Improving satellite-based PM_2.5 estimates in China using Gaussian processes modeling in a Bayesian hierarchical...

Area of Science:

• Environmental Science
• Atmospheric Science
• Statistics

Background:

• Satellite-derived Aerosol Optical Depth (AOD) is crucial for estimating ground-level PM2.5 in unmonitored areas.
• Traditional models like Linear Mixed Effects (LME) and Geographically Weighted Regression (GWR) have limitations in capturing complex spatial variations.

Purpose of the Study:

• To develop and evaluate a novel Bayesian hierarchical regression model using Gaussian processes for PM2.5 estimation.
• To compare the accuracy of the proposed Gaussian process model against conventional LME and GWR models.

Main Methods:

• Developed a Bayesian hierarchical model incorporating Gaussian processes to account for spatial stochasticity in PM2.5 concentrations.
• Included AOD, spatial, and non-spatial random effects in the model.
• Evaluated model performance using within-sample fitting and out-of-sample cross-validation (CV).

Main Results:

• The Gaussian process model achieved a cross-validation R² of 0.81.
• This accuracy is superior to GWR (R² = 0.74) and LME (R² = 0.48).

Conclusions:

• Gaussian process models demonstrate significant potential for enhancing the accuracy of satellite-based PM2.5 estimations.
• The proposed model offers a more robust approach to spatial modeling of air quality data.