Predicting fine-scale daily NO₂ over Mexico City using an ensemble modeling approach

Area of Science:

• Environmental Science
• Atmospheric Chemistry
• Epidemiology

Background:

• Accurate air pollution data is vital for epidemiological studies, but localized prediction models are scarce outside the US and Europe.
• New satellite instruments like TROPOMI offer opportunities for improved air quality monitoring.
• Nitrogen dioxide (NO2) is a key air pollutant with significant health implications.

Purpose of the Study:

• To develop and validate a localized, fine-scale air pollution prediction model for daily ground-level NO2 concentrations.
• To utilize novel remote sensing data and advanced machine learning techniques for NO2 estimation.
• To provide high-resolution NO2 data for epidemiological research in the Mexico City Metropolitan Area.

Main Methods:

• A four-stage approach was employed, integrating data from the Ozone Monitoring Instrument (OMI) and TROPOMI.
• Random Forest (RF) and Extreme Gradient Boosting (XGBoost) were used for data imputation, calibration, and residual modeling.
• A Generalized Additive Model (GAM) was used to ensemble predictions, achieving a 1-km² grid resolution.

Main Results:

• The ensembled model achieved a cross-validated R² of 0.87.
• The cross-validated root-mean-squared error (RMSE) for the GAM was 3.95 μg/m³.
• The model successfully reconstructed fine-scale NO2 estimates from 2005 to 2019.

Conclusions:

• The developed multi-stage model demonstrates high accuracy and provides valuable fine-scale NO2 estimates.
• This approach leverages new satellite data and advanced modeling for improved air quality assessment.
• The reconstructed NO2 data can significantly enhance future epidemiological studies in Mexico City.