A method to improve MODIS AOD values: Application to South America

Abstract

We present a method to correct aerosol optical depth (AOD) values taken from Collection 6 MODIS observations, which resulted in values closer to those recorded by the ground-based network AERONET. The method is based on machine learning techniques (Artificial Neural Networks and Support Vector Regression), and uses MODIS AOD values and meteorological parameters as inputs. The method showed improved results, compared with the direct MODIS AOD, when applied to nine stations in South America. The percentage of improvement, measured in terms of R2, ranged from 2% (Alta Floresta) to 79% (Buenos Aires). This improvement was also quantified considering the percentage of data within the MODIS expected error, being 91% for this method and 57% for direct correlation. The method corrected not only the systematic bias in temporal data series but also the outliers. To highlight this ability, the results for each AERONET station were individually analyzed. Considering the results as a whole, this method showed to be a valuable tool to enhance MODIS AOD retrievals, especially for locations with systematic deviations.