Improving satellite-driven PM_2.5 models with Moderate Resolution Imaging Spectroradiometer fire counts in the...

Area of Science:

• Environmental Science
• Atmospheric Science
• Remote Sensing

Background:

• Surface PM2.5 prediction models commonly use satellite aerosol optical depth, meteorological, and land use data.
• Satellite-retrieved fire information has not been extensively utilized in statistical PM2.5 prediction models.
• Fires are known significant contributors to ambient PM2.5 concentrations.

Purpose of the Study:

• To investigate the utility of remotely sensed fire count data for improving PM2.5 concentration prediction.
• To evaluate the impact of fire data on PM2.5 prediction accuracy within a spatial statistical model framework in the southeastern U.S.

Main Methods:

• Developed spatial statistical models for PM2.5 prediction using satellite-derived aerosol optical depth, meteorological, land use, and fire count data.
• Conducted a sensitivity analysis to determine optimal buffer zone radius (75 km) for fire count data.
• Performed cross-validation (CV) to assess model performance, including R-squared, mean prediction error, and root-mean-square prediction errors (RMSPE).

Main Results:

• Incorporating fire count data improved PM2.5 prediction accuracy, with an R-squared of 0.69 and RMSPE of 4.29 µg/m³.
• Prediction accuracy gains were more substantial at sites with higher fire counts, showing up to a 13.4% improvement.
• Fire count data demonstrated better performance in southern Georgia and during the spring season, correlating with higher fire occurrence.

Conclusions:

• Remotely sensed fire count data provide a measurable improvement in PM2.5 concentration estimation.
• The inclusion of fire data is particularly beneficial in areas and seasons characterized by frequent fire events.
• Fire count data represent a valuable predictor for enhancing the accuracy of PM2.5 monitoring and forecasting.