Comparison of contrast reduction based MODIS AOT estimates with AERONET measurements

Current satellite aerosol retrieval products could be complemented by contrast reduction methods to overcome limitations regarding highly reflective or heterogeneous surfaces such as urban, desert or snow covered areas. Algorithms based on the contrast reduction principle, define contrast loss in an image, inside a pre-determined window size, as an exponential function of the Aerosol Optical Thickness (AOT) difference between two images (a reference and a polluted) acquired under similar observation geometry conditions. This paper presents a contrast reduction algorithm designed for the MODIS sensor, based on the Differential Texture Analysis (DTA) approach. It focuses on algorithm optimization by: a) determining an optimal AOT spatial resolution; b) constraining the relative observation geometry differences between polluted and reference images; and c) assessing the influence of several land cover classes on the accuracy of the retrievals. A comparison of the results obtained for 192 images acquired for the year 2005 with data from five European AERONET stations is performed to assess overall algorithm accuracy as well as the impact of the proposed improvements. Comparative analysis of the results for the various sites showed an optimal algorithm performance for MODIS images using a 39 pixel distance window, composed of only forest and urban pixels. Comparison with AERONET AOT data showed a good agreement with a correlation coefficient of 0.78. A similar correlation is found when comparing AERONET measurements and MODIS aerosol standard product. This research supports the establishment of contrast reduction methods as a potential complement to other aerosol retrieval methodologies. Future work will aim at removing the residual aerosol influence from reference images, including BRDFs to better reproduce surface heterogeneity and observation geometry influences and expanding the scope of this study to other AERONET sites so as to further test the algorithm at a global scale.