Can we Quantify the Uncertainty in Aerosol Optical Depth Retrieved by the Dark Target Aerosol Retrieval Algorithm?

The “Dark Target” (DT) algorithm uses the look-up table (LUT) approach for retrieving aerosol properties from satellite observations of spectral reflectance. Specifically, the DT algorithm has been routinely used to retrieve AOD from MODIS –observed reflectance, resulting in decade-plus long records from MODIS instruments on Terra and Aqua. The retrieved AOD product has been extensively validated, by comparing against collocated AOD from ground-based AERONET sunphotometers. These studies have shown that globally, and for best quality data, ~67% of MODIS-retrieved AOD agrees with AERONET-observed AOD to within expected error (EE) envelopes of ±(0.03+10%) over ocean and ±(0.05+15%) over land. In addition, these studies have found where, when, and sometimes why MODIS and AERONET do not agree within EE.

The process of “validation” determines the EE for the global population, but does not tell a user the uncertainty of an individual retrieval. There is uncertainty in retrieved AOD owing to non-linearities in the LUT approach. There is uncertainty due to various assumptions made prior to and during the retrieval, including errors in the observed reflectance (calibration or missing data), standard deviation of the reflectance over the retrieval box, non-aerosol (e.g. gas absorption and Rayleigh) atmospheric corrections, and characterization of the surface reflectance. These sources of error are some of the factors that limit the agreement between retrieved and AERONET-observed AOD.

As a next step forward, the DT retrieval team is developing a methodology for estimating the overall uncertainty on a per-retrieval (per-pixel) basis. This means quantifying each of the above mentioned error sources, and determining how these error sources will impact the final aerosol retrieval. It is expected that overall uncertainty will be reported in a future aerosol product. In this paper, we will present some example results and explore importance of different sources of uncertainty in AOD retrievals.

This uncertainty information is especially useful for modeling studies that assimilate satellite AOD. It is also important for studies that use satellite observations (such as AOD) for application oriented studies such as assessing the radiative effects of aerosols. Quantifying the uncertainty in AOD retrievals will help to assess the uncertainty in radiative forcing estimates. In addition to application to MODIS data, this uncertainty estimation can be used for any present or future instruments (e.g. VIIRS) that retrieve aerosol properties using the DT algorithm.