A GeoML-XAI framework for identifying high PM_2.5 areas and source attribution: Application to an agricultural...

Area of Science:

• Environmental Science
• Air Quality Monitoring
• Agricultural Management

Background:

• Air pollution, particularly fine particulate matter (PM2.5), is a significant environmental health issue.
• Limited research exists on spatial assessment and source attribution of PM2.5 in agricultural landscapes.

Purpose of the Study:

• To develop and apply a GeoML-XAI framework for identifying agricultural areas with elevated PM2.5.
• To attribute PM2.5 concentrations to specific emission sources in agricultural regions.

Main Methods:

• Integrated Internet of Things (IoT) microsensor networks, machine learning (Extreme Gradient Boosting Regression - XGBR), and explainable artificial intelligence (XAI).
• Calibrated IoT microsensor data against fixed monitoring stations.
• Incorporated meteorological factors, co-pollutants, and land-use predictors into the XGBR model.

Main Results:

• Achieved high predictive accuracy for PM2.5 concentrations (R² = 0.93; RMSE = 1.94 μg/m³).
• Identified elevated PM2.5 levels in western and southwestern coastal agricultural regions, particularly those with dry farmland and rice fields.
• XAI analysis revealed co-pollutants, meteorological conditions, and dry farmland as primary contributors to PM2.5 variability.

Conclusions:

• The GeoML-XAI framework effectively identifies high-priority PM2.5 emission control sites in agricultural areas.
• Findings support data-driven, targeted air quality management strategies for rural and peri-urban environments.
• Highlights the importance of agricultural practices and associated factors in regional air quality.