Quantifying Multi-pollutant Co-exposure via Deep Learning-Based Simultaneous Prediction Using Geostationary Satellite...

Area of Science:

• Environmental Science
• Data Science
• Public Health

Background:

• Accurate hourly assessment of multiple air pollutants is crucial for understanding human co-exposure and health impacts.
• Existing monitoring methods face spatial and temporal limitations, hindering comprehensive analysis.
• Pollutant mixtures pose complex health risks that require advanced analytical tools.

Purpose of the Study:

• To develop and validate Deep Learning for Multiple Air Pollutant analysis (DeepMAP), a novel deep learning framework.
• To enable simultaneous prediction of six major air pollutants (PM10, PM2.5, O3, NO2, CO, SO2) at hourly resolution.
• To identify co-pollution hotspots and characterize co-exposure patterns for targeted public health interventions.

Main Methods:

• Developed DeepMAP, a deep learning framework for simultaneous multi-pollutant prediction.
• Validated the model's performance across diverse regions and pollutant types.
• Utilized a novel co-exposure index to pinpoint regions with significant pollutant contributions.

Main Results:

• DeepMAP achieved robust performance, with normalized RMSE below 0.36 for all pollutants during high-concentration episodes.
• Identified frequent co-exceedance events: PM10-PM2.5 (91 days/year) in East Asia, followed by PM10-PM2.5-NO2, PM2.5-O3, and PM10-PM2.5-O3.
• Pinpointed hotspots for PM10-PM2.5-NO2-O3 co-exceedance in the North China Plain, East China, and South Korea.
• Revealed distinct hotspot regions where NO2 contribution to co-exposure was significantly higher.

Conclusions:

• DeepMAP offers a high-resolution, data-driven solution for analyzing multi-pollutant dynamics.
• The framework accurately captures complex co-pollution episodes and identifies critical exposure areas.
• Findings support enhanced air quality management strategies and public health protection efforts in vulnerable regions.