Due to the sensitivity of the notification process, the significance of the follow-up epidemiological studies, and the intense interest of the atmospheric science community, the DHS convened a panel of experts in the fields of meteorological and turbulent diffusion modeling to review the DoD and CIA work. The panel, consisting of Richard Anthes, University Corporation for Atmospheric Research; Bruce Hicks and Will Pendergrass, National Oceanic and Atmospheric Administration; and Steven Hanna, George Mason University, participated in a meeting at George Mason University in Fairfax, Virginia on November 4-5, 1997. The primary purpose of the panel discussion was to assess the scientific credibility of the Khamisiyah modeling results publicly released in July 1997, with an additional assessment of the modeling methodology and model selection. The panel members received a copy of the draft technical report one week before the meeting.

At the meeting, a CIA contractor briefed the Pit demolition, the Dugway rocket tests, and the Edgewood and Dugway agent evaporation tests. Meteorological modelers briefed their models’ capabilities, supporting publications, configurations, and modeling results and performance measures. Dispersion modelers from the CIA (NUSSE4), NSWC (VLSTRACK), and DTRA (HPAC/SCIPUFF) presented their analyses of the Pit release and an overview of their model’s validation.

The panel prepared a December 11, 1997, report, "Comments by Peer Review Panel on Khamisiyah Modeling Report and Presentations on 4-5 November 1997" (Anthes et al., 1997). On the question of using outputs from several independent meteorological and dispersion models to determine the area within which ground personnel were at risk of exposure, the panel stated: "The panel endorses the DoD decision to base its estimates of potentially-affected personnel on the union of all the model outputs."

The panel discouraged using computationally inexpensive operational models and suggested using state-of-the-art high-resolution models with the fewest physical limitations and assumptions. In addition, the panel commented on the significance of several topics, including spatially- and temporally-varying atmospheric stability, accurate mixing-layer heights, and inconsistencies in the dispersion models’ results using the same nominal meteorological input data. On modeling strategy and the selection of models, the panel noted:

• COAMPS also is credible, with supporting peer-reviewed articles. Its reanalysis is comprehensive and the error diagnostics indicate the baseline run meets or exceeds the performance expectations for mesoscale meteorological models in general.
• OMEGA, the DTRA-sponsored model, does not yet have corresponding published peer-reviewed articles as of 1997. In addition, the panel found some OMEGA solutions questionable (e.g.,10� C temperature errors and wind speed errors of more than 20 knots).
• MM5 is a very credible model (with about 300 peer-reviewed articles) and was run by one of the model’s developers. However, the panel found that the 1997 model simulations were not as complete compared to the DoD mesoscale analyses.
• Comparing the meteorological model results with observed trajectories of the oil-well fire smoke plumes may be misleading because of the uncertainties about the elevations of the observed plumes.
• Since material transported above the mixing layer may have dominated the oil-well fire smoke plumes, the panel assigned greater importance to comparisons with local soot patterns at Khamisiyah.

The panel also issued these findings on the dispersion models:

• Of the three dispersion models that were used, only HPAC/SCIPUFF had been subject to a thorough peer-review process and published extensively in refereed journals.
• VLSTRACK and HPAC/SCIPUFF assumed different transport speeds for near-surface puffs, potentially leading to large inconsistencies.
• VLSTRACK and HPAC/SCIPUFF used different source inputs (i.e., internally versus externally specified), which complicated comparisons.
• Other US agencies sometimes also use Eulerian or hybrid Eulerian-Lagrangian. HPAC/SCIPUFF and VLSTRACK are Lagrangian models.

In addition to the issues raised by the George Mason University panel, agent removal effects (e.g., decay and vapor deposition) also required serious consideration. The July 1997 results did not fully represent the agent removal effects because empirical data was lacking and more inclusive hazard areas were preferred. The Senate Special Investigative Unit on Gulf War Illnesses also identified these omissions in its August 1998 report. Other deficiencies in the 1997 modeling are as follows:

• Inconsistencies In Dispersion Calculations For Same Source And Same Meteorology

  Dosage patterns for the July 1997 VLSTRACK and HPAC/SCIPUFF model runs based on the same meteorological input showed a discrepancy (Figures A-58 and A-59). Dispersion results strongly depend on the puff tracking heights for near-surface releases. For all runs conducted prior to the November 1997 George Mason University panel review meeting, VLSTRACK used either the puff centroid or the height of interest as the effective tracking height, while HPAC/SCIPUFF used the maximum of the centroid and 0.6 cloud vertical sigma. Subsequent to the review meeting, NSWC ran VLSTRACK using the HPAC/SCIPUFF near-surface tracking algorithm, which resulted in plumes with trajectories quite consistent with the corresponding HPAC/SCIPUFF output (Figures A-58 and A-60). NSWC has since proposed a vertical tracking and splitting algorithm using the maximum of the centroid and 0.6 cloud sigma.

  

  Figure A-58. HPAC/COAMPS4 predicted four-day cumulative
  dosage contours

  

  Figure A-59. VLSTRAC/COAMPS4 predicted four-day cumulative
  dosage contours (original tracking height)

  

  Figure A-60. VLSTRAC/COAMPS4 predicted four-day cumulative
  dosage contours (revised tracking height)

• Inconsistency in the Source Term

  Because of the pressure to release modeling results quickly, the July 1997 simulations used inconsistent source terms (e.g., the internal evaporation algorithm for VLSTRACK versus the DPG evaporation curve for HPAC/SCIPUFF). A common source characterization, the DPG evaporation curve, was used for all subsequent simulations.

Therefore, the DHS intensified its efforts to obtain more refined exposure assessments. The inconsistencies in the puff tracking algorithm and the source term mentioned above were resolved for the refined (2000) modeling. In addition the following sections describe in detail the improvements made to meteorological modeling and the unit location and personnel data. The results are more refined estimates of possible exposure to servicemembers from the demolition in the Khamisiyah Pit.

a. Models’ Improvements

Since 1997, several new simulations of MM5, OMEGA, and COAMPS were made to update the description of the mesoscale meteorological conditions near Khamisiyah. The large-scale fields used to initialize the mesoscale models and to provide the lateral boundary conditions are largely the same as those used in the 1997 modeling. The improvements made to COAMPS, OMEGA, and MM5 are described below.

1) COAMPS. As mentioned before (Table A-9) in 1997, in addition to the baseline analysis (Run COAMPS 7), alternative COAMPS analyses were performed that consisted of (Westphal et al., 1999):

• An analysis where all observations within the COAMPS 15-km grid were denied to the data assimilation process (Run DD15 or COAMPS 3);
• NOGAPS and COAMPS analyses where all observations within the COAMPS 45-km grid were denied to the data assimilation process (Run DD45 or COAMPS 4); and
• NOGAPS and COAMPS analyses where random, bounded errors were introduced to the global observation database according to the type and altitude of the observation (Run RPO or COAMPS 6).

The baseline analysis was used in the 2000 modeling. In 1997, however, COAMPS 4 was chosen for dispersion modeling based on comparison to the ground soot patterns due to bunker explosions. COAMPS 4 is equivalent to running COAMPS in pure forecast mode with no data assimilation. The peer review panel questioned the decision because intuitively a data-denial run should give inferior results. Furthermore, the COAMPS developers also recommended the use of the baseline analysis. Therefore, COAMPS 7 was used in the 2000 Modeling.

2) OMEGA. The following improvements incorporated into OMEGA version 3.7 used for the 2000 modeling:

• OMEGA 3.7 now runs with 12-hour intermittent FDDA. Previous OMEGA runs were made in a pure forecast mode, i.e., the observations were assimilated only once at the start of the integration;
• The OMEGA 3.7 grid generator is capable of producing higher resolution around selected points (i.e., enabling observational sites, thus improving the incorporation of those observations into the initial conditions);
• OMEGA 3.7 uses a time-split advection solver, which groups the cells according to the maximum advective time-step possible in the cells. Thus, the high-resolution portions of the domain can be integrated with a smaller time-step, while the low-resolution regions use longer time-steps. This increases the computational efficiency of the model;
• OMEGA 3.7 has an improved semi-implicit solver to damp acoustic modes, resolving acoustic problems observed with OMEGA 2.0. In addition, near the top of the model domain, OMEGA 3.7 has several model layers (sometimes called sponge layers) with the damping factors chosen to further prevent spurious waves from being generated and propagated downward;
• OMEGA 2.0 used a fairly simplistic first-order turbulence closure scheme (the O’Brien scheme) for the Khamisiyah simulations. OMEGA 3.7 uses a more complete, higher-order k-e turbulent kinetic energy formulation;
• OMEGA 3.7 includes a spatially inhomogeneous ground surface characterization in its PBL scheme. OMEGA 2.0 assumed surface characteristics to be spatially uniform;
• OMEGA 3.7 contained additional data ingest routines to allow more flexibility in the input data formats. However, this is not relevant to the 2000 exercise, as the same observational data were used to run both OMEGA 2.0 and 3.7.

See Bacon et al., (2000) for OMEGA 3.7’s application to the Khamisiyah analysis.

3) MM5. The new MM5 reanalysis consisted of four six-day simulations using continuous data assimilation (see Table A-13). As with the spring 1997 reanalysis, NCAR performed these simulations with different options for the PBL parameterizations and global data for lateral boundary conditions and assimilation. In addition, a simulation was made with a variation in the surface roughness length. This simulation was motivated by the fact many mesoscale models historically have employed a default roughness length of about 10 cm for desert. This value may be appropriate for some southwestern US deserts with large brush and scrub trees, but surface roughness less than 1 cm probably is more appropriate for barren desert with scattered, limited vegetation (e.g., Oke, 1987). Thus, in Simulations 1, 2, and 3, the lower values of desert roughness length were employed to be consistent with the surface conditions of the Arabian desert (1 cm for vegetated desert and 0.5 cm for unvegetated desert). In Simulation 4, otherwise the same as Simulation 1, 10 cm was used.

Systematic performance evaluation shows Simulation 2 performed slightly better than the rest. To quantify model performance, the mean error (ME), the mean absolute error (MAE), and the root mean square error (RMSE) are calculated for predictions of wind speed, wind direction, temperature, and dew point temperature:

	(Equation A-34)
	(Equation A-35)
	(Equation A-36)

where M_i and O_i are the i^th pair of model prediction and observation, and N is the total number of pairs. Separate statistics were calculated for

• Grids 1, 2 and 3;
• Low-level (lower than 850mb, or roughly 1500 m) and upper-air levels (lower than 5000 m).

In addition, NCAR made the following improvements to MM5:

• The resolutions for land-use properties for the old and new runs were 18 and 1 km, respectively;
• The old runs covered 40 hours of simulation time, from 0000 UTC March 10 to 1600 UTC March 11, 1991. The new runs covered 144 hours, from 1200 UTC March 9 to 1200 UTC March 15, 1991;
• The first-guess fields and lateral boundary conditions used in the 1997 model runs were based on either the NCEP AVN model forecasts or the global reanalysis ECMWF specifically produced for the Tropical Ocean Global Atmosphere (TOGA) project. The 2000 model runs used either the NCEP global reanalysis or TOGA/ECMWF reanalysis; and
• All 2000 model runs used now-declassified observations, many of which were still classified in 1997, for model initialization and data assimilation.

Warner and Sheu (2000) provide further details of how NCAR applied MM5 to the Khamisiyah analysis.

Figures A-61 through A-72 show the surface wind vectors the three mesoscale models produced for the 2000 Khamisiyah analysis every 24 hours for 1200 UTC March 10, 1991, through 1200 UTC March 13, 1991. Wind vectors are plotted only at every half degree for legibility, although the model resolution is much higher. Moreover, the surface wind fields refer to 10, 40, and 30 m above the ground for COAMPS, MM5, and OMEGA, respectively.

At the time of the release (near 1200 UTC March 10, 1991), all three models produce near-surface winds out of the northwest over Khamisiyah (Figures A-61, A-62, and A-63). Furthermore, these figures also show the sea breeze effect that drives wind flow counter to the prevailing circulation over the Persian Gulf’s northern coast. Consistent with the earlier predictions, the low-level flow south of Khamisiyah shifts from the northwest to the north-northeast over the next 24 hours (i.e., 1200 UTC March 10, 1991, through 1200 UTC March 11, 1991). Figures A-64, A-65, and A-66, reflect this shift, which is consistent with the evolving large-scale conditions. Comparing the individual plots, we see the near-surface flow fields are fairly consistent with the plume trajectories generated in 1997, and individual model differences are generally limited to locations exterior to the hazard area footprint.

Figure A-61. 2000 MM5 Grid 3 predicted
wind fields for 1200 UTC March 10, 1991

Figure A-62. 2000 COAMPS Grid 3 predicted
wind fields for 1200 UTC March 10, 1991

Figure A-63. 2000 OMEGA Grid 3 predicted
wind fields for 1200 UTC March 10, 1991

Figure A-64. 2000 MM5 Grid 3 predicted
wind fields for 1200 UTC March 11, 1991

Figure A-65. 2000 COAMPS Grid 3 predicted
wind fields for 1200 UTC March 11, 1991

Figure A-66. 2000 OMEGA Grid 3 predicted
wind fields for 1200 UTC March 11, 1991

Between 1200 UTC March 11, 1991, and 1200 UTC March 12, 1991, the low-level pressure gradient over the Khamisiyah region intensified, shifting the winds more easterly (Figures A-67, A-68, and A-69). Over the next 24 hours (1200 UTC March 12, 1991, through 1200 UTC March 13, 1991), the models show weak cyclonic circulation driven by the surface low pressure over the area (Figures A-70, A-71, and A-72).

Figure A-67. 2000 MM5 Grid 3 predicted
wind fields for 1200 UTC March 12, 1991

Figure A-68. 2000 COAMPS Grid 3 predicted
wind fields for 1200 UTC March 12, 1991

Figure A-69. 2000 OMEGA Grid 3 predicted
wind fields for 1200 UTC March 12, 1991

Figure A-70. 2000 MM5 Grid 3 predicted
wind fields for 1200 UTC March 13, 1991

Figure A-71. 2000 COAMPS Grid 3 predicted
wind fields for 1200 UTC March 13, 1991

Figure A-72. 2000 OMEGA Grid 3 predicted
wind fields for 1200 UTC March 13, 1991