Population exposure across central India to PM2.5 derived using remotely sensed products in a three-stage statistical model

Abstract

Surface PM2.5 concentrations are required for exposure assessment studies. Remotely sensed Aerosol Optical Depth (AOD) has been used to derive PM2.5 where ground data is unavailable. However, two key challenges in estimating surface PM2.5 from AOD using statistical models are (i) Satellite data gaps, and (ii) spatio-temporal variability in AOD-PM2.5 relationships. In this study, we estimated spatially continuous (0.03° × 0.03°) daily surface PM2.5 concentrations using MAIAC AOD over Madhya Pradesh (MP), central India for 2018 and 2019, and validated our results against surface measurements. Daily MAIAC AOD gaps were filled using MERRA-2 AOD. Imputed AOD together with MERRA-2 meteorology and land use information were then used to develop a linear mixed effect (LME) model. Finally, a geographically weighted regression was developed using the LME output to capture spatial variability in AOD-PM2.5 relationship. Final Cross-Validation (CV) correlation coefficient, r2, between modelled and observed PM2.5 varied from 0.359 to 0.689 while the Root Mean Squared Error (RMSE) varied from 15.83 to 35.85 µg m−3, over the entire study region during the study period. Strong seasonality was observed with winter seasons (2018 and 2019) PM2.5 concentration (mean value 82.54 µg m−3) being the highest and monsoon seasons being the lowest (mean value of 32.10 µg m−3). Our results show that MP had a mean PM2.5 concentration of 58.19 µg m−3 and 56.32 µg m−3 for 2018 and 2019, respectively, which likely caused total premature deaths of 0.106 million (0.086, 0.128) at the 95% confidence interval including 0.056 million (0.045, 0.067) deaths due to Ischemic Heart Disease (IHD), 0.037 million (0.031, 0.045) due to strokes, 0.012 million (0.009, 0.014) due to Chronic Obstructive Pulmonary Disease (COPD), and 1.2 thousand (1.0, 1.5) due to lung cancer (LNC) during this period.