TAB E Methodology for Modeling a Possible Chemical Warfare Agent Release
In 1996, the Central Intelligence Agency (CIA) reported on computer modeling it had used to simulate possible releases of chemical warfare agents from several sites. Because the CIA used only a single model approach, their results reflected the strengths and weaknesses of only that model. On November 2, 1996, to improve computer modeling over the earlier CIA results, the DoD asked the Institute for Defense Analyses (IDA) to convene an independent panel of experts in meteorology, physics, chemistry, and related disciplines. The panel reviewed previous modeling analyses and recommended using multiple atmospheric models and data sources for future modeling to generate a more robust result than produced by a single model. The Special Assistant for Gulf War Illnesses agreed to new modeling to estimate the areas of possible exposure to chemical warfare agents that may have been released during Gulf War air operations and post-war demolitions.
The Special Assistant used computer simulations because on-site measurements of chemical agent exposure were unavailable and the local weather conditions were not always measured or recorded. To implement the recommendations of the IDA panel, the DoD and CIA asked other agencies with extensive modeling experience to participate in the modeling process. The modeling team consists of scientists from the Defense Threat Reduction Agency (DTRA) (formerly the Defense Special Weapons Agency); the Naval Research Laboratory (NRL); the Naval Surface Warfare Center (NSWC); the National Center for Atmospheric Research (NCAR); and Science Applications International Corporation (SAIC) (supporting the CIA and DTRA). The team used existing sophisticated computer models, as recommended by the IDA panel, to develop potential exposure areas specific to each incident under investigation. The team combined these models (referred to as an ensemble) to compensate for the bias that is inherent in each model, that is, to produce a more robust result by maximizing the strengths of each model and minimizing its weaknesses.
We adopted the IDA panel recommendation to use an ensemble of weather and dispersion models combined with global data sources to assess the possible dispersion of chemical warfare agents. The methodology for modeling the release of agent is a process that uses:
Figure 54 depicts the methodology we use to estimate possible hazard areas and, in the process, possible exposures of military units.
Figure 54. Process for modeling possible chemical warfare agent releases
The methodology uses two types of models: weather models and dispersion models. The weather models allow us to simulate the weather conditions in specific areas of interest by approximating both global and regional weather patterns. Based on the weather generated by a global model, a regional weather model predicts the local weather conditions in the vicinity of a possible chemical warfare agent release. Both the global and regional weather models are supplemented by actual, although quite limited, weather measurements from the Persian Gulf and surrounding regions.
The dispersion models allow us to simulate how chemical warfare agents may move and diffuse in the atmosphere given the predicted local weather conditions. These models combine the source characteristics of the agentincluding the amount of agent, the type of agent, the location of the release, and the release ratewith the local weather from the regional models to predict how the agent might disperse. Running one dispersion model with the weather conditions predicted by one regional model results in a prediction of a unique downwind hazard area. Running each dispersion model with the weather from each of the different regional weather models results in a set of unique hazard areas. We overlay all of these hazard areas to create a union, or composite, of the various projections. The resulting composite graphic provides the most credible array of potential agent vapor hazard areas for determining where military units might have been exposed. This is the basic process for all of our modeling efforts.
C. Global Models and the Database
The global weather models forecast atmospheric conditions over the globe. To perform calculations, the models mainly rely on observations collected from a global network of land-based weather stations, sea-based ships or buoys, and remote sensing satellites. Agencies such as the World Meteorological Organization distribute these data for the use of civilian agencies. Because of the vast domain (i.e., the whole globe), the global models can only provide general weather information.
We use three sources for global weather data:
These sources collect and process weather observations differently, and maintain historical archives. Their archives provided historical and simulated global weather data for our analyses. The only exception was that we reran the NOGAPS model specially for our applications.
D. Regional Models
Regional weather models, sometimes called mesoscale meteorological models, generate the estimated local weather conditions in the detail required by the dispersion models. To predict detailed local atmospheric conditions, regional models take the outputs from global models to yield weather estimates where the resolution can be reduced to a few kilometers.
We use three regional models:
These models simulate atmospheric conditions for advanced, high-resolution weather forecasting. They have the ability to simulate local weather conditions to a few square kilometers. Although these models operate differently, they all produce the detailed meteorological data needed to run the dispersion models. Because of differences in input data, modeling processes, and the physical process assumptions, each regional model produces different results. However, careful analysis of each simulation has shown that the outputs among the regional models are generally consistent.
E. Dispersion Models
Dispersion models predict possible downwind hazard areas. These hazard areas indicate how the chemical warfare agent disperses over time due to prevailing local atmospheric conditions. Dispersion model results depend on both the local weather descriptions, created by the regional weather models, and other modeling assumptions, including:
We use two dispersion models:
Because of their different inherent assumptions, these two dispersion models generate slightly different results even with the same weather inputs and source characterization. An analogy would be different weather models produce different forecasts even when using the same observed data. As a result, combining all regional weather models with each dispersion model can create multiple distinct hazard areas.
The IDA technical review panel hypothesized that because of the uncertainty of modeling and because of the consistency of the simulations, the results of each simulation are equally likely to accurately reflect what may have really occurred. Therefore, we combine all hazard projections generated by HPAC or VLSTRACK to create a single hazard area combining all exposure areas from all of the dispersion model runs. This approach provides a high probability that the exposure area included all units possibly exposed. Figure 55 depicts the process for creating a composite hazard projection.
Figure 55. Example of process for creating a composite hazard area
F. Unit and Personnel Locations
We send the hazard projection graphics derived from the dispersion models to the US Army Center for Health Promotion and Preventive Medicine (CHPPM.) CHPPM overlays the hazard projection graphics with data on US unit locations to create an exposure plot showing the areas and levels of possible exposure.
Two Department of Defense organizations, the US Armed Services Center for Unit Records Research (USASCURR) and the Defense Manpower Data Center (DMDC), provide the databases used to determine unit locations during the Gulf War and who was in each unit in the possible exposure hazard areas. We cross-reference, validate, refine, and update the unit location data to ensure the accuracy of whether a unit was in an exposure area at a specific time.
G. Modeling Considerations
Weather models represent our best attempts to approximate actual atmospheric conditions. They do not replicate reality with absolute certainty, but modern modeling techniques enable us to generate reasonably close approximations. We simulate regional weather conditions using weather modeling. Dispersion models then work with the simulated weather to project the overall agent distribution in the area. The composite hazard projection area represents an average picture. Since the atmosphere is inherently turbulent in nature, the actual concentration of agent within the hazard area might not be the same throughout the projected area. As a result, modeling predicts that the concentration of chemical warfare agent is at the exposure threshold throughout the hazard area, even though the agent may not necessarily be everywhere in the area. We can only conclude that individuals within the hazard projection area may have been exposed to the calculated concentration of agent multiplied by the time of exposure across the entire hazard area.
CIA estimated the quantity, type, and storage configuration of chemical warfare agents stored at the sites under investigation. The source characterizations tend to overstate the size of the release to minimize risk of failing to identify all of the agent that might have been released. Presenting a composite of the different modeling results is another method to minimize the risk of missing veterans who may have been in a hazard area. This is because, if we are to err, we would prefer to identify a veteran incorrectly as possibly exposed rather than fail to recognize a veteran who was exposed.