DEQ Comment Response on STATS TM 01202017

1/20/2017

1
The following informal comments regarding the draft Technical Memorandum, Statistical Methods for
Developing Reference Background Concentrations for Groundwater at Coal Ash Facilities, HDR, October
2016 were provided by DEQ through email from Steve Lanter dated 11/22/2016. Responses prepared
by Duke Energy consultants HDR and SynTerra are noted below each comment.

DEQ Comment #1
Ensure that HDR and SynTerra use the exact same technical approach to determine preliminary
provisional background threshold values (PPBTVs) for groundwater.

Duke Response #1
HDR and SynTerra will use the same technical approach to determine PPBTVs for groundwater. The
draft version of the TM Statistical Methods for Developing Reference Background Concentrations for
Groundwater at Coal Ash Facilities was developed jointly by HDR and SynTerra and the revised
version was jointly developed by HDR and SynTerra but was developed to be a guidance document
issued by Duke to be used to develop PPBTVs for groundwater and soil (Statistical Methods for
Developing Reference Background Concentrations for Groundwater and Soil at Coal Ash Facilities,
January 2017- DRAFT). Although the same technical approach will be used by both HDR and
SynTerra, due to site differences, the methodology was developed to provide for technical flexibility
if needed due to site differences.

DEQ Comment #2
Low probability, extreme upper tail outliers should not be used to develop reference background
concentrations. As ProUCL-V5.0 Technical Guide (page 17, paragraph 1) indicates, since the
presence of outliers in a data set tends to yield distorted values of decision making statistics,
elevated outliers should not be included in background data sets and estimation of background
threshold values. The objective here is to compute background statistics based upon the majority of
the data set representing the main dominant background population and not to accommodate a
few low probability high outliers that may also be present in the background data set.
Duke Response #2
Part II of the methodology describes the process that will be used to identify outliers for
groundwater results. While there are several tests available to test for possible outliers, Dixon’s or
Rosner’s Outlier tests are specifically identified in the Unified Guidance (USEPA 2009) for providing
requisite statistical strength and power necessary to appropriately identify potential outliers.
Additionally, visual methods such as Q-Q plots and box plots will be utilized in the identification of
outliers.

Statistical outliers will be tabulated as part of the background determination report. If statistical
outliers have been detected, the project scientist will review the outliers to determine if the values
should be removed or if they are representative of background and should be retained for statistical
analysis. Reasons as to why a particular statistical outlier should be included or excluded from the
background dataset will be documented as part of the final background determination report.

1/20/2017

2

DEQ Comment #3
Duke should adopt and use the protocols (“rules”) that were internally agreed upon by Division staff
(no samples with pH > 8.5, no samples with turbidity > 10 NTUs, and no samples with detection or
reporting limits that are higher the applicable standard).
Duke Response #3
Unless site specific conditions exist that dictate exceptions, Duke will use the following protocol in
determining if groundwater sample results are to be considered valid for determining PPBTVs:
• Sample pH is less than or equal to 8.5 standard units (S.U.);
• Sample turbidity is less than or equal to 10.0 Nephelometric Turbidity Units (NTUs); and
• Result is not a statistical outlier identified for background sample data.

Professional judgment can be used to retain data that does not meet these criteria (such as pH);
however, the decision to retain data that does not satisfy these criteria must be documented.

Should sample results with detection or reporting limits greater than the applicable standard impact
the data pool such that sufficient data is not available for producing PPBTVs, NCDEQ will be
consulted to discuss alternative evaluation options for the particular constituents, such as modeling.

DEQ Comment #4
Because it is difficult to know how many comparisons are going to be made to background (that is, is
difficult to know the number of down- and side-gradient wells and sample events that will be
compared to the computed background concentration), it is recommended that prediction limits not
be used since that knowledge is needed to calculate the upper prediction limit. Instead, it is
recommended that a 95-95 UTL be used. The log normal distribution should not be used to
compute the UTL unless the data are not heavily skewed and the standard deviation is <=
1. Preference should be given to a gamma distribution over that of the log normal distribution as
discussed on pages 22-23.
Duke Response #4
As described in Part IV of the methodology, the upper tolerance limit (UTL) will be used to develop
the PPBTVs. As described in the ProUCL guidance document (Section 3.4), the UTL95-95 is designed
to provide coverage for 95 percent of all observations from the background population with a
confidence coefficient (CC) of 0.95. A UTL95-95 will be exceeded by all (current and future) values
coming from the background population less than 5 percent of the time with a CC of 0.95, that is 5
exceedances per 100 comparisons (of background values) can result just by chance for an overall CC
of 0.95. Unlike a UPL95, a UTL95-95 can be used when many, or unknown number of current or
future onsite observations need to be compared with a BTV. The actual confidence coefficient for a
given sample population is determined by the sample size and the order statistic. For sample sizes
as low as 11 samples, the coverage limit may be 85 percent while maintaining a confidence limit of
95 percent. Part IV of the methodology describes the general approach to be utilized when samples
sizes are not sufficient to calculate UTL95-95 values. For constituents with insufficient data (i.e., <8
samples), continued sampling of background wells will be performed until sufficient sample size is
achieved.

1/20/2017

3
As described on the referenced pages, preference will be given to the gamma distribution over the
log normal distribution, with acknowledgement that in the end, the goal is to compute the correct
statistics of practical merit representing the data set. As additional data is collected over time, the
background data set distribution and UTL will be re-evaluated.

DEQ Comment #5
If the UPL method is employed, many comparisons may need to be made in order to make a
determination of PPBTVs. Since PPBTVs will be determined with limited data sets, this approach
may be difficult to defend.
Duke Response #5
As noted in Duke Response #4, PPBTVs will be calculated using UTLs.

DEQ Comment #6
As a UPL based upon the Chebyshev Inequality tends to yield higher estimates of BTVs than other
methods, HDR proposed to use Chebyshev’s non-parametric estimate at a lower confidence
coefficient, 85% or 90% (page 23). If the UPL approach is used, please specify when 85% is used and
when 90% is considered. Otherwise, to be conservative, 85% should be considered.
Duke Response #6
As noted in Duke Response #4, PPBTVs will be calculated using UTLs.

DEQ Comment #7
Page 2. “If the concentration in the future sample is higher than the estimated UPL95, then steps
are taken to validate the observation by means of quality assurance checks and/or by re-sampling
and/or to investigate for possible natural or anthropogenic reasons for observing a higher
concentration than expected.” This implies that we are in a “detection monitoring” stage, which we
are not. We are in an “assessment monitoring” stage in which down- and side-gradient wells are
compared to 2L or background to map the extent of contaminated groundwater.
Duke Response #7
The methodology was developed to provide guidance on statistical methods to use used in
calculating PPBTVs, therefore the referenced text describing re-sampling was removed.

DEQ Comment #8
Pooling of background data should not be done across flow systems. That is, shallow, deep, and
bedrock wells should not be pooled together for purposes of background determination.
Duke Response #8
Duke will not pool background data across groundwater flow layers. Separate PPBTVs will be
calculated for the site flow layers (i.e., shallow, transition zone, deep or bedrock, or others as
appropriate by site). There may be specific sites where geologic and geochemical conditions vary by
depth within a flow layer and these conditions may warrant the calculation of separate PPBTVs
within a particular flow layer.

1/20/2017

4
For example, at Cape Fear, background well results for the shallow (surficial) and bedrock flow zones
show geochemical differences within the flow zones. In the surficial zone at Cape Fear, differences
are observed in groundwater geochemistry between the upper and lower surficial zone. Differences
in background groundwater geochemistry are also observed in the bedrock layer at Cape Fear
between wells screened in sandstone and those screened in mudstone. For those cases, separate
PPBTVs may be calculated for separate conditions within those flow layers.

DEQ Comment #9
Page 3. “If distinct sub-groups exist as which may be the case across different hydrostratigraphic
units, HDR and SynTerra will develop separate background concentrations for each distinct sub-
group.” Since background determinations are needed to continue and complete assessments and
corrective action plans, sufficient quantities of data for these sub-group tests generally will not be
available. Sub-group dataset sizes of generally 8 to 10 are needed to conduct the sub-group tests
which include the testing of seasonal effects, etc. Note that on page 15 HDR recommends that
sample sizes per sub-group should be at least 30. In many or most cases this would result in
significant delays in the background determinations that are not feasible given CAMA
deadlines. The Division reserves the right to re-visit sub-group testing once adequate data are
available.
Duke Response #9
Sub-group analysis will be performed as needed, where sufficient datasets of at least eight samples
are available for each potential sub-group .

DEQ Comment #10
Under no circumstances should a maximum detection limit be used to establish a background value.
Duke Response #10
PPBTVs will be based on UTLs and not based on maximum detection values.

DEQ Comment #11
In the meeting, it was agreed that Duke Energy would email the Regional Hydrogeologists a
spreadsheet for each facility that contains all background well data and the actual culled data (see
comment 2 and 3 above) that they plan to use for background determinations. The Regional
Hydrogeologists would then review the dataset to ensure that the Division and Duke Energy are
using the same data for the background determinations.
Duke Response #11
Duke has provided comprehensive summary spreadsheets of the background groundwater data set
by flow layer to DEQ.