General:

The quality of the L3 products are dependent upon the quality of the L2 input products. Any problem found in the L2 products will be propogated into the L3 product.

Information provided by Paul Hubanks, NASA Goddard Space Flight Center


Aerosol Products:

All Aerosol parameters with a 4D Confidence Histogram, which were processed with PGE v 4.2.0, have completely empty (blank) Mean, QAMean, StdDev, and QAStdDev arrays. This bug was corrected in PGE v 4.2.1. This corresponds to Eight Day and Monthly data files which cover the period from January through February 2004 (both Terra and Aqua); and the period from July through December 2002 (for Aqua only). All Eight Day and Monthly data which contain this bug will be reprocessed by MODAPS sometime during the 2004 calendar year.

The 8-day and monthly mean aerosol products are provided to the user for qualitative imaging. We do not recommend quantitative use of these products. It is NOT recommended to use monthly or 8-day mean aerosol optical thickness to derive aerosol radiative forcing, as the results are likely to be WRONG and MISLEADING. See explanation and recommendations below.

If only one10-km aerosol retrieval is made within a 1-degree L3 grid square, the L3_daily product will report the value of that single retrieval as representative of the entire 1-degree square. If no other retrievals are made in that grid square for the entire remainder of the month, that single 10-km retrieval made on a single day is reported as representing the aerosol for the entire grid square for the entire month. Why would there be only a single aerosol retrieval made in a particular grid square for an entire month? Most likely the region is either perpetually cloudy, snow covered, or land surface is too bright.

We expect up to 1% of the aerosol retrievals over the ocean to have residual cloud contamination. The number is somewhat higher over land. With sufficient data on a monthly basis, the effect of that contamination is minimal. However, in habitually cloudy regions (cloud fraction > 90%) few aerosol retrievals are possible. There is a high probability that the single 10-km retrieval that is now representing the entire monthly mean for the entire grid square is contaminated by clouds. Similar contamination by snow or surface features can affect lone retrievals over land.

Recommendations: Users should construct their own monthly and 8-day aerosol statistics from the L3_daily QA_WEIGHTED data. They should make sure sufficient pixel counts exist in each daily grid square to properly represent the grid square and that sufficient days of daily data exist to form a representative 8-day or monthly mean. The mean should be weighted by pixels counts. Additionally, weighting by (1 - cloud fraction) will reduce residual cloud contamination, though, depending on the cloud mask used may remove some of the highly concentrated aerosol. Aerosol size parameters (Angstrom exponent, effective radius and fine mode ratio) should be weighted by the optical thickness values as well as by pixel counts and cloud free area. This produces monthly mean values of size parameters in better agreement with the monthly mean values of aerosol optical thickness.

We refer the user to the 04_L2 disclaimer for additional detail.

Information provided by Lorraine Remer, NASA Goddard Space Flight Center


Cloud Optical Property Products:

All Cloud Optical parameters are not being computed as advertised for PGE versions prior to 4.1.0. The ojective was to create multiday values based on pixel count weighted daily values, where the pixel counts are based on the count in each cloud phase category at the bottom (end) of the decision tree used in 06_L2. Instead the information passed to the L3 algorithm from 06_L2 reflected the cloud phase category at the top (beginning) of the decision tree used in 06_L2, which does not take into account failed retrievals. Therefore the weighting factors more closely resemble results from the 35_L2 cloud mask, rather than the cloud optical retrieval algorithm. In addition the Cloud_Fraction_phase parameters also reflect (more closely) results from the 35_L2 cloud mask, rather than the cloud optical retrieval algorithm. This problem was corrected in the L3 PGE version 4.1.0, which went into operations in in late May 2003 and can be identified by a parameter name change -- the parameter Cloud_Fraction_phase was replaced by Successful_phase_Cloud_Retrieval_Fraction, which provides a more precise identification.

All Cloud Optical Thickness logarithmic statistics are set to fill values in Collection 003 and earlier HDF files. This was a software bug and has been resolved. All Collection 004 HDF files have this error corrected. Collection 004 HDF files can be identified by an ".004." version field in the HDF filename.

Information provided by Paul Hubanks, NASA Goddard Space Flight Center


Atmosphere Profile Products:

The infrared water vapor products (SDS names of the form Atmospheric_Water_Vapor*) should be considered invalid for any retrieval location where the total precipitable water vapor (SDS Atmospheric_Water_Vapor_Mean) exceeds 7.0 centimeters (especially obvious at times over the African Sahara) in Collection 003 and earlier HDF files. Invalid data may be masked though in regions where a mix of valid and invalid data yield statistical results less than 7.0 centimeters. In addition, the total precipitable water vapor in these products has a wet bias on the order of 10-20%. The Level 2 infrared water vapor algorithms and products have not been corrected in order to maintain consistency in this first year of MODIS Collection 3 products. Spatial and temporal gradients are believed to be representative and useful. Corrections will be in place for data in Collection 004.

Information provided by Liam Gumley and Paul Menzel, University of Wisconsin