1. Initial CIA Analysis

In the summer of 1996, CIA analyzed the possible hazards caused by the demolition of Bunker 73 at the Khamisiyah Ammunition Supply Point on March 4, and in the Pit on March 10, 1991. The CIA analysis used the OMEGA mesoscale weather model to produce wind fields for the NUSSE4 and VLSTRACK transport and dispersion models. OMEGA initially produced 12 hours of meteorological reconstruction CIA used to provide meteorological fields to drive diffusion and transport calculations. The modeling was completed using the NCEP Global Data Assimilation and Forecast System (GDAS) archived at NCAR without additional declassified observations or detailed surface characteristics. CIA used mean meteorological conditions for NUSSE4, a Gaussian dispersion model, since it assumes uniform wind fields. The mean meteorological conditions were estimated by inspecting the VLSTRACK results, where hourly profile data for the nearest OMEGA grid point were used and by inspecting the particle trajectories predicted by OMEGA. The CIA reported the key findings on August 6, 1996.

When the CIA released its report (CIA, 1996), considerable uncertainty existed about the purported release of chemical warfare agent from the open pit at Khamisiyah on March 10. Imprecise data characterized the source parameters (e.g., the number of rockets, the mixture ratio and purity of the two agents, the exact position of the rocket stacks) and meteorological data. Therefore, active modeling of the March 10 demolition was suspended and CIA and DoD vigorously redirected their investigations toward resolving these questions. In addition, a fundamental lack of detonation characteristic data (e.g., the number of possible flyouts, the rate of leakage into spills, the initial liquid-to-vapor ratio, the droplet size distribution, and the size of the primary vapor cloud) strongly indicated that field tests be conducted to reduce these uncertainties.

The CIA conducted more detailed meteorological modeling in September and October 1996. The modeling results consisted of a series of 48-hour OMEGA runs containing a progressively richer set of observations used, including:

OMEGA’s embedded Atmospheric Dispersion Model tracked neutral particles that outlined the area over which the plume may have been advected.

Also, in mid-October 1996, the NRL performed a global meteorological reanalysis of the period from January 15, 1991, to March 15, 1991, using NOGAPS. Assimilated data included the standard operational set, the delayed Saudi Arabian observations, and the USAF Special Operations observations. The NRL, using the reanalyzed NOGAPS fields as the lateral boundary conditions, then made COAMPS simulations with 30 vertical levels and a nested grid at resolutions of 45, 15, and 5 km at 12-hour intervals starting March 8, 1991 at 1200 UTC. The CIA combined the OMEGA and COAMPS results with NUSSE4 to yield a sector of concern. The sector has a center bearing of 170 degrees and a width of 70 degrees. The sector radius varied with assumptions on the amount of agent released. Among all the cases, the maximum distance the chemical warfare agent traveled was approximately 275 kilometers, based on dispersion simulations lasting up to 48 hours.

2. IDA Panel Recommendations

As a result of the extensive uncertainties characterizing the Khamisiyah Pit release and significant potential hazards it may have posed to Coalition forces stationed downwind, the Deputy Secretary of Defense and the Director of Central Intelligence asked the Institute for Defense Analyses (IDA) to convene an independent panel of experts in meteorology, physics, chemistry, and related disciplines to review the CIA analyses of the demolitions in the Khamisiyah Pit and subsequent movement of agent to:

The panel members met from November 18 to 20, 1996, and February 13, 1997. At the first meeting, the panel reviewed the then-known facts about demolitions in the Pit, the CIA’s analyses, and DoD’s and the Department of Energy’s existing modeling capabilities. The CIA, together with its contractors, and the NRL briefed the panel on their analyses of demolitions in the Pit. In addition, their briefings detailed the significance of several assumptions, including the number of rockets the demolition could have affected, the number of flyouts, agent purity and other characteristics, and the mechanism that may have released the chemical warfare agent.

After reviewing the initial analyses and receiving briefings on other models applicable to this problem, the panel proposed these recommendations (IDA, 1997):

The panel also suggested various additional organizations to perform at least some of these analyses. The panel asked the NRL to team with the NSWC to link the COAMPS meteorological model with the VLSTRACK dispersion model; the LLNL ARAC to link its MATHEW diagnostic meteorological model with the ADPIC dispersion model, and the DTRA (formerly DSWA) to run the OMEGA prognostic meteorological model linked to the HPAC/SCIPUFF dispersion model.

At the February 13, 1997, meeting, the presented results of the dispersion assessments differed greatly. The DTRA and LLNL models showed similar initial directions for agent transport, but the DTRA models predicted the agent eventually turned west, while the LLNL models showed the agent continuing southeasterly towards the Persian Gulf. The NRL/NSWC models showed an initial trajectory toward the southeast, but the agent contours turned toward the west and then back toward the north as the wind changed. A review of modeling methodologies suggests that the coarse meteorology (2.5� -by-2.5� NOAA/NCEP reanalysis fields) LLNL failed to resolve the mesoscale features. As a result the panel viewed the LLNL results as probably less accurate.

To reduce the uncertainties in the Khamisiyah analysis, the IDA panel suggested several approaches. First, it is well established that boundary conditions supplied by the global-scale analysis are likely to dominate long-range (more than 12 hours) prediction and analysis from a limited area model. To assess the stability of meteorological reconstructions based on sparse observations, the panel asked for analyses with perturbed or denied weather observations. The NRL performed several reassessments of the meteorology with COAMPS, including "data denial" runs, which ignored observations near Khamisiyah; a constant baseline run, which held constant the COAMPS lateral boundary conditions (versus the standard 12-hour update); and a "random perturbation" run, which randomly perturbed local observations to represent observational error.

As stated above, while concurring with the potential importance and benefits of field tests, the panel emphasized a parametric approach to treating the uncertainty in the source term. In particular, estimates and bounding values for the following characteristics should describe the range of possible outcomes:

The panel recognized that another important source of uncertainty emerged from the differences in the algorithms and parametizations used in various models. The panel recommended using mesoscale meteorological models with these characteristics: (1) possess sophisticated physics, (2) apply FDDA methods to correct model errors by incorporating observations into the solutions during the simulation, (3) are adequately validated, and (4) are widely accepted in their professional communities. The panel also suggested applying an ensemble of models, including several dispersion models, to address the question of model-induced bias. The NRL was able to secure 48 hours of meteorological reconstruction from NCAR’s MM5 mesoscale model. MM5’s results were generally consistent with both COAMPS and OMEGA, although the three models predicted somewhat different areas of agent coverage. MM5 is a long-established model in the civilian meteorological community and has a larger user base than COAMPS and OMEGA.

The primary output of dispersion studies is the mean concentration as a function of space and time. However, with the Khamisiyah modeling results, the epidemiologists were primarily concerned with the time-integrated concentration (or dosage) at specific servicemember locations. These dosage levels can be used to compute the possible exposures to these forces. Dosages for the release’s full duration are useful for indicating static possible hazard, but can overestimate the risk to forces who were not at a stationary location over the release duration. As the panel suggested, the most effective presentations are periodic integrated dosage contours using a time period equal to the periods for which servicemember locations are known. For example, if servicemember locations are known only every 24 hours, then a set of 24-hour integrated dosage contours probably best summarizes the results.


1. Detailed Source Characterization

The lack of data describing the chemical agent release impeded initial modeling efforts. Fulfilling the IDA panel’s specific recommendations in most regards, the CIA and DoD further expanded their investigations to attempt to reduce uncertainty in the source term. As an alternative to producing a large matrix of possible modeling outcomes (resulting from extensive parametric studies), the CIA and DoD embarked on an aggressive effort to establish the modeling parameters characterizing the detonations. This effort consisted of in-depth interviews with five soldiers who helped rig the stacks in the Pit, extensive examination of UNSCOM records, interviews with UNSCOM personnel, and field and laboratory testing.

The CIA and DoD cooperated in designing several tests intended to obtain source characterization data for the Khamisiyah Pit detonation. In May 1997, The Dugway Proving Ground in Utah, conducted seven field trials using a total of 33 122mm chemical-filled rockets. The testing involved some trials that detonated individual rockets and other trials that detonated stacks of rockets. The test team used the information from the soldiers who placed the charges at Khamisiyah Pit to recreate the incident as closely as possible, except for the number of rockets and the use of a simulant rather than the actual chemical warfare agent. The 122mm rockets, the shipping crates, the type and location of the demolition charges, and the stacking method tests were accurately recreated. The crate construction was based on precise measurements and UNSCOM photographs. A chemical agent simulant, triethyl phosphate (TEP), that closely simulates cyclosarin’s volatility was used. The rockets were stacked as the soldiers involved in the Pit demolition described. Nudell et al., (2000) describe the tests in detail.

The tests used rocket motors identical to those detonated in the Pit. The warheads were based on UNSCOM’s detailed design parameters, including precise wall thicknesses, materials, and type of burster tube explosive. The 122mm rockets were of foreign design, with warheads manufactured in the United States specifically for this test. Each warhead was filled with an average 5.78 kg (5.4 L) of TEP, with oil red dye 2144 used as an indicator.

a. Single Rocket Detonation Tests. The test team conducted five single-rocket trials. For three trials, the testers placed one rocket in a shipping crate and detonated it in various configurations. Two of the warheads detonated normally; the other split in two. The first trial, which served as the operational ready inspection trial, produced limited data. To obtain the maximum simulant release data, the test team detonated a fourth TEP-filled warhead without a crate or rocket motor by placing C4 in the fuze well. This warhead also detonated normally. To examine the likelihood of burster function as a result of landing from flyouts, the fifth trial involved attaching one warhead to an empty motor assembly and dropping it from a helicopter at about 1500 m (5000 ft) above ground level. Only the motor assembly and a small portion of the warhead were recovered; the remaining warhead section was more than 9 m (30 ft) below the surface and could not be recovered.

b. Multiple Rocket Detonation Tests. The test team conducted two multiple-rocket trials; one with nine live rockets and three dummy rockets; and one with 19 live rockets, five dummy rockets, and one motor. In the first trial, one burster functioned, releasing TEP. An adjacent motor burned and low-order detonated; all other warheads leaked, releasing TEP. In the second trial, the burster functioned in both warheads with C4 charges. All warheads next to a C4 charged warhead leaked, with one exception. Again, Nudell et al., (2000) provide more detail on the tests.

c. Blast Modeling. The Dugway testing provided a physical basis from which to estimate the effect of charge placement on surrounding rockets in the Khamisiyah Pit. Based on these rocket tests, the DoD/CIA team hypothesized that during the Khamisiyah Pit detonation, the only warheads that burst and aerosolized agent were those on which soldiers placed charges just beyond the warhead’s nose, and that only the warheads immediately adjacent to the charges leaked agent. Even the rocket dropped to simulate a flyout did not disperse any simulant; it buried itself more than 9 m below the surface.

To substantiate these empirical findings and develop criteria for the onset of warhead leakage, sympathetic detonation of burster function in other rockets, and rocket motor initiation, CIA performed numerical calculations for the single- and multi-rocket cases with the Radiative Adaptive Grid Experiment (RAGE) hydrodynamic model, an explicit Eulerian code with adaptive mesh. For the multi-rocket case, CIA ran two-dimensional plane (a plane normal to the axis of the warhead, with the origin at the center of the warhead) calculations to determine the environment produced by the demolition and detonated burster charge. CIA modeled the detonated bursters as 2.7 cm diameter line charges surrounded by 12.2 cm diameter rings of water. All other rockets in the stack were treated as fixed internal boundaries in the computational grid. The presence of crates were ignored because, relative to the massive rockets, these crates blow away in short order. These calculations confirmed the results of the rocket detonation testing: if individual C4 charges are placed on several individual warheads in a multi-rocket stack, the likelihood of producing a sympathetic burster detonation, even in the unlikely event a warhead was positioned between two detonated warheads, is very low.

The RAGE calculations also showed that in a rocket whose burster has detonated, the greatest threat to initiate the motor is if the exploding burster penetrates the motor axially. Single-rocket calculations of the pressure outside the motor casing indicated that it was very unlikely that the exploding burster penetrated the motor axially.

CIA also performed RAGE calculations to predict the environment of full-scale configurations. According to the Dugway test results, and confirmed by CIA’s blast modeling, the primary release mechanism for nerve agent at the Khamisiyah Pit consisted of pooled liquid spilled from ruptured warheads and the evaporation of agent from the saturated wood crating. The warhead damage was more likely to lead to agent leakage caused by impulsive loads originating in adjacent detonations. There were relatively few warheads that burst due to direct placement of charge. Based on the Dugway tests, CIA developed various criteria for warhead leakage depending on the warhead’s location relative to the nearby C4 demolition charges, the warhead’s orientation, and whether the rocket was located on the edge of a stack. Subsequently, CIA generalized these criteria to the larger stack configurations found at the Khamisiyah Pit in March 1991.

d. Rocket Damage Modeling. CIA’s computational studies allowed the DoD/CIA team to develop a computer model to determine the rocket damage resulting from various placements of charges and orientations of rockets in an arbitrary stack configuration. In effect, this model established the total number of damaged rockets that released agent into the atmosphere and the distribution of released agent as vapor, liquid droplets, spills on soil, and penetration into wood. The model’s logic incorporated the generalization of Dugway’s test results into CIA’s numerical calculations. For a particular warhead positioned in a stack, the model algorithm distinguishes among:

The agent release model is deterministic, reflecting the Dugway test results and supporting calculations. The code is sensitive to the placement of the C4 demolition charges and the orientation of the stacked rockets, i.e., either facing forward or backward.

The pie chart in Figure A-22 shows the model’s predicted distribution of agent as aerosolized vapor, droplets, spilled into soil and wood, burned, and unaffected. The demolition released only about 32% of the agent, mostly by leaking into the soil and wood. A total of 18% became part of the plume-2% through aerosolization and 16% through evaporation (5.75% from soil and 10.4% from wood).

Figure A-22. Pie chart of Dugway tests showing partitioning of the source term

Figure A-22. Pie chart of Dugway tests showing partitioning of the source term

e. Evaporation Studies. The Dugway field tests and other supporting calculations indicate a large percentage of the agent eventually leaked into the soil and the wood crates’ debris. This result underscored the importance of additional work conducted at Dugway and Edgewood laboratories to study the evaporation of agent from different media. The two laboratories initially planned to perform tests on soil. However, based on the Dugway test results, the tests were expanded to include wood. These tests involved saturating wood and soil with a mixture of sarin and cyclosarin and then measuring the rate at which the agent evaporated. The tests also were designed to closely replicate conditions in the Pit, including:

Figure A-23 presents the Dugway laboratory test results. Of particular interest, most of the chemical warfare agent was predicted to have evaporated during the first ten hours. Thereafter, the evaporation rate was slow and large portions of the agent stayed in the soil and wood. In addition, soil tests at Edgewood indicated about 12.5% of the agent degraded in the soil in the first 21 hours.

Figure A-23. Dugway test evaporation curve

Figure A-23. Dugway test evaporation curve


| First Page | Prev Page | Next Page |