New customized methods for improvement of the MODIS C6 Dark Target and Deep Blue merged aerosol product

The Moderate resolution and Imaging Spectroradiometer (MODIS) level 2 operational aerosol products (MOD04/MYD04), based on the Dark Target (DT) and the enhanced Deep Blue (DB) algorithms, have been providing daily global aerosol information. The MOD04/MYD04 has different data coverage for the DT and the DB algorithms due to differences in their retrieval methods. Recently in the collection 6 (C6), a DT and DB merged (DTBC6) AOD product has been introduced based on different thresholds of the Normalized Difference Vegetation Index (NDVI) to improve the data coverage. The main objective of this study is to increase the data coverage and reduce the error in the merged DTBC6 AOD product by introducing three new customized methods (DTBMX): (i) DTBM1: “use an average of the DT and DB AOD retrievals or the available one for all the NDVI values ”, as it is independent of the NDVI values, (ii) DTBM2: “use an average of the DT and DB AOD retrievals or the available one for NDVI≥0.2, and use the DB retrievals for NDVI<0.2”, and (iii) DTBM3: “use AOD retrievals from the DB algorithm for NDVI>0.3, and use an average of the DT and DB retrievals or the available one for NDVI≤0.3”. Validation of the DTBMX is conducted at a global scale from 2004 to 2014 using AERONET AOD measurements from 68 sites located in Asia (9), Europe (22), Southern Africa (8), and North (23) and South America (8), and for comparison purpose, the DTBC6 is evaluated for the same period. Results showed that the number of coincident observations of the DTBMX methods compared to the DTBC6 is increased by 31%, 41% 30%, and 108% for Asian, European, Southern African, and North and South American sites, respectively. At global scale, the number of coincident observations are increased for the DTBM1, DTBM2, and DTBM3 from 29,088 for DTBC6, to 45,937, 45,028, 37,393 which are 58%, 57%, and 29%, respectively more than the for DTBC6 observations. For an equal number of coincident observations, the percentage of retrievals within the EE is increased by between 17% to 20% and the RMSE is decreased by up to 15% for the DTBMX methods, but R is the same as for the DTBC6. The percentage above the EE is also decreased by between 43% to 55% due to greater contribution of the DB retrievals. Overall, the performance of all the DTBMX methods is much better than the DTBC6, but the DTBM1 is the most robust as it is independent of NDVI values, and significantly increases the data coverage. Therefore, it can be used operationally for global merged AOD retrievals.