Traffic-related air pollution backcasting using convolutional neural network and long short-term memory approach

Area of Science:

• Environmental Science
• Epidemiology
• Data Science

Background:

• Accurate air pollution backcasting, particularly for nitrogen dioxide (NO2), is essential for understanding long-term health effects in epidemiological studies.
• Urban NO2 concentrations are influenced by traffic, vehicle technology, and regional sources, necessitating sophisticated modeling approaches.

Purpose of the Study:

• To develop and validate a spatiotemporal model for backcasting nitrogen dioxide (NO2) levels in urban environments.
• To assess historical NO2 exposure trends and their correlation with socioeconomic factors in Toronto, Canada.

Main Methods:

• Integrated Convolutional Neural Network (CNN) for spatial variability and Long Short-Term Memory (LSTM) for temporal dynamics.
• Utilized traffic-related predictors like nitrogen oxides (NOx) emissions and Annual Average Daily Traffic (AADT) derived from a Traffic Emission Prediction scheme (TEPs).
• Trained the model using NO2 measurements from the Urban Scanner mobile platform (2020-2021) to backcast NO2 from 2006-2020.

Main Results:

• The model successfully estimated spatiotemporal NO2 levels across Toronto from 2006 to 2020.
• Despite an overall decrease in NO2, the study identified an increase in exposure disparities over the period.
• Analysis revealed that marginalized communities experienced a disproportionate rise in NO2 exposure, indicating environmental injustice.

Conclusions:

• The developed CNN-LSTM model provides accurate spatiotemporal backcasting of urban NO2.
• Historical NO2 trends in Toronto show increasing exposure inequality, with vulnerable populations bearing a greater burden.
• This research highlights the need for targeted environmental policies to address health disparities linked to air pollution.