As members of the MISR instrument team responsible for contributing data to the EOS project, we have been asked to develop a "Data Quality Assessment" plan for our Level 2 products. We have begun to develop an approach to creating what we hope will be a useful Quality Assessment (QA) product. We view this note as a contribution to the discussion about how EOS instrument teams will choose which quantities to report as Quality Flags.

The main goal of the MISR Level 2 retrieval algorithms is to obtain, as accurately as possible, the physical meaning of the Level 1 MISR radiances. An important tool in achieving this goal is to apply tests and external constraints to the data at various stages of processing, and to report the results in the form of "quality indicators." Some constraints may be applied as pre-processing steps, such as choosing a processing path based on surface terrain type, or rejecting cloudy areas. Others may affect post-retrieval analysis, such as our "ClimLikely" aerosol climatology data used to select among those aerosol models that have the lowest residuals in the fits between models and the measurements. Some will simply be passed as ancillary data without further action, in anticipation that they may prove to be of value for future analysis of the data.

For practical reasons, these constraints will be applied within the standard processing data stream for MISR Level 2 data. The resulting indicators will then be reported routinely as an aid to science users of the standard data product and for convenient monitoring at the DAAC for production control and quality assurance purposes.

Note that data often will be useful even if a quality flag is set to a non-optimal value. In some cases, such occurrences may help identify important discoveries. In developing these constraints, we found it useful to organize our effort around situations arising in at least four areas:

Instrument Performance

The instrument performance area is largely a matter of tracking the instrument behavior indicators that affect spectral, radiometric, and geometric performance. These are monitored for engineering purposes as well, and to effect updates to the instrument calibration parameters. As part of the Level 2 data stream, some subset of the Level 1 indicators will be compared with sets of limits, and the relevant performance implications encoded into data quality flags. A few examples of instrument performance areas relevant to the MISR Level 2 data are: up-to-date radiometric uncertainty estimates, dropped lines, and missing pixels.

Physical Constraints

There are many physical constraints that can be applied to the retrieval results, some of which can be used as indicators of the data quality. Some examples are: the requirement of non-negative radiances, albedo within the range of zero to one, an upper bound on the total aerosol optical depth based on the darkest pixel in the scene, etc.

Algorithmic Constraints

Since keeping track of the assumptions and numerical behavior of the algorithm is part of the development effort anyway, these constraints are relatively easy for us to identify. They include such items as: (1) performance bounds on the convergence characteristics of numerical methods (residuals and number of iterations), (2) the limits of intrinsic assumptions made in the parameterizations used (such as an ocean surface roughness model that is meaningful only within a certain range of wind speeds), (3) case limitations (such as treating pixels that may cross radically different terrain types, e.g., coasts, if the algorithm is designed to assume an "average" terrain type), and rejecting pixels that are too cloudy or with terrain too rough for the retrieval to work.

Climatological Constraints

Climatological data provide statistical constraints that could result in "warnings" rather than "errors". An "unlikely" result may mean a misinterpretation of the data, or a discovery. Flags based upon such constraints will be very helpful for the first-order analysis of the MISR Level 2 results. The ClimLikely aerosol data, for example, may indicate that it is more likely to find biomass burning particles than mineral dust particles over a tropical rain forest. Placing climatological constraints on other MISR-retrieved physical parameters (surface albedo and view-dependent reflectances, cloud cover, etc.) would require obtaining similar "climatologies" for these quantities.

The matter of anticipating which indicators will be most valuable to users once scientific analysis of the Level 2 data begins seems to warrant further discussion. Clearly we can not plan for all contingencies, and storing every conceivable parameter at every stage of data processing is impractical. We are currently planning a modest but reasonable selection of indicators in each of the four categories, but we welcome further discussion of this subject.

Acknowledgments: We thank our colleagues Lucien Froidevaux and Lee Elson of the Upper Atmosphere Research Satellite/Microwave Limb Sounder Team for useful discussions.

[Table of Contents] [Previous] [Next]