The ocean algorithm attempts to match calculated reflectances to measured reflectances. To do this the ocean algorithm selects all combinations of the four fine mode and five coarse mode aerosol models and varies the amount of each to find solutions within pre-defined error limits.  All solutions within the error limits are used to calculate an “average” value.  The solution with the lowest error is used for the “best” value.  We recommend that users select the “average” value for ocean products.

Both of these variables are merged land and ocean AOD at 0.55 µm from the dark target algorithms. The “Optical_Depth_Land_And_Ocean” SDS only contains high quality pixels (QA=3 over land, QA > 0 over ocean). The “Image_Optical_Depth_Land_And_Ocean” SDS contains all valid (non-fill) pixels; this product can be used to trace aerosol plumes and make aerosol maps, but should not be used for quantitative study.

Ocean and land products use completely different algorithms. For any dark target retrieval algorithm we must accurately account for surface effects and build correct models of the aerosols.  Due to the complexities of the land surface it is more difficult for the land retrieval to account for surface effects. In addition aerosol models are fixed for the land product based on season and location.  The ocean product selects coarse and fine mode models dynamically.

These two factors account for several differences in the two products.

Quality (QA) flags are an indicator of the algorithm team’s assessment of the quality of the data.  QA flag values range from 0 – 3 where 0 is lowest quality. We advise using only highest quality data (QA=3) for land products.  For ocean products we advise using anything above QA zero.  In collection 6 data this information is included in the SDS “Land_Ocean_Quality_Flag”.

 It is normal and expected that the response of some of the MODIS detectors will degrade over time.  Due to some hardware issues with Terra the response of some of the detectors have degraded at an accelerated rate.  The MODIS calibration team analyses the changes in response over time to adjust the calibration of the sensor.  The calibration changes are applied to the data so that the final products should only be minimally affected by the changes in response.  One main reason it is taking longer to create the Terra collection 6 product

Users need to make this determination for themselves based on the particular needs of their study.  For Collection 5 we feel Aqua is the more reliatble sensor due to the issues discussed in the FAQ "Is there a difference between Terra vs Aqua AOD?"   However for many types of studies data from either sensor can be used. Collection 6 data will require further study but we expect Aqua to be the more reliable sensor.

Although sensor and algorithm design are identical between the two sensors, over time the sensors have decayed at different rates.  The rate of decay of Terra has been greater than Aqua. Consequently there is a small offset between Terra and Aqua AOD (.015 in Collection 5). Additionally, there is a decreasing trend of global land AOD in Terra Collection 5 (Levy et al., 2010), which is not observed in Aqua. This has been traced to a calibration issue (e.g. Wang et al., 2012), and is expected to be corrected in Collection 6.

At this point, the MODIS aerosol team has not evaluated the reliability of other parameters, including certain derived parameters (e.g. Mass Concentration) and diagnostic parameters (cloud fraction). The user is cautioned that there is no expected quantitative value to these parameters. 

In Collection 6, the preliminary estimate error is 0.45; pixels with an AOD > 0.2 are expected to have a more accurate Angstrom Exponent.

In collection 5 the estimated errors are:

   Land:   AOD * 0.15 +/- 0.05

   Ocean: AOD * 0.05 +/- 0.03

 These are global estimates from the entire data set.  Errors in the product vary from region to region and season to season. For a site by site analysis see the maps in our “Validation” section. The Collection 6 product is currently being analyzed and error estimates have not yet been published.