The Panel also recommends assessing the magnitude of data-induced uncertainties through ensemble and perturbation methods, which systematically vary input data, and/or data-denial methods, which examine the impact of removing certain observations from the analysis. The combination of using more than one model and of varying the inputs provides a comprehensive approach to understanding the uncertainties contributed by the reconstruction of the meteorology at the time of the demolition.

The Naval Surface Warfare Center also investigated meteorological variations by systematically offsetting the detonation in time (plus or minus 1 hour) and space (60 km north, south, east, and west), producing transport and diffusion results and recentering them on the actual Khamisiyah coordinates. This is somewhat less computer-intensive than varying the meteorological models, and can be used to provide additional insight into possible variations.

Estimating Transport and Diffusion

A variety of models are available for estimating agent transport and diffusion given a meteorological reconstruction. Ideally, a model used for this should account for spatially and temporally varying horizontal and vertical mixing. Rather than rely on Pasquill-Gifford nomogram estimates of atmospheric stability, transport and diffusion models should compute this mixing directly from atmospheric parameters. They should also be configured for medium- and long-range transport. Among the models which have these features are SCIPUFF, HYSPLIT 4, HYPACT, and LPDM. Gaussian models using Pasquill-Gifford stability categories such as VLSTRACK are less preferred, although VLSTRACK has the advantages that it is configured with a chemical agent data base, can interface easily with the COAMPS model, and is widely used within the Department of Defense. The NUSSE4 model does not accept varying meteorological input but does contain a number of sophisticated algorithms dealing with, among other things, mixtures of chemical agents and evaporation, that are important elements of the Khamisiyah release.

Dealing with Uncertainty/Presenting the Results

We have discussed the unknowns, uncertainties, and variabilities in the source and meteorological data. These can be addressed through systematic parametric variation of these data. A second source of uncertainty emerges from the differences among the models. Even models that simulate the same physical phenomena can differ in their outputs because of the use of different solution algorithms, data gridding, and other factors.

One way to address these concerns is to apply several appropriate models. If they agree, then there should be less concern for model-induced bias.

| Previous Page | | Next Page |

| First Page | Prev Page | Next Page | Back to Text |