ANNOUNCEMENTS
Satellite observations offer high spatio-temporal resolution data, which is of great relevance for air quality characterisation, climatological and epidemiological studies. Satellite retrieval of “aerosol optical depth” (AOD) has been extensively used in predicting ambient PM2.5. For this research we used “1 km ×1 km resolution MODIS-MAIAC AOD” to predict PM2.5 concentrations over Delhi region for the year 2019. In addition to the AOD, other predictors used in the study include columnar water vapour (CWV), land use proxies (elevation and NDVI), and various meteorological parameters. These variables were used to train various linear and non-linear statistical models for predicting PM2.5. Statistical models include simple linear regression (SLR), multivariate linear regression (MLR) and generalized additive model (GAM). It was observed that the model prediction accuracies improved as we move from SLR to GAM. The comparison between the measured and predicted daily mean PM2.5 depicted an R-square (coefficient of determination) value of 0.2 6for SLR, which improved to 0.75 for the GAM. The observed RMSE values are 66.6 μg m-3 and 38.2 μg m-3 for SLR and GAM respectively. This study highlighted the superior performance of an advanced statistical model (GAM) in predicting PM2.5 from satellite AOD.
Keywords: PM2.5, AOD, Simple Linear Regression, Multivariate Linear Regression, Generalized Additive Model.
