In-Depth Evaluation of Aqua MODIS Collection 6 AOD Parameters Over the Contintinental U.S. Via Comparison to Both Ground-Truth and Modeled Data

Abstract:

We evaluated all four MODIS Collection 6 aerosol AOD parameters: 10 km Dark-Target, 3 km Dark-Target, 10 km Deep-Blue, and 10 km merged Dark-Target and Deep-Blue over the continental U.S. for the years 2011-2013 using AERONET observations. General results of this evaluation are illustrated in the attached figure, which includes data from 84 permanent AERONET sites and 64 DRAGON sites. There are indications of positive retrieval error in the AOD over the continental U.S. for Dark-Target and merged AOD parameters, such that slopes are greater than one, and the percentage of observations above the error envelope (EE, ±(0.05 + 0.15*AERONET AOD) is greater than the percentage below. In contrast, Deep-Blue has a large number of values within the error envelope. However, the correlation with ground observations is poor (r=0.73), the bias is relatively high (0.03) and the slope is below 1 (0.77). While coverage for Deep-Blue retrievals has been improved in Collection 6, the 10 km merged parameter, while partially dependent on Deep-Blue retrievals, performs poorly with regards to coverage, particularly for lower confidence values. For this parameter, an average of only 40.2% of pixels in a valid AERONET-MODIS collocation has any retrieved values. This is in comparison to 72.9% of Deep-Blue pixels and 59.5% of Dark-Target pixels in the same 10 km product. Correlation coefficients between MODIS and AERONET AOD over the Western U.S. are significantly lower (between 0.67 and 0.71) than those in the East, (between 0.84 and 0.93). However, Dark-Target and merged AOD parameters from the West do not show overall positive retrieval errors, and have regression slopes against AERONET observations between 0.98 and 1.02. MODIS aerosol products are further combined with information from the MODIS 16-day gridded NDVI (Normalized Difference Vegetation Index) product, Global Multi-resolution Terrain Elevation Data (GMTED2010), and the National Land Cover Database (NLCD) to elucidate ground conditions which could affect both the accuracy and coverage of AOD retrievals, providing a more nuanced picture of MODIS AOD data over the US. In addition, the spatial pattern of AOD at the national scale is evaluated against simulated AOD by atmospheric chemistry models GOCART and GEOS-Chem.