Publications

Singh, MK; Gautam, R; Venkatachalam, P (2017). Bayesian Merging of MISR and MODIS Aerosol Optical Depth Products Using Error Distributions From AERONET. IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING, 10(12), 5186-5200.

Abstract
Aerosol Optical depth (AOD) routinely retrieved by spaceborne sensors (e.g., MODIS and MISR) is widely used in studies focused on atmospheric aerosol characterization, their variability/trends, and environmental/climate impacts. Despite significant advancement in the understanding of aerosol loading patterns, there exist global/regional differences among AOD products from various satellite sensors; owing to differences in retrieval algorithms, spatial/temporal resolution and sampling, and cloud-screening schemes. A merged AOD dataset combining multiple satellite retrievals is therefore desirable, which utilizes the strengths of individual products, and at the same time reduces biases relative to ground-based measurements (e.g., from aerosol robotic network (AERONET)). In view of this, a Level-2 "merged" AOD dataset based on MODIS and MISR retrievals is developed in this study using Bayesian principles, which takes into account the error distribution of AOD from AERONET data. The merged AOD dataset is demonstrated over the Indo-Gangetic Plains, in southern Asia, and is intercompared with existing satellite and AERONET data. The RMSE of the merged AOD data (0.08-0.13) is lower than the MISR and MODIS retrievals. Additionally, the merged AOD data have higher correlation with AERONET data (r within 0.89-0.93), compared to MISR (0.82-0.89) and MODIS (0.76-0.77) data. In terms of the expected error (EE), the accuracy of merged data is found to be higher, with larger percent of merged AOD within the EE envelope (74.7%-88.7%) compared to MISR (62.7%-84.9%) and MODIS (67.9%-69.8%) data. The merging methodology and resulting dataset are especially relevant in the scenario of fusing multisensor retrievals for producing long-term satellite-based climate data records.

DOI:
10.1109/JSTARS.2017.2734331

ISSN:
1939-1404