Estimation of PM2.5 concentrations using aerosol optical depth derived from satellite remote sensing

Abstract

Satellite-derived aerosol optical depth (AOD) may provide information on atmospheric aerosols for entire Bangladesh. However, estimating particulate matter (PM) concentration using AOD has some challenges. Relation between PM and AOD is region-specific and depends on meteorological variables. In the present study, statistical models were developed for estimating PM concentration over several sites in Bangladesh using AOD and meteorological parameters. Univariate model (AOD) showed poor performance (R2 <0.1) in PM estimation. The inclusion of meteorological parameters improved the accuracy of the model. Multivariate linear model with AOD and surface meteorology explains variability up to 39% for PM2.5 and 33% for PM10. AOD, temperature, and relative humidity account for 97% of total variability explained by the model for both PM2.5 and PM10. Dominance analysis shows that, for PM2.5 estimations, the temperature variable is more significant than relative humidity during the dry season and the opposite during the wet season. On the other hand, relative humidity is the most important variable for PM10 estimation during both dry and wet seasons for PM10 estimation. The model’s accuracy significantly varied at different sites (R2 value ranging from 0.51 to 0.21) which indicate geographical influence is present. Site-wise validation of the model revealed that its accuracy is higher at sites located in Dhaka, Chittagong, and Barisal and lower in Narayangonj and Khulna. Validation over the entire study period shows that the model tends to underestimate PM at higher concentrations. However, the inclusion of first-order interaction terms did not result in enhanced accuracy. For monthly PM estimation, the multivariate linear model showed good performance (R2 = 0.77 for PM2.5 and 0.74 for PM10) although, the contribution of AOD is less than the contribution of AOD for daily PM estimation. For both daily and monthly PM models, AOD has a positive association with PM whereas both temperature and humidity have negative correlations. These models have the potential to explain the temporal and spatial variability of daily and monthly PM over Bangladesh. More ground-based PM monitoring stations may improve the model so that it can be used to predict PM with greater accuracy.