Generating High Spatial Resolution Exposure Estimates from Sparse Regulatory Monitoring Data

Area of Science:

• Environmental Science
• Data Science
• Public Health

Background:

• Random Forest models are widely used for estimating ambient air pollutant concentrations.
• Model accuracy can be limited by extrapolation issues due to sparse measurement data.
• Enhancing model performance in data-scarce regions is crucial for accurate air quality assessment.

Purpose of the Study:

• To develop and evaluate novel approaches for improving the extrapolating ability of Random Forest models.
• To incorporate low-cost sensor data to enhance PM2.5 concentration estimations in areas with sparse monitoring.
• To assess the performance of improved models using urinary 1-hydroxypyrene as a biomarker.

Main Methods:

• Developed two approaches: a two-step backward selection and a regression-enhanced Random Forest method.
• Utilized daily PM2.5 concentrations from NAMS/SLAMS sites as the response variable.
• Incorporated satellite, meteorological, land-use, and low-cost sensor data as predictors.

Main Results:

• The two-step approach improved external validation R² from 0.49 to 0.65 and decreased RMSE from 3.56 to 2.96 μg/m³.
• The regression-enhanced Random Forest model achieved an external validation R² of 0.54 and RMSE of 3.40 μg/m³.
• Both improved models showed significant and comparable relationships with urinary 1-hydroxypyrene levels.

Conclusions:

• The proposed strategies effectively enhance the extrapolating ability of Random Forest models in areas with sparse monitoring data.
• Incorporating low-cost sensor data is a viable method for improving air pollutant concentration estimations.
• The developed PM2.5 estimation strategy offers a valuable tool for environmental health research in data-limited regions.