Flexible Bayesian Ensemble Machine Learning Framework for Predicting Local Ozone Concentrations.

Area of Science:

• Environmental Science
• Atmospheric Chemistry
• Data Science

Background:

• 3D-grid chemical transport models like CMAQ are vital for air pollutant prediction.
• Nationwide CMAQ simulations lack the resolution for local exposure and environmental justice analysis.

Purpose of the Study:

• To develop a Bayesian ensemble machine learning (BEML) framework for downscaling CMAQ ozone estimates.
• To achieve census tract-level resolution for improved human exposure and environmental justice assessments.

Main Methods:

• Integrated 13 machine learning algorithms into a BEML framework.
• Employed three-stage hyperparameter tuning and targeted validations for accuracy.
• Utilized Shapley values to interpret subgrid gradient drivers.

Main Results:

• Successfully downscaled CMAQ ozone data to the census tract level for the contiguous US in 2011.
• Demonstrated model transferability by estimating fine-scale concentrations for 2012-2017 without retraining.
• Validated BEML's feasibility in data-limited scenarios, including future climate change impacts.

Conclusions:

• The BEML framework provides a robust method for high-resolution air quality modeling.
• This approach enhances the accuracy of human exposure and environmental justice assessments.
• BEML offers a flexible and transferable solution for air pollution downscaling, even in data-limited future scenarios.