Improve ground-level PM_2.5 concentration mapping using a random forests-based geostatistical approach.

Area of Science:

• Environmental Science
• Geospatial Analysis
• Public Health

Background:

• Accurate ground-level PM2.5 measurements are vital for health studies.
• Satellite-derived PM2.5 datasets offer broad coverage but often lack spatial detail.
• Current methods may overlook geographic and socioeconomic influences on PM2.5.

Purpose of the Study:

• To refine satellite-derived PM2.5 datasets for the contiguous US.
• To improve spatial resolution and accuracy of PM2.5 concentration mapping.
• To integrate ground-based measurements with geographic variables using advanced statistical methods.

Main Methods:

• Employed a random forests-based geostatistical (regression kriging) approach.
• Combined machine learning (random forests) with geostatistical methods (kriging).
• Integrated ground-based PM2.5 data with geographic variables like nighttime lights, NDVI, and elevation.

Main Results:

• Achieved a refined spatial resolution (0.01°) and enhanced accuracy for PM2.5 datasets.
• Demonstrated the effectiveness of geographic variables in long-term PM2.5 mapping.
• Quantified the contribution of nighttime lights, NDVI, and elevation to PM2.5 spatial distribution.

Conclusions:

• The proposed geostatistical approach significantly improves PM2.5 data quality.
• Geographic variables are crucial for accurate and high-resolution PM2.5 mapping.
• The enhanced dataset provides a valuable resource for human and environmental health research.