Estimating PM2.5 concentrations in a central region of China using a three-stage model

Owing to uneven environmental monitoring site distribution, there are significant spatial data gaps for concentrations of ambient fine particles with diameters <= 2.5 mu m (PM2.5) obtained using traditional monitoring methods. Satellite products are an alternative data source for locations where monitoring sites are unavailable. The Moderate Resolution Imaging Spectroradiometer (MODIS) aerosol optical depth (AOD) product has been widely used in PM2.5 assessment for years; however, it has obvious data gaps in winter. Here, the Visible Infrared Imaging Radiometer Suite (VIIRS) AOD was applied to supplement MODIS AOD data to obtain a fused AOD dataset. A three-stage model consisting of a corrected AOD model, mixed effects model, and geographically weighted regression model was developed and used with meteorological and vegetation factors to estimate PM2.5. Results showed overall model fitting by cross-validation (CV) with an R-2 of 0.92, mean absolute error of 5.72 mu g/m(3), and root mean square error of 7.15 mu g/m(3). The combination of MODIS AOD and VIIRS AOD was a suitable method for enhancing AOD coverage. The CV R-2 value of the three-stage model (0.92) was higher than that of the two-stage model (0.9). Hence, the three-stage model could achieve a better fit in estimating PM2.5 on a regional scale.

Automated summary

Due to the uneven distribution of environmental monitoring sites, there are data gaps in concentrations of PM2.5 obtained using traditional methods. Satellite products, such as MODIS AOD, can be used as an alternative data source. However, there are data gaps in winter. This study used VIIRS AOD to supplement MODIS AOD and developed a three-stage model to estimate PM2.5 with high accuracy.