DSHW-2006-007254 - 0901a06880161bef

Paul Hancock, Manager
Environmental Remediation
ATK Thiokol Inc. – Promontory
P.O. Box 707
Brigham City, UT 84302-0707

RE: Groundwater Flow and Contaminant Transport Model
ATK Thiokol – Promontory Facility, EPA ID #009081357

Dear Mr. Hancock:

The Division of Solid and Hazardous Waste (the Division) has completed its review of the Groundwater Flow and Contaminant Transport Model for the ATK Thiokol – Promontory (Thiokol) facility.
The model was constructed so that it could be used to assist with decision making on remedial actions and as a tool for assessing the risk posed by contaminated groundwater at the site.

Enclosed with this letter are our comments and questions on the model that was submitted. Most of the comments and questions were previously provided to Thiokol and EarthFax. However,
there are a number of new comments that apply primarily to the geologic framework for the model and the geophysical survey that was conducted. The model requires approval before it
may be used. Therefore, please provide a written response to each question and comment within 45 days after receipt of this letter. If you have any questions, please contact Jeff
Vandel at 538-9413.

Sincerely,

ORIGINAL DOCUMENT SIGNED BY DENNIS R. DOWNS ON 7/19/06
Dennis R. Downs, Executive Secretary
Utah Solid and Hazardous Waste Control Board

DRD\JV\tm

Enclosure

c: Lloyd C. Berentzen, M.B.A., Health Officer/Director, Bear River Health Dept.
Nancy Morlock, USEPA Region VIII

Division of Solid and Hazardous Waste
Comments on the ATK Thiokol – Promontory Facility
Groundwater Flow and Contaminant Transport Model
July 2006

Section 2.3, Model Geologic Framework: The text mentions that, based on well and outcrop data, four easily identifiable layers exist in the subsurface of the study area. Is the delineation
of layers three and four actually based on well and outcrop data, or is it a hypothetical boundary, with layer four being created to provide more stability during model runs (as stated
in Section 3.1, page 3-2)? Please clarify this and provide more detailed information to support the delineation of layers three (fractured bedrock) and four (unfractured bedrock).

Geologic cross-sections prepared by Miller et al. (Utah Geological Survey Map 136, 1991) show that the low-angle normal fault, discussed in the report, and a thick package of Manning
Canyon Shale underlie the study area approximately at the elevation of layer four beneath the burn grounds. This interpretation of the geology of the area appears to be supported by
some well logs. What impact would the presence of these features have on the designation of layers three and four?

What were the criteria used for the delineation of the unconsolidated fine-grained (layer one) and unconsolidated coarse-grained (layer two) sediments?

Section 2.3, Model Geologic Framework: The text acknowledges the existence of a perched aquifer at the eastern edge of the facility. However, it appears that this perched aquifer has
not been modeled directly (even though the area of the perched zone is enclosed in the model grid, as shown on Plate 3-1, or is it not?). Please elaborate. For clarity, we suggest
to post all groundwater monitoring wells used for the calibration of the model onto Plate 3-1. Can constant head cells with specific head values in that eastern area accurately reflect
the influence of the perched aquifer on the regional aquifer, especially since fluxes aren’t known (and aren’t even demonstrated to exist)?

3) Section 2.4.2, Faults Identified by Geophysical Survey: The results of the VLF survey are difficult to evaluate based on the information provided in the report. It would be helpful
to provide more discussion on the interpretation of the VLF data. Are anomalies consistent from one profile to the next? This is difficult to determine since the profile displays can’t
be correlated to the traverses plotted on the geologic map. The scale presented on the profiles does not correlate to the map traverses. For example, Profile 0000N in the South Test
Area is about 300 meters (985 feet) long, but on the map it appears that the traverse is about 1,500 feet long. In addition, the directions of the profiles are plotted inconsistently.
The locations of profiles 0000E and 0210W in the South Test Area are not plotted on the map.

Please present the data on the profiles and map so that the locations of anomalies can be determined. It would also be helpful to label the anomalies on the profiles that have been
interpreted to be faults or fracture zones and label distances on the map traverses that correspond to the profiles.

Was only one filter depth applied to the data? The case for interpreting geologic structures from VLF data can be strengthened by applying additional filter depths to see if anomalies
are consistent with depth. How was a depth of 60 feet selected?

Can any indication of geologic structure geometry be interpreted from the VLF data?

VLF sources are most beneficial when aligned with the strike of the structural feature. The sources used for this project are not located along the projected strike of the north/south
structures. How does this impact the data and do corrections need to be made?

A description of on-site instrument calibration methods and the instrument configuration should be included in this section of the report.

Section 2.5, Study Area Hydrology: The text states that water levels do not exhibit significant seasonal variations; however, they do reflect changes according to general climatic conditions.
Exactly how was this statement derived (e.g., was a statistic analysis for seasonality or trend conducted)? Please clarify.

5) Section 2.6, Groundwater Quality: In the fourth paragraph, page 2-10, it is stated that “it appears that contaminants reach these locations via flow along high-angle normal faults
and the low-angle thrust fault that passes through areas of contamination northwest of the wells.” This statement is confusing as it is the low-angle normal fault that likely impacts
the areas of contamination northwest of wells G-6 and G-8, not the thrust fault. Please clarify.

Section 3.1, Hydrogeologic Setting: The text states that two long-term pumping tests were conducted at the site. How many observation wells were used in each case to determine presence
or absence of anisotropy? Please identify all wells (pumping and observation wells), and state the methodology used for determining anisotropy. Because anisotropy has apparently been
detected in monitoring well T-2, was this information used for starting values regarding transverse dispersivities in that area?

Section 3.1, Hydrogeologic Setting: The text mentions that spring flows were used to determine an approximation for groundwater sinks in the model. However, although that data is
listed in Section 3.5.4, the text should be more specific as to how the data was determined (e.g., by having calibrated flumes, buckets, etc.).

Section 3.5, Parameter Zonation: The text states that initial input parameters were modified within reason, using an iterative (and presumably subjective) process. Was the PEST add-on
package (integrated with Version 4.0) of Visual Modflow used as well, in order to gain a more systematic understanding of input parameter sensitivities? From a perusal of Section 3.6.1,
it appears that this wasn’t the case, but please confirm.

Section 3.5.1, Specific Storage and Specific Yield: The text states that the specific storage coefficient (for the confined portion of the aquifer), as well as the specific yield (for
the unconfined portion of the aquifer), were estimated based on literature values. However, pumping data does exist, as displayed in Table 3-1, and could have been used to numerically
estimate specific storage. Why was that data apparently used only for hydraulic conductivity estimation?

Section 3.5.3, Recharge: The text states that three inches/year was used to simulate recharge into the uppermost active layer of the model. Was this value also double-checked with
any pertinent, empirical rainfall-infiltration models (e.g., Horton, Holtan, etc.)?

Section 3.6, Groundwater Model Flow Calibration: The text states that the model was not calibrated to any transient conditions, and we agree that the calibration to assumed steady-state
conditions (namely, the potentiometric surface 2003/2004) was sufficient. However, how exactly were the initial water levels (as listed in Table 3-2) derived (e.g., are they averages
of two separate sampling events)? In addition, if, for example, groundwater levels from monitoring well TCC6 were available and used for modeling purposes, why weren’t head isocontours
extended to that area (see Plate 2-2)? Please clarify.

Section 3.6.2, Calibration Targets: The text states that the two statistical goodness-of-fit models used (namely, mean absolute error (MAE), and normalized root mean squared error (NRMS)),
are represented in Anderson and Woessner (Applied Groundwater Modeling, 1992). However, we were only able to verify the use of the MAE, and a non-normalized version of the NRMS, the
root mean squared error (RMS), in Chapter 8.4 of that book. We understand that the only difference between RMS and NRMS is the added division by the term (maxhead-minhead). Why is
this term necessary, and does it apply to the actual (observed) or modeled heads? And, strictly speaking, since the total NRMS value is reported in percent, should (maxhead-minhead)
be written as an absolute value, in order to avoid negative percentages?

Section 3.6.3, Calibration Results: Please specify which numerical solver was used in order to achieve convergence (e.g., WHS, PCG, SIP, SOR, etc.). Furthermore, it is not clear
if during the actual modeling run(s), any re-wetting was necessary for cells which might have gone dry. Were any oscillations or numerical instability, which, if present, could shed
some light on inadequate grid or nodal spacing, observed in any aborted, or even the final solution run? Please elaborate in more detail.

Section 4.2, Transport Parameters: The text states that the acceptable range for longitudinal dispersivity is 1 to 50 ft, according to Zheng and Bennett (Applied Contaminant Transport
Modeling, 1995). However, we were unable to verify that claim. For example, in Section 9.3 of that book, reliable ranges for longitudinal dispersivity appear to congregate around 100m
(330 ft). Also, is the bulk density used (48.13 kg/ft^3) held constant over the entire model domain? Please clarify.

Section 4.3, Sources and Sinks: The text states that the area east of the Burning Grounds (the perched aquifer) was not included for the transport simulation because cells, modeled
as constant concentration cells, actually removed contaminant mass. Was there no other way of circumventing this problem (e.g., by assigning other cell types at the boundary in question

with assigned fluxes (Neumann or Cauchy-type conditions))? Also, with a TCE concentration above 2,500 pbb in monitoring well J-4, has the contaminant plume even been delineated? Please
elaborate.

Section 4.3, Sources and Sinks: The text states that volatilization is likely not occurring (due to the depth to groundwater). Is this statement also true in the alluvial, western
model domain (layer 2), where groundwater is generally shallower, and influences from Blue Creek can be observed (based on the hydraulic head, isopotential patterns along Blue Creek
on Plate 2-2)?

Section 4.5, Grid: The text states that the flow model grid spacing (200 ft x 200 ft) remained the same for the transport model. Usually, in order to achieve better numerical stability
(minimize numerical dispersion), the grid is refined for the transport model. Since this was apparently not necessary, what exactly was the grid Peclet number? Also, regarding the
transport model’s temporal discretization, the transport step Courant number should be listed (perhaps in Section 4.6.1).

Section 4.8.2, Results: The text mentions a mass balance error of 0.01% for perchlorate (presumably indicating that more mass enters the system than leaves the system). Does this really
make sense? Please explain. Also, on Plate 4-2, there appears to be an error in the legend, since both green and red isocontours predict the TCE extent in layer 2. Please correct.

Section 5.2, Sensitivity Analysis Procedure: The text states that sensitivity analyses were not performed on the TCA/DCE and perchlorate models, since the TCE model’s sensitivity analysis
was deemed representative. Please explain this in more detail, particularly since the TCA/DCE model run appears to have used the reactive transport code RT3D. Generally speaking, it
is good, common modeling practice to conduct sensitivity analyses for all relevant transport scenarios.

Section 5.4, Contaminant Transport Model Sensitivity: On page 5-4, the term “activity coefficient” should be explained. We assume you mean the fraction of organic carbon, but this
is not clear to the casual reader.

General comment: Statistical, goodness-of-fit figures, appear to be missing. These figures, readily available from Visual Modflow, are patterned after Figures 8.9 and 8.13a/b in Anderson/Woessner,
1992. While all relevant simulation output data is presented in the Tables of this report (organized by wells), a graphical presentation would be helpful, as the reader could assess
computed vs. observed heads over the entire model domain at one glance in one figure, or assess head residuals spatially on one map.

1

July 19, 2006

TN200600806.doc
288 North 1460 West • PO Box 144880 • Salt Lake City, UT 84114-4880 • phone [phone] • fax [phone]
T.D.D. [phone] • www.deq.utah.gov

State of Utah

Department of Environmental Quality

Dianne R. Nielson, Ph.D.
Executive Director

DIVISION OF SOLID AND HAZARDOUS WASTE
Dennis R. Downs
Director

JON M. HUNTSMAN, JR.
Governor

GARY HERBERT
Lieutenant Governor