Meteorological and traffic effects on air pollutants using Bayesian networks and deep learning

Area of Science:

• Environmental Science
• Atmospheric Chemistry
• Data Science

Background:

• Traffic emissions are a primary source of urban air pollution.
• Predicting air quality requires understanding complex, non-stationary traffic impacts.

Purpose of the Study:

• To investigate traffic factor causality on air quality using big data analysis.
• To develop an accurate air pollutant concentration prediction model.

Main Methods:

• Bayesian network probability model integrated with rainfall event data.
• Generalized Additive Model (GAM) for non-linear relationships.
• Long Short-Term Memory (LSTM) network for prediction.

Main Results:

• Optimal conditions: vehicle speed > 40 km/h, traffic flow < 1200 vehicles/hour.
• Four rainfall event types identified, with Type I (long-duration heavy rain) most impacting air quality via wet deposition.
• LSTM model achieved R² > 0.9 for CO, NO, NO₂, NOx and R² > 0.8 for O₃, PM₁₀, PM₂.₅.

Conclusions:

• Traffic management strategies can significantly improve air quality.
• Rainfall characteristics are crucial factors in air quality dynamics.
• The proposed LSTM model offers high accuracy for air pollutant prediction.