Evaluation of MODIS Collection 6 aerosol retrieval algorithms over Indo-Gangetic Plain: Implications of aerosols types and mass loading

This study evaluates the performance of MODerate resolution Imaging Spectroradiometer (MODIS) Collection 6 (C6) AOD retrieval algorithms, including Dark Target (DT) aerosol optical depth (AOD) at 3 and 10 km spatial resolutions, Deep Blue (DB) AOD at 10 km, and the merged DT-DB AOD at 10 km across the Indo-Gangetic Plain (IGP), South Asia. A total of 14,736 collocated Aqua MODIS C6 AOD at 550 nm were evaluated against AOD from six AERONET stations over IGP, measured during the satellite overpass (+/- 1 h) from 2006 to 2015. The effects of aerosol heterogeneity, in terms of both aerosol loading and the aerosol type, on the uncertainty of the satellite-borne AOD retrieval were examined. The DT algorithm at both resolutions (3 km and 10 km) overestimated the AOD by 14-25%, with only 51.37-61.29% of the retrievals falling within the expected error (EE). The DT 3 km algorithm underestimates the surface reflectance in comparison to the DT 10 km, with the latter outperforming the former both in terms of number of collocations and retrieval accuracy, especially over urban areas. The DB 10 km was able to retrieve AOD over both arid/desert regions and vegetated surfaces even under low aerosol loading conditions. Yet, its performance was still poor, with retrieval accuracy of 53.76%, low RMSE (0.214), and generally underestimated AOD across the IGP. The merged DT-DB AOD product was mostly dominated by DT retrievals (73%-100%), except over bright land surfaces and 56.03% of the merged DT-DB retrievals fell within the EE. The retrieval accuracy of MODIS C6 products was found to be strongly dependent on the estimated surface reflectance and the aerosol type. Across IGP, DB predicted the surface reflectance better while DT at both resolutions overestimated the surface reflectance at varying extent. For high aerosol loading conditions with varying aerosol size, retrieval accuracy of DT 10 km poses lower sensitivity while DT at 3 km exhibits larger uncertainty in estimating surface reflectance. In contrast, DB 10 km shows greater bias that depends on the aerosol size. For very high aerosol loading conditions, dominated by fine or mixed aerosols, all the algorithms have errors in the aerosol model. The DT 10 km, DB 10 km and the merged AOD performed almost equally within the threshold level while the DT 3 km showed the poorest performance in terms of retrieval accuracy and RMSE. We conclude that across IGP, DB 10 km AOD has the highest accuracy in retrieving fine mode aerosols while DT 10 km AOD has almost identical accuracy in retrieving varying aerosol types. For coarse dominated aerosols, when the dissimilarity between DT and DB remains highest, the merged AOD is found to have higher accuracy in retrieving AOD across IGP.