pollution exposure disparities.

Main Methods:

• Developed a machine learning model utilizing dynamic traffic data.
• Estimated year-long concentrations of nitrogen dioxide (NO2), MDA8 ozone (O3), and fine particulate matter (PM2.5).
• Validated model performance with and without real-world traffic data.

Main Results:

• Incorporating dynamic traffic data improved spatial fidelity by 47% for NO2, 4% for MDA8 O3, and 15% for PM2.5.
• The model successfully identified PM2.5 exceedances linked to heavy traffic and mapped exposure disparities.
• Models lacking real-time traffic data underestimated PM2.5 exposure, especially in disadvantaged communities.

Conclusions:

• Machine learning models leveraging dynamic traffic data enhance the accuracy of air pollution assessment.
• Real-time traffic data is essential for capturing traffic-induced exposure disparities.
• Accurate exposure assessment is critical for addressing environmental justice and public health inequities.