REPORT OF THE HUDSON RIVER PCBs SITE
MODELING APPROACH PEER REVIEW
—Final Report—
Preparedfor:
U.S. Environmental Protection Agency, Region II
Emergency and Remedial Response Division
290 Broadway, 18th Floor
New York City, NY 10007-1866
EPA Contract No. 68-W6-0022
Work Assignment No. 3-12
Prepared by:
Eastern Research Group, Inc.
110 Hartwell Avenue
Lexington, MA 02421
November 10,1998

-------
REPORT OF THE HUDSON RIVER PCBs SITE
MODELING APPROACH PEER REVIEW
—Final Report—
Preparedfor:
U.S. Environmental Protection Agency, Region II
Emergency and Remedial Response Division
290 Broadway, 18th Floor
New York City, NY 10007-1866
EPA Contract No. 68-W6-0022
Work Assignment No. 3-12
Prepared by:
Eastern Research Group, Inc.
110 Hartwell Avenue
Lexington, MA 02421
November 10, 1998

-------
NOTE
This report was prepared by Eastern Research Group, Inc. (ERG), an EPA contractor, as
a general record of discussion for the peer review meeting. This report captures the main points
of scheduled presentations and highlights discussions among the reviewers. This report does not
contain a verbatim transcript of all issues discussed during the peer review. Additionally, the
report does not intend to embellish, interpret, or enlarge upon matters that were incomplete or
unclear. EPA will evaluate the recommendations developed by the reviewers and determine what,
if any, modifications are necessary to the current modeling approach. Except as specifically
noted, no statements in this report represent analyses or positions of EPA or of ERG.

-------
TABLE OF CONTENTS
LIST OF ABBREVIATIONS	iii
EXECUTIVE SUMMARY	 v
1.0 INTRODUCTION	1-1
1.1	Background 	1-1
1.2	Scope of the Peer Review	1-2
1.3	Meeting Organization and Agenda 	1-4
1.4	Summary of Opening Remarks at the Meeting 	1-4
1.5	Report Organization	1-5
2.0 DISCUSSION ON QUESTION A: On the Appropriateness of Using the
Models, Datasets, and Assumptions to Make Scientifically Credible
Decisions	2-1
3.0 DISCUSSION ON QUESTION B: On the Ability of the Models to Answer the
Principal Study Questions of the Preliminary Model Calibration
Report 	3-1
3.1	Predicting Future Levels of PCBs in Fish 	3-1
3.2	Evaluating Remedies Other Than "No Action"	3-4
3.3	Characterizing the Fate of "Reactivated" Sediments Following Maj or
Floods	3-5
4.0 DISCUSSION ON QUESTION C: Specific Questions Regarding the Models . 4-1
4.1	Developing Quantitative Relationships Between Forcing Functions and PCB
Concentrations	4-1
4.2	Adequacy of Spatial and Temporal Scales	4-2
4.3	Use of Several Bioaccumulation Models	4-4
4 .4 Adequacy of the Models' "Level of Process Resolution"	4-5
4.5 Usefulness of Modeling Results for Risk Assessment 	4-7
5.0 DISCUSSION ON QUESTION D AND E: Recommended Changes to, and Serious
Flaws in, the Proposed Modeling Approach	5-1
i

-------
TABLE OF CONTENTS (Continued)
6.0 REVIEWERS' OVERALL RECOMMENDATIONS 	6-1
6.1	Discussions That Led to Identifying Major Recommendations	6-1
6.2	Peer Reviewers' Major Recommendations 	6-4
6.3	Peer Reviewers' Final Statements	6-5
7.0 REFERENCES 	7-1
APPENDIX A
List of Expert Peer Reviewers
APPENDIX B
Charge to Expert Peer Reviewers
APPENDIX C
Premeeting Comments, Alphabetized by Author
APPENDIX D
List of Registered Observers of the Peer Review Meeting
APPENDIX E
Agenda for the Peer Review Meeting
APPENDIX F
Summaries of Observers' Comments
ii

-------
LIST OF ABBREVIATIONS
DOC
dissolved organic carbon
EMC
environmental management council
EPA
U.S. Environmental Protection Agency
ERG
Eastern Research Group, Inc.
GE
General Electric Company
PCB
polychlorinated biphenyl
PMCR
"Preliminary Model Calibration Report"
QEA
Quantitative Environmental Analysis, LLC
TEQ
toxic equivalent
TSS
total suspended solids
iii

-------
EXECUTIVE SUMMARY
Seven independent peer reviewers critiqued the October 1996 "Preliminary Model
Calibration Report" (PMCR), a document prepared as part of the U.S. Environmental Protection
Agency's (EPA's) reassessment of the Hudson River PCBs Superfund site. After discussing at
length the scientific rigor of the proposed modeling approach, most of the reviewers commended
EPA on its extensive modeling efforts, but they unanimously agreed that the modeling approach
described in the PMCR was "acceptable with major revision." At the close of the peer review
meeting, the reviewers developed a short list of ma jor findings and recommendations, which are
summarized below. Except as noted otherwise, all of the peer reviewers agreed with these major
findings and recommendations. Specific examples of other suggested revisions and
recommendations made by the peer reviewers can be found throughout this report.
•	The reviewers recommended that EPA make the following improvements in the
description of sediment resuspension and deposition processes in the fate and transport
models: address the fate of resuspended material; address the role of uncovered,
potentially contaminated surfaces; address the issue of non-cohesive sediment
resuspension; assure consistency in resuspension rates between the TIP and HUDTOX
models; and identify the effect of flood resuspension on the rate of long-term recovery of
the Hudson River. Some reviewers indicated that these changes could be made within the
existing modeling framework (i.e., with several different fate and transport models that are
linked), while other reviewers thought these changes should be made by incorporating
sediment transport mechanisms directly into the HUDTOX model, instead of keeping the
models separate.
•	The reviewers recommended that EPA employ time- and space-dependent mechanistic
models that reflect the abiotic and biotic dynamics of the Hudson River system.
The reviewers indicated that the models should consider bioavailability and sediment
sequestration with respect to congener sorption/desorption and transformation kinetics,
sediment particle characteristics, and biotic characteristics.
The reviewers recommended that EPA link, to the greatest extent possible, the spatial and
temporal scales of the different fate and transport and bioaccumulation models.
•	The reviewers recommended that EPA clearly identify in its modeling approach the risk
assessment targets as related to forms and concentrations of PCBs (e.g., to what
v

-------
guidelines or advisories will the modeled concentrations be compared). More specifically,
the reviewers thought risk assessors and managers should be involved with the
development of the transport and fate and bioaccumulation models to ensure that the
model outputs will generate the data needed for completing human health and ecological
risk assessments.
The reviewers recommended that EPA develop a mechanistic food web model based on
exposure dynamics of the identified forms of PCBs relevant to risk quantification, and that
EPA identify appropriate data needs for the fete and transport models.
The reviewers recommended that EPA develop an explicit plan for model calibration and
independent validation that includes criteria for validation, makes use of numerous data
sets that span a long time period, and includes chemicals in addition to PCBs.
Noting that analytical methods have improved and will continue to improve, the reviewers
recommended that EPA develop, and agree on how to use, a method for interpreting
historical PCB monitoring data, including data quality factors.
Noting that EPA has already addressed or considered many of the recommendations listed
above, the reviewers suggested that EPA hold an open workshop (involving all interested
parties as well as independent reviewers) to evaluate current modeling efforts.
vi

-------
1.0	INTRODUCTION
This report summarizes a peer review by seven experts of the site modeling approach that
the U.S. Environmental Protection Agency (EPA) proposed for reassessing the Hudson River
PCBs Superfund site. The peer reviewers addressed information provided in three documents:
The October 1996 review copy of the "Preliminary Model Calibration Report" (PMCR)
(Limno-Tech et al., 1996)
The July 1998 draft copy of the revised scope of work for baseline modeling (Limno-Tech
et al., 1998)
Written responses to selected comments that stakeholders made following the initial
release of the PMCR
The seven reviewers participated in the peer review meeting, which took place in Saratoga
Springs, New York, on September 9-10, 1998. Eastern Research Group, Inc. (ERG), a
contractor to EPA, organized the expert peer review and prepared this summary report. This
introductory section provides background information on several topics relevant to this report,
including a brief background of the Hudson River PCBs site, the scope of the current peer review,
and the organization of this report.
1.1	Background
In 1983, EPA classified approximately 200 miles of the Hudson River in the state of New
York as a Superfund site, due to elevated concentrations of polychlorinated biphenyls (PCBs) in
sediments. The sediments are believed to have been contaminated by discharges of PCBs over
approximately 30 years from two General Electric (GE) capacitor manufacturing plants, one in
Hudson Falls and the other in Fort Edward. The Superfund site runs from Hudson Falls to the
Battery in New York City. After an initial site assessment, EPA issued an "interim No Action
decision" in 1984 for the contaminated sediments at the Hudson River PCBs site.
1-1

-------
Since 1990, EPA has been reassessing its "interim No Action decision" to determine
whether the PCB contamination in the Hudson River necessitates a different course of action.
EPA proposed to complete this reassessment by compiling and analyzing existing data ("Phase
I"), collecting additional data and using models to conduct human health and ecological risk
assessments ("Phase IT'), and performing a feasibility study of selected remedial alternatives
("Phase HF'). As part of "Phase IT' of the site reassessment, EPA's contractors completed the
PMCR, which is the subject of the current peer review. The purpose of the PMCR was to
describe models that EPA will use to characterize the fate and transport of PCBs in sediments,
water, and biota in the Hudson River. More specifically, the PMCR was prepared "to provide
interested parties with information about the data and assumptions that are being used in the
models, prior to completion of the actual modeling work" (Limno-Tech et al ., 1996). Several
parties provided comments on the PMCR during the report's public comment period in 1996.
To ensure that the assumptions and preliminary findings presented in the PMCR are based
on sound scientific principles, EPA decided as per policy to obtain an expert peer review of the
document. The remainder of this report describes the scope and findings of the peer review of the
PMCR
1.2 Scope of the Peer Review
To organize a thorough review of the PMCR, ERG selected seven independent peer
reviewers who are engineers or senior scientists with demonstrated expertise in any combination
of the following areas:
Transport and fate models for sediments and the water column
•	Fish body burden models
•	Calibration and validation of models
•	Sensitivity analysis of models
•	Familiarity with PCBs or other compounds that bioaccumulate
1-2

-------
To ensure the peer review's independence, ERG considered only individuals who could
provide an objective and fair critique of EPA's work. ERG did not consider in the reviewer
selection process individuals who were associated in any way with preparing the PMCR or
individuals affiliated with GE or any other specifically identified stakeholder.
Appendix A lists the seven reviewers ERG selected for the peer review meeting. Brief
summaries of each reviewer's area of expertise can be found in Appendix C. Recognizing that
few individuals specialize in every topic area listed above, ERG ensured that the collective
expertise of the selected peer reviewers covers the topic areas listed above (i.e., at least one
reviewer has expertise in transport and fate models, at least one reviewer has expertise in fish
body burden models, and so on).
To focus peer reviewer comments, ERG worked with EPA to develop written guidelines
for the technical review. These guidelines (commonly called a "charge") asked reviewers to
address at least the following topics: if the proposed models could be used to make scientifically
credible decisions; whether EPA's proposed models, datasets, and assumptions could answer the
principal study questions of the PMCR; and if the modeling approach had any serious flaws that
would invalidate its conclusions. A copy of the charge, which includes many additional topics and
questions, is included in this report as Appendix B.
Several weeks before the meeting, ERG distributed copies of the PMCR, the revised
scope of work for baseline modeling, and the responses to selected stakeholder comments to the
reviewers and asked them to prepare written responses to the charge questions, based on their
initial reviews of the documents. ERG compiled these "premeeting comments," distributed them
to the reviewers, and made copies available to observers during the peer review meeting. These
initial comments are included in this report, without modification, as Appendix C. It should be
noted that the premeeting comments are preliminary in nature and some reviewers' technical
findings may have changed based on discussions during the meeting. The premeeting comments
should not be considered as the reviewers' final opinions.
1-3

-------
It should also be noted that the reviewers were asked to base their reviews on the written
materials distributed by ERG: the PMCR, the revised scope of work, and the responses to public
comments. Even though EPA has currently completed several "Phase IF reports in addition to
the PMCR, the reviewers were not asked to consider these additional reports, which will be (or
already have been) subject to public comment and separate peer reviews. Though not required
for this peer review, some reviewers may also have researched site-specific reports they obtained
from other sources.
1.3	Meeting Organization and Agenda
The peer review meeting, which was held at the Sheraton Saratoga Springs Hotel and
Conference Center in Saratoga Springs, New York, on September 9-10, 1998, was attended by
the seven expert reviewers and at least 29 observers. Appendix D lists the observers who
confirmed their attendance at the meeting registration desk. The schedule of the peer review
meeting generally followed the agenda, presented here as Appendix E. As the agenda indicates,
the meeting began with introductory comments both by the designated facilitator and by the
designated chair of the peer review meeting. (Section 1.4 of this report summarizes these and
other introductory comments.) The rest of the meeting consisted of discussions that focused on
responding to the questions in the charge. During the technical discussions, the reviewers
provided many comments, observations, and recommendations. The agenda included two time
slots for observer comments, which are summarized in Appendix F of this report. An ERG writer
attended the meeting and prepared this summary report.
1.4	Summary of Opening Remarks at the Meeting
On the first day of the meeting, Jan Connery of ERG—the designated facilitator of the
review—welcomed the seven reviewers and the observers to the 2-day meeting. In her opening
remarks, Ms. Connery introduced Dr. A1 Maki (a peer reviewer and the technical chair of the
meeting), stated the purpose of the peer review meeting, and identified the documents under
review. To ensure that the peer review remained independent, Ms. Connery asked reviewers to
discuss technical issues among themselves during the meeting and to consult EPA only for
1-4

-------
necessary clarifications. Ms. Connery explained the procedure observers should follow to make
comments. Finally, she reviewed the meeting agenda, which appears in this document as
Appendix E.
Following Ms. Connery's opening remarks, the peer reviewers introduced themselves,
noted their affiliations, identified their areas of expertise, and stated that they had no conflicts of
interest in conducting the peer review. Selected representatives from EPA and from EPA's
contractors then introduced themselves and identified their roles in the site reassessment. To
orient the peer reviewers to EPA's ongoing site reassessment efforts, Mr. Doug Tomchuk (EPA)
then gave a presentation describing the history, current status, and planned future activities for the
Hudson River PCBs site. Mr. Tomchuk explained how the PMCR relates to EPA's overall site
reassessment, though he did not interpret, or expand on, the assumptions and findings
documented in the report.
As a transition into technical discussions, Dr. Maki reviewed the 11 questions in the
charge and identified several common themes among the peer reviewers' premeeting comments.
For the remainder of the meeting, the peer reviewers discussed the questions in the charge,
following the agenda. This report summarizes the peer reviewers' discussions and documents
their major findings and recommendations.
1.5 Report Organization
The structure of this report reflects the order of questions in the charge to the reviewers:
Section 2 of this report summarizes the reviewers' discussions on Question A in the charge,
Section 3 summarizes the discussions on Question B, and so on. Section 7 of this report lists all
references cited in the text.
As mentioned earlier, the appendices to this report include a list of the peer reviewers
(Appendix A), the charge to the reviewers (Appendix B), the premeeting comments organized by
1-5

-------
author (Appendix C), a list of the observers present at the meeting (Appendix D), the meeting
agenda (Appendix E), and summaries of the observer comments (Appendix F).
1-6

-------
2.0 DISCUSSION ON QUESTION A: On the Appropriateness of Using the Models,
Datasets, and Assumptions to Make Scientifically Credible Decisions
The peer reviewers opened their technical discussions by addressing the first question in
the charge: "Is EPA using appropriate models, datasets and assumptions on which to base a
scientifically credible decision?" To answer this question, each reviewer presented his or her
initial thoughts and comments, which the reviewers as a group then further discussed.
Summarizing the discussions, the technical chair suggested that the reviewers thought some of the
selected models were appropriate, but that the reviewers also thought EPA could achieve its
ultimate goals only after making major revisions. A general record of the peer reviewers'
discussion on this question, organized by topic, follows (but, it should be noted, that Sections 6.2
and 6.3 present the reviewers' final list of major recommendations):
•	Limitations posed by the question. When responding to Question A, several peer
reviewers commented that their responses may have been limited by ambiguities in the
question. For example, finding the question too broad, three reviewers indicated that they
would present only general comments in response to this question but would provide
detailed comments in response to other questions in the charge. Further, several reviewers
thought the question could be interpreted in many ways: What kinds of scientific
decisions will EPA make? What makes a decision "good enough" to be considered
scientifically credible? Will the decisions be used in legal proceedings or regulatory
determinations? To clarify the intent of the question, representatives from EPA explained
that they were particularly interested in whether the reviewers thought the models could
answer the three "principal study questions" of the PMCR Recognizing that Question B
of the charge addresses the principal study questions (see Section 3.0), several reviewers
again indicated that they would provide their detailed comments in response to Question B
and other questions that focus on more specific topics.
•	The proposed bioaccumulation models. Three reviewers recommended (and all of the
reviewers later agreed) that EPA should use predictive or dynamic models to address
bioaccumulation, instead of using the descriptive models proposed in the PMCR. More
specifically, the reviewers suggested that EPA use a more complex model based on a
mechanistic understanding of bioaccumulation, rather than use a simpler model based on
empirical relationships. The reviewers indicated that they would provide more detailed
comments on the bioaccumulation models when responding to other questions in the
charge.
2-1

-------
•	The proposed transport andfate models. The two reviewers who commented on the
proposed transport and fate models (in response to Question A) found them to be
generally appropriate. One reviewer provided detailed comments, noting that he approved
of the proposed grid sizes, model calibration, and selected steady-state assumptions.
Regarding improvements to the models, this reviewer recommended that the "depth of
scour" model for the Thompson Island Pool consider erosion of non-cohesive sediments,
that the models consider how turbulent flows affect erosion, and that EPA reconsider the
assumption that bed shear stresses reach their maximum values instantaneously.
•	Involvement of risk assessors. Several reviewers emphasized the importance of ensuring
that EPA's proposed modeling approach can meet the data needs of the human health and
ecological risk assessments planned for the Hudson River PCBs site. As an example, one
reviewer indicated that the modelers should know whether the risk assessors intend to
evaluate exposures to total PCBs, to a subset of Aroclors, or to individual congeners. To
demonstrate a clear link between the PMCR and the risk assessments, the reviewer
suggests the modeling reports should describe "acceptable risk levels" and other concepts
relevant to the planned risk assessments. Another reviewer wondered if the risk assessors
might evaluate environmental contaminants other than PCBs, thus necessitating modeling
for a greater number of contaminants.
•	Consideration of bioavailability. One peer reviewer indicated that the models should
more prominently acknowledge the issue of bioavailability, including kinetics of adsorption
and desorption, though the reviewers reserved specific comments on this topic for later in
the peer review meeting.
•	Links between the proposed models. Several reviewers did not think the PMCR
adequately addressed how EPA plans to interface the different transport and fete models
and bioaccumulation models. A reviewer noted that the modeling effort could face future
problems if the temporal and spatial scales of the various sub-models are not linked.
•	Consideration of other relevant data sources. A peer reviewer gave two examples of
how the modeling approach in the PMCR can be strengthened by considering data from
other relevant sources. First, the reviewer was surprised that the PMCR did not compare
any of the monitoring data or preliminary model predictions to data and predictions from
PCB modeling efforts in other river systems. Second, the reviewer thought EPA should
make use of data sets for pollutants other than PCBs. As an example, the reviewer
indicated that monitoring data for other chemicals (e.g., inorganics) could illustrate fete
and transport in the Hudson River over the long term. The reviewer acknowledged that
the relative importance of various fete and transport processes differ from one chemical to
the next, but noted that the same processes result in the transport of all chemicals. Thus,
the rigor of the model testing and validation could be greatly improved by considering
other chemicals.
2-2

-------
•	Use of historical monitoring data Several reviewers were concerned about the
appropriate use of historical monitoring data for the Hudson River. For instance, one
reviewer stressed that historical monitoring data for PCBs are particularly difficult to
interpret because analytical methods and sensitivities have changed significantly in the last
20 years—an issue the reviewer intended to describe in greater detail during later
discussions. According to another reviewer, the fact that many different agencies and
parties have collected Hudson River samples for different purposes may also complicate
efforts to understand historical data. This reviewer wondered whether some unique trends
in the historical data (e.g., relatively high concentrations of BZ4 in the distribution of
PCBs) reflect actual levels of environmental contamination or indicate evolving analytical
techniques.
•	Links between the aquatic, terrestrial, and atmospheric systems. One reviewer was
concerned because the proposed modeling approach apparently ignores possible physical
or biological links between the aquatic, atmospheric, and terrestrial systems. Two
reviewers cited examples of other systems in which they thought such links have been
addressed. The examples were PCB transport between the atmospheric and aquatic
systems for the Great Lakes and PCB transport between the aquatic and terrestrial
systems for the St. Lawrence River. The reviewers did not provide specific references for
these examples during the meeting.
•	Impact of uncertainty. One reviewer wondered if there might be enough uncertainty in
the proposed models to limit their use in making scientifically credible decisions. This
reviewer stressed that the uncertainty in the models' predictions does not necessarily imply
that the selected models, datasets, or assumptions are inappropriate or invalid. Rather, he
indicated that current "quantitative understanding" of the transport and fate and
bioaccumulation of PCBs in river systems may be too limited for models to predict future
river conditions with "sufficient certainty."
•	Inability to verify references to the "gray literature." One reviewer found references in
the PMCR to the "gray literature" troubling in two regards. First, the reviewer thought
EPA should cite peer-reviewed literature to the greatest extent possible. Second, the
reviewer was frustrated that the public generally cannot access much of the "gray
literature" that was cited. As a result, the reviewer suggested that EPA encourage wider
dissemination of the "gray literature," in cases where reference to such sources is truly
necessary. Another reviewer was frustrated that details of the Thomann model—a major
feature of the overall modeling effort—were not readily accessible and possibly not
complete.
•	Incomplete information on the available data. Several reviewers commented on different
topics related to the monitoring data available for the Hudson River. For example, some
reviewers were unsure of exactly how much monitoring data are available for the site and
exactly what subset of these data EPA plans to consider in its modeling analysis. As a
2-3

-------
result, one reviewer said he did not have enough information to determine whether EPA
was using appropriate data sets. Two reviewers thought that, to address concerns
regarding the use of available data, the PMCR should not only list the available data
sources but should also state the criteria used to include and exclude these sources from
the modeling analysis. On a related topic, one reviewer thought the extent to which EPA
considered data outliers was unclear; the reviewer recommended that outliers be
considered in the modeling analysis, because they may provide clues to unique trends and
patterns in the river system. Finally, one reviewer thought giving reviewers access to the
monitoring data might have helped the reviewers evaluate the PMCR more
comprehensively.
•	Assumptions in the modeling approach. One reviewer noted that the modeling approach
documented in the PMCR essentially assumes "ecological steady state" for the Hudson
River. The reviewer presented several scenarios (e.g., development of wetlands in the
area, invasion of the Hudson River by foreign species, and so on) that could invalidate this
assumption. When reviewing these discussions at the end of the peer review meeting,
however, the group of peer reviewers unanimously agreed that the modeling assumption
of "ecological steady state" was neither a "nuyor" or a "minor" flaw in the proposed
modeling approach.
•	Location of the upstream modeling boundary. In response to Question A, two reviewers
recommended that EPA consider moving the upstream boundary of its models to a
location upstream from Hudson Falls. The reviewers thought this change would not only
allow the models to provide a more complete mass balance of PCBs in the Hudson River
but would also allow the models to characterize "background" levels of PCBs at locations
upstream from the major sources. As Sections 5 and 6 describe, the peer reviewers'
recommendations on this topic changed during the meeting, and they eventually decided
that moving the upstream boundary may not be necessary
•	Insufficient time for the peer review. Noting that, in the time prior to the peer review
meeting, the reviewers had to (1) become familiar with the Hudson River system and
(2) research and critique the proposed models, one reviewer thought EPA should have
given the group of reviewers more time and more background information earlier in the
peer review process to critically evaluate the PMCR.
2-4

-------
3.0	DISCUSSION ON QUESTION B: On the Ability of the Models to Answer the
Principal Study Questions of the Preliminary Model Calibration Report
The peer reviewers continued their technical discussions by addressing Question B in the
charge: "Will the models, with the associated datasets and assumptions, be able to answer the
following principal study questions as stated in the PMCR:
1.	When will PCB levels in the fish population recover to levels meeting human
health and ecological risk criteria under No Action?
2.	Can remedies other than No Action significantly shorten the time required to
achieve acceptable risk levels?
3.	Are there contaminated sediments now buried and effectively sequestered from the
food chain which are likely to become 'reactivated' following a major flood,
resulting in an increase in contamination of the fish population?"
At the beginning of the discussions, the meeting facilitator (Jan Connery, ERG)
emphasized that Question B does not ask the reviewers to answer the three principal study
questions, but rather asks the reviewers whether the models will be able to answer the three
principal study questions. The following subsections summarize the reviewers' discussion on
Question B. Their comments are organized by the principal study question under discussion. It
should be noted, however, that Sections 6.2 and 6.3 present the reviewers' final list of major
recommendations.
3.1	Predicting Future Levels of PCBs in Fish
The reviewers made the following comments, observations, and recommendations
regarding Question B. 1 of the charge:
• Comments on the question. One reviewer noted that some aspects of Question B. 1
overlap with aspects of Questions B.2 and B.3. For example, the reviewer thought the
ability of the models to predict sediment resuspension during floods (Question B .3)
significantly affects the ability of the models to predict levels of PCBs in fish (Question
B. 1). As a result, the reviewers occasionally indicated when their specific responses
3-1

-------
applied to more than one question. As described later in this subsection, another reviewer
wondered exactly what "human health and ecological risk criteria" EPA will consider in its
risk assessments.
•	Improvements to the bioaccumulation models. Much of the reviewers' discussion focused
on the bioaccumulation models' inadequacy to predict future levels of PCBs in fish.
Consistent with the responses to Question A, several reviewers reiterated the importance
of using dynamic and mechanistic bioaccumulation models, instead of using empirical
models that have no predictive capacity. Elaborating on the inadequacies of the models,
one reviewer explained that the bivariate and probabilistic models provide only "pairwise
comparisons" of concentrations of PCBs in different media, without mechanistically
linking the different media. For instance, the proposed models assume that if the
concentrations of PCBs in the sediment and in the water column decrease by a factor of
two, then the concentrations of PCBs in the biota will also decrease by a factor of two.
The reviewer cited examples from Lake Ontario where such simple "pairwise
comparisons" are not observed in the food web. In short, the reviewer noted that
concentrations of PCBs in the different media—biota, sediment, and water—will change
at different rates, a scenario that the proposed bioaccumulation models cannot consider.
To improve the bioaccumulation models, the reviewers offered several recommendations.
First, several reviewers recommended that EPA use a bioaccumulation model, such as one
of the Gobas models, with mechanisms for predicting how concentrations of PCBs in biota
will respond to changing levels of PCBs in the river system. The reviewers indicated that
such mechanistic models can explore different pathways of exposure and uptake. Second,
another reviewer recommended that the bioaccumulation models link with the temporal
scales of the HUDTOX model and that these temporal scales be short enough to address
key life cycle stages in biota. Third, another reviewer thought the bioaccumulation models
should consider, to the extent possible, congener-specific rates for PCB metabolism,
uptake, and excretion. Finally, yet another reviewer recommended holding a separate
workshop to review the available bioaccumulation models before selecting the most
appropriate model for the Hudson River PCBs site.
•	Modeling the appropriate compounds for the risk assessment. One reviewer was
concerned that the PMCR did not specify the form of PCBs that will be considered in the
human health and ecological risk assessments. To illustrate his concern, the reviewer
wondered whether the risk assessments will consider total PCBs, a subset of Aroclors,
individual PCB congeners, or possibly a toxic equivalency (TEQ) analysis. Asserting that
the modeling outputs will eventually be critical inputs to the risk assessments, the reviewer
recommended that EPA ensure that the fete and transport and bioaccumulation models
evaluate the same forms of PCBs that the risk assessments will consider.
•	Models' consideration of bioavailability. Several reviewers suggested that the fete and
transport models should place a greater emphasis on bioavailability (in contrast to simple
3-2

-------
chemical detectability). As an example, one reviewer thought the models should not only
predict the amounts of sediments that are resuspended, but also should consider the
sorption and desorption kinetics of PCBs in these resuspended sediments (i.e., how much
is desorbed or becomes bioavailable as a result of resuspension). The reviewer
recommended using available congener-specific sorption and desorption kinetics, or
estimates of these parameters if they are not available, in the fate and transport models,
especially because some PCB congeners (higher-chlorinated PCBs) tend to bioaccumulate
in fish more than lower chlorinated PCB congeners. A reviewer also noted that those
congeners that sorb most to sediments also tend to bioaccumulate. Therefore, the models
need to address the bioavailability of PCBs sorbed to resuspended sediments.
•	Long-term validation of models. One reviewer acknowledged that long-term PCB fate
and transport modeling is a difficult, and often unsuccessful, endeavor. As a result, the
reviewer recommended that EPA perform long-term validation studies on levels of PCBs
in the fish and in the river, and possibly even on levels of other chemicals for which data
are available. The reviewer thought the results of such long-term validation studies might
best indicate whether the models will be able to answer the first principal study question of
the PMCR.
•	Benefits of conducting uncertainty and sensitivity analyses. One reviewer thought it
would be useful for EPA to develop a plan for conducting uncertainty analyses and
sensitivity analyses on the transport and fate and bioaccumulation models. The reviewer
suggested that these analyses may help EPA identify which variables and parameters have
the greatest impact on future river conditions.
•	Inconsistently predicted sediment resuspension rates. Citing a significant inconsistency
between the sediment resuspension rates predicted for flood events by HLJDTOX and the
rates for flood events predicted by the "depth of scour" model, one reviewer wondered if
the models can characterize sediment transport during flood events accurately. The
reviewer further noted that an inability to characterize this sediment transport would
ultimately compromise the models' ability to forecast concentrations of PCBs in fish. As a
result, the reviewer recommended that EPA explain why the HUDTOX and "depth of
scour" models seem to predict extremely different sediment resuspension rates during
floods.
•	Location of the upstream modeling boundary. One reviewer thought the ability of the
models to forecast future levels of PCBs in fish is limited by the primary source of PCB
loadings to the Hudson River being upstream of the modeling domain. More specifically,
the reviewer indicated that if EPA cannot predict how the source of PCBs in the Upper
Hudson River will change in the future, then the models certainly would not be able to
predict future river conditions. Upon request for clarification, EPA explained that
remedial actions at the GE Hudson Falls plant have greatly reduced the upstream loading
of PCBs into the HUDTOX modeling domain since 1993. Further, EPA noted that
3-3

-------
monitoring upstream of the Hudson Falls plant consistently fails to detect PCBs in the
water column (with a detection limit of 11 parts per trillion). As Sections 5 and 6
describe, the peer reviewers' recommendations on the placement of the upstream
boundary for modeling changed during the meeting, and they eventually decided that
moving the boundary may not be necessary.
3.2 Evaluating Remedies Other Than "No Action**
The reviewers made the following comments, observations, and recommendations
regarding Question B.2 of the charge:
•	Comments on the question. Because neither the PMCR nor the charge to the reviewers
specify what remedial options EPA is considering for the contaminated sediments, several
reviewers found it difficult to comment on whether the models can determine if remedies
other than "No Action" could accelerate the rate at which environmental concentrations of
PCBs might reach "acceptable risk levels." In addition, one reviewer was uncomfortable
answering the question without knowing exactly what the "acceptable risk levels" are and
how EPA derived them (e.g., are the acceptable levels expressed for total PCBs, selected
Aroclors, or individual congeners?). Further, a reviewer said it was difficult to comment
on whether remedial actions other than "No Action" could shorten the time for PCB
contamination in the river to reach "acceptable risk levels," without first knowing how
soon such levels would be reached under a "No Action" scenario. This reviewer also
found the question difficult to answer because specific remedial actions, such as dredging
sediments, can affect the ecological balance of the river system, regardless of how the
actions change levels of PCBs in the river.
•	Significance of higher concentrations of PCBs in the Thompson Island Pool. Citing the
models' inability to identify the mechanisms causing the increase of PCB loading across
the Thompson Island Pool, several reviewers wondered if the models would have the
predictive capacity to conduct longer (e.g., 20-year) simulations, such as evaluating
outcomes of different remedial actions. Elaborating on this point, a reviewer noted the
calibration described in the PMCR failed to explain the significant increase of PCB
concentrations over a relatively short calibration period within a relatively short section of
the Hudson River. Until the models can characterize the unique observations during the
model calibration period, some reviewers feared the models might not be able to answer
the second principal study question. For this same reason, one reviewer thought the
models may not be able to answer the first principal study question either.
•	Emphasis on source attribution. Related to the previous topic, a reviewer thought the
modeling should place a greater emphasis on source attribution. As an example, another
reviewer suggested that the models should be able to determine whether PCB loads
3-4

-------
originate from recent discharges or from historical sediment sources. Referring to page
4-2 of the PMCR, this reviewer noted that the models currently might not be able to
differentiate between these sources ofPCBs. These reviewers indicated that
understanding the sources ofPCBs during different flow conditions is critical to evaluating
different remedial options.
•	Needfor long-term model validation Reiterating a concern raised in response to
Question B. 1, a reviewer thought EPA should conduct long-term validation studies on the
proposed models under different flow conditions. The reviewer recommended that EPA
conduct these validation studies both for PCBs and for other chemicals with sufficient
historical data. Successful validation, according to the reviewer, could ultimately increase
confidence in the models' predictive capacity.
•	Better hydrogeological characterization of the Thompson Island Pool area. Noting that
the PMCR hypothesizes that groundwater advection might explain the observed increased
PCB loads across the Thompson Island Pool, one reviewer suggested that greater
hydrogeological characterization in this section of the Hudson River is needed to verify
this hypothesis. Subsequent discussions following the observer comment period suggested
that this may not be necessary as the apparent increase across the pools may be an artifact
of bias in a downstream monitoring location, as an observer suggested.
•	Consideration of two-stage partitioning models. On several occasions, one reviewer
recommended the possibility of using simple, two-stage partitioning models instead of the
more elaborate fate and transport and bioaccumulation models described in the PMCR.
The other reviewers commented on this proposition during their discussions on Question
D and E (see Section 5).
3.3 Characterizing the Fate of "Reactivated" Sediments Following Major Floods
The following discussion summarizes the reviewers' comments, observations, and
recommendations regarding Question B.3 of the charge:
•	Characterization of sediment erosion. One reviewer listed three areas in which the
models could better characterize sediment erosion. First, the reviewer recommended that
the models account separately for how cohesive and non-cohesive sediments erode,
instead of considering erosion of only cohesive sediments. Second, noting that turbulent
flow can enhance sediment erosion, even during periods of "average" river flows, the
reviewer suggested that EPA consider incorporating the role of turbulence in its models.
Third, noting that sediments can erode during "average" flows as well as during flood
events, the reviewer wondered whether the cumulative amounts of sediment eroded during
"average" flows might be of greater concern than the large amounts of sediment eroded
3-5

-------
during 100-year floods. Other reviewers agreed with these findings, and one reviewer
thought the relative significance of "average" flow and 100-year flood events was
particularly important for EPA's modeling efforts (as is discussed in greater detail in the
following bulleted items).
•	Significance of a 100-year flood event. The reviewers had difficulty reconciling the
models' predictions of sediment resuspension during a 100-year flood event in the Hudson
River and observed sediment resuspension during major flood events in other rivers. For
example, one reviewer noted that the "depth of scour" model predicts that SS pounds of
PCBs would be eroded from the Thompson Island Pool during a 100-year flood event in
the Hudson River. The reviewer noted further that SS pounds of PCBs is several orders
of magnitude less than the amount of PCBs believed to be originally discharged to the
river. Based on these observations, the reviewer suggested that a 100-year flood event in
the Hudson River may not cause significant erosion of sediments. On the other hand, two
reviewers provided quite different accounts of sediment erosion during major floods in
other river systems: one reviewer said tropical storm Charlie caused floods that almost
"completely scoured" the sediments in parts of the Rio Grande, and another reviewer
mentioned that rains from hurricane Agnes diluted the entire salt-water content of the
Chesapeake Bay. By these accounts, the reviewers implied that 100-year flood events
could quite effectively remove contaminants in sediments from rivers. To make sense of
these very different outcomes, a reviewer asked EPA to describe what conditions are
expected during a 100-year flood in the Hudson River. In response, one of EPA's
contractors indicated that, due in part to the feet that the Upper Hudson River has several
"controls" (e.g., locks and dams), water levels during a 100-year flood event for the
Upper Hudson River would not be significantly higher than those during a typical spring
high-flow event. The EPA contractor further noted that PCBs remain in the Hudson River
even though several notable high-flow events have occurred over the past 20 years. The
following three bulleted items list the reviewers' recommendations that evolved from this
discussion of 100-year floods.
•	Emphasis on modeling 100-year floods. Assuming that sediment erosion occurs during
"average" flows (see the first bulleted item under Section 3.3) and that erosion during
100-year flood events resuspends only a small fraction of the "sequestered" PCBs (see the
previous bulleted item), one reviewer wondered if EPA should place less emphasis on
evaluating the effects of a 100-year flood event and more emphasis on modeling the daily
"average" flows. Another reviewer cautioned that this suggestion assumes that 100-year
flood events do not erode significant quantities of sediments—a finding that is based on
only preliminary modeling results.
•	Role of freshly uncovered sediments. Two reviewers were concerned that the PMCR
focuses on modeling the amounts of resuspended sediments and not on characterizing
freshly exposed, and possibly highly contaminated, surfaces following flood events.
Unsure whether these freshly uncovered sediments are a more significant source of PCBs
3-6

-------
to fish than the resuspended sediments, the reviewers recommended that EPA examine the
significance of uncovered sediments in greater detail.
•	Role of resuspended sediments. One reviewer raised the possibility that resuspended,
PCB-contaminated sediments during flood events may act as a "source" (e.g., the
sediments that were previously sequestered from the environment may become
bioavailable) or they may act as a "sink" (e.g., flood waters may flush some contaminated
sediments from the Hudson River altogether). The reviewer recommended that EPA
clarify which scenario is most likely.
•	Links between the proposed models. One reviewer noted that the mass balance in the
HUDTOX model would not be complete if the model does not consider the amounts of
sediments resuspended during a 100-year flood. Thinking the ultimate fate of resuspended
sediments is just as important as the amount of the sediments that are resuspended, the
reviewer recommended that EPA link the HUDTOX and the "depth of scour" models in a
time-dependent fashion to ensure that HUDTOX models the transport and fate of
sediments that are resuspended during flood events.
•	Influence of other factors on PCB loadings. One reviewer indicated that factors other
than 100-year flood events can significantly increase PCB loadings in the Hudson River.
As examples, the reviewer suggested that bioturbation, "propeller backwash" from ships
and boats, and even earthquakes (which might damage or destroy containment systems) all
can affect sediment erosion, possibly significantly. The reviewer thought that EPA should
acknowledge these factors in its modeling efforts.
•	Better characterization of suspended solids. Citing information documented in other
reviewers' pre-meeting comments, one reviewer recommended that the models
incorporate a better physical characterization of resuspended sediments (e.g., particle size
and fraction of organic carbon).
3-7

-------
4.0 DISCUSSION ON QUESTION C: Specific Questions Regarding the Models
The peer reviewers continued their discussions by answering the five specific questions
listed in Question C of the charge. For these five questions, the following subsections restate a
specific question in the charge and summarize the reviewers' relevant discussions. When
answering the five specific questions, some reviewers referred to their comments from Questions
A and B. The following subsections note these references, where appropriate. It should be
noted, however, that Sections 6.2 and 6.3 present the reviewers' final list of major
recommendations.
4.1 Developing Quantitative Relationships Between Forcing Functions and PCB
Concentrations
Question C. 1 of the charge asked the peer reviewers:
"Are the modeling approaches suitable for developing quantitative relationships
between external forcing functions (e.g., hydraulic flows, solids and PCB loads,
sediment initial conditions, etc.) and PCB concentrations in the water column,
sediments and fish? Are the models adequate for discriminating between
water-related and sediment-related sources of PCB s?"
The following bulleted items summarize the reviewers' comments, observations, and
recommendations regarding this question:
•	Evolution of analytical methods for PCBs. Because laboratories, over the last 20 years,
have used different methods with different sensitivities to measure levels of PCBs, one
reviewer recommended that EPA make the issue of resolving historical differences in
analytical chemistry a priority for the ongoing modeling efforts. As an example, noting
peculiar trends in the distribution of congeners in selected sediment samples (e.g.,
relatively high levels of BZ4 in comparison to levels of other congeners), another reviewer
wondered if such peculiarities accurately reflect actual levels of environmental
contamination or possibly result from EPA misinterpreting historical monitoring data.
•	Improvements to the bioaccumulation models. Two reviewers thought, and all of the
reviewers later agreed, that only a mechanistic bioaccumulation model can discriminate
between "water-related and sediment-related sources of PCBs." Noting that the statistical
and probabilistic models proposed in the PMCR do not address mechanisms of
bioaccumulation, one reviewer recommended that EPA use mechanistic models that, to
4-1

-------
the greatest extent possible, address congener-specific rates of uptake, metabolism,
excretion, and storage of PCBs in biota.
Inputs from the watershed. One reviewer thought PCB inputs from the watershed might
be a relevant forcing function for the models, especially for long-term forecasting, but the
reviewer indicated that the PMCR currently assumes such inputs are negligible. The
reviewer acknowledged that current understanding of PCB inputs from watersheds may be
too limited to incorporate in EPA's modeling approach. Nonetheless, another reviewer
thought the PMCR should at least compare PCB inputs from the Hudson River watershed
to other PCB loads in the system. This reviewer indicated that EPA may be able to
characterize watershed loadings by moving the boundary of the modeling domain to
locations upstream of Hudson Falls—a recommendation which the next bulleted item
describes in greater detail.
Location of the upstream modeling boundary. As the reviewers noted when discussing
Questions A and B, the proposed location of the upstream modeling boundary may limit
the models' ability to relate external forcing functions to concentrations of PCBs in the
Hudson River. One reviewer indicated that moving the modeling boundary upstream of
Hudson Falls will allow EPA to characterize the background levels of PCBs that enter the
modeling domain. (As Sections S and 6 describe, the peer reviewers' recommendations
on the placement of the upstream boundary for modeling changed during the meeting, and
they eventually decided that moving the boundary may not be necessary.)
Consideration of "apparent" dissolved phase PCBs. One reviewer thought that PCB
measurements that do not distinguish "truly dissolved PCBs" from PCBs that are bound to
dissolved organic carbon (DOC), such as the "Phase 2" measurements made in the Upper
Hudson River, can be a major problem for the modeling results. The reviewer
recommended that EPA make a better distinction between these two forms of PCBs.
Further, another reviewer thought the PMCR should better characterize the DOC typically
found in the Hudson River.
Bias in modeling results. Based on a comparison of the modeling results during low-flow
and high-flow events, one reviewer thought the predicted concentrations of PCBs and
concentrations of total suspended solids (TSS) may have been biased. The reviewer
recommended that EPA closely examine the relationship between flow, total suspended
solids, and PCBs, and explain any peculiar trends; rather than dismiss outliers, as was
done.
Adequacy of Spatial and Temporal Scales
Question C.2 of the charge asked the peer reviewers:
4-2

-------
"Are the spatial and temporal scales of the modeling approaches adequate to
answer the principal study questions? If not, what levels of spatial and temporal
resolution are required to answer these questions? What supporting data are
required for calibration/validation of these spatial and temporal scales?"
The following discussion summarizes the reviewers' comments, observations, and
recommendations regarding this question:
•	Spatial resolution of the HUDTOX and "depth of scour " models. The reviewers agreed
that the HUDTOX and "depth of scour" models have adequate spatial resolution for
modeling flows in the Hudson River. Further, one reviewer thought that use of finer
resolution in these models will significantly increase the time needed for computation,
without providing much greater insight into fate and transport of PCBs in the Hudson
River.
•	Links between the models. Reiterating comments they made when discussing Questions A
and B, several peer reviewers recommended that EPA link the spatial scales and temporal
scales of every proposed model (e.g., the HUDTOX model, "depth of scour" model,
Thomann model, and bioaccumulation models). The reviewers thought these links were
important for modeling the Hudson River over the long term.
•	Boundaries of the models. The reviewers commented on two aspects of the boundaries of
the proposed models. First, two reviewers wondered if EPA could use a single model to
characterize PCBs in the entire Hudson River, instead of using the Thomann model for the
Lower Hudson River and the HUDTOX model for the Upper Hudson River. In the event
that EPA continues to use separate models, these reviewers thought the models should at
least be properly linked (see previous bulleted item). Second, consistent with comments
made on Questions A and B, several reviewers again noted that the models currently
neglect segments of the Hudson River upstream from Hudson Falls. (However, as
Sections 5 and 6 describe, the peer reviewers' thinking on the placement of the upstream
boundary for modeling changed during the meeting, and they eventually decided that
moving this boundary may not be necessary.)
•	Temporal scale of the bioaccumulation models. Several reviewers noted that the
proposed bioaccumulation models have no temporal component, thus preventing them
from predicting how changing levels of PCBs in the Hudson River will affect levels of
PCBs in fish. One reviewer explained that the proposed models do not account for
temporal changes in fish feeding patterns, which are known to vary daily and seasonally,
or sensitive life stages. Citing these temporal changes, another reviewer thought time-
dependent concentrations of PCBs in the water column may be important for
characterizing PCB uptake by fish. This reviewer recommended that EPA use a
4-3

-------
"bioenergetically-based food chain model" with adequate temporal components. A
different reviewer noted that EPA may have already started using such a food chain
model.
Temporal scale of the "depth of scour" model. One reviewer thought the temporal scale
in the "depth of scour" model was generally adequate, but he disagreed with the model's
assumption that bed shear stresses reach their maximum values instantaneously. However,
the reviewer did not recommend alternative assumptions for addressing temporal
variations in bed shear stresses.
Model calibration issues. The reviewers made several comments regarding how EPA
proposes to calibrate its models. For example, noting that calibration data were available
for only 6 of the 13 river segments in the HUDTOX model, one reviewer thought EPA
should reconcile the number of modeling segments with the available data before
conducting the final model calibration. Regarding the model calibration period, another
reviewer was discouraged by the statement in the PMCR that "it is not yet clear whether
the PCB dynamics operative during this simulation period are fully representative of
historical PCB dynamics, or whether they will be representative of PCB dynamics under
future conditions" (Limno-Tech et al., 1996). Yet another reviewer recommended that
EPA calibrate only a certain set of parameters in the models, rather than calibrating "the
entire process."
Data validation issues. Several reviewers discussed the importance of data validation,
particularly with respect to levels of PCBs in fish. One reviewer suggested that data
validation should consider concentrations of PCBs in sediment, the water column, and
fish, but two reviewers stressed that EPA should place much greater emphasis on
validating fish body burdens (e.g., average concentrations of PCBs and standard
deviations of the concentrations). One of these reviewers thought using the models to
estimate fish concentrations back to 1977 is also important, but only after EPA resolves
historical differences in PCB analytical methods. This reviewer further recommended that
validation not only consider PCBs, but also other chemicals with sufficient data. Finally,
one reviewer emphasized that EPA should develop specific validation criteria (e.g., ranges
of concentrations of PCBs in fish that would be considered "acceptable") before it actually
performs the data validation.
Use of Several Bioaccumulation Models
Question C.3 of the charge asked the peer reviewers:
"It is contemplated that PCB concentrations in fish will be estimated using several
modeling approaches: an empirical probabilistic model derived from Hudson River
data, a steady state model that takes into account mechanisms of bioaccumulation
body burdens, and a time-varying mechanistic model (not included in the PMCR).
4-4

-------
A bi-variate statistical model may also be used to provide insight into
accumulations. This multi-model approach is being contemplated because of the
uncertainties associated with any individual model. Is this a reasonable approach
or should predictions be made using a single 'best' model?"
The following bulleted items summarize the reviewers' comments, observations, and
recommendations regarding this question:
•	Use of multiple bioaccumulation models. After lengthy discussions on the topic, the peer
reviewers agreed that EPA's proposed use of multiple bioaccumulation models to estimate
concentrations of PCBs in fish is advisable, even though the reviewers thought the best
option ultimately will be to use a time-dependent, mechanistic model. One reviewer noted
that comparing predictions from multiple models has been a productive exercise for other
modeling endeavors, even when predictions from the individual models differ. Another
reviewer urged EPA to review the assumptions of the candidate models and to develop
expectations of what the different models will predict, before comparing results from the
different models. This reviewer also suggested that EPA consider the quality of each
model's input data as a factor when selecting which model is most appropriate.
•	Input from risk assessors. Recognizing that risk assessors will ultimately use the outputs
from the bioaccumulation models as inputs for risk modeling, two reviewers stressed the
importance of early and frequent involvement of risk assessors in the transport and fate
and bioaccumulation modeling, including the selection of appropriate models.
•	Use of experimental data. To provide greater insight into bioaccumulation of PCBs in the
Hudson River system, one reviewer thought, EPA should consider conducting simple
laboratory studies designed to mimic the Hudson River. Although another reviewer
agreed that additional laboratory studies might help EPA parameterize its models better,
two other reviewers were not convinced that additional experimental studies should
precede EPA's modeling efforts: one of these reviewers noted that models can be used to
identify where experimentation is needed, and the other reviewer indicated that laboratory
experiments probably would not provide useful information in the short term, due to the
long time scales needed to observe bioaccumulation of PCBs in fish.
4.4 Adequacy of the Models' "Level of Process Resolution"
Question C .4 of the charge asked the peer reviewers:
"Is the level of process resolution in the models adequate to answer the principal
study questions? If not, what processes and what levels of resolution are required
4-5

-------
(e.g., endangered species and importance for fishing) when selecting fish species for
modeling, another reviewer thought the six selected species were adequate and suggested
that modeling fewer species might lower process resolution.
•	Level of resolution neededforfinal decisions. One reviewer found the question difficult
to answer without knowing what level of process resolution EPA requires to make its
decisions on remedial actions. As an example, the reviewer explained that the level of
process resolution for the models would have to be extremely high if EPA needed to
predict concentrations of PCBs in fish with very tight error bounds; lower levels of
process resolution might be adequate if broader error bounds in the predicted
concentrations of PCBs in fish are acceptable.
4.5 Usefulness of Modeling Results for Risk Assessment
Question C.5 of the charge asked the peer reviewers:
"The results of the modeling effort will be used, in part, to support human and
ecological risk assessments. In your judgment, will the models provide estimates
adequate for this purpose?"
The following discussion summarizes the reviewers' comments, observations, and
recommendations regarding this question:
•	Involvement of risk assessors. Consistent with comments made to Questions A and B and
C.3, two reviewers emphasized the importance of early and frequent involvement of risk
assessors in EPA's ongoing modeling efforts. The reviewers explained that this
involvement should help ensure that outputs from the fate and transport and
bioaccumulation models will coincide with the desired inputs for the risk assessments.
•	Relevant forms of PCBs. Also consistent with comments raised earlier in the peer review
meeting, one reviewer was not convinced that the proposed models and the risk
assessments will consider the same forms of PCBs. As a result, the reviewer suggested
that the models described in the PMCR might need to characterize concentrations of only
total PCBs, instead of characterizing levels of individual PCB congeners, selected PCB
homologues, or selected Aroclors. Noting that some health outcomes are linked to
specific PCB exposures (e.g., exposure to lower homologues of PCBs may be a more
likely cause of neurotoxic developmental effects in children than exposure to higher
homologues of PCBs), another reviewer did not agree that modeling only total PCBs was
advisable. These reviewers agreed, however, that EPA should model fate and transport
4-7

-------
and bioaccumulation of at least those forms of PCBs that the human health and ecological
risk assessments will consider.
•	Adequacy of the proposed bioaccumulation models. Again reiterating a topic from earlier
discussions, one reviewer emphasized that the proposed bioaccumulation models may not
generate data adequate for the risk assessment. The reviewer recommended that EPA use
a dynamic, mechanistic bioaccumulation model instead of the bivariate and probabilistic
models.
•	Specifics for the ecological risk assessment. Two peer reviewers suggested that the
ecological risk assessment should consider mink, and possibly snapping turtles, as sentinel
species, since, once revised as suggested, the models described in the PMCR would
provide estimates adequate for assessing risks to these species and their consumers or
others (e.g., otter). (The reviewers agreed that the models should provide data from
which exposures relevant to the ecological risk assessment can be estimated.)
4-8

-------
5.0 DISCUSSION ON QUESTION D AND E: Recommended Changes to, and Serious
Flaws in, the Proposed Modeling Approach
The peer reviewers continued their technical discussions by answering Questions D and E
in the charge:
"D. Are there any changes to the work effort outlined in the revised work plan that
would significantly improve the outcome?
E. In terms of evaluating the overall and specific effects and behavior of PCBs in the
Hudson River, are there any serious flaws in the modeling approach (theory,
structure, physical parameters, etc.) that would limit or invalidate any conclusions
or further work based upon the results of these models?"
The peer reviewers decided to address these questions at the same time, because they
thought serious flaws in the modeling approach (i.e., responses to Question E) would imply the
need for revising the modeling work plan (i.e., responses to Question D). Before answering these
questions, however, some peer reviewers indicated that they were uncomfortable commenting on
improvements EPA has made since the PMCR was published. Other peer reviewers noted that
their responses to Questions D and E would be quite similar, if not identical, to their responses to
Questions A, B, and C.
The reviewers made the following comments, observations, and recommendations
regarding Questions D and E of the charge (but, it should be noted that Sections 6.2 and 6.3
present the reviewers' final list of major recommendations):
• Improved characterization of sediment transport. Several peer reviewers' responses to
Questions D and E focused on issues related to sediment transport. For instance,
consistent with responses to Question B.3, one peer reviewer thought the proposed
models could be enhanced by considering how non-cohesive sediments erode and how
turbulence affects sediment resuspension. The reviewer noted that these factors have been
considered in other recent modeling efforts (Cao, 1997). Further, this reviewer
emphasized that sediment erosion can occur during low-flow conditions, thus potentially
causing the cumulative effects of sediment erosion during "average" flows to be just as
5-1

-------
significant as effects of sediment erosion during floods. Building on these findings,
another peer reviewer suggested that EPA can better characterize sediment transport by:
(1) providing some justification for assuming that the "active layer" of sediments is 5
centimeters thick; (2) accounting for the ultimate fate of resuspended sediments;
(3) considering the significance of freshly uncovered, potentially contaminated sediments;
and (4) incorporating semi-empirical sediment resuspension algorithms in HUDTOX,
rather than characterizing sediment resuspension rates by model calibration.
Although the reviewers generally agreed on how the models should characterize sediment
transport, the reviewers did not agree on an approach EPA should take to revise its
models. One reviewer recommended that EPA incorporate the "depth of scour" model
directly into the HUDTOX model, rather than leaving the models decoupled. This
reviewer thought the use of a single, comprehensive model is needed to track the fete of
resuspended sediments. Another reviewer disagreed, however, thinking EPA can improve
its characterization of sediment transport while still using two different models. More
specifically, noting that the HUDTOX and "depth of scour" models were developed to
address two different principal study questions, this reviewer thought it may be
appropriate to keep the models separate. In any case, this reviewer thought, it was
sufficient for the peer reviewers simply to identify the shortcomings in the proposed
modeling approach and to let EPA decide how to improve its models.
•	Statistical analysis of the modeling results. Two reviewers thought the use of Student's
t-tests to evaluate the performance of the HUDTOX model (e.g., see page 4-20 of the
PMCR) was statistically invalid. Both reviewers agreed that these invalid analyses may
have biased conclusions in the PMCR regarding performance of the models. The two
reviewers suggested that EPA instead use regression analyses or simple plots of modeling
results against observed values.
•	Link between the Thomarm and HUDTOX models. One reviewer thought the use of two
models with different assumptions to address PCB transport in two different regions of the
Hudson River could be a "serious flaw." The reviewer suggested that there should be
more agreement between how HUDTOX characterizes PCB transport in the Upper
Hudson River and how the Thomann model characterizes PCB transport in the Lower
Hudson River.
•	Use of congener-specific data. Citing an example of how Henry's Law constants may
differ for individual congeners within a PCB homologue, one reviewer recommended that
EPA consider, to the greatest extent possible, congener-specific data for Henry's Law
constants, fish uptake and metabolic rates, degradation rates, and other relevant physical,
chemical, and biological mechanisms.
•	Use of predictive bioaccvmulation models. Consistent with comments made on
Questions A through C, two reviewers emphasized the importance of using predictive,
5-2

-------
• Topics discussed earlier in the meeting. At the end of their discussions on Questions D
and E, the peer reviewers mentioned that they could have addressed several additional
topics from their earlier discussions on Questions A through C. These additional topics
include, but are not limited to, incorporating the risk assessment paradigm, involving risk
assessors early and often, validating the model, addressing factors that affect
bioavailability, and resolving analytical chemistry issues. Rather than repeat their
discussions on these topics, however, the peer reviewers instead decided to classify their
findings into "major recommendations'' and "minor recommendations," as described in
Section 6.
5-4

-------
6.0	REVIEWERS' OVERALL RECOMMENDATIONS
After completing their technical discussions on Questions A through E, the reviewers
spent the final hours of the peer review meeting identifying their major findings and
recommendations for the proposed PCB modeling approach. The peer reviewers generally
separated initial findings from their earlier discussions into three categories: major
recommendations, minor recommendations (or suggested revisions), and findings they no longer
considered important. This section summarizes the reviewers' discussions that led to identifying
major recommendations (Section 6.1), then lists the reviewers' major recommendations (Section
6.2), and finally documents the peer reviewers' closing statements to EPA (Section 6.3).
6.1	Discussions That Led to Identifying Major Recommendations
As the first step in identifying their major recommendations, the peer reviewers engaged in
a detailed discussion on the relative importance of the many recommendations presented earlier in
the peer review meeting. During these discussions, the peer reviewers noted that their
understanding of the modeling approach had evolved significantly since the time when they
prepared their premeeting comments (see Appendix C), and even had evolved since the first day
of the peer review meeting. The peer reviewers also consistently acknowledged that the issues
they brought up may have already been addressed, since 2 years have passed since the PMCR was
published.
When trying to identify their most important findings for the proposed PCB modeling
approach, the reviewers made the following comments, observations, and recommendations. It
should be noted, however, that Sections 6.2 and 6.3 present the reviewers' final list of major
recommendations.
• Importance of involving risk assessors. The peer reviewers discussed at length the
importance of involving risk assessors early and often in EPA's ongoing modeling efforts.
For example, several reviewers emphasized as a "key message" of the peer review that the
transport and fate and bioaccumulation modelers must be aware, if they are not already so,
of the different exposure endpoints and forms of PCBs that the risk assessors will
consider. These reviewers thought the modeling approach would be "fatally flawed" if the
6-1

-------
modelers were unaware of the ultimate ride assessment targets. These reviewers were
concerned that even scientifically rigorous transport and fate and bioaccumulation models
will be rendered useless if their outputs do not interface with the inputs for risk
assessment. To enhance risk assessor involvement, one reviewer suggested, EPA should
hold a workshop or series of workshops to identify bioaccumulation models that provide
data sufficient for conducting the planned human health and ecological risk assessments.
When discussing this topic, one reviewer was uncomfortable making a strong
recommendation to EPA without knowing the extent to which EPA has involved, and
continues to involve, risk assessors in the modeling efforts. Responding to this concern,
several other reviewers indicated that the PMCR shows no evidence that risk assessors
provided input to the modeling approach. These reviewers listed several specific
questions that risk assessors should have answered at the onset of EPA's modeling efforts:
What forms of PCBs (e.g., congeners, Aroclors, homologues, total PCBs) will the risk
assessors consider? Will the risk assessors conduct a TEQ analysis? What exposure
pathways will be considered? What species will be considered in the ecological risk
assessment? What durations of exposure will the risk assessments consider? Without
finding any indication in the PMCR that these issues have been resolved, most of the
reviewers assumed that risk assessors have had limited involvement in EPA's ongoing
modeling efforts.
Due to limitations in the transport and fate and bioaccumulation models, one reviewer
thought a compromise may eventually be necessary between what the risk assessors would
ideally like to consider and what the modelers can actually provide. More specifically, this
reviewer noted that insufficient data may be available to calibrate the transport and fate
models for every PCB congener, possibly even for congeners selected for analysis in the
risk assessment. Another reviewer disagreed, noting that modelers can consider
chemically similar PCB congeners to evaluate transport and fate and bioaccumulation of
congeners that cannot be validated or tested directly. In any case, all of the reviewers
agreed that the modeling efforts will be most useful if risk assessors specify at least (1) the
types of species the bioaccumulation models should consider, (2) the forms of PCBs that
need to be quantified, and (3) the durations of exposure that will be evaluated (i.e., acute
or chronic).
• Improvements to the sediment transport algorithms. Noting that sediment transport can
play a critical role in the bioaccumulation of PCBs in fish, the peer reviewers listed several
areas in which EPA could improve its characterization of sediment transport. For
instance, one peer reviewer thought the HUDTOX and "depth of scour" models should
address how non-cohesive sediments erode, but he acknowledged that EPA may have
considered this issue since the PMCR was published. This reviewer did not identify any
"fetal flaws" with the HUDTOX and "depth of scour" models from a purely scientific
basis, but the reviewer indicated there may be problems with how EPA intends to use
them (e.g., how EPA will calibrate and link the models). Another reviewer echoed this
6-2

-------
concern by indicating that the models will be only as good as the data against which they
are calibrated. As an example, this reviewer indicated that the apparent increased loadings
of PCBs in the Thompson Island Pool may have resulted from a sampling bias at a
downstream monitoring location as suggested during the observer comment period (see
Appendix F. To prevent biased monitoring data from invalidating the modeling results,
this reviewer recommended, EPA should carefully verify the accuracy of all data used for
calibration. Yet another reviewer commented on sediment transport by recommending
that HUDTOX include a more mechanistic treatment of sediment resuspension and track
the fate of resuspended sediments.
•	Use of mechanistic bioaccumulation models. Several peer reviewers agreed that use of
only the bivariate and probabilistic food chain models (i.e., the models described in the
PMCR) would be a "fetal flaw" in EPA's effort to evaluate bioaccumulation of PCBs.
These reviewers thought use of a mechanistic, predictive bioaccumulation model would be
necessary to answer the principal study questions listed in the PMCR. These reviewers
also acknowledged that EPA currently plans to include a mechanistic bioaccumulation
model in its future work.
•	Links between individual models. One reviewer thought EPA's proposed modeling
approach may ultimately be rejected if the HUDTOX and Thomann models are not linked
"in a seamless manner" and if these models do not share common assumptions. This
reviewer thought the links were necessary (1) because PCBs will eventually flow from the
Upper Hudson River (the domain of HUDTOX) to the Lower Hudson River (the domain
of the Thomann model) and (2) because EPA will eventually assess human health and
ecological risks to receptors throughout the entire Hudson River valley.
•	More sophisticated consideration of bioavailability. Citing comments raised in responses
to Questions A through C, the peer reviewers recommended that the modelers consider
bioavailability in greater detail, possibly by developing chemical fate parameters (e.g.,
sorption and desorption kinetics) and biological fate parameters (e.g., uptake and
metabolic rates), and incorporating these parameters into a bioaccumulation model—an
approach that is not currently proposed in the PMCR
•	Involvement of stakeholders. One reviewer suggested that EPA's ongoing modeling
efforts would benefit from the involvement of multiple stakeholders. This reviewer
thought such involvement would ensure not only the scientific rigor of modeling results,
but also the appropriateness of selected remediation options. The reviewer noted that this
suggestion was related more to EPA's site reassessment process than to the technical
validity of the models described in the PMCR. During this discussion, EPA explained that
stakeholder involvement (including community interaction programs) has been a critical
element of the site reassessment process, even though the PMCR did not describe this
involvement.
6-3

-------
• Comprehensive evaluation of different remedial options. Noting that the models
described in the PMCR can only predict the effects of remedial options on concentrations
ofPCBs in the Hudson River and in fish, one reviewer recommended that EPA should
also focus its resources on conducting a "life-cycle type of assessment" in order to
evaluate the overall environmental burden of the different remedial options. This reviewer
acknowledged that EPA already uses several criteria to evaluate the feasibility of different
remedial options. (Another reviewer clarified that a "life-cycle type of assessment"
considers several technical, economic, and political criteria for making decisions.)
6.2 Peer Reviewers* Major Recommendations
The peer reviewers briefly discussed selected responses to Questions A through E before
identifying their major recommendations for the modeling effort. During this discussion, the
reviewers agreed that they no longer considered moving the boundary of the modeling domain to
a location upstream of Hudson Falls as an important issue, and they identified selected key
findings, such as ensuring that the individual models will be linked, incorporating the modeling
approach into the risk assessment paradigm, and considering bioavailability in greater detail.
To prepare a final list of their overall recommendations to EPA, the peer reviewers
decided to have each peer reviewer list his or her major recommendations and then to characterize
the extent to which the peer reviewers agreed or disagreed with each individual's
recommendations. The comprehensive list of the reviewers' major findings and recommendations
is provided below, as well as in the Executive Summary. Unless noted otherwise, all of the peer
reviewers agreed with these major recommendations.
The reviewers recommended that EPA make the following improvements in the
description of sediment resuspension and deposition processes in the fate and transport
models: address the fete of resuspended material, address the role of uncovered,
potentially contaminated surfaces; address the issue of non-cohesive sediment
resuspension; assure consistency in resuspension rates between the "depth of scour" and
HUDTOX models; and identify the effect of flood resuspension on the rate of long-term
recovery of the Hudson River. Some reviewers indicated that these changes could be
made within the existing modeling framework (i.e., with different fate and transport
models that are linked), while other reviewers thought these changes should be made by
incorporating sediment transport mechanisms directly into the HUDTOX model, instead
of keeping the models separate. A reviewer added that EPA should analyze error
6-4

-------
propagation in the linked models and place a greater emphasis on "source attribution of
PCBs to fish."
•	The reviewers recommended that EPA employ time- and space-dependent mechanistic
models that reflect the abiotic and biotic dynamics of the Hudson River system.
The reviewers indicated that the models should be based on a consideration of
bioavailability and sediment sequestration with respect to congener sorption/desorption
kinetics, sediment particle characteristics, and biotic characteristics.
•	The reviewers recommended that EPA link, to the greatest extent possible, the spatial and
temporal scales of the different fate and transport and bioaccumulation models.
•	The reviewers recommended that EPA clearly identify in its modeling approach the risk
assessment targets as related to forms and concentrations of PCBs (e.g., to what
guidelines or advisories will the modeled concentrations be compared). More specifically,
the reviewers thought risk assessors and managers should be involved with the
development of the transport and fate and bioaccumulation models, such that the model
outputs will generate the data needed for completing human health and ecological risk
assessments.
The reviewers recommended that EPA develop a mechanistic food web model based on
exposure dynamics of the identified forms of PCBs relevant to risk quantification, and that
EPA identify appropriate data needs for the fate and transport models.
•	The reviewers recommended that EPA develop an explicit plan for model calibration and
independent validation that includes criteria for validation, makes use of numerous data
sets that span a long time period, and includes chemicals in addition to PCBs.
•	Noting that analytical methods have improved and will continue to improve, the reviewers
recommended that EPA develop, and agree on how to use, a method for interpreting
historical PCB monitoring data.
Noting that EPA has already addressed or considered many of the recommendations listed
above, the reviewers suggested that EPA hold an open workshop (involving all interested
parties as well as independent reviewers) to evaluate current modeling efforts.
6.3 Peer Reviewers' Final Statements
The peer review meeting concluded with each peer reviewer providing closing statements
on the proposed modeling approach, including an "overall recommendation" in response to the
final question in the charge: "Based on your reading and analysis of the information provided,
6-5

-------
please identify and submit an explanation of your overall recommendation for the modeling effort
for the Hudson River PCB Reassessment RI/FS:
1.	Acceptable as is
2.	Acceptable with minor revision (as indicated)
3.	Acceptable with major revision (as outlined)
4.	Not acceptable (under any circumstance)"
During their final statements, several peer reviewers commended EPA on its ongoing
efforts to model PCBs in the Hudson River, and some reviewers applauded EPA for using peer
review to test the scientific rigor of the site reassessment process. Every reviewer found the
overall modeling approach, as described in the PMCR, to be "acceptable with major revision";
however, one reviewer found the HUDTOX and "depth of scour" models to be "acceptable with
minor revision." Noting that this classification is based primarily on the information documented
in the PMCR, several reviewers acknowledged that EPA may have already addressed several of
the major recommendations in the past 2 years.
6-6

-------
7.0 REFERENCES
Cao, Z. 1997. Turbulent Bursting-Based Sediment Entrainment Function. Journal of Hydraulic
Engineering, 123: 3, 233-236.
Limno-Tech et al., 1996. Preliminary Model Calibration Report. Prepared by Limno-Tech, Inc.,
Menzie Cura & Associates, Inc., and The CADMUS Group, Inc. Prepared for U.S.
Environmental Protection Agency Region II.
Limno-Tech et al., 1998. Preliminary Model Calibration Report, Revised Appendix B. Prepared
by Limno-Tech, Inc., Menzie Cura & Associates, Inc., and The CADMUS Group, Inc.
Prepared for TAMS Consultants, Inc.
7-1

-------
APPENDIX A
LIST OF EXPERT PEER REVIEWERS

-------
United States
LailimCjk Environmental Protection Agency
m\ Region 2
Hudson River PCBs Site
Modeling Approach Peer Review
Sheraton Saratoga Springs Hotel and Conference Center
Saratoga Springs, NY
September 9-10,1998
Peer Reviewers
Ellen Bentzen
Research Scientist
Environmental Studies
Trent University
Environmental Modeling Center
Peterborough, Ontario CANADA K9J 7B8
705-748-1645
Fax: 705-748-1569
E-mail: ebentzen@trentu.ca
G. Douglas Haffner
Professor
Biological Sciences
University of Windsor
Windsor, Ontario CANADA N9B 3P4
519-253-4232, Ext. 3449
Fax: 519-971-3616
E-mail: haffner@uwindsor,ca
Miriam Diamond
Associate Professor
Department of Geography
University of Toronto
100 St. George Street
Toronto, Ontario CANADA M5S 3G3
416-978-1586
Fax: 416-978-6729
E-mail: diamond@geog.utoronto.ca
Alan Maki
Environmental Advisor
Exxon Company, USA
800 Bell Street-Suite 4111
Houston, TX 77002
713-656-3945
Fax: 713-656-9430
E-mail: al.w.maki@exxon.sprint.com
James Gillett
Professor of Ecotoxicology
Cornell University
216 Rice Hall
Ithaca, NY 14853
607-255-2163
Fax: 607-255-0238
E-mail: jwg3@comell.edu
Thanos Papanicolaou
Assistant Professor
Department of Civil Engineering
Washington State University
101 Sloan Hall
Pullman, WA 99164-2910
509-335-2144
Fax: 509-335-7632
E-mail: apapanic@wsu.edu
Frank Wania
Wania Environmental Chemists Corporation
280 Simcoe Street - Suite 404
Toronto, Ontario, CANADA M5T 2Y5
416-977-8458
Fax: 416-977-4953
E-mail: frank.wama@utoronto.ca
Printed on Recycled Paper
*ERC

-------
APPENDIX B
CHARGE TO EXPERT PEER REVIEWERS

-------
United States
JMU ImMlj Environmental Protection Agency
VU #m Region 2
Hudson River PCBs Site
Modeling Approach Peer Review
Sheraton Saratoga Springs Hotel and Conference Center
Saratoga Springs, NY
September 9-10, 1998
REVISED CHARGE TO REVIEWERS
Members of this peer review will be tasked to determine whether the models being used to
support the decision-making process for the Reassessment, and the assumptions therein,
are appropriate. The peer reviewers will base their assessment on the review the
Preliminary Model Calibration Report (PMCR), an updated Technical Scope of Work for the
Baseline Modeling Report (Appendix B of the PMCR) and the responses to selected
comments received from stakeholders during the public comment period on the PMCR.
In October 1996, EPA released the Preliminary Model Calibration Report (PMCR), which
described the models, datasets and assumptions being used as part of the Hudson River
PCB Reassessment RI/FS. The PMCR represents the status of the preliminary PCB
modeling effort as of Fall 1995. Datasets, database corrections and other pertinent
information which became available after October 1995 were not incorporated within the fate
and transport modeling presented in the PMCR. The PMCR was an interim document
prepared to describe work in progress and was not intended to be a conclusive report. In
particular the HUDTOX model presented in the PMCR was not intended to be used as a
predictive tool to assess remedial action scenarios. In addition, while time-varying
mechanistic models of bioaccumulation will be used along with other models to predict fish
body burdens, these models are not described in the PMCR.
The PMCR was not formally peer reviewed at the time of publication, but was distributed to
interested parties who were invited to submit comments and questions. Written responses
were made to all of these comments and questions. In addition, the work plan contained in
Appendix B of the PMCR has been revised to reflect the ongoing work being conducted as
part of the Baseline Modeling effort. Results from this effort will be presented in a Baseline
Modeling Report that will be formally peer reviewed.
The peer reviewers are requested to determine whether the models being used to support
the decision-making process for the Reassessment RI/FS, and the assumptions therein, are
appropriate. The peer reviewers are not being asked whether they would conduct the work
in the same manner, only whether the work being conducted will yield scientifically credible
conclusions.
It is suggested that the reviewer first read the PMCR. The Responses to Comments

-------
provides information on the context of the PMCR within the overall modeling effort and
additional details beyond the PMCR results. The current work plan as revised in June 1998
reflects the ongoing Baseline Modeling effort and revisions to some of the original modeling
tasks proposed in Appendix B of the PMCR. In addition, the USEPA/TAMS Phase 2
database has been considerably revised. New datasets have been added and some earlier
datasets have been extensively revised.
The peer reviewers are asked to comment on the following:
A.	Is EPA using appropriate models, datasets and assumptions on which to base a
scientifically credible decision?
B.	Will the models, with the associated datasets and assumptions, be able to answer the
following principal study questions as stated in the PMCR:
1.	When will PCB levels in the fish population recover to levels meeting human
health and ecological risk criteria under No Action?
2.	Can remedies other than No Action significantly shorten the time required to
achieve acceptable risk levels?
3.	Are there contaminated sediments now buried and effectively sequestered from
the food chain which are likely to become "reactivated" following a major flood,
resulting in an increase in contamination of the fish population?
C.	Specific questions:
1.	Are the modeling approaches suitable for developing quantitative relationships
between external forcing functions (e.g., hydraulic flows, solids and PCB loads,
sediment initial conditions, etc.) and PCB concentrations in the water column,
sediments and fish? Are the models adequate for discriminating between
water-related and sediment-related sources of PCBs?
2.	Are the spatial and temporal scales of the modeling approaches adequate to
answer the principal study questions? If not, what levels of spatial and
temporal resolution are required to answer these questions? What supporting
data are required for calibration/ validation of these spatial and temporal
scales?
3.	It is contemplated that PCB concentrations in fish will be estimated using
several modeling approaches: an empirical probabilistic model derived from
Hudson River data, a steady state model that takes into account mechanisms
of bioaccumulation body burdens, and a time-varying mechanistic model (not
included in the PMCR). A bi-variate statistical model may also be used to
provide insight into accumulations. This multi-model approach is being
contemplated because of the uncertainties associated with any individual
model. Is this a reasonable approach or should predictions be made using a
single "best" model?

-------
4.	Is the level of process resolution1 in the models adequate to answer the
principal study questions? If not, what processes and what levels of resolution
are required to answer these questions? What supporting data (such as data
to support specifications of a mixed depth layer, solids and scour dynamics,
groundwater inflow, etc.) are required for these processes and levels of
resolution?
5.	The results of the modeling effort will be used, in part, to support human and
ecological risk assessments. In your judgment, will the models provide
estimates adequate for this purpose?
D.	Are there any changes to the work effort outlined in the revised work plan that would
significantly improve the outcome?
E.	In terms of evaluating the overall and specific effects and behavior of PCBs in the
Hudson River, are there any serious flaws in the modeling approach (theory, structure,
physical parameters, etc.) that would limit or invalidate any conclusions or further work
based upon the results of these models?
Recommendations
Based on your reading and analysis of the information provided, please identify and submit
an explanation of your overall recommendation for the modeling effort for the Hudson River
PCB Reassessment RI/FS:
1.	Acceptable as is
2.	Acceptable with minor revision (as indicated)
3.	Acceptable with major revision (as outlined)
4.	Not acceptable (under any circumstance)
1. The "level of process resolution" refers to the theoretical rigor of
the equations used to describe the various processes affecting PCB
fate and transport such as: settling, resuspension, volatilization,
biological activity, partitioning, etc. An example of low process
resolution is use of a constant value for the solids resuspension
rate. A higher level of process resolution is use of a complex
mathematical description of the physics involved in remobilizing
bedded sediment particles (such as cohesive forces, bed shear
stresses, etc.)

-------
APPENDIX C
PREMEETING COMMENTS, ALPHABETIZED BY AUTHOR

-------
Hudson River PCBs Site
Modeling Approach Peer Review
Premeeting Comments
Saratoga Springs, NY
September 9-10, 1998
Prepared by:
Eastern Research Group, Inc.
110 Hartwell Avenue
Lexington, MA 02421
^ERS

-------
Table of Contents
Charge to Reviewers 	i
Peer Reviewer Comments
Ellen Bentzen	 1
Miriam Diamond	 11
James Gillett	 23
G. Douglas Haffner	 31
Alan Maki 	 39
Thanos Papanicolaou	 51
Frank Wania	 61

-------
CHARGE TO REVIEWERS
Members of this peer review will be tasked to determine whether the models being used to suppor
the decision-making process for the Reassessment, and the assumptions therein, are appropriate.
The peer reviewers will base their assessment on the review the Preliminary Model Calibration
Report (PMCR), an updated Technical Scope of Work for the Baseline Modeling Report (Appendi>
of the PMCR) and the responses to selected comments received from stakeholders during the
public comment period on the PMCR.
In October 1996, EPA released the Preliminary Model Calibration Report (PMCR), which describe!
the models, datasets and assumptions being used as part of the Hudson River PCB Reassessmer
RI/FS. The PMCR represents the status of the preliminary PCB modeling effort as of Fall 1995.
Datasets, database corrections and other pertinent information which became available after
October 1995 were not incorporated within the fate and transport modeling presented in the PMCF
The PMCR was an interim document prepared to describe work in progress and was not intended
be a conclusive report. In particular the HUDTOX model presented in the PMCR was not intendec
to be used as a predictive tool to assess remedial action scenarios. In addition, while time-varying
mechanistic models of bioaccumulation will be used along with other models to predict fish body
burdens, these models are not described in the PMCR.
The PMCR was not formally peer reviewed at the time of publication, but was distributed to
interested parties who were invited to submit comments and questions. Written responses were
made to all of these comments and questions. In addition, the work plan contained in Appendix B
the PMCR has been revised to reflect the ongoing work being conducted as part of the Baseline
Modeling effort. Results from this effort will be presented in a Baseline Modeling Report that will b
formally peer reviewed.
The peer reviewers are requested to determine whether the models being used to support the
decision-making process for the Reassessment RI/FS, and the assumptions therein, are
appropriate. The peer reviewers are not being asked whether they would conduct the work in the
same manner, only whether the work being conducted will yield scientifically credible conclusions.
It is suggested that the reviewer first read the PMCR. The Responses to Comments provides
information on the context of the PMCR within the overall modeling effort and additional details
beyond the PMCR results. The current work plan as revised in June 1998 reflects the ongoing
Baseline Modeling effort and revisions to some of the original modeling tasks proposed in Appendi
B of the PMCR. In addition, the USEPA/TAMS Phase 2 database has been considerably revised.
New datasets have been added and some earlier datasets have been extensively revised.
The peer reviewers are asked to comment on the following:
A.	Is EPA using appropriate models, datasets and assumptions on which to base a scientifica
credible decision?
B.	Will the models, with the associated datasets and assumptions, be able to answer the
following principal study questions as stated in the PMCR:
i

-------
1.	When will PCB levels in the fish population recover to levels meeting human health
and ecological risk criteria under No Action?
2.	Can remedies other than No Action significantly shorten the time required to achie\
acceptable risk levels?
3.	Are there contaminated sediments now buried and effectively sequestered from the
food chain which are likely to become "reactivated" following a major flood, resulting
in an increase in contamination of the fish population?
C. Specific questions:
1.	Are the modeling approaches suitable for developing quantitative relationships
between external forcing functions (e.g., hydraulic flows, solids and PCB loads,
sediment initial conditions, etc.) and PCB concentrations in the water column,
sediments and fish? Are the models adequate for discriminating between water-
related and sediment-related sources of PCBs?
2.	Are the spatial and temporal scales of the modeling approaches adequate to answi
the principal study questions? If not, what levels of spatial and temporal resolution
are required to answer these questions? What supporting data are required for
calibration/ validation of these spatial and temporal scales?
3.	It is contemplated that PCB concentrations in fish will be estimated using several
modeling approaches: an empirical probabilistic model derived from Hudson River
data, a steady state model that takes into account mechanisms of bioaccumulation
body burdens, and a time-varying mechanistic model (not included in the PMCR). .
bi-variate statistical model may also be used to provide insight into accumulations.
This multi-model approach is being contemplated because of the uncertainties
associated with any individual model. Is this a reasonable approach or should
predictions be made using a single "best" model?
4.	Is the level of process resolution1 in the models adequate to answer the principal
study questions? If not, what processes and what levels of resolution are required
answer these questions? What supporting data (such as data to support
specifications of a mixed depth layer, solids and scour dynamics, groundwater inflo
etc.) are required for these processes and levels of resolution?
5.	The results of the modeling effort will be used, in part, to support human and
ecological risk assessments. In your judgment, will the models provide estimates
adequate for this purpose?
1. The "level of process resolution'' refers to the theoretical rigor of the equations used to describe the various process
affecting PCB fate and transport such as: settling, resuspension, volatilization, biological activity, partitioning, etc. An
example of low process resolution is use of a constant value for the solids resuspension rate. A higher level of process
resolution is use of a complex mathematical description of the physics involved in remobilizing bedded sediment particl
(such as cohesive forces, bed shearstresses, etc.)
ii

-------
D.	Are there any changes to the work effort outlined in the revised work plan that would
significantly improve the outcome?
E.	In terms of evaluating the overall and specific effects and behavior of PCBs in the Hudson
River, are there any serious flaws in the modeling approach (theory, structure, physical
parameters, etc.) that would limit or invalidate any conclusions or further work based upon
the results of these models?
iii

-------
Peer Reviewer Comments

-------
Ellen Bentzen

-------
Ellen Bentzen
Ellen Bentzen has a Ph.D. (1990) and an M.Sc. (1986) in aquatic ecology from the University of
Waterloo and a B.Sc. (1982) in limnology from McGill University. For the past eight years, she
has worked as an applied aquatic ecologist/environmental toxicologist Research Associate at
Trent University, Peterborough, ON. Her initial research project at Trent was a study of how
aquatic food web structure influences the concentration of persistent organic pollutants (POPs)
in lake trout from Ontario lakes. This work instigated a number of related projects ranging from
field studies of POPs in the lower part of aquatic food webs and food web structure to
development of contaminant bioaccumulation models (ongoing research). She also has
examined the role of food web structure and dissolved organic carbon in lake water on
bioaccumulation of mercury in lake trout and other fish species. Distinct from these projects,
she also has been collaborating with scientists from Texas A&M University on studies of
microbial nutrient dynamics in the Sargasso Sea near Bermuda.
Ellen Bentzen is associated both with the Environmental Modeling Centre at Trent University,
Peterborough, ON, working with Dr. Don Mackay, and with the St. Lawrence River Institute,
Cornwall, ON (affiliated with the University of Ottawa, Ottawa, ON) working with Dr. David Lean.
Both the Environmental Modeling Centre and the St. Lawrence River Institute have associations
with industry and government agencies. She is currently completing a research paper which is
a review of POPs in Lake Ontario biota for submission to Environmental Reviews. This paper
includes temporal data for a number of organisms from Lake Ontario and an assessment of
recent trends in contaminant concentrations. This review is under contract for the Canadian
Chlorinated Chemical Council. A companion review is being prepared for POPs in St.
Lawrence River biota. She also has been collaborating on the development of contaminant
bioaccumulation models both for benthic invertebrates and for lake trout residing in different
aquatic food webs. Results from her research have been presented at Society for
Environmental Toxicology and Chemistry (SETAC), International Association for Great Lakes
Research (IAGLR) and American Society of Limnology and Oceanography (ASLO). Recent
research papers include: The role of atmosphere in Great Lakes contamination; Nutrient-
limited bacterioplankton growth in the Sargasso Sea; Role of food web structure on lipid and
bioaccumulation of organic contaminants by lake trout (Salvelinus namaycush); and Size-
structure and species composition of plankton communities in deep Ontario lakes with and
without Mysis relicta and planktivorous fish.
2

-------
Ellen Bentzen
Peer Review of Hudson River PCBs: RI/FS Phase 2 Reports: Preliminary Model
Calibration Report: Ellen Bentzen, Environmental Modelling Centre, Environmental &
Resource Studies, Trent University, Peterborough, Ontario.
General comments to the PCMR:
There is a tremendous amount of important work described in the preliminary model
calibration report (PCMR). A common problem of large projects entailing many collaborators
(especially when from several agencies) is the project gains momentum in several directions and
which then becomes a significant challenge to channel it into a single, cohesive directive. This
appears to be the case with parts of this Hudson River assessment. Specifically, it is not clear
how some of the various "submodels" or compartments will be unified, particularly with the goal
of identifying future trends. There is an indication (although this may be an artifact of the
somewhat unclear presentation of information) that a rigorous scientific protocol was not
adopted which would a priori establish the appropriate tests or hypotheses, determine the best
approach to examine these hypotheses and then follow by collecting the appropriate data. For
example, despite the Phase 2 sampling was identified as a large effort, quite often the calibration
of the models in the PMCR were hampered by the lack of appropriate data. A specific example:
the upper Hudson flow calibration had water data for 6 of the segments and these appeared to be
at limited times (low flow periods: however, the graphs were difficult to read clearly).
A problem with the PMCR report was the overall presentation. The organization was difficult
to follow, and while it was explained that this is a standard format, it prevents an efficient and
effective review process. Specifically, the style of introducing the different models in Chapter 3
(which were not completely presented), then following with the calibrations in later chapters
made it necessary to return to earlier passages because it is not easy to remember all the salient
information when one is not intimately familiar with the models or the system. Sources for
many details and important assumptions used in the model development were identified as grey
literature and inaccessible to external reviewers. It would have been useful to see more of the
actual data, and as the data were summarized in a report released in 1997, this should have been
possible. Figure and table titles did not identify all the pertinent details in the figures or tables:
these should always be able to stand on their own with clearly identified contents. Units were not
consistent throughout the report. On a positive note, chapters 8,9 and 10 were fairly well
written. It is noteworthy that these sections had numerous references to published studies.
Note: use of the word "METHODOLOGY". This is NOT synonymous with "method" but refers
to the study of processes, just as BIOLOGY is not synonymous with biota (try looking up
methods and methodology in the word perfect thesaurus)!
Questions to reviewers;
A. Is EPA using appropriate models, datasets and assumptions on which to base a scientifically
credible decision?
This question is difficult to evaluate for several reasons. The revised scope of work for the
Baseline Modeling identified that the models presented in the PCMR are being updated, and
these updates include applying the HUDTOX model to a new segmentation scheme, further
characterization of the Thompson Island Pool (TIP), modifications to the Lower Hudson River

-------
Ellen Bentzen
Model, and use of further data, etc. Based upon the descriptions in the PCMR, the Upper
Hudson River Mass Balance Model (HUDTOX) and TIP models do appear to be conceptually
appropriate, however, the mathematical relationships were not given for the HUDTOX model
nor were all assumptions identified; references to grey literature are not useful (e.g. Ambrose et
al.; formulations in Thomann et al. 1989, etc). The fish body burden models as described in the
PCMR may not be adequate to base scientifically credible decisions because they lack true
predictive capacity (see below for further comments).
Datasets: The data used in various parts of the PCMR were not all identified or described. Data
were available from the USEPA, NYSDEC, G.E., and USGS and also from private and academic
studies. The Hudson River Database was described as very extensive, including 750000 records,
with nearly half collected by the USEPA Phase 2 work plan. What constitutes a record? Are
these 750000 data points (excluding the date, time, depth, etc, information)? The review process
would have been facilitated by better understanding of the available data, especially the Phase 2
sampling program. How was the sampling program designed? There is an impression that
sampling was ongoing at the same time that the models were being developed, instead of model
development identifying sample needs. For example, 13 spatial segments were identified to
represent the Upper Hudson, but water data were only collected from 6. The Phase 2 flow-
averaged data appeared to be collected only during the low-flow event period (section 4.7.2).
Granted, as this report is already two years old, much of our information is out of date at time of
review.
B. Will the models, with the associated data sets and assumptions, be able to answer the
following principal study questions as stated in the PCMR:
1.	When will PCB levels in the fish population recover to levels meeting human health and
ecological risk criteria under NO ACTION?
The current models do not incorporate any temporal component, nor are the fish body burden
models actually predictive. An integrated, time-dependent environmental fate and food chain
bioaccumulation model, such as described by Gobas et al. (1995: EST 29: 2038) is needed to
simultaneously characterize the time response of PCBs in water, sediments, and biota to
predicted changes in PCB concentrations in the river (from a modified HUDTOX model with
temporal component). This will be more complex for the Hudson River because PCB declines
may be more variable over time than in a lake such as Lake Ontario. Loadings from the
sediments likely are more important than from a deep lake, and also more variable following
high flow events.
An important variable to determine is the possible rate of PCB decline in the water column and
interaction with sediments. While concentrations remain as high as shown for 1993, fish
concentrations will also remain elevated.
2.	Can remedies other than NO ACTION significantly shorten the time required to achieve
acceptable risk levels?
No information was provided to discuss possible remedies. It would be preliminary to offer
conclusions until the complete mass balance for the river is finalized and all sources and sinks
for PCBs identified. However, remedies such as dredging can have disturbing consequences to
4

-------
Ellen Bentzen
the ecosystem, which would need to be characterized independent of the PCB fate models.
Acceptable risk levels need to be identified in this context: acceptable to human consumption
or to wildlife. If the latter, levels which are 'not safe' need to be identified.
3. Are there contaminated sediments now buried and effectively sequesteredfrom the food
chain which are likely to become "reactivated" following a major flood event, resulting in an
increase in contamination of the fish population?
The results from the TIP model would suggest it is possible that buried, contaminated
sediments may be sufficiently disturbed by flood events to result in increased loading of PCBs
back into the water column (resuspension). Different severities of flood events may be
examined (e.g. 5 year or 100 year floods). However, whether the resuspended material is
bioavailable to the food chain is not identified. The preliminary conclusions that even a large
flood contributes a small percent of the PCB load is intriguing, but is this reconciled with the
implication that TIP contributes 50% of the PCB load down river?
C. Specific questions:
1. Are the modeling approaches suitable for developing quantitative relationships between
external forcing functions and PCB concentrations in the water, sediments andfish? Are the
models adequate for discriminating between water related and sediment related sources of
PCBs?
There are a number of issues to address with this question. The sources of PCBs to the
river have not been fully quantified or identified. For example, what is the source of PCBs
from the Mohawk River (17% loading into the Hudson). Atmospheric loading has not been
quantified. It was mentioned there is 'much uncertainty in the tributary PCB loadings for the
HUDTOX calibration period because of.... just 6 sample collections over the 9-month 1993
Phase 2" sampling. Have further samples been collected and if so, do these change the
contribution of the Mohawk? If the Mohawk has no known sources for PCBs and the 17%
estimate inaccurate, thus all the results may have high error. 74% of the loading was across
the Fort Edward boundary, presumeably from G.E. I find it curious that no data was collected
from a segment uptream from Hudson Falls, to characterize base loading of PCBs (this would
largely be a measure of atmospheric deposition onto the river and the watershed if there are no
known industrial sites). What is the loading source of PCBs from GE? Is this from
contaminated sediments near the sites, or seepage from inadequate containment?
—Section 4.4.2: Solids & PCB loads: TSS and flow and PCBs and TSS were correlated with
the USGS and USEPA data but not GE data, the latter which were quite variable. Does this
suggest that the GE data are less reliable (perhaps because of differences in analytical protocol,
p. 4-3?). This also seems to be the case for the sediment data discussed on p. 4-13. How do
these influence the model estimates? (Note also that no descriptive statistics for these
correlations were given).
-Section 4.4.5, p. 4-13: If the estimate of PCB congener #4, derived as 78% of the estimated
sum of PCB congener BZ#4+10, which in turn was estimated as double the GE Peak #53
concentration, was sometimes greater than the estimated total PCB concentrations, this casts
doubt over the calibration process used to convert any of the GE data to be comparable to
Phase 2 data. Or, the ratio of BZ#4 to BZ# 10 in sediments is not the same as for water?
5

-------
Ellen Bentzen
These are some factors which influence quantifying relationships for PCBs in water and
sediments. Discriminating between water and sediment related sources for fish PCBs cannot
be adequately addressed by the described fish body burden models. A bioenergetics based
bioaccumulation model is required to identify relative contributions of water and sediments to
fish PCB burdens. This point is elaborated below (Question E).
2.	Are the spatial and temporal scales of the modeling approaches adequate to answer the
principal study questions? If not, what levels of spatial and temporal resolution are required to
answer these questions? What supporting data are requiredfor calibration/validation of these
spatial and temporal scales?
As already discussed, no temporal scales have been incorporated into the models in the
PCMR, although the revised scope for baseline monitoring indicates this is underway. Overall,
the spatial scales are well delineated, although further resolution in Thompson Island Pool is
planned and should be useful. However, if the data are not available to match the model
segments, further detail in the model cannot be validated. As discussed previously, the model
segmentation and sampling (Phase 2) were not well matched for the HUDTOX model, both in
terms of lack of available data in all segments and characterizing the flow/low flow events.
The principal study area for the Upper Hudson River begins at Fort Edward. Why are there
not data collected upstream from Hudson Falls? These would be useful as baseline data (and
would incorporate the contribution of atmospheric loading which currently was unknown and
data from Green Bay was used). It was noted on page 4-10 that this is indeed a source of
uncertainty in the HUDTOX model.
3.	Estimating PCB concentrations in fish using several models: empirical probabilistic model,
steady-state bioaccumulation model (not described in the PCMR), time-varying mechanistic
model (not included in the PCMR); the bivariate statistical model may be used to provide
insight. Use of multiple models may be useful because of uncertainties in any one model. Is
this a reasonable approach or should predictions be made using a single "best" model?
It is a useful exercise to compare the results of different models because they differ in their
assumptions and in the constant parameters used. However, it should be established a priori
the purpose of each model, how each model compares and differ in their assumptions and
parameterizations and their limitations. Also, available data need to be considered; the best
model may not be useful if the data are not adequate. Otherwise, if a number of models are
run and produce different outcomes, then post priori explanations may not be adequate.
4.	Is the level of process resolution in the models adequate to answer the principal study
questions? If not, what processes and what levels of resolution are required to answer these
questions? What supporting data are requiredfor these processes and levels of resolution?
All assumptions and parameters used in these models need to be clearly identified. For
example (but not inclusive), growth and respiration rates for white perch and striped bass, fish
lipid content*, bioconcentration factors, gill efficiency transfer, oxygen consumptions, PCB
excretion rates, PCB loss rates across gills, chemical assimilation efficiencies, feeding rates,
estimates of dissolved, available PCBs vs total water column PCBs & role of water column
carbon-based solids in reducing bioavailability of PCBs
6

-------
Ellen Bentzen
p. 8-9: there is some confusion here (as elsewhere)- "variability may also reflect differing lipid
compositions, with correspondingly different rates of uptake of lipophilic compounds, between
different fish species". Concentration of PCBs in the food, PCB assimilation efficiency, food
assimilation efficiency, and bioenergetics of the individual fish species (and hence
physical/chemical factors which influence bioenergetics): these factors influence rates of
uptake of PCBs. FAT content of the fish influences primarily the rate of depuration (loss) of
the contaminant. Hence, fatter fish retain PCBs to a greater extent than skinny fish.
"Note: no where is there a discussion of fish lipid estimates and yet this is a parameter
which is highly critical to both the fish body burden approaches discussed here (this is a
serious problem but which is rarely addressed elsewhere as well). How was lipid estimated,
using what solvents? In addition, no where was it described how lipids (and PCBs) were
measured in the fish species: are these whole fish or fish muscle (skin on or off). Whole fish
versus muscle can have a consequence when comparing among fish because some fish store
fats in muscles (such as salmonids) while others store them predominantly around viscera
(such as pike). What solvents were used in the extraction of PCBs and lipid? Randall et al.
(1991, 1998) demonstrated huge and significant differences in lipid estimation depending upon
choice of solvents and 1 have noted this in comparing Lake Ontario lake trout estimates
between analyses conducted by American or Canadian agencies. Extraction procedures also
influence PCB estimates. Any QAQC (quality assurance quality control) done with other
laboratories?
Section 4.7: Calibration Results;
The constant solids gross settling velocity of 2.0 m/day was chosen but not very well
justified (this includes the response to the comments). Also, (p. 4-19, top) long term solids
deposition rates ranged from 0.5 to 5 cm/year, but a value of 0.22 cm/yr was used in the model
for burial velocity. The derivation of this value was not clear.
D. Are there any changes to the work effort outlined in the revised work plan that would
significantly improve the outcome?
Some aspects of this question have been addressed above. Another proposed effort is to use
the Gobas food chain bioaccumulation model, which is a valid idea (although note that model
has assumptions which are not all correct, either; these, however, may be specific to the Lake
Ontario ecosystem). The description of the Gobas (1993) food chain model (p. 8-2 and in the
revised scope) is slightly confused. The model does not "focus specifically upon food digestion
and absorption in the gut". This latter mechanism refers to another paper by Gobas which
describes how food is digested and contaminants absorbed against a fugacity gradient and which
results in biomagnification of the contaminant in the predator organism relative to their prey,
hence the fugacity of the predator exceeds prey (=biomagnification). A key factor of the Gobas
food chain bioaccumulation model which is not incorporated into the probabilistic model is that
the food web is run as a unit which combines the toxicokinetics of PCB uptake, elimination and
bioaccumulation in individual organisms with the trophodynamics of the food webs. Thus any
changes in the food web structure or physical characteristics (due to changes in population
abundance, diet of predators, temperature, etc) will modify the response in other members.
Pieces of this approach are suggested in the probabilistic model but the final model was not
described thus it is not possible to evaluate the effectiveness of the model. Note the 1993 Gobas
7

-------
Ellen Bentzen
food chain model is available in an easy to use format (Visual Basic, with plug in values for diet
composition, lipid content, number of predator and prey species in the food chain, plus water &
sediment concentrations of contaminants, etc) and produces lovely graphs of the outcome, but it
is his more recent temporally-based model (1995) which may be of more use to adapt to the
Hudson.
E. In terms of evaluating the overall and specific effects and behaviour of PCBs in the Hudson
River, are there any serious flaws in the modelling approach that would limit or invalidate any
conclusions or further work based upon the results of these models?
Calibration of the HUDTOX model: (p. 4-20) The goodness-of-fit test results for both TSS
and PCBs are curious. Looking at the predicted vs observed plot suggests a positive bias; this is
also shown for the means for each segment shown in Table 4-14. I perhaps have missed
something in the derivations of these data: why does the model have high variance? In theory,
(not including Monte Carlo or sensitivity analyses) the model has no variance. Would this not be
variability on a temporal scale as opposed to error, while the variance associated with the
observed data is from repeat measures for that segment. Therefore, the variability estimated with
the data are different from the variability used for the model and a t-test is not appropriate. The
regression plot of predicted vs observed is the most useful comparison.
Cesium was used as a model calibration tracer for settling solids and resuspension velocities
in the Lower Hudson River Model (Thomann; Chapter 7). No calibrations were identified for
the upper River models.
Food chain models (p. 7-5): Zooplankton growth and respiration rates were based upon
published data for Gammarus, which is a macroinvertebrate.
Component analysis: the data shown in Table 7-2 are not clear. Specifically, loss rates of
PCBs exceed total uptake in Segments 3, 4 & 5, which is not possible. This suggests a failing of
the model altogether, or some pertinent information have not been included in the text to help me
to understand this table.
The Bivariate statistical model for fish body burdens is described as a statistical model with
two independent variables, water and sediment PCB concentrations. However, even if water and
sediment concentrations are not in equilibrium, they are not independent of one another and
hence the bivariate model is not based upon independent variables. This has both important
ramifications both statistically and mechanistically for the interpretation of sediment vs water
contributions to the fish body burden of PCBs. An important assumption of multiple regression
models is that the independent variables are independent (not correlated). If they are correlated,
then this affects the regression coefficients because they explain overlapping information.
Because both sediment and water concentrations drive concentrations in the food chain, it is
perhaps not surprising that the bivariate model was able to roughly estimate the concentration of
PCBs in the fish but the unexplained variability is still quite large. It might be expected, for
example, that as older, larger fish with more complex diets are examined, the difference between
estimated concentration from the partial bioaccumulation factors derived for water and sediment
and observed concentrations would increase. And yet, for human consumption purposes, it is
desired to predict body burdens in the larger fish with greater confidence. As stated previously,
deriving the relative contribution from sediment vs water sources needs a bioenergetically based
food chain model. Also note that the relative contribution of prey items in a pelagic predator that
are described as water-based (eg yellow perch consumption of small fish; young yellow perch
consuming pelagic zooplankton) does not follow that water concentrations are the driving force
8

-------
Ellen Bentzen
behind the contaminant burden in the pelagic predator. Gobas (1993) noted that sediments are
the main source for contaminants for pelagic salmonids in Lake Ontario because of a benthic link
at the base of the food web. We also have suggested that the sediment-link is important for lake
trout PCB burdens after examining how food web structure influences lake trout PCBs (Bentzen
et al. 1996). Specifically, presence of the sediment-pelagic invertebrate, Mysis relicta, results in
substantially elevated PCBs in lake trout, whether or not they directly consume mysids. This is
an important and interesting aspect of that temporal changes in water concentrations might be
quite rapid once the "tap is turned off but is followed by relatively slower changes in sediment
concentrations. This is another point which should be addressed for the river, to predict what the
temporally changing relationship between sediment and water concentrations will be (fugacity-
based models would be very useful here).
More general comments (not in any particular order of importance):
-Figure 3-9: details of these maps are not identified (e.g. what is floodplain vs the TIP channel?)
-presentation of Figures 4-63 and 4-64 are somewhat confusing.
-it is not easy to differentiate the data shown in Fig. 4-10, Segment 12.
-units of both ft/day and cm/yr given for solids settling & sedimentation velocities... use standard
units!
-the description of the data in Tables 9-9 and 9-10 are obtuse. For example, the coefficients are
the partial BAF's and the r2 is from the predicted vs observed fit. This information should be
identified on the tables.
-Figures 9-8 to 9-13: the 1:1 line should be shown, and include the r2 and the probabilities, on the
figure. Some of the predicted/observed fits show a lot of scatter. A relatively high r2 can still be
insignficant.
-the fraction of organic carbon in the suspended solids (phytoplankton, zooplankton) is probably
relatively constant; were data not available from Thomann's modeling in the lower river, or use
the study of Cole et al. which produced an average estimate of 40%? (note this value was used in
the other Upper Hudson River model).
-the statement on p. 8-17 (bottom), that for the lighter chlorinated congeners, bioaccumulation is
driven primarily by direct uptake from dissolved phase, but food consumption is more important
for the higher chlorinated congeners. Note that other studies have documented this observation,
-note that forage fish is not synonymous with planktivorous fish. Forage fish refers to prey fish
for the piscivores.
-many layers to the sediment based upon GE sediment core data, but not used?
-Calibration of the Upper Hudson River PCB Model: chapter 4, fig. 4-1: what do the open and
closed circles and diamonds represent in the TSS and PCB panels? These should be identified on
the figure. It might be useful to see either an annual average (or warm seasonal average) plotted
against year. What are the set of data at 5 ng/L when values were in excess of 1000 ng/L in the
early years? Why are there many data equal to 100 ng/L? Any comments about the differences
in estimated water concentrations by the different surveys?
-Huestis et al. (1996) compared historical data for PCBs in Lake Ontario lake trout to reanalyzed
values on archived fish and observed a reasonable agreement between the annually estimated
historical values and the values based upon newer protocal (congener-based, etc). Did any
agency archive samples that could be reanalyzed to validate the conversions used to standardize
the Phase 2 data with the NYSDEC data? Note that other aspects of analytical protocol,
including but not inclusive, fish size, fat content, solvents, etc, also will influence temporal data.
Population characteristics, including population abundance, influence reproductive strategies and
9

-------
Ellen Bentzen
success; population characteristics are influenced by fishing and other predation pressures and
competition for food resources.
RECOMMENDATIONS
Overall, I recommend that modelling effort for the Hudson River PCB Reassessment RI/FS is
acceptable with major to minor revisions. This latter is somewhat widely interpreted because
my recommendation is based upon the PC MR (the models described in which are still
preliminary but essentially sound as such) and the promise of what is currently underway
(Revised Scope for Baseline Modelling).
To facilitate the review process:
•	provide the data for evaluation alongside the models
•	make a summary list of available data (dates, locations, compartments analyses, analysis
methods, etc)
•	allow greater lead time for the review process; it is rather curious that we were given one
month for a report that has been available for 2 years. All of us have multiple
commitments and scheduling time is not easy without adequate notice for such a major
effort.
•	there is a substantial amount of detail in this report; I could have used more time in the
evaluation. Time was partly "lost" because of the difficult format of the material. It
would be more efficient to have all the material for each model as a complete package
(as scientific work normally is presented), and then a summary chapter identifying how
the models will be linked together (a flow-chart or diagram may help). The jumbled
together format of the PCMR made it more difficult to retain all the necessary
information during subsequent sections
For the models:
•	a review of the literature seems warranted. While there is limited work on contaminant
fate models for rivers, there are nonetheless several publications (e.g. Larsson et al.
1990; Rice and White 1987; Chevreil et al. 1987, work from Green Bay and the Fox
River, WI). There is also a wealth of information based around the Great Lakes, with
several other time-dependent models, sediment/water PCB exchanges, role of
atmospheric deposition in mass balance, etc, etc.
•	for each model/section of work, identify a list of specific questions to be addressed,
assumptions made, available data, wish-list for data or information, concerns about the
model formulation, calibration, etc. Having this neatly organized would more readily
generate specific contributions from reviewers and other collaborators; workshops and
seminars are also useful.
10

-------
Miriam Diamond

-------
Miriam Diamond
Miriam Diamond holds a B.Sc. in Biology from University of Toronto, an M.Sc. in Zoology from
University of Alberta, an M.Sc.Eng. in Mining Engineering from Queen's University, and a Ph.D.
in Chemical Engineering and Applied Chemistry from University of Toronto. She has worked as
a limnologist with the Ontario Geological Survey and a private consultant with several
governmental agencies. She has over 15 years of research experience in the areas of water
quality and chemical fate and transport.
Miriam Diamond is currently employed by the University of Toronto as an Associate Professor
in the Department of Geography with cross-appointment to Department of Chemical
Engineering and Applied Chemistry. Her main research focus is the fate and transport of
organic and inorganic contaminants in aquatic systems, and more recently urban areas. The
research methods used vary from mathematical modelling to laboratory and field investigations
from the Arctic to the southern United States. A second research area is Life Cycle
Assessment and its application to site remediation options. She has sat on the Technical and
Publications Committees of the Association for Great Lakes Research, is a board member of
the Canadian Environmental Law Association, member of the review committees of the Niagara
River Remedial Action Plan and Lake Michigan Mass Balance program, vice-chair of the
Canadian Standards Association (CSA) Committee on the Life Cycle Impact Assessment of the
Pulp and Paper Production Phase and member of the Canadian Raw Materials Database
Technical Committee (also CSA). Along with her graduate students, Dr. Diamond has
published over 30 papers and reports in the scientific and technical literature.
12

-------
August 21,1998
Peer Review of Hudson River PCBs
RI/FS Phase 2 Reports
Preliminary Model Calibration Report
Miriam Diamond
Department of Geography
University of Toronto
General Comments
This review concerns three components of the U.S. EPA's effort to address elevated PCB
concentrations in the Hudson River: Preliminary Model Calibration Report (PMCR), the Revised
Scope of Work for Baseline Modeling Report, and the Responses to Selected Comments to the
PMCR. According to the "Charge to Reviewers", my comments are directed towards the main
question of whether the approach being taken will yield scientifically credible conclusions upon
which to base management decisions. Generally, the approach being taken appears to be sound with
several caveats and qualifiers discussed below.
First, I would like to couch my comments within the context of the review process. It is difficult to
obtain a clear understanding and overview of the models, assumptions and calibration results from
the reports. As several of the "Selected Comments" pointed out, the PMCR is a difficult document
to digest. This is particularly true in the absence of data reports and a clear summaiy that places all
work in the overall scope of the Hudson River effort. I suggest that a more productive review
process should incorporate oral presentations of the work to be reviewed along with interactions
among reviewers, modellers and regulatory officials, from which we would develop our
commentary. Receiving all materials well in advance of the review would be beneficial.
Specifically, the following would aid in understanding the process and approach:
•	a graphical presentation of the modeling components;
•	timelines for each component;
•	a matrix of the consultants involved along with their responsibilities and indications of the
cross-overs between models (e.g., model outputs that become inputs of the subsequent
model); and
•	a table listing the data available, time and conditions under which data were collected,
agency involved and annotations (e.g., QC/AC of chemical analyses, supporting data for
PCB fish concentrations such as size class and sexing).
Other specific criticisms of the documentation are:
•	too much reliance on the gray literature rather than concisely written and peer reviewed
scientific literature;
•	many unsubstantiated comments as pointed out in the "Selected Comments";
13

-------
Miriam Diamond
•	too much data with too little interpretation and clear summary of the main conclusions;
•	poor graphical presentation of the results;
•	poor representation of the geographic area and model segmentation (recommend a
comprehensive single map indicating the location of GE plants, tributaries, segment
boundaries, relative water and TSS loads from tributaries, and a second map indicating
sediment PCB concentrations); and
•	minimal comparison with other systems, e.g., what do other systems tell us about the
mobility and bioavailability of pollution?
A.	Is EPA using appropriate models, datasets and assumptions.
In terms of fate and transport, to which stage is this question addressed? The Workplan documents
states that the Lower HudsonRiver model is undergoing refinement and the HUDTOX model will
be applied to a new segmentation scheme. From these expanded plans, the EPA appears to be using
appropriate models with sufficient temporal and spatial resolution to address the question of
remedies. The models and their resolution also appears appropriate to the system and data.
B.	Will the models, with the associated datasets and assumptions, be able to answer the following
principal questions as stated in the PCMR:
1.	When will PCB levels in the fish population recover to levels meeting human health and
ecological risk criteria under No Action?
This question can only be answered using a fully time dependent bioaccumulation model that
includes fish age classes, a decoupled bioenergetic treatment of uptake and depuration, and
an accurate food preference matrix that allows discrimination of benthic versus pelagic
routes of chemical uptake. The statistical and empirical biotic models will not supply the
necessary time dependent information.
In order for a bioenergetically-based bioaccumulation model to answer this question, it must
rely on sufficiently spatially and temporally resolved input data, namely PCB water and
sediment concentrations. The spatial resolution must capture migratory and seasonal fish
distribution patterns (have not seen a discussion of this). The input must also capture
temporal patterns such that accurate water concentrations coincide with main life events such
as spawning, times of high activity, etc. Sufficient information is necessary to track changes
in dietary consumption patterns that may also require spatially and temporally resolved data.
2.	Can remedies other than No Action significantly shorten the time required to achieve
acceptable risk levels?
The first step towards answering this question lies in quantifying the source contributions to
14

-------
Miriam Diamond
PCBs in the river, specifically the "GE" versus "non-GE" PCBs. The contribution of"non-
GE" PCBs is not clearly discernable from the report. For example, p.2-4, #4. states that the
"principal external loadings" were 74% across the upstream boundary at Fort Edward (this
presumably includes the GE contribution?) and 18% from the Mohawk River (is there a GE
contribution to this 18%?). Moreover, of the GE contribution, how much is currently
seeping into the river, or are most PCBs found in contaminated sediments (i.e., what is the
specific source upstream of Fort Edward?). One can not answer management questions
without a sound understanding of the source terms since management actions differ widely
depending on the nature of inputs.
Secondly, it is difficult to answer this question without specific information on the options
under consideration. The next step is to evaluate whether the effect(s) of these options can
be simulated within the models. For example, if dredging the sediments of the TIP is
contemplated, then the model must simulate the disruptive effect of the dredging process
itself followed by the sediment reconsolidation. This simulation would require considerable
thought to reconstruct, but may be possible within the modeling framework.
3. Are there contaminated sediments now buried and effectively sequestered from the food
chain which are likely to become "reactivated" following a major flood, resulting in an
increase in contamination of the fish population?
This question can be restated more explicitly as: Can PCBs sorbed to buried sediment or in
the pore water at depth re-enter the water column at sufficient levels (mass) to affect
downstream concentrations of bioavailable PCBs?
There are two parts to the questions. The first refers to the physical process of sediment bed
resuspension and disturbance of the sediment profile. The TIP scour model aims to address
the question of sediment resuspension, with modifications to include cohesive and
uncohesive sediments. Insufficient details are available in the PMCR to determine the
efficacy of the model. I question the legitimacy of a steady-state assumption since
resuspension is event driven and dependent on antecedent conditions (e.g., consolidation
time). Will sufficient empirical work be available to adequately parameterize and validate
the model since resuspension models are semi-empirical and rely heavily on empirically
derived coefficients? What is the fate of the resuspended material? This question was raised
in the "Selected Comments" but not adequately answered. What is the fate of the PCBs in
pore water exposed due to sediment scouring?
The second part of the question concerns the bioavailability of the resuspended material. The
fate and transport model assumes equilibrium conditions to describe the truly dissolved,
DOC-bound and particle-bound fractions of PCBs. The literature contains several studies
that question this assumption, supported by findings of slow and biphase desorption kinetics.
This returns us to the question of the fate of the scoured PCBs. Will the resuspended PCBs
have sufficient time in the water column to re-equilibrate, including desorption to a
15

-------
Miriam Diamond
bioavailable form? Will the resuspended material scavenge dissolved PCBs from the water
column? The work of DePinto and Clarkson may be informative here. Similar questions can
be posed for the sediment bed - will recently exposed PCBs as a result of a major storm event
be more bioavailable than the previously weathered and armoured bed? How can the model
be used to address these questions?
C. Specific questions:
1.	Are the modelling approaches suitable for developing quantitative relationships between
external forcing functions and PCB concentrations in the water column, sediments andfish?
Are the models adequate for discriminating between water- and sediment-related sources
ofPCBs?
Again, there are several components to this question. The first concerns the interface among
models, as the question implies. The reports contain minimal information on how the
models will be interfaced. What criteria will be used to determine if the model results, when
coupled (e.g., fate and transport, intermittent sediment scour, bioaccumulation) are sensible?
How do errors and uncertainty propagate from model-to-model? For example, the
uncertainties in model estimates from the fate and transport model differ according to degree
of chlorination since the movement of lower chlorinated congeners is dominated by different
pathways (e.g., diffusion, air-water exchange) than the higher chlorinated congeners (e.g.,
sediment deposition and resuspension). With respect to fish, we are concerned primarily
with the higher chlorinated compounds.
A second concern is discriminating between water and sediment sources of PCBs. This can
be extended to the discrimination between "non-GE" versus "GE" PCBs (i.e., the source of
PCBs upstream of the northern river boundary is unclear to me and contributes significantly
to the overall budget). Is a method being formulated to make this distinction within the fate
and transport model? See Diamond (1995 Environ Sci. Technol. 28:29-42) for a discussion
of a modelling method that can be used). It may be desirable to employ a multivariate
statistical technique such as factor analysis or discriminate functions analysis, to tease out
the congener patterns according to medium. This method has been used to determine sources
of dioxins, PAHs, etc. This analysis should include the congener pattern of gas and particle
phase PCBs within the atmosphere.
The final part of the question relates to the model used to estimate fish concentrations.
Again, a mechanistic, bioenergetically-based model is necessary to discriminate between
water and sediment chemical sources.
2.	Are the spatial and temporal scales of the modeling approaches adequate to answer the
principal study questions? Additional data?
16

-------
Miriam Diamond
Since the time scale for the mechanistic bioaccumulation model is unclear, it is difficult to
answer this question. The temporal linkage between the fate and transport and mechanistic
bioaccumulation model is important given the seasonality in both. For example, can the
models address the temporal relationship between concentrations and critical life events?
The greatest PCB concentrations and loadings occur during spring freshet. Does this
coincide with spawning or a time of important dietary changes? I have discussed other
aspects of this question under B1.
There is concern that the water and sediment sampling regimes do not coincide with the
model segmentation (question of spatial resolution). Temporally, there is some concern
about the data collection adequately capturing storm events (which are difficult to capture)
versus "average" events. The importance of storm events, or at least high flow conditions,
is apparent from the results presented in the PMCR. In terms of sediment resuspension and
bulk PCB movement, adequately characterizing storm events is clearly critical.
Some additional data should be collected from the unmonitored tributaries to verify that their
assumed hydraulic, TSS and PCB contributions are accurately estimated.
Overall, the degree to which the models can be validated raises some concerns. I suggest
that the models be calibrated and tested with chemicals in addition to PCBs to increase the
rigor of the testing procedure. I have elaborated on this below.
Plans to run HUDTOX for decadal-scale periods and testing hindcasts from 1977-1997 are
certainly necessary. The success of this effort rests, in part, on "normalizing" PCB
measurements relative to the analytical methods used. This has been addressed in other parts
of the study.
3. Multiple versus single models to estimate fish concentrations?
As discussed above, a mechanistic bioaccumulation model based on bioenergetic
considerations is needed to obtain answers to the temporal questions being posed. Time
response information can not be obtained reliably using empirical or statistical models.
However, it is reasonable to use several models to estimate steady-state conditions and test
the mechanistic model, with the caveat that the agreement among these approaches does not
validate the time response information produced by the bioaccumulation model. Overall,
using multiple approaches to examine fish concentrations lends credibility to the modelling
process.
As I suggested for the fate and transport model, a more rigorous test of the models would
come from applying the models to other chemicals that differ in their main exposure route
(e.g., water versus food, pegalic versus benthic food chains) and time response.
17

-------
Miriam Diamond
4. Level of process resolution?
The fate and transport model could benefit from either the inclusion or improved
parameterization of several processes:
•	Air-water exchange rate: more accurate determination of congener abundances in gas
phase rather than assuming (erroneously) that the gas phase congener pattern is the
same as that in the water (p 4-11);
•	Unclear how the air-water exchange mass transfer coefficient is obtained and its
dependence on temperature and air and water velocity;
•	Sediment-water diffusion: whereas the treatment of diffusivities is explained, the
stagnant boundary layer estimate is not apparent but can potentially alter rates of
sediment-water diffusion by orders-of-magnitude, i.e., what is the stagnant boundary
layer thickness, how was it obtained? This is important given the uncertainty in rates
of diffusion from the contaminated sediments of TIP and the need to invoke an
explanation of groundwater discharge to reconcile observed concentrations.
•	Sediment-reworking: how is vertical reworking due to mechanical perturbation
treated? Presumably sediment scour can result not only in sediment entrainment, but
in mixing sediments vertically and horizontally. Evidence for this would come from
disturbed core profiles for which dating using, for example, Pb-210 or Cs-137, is not
possible. The effect of sediment reworking would be enabling interaction between
deep, formerly buried sediment and the water column. We have been examining this
hypothesis in a mercury contaminated system where, we suspect, sediment reworking
may be maintaining elevated sediment and water concentrations (Diamond et al., in
press).
•	Application of primary production estimates from the Lower to Upper Hudson River
sections. How valid is this? What level of uncertainty is introduced by this
assumption?
•	Groundwater discharge/recharge effects: groundwater discharge in TIP has been
invoked to account for observed elevated concentrations of lower congener PCBs.
The treatment of this process was substantiated by some hydrogeological
interpretation. Are plans underway to conduct more rigorous testing of this
assumption and look for other zones of discharge/recharge along the river? See
comments below. How does the incorporation of groundwater discharge affect the
water balance for which a surplus appears to exist?
•	Watershed export: the question of the proportion of PCBs contributed by atmospheric
deposition to the watershed followed by watershed runoff has not been addressed.
I also did not see a discussion of other potential sources. A simple calculation may
suggest that watershed export may contribute negligibly to the mass balance,
however it would be useful to explore this source term briefly.
•	PCB degradation in sediments: there is an ongoing controversy about the extent of
degradation through aerobic and anaerobic microbial processes. Degradation has not
been included in the fate and transport model presented in PMCR, but should be
1 8

-------
Miriam Diamond
included to improve the comprehensiveness of the model and to determine the
relative importance of this process, particularly over the decadal simulations.
5. Will the results from the modeling effort be adequate to support human and ecological risk
assessments?
Generally, the results of the extended modelling effort (including results obtained from
activities outlined in the "Revised Scope" document), appear to be adequate to support
human and ecological risk assessments as presently conducted by the EPA. However, the
key lies in the results of model testing and validation, particularly over the long term (e.g.,
the hindcasting exercise).
In addition to model validation, a critical component of the analysis that I have not yet
addressed, is how PCBs are treated, e.g., - Arochlor equivalents, specific congeners or
SPCBs. Modelling SPCBs will not provide sufficiently accurate information on fate and
transport nor bioaccumulation for risk assessment purposes. Justification is lacking on the
choice of PCB congeners and that those chosen do not extend beyond hexachlorobiphenyl
(why this limitation?).
D.	Are there significant changes to the work effort outlined in the revised work plan that would
improve the outcome?
See recommendations below.
E.	Are there serious flaws in the modeling approach that would limit or invalidate model
conclusions?
The following flaws should be rectified:
•	inclusion of segments upstream of GE contributions;
improved delineation of groundwater discharge and recharge zones;
•	justification of choice of PCB congeners for analysis, choice of their physical-chemical
properties, and lack of consideration of congeners beyond hexa's;
•	improved treatment of air-water exchange, including SPCB and congener air concentrations
and velocity-dependent mass transfer coefficients;
•	linkage between sediment scour and HUDTOX models to track the fate of resuspended
material;
•	determine bioavailability of resuspended PCBs by including sorption/desorption kinetics
rather than assuming equilibrium conditions for PCB partitioning; and
•	address bias in model predictions that lead to overestimations of TSS at low flows and
underestimation at high flows (p.4-20 focuses on the results of an overall T-test indicating
19

-------
Miriam Diamond
goodness of fit between TSS and discharge that belies the bias apparent in Figure 4-13).
Recommendations
Overall, I recommend that the modelling effort for the Hudson River PCB reassesment RI/FS is
Acceptable with minor revisions. This statement is based on the information provided, which, I
suggest, is incomplete, dated (PMCR) and difficult to follow.
In addition to recommendations stated above, I offer the following additional recommendations that
are intended to strengthen the approach taken:
•	re-design the spatial boundaries to include a segment upstream of the GE plants in order to
provide a "control" segment that is useful for understanding the contribution of "non-GE"
PCBs to the river (albeit at the upstream end);
•	consider estimating the contribution from atmospheric deposition to the watershed, input
through watershed export;
•	place greater emphasis on determining source contributions (e.g., "GE" versus "non-GE
PCBs, those in contaminated sediments versus those entering from land-based sources) (see
comments above);
•	justify the choice of PCBs modelled, e.g., particular congeners, how EPCBs are treated;
•	document information on the physical-chemical properties of EPCBs and congeners
modelled, including temperature corrections for vapour pressure and Henry's law constants;
•	place much greater emphasis on model testing and validation,
•	develop a method and decision process to judge the performance of each model and
the results obtained from models linked in sequence;
•	use of chemicals in addition to PCBs that span a range of physical-chemical
properties and hence pathways;
•	use of chemicals that have a strong time response for hindcasting (e.g., lead); and
•	use of chemicals with a long time record, thereby circumventing difficulties in using
historical and recent PCB analyses;
•	use differences in data sets (e.g., USGS versus GE differences in the relationship between
PCB concentrations and TSS) and the occurrence of outliers to probe the behaviour of the
system rather than discounting differences (discrepancies can tell us more about systems and
better test hypothesis than data that conform to predictions);
20

-------
Miriam Diamond
more sophisticated treatment of non-detects in the PCB data sets rather than setting non-
detects to zero or one half the detection limit (e.g., use of statistical methods such as least
squares estimators developed by El-Sharaawi);
test importance of assumption of constant DOC (does this imply temporal as well as spatial
consistency?); and
test the hypothesis of groundwater discharge as the mechanism responsible for contributing
"additional" PCB loads, e.g., piezometer studies (deployed in sediments, piezometer nests
in bank sediments to determine flow path, use of benthic chambers, measurements of stable
isotopes (0-18) to "age" the water).
21

-------
James Gillett

-------
James W. Gillett
James W. Gillett has a B.S. in Chemistry with Honors from the University of Kansas and a
Ph.D. in Biochemistry from the University of California-Berkeley. He has worked as an
environmental toxicologist in Entomology & Parasitology at the University of California and in
Agricultural Chemistry at Oregon State University, and then as a research
ecologist/environmental scientist at the U.S. Environmental Protection Agency in a nearly
four-decade career.
James W. Gillett is a Professor of Ecotoxicoiogy and director of the Cornell Superfund Basic
Research & Education Program (SBREP) at Cornell University. In addition to teaching courses
in environmental toxicology, ecotoxicoiogy, ecological risk assessment, and natural resource
management, he coordinates research on bioavailability in the SBREP, works on multi-pathway
exposure assessment, and is evaluating the impact of public participation on ecological
resources in site clean-ups. He has served on the Biotechnology Science Advisory Committee
(U.S. EPA), the NRC-NAS Committee on Prioritization of Superfund Site Clean-Ups, the New
York State Ecological Risk Reduction Task Force, and numerous other advisory panels for U.S.
EPA and other federal and state (Oregon and New York) agencies, non-governmental
organizations, and industrial associations.
24

-------
Comments on
"Preliminary Model Calibration Report"
and Hudson River Phase 2 Reports
James W. Gillett
Cornell University
A.	Is EPA using appropriate models, datasets, and assumptions on which to base a
scientifically credible decision?
The existence of the Hudson River Superfund site as a continuous entity 175 mi long on which
literally billions have been spent on a whole range of pertinent research and management issues creates
a mind-boggling challenge encompassing this lead question. The short answer is, 'Yes, but....", and then
comes the long answer. The most useful databases (consistency of purpose, acquisition and handling,
quantitation) are reasonably robust, yet still have huge gaps in temporal and spatial resolution. What is
the largest quantity of as-yet undetected PCB contamination which might be mobilized to make GE, EPA
and us look like fools? How do the various components (water column, DOC, interstitial water,
consolidated sediment, and non-consolidated sediment) relate to each other and measured values in
terms of bioavailable, slowly available, and non-available (fully sequestered) loads? The absence of
terrestrial data on biota and soils within the watershed assumes the outputs to and inputs from these sinks
(with their own external links and forcing functions) are not relevant or can be lumped at much coarser
levels of resolution. That the models discussed all assume these links to be negligible very much limits
long-term and long-range utility of the approach.
The models seem basically sound, but the use of four contractors with numerous technical tasks
illustrates the breadth of knowledge required to make any global statement about suitability. Use of a
combination of modeling approaches appears wise, but hardly fool-proof and ultimately unsatisfactory in
determining the likely outcome of management actions. Under EPA's new "Proposed Guidance for
Ecological Risk Assessment" [EPA/630/R-95/002B, Risk Assessment Forum, U.S. Environmental
Protection Agency, Washington, DC; see also Fed. Register 61:475552+ (Sept. 9,1996)], eco risk is
evaluated by simultaneously considering exposure and effects and their interactions. In a recent review of
EPA's new Multi-Pathway Exposure Analysis Methodology, it was clear that similar approaches will be
incorporated into human health-based assessments as well. The subject assessment does use part of
the multi-pathway exposure assessment (a good part of which was developed by various contractors and
researchers in the Hudson River system), but falls short of embracing a more holistic view and
methodology.
Finally, however suitable the models, etc., might be in regard to scientific credibility of a decision
(useful in a court of law or administrative action), that may have come so late in the process as to be
functionally useless for regulators, the regulated community, and various public interests inter alii. The
use of the models by scientists and engineers is fine; the other publics will probably not understand the
complexities and uncertainties.
B.	Will the models, with the associated datasets and assumptions, be able to answer the
following principal study questions as stated in the PMCR:
1. When will PCB levels in the fish population recover to levels meeting human health and
ecological risk criteria under No Action?
Increasingly, this judgment relies upon congener-specific exposure assessment and toxic
equivalent functional analysis. For highly mobile populations of people and mink which may or may not
feed on the fish represented by Hudson River sampling schemes and models derived therefrom, it would
seem to be prudent to pay attention to congener-specific transformations as a part of chemodynamics and
bioaccumutation. We have almost no knowledge of congener-specific dose-response relationships in
25

-------
James Gillett
people, but we are making some inroads with mink. Nevertheless, at the rate of gain of knowledge about
the multi-faceted parameters of such interactions, we may best be served by looking to when mink can be
self-sustaining in the watershed. Then we might assume that the situation is safe for people as well.
When, if ever, can the models tell us this? First, the models need to be at least sensitive to congener
structure on a basis other than log Kow or molecular weight/chlorine number. This includes all
chemodynamic parameters (metabolism, uptake, storage and excretion) in each trophic level and
target The Thomann et al. (1989) model, central to many efforts, uses only homologue categorization,
without regard to chlorine atom placement and all the implications for effects on metabolism and toxicity.
Alternatively, more complete and persuasive evidence of homology of these parameters under differing
biotic and abiotic situations would be helpful. Second, they need to be more holistic in the species
addressed (inc. terrestrial), incorporating biology (e.g., feeding range, preferences vs availability) and
seasonality (winter depuration). Third, they need to be evaluated longitudinally with fish of known age and
residency, with particular attention to the nature and quantities passed on in the roe. Long-lived sturgeons
and species such as black and rock bass, for which there are already some data, might be helpful.
The difficulty right now is that the models present a sort of minimal upper bound of potential
exposure which is far greater than might be realized by a nominal subject (of whatever species), much
less than might be happening for the upper 99th percentile, and without a clear basis forjudging the merit
of a given management decision, e.g., No Action, removal, capping, etc.
2.	Can remedies other than No Action significantly shorten the time required to achieve
acceptable risk levels?
The reason that a demonstration of longitudinal validity in the modeling is so critical is that
assumptions about the long-term outcome of management actions remain untested. (The No Action
alternative, of course, is tested by long experience, almost three decades.) The peak loading of some
food web components will occur substantially after the primary known source(s) may be removed and
pathways eliminated, either because unknown and unmitigated sources are exposed by natural processes
(e.g., flooding), because of external inputs (via atmosphere and soil), or because of the lags in equilibrium
posed by chemicals such as PCBs and DDT-R. We still do not know if this long time scale problem is
attributable to slow release consequent to sequestration (short-term non-bioavailable, but chemically
detectable residues), cumulative impacts, or other mechanisms to be discovered. As it stands right now
our knowledge and assumptions are either in error or incomplete, or both, to a significant degree with
respect to long term consequences.
This set of issues becomes critical when amassed at the scale of the Hudson River. What might
work in theory at a smaller site may be fairly dubious at this grander scale. The first dredging of Foundry
Cove increased measurable Cd therein by about 25%! Rochester Gas & Electric hit the mother load' of
coal tars at their Genesee River site coincidental to other actions, necessitating a massive restructuring of
not just that effort, but of numerous MCG sites in New York State. If one multiplies the potential for
adverse outcomes of site remediation activities due to other pollutants in the watershed being disturbed
and/or redistributed consequent to dredging, for example, then this question demands more wisdom than I
believe we have.
An alternative to dredging could be capping or in-situ immobilization, or accelerated efforts at
in-situ bioremediation (sort of augmenting part of the No Action alternative). My understanding is that
these all have or are being applied in the Hudson Basin or elsewhere, but I have seen no estimates of
when or whether we "significantly shortened] the time required to achieve acceptable risk levels." There
have been more numerous instances of reported fraud and failure that success in this matter. In the St
Lawrence, the Canadian stretches of the river and the numerous downstream embayments have yet to be
addressed, in no small part be cause of the issues of scale, but also because of inadequate assessment
3.	Are there contaminated sediments now buried and effectively sequestered from the
food chain which will likely become "reactivated" following a major flood, resulting in an
26

-------
James Gillett
increase in contamination of the fish population?
The kind judgment would be that we just don't know, but in point of fact we are virtually clueless.
There is no reason for there not to be, because, as noted earlier, we need to define how large a load that
might possibly be. Could a PCB-based (or heavily contaminated) DNAPL be snaking its way through the
sediments between sediment cores? Are water column and sediment sampling methods robust enough to
tell us? There is an additional problem here. The mechanisms and models seem to be in place to
ascertain where flood-mobilized sediments will be deposited, but no effort in the examined documents
really spelled it out in a manner that would ultimately permit use in an exposure assessment model leading
to people or a critical ecological resource. The phrase, "increased contamination of the fish population," is
the generic type of statement used (incorrectly) to justify regulation of land-applied wastewater sludges,
assuming that erosion of deposited material makes contaminants available for bioaccumulation. If we
cannot attribute contamination in the long term to known sources, how will we know that an as-yet
undiscovered is affecting anything?
Please note that my use of the terms sequestered and sequestration may differ from that of the
authors of the PMCR. I refer you to Alexander [Alexander, M. (1995). A small circle of knowledge, a large
circle of ignorance. Environ. Health Perspect. 103(Suppl. 5):121-123 ], wherein materials are sorbed or
intercalated into mineral or organic matter in pores too small for microbial action and thereby limiting
back-diffusion (formerly called soil hysteresis). Some samples of experimentally DDT-treated soils have
sequestered non-bioavailable compounds for several decades. What fraction will be released with acid
rain? photoxidation of a carbon matrix? accidental exposure to another solvent of greater polarity and
smaller size? Will releases be bioactive (i.e., present in an effective dose to a vulnerable and valuable
receptor)? Almost all of the data in the PMCR et al. concerns chemically detectable residues;
sequestered material is not determined, say, by selective extraction or in-situ bioassay.
C. Specific Questions:
1. Are the modeling approaches suitable for developing quantitative relationships between
external forcing functions (e.g., hydraulic flows, solids and PCB loads, sediment initial
conditions, etc.) and PCB concentrations in the water column, sediments and fish?
Much of that is beyond my expertise. I had trouble with a number of the equations just in
identifying the components and what terms meant (e.g., the term m in Eq. 3-9,3-10; the third and fourth
terms in Eq. 3-11). [Why is there a term for "net sedimentation" on top of terms for "settling" and
"resuspension" in Eq. 3-11?. It kind of reminded me of my Russian language exam at Berkeley, which
was presented to me by the late Prof. Zev Hassid in Old Church Cyrillic on a subject matter out of step
with our understanding of chemistry. It really is hard to translate what you can't believe ] However, I am
fairly well versed in the modeling of relationships between water, sediments and biota. The Thomann-type
model is archetypical; many of us use it or suitable modifications. Theresulting variations may be
improvements, but typically only a little new information is brought to bear. Ram and Gillett (1992, 1993)
[Ram, R.N. and J.W. Gillett (1992). An aquatic/terrestrial foodweb model for polychlorinated biphenyls
(PCBs). IN: J. Hughes, W. Landis and M. Lewis (eds) Environmental Toxicology and Risk Assessment.
First ASTM Symposium on Ecological Risk Assessment, Atlantic City, NJ. American Society for Testing
and Materials, Philadelphia, PA. pp. 192-212; Ram, R.N. and J.W. Gillett (1993). Comparison of
alternative models to predict the uptake of chlorinated hydrocarbons by oligochaetes. Ecotoxicol. Environ.
Safety 26:166-180 ] estimated numerous parameters by analogy, bioenergetics, and simulation theory.
Interestingly, the natural history database upon which so much of the St Lawrence-Great Lakes effort
(Ram 1990) was based is horribly out-of-date, since professional journals and graduate programs stopped
accepting that sort of descriptive science as "state of the art." The Hudson has fared a little better, due to
the efforts of the Hudson River Foundation, beneficiaries of the court award derived from power plant
releases of PCBs into the subject body of water.
27

-------
James Gillett
The most important assumptions seem to be that PCB bioaccumulation has no significant effect
on mortality, morbidity, behavior or fecundity of fish or their prey. These outcomes may account for the
high degree of variability offish residues (e.g., lognormal distribution). Possible changes in behavior (or
lack thereof, such as attracting or repelling predators, perhaps by organoleptic processes) would seem to
be critical, but are undocumented. By the same token, underlying biochemical and physiologic shifts in
lipid metabolism and storage appear to have been studied a bit more thoroughly, albeit still inadequately.
Are the models adequate for distinguishing between water-related and sediment-related
sources of PCBs?
We believe they are (Ram and Gillett 1993), if one pays careful attention to fbodweb components
derived from one or the other. That is, one must distinguish between sediment and water sources both for
02 and PCB-contaminated food. Unfortunately, we were unable to do this on a congener-specific basis
because of a shortage of uptake and metabolism rates for specific congeners. Much data has been
generated in the interim and a much better job may be possible now.
2. Are the spatial and temporal scales of the modeling approaches adequate to answer the
principal study questions?
Sadly, no. The spatial scales in the figures laid out over the entire course of the river are simply
too coarse to protect us from "hidden" pockets and untoward shifts. There may be 20 yrs of data, but the
data are uneven, poorly correlated (to water, sediment, biota, organic carbon) and not as well conceived
as we have had in more recent times (i.e., since TSCA and CERCLA, regulating PCBs). Someone is
always having to make a correction that appears to the public or outsiders as a "fudge factor". All the
uncertainties pile up as ignorance and incompetence (Johnson & Slovak 1995). [Johnson, B.B. and P.
Slovic (1995). Presenting uncertainty in health risk assessment: Initial studies of its effects on risk
perception and trust Risk Analysis 15:485-494] How many factors modify the value at a given point in
time and space? What were they for each sample? Do we just pretend that they don't matter? One
approach is to do as was done in parts of the PMCR: get all the models you can to describe the
proverbial elephant. Even though the assumptions may be incompletely examined and tenuous, you can
see some of the convergent patterns. Thus far, however, even that approach does not inspire much trust
We are asked to wait for the hindcast, the next round of recalibrated samples, the new sampling
paradigm, and so forth. If an educated professional is confused, I would suspect the public to be very
much put off by this difficulty.
If not, what levels of spatial and temporal resolution are required to answer these
questions? What supporting data are required for calibration/validation of these spatial
and temporal scales?
The main object would be to sample until the increase in variability or inhomogeneity is negligible
(i.e., the derivative -> 0). A statistician then should be able to establish the appropriate power of a
sampling network in space and time in which the driving variables of the data distribution could be
ascertained. Without access to the original sampling designs and rationales, it is hard to know if this is as
good as we can get for the dollars available in relation to the attendant risks. With a high proportion of
non-detects the results are invariably contentious. Rather than assume some arbitrary value, why not use
the known portion of the data distribution to represent a lognormal distribution of the non-detects? On the
other hand there is more risk to a consumer of that one outlier fish than in all the non-detects. Therefore,
our regulatory goal would be to have no fish (prob. <01) presenting more risk than a population of fish
50% of which are non-detects. What power does the fish monitoring have to have to detect the likelihood
of the upper 99th percentile?
28

-------
James Gillett
3.	It is contemplated that PCB concentrations in fish will be estimated using several
modeling approaches: an empirical probabilistic model derived from Hudson River data, a
steady state model that takes into account mechanisms of bioaccumulation body burdens,
and a time-varying mechanistic model not included in the PMCR). A bi-variate statistical
model may also be used to provide insight into bioaccumulation. This multi- model
approach is being contemplated because of the uncertainties associated with any
individual model. Is this a reasonable model approach or should predictions be made
using a single "best" model?
As noted above, this is the wisest course, albeit not necessarily that which may find "truth". The
integration of the models is tricky, since they are each incomplete and in contradiction over some details.
The empirical probabilistic model can suggest overall or summary process rate constants, but probably
glosses over species- and congener-specific parameters. The deterministic/mechanistic model needs to
identify the range of possible parameter values and either generate a set of probabilistic outcomes for
individuals over those ranges or otherwise represent the spectrum of responses (especially that 99th
percentile). The realism of the models is not the same as the realism of the sample data, and likely
neither is "truth." Adam Finkel (1990) termed this "model uncertainty," in that we don't know which model
should be expected to not only represent the data but also to predict a range of outcomes from which the
risk manager must make a decision. The risk manager has to decide in advance what questions a model
may address effectively and efficiently. Therefore, if all models in hand are false, the search for a true
model is yet worthwhile. Ram & Gilletfs model (Ram & Gillett 1992) is true if you can surmise the history
of individual organisms sampled for residues, but we realized it was false for migrants, casual visitors, and
those with different lifestyles than whatever norm was expected of a species or population. The criteria
we used were established in advance: 'Criterion 1: 90% of the measured values for each species would
be within the 95% confidence interval of the log-normal distribution predicted, and 'Criterion 2: 90% of the
species with three or more specimens would meet Criterion 1. [Valid for 41 of 46 spp. or species clusters
derived from prey use.] These criteria held up for PCBs in the St. Lawrence Basin, eastern and western
Lake Ontario deepwater food webs, and each of the other Great Lakes, but did not account for growth
dilution and toxicodynamics in very large salmonids subjected to a variety of adverse conditions (alewife
die-off, hypo-biotinosis in lake trout, invading invertebrates, etc.).
If a combination of models can lead to the construction of a "best" model for its pre-determined
criteria, then working toward that goal is certainly the right course. I just don't want the present models
singled out for "Eureka! We've got it!"
4.	Is the level of process resolution adequate to answer the principal study questions? If
not, what processes and what levels of resolution are required to answer these questions?
What supporting data (such as data to support specifications of a mixed depth layer,
solids and scour dynamics, groundwater flow, etc.) are required for these processes and
levels of resolution?
One is tempted to believe that the seemingly robust datasets from which the PMCR is proceeding
or will have demand the highest level of process resolution. The biologists, I sure, would love to have the
physical environment "tamed" by high resolution equations, well-described parameters, and coordinated
datasets. The engineers and physical scientists would appreciate any improvements in the mathematical
organization of the biosphere. Except where remote sensing provides a continuum of data over large
areas, I'm not well impressed with our ability to get ground truth by sampling along 250-m transects and
such. Data resulting from a monitoring survey - "Are PCBs moving from the Upper Hudson to the Lower
Hudson?"- are inherently different from data based on a sampling grid derived from a high-resolution
deposition/scour model implying where PCB-bearing particulates might be trapped. Maybe, as was
suggested, intensive sampling over time using the high resolution model would answers the movement
question in a more definitive manner.
29

-------
James Gillett
The sediment deposition and depth of scour questions may be the only ones where the highest
level of resolution obtainable seems required. That would also add mechanical turbation (prop wash of
large vessels, episodic recreational boating uses) and bioturbation to the processes described. There is a
general tendency for CHCs to be degraded in inverse proportion to CI content or the log Kow and in
proportion to water solubility and frequency of sample presence. Three- to five-fold variation in metabolic
rate of a congener in or between species seems common, however. Slight shift in prey selection,
residency, and level of feeding might cause as large a change, although sensitivity analyses of individual
factors do not In any case the assumption that conditions continue from one season to the next and from
year to year is belied by the physical data. When you do a full physiologically based, pharmacokinetic
(PBPK) model, it is easy to represent all of the compartments in considerable detail. Even though the
PBPK model is regularly employed now in drug and carcinogen assessment quite a bit of utility is derived
from far simpler expressions (e.g., one- or two-compartment models).
5. The results of the modeling effort will be used, in part, to support human and ecological
risk assessments, in your judgment, will the models provide estimates adequate for this
purpose?
That*s an interesting question. The models certainly will assist enormously in planning the
assessment design and targets of impact. That will help the risk assessor describe issues for the risk
managers in terms which are more readily understood by the public. The absence of plant uptake,
terrestrial-aquatic linkages, and other inter-media transfers blunt the aforesaid usefulness, at least for
primary ecological risks. For health effects the models seem even less useful or more limited, depending
on your point of view.
D.	Are there any changes to the work effort outlined in the revised work plan that would
significantly improve the outcome?
The work needed on air pathways and plant uptake (both rooted macrophytes and riparian
vegetation) is considerable. There are few links to the terrestrial components, so almost all of the effort is
directed at the water column relationships. Most of the planned effort is promise as partial solutions to
present problems. That is easy to support. I'm not sure where the Gobas model fits in.
E.	In terms of evaluating the overall and specific effects and behavior of PCBs in the
Hudson River, are there any serious flaws in the modeling approach (theory, structure,
physical parameters, etc.) that would limit or invalidate any conclusions or further work
based on the results of these models?
For the most part I see no quarrel based on the models and their makers/users per se. The
tendency to use mean values as point estimates is not productive. As pointed out above, there are some
deficiencies and insufficiencies, but probably nothing radical. The argument that sequestered PCBs are
totally immobile denies their pervasive (as well as persistent) character.
Recommendations
1.	Acceptable as is
2.	Acceptable with minor revision (as indicated) /
3.	Acceptable with major revision (as outlined)
4.	Not acceptable (under any circumstances)
30

-------
G. Douglas Haffner

-------
G. Douglas Haffner
Doug Haffner received his B.Sc. In biology/chemistry at Queen's University, Ontario and
completed a Ph.D. at the University of London, England in aquatic ecology. He has worked
with the Ontario Ministry of the Environment, the International Joint Commission and
Environment Canada in the area of water quality management and monitoring. He has also
been the President of the International Association for Great Lakes Research, and is currently a
senior science advisor to the Indonesian Institute of Science (LIPI). He has chaired the Detroit
River Rap and is serving as chair of Contaminated Sediments Subcommittee on the Detroit
River
Currently at the University of Windsor, Dr. Haffner is a professor of biological science with a
research specialty in ecotoxicology. Dr. Haffner also served as director of the Great Lakes
Institute for Environmental Research. His laboratory is the only Canadian university-based
facility that is certified under the Canadian Association of Environmental Analytical Laboratories.
Research interests are the trophodynamics of PCBs and PAHs. His laboratory has developed
models of PCB dynamics in Lake Erie, Lake Ontario and the Detroit River. Other research
interests include direct assessment of human health hazards of chemicals in the environment
using in vitro assays and the effects of chemicals on amphibians.
32

-------
PEER REVIEW REPORT OF HUDSON RIVER PCB RI/FS PHASE 2 REPORTS
PRELIMINARY MODEL CALIBRATION REPORT
G.D. HAFFNER
Introduction: There are two separate modelling efforts decribed in the PMCR. It is understood
that at a later date, these models will be integrated in order to address the study objectives of
determining system recovery times under a no action scenario, identifying potential remedial
actions and the potential effects of flood events. The first set of models quantify transport and
fate of PCBs in the Hudson River (Hudtox, TIP & SCOUR), and are relatively well advanced.
The second set of models (Bivariate, Probabilistic &Mechanistic) quantify bioaccumulation of
PCBs in order to predict chemical concentrations in fish as a means to determine human and
environmental health risk, by comparing predictions with consumption guidelines.
General Comments: As is often the case with many environmental studies, the monitoring
programs that contributed to the existing PCB data base in the Hudson River were implemented
by different agencies and with different study objectives. The compilation of the data base for a
comprehensive review of spatial and temporal trends of PCBs in the river leaves many questions
of quality assurance, and such questions can effect the ability of the models to make accurate,
precise predictions. Other aspects of the report that can be modified to improve its scientific
credibility include;
1.	Use of a consistent set of units. The switching back and forth from metric causes considerable
problems to compare sections and will only delay the final integration modelling results.
2.	Do not invent new terminology. The IUPAC numbering system should be adhered to. BMFs
are not assimilation efficiencies. I am not certain as to the intended audience, but the use of
terms such as consolidated sediment will confuse the lay person, why not simply refer to
depositional and non depositional sites which are really the issue being discussed.
3.	As noted above, the quality of the data is very suspect (P4-13) where very different methods
and instrumentation have been used to generate the data base. There is reference to the use of
both validated and unvalidated data (P3-4) yet no information as to what criteria were used to
validate data. These criteria must be specifically referenced.
4.Chapter	4 was presented back to front in the text provided.
Transport and Fate Models: The models (Hudtox, TIP SCOUR) are quite acceptable for
addressing the specific study objectives. The emphasis on low chlorinated congeners (IUPAC 4,
28, 52,90/101 and 138) tends to be a very biased manner of quantifying PCB transport fate and
effects. The underlying assumption of all the models used in the study is that these congeners
equally persist in the environment, yet it is well known that any PCB with adjacent,
unsubstituted carbons is very susceptible to metabolism. PCB 138 is the only selected congener
with a known resistance to metabolism and thus truly persistent in the environment. At times
Aroclors are modelled to describe PCB dynamics, yet there is no mention as to how the Aroclors
33

-------
G. Douglas Haffher
were estimated. For example, some agencies use only congener 138 to estimate concentrations
of PCB as Aroclor 1254, thus spurious correlations are readily encountered when comparing
dynamics of PCB in aquatic ecosystems. Furthermore, it is difficult to make spatial comparisons
when the upper Hudtox model uses different PCB measurement than the lower Hudtox model.
Another assumption of the choice of PCB to model is that the hazard of PCBs in the
environment is based on bioaccumulation. A better selection would be to address transport and
fate of congeners that persist, bioaccumulate and are toxic (eg non-ortho PCBs). Future work
should include a meeting with those responsible for the biological/risk models to identify if these
chosen congeners will adequately address the specific study objectives.
On P 4-2, the conclusion is made that it is currently not possible to distinquish between
historical sediment load input and recent (ongoing) discharges. This is based on confusing
results with PCB #4, that can seriously affect the overall quality of the modelling efforts. In my
experience with PCBs, the presented relative distributions of PCB congeners look very
questionable. Consider for example that dichloro-PCBs were only 20% of Aroclor 1016 and
<1% of Aroclor 1254. If these Aroclors were chosen because they were in commercial use
during the time that discharges were made to the environment, then the concentrations of PCB#4
appear to very suspect (even with groundwater seepage into the system). I suggest this is a
critical example of where data quality is affecting the power of the models .
TIP appears to be a good predictive tool of velocities and integrates with SCOUR.
Predictions of the the models provided in the report are reasonable.
Models &Data: The transport models are very adequate to address study objectives 2 and 3. If
the ultimate goal is to quantify risk of PCBs in the river, under different scenarios, then there will
need to be stronger rationale given for the various forms of PCB being modelled. The current
and recommended biological models (discussed later) stress bioaccumulation, yet many of the
forms modelled do not persist sufficiently to bioaccumulate, or are too water soluble to
bioaccumulate beyond that expected by simple bioconcentration processes. There appears to be
an important inconsistency between the variables modelled and the specific objective of the
study regarding human and environmental health.
As mentioned earlier, PCB data are, in my opinion, somewhat questionable. Not only are
relative abundances very different from those usually encountered in aquatic systems, but so are
the relative concentrations in the various biological compartments. For example;
Sediment Table 4-8 10,000 -86,000 ug/g (WW?)
Water	Table 9.8 0.3-0.8 ug/L
Table 9.9 1366-6184 ug/L (incorrect units? as well beyond saturation)
Fish(aroclor) Table 10.3 l-9mg/kg
Such distributions are difficult to comprehend given the physical/chemical properties of PCBs
and their partitioning in the environment.
Specific Questions:
1. Forcing functions are appropriate, although some assumptions might be further tested with
respect to model sensitivity.
a)constant organic carbon in suspended sediment
34

-------
G. Douglas Hafiher
b)settling velocities of 2m/d in a well mixed, shallow riverine system.
2.Spatial/Temporal Trends will not be limited by the use and resolution of the models but by
questionable data quality assurance.
3.	Not applicable.
4.	Relation to risk assessment is limited due to the choice of forms of PCBs modelled. This
choice, if continued, must be justified with respect to the study objectives.
Revised Work Plan This appears to be appropriate save for concerns stated above. There
needs to be a better balance in the development of the physical models and the biological models.
The latter are less well developed, thus opportunities for strategic course correction to address
human and environmental risk might be lost.
Additional Questions: None.
BIOACCUMULATION MODELS
Appropriate Models: The emphasis on bioaccumulation might lead to a limited evaluation of
the hazard of PCBs in the Hudson River. Both the bivariate and probabilistic models are good
descriptive approaches, but lack the robustness of more recently developed food web models.
The suggestion in the text of moving towards steady state models should be seriously considered.
Models that more realistically deal with processes of benthic/pelagic coupling might be more
appropriate for the food web described for the Hudson River.
Models and Data There is no mention as to the use of the TEQ approach to quantifying the
hazard of dioxin-like compounds such as PCBs. This might be a function of the limited data
base available, but if so, this should be explicit in the report. I am surprised the work of Hong et
al. (1992)on TEQs in the estuary of the Hudson River was never mentionned, if even to support
the argument that total PCB predictions might be related to TEQ estimates as done for Lake
Michigan.
The food web description is less than encouraging. Without relative abundance of predator/prey
interactions and trophic levels (e.g. stomach contents, isotope data etc) the food web is very
speculative. More effort should be expended in this area.
Special Questions: This aspect of the study has not advanced as well as the transport/fate
models, and new approaches are suggested in the text (steady state, fugacity models). Thus it is
difficult to respond to the specific questions provided.
Forcing functions: Simple descriptive models assuming equilibrium (BAF, BSAF) are not
35

-------
G. Douglas Haffner
realistic for a highly contaminated system like the Hudson River. Food web knowledge at this
time is very descriptive and can limit the development of more appropriate steady state models.
Spatial/Temporal Scales: Will be appropriate based on successful integration with the physical
transport models.
A single best model (most scientifically defensible) would be the best approach, as the
assumptions driving the other models (equilibrium vs nonequilibrium; steady state) are quite
different.
Ecological/human health risks are being estimated by comparing predicted fish concentrations
with consumption guidelines and wild life protection guidelines. This approach can result in
potential management decisions to protect one (human) and not the other (ecological). The
ecological (if you accept that bioaccumulation is the most important aspect of PCB hazard) will
prove to be more stringent, yet most difficult to quantify and possibly enforce. The high degree
of contamination of system suggests the need for immediate action, and maybe the human health
component might be considered to be given a priority.
Suggested Work The Gobas model would be an improvement, but since then more realistic
food web models have been developed. It might be best to host a workshop to review models
currently available and select the most appropriate model for the selected approach to quantify
risk (bioaccumulation vs TEQ or a combination thereof).
Specific Questions Has EPA specified the information it needs to support a management action
plan? The lack of focus on the various forms of PCBs, the lack of a TEQ approach and the lack
of direction in the biological models suggests that there needs to be a better specified managment
approach. What forms of PCBs are the guidelines based on? Can the other data be adequately
transformed to such estimates? Ecological risk is a very longterm goal, yet the project suggests
the need for decisions today, should consumption guidelines be the main focus of body burden
predictions.
Recommendations: The study should definitely continue, but the biological component might
be stressed. There is not sufficient information to recommend a course of action, but a workshop
on risk models is recommended as a high priority. The workshop must identify the best model,
the data needed for risk assessment and discuss the integration of the biological and phyical
models required to fulfill the need of managers to make appropriate decisions.
36

-------
G. Douglas Haffiier
HUDSON RIVER PCB SITE MODELING APPROACH PEER REVIEW
RESPONSES TO SELECTED COMMENTS PMCR
I have reviewed the second document as noted above. Generally there is considerable overlap
between issues raised in my first review and the comments made by the various parties such as;
-integration of the upper and lower Hudson River models
-organization awkward, I still think putting key figures in the text, and supplementary
figures and tables in an appendix would produce a more cohesive document.
-the issue of pore water being flushed upward relates to the noted inability to separate
current and historic sources, which I still think is a data quality problem.
-semantic problems of identifying PCB congeners, areas of deposition/resuspension
-biodegradation must be considered, especially with the PCB #4 data
-settling velocities must be justified, I still think the value used is very high for a shallow,
flowing river.
-I am somewhat concerned when asked as a reviewer if the three biological models
should be used or one state of the art model, when it is obvious (page 34) a choice has
already been made, based on the 'belief of those developing the models. This 'belief is
based on a weight of evidence approach that assumes the models are equal in their
predictive powers for the different questions being asked. I think models should be tested
and the most rigorous model used for each specific question. I doubt if equilibrium
models will be of much use in a highly contaminated system like the Hudson River,
-inconsistent units, metres vs feet, lipid corrected data, wet wgt data etc. If all this work
is ever going to be integrated, there must be standardized reporting mechanisms!
-feeding ecology of forage fish (see Hebert/Haffner CJFAS 1991) is very different within
cyprinids.
I now have even greater concern regarding the quality assurance associated with the data base.
GE questions the conversion of their own data set, and this is poorly responded to by those who
made the conversion (no statistical justification given at all). The lack of congener 138 (very
persistent and a dominant congener in Aroclor 1254- page 18) is quite an anomalous relative
distribution of PCBs. I agree a water concentration of 1086 ng/L is high (near saturation), but
think most of the water data are very high with respect to the observed concentrations in fish.
Lastly, I am at least confused with the response on Page 38 that 'from a management perspective
only total PCB and Aroclors are of interest'. If this is true, then obviously those designing the
study were not aware of this management decision, and much of the work done is not relevant to
management decisions on hazard assessment and remediation scenarios. This response has
major implications for those doing the biological models to quantify risk. It would appear that a
decision has already been made to use guidelines developed in the late 60s/early 70s to quantify
risk/hazard of PCBs.
37

-------
Alan Maki

-------
Alan Maki
Alan Maki holds a Ph.D. in wildlife and fisheries management from Michigan State University.
He is currently Environmental Advisor for Exxon Company, U.S.A., and served as Senior
Environmental Scientist for Exxon in Alaska from 1986 to 1991. Following the Exxon Valdez oil
spill, he was responsible for managing Exxon's wildlife rescue and rehabilitation program and
for organizing the company's scientific assessment of ecological damage and recovery. Dr.
Maki has authored and co-authored over 160 publications and reports and 6 books on
numerous aspects of environmental quality, ecological risk assessment, toxicology and aquatic
biology.
Active in a wide range of professional organizations, Dr. Maki is currently a member of the
Environmental Protection Agency - Science Advisory Board and has served on numerous
advisory panels for EPA Office of Research and Development. He is former President of the
Society of Environmental Toxicology and Chemistry, and is serving on a National Academy of
Sciences panel concerned with the assessment and management of ecological risks.
40

-------
Peer Review of Hudson River PCB Reassessment
Preliminary Model Calibration Report
Review Comments
Dr. Alan W. Maki
August 21, 1998

-------
A. W. Maki
General Comments: The project managers are to be congratulated for the broad-based state of
the science approach detailed within the Preliminary Model Calibration Report (PMCR). The
contemporary literature and key experts in the chemical fate modeling arena have been
effectively combined to yield a promising program that will undoubtedly address the key
objectives requested by EPA. However, there remain two key areas that I feel will ultimately
limit the success of the program: a) failure to incorporate guidance from the human health and
ecological risk assessment paradigms early in the original design of the program; b) costs
associated with the need for optimal field validation and empirical data directly from the Hudson
River may ultimately preclude the need for much of the efforts associated with model
development. Further comments on both of these areas are contained in the responses to peer
review charge questions detailed below.
A- Is EPA using appropriate models to support scientifically credible decisions?
The direct answer to this question must be evaluated in light of the original decision to do the
reassessment. The information provided to the peer review panel is not sufficient to determine
exactly what were the decision criteria used to re-open the earlier 1984 decision. At that time, the
No Action alternative was collectively determined to be the best alternative with the thought that
natural attenuation of PCB levels would ensue and that natural sedimentation would effectively
cap the PCB-laden sediments in place, thus rendering them non-biologically available. It was not
clear to this reviewer exactly how or why it was determined that this earlier decision required a
reassessment. A clear discussion of exactly what were the decision criteria or concerns that led to
the need for a reassessment will help the review panel with the much-needed perspective to
answer the question of appropriate models.
Risk Paradigm - The first model that should have been employed to define the ultimate construct
of the entire program is the model for human health and ecological risk assessment paradigms.
As currently designed, the program reads as though "we are going to do all this state-of-the-
science fate and effects modeling for PCB concerns in the Hudson River and then we'll hand it
over to the risk assessors to do their assessment." Unfortunately, this is much more the norm for
conduct of risk assessments than it should be. Closer adherence to the guidance provided by risk
assessment literature would help to focus the entire effort and ensure that limited resources are
being maximized to address the key questions. For example, much of the analytical chemist's
42

-------
A. W. Maki
concerns over subtle differences in environmental fate of closely related PCB congeners is moot
since biologists and toxicologists are unable to determine differential biological effects of many
of these related congeners.
NEBA Concept - Restoration ecologists have developed the Net Environmental Benefits
Assessment (NEBA) concept to provide guidance on when remediation efforts are needed and
will help advance the rate of ecosystem recovery versus when further human intervention or
remedial activity will not add a Net Environmental Benefit to the recovery process. For the
Hudson River PCB clean-up, the issue of PCB interactions with benthic sediments is the main
determinant defining the need for additional actions beyond natural attenuation and thus the
pivotal action defining a Net Environmental Benefit Assessment.
Sorption, desorption, and resorption of PCB congeners from Hudson River sediments define the
subsequent bioavailability of residual PCBs to aquatic biota. In a very real sense, "BCF = 0.8
KOC" is a hypothetical simplistic model relating sorption kinetics and thus bioavailability to
organic carbon content of sediments. This reviewer is a strong believer in the application of
simplistic models with minimal variables to provide the most useful guidance. As such, I am
concerned that the PMCR provides only minimal discussion of sorption kinetics and
relationships to bioavailability. It seems that more attention to the fundamental details of these
pivotal relationships would help ensure that scientifically credible decisions would indeed result
from the project.
B.l. Will the models forecast when PCB levels in fish will recover to levels meeting human
and eco-risk criteria?
The models will almost certainly develop a forecast or prediction of the rate of PCB attenuation
in Hudson River sediments and biota. The question remains whether this prediction will have
any relation to the actual future conditions in the river. To date, the system has behaved within
relatively predictable bounds as forecast by simplistic partitioning models. PCBs are found
sorbed to fine particulate sediments with high organic content. These sediments are found in net
depositional environments such as TIP and movement into the food web occurs through
sediment-associated benthos and pore-water partitioning. Relatively simple, two-stage
partitioning models demonstrate this well. Under the current scenario, it appears that benthic
43

-------
A. W. Maki
sediments will serve as a reservoir providing for the slow release of PCBs into Hudson River
biota for many years into the future. However, none of these modeling efforts capture the most
likely long-term determinant, which will be the 50-year or 100-year flood event. As a strong
proponent of ecological chaos theory, this reviewer is convinced that at some point in the future,
an extremely wet summer/fall period will be followed by an eastern seaboard hurricane which
will bring about high water conditions causing flooding and subsequent scouring of sediments
from the depositional areas of the river. That scenario or a winter of heavy snowfall combined
with a wet spring could also cause severe flooding with the same effect on the scouring of PCB-
laden sediments. Any of these chaotic flooding events will likely scour the sediments and
literally flush most PCB residuals downstream and into the marine environment where further
capping by natural marine sedimentary processes will ensue. Obviously, none of these flooding
scenarios are directly predictable by any of the models currently under development, yet it is this
random flooding event that will bring about the most rapid change and recovery of PCB
residuals to background levels.
B.2. Will the models help identify other remedial actions to shorten the time required to
achieve acceptable risk levels?
The models will help to describe the chemical, physical, and biological interactions that control
the distribution of residual PCBs within the Hudson River ecosystem. To the extent that the
interactive models help to identify PCB reservoirs or sinks, they will be useful to help identify
alternative remedial actions. However, this reviewer feels that existing monitoring data and
simplistic partitioning models provide the basic understanding of environmental
compartmentalization of PCBs in the Hudson River ecosystem and no magical alternatives are
likely to evolve from more complex modeling. Existing data show us that PCBs are mainly
found associated with fine sediments of high organic content in the net depositional areas of the
river bottom. This PCB reservoir appears to be the source of food web contamination and will
likely be a continuing source of PCBs for some time into the future. Therefore, the only relevant
remedial strategies involve either removal or sequestering of PCB residuals from these
depositional environments such as TIP.
B.3. Will contaminated sediments become 'reactivated" following a major flood event?
44

-------
A. W. Maki
It is this reviewer's opinion after reviewing the monitoring and modeling data contained in the
PMCR that a major flood event would likely be the best long-term remedial action currently
available. Assuming a sufficient flood-stage water level, major scouring of PCB-laden sediments
would occur from even the depositional areas such as TIP and other slough or back-water areas
where PCB levels are highest. During the flood event, the river would also be carrying a major
load of silt and sediments from upstream areas and the river water color would be chocolate-
brown with turbidity. Resuspension of PCB-laden sediments would undoubtedly occur; however,
due to the flood-stage level of silt and sediments, re-solubilization of PCBs into the water
column would be minimal since ample silt would be present to ensure rapid resorption and
downstream movement of PCB-laden sediments. The water dilution and sediment loading from a
50-year or 100-year flood event would be sufficient to mobilize PCB sediment complexes
downstream to the marine environment of the lower Hudson/Raritan estuary where flood-
associated sediments combined with dilution in the marine environment would further attenuate
PCB concentrations to extremely low ecological risk levels.
C.l. Are the models suitable for quantifying external forcing functions?
There is no doubt that the modeling components are indeed state-of-the-science and that they
incorporate the key variables that influence environmental partitioning of PCBs in the Hudson
River ecosystem. One key point that needs better treatment in the models is the bioavailability
issue. While the physical transport models do an adequate job of describing scouring and
sediment movement, the partitioning relationships that drive sorption/desorption and ultimately
bioavailability need to be better considered in the models. It is not just pure physical transport of
PCB-laden sediments that is important, but it is also the availability of those residual PCBs that
defines their food web mobility.
On page 3-6, a discussion of the Solids Submodel and Toxic Chemical Submodel underscores
the importance of organic carbon content of sediments as a key determinant of PCB fate in
sediments. It was not clear to this reviewer that organic carbon content was adequately
considered in the subsequent model development.
45

-------
A. W. Maki
This discussion again appears on page 8-9 where DiToro's work showing a relationship between
partitioning and foe is shown. It is not clear that this key literature has been fully incorporated
into the PMCR effort. Also on page 8-12, we are told that foe data were not available and the
partial BAF must be expressed on a whole water basis. This is a source of significant error and
calls into question the validity of the entire model.
Also on page 3-6, the statement is made that "The Phase 2 database does not distinguish DOC-
bound PCB's from truly dissolved PCB's but measures these together as 'apparent' dissolved
phase PCB's." This is a major problem and potentially could render model results useless from a
predictive standpoint. This must be clarified and proper distinctions between these phases must
be corrected.
Analytical Chemistry. Pages 9-3 to 9-16 discuss the analytical chemist's attempts to resolve the
historical database for PCB peak resolution. Clearly, the historical trend analysis is confounded
by different analytical techniques and resolution levels. This problem has hounded chemists
since the Swedes first found these unidentified peaks in sediments and biota from the Baltic Sea
in the early 1970s. Historical trend analysis in the Hudson River and, indeed, the success of the
entire modeling effort, are tied to the ability to successfully resolve the earlier analyses.
Chemists must develop a method and agree on its use to interpolate the earlier data. Analytical
methods in the future will continue to change and improve; unless agreement is achieved on a
"best approach," we will continue the endless debate on PCB peak resolution.
C.2. Are spatial and temporal scales of the models adequate? What data are required for
val idation/cal ibrat ion?
These are two very different questions requiring significantly different answers. This reviewer
finds the second question regarding validation as perhaps the major issue facing the future of the
entire modeling effort.
First, regarding the issue of scale, I would like to see more attention to the entire riverine
ecosystem including the lower estuary. It is not clear why marine fish and marine concerns were
not included beyond the Thomann model, since gradual downstream movement of PCB
sediments appears to be occurring. If attenuation to low risk levels is occurring in the lower
46

-------
A. W. Maki
river, this is important to include in the model.
Validation/Calibration. On page 2-3 we are told that the HUDTOX model provides a "reasonable
representation" of PCB dynamics during a nine-month period in 1993. This is encouraging, but it
seems incredible that validation is only available for that short period. However, on page 4-2 we
are told that it is not yet clear whether the PCB dynamics are historically accurate or whether
they are representative of future dynamics. This level of ambiguity seems to characterize the
discussion of validation throughout the PMCR. The modelers involved with each section of the
entire PMCR need to include a specific discussion section stating exactly how model parameters
are to be calibrated and validated against real-world data. Without validation, the models remain
as simple cartoons of reality, open to speculation and challenge.
C.3. Is it better to develop several models of PCB dynamics or use a single "best" model?
Chapter 8 discusses the various modeling approaches under development. The optimized answer
to the question should come from the risk assessors who will be using the model output in the
risk assessment. They should be much more strongly involved in the entire modeling process at
all steps to help ensure the focus and ultimate utility of the models. With their involvement, I am
virtually certain that the empirical probabilistic model derived from the real-world data would be
of most value. Following this approach the model is essentially calibrated from the outset and
can then be readily validated by comparison with historical data. The 'best" model selected for
development must fully consider sediment/water interactions, partitioning theory, and
solubilization as the main exposure route and food web sources as secondary routes for fish body
burdens.
In many places in the PMCR, vague promises or allusions are made to the model's ability to
provide further insight into the role of water versus sediment versus forage fish as sources of
PCBs (page 8-6). These allusions need to be tempered with a strong dose of reality. This
reviewer does not feel that the models will elucidate these complex relationships. Either
carefully designed and controlled laboratory experiments or major field monitoring programs are
needed to accurately describe PCB dynamics.
47

-------
A. W. Maki
C.4. Is the level of process resolution adequate to describe PCB dynamics?
This question is not substantively different from C.2 and my answers there pertain equally well
to this question. Indeed, much of the detail included in many of the modeling sections seems
superfluous and subject for later "tuning" once the basic model is running. In many places, it
appears that the modelers are attempting to "micrometer a brick." Examples include page 5-5
where concerns are expressed over the localized effect of bridges on water flow measurements.
Also, page 8-2 and 10-27 where the six main fish species to be modeled are discussed. I don't
see where we need to spend a great deal of effort to distinguish between pumpkinseed, yellow
perch and white perch. They are functionally very similar and should show similar PCB
dynamics. Use of a shiner, perch, bullhead, and possible addition of the striped bass would
simplify resolution issues as the models are developed.
C.5. Will the models be useful for human and eco-risk assessments?
Yes, but as stated in responses to the previous questions, the risk assessors need to be involved
much more rigorously in the entire model development process. The PMCR simply reads as a
project to generate the models and 'when we're done we'll give it all to the risk assessors to see
what they can make of it." Obviously, this is not correct and the risk assessors need to be
involved much more interactively at all stages of the process. The program managers should
regularly consult the risk assessment paradigm to help guide decisions throughout the model
development process.
D & E. What changes would improve the outcome? Are there serious flaws that will limit use of
the models?
This reviewer is unable to distinguish different responses to questions D & E since they are re-
statements of the same question and ask for a summary of the changes we recommend. My
comments on the program are summarized in responses to all previous questions and are thus
summarized in bullet form here:
• Incorporate and follow the risk assessment paradigm in all stages of model development.
48

-------
A. W. Maki
Involve the risk assessors not only at the end but throughout the model development.
Validate all steps of the model and model outputs with real-world monitoring data from the
Hudson river ecosystem.
Give better consideration to the factors controlling bioavailability of PCBs. These include
sorption/desorption kinetics, solubilization and organic content of sediments and equilibrium
partitioning.
Make better use of simplistic, two-stage partitioning models to model PCB dynamics before
developing the complex, multi-variable models.
Chaos theory would say to plan for the 50-year or 100-year flood event as the most
reasonable remedial action.
Resolve the analytical chemistry to ensure full consideration of historical PCB monitoring
data.
Consider the Net Environmental Benefits Assessment (NEBA) before recommending
additional remedial actions.
Models must be able to distinguish between soluble and sediment/DOC-bound PCB's. These
cannot simply be lumped together since they represent fundamentally different PCB
reservoirs.
Don't spend a lot of resources attempting to "micrometer a brick." Target for simplistic,
reasonably accurate outputs before involving too many extrinsic variables on too many
species.
Do not hold unrealistic expectations of the modeling capability. There is very little hope that
much insight will be gained from modeling into the complex interactions of PCB bio-
dynamics.
Do not let the analytical chemists get ahead of the toxicologists. The risk assessment will
gain little value from chemists' resolution of each substituted PCB congener if toxicologists
cannot explain the biological/toxicological significance of these congeners.
49

-------
Thanos Papanicolaou

-------
Thanos Papanicolaou
Dr. Papanicolaou, Assistant Professor of Civil and Environmental Engineering, Washington
State University, received his M.Sc. and Ph.D. in Hydraulic Engineering from Virginia
Polytechnic Institute and State University, Blacksburg, VA in 1992 and 1997, respectively and
his B.Sc. degree with honors from the Department of Civil Engineering in Aristotels University of
Thessaloniki in 1990.
He has been a Research Assistant of the Department of Civil Engineering at Virginia Tech from
1991-1997. During this period his research received support from the International program of
NATO Fellowships and from the International organization A. ONASSIS. In addition, he
received support from two different projects sponsored by the National Science Foundation and
U.S. Geological Survey. From 1989 until 1990 he worked as a consultant to the engineering
firm, THEMELIODOMI. His main area of specialization is sediment transport and experimental
fluid mechanics.
He has served as a reviewer for the Journal of Hydraulic Engineering. Computing Engineering,
and Engineering Mechanics of American Society of Civil Engineers (ASCE) and he is member
of the ASCE, American Water Resources Association (AWRA), American Geophysical Union
(AGU), and of the Greek chamber of Civil Engineers. Dr. Papanicolaou is the author of more
than 35 research papers and serves as a member of the American Society of Mechanical
Engineers (ASME) turbulence committee.
Dr. Papanicolaou currently supervises two graduate and two undergraduate students. He is
involved in cohesive sediment studies examining the role of turbulence on the entrainment of
sediment. Currently, he is involved in a project focusing on different fish bypass designs. He
taught the open channel flow course at Virginia Tech and the Sediment transport, Experimental
Hydraulic Engineering, and Fluid Mechanics courses at Washington State University.
52

-------
Thanos Papanicolaou
Final Comments
of
Thanos N. Papanicolaou
Title: Thompson Island Pool Depth Modeling
Question A: Is EPAusing appropriate models, datasets and assumptions on which to base
a scientifically credible decision?
I feel comfortable to comment on the efficiency of the hydrodynamic model that is used
here to calculate the erosion rate of sediment in Hudson River.
The hydrodynamic model that is used in the present study is the RMA-2V model. This
model has significant capabilities comparatively to other numerical hydrodymanic
models (e.g. HEC-6). First, the RMA-2V is a two dimensional finite element model and
provides the stream local velocity in the longitudinal and transverse direction of the
stream at different time instants. Second, the RMA-2V model is one of the few models
that is supported by other commercial software that are used for the pre-processing of the
input data file (i.e. the program Fast Tabs developed by the Boss Corp.).
In the present study, the quasi-steady aspect of the problem is examined. The RMA-2V
finite element model is solved for sufficient large time steps to satisfy the conditions
needed for the quasi-steady solution of the problem. The solution of the model, viz. the
53

-------
Thanos Papanicolaou
velocities in the longitudinal and transverse directions, are used to calculate the bed shear
stress at different locations along the stream cross section. Knowledge of the magnitude
of the bed shear stress is important to determine the erosion rates of sediment.
There are two types of sediment types within the Thompson Island Pool, cohesive and
non-cohesive. The erosion of the cohesive sediment is studied here by using the Lick et
al. (1995) model (i.e. equations (3-6) and (5-2) in the report). No method has been
adopted for the study of the cohesion-less sediment erosion process.
At this point, the reviewer would like to comment on the effectiveness of the cohesive
erosion model. The Lick et al. model (1995) is used here to describe the entrainment of
cohesive sediments. Specifically, this model considers that the sediment erosion rate is
well described by a power law. The parameters considered in the model are: the bed
shear stress, and the time after the last sediment deposition took place. While the above
model constitutes the latest development in the area of cohesive sedimentation, it presents
some limitations since it is applicable to cases that the flow conditions are not extremely
intensive. The present model is limited to flow conditions under which the average bed
shear stress does not exceed more than 20 dynes/cm2. During flood conditions the bed
shear stress obtains values that are close to 50-100 dynes/cm . Under these conditions the
erosion of the sediment bed is not sporadic but reaches an equilibrium value. Despite the
above limitations, the existing cohesive scour model (equation (3-7)) provides accurate
results when the average bed shear is not greater than 20 dynes/cm2 (this applies here).
54

-------
Thanos Papanicolaou
In the present study, determination of the entrainment of the cohesion-less sediment is
important since "42 percent of the PCB mass reservoir as of 1984 was located in larger
sediment areas consisting of non-cohesive sediment". Another reason for determining the
entrainment rate of non-cohesive material is the notion that the existence of large
quantities of non-cohesive sediment may alter the sedimentation behavior of cohesive
sediments. The present scour model does not take into consideration the entrainment of
non-cohesive sediments. The reviewer strongly believes that knowledge of the erosion
rate of non- cohesive sediments is very important in order to determine the total rate of
sediment that will be likely "reactivated" following a major flood. The research team
involved in this project intends in the near future to consider in their study the erosion of
non-cohesive sediments.
The reviewer would like to provide some information about the existing state-of-the-art
knowledge in the area of non-cohesive sediment entrainment. There are two types of
models, the deterministic and stochastic models. The deterministic models consider the
average bed shear stress as responsible for the entrainment of sediment while the
stochastic models take into consideration the occurrence of turbulent bursts. The latest
developments in the area of sediment-water interaction suggest that the erosion process,
especially for low flow conditions, is a random process. Unfortunately, use of the
stochastic models at this point is not possible due to the complex nature of the erosion
process. For this reason, the reviewer suggests the use of a semi-deterministic model.
55

-------
Thanos Papanicolaou
This model takes into account the frequency of occurrence of bursts, the sediment
availability, the sediment size and weight, and the average bed shear stress and was
developed by Cao (1997).
The Cao's entrainment model is described by the following equation:
E„= Pd1'5 (F/f-1) F
Where,
En=normalized entrainment rate
p _ IC.jsg)"
vTb
where X is the averaged area of all bursts per unit stream bed area=0.02
C0 is the sediment packing density that is equal to 0.6
s is the sediment specific gravity that is equal to 1.65
g is the acceleration of gravity, equal to 9.8 m/s
v is the Kinematic viscosity and is approximately equal to 10"6 m2/s
Tb is the average bursting period
U?
F is the Shields parameter and is defined as F = —-
sgd
Where U. is the friction velocity and d is the average sediment particle diameter.
56

-------
Thanos Papanicolaou
It is expected that the above model can be incorporated into the hydrodynamic model
suggested here by the research team. In conclusion, the hydrodynamic model suggested
here is acceptable with minor revision about the erosion of the cohesion-less sediment.
References:
Cao, Z. (1997). "Turbulent Bursting-Based Sediment Entrainment Function", Journal of
Hydraulic Engineering, Vol. 123, No.3, pp. 233-236.
Question B:
1.2:1.am not familiar with these issues.
3. Are there contaminated sediments now buried and effectively sequested from the food
chain which are likely to become "reactivated" following a maior flood, resulting in an
increase in contamination of the fish population.
The rate of sediment that will be likely eroded within the TIP strongly depends on the
variation of the bed shear stress values. For a 100-year flood event the hydrodynamic
model used here predicts that the mean bed shear stress value is equal to 19.5 dynes/cm2.
This value of stress is considered as sufficient to yield the erosion of the less compacted
layer located atop the river bed. There are typically two threshold erosion values for the
shear stress, one for the newly deposited sediment and one of the consolidated layers.
When the bed shear stress exceeds the value of 10 dynes/cm then erosion of the more
57

-------
Thanos Papanicolaou
compacted layers of sediments may occur. The average depth of scour calculated here for
the 100-year flood event is within the range of 0.003 cm to 0.97 cm.
Moreover, the shear strength of sediment is largely function of the moisture content of
sediment. Knowledge of the moisture content of sediments is required to provide a
definite answer to the above question.
Question C:
1. Are the modeling approaches suitable for developing quantitative relationships
between external forcing functions and PCB concentrations in the water column.
sediments and fish? Are the models adequate for discriminating between water-related
and sediment-related sources of PCBs?
The present work involves the use of different models for the transport and fate of PCBs
in the water column and bedded sediments, and for PCB body burdens in fish. The
modeling approach is suitable for the present problem since it incorporates 3 separate
mass balances, viz., a water balance, a solids balance and a PCB balance. The solids
balance provides information about the amount of PCBs absorbed to the sediments, the
water balance is important since it provides the rate of PCB's transported by water, and
PCB balance provides the amount of PCBs as a function of sediment-water and air-water
exchanges.
58

-------
Thanos Papanicolaou
2.	Are the spatial and temporal scales of the modeling approaches adequate to answer the
principal study questions?
I feel comfortable to comment on the spatial temporal scales of the hydrodynamic and
scour model. For the scour model a uniform size grid was defined by using a fine scale
Geographical Information system approach. The cell spacing that was chosen here is 10
feet. This level of spatial resolution is adequate in capturing the level of erosion that
takes place within the TIP.
The spatial scaling of the hydrodynamic model (i.e., RMA-2V) is found reasonable. The
size of the grid depends on the magnitude of the velocity. Therefore, the number of
elements in the finite element mesh should be adjusted in accordance to the current flow
conditions at the site.
The reviewer would like to express his concern about the hypothesis used here that the
maximum bed shear stress value is established instantaneously. This may not be true in
all cases.
3.	It is contemplated that PCB concentrations in fish will be estimated using several
modeling approaches: an empirical probabilistic model derived from Hudson River data,
a steady state model that takes into account mechanisms of bioaccumulation body
burdens, and a time-varving mechanistic model. A bivariate statistical model mav also be
used to provide insight into accumulations.	This multi-model approach is being
contemplated because of the uncertainties associated with anv individual model. Is this a
reasonable approach or should predictions be made using a single "best" model?
59

-------
Thanos Papanicolaou
The reviewer would like to compliment the research team focusing on the study of the
PCB concentrations in fish. The idea of using a bi-variate statistical model is very
interesting. Similar approach has been used in the past in studying the flow-sediment
interaction problem. The multi-model is acceptable as is.
4. Is the level of process resolution in the models adequate to answer the principal study
questions? If not, what processes and what levels of resolution are required to answer
these questions? What supporting data are required for those processes and levels of
resolution?
The calibrated RMA-2V model is a reasonable representation of the hydraulic conditions
that exist within TIP. The fact that the model provides velocity values that do not deviate
significantly for the values provided by the USGS indicates that the calibration procedure
that was followed during the course of this study is pretty accurate. The reviewer
suggests here that the value of the average equivalent sand roughness is equal to 3
d50.(VanRijn 1984).
P. Are there anv changes to the work effort outlined in the revised work plan that would
significantly improve the outcome?
The scour model should be modified to account for the erosion rate of non-cohesive
sediments. This will make the scour model more general and applicable to different
sediment types (review the suggested model for non-cohesive sediments (see the answer
for question 3(B)).
60

-------
Frank Wania

-------
Frank Wania
Frank Wania has a "Diplom in GeoSkologie" from Bayreuth University in Germany and a Ph.D.
in chemical engineering and applied chemistry from the University of Toronto. He has worked
for one year as a research assistant at the GSF - Centre for Health and Environment in Munich,
and for two years as a research scientist with NILU - the Norwegian Institute for Air Research in
Tromse. He has several years of research experience in the fields of describing the fate of
persistent organic pollutants and mercury on a global, regional and local scale by developing
and applying compartmental multi-media models, interpreting measurements of chemical
behavior in the environment, and measuring physical-chemical properties of organic chemicals.
Frank Wania is an independent research scientist in Toronto, Canada. His clients include
industry, academia, governments, and non-governmental organizations from Canada, the USA,
Norway, Sweden, and other European countries. He has served as a member on the North
American Expert Advisory Panel on Continental Pollutant Pathways and is presently a member
of the editorial board of "AMBIO a Journal of the Human Environment," published by the Royal
Swedish Academy of Sciences. He is the author of more than 50 papers, including more than
20 peer-reviewed articles in scientific journals and books.
62

-------
Peer Review of Hudson River PCBs Reassessment
Frank Wania, Page 1
Peer Review of Hudson River PCBs Reassessment
RI/FS Phase 2 reports
	Preliminary Model Calibration Report	
by
Frank Wania
WECC Wania Environmental Chemists Corp.
280 Simcoe Street, Suite 404, Toronto, Ontario, Canada M5T 2Y5
A. Is EPA using appropriate models, datasets and assumptions on which to base a
scientifically credible decision?	
The approach of using models of the physical environment of the Hudson River to
predict water and sediment concentrations, and then use these as input for models of
the uptake in biota appears sensible and appropriate. The models of PCB transport and
fate in the Hudson River are based on well-established and accepted methodologies
for the quantitative description of hydrophobic organic substances in aquatic systems.
They are using an adequate spatial and temporal resolution and include most of the
relevant fate processes. The presented models to predict levels in fish so far rely in my
opinion too much on empirical data and too little on a mechanistic understanding of the
processes of bioaccumulation.
The amount of data gathered on PCB in the Hudson River and used in this
investigation is unique and impressive in terms of both their spatial and temporal
coverage, and the number of compartments investigated. The Hudson is possibly the
most comprehensively investigated river with respect to PCB contamination. The extent
of work devoted in this project to making historic and diverse data sets on PCB
concentrations comparable is well-spend, because crucial for the success of the
project.
Despite the models being state-of-the-art, there is a possibility that their usefulness for
answering the principal questions will be limited, because of the large uncertainty
attached to their predictions. This is not because the models, datasets or assumptions
are unsuitable for the task they were designed to address. Our quantitative
understanding of the behaviour of PCBs in river systems and the fresh water food web
may simply be so limited that it does not enable us to predict with sufficient certainty
what the levels of PCBs in fish in the Hudson River are going to be two decades from
now.
63

-------
Peer Review of Hudson River PCBs Reassessment
Frank Wania, Page 2
B. Will the models, with the associated datasets and assumptions, be able to
answer the principal study questions as stated in the PMCR
In order to answer this question, it is necessary to predict water and sediment
concentrations in the future on the time scale of decades. This involves the prediction
of (1) the future release of PCBs from the contaminated sediments, and (2) the future
import of PCBs to the river from the drainage basin.
In order to accomplish task (1), it is necessary to:
•	explain and quantitatively describe what processes are responsible for the observed
increase in PCB water concentration across the TIP and presumably also upstream
of the TIP.
•	predict how that process is likely to develop in the future.
I am not convinced that it can presently be stated with confidence that the HUDTOX
model has a full quantitative understanding of the mechanisms of PCB release from the
sediments in TIP. This is illustrated by the need to invoke contaminated advective pore
water inflow to explain the measured increase of the lower chlorinated PCB congeners
across TIP.
Particularly worrisome is that there appears to be a significant discrepancy between the
results of the HUDTOX model and the TIP Depth of Scour model in their estimation of
how much PCB is released with resuspended sediments during the model calibration
period.
According to Figure 4-39, the HUDTOX model calculates that during the model
calibration period (i.e. the first nine month of 1993) 405.7 kg of PCBs were transferred
to the water column with resuspended sediments within TIP. Most of that transfer
occurred during the spring run-off event period from 3/26/93 - 5/10/93 (Figure 4-40).
That period saw its peak flow on April 12, 1993 with 20,300 ft3/s (Table 5-1). The TIP-
Depth of Scour model does not report specific estimations of the mass of PCB eroded
from TIP during the spring event 1993. However, the spring 1994 event with a peak
flow of 28,000 ft3/s is predicted to have eroded 6.58 kg. The spring 1992 event with
19,000 ft3/s is predicted to have eroded only 1.57 kg (Table 6-5). There is thus a huge
discrepancy between the estimated mass of PCB eroded from TIP during the 1993
64

-------
Peer Review of Hudson River PCBs Reassessment
Frank Wania, Page 3
flood event by the TIP-Depth of Scour model (approx. 2 kg) and the HUDTOX model
(approx. 300 kg).
I see several potential explanations for that discrepancy: Maybe the "mass of PCB
eroded" estimated by the TIP depth of scour model does not apply to the entire flood
event, but only to a relatively short time span (e.g. the one day period of maximum
flow). If that is the case the estimates of PCB eroded as reported in Table 6-5 could be
very misleading. The mass released during the entire flood period could be potentially
much higher than what is listed in that Table. Another potential explanation is that the
TIP-Depth of Scour model does not yet include erosion of non-cohesive sediments.
However, the mass of PCB eroded from non-cohesive sediments is unlikely to be so
large as to resolve this large discrepancy.
The discrepancy between the two models applies similarly to the solids. The HUDTOX
model calculates a resuspension of 7.56-10+6 kg from TIP during the entire calibration
period (Figure 4-36), and most of it during the 1993 spring flood event with 20,300 ft3/s
(Figure 4-37). The TIP-Depth of Scour model estimates only 5.53 10+4 kg for a event
with 19,000 ft3/s and 2.20-10+5 kg for a event with 28,000 ft3/s.
Presumably the resuspension rates in the HUDTOX model are that high in order to
explain the increase in PCB concentrations across TIP. That discrepancy needs to be
resolved because this increase of PCB concentrations in the TIP is at the very core of
the problem to be addressed! (After writing this, I read on page 5 of the revised
Appendix B, that "for the 1993 and 1994 flow events, cumulative gross resuspension
estimates from the TIP resuspension model will be compared with cumulative gross
resuspension results from the TIP portion of the revised HUDTOX model." This is
exactly what I was trying to do in a "back-of-the-envelope" fashion above.)
As long as the observed increases in PCB water concentrations in TIP can not be
described in HUDTOX with sediment resuspension rates consistent with the Depth of
Scour model, it is unlikely that it can predict the release of PCBs from sediments in the
future. In addition to an understanding of the release processes and their kinetics, the
dynamics of this future release would obviously require the capability to predict the rate
by which the PCB reservoirs in the sediments are being buried. The planned long-term
hindcasting calibration should give some indication of the capability of the HUDTOX
model to describe the rate of this process.
65

-------
Peer Review of Hudson River PCBs Reassessment
Frank Wania, Page 4
As mentioned above, the prediction of future exposure concentrations requires also an
estimate of the future import of PCBs from across the upstream model boundary. In the
revised Appendix B it is stated that for the predictive No Action scenario modelling,
"long term -time series must be constructed for [...] external loadings of [...] PCBs."
(Page 5, 1. Paragraph). I wonder how this should be reliably done, considering that
there is some clear indication that the inflow of PCBs at the Northern model boundary
derives not only from the watershed, but that there are upstream sources of PCBs
other than the run-off of atmospherically deposited PCBs.
During the model calibration period:
Upstream	Tributaries	U/(U+T)	Source
Average water flow 5418 ft3/s	9465 ft3/s	36.4 %	Table 4-2
TSS load	3.55-10*7 kg	3.85-10+skg	8.4%	Figure 4-35
PCB load	352.0 kg	121.46 kg	73.4%	Figure 4-38
This table shows that the Hudson River upstream of Ft. Edward supplied only 36.4 % of
the total water input to the Upper Hudson River, and less than 10 % of the solids (the
rest being supplied by the other tributaries). Nevertheless, it imported almost 73.4 % of
the external PCB load. Because there is little reason to believe that the watershed input
from atmospherically derived PCBs is substantially different between the drainage
basins of the Upper Hudson River above Fort Edward and the other tributaries, such as
the Mohawk River, these data clearly indicate that there are other sources of PCBs in
the upstream Hudson River.
Some of the presented data (e.g. Figures 10-1 to 10-8) show that sediments upstream
of river mile 195 are contaminated with PCBs. Though levels tend to be lower than in
TIP, they are consistently higher than in the lower Hudson or in the sediments upstream
of Hudson Falls (river mile 200). What was the rationale for starting the simulation a
few kilometers downstream of the original PCBs discharge points, namely why was the
section from Hudson Falls to Fort Edward (river mile 195 to 200) not included in
HUDTOX and the depth of scour model?
Inclusion of that section in the model would make a clearer distinction between
watershed PCB inputs and internal river source possible. For a prediction of future
development of the river concentration, the future development of the PCB
66

-------
Peer Review of Hudson River PCBs Reassessment
Frank Wania, Page 5
concentrations at the upstream boundary are of obvious importance. It is difficult
enough to predict what those concentrations are likely to be if they are only determined
by atmospherically derived watershed inputs. It will be considerably more difficult if they
are influenced by river-internal sources outside of the model boundaries.
B. Will the models, with the associated datasets and assumptions, be able to
answer the principal study questions as stated in the PCMR
The PMCR never details what the alternatives to "No Action" are beyond vaguely
referring to "selected dredging and/or containment" (page 13 of revised Appendix B).
How should it be possible to judge whether the models can evaluate a scenario, if that
scenario is not specified? If the scenarios involve shutting off future release of PCB
from certain contaminated sediment sections, it may be possible to calculate the effect
on fish concentrations, if the models spatially resolves these sections.
B. Will the models, with the associated datasets and assumptions, be able to
answer the principal study questions as stated in the PCMR
The combination of the hydrodynamic model and the Depth of Scour model for TIP
should be able to address the question of how much of the PCBs buried in the
sediments could be reactivated during a "flood event". (The discrepancy in the
estimated amount of resuspended sediment/PCB between the two models approaches,
which is mentioned above, needs to be resolved.) What has not been laid out in detail
is how the result of that calculation would feed into the models of food chain
uptake/accumulation. Such a link would be necessary to answer the question of the
impact of that "reactivation" on PCB levels in fish. The presented biological model
approaches assume some sort of steady-state exposure concentration and are not
designed to describe fish uptake (let alone the effects on fish) during short periods of
greatly elevated exposure as it would be caused by a flood event stirring up previously
buried sediments. This may be the task assigned to the time-varying model, that is
referred to, but not presented in detail, in the PMCR.
67

-------
Peer Review of Hudson River PCBs Reassessment
Frank Wania, Page 6
C. Specific Questions
This question was already addressed in part above.
Why is it important to discriminate between water and sediment-related sources of
PCBs to biota? Is it even conceptually sound to distinguish between the two? Aren't
present day PCB concentrations in the water column "sediment-related", in that the
supply of PCBs to the water column is almost exclusively from the sediments. Even
though PCBs in water and sediment may not be in equilibrium, there is likely to be a
clear relationship between them.
In that respect it would be a very worthwhile undertaking to investigate the equilibrium
status of PCBs (total PCBs as well as individual congeners) between water column and
sediment in various river sections. This could be done e.g. by calculating fugacity ratios
between water and sediment or by comparing "apparent" or Uin-situ" water-sediment
partition coefficients derived from measured water and sediment concentrations. (A lot
of effort has been spend on comparing carbon and lipid-normalised concentrations in
water and biota and in sediment and biota. Something similar should be done for water
and sediment.) An analysis of the spatial and temporal variability of that equilibrium
status would be very instructive in identifying the relative importance of water and
sediments as an exposure medium, and how they influence each other.
C. Specific Questions
I think the spatial and temporal scales are adequate. A higher spatial resolution (e.g. of
the HUDTOX model) would not be justified because the measured concentration data
needed for calibration and validation are not available in a higher spatial resolution.
Even at the present model resolution there are river segments for which no
measurements exist.
68

-------
Peer Review of Hudson River PCBs Reassessment
Frank Wania, Page 7
An issue of spatial scale was already mentioned before: If an alternative to No Action
involves that certain sediment sections are being dredged or contained, the sediment
compartmentalisation in the model should be able to resolve these sediment sections.
By adding a second dimension to the description of TIP in HUDTOX, this may have
been accomplished.
Two issues related to temporal scales, which were already mentioned before, may be
worth repeating here:
1.	It is not clear to me, to what temporal scale the Depth of Scour model predictions
(mass of sediment and PCB eroded) refer to. Do they refer to the entire flood event
or only to the day (or hour ?) of peak flow?
2.	The question of the impact of flood-related reactivation on PCB levels in fish could
only be addressed with a time-variant model of food chain uptake, resolving time
scales of less than a month.
C. Specific Questions

The PMCR primarily presents the two empirical approaches (bi-variate statistical model
and probabilistic model) for estimating PCB concentrations in fish, whereas it is rather
vague on the more mechanistic approaches.
Mechanistic and empirical approach have advantages and disadvantages:
The empirical approach:
+ is very specific to the studied system, as accumulation factors are estimated from
concentrations measured in the Hudson River.
- is only as good as the available data (As becomes clear from reading the PMCR the
necessary data are often missing ("numerous data gaps" page 12 in revised
workplan). It is even suggested to use model-predicted concentration in water and
69

-------
Peer Review of Hudson River PCBs Reassessment
Frank Wania, Page 8
sediment for deriving "empirical" accumulation factors. I am very skeptical about
employing calculated data as substitutes for measured data in a supposedly
empirical model)
-	provides only a limited understanding of how the system works (E.g. a accumulation
factor derived from measurements is taken as it is, there is no attempt at
understanding the underlying processes). If there is a change in how the system
operates, the empirical approach is likely to fail (e.g. a profound system change
would occur if PCB concentrations in the water had been driven by upstream
loadings of PCBs for some time, but as these loadings decrease is increasingly
controlled by release from contaminated sediments). Are the empirical relationships
still valid after potential remedial measures, e.g. dredging?
The mechanistic approach:
+ may be able to describe the system even after a major change in system properties,
because it simulates and predict the system from an understanding of the underlying
chemical and physical principles.
-	may fail to include all relevant processes (e.g. pore water advection). Also, the
quantitative understanding of the processes may be poor.
Although specified in the question above, I see little in the PMCR or the revised
workplan that shows a commitment to seriously pursuing a mechanistic modelling
approach ( "is being explored", "is being evaluated"). I suggest that it is extremely
important that a mechanistic food chain accumulation model be used to complement
the (semi-)empirical models described in the PMCR. Models such as those by Thomann
et al. (1992), Gobas (1993), or Campfens and Mackay (1997) are publicly available and
should be well suited for the Hudson River
Thomann, R.V., Connolly J.P., and Parkerton T.F. 1992. An equilibrium model of
organic chemical accumulation in aquatic food webs with sediment interaction.
Environmental Toxicology and Chemistry 11, 615-629.
Gobas, F. 1993. A model for predicting the bioaccumulation of hydrophobic organic
chemicals in aquatic food webs: application to Lake Ontario. Ecological
Modelling 69: 1-17
Campfens, J. and Mackay, D. 1997. Fugacity-Based Model of PCB Bioaccumulation in
Complex Aquatic Food Webs. Environ. Sci. Technol. 31: 557-583.
For a recent comparative discussion of the models by Thomann et al. and Gobas see:
70

-------
Peer Review of Hudson River PCBs Reassessment
Frank Wania, Page 9
Burkhard, L.P. 1998. Comparison of two models for predicting bioaccumulation of
hydrophobic organic chemicals in a Great Lakes food web. Environ. Toxicol.
Chem. 17, 383-393.
In brief, it is very important to use more than one model and check the validity of
predictions by comparing the results of several modelling approaches. But these
approaches should go beyond the two presented in the PMCR.
C. Specific Questions
As the authors of the PMCR point out themselves, advective flow of pore water is
another process potentially transferring PCBs from sediments to the water column.
Pore water diffusion of PCBs tends to be fairly low and resuspension of particle-bound
PCBs is mostly an episodic phenomenon, so ground water inflow may be a significant
process explaining the increase of PCB water concentrations across TIP during periods
without flood events. Whereas the authors mention only advective pore water inflow of
truly dissolved PCBs (which is most pronounced for the most water soluble congeners
such as BZ#4), it is conceivable that also the less water soluble congeners could be
transferred by this route, if DOC - and thus DOC-bound PCBs - is advected out of the
sediments.
C. Specific Questions
Human and ecological risk assessment is outside my area of expertise and I thus see
myself not in a position to pass a judgment on this issue.
D. Are there any changes to the work effort outlined in the revised work plan that
would significantly improve the outcome?		
As mentioned before, I suggest
71

-------
Peer Review of Hudson River PCBs Reassessment
Frank Wania, Page 10
•	to include river mile 195-200 in both the HUDTOX model and the hydrodynamic and
depth of scour model.
•	to be careful in not relying solely on the empirical models of PCB uptake in fish,
especially considering the gaps in the available data. The use of a mechanistic food
chain accumulation model is imperative.
E. In terms of evaluating the overall and specific effects and behaviour of PCBs in
the Hudson River, are there any serious flaws in the modeling approach
(theory, structure, physical parameters, etc.) that would limit or invalidate any
conclusions of further work based on results of these models?
The shortcomings that I pointed out above are not of such fundamental nature that they
invalidate the entire approach, yet addressing them may increase the usefulness of the
models.
I am bit puzzled to see the word "effects [...] of PCBs" in this question. None of the
presented methodologies addresses the question of effects. The focus of the
presented work is entirely on exposure.
Other Issues
Though I am not a statistician, I am a bit skeptical about the appropriateness of various
statistical procedures employed in the PMCR:
Comparison between model results and observed data with t-tests
In checking the HUDTOX model performance, t-tests are employed comparing the
mean of the measured concentrations with the mean of the calculated concentrations
If the test shows no significant difference, it is taken as a sign that the model performs
well. I think there are two problems with this approach:
1: The observed and the modelled concentrations are not random samples of a
population, but refer to particular points in time. When comparing the means rather
than the observed and modelled concentrations which apply to the same point in time,
that aspect is ignored. For illustration, see example below: The modelled and observed
concentrations have the same mean and a large variance, and a t-test on the means
72

-------
Peer Review of Hudson River PCBs Reassessment
Frank Wania, Page 11
will thus show definitely no significant difference. Yet the model-measurement
agreement is very poor.
e
o
c
V
o
c
o
. observed
modelled
time
2: A closer analysis of the comparison shows that often the t-test indicates "no
significant difference" not because the means are close to each other, but because the
variance of the means is so high. To say that two means are not significantly different is
not the same as saying that the two means are actually similar. No judgment on model
performance can thus be derived from a test result suggesting "no significant
difference".
Again, the example below may illustrate the point: the comparison to the left will show
"no significant difference" because of the high variance, whereas the comparison to the
right will show "significant difference" even though the means of modelled and
observed concentrations are actually closer.
observed
modelled
c
o
P3
*3
C
(U
o
c
o
a
observed
modelled
time
time
Need to distinguish between development and testing of an empirical model
A similar sort of problem, I have with the bi-variate statistical model presented in
Chapter 9. The goodness of fit between "predicted" concentrations and measured
concentration as shown in Figure 9-8 through 9-19 is seen as providing "good
73

-------
Peer Review of Hudson River PCBs Reassessment
Frank Wania, Page 12
explanatory power in predicting annual mean body burden..." (page 9-15). This is not a
correct statement. In fact, the predictive capabilities of this model have not been tested
at all. These graphs (and the r2 in Table 10) only give an indication of how well
measured fish body burdens and measured water and sediment concentrations are
correlated. The predictive power of the regression model would have to be tested on
independent data pairs, i.e. on measured fish and water/sediment concentrations which
were not used in the derivation of the regression equations. A distinction between a
calibration data set and a test data set is necessary.
Derivation of distributions for accumulation factors
In deriving the distributions of accumulation factors, the individual measured
concentrations in biota are divided by the mean of the water or sediment
concentrations. What is the rationale for using the mean of the water and sediment
concentrations rather than their distributions?
Other Issues
Many of the presented BSAF (Figure 10-9 to 10-40) are close to 1, i.e. indicate
equilibrium partitioning between sediments and benthic invertebrates. Deviations from
this for some locations and some species are variable and uncertain, i.e. can not really
be explained satisfactorily (E.g. what is the mechanism explaining higher than
equilibrium partitioning in zooplankton? Gastrointestinal biomagnification at the stage of
the zooplankton? Also, the combination of water column data with zooplankton data
requires often quite strenuous assumptions.)
Is there really something gained by going beyond an assumption of equilibrium
partitioning? Isn't it rather introducing a lot of uncertainty. The Gobas model predicts
concentrations in benthic and pelagic zooplankton (and thus also for phytoplankton)
from simple equilibrium partitioning. Can it be tested whether the BSAFs are actually
significantly different from 1.
Other Issues
Page 10-14, 2nd paragraph and page 10-18, alternative approach 6
74

-------
Peer Review of Hudson River PCBs Reassessment
Frank Wania, Page 13
"Skoglund et al. found that phytoplankton accumulate more PCB than would be
predicted be equilibrium partitioning alone"
"equilibrium model significantly underestimates observed accumulation"
This is not correct, but rather the opposite is true. The model by Skoglund et al. shows
that during periods of rapid growth the accumulation is less than what equilibrium
partitioning would predict, because the kinetics of PCB uptake a slower than the
kinetics of growth! Also, as far as I know the model by Skoglund et al. applies to
phytoplankton and not zooplankton!
Other Issues



* Si n ? [v i ^ is ^:


j
til
WW
¦
;.S
>7
Page 6.2: Referring to TIP sediments, it is stated that "based on the vertically-
integrated coverage, the inventory of PCBs in the cohesive areas was 3208 kg (28.7
%), as compared to 7974 kg (71.1 %) in the non-cohesive areas." On the other hand,
page 5 of revised Appendix B states that 58 % of the PCB mass in TIP is in cohesive
type sediments, and 42 % in non-cohesive type sediments. Why is there such a
discrepancy in the relative importance of various sediment types for storing PCBs in the
TIP?
I would like to point the attention of those involved in this project on PCBs in the
Hudson River to a similar study conducted on the Eman River in Southern Sweden.
Though the river involved is considerably smaller than the Hudson River, the
contamination situation was similar in that PCB contaminated sediments had for years
provided a constant source of PCBs to the water and the biota living in that river. Also
the climate and the seasonality of run-off conditions are not unlike those in the upstate
New York. This river has now been subject to remediation, which involved the dredging
and land-filling of the contaminated sediments. A comprehensive description of that
study, and particularly the results of the environmental monitoring before, during and
after site remediation can be found in a Ph.D. thesis:
Bremle, G. 1997. Polychlorinated Biphenyls (PCB) in a River Ecosystem. Doctoral
dissertation, Lund University, Lund, Sweden, 144 pages (ISBN 91-7105-085-X)
75

-------
Peer Review of Hudson River PCBs Reassessment
Frank Wania, Page 14
For a copy of the thesis you may want to approach.
Dr. Gudrun Bremle
Lund University
Department of Ecology
Chemical Ecology and Ecotoxicology
Ecology Building
S-223 62 Lund
Sweden
E-mail: Gudrun.Bremle@ecotox.lu.se
Several aspects of that thesis - though by far not all of them - are also accessible as
peer-reviewed scientific publications:
Bremle, G., Okla, L., and Larsson, P. 1995. Uptake of PCBs in fish on a contaminated
river system: Bioconcentration factors measured in the field. Environ. Sci.
Techno!. 29, 2010-2015.
Bremle, G., and Ewald, G. 1995. Bioconcentration of polychlorinated biphenyls (PCBs)
in Chironomid larvae, oligochaete worms and fish from contaminated lake
sediment. Mar. Freshwater Res. 46, 267-273.
Bremle, G, and Larsson, P. 1998. PCB in the air during landfilling of contaminated lake
sediment. Atmos. Environ. 32, 1010-1019.
76

-------
APPENDIX D
LIST OF REGISTERED OBSERVERS OF THE PEER REVIEW MEETING

-------
United States
Jfc IhMU Environmental Protection Agency
%^Lal n Region 2
Hudson River PCBs Site
Modeling Approach Peer Review
Sheraton Saratoga Springs Hotel and Conference Center
Saratoga Springs, NY
September 9-10,1998
Observers
David Adams
Saratoga County Environmental Management Council
216 Stage Road
Charlton, NY 12019
518-399-1690
Fax: 518-399-1690
Mark Behan
President
Behan Communications
13 Locust Street
Glens Falls, NY 12801
518-792-3856
Fax: 518-745-7365
Kathy Cooke
Program Research Associate
Office of Fiscal Research - Policy Analysis
NVS Comptroller's Office
5th Floor AESOB
Albany, NY 12236
518-486-5433
Fax: 518-473-1900
Jonathan Butcher
Principal Hydrologist
Tetra Tech, Inc.
P.O. Box 14409
Research Triangle Park, NC 27709
919-485-8278
Fax: 919-485-8280
E-mail: jo3n@email msn.com
Thomas Echikson
Attorney
Sidley & Austin
1722 Eye Street, NW
Washington, DC 20006
202-736-8161
Fax: 202-736-8711
Leigh Foster
Executive Director
Arbor Hill Environmental Justice Corporation
200 Henry Johnson Boulevard
Albany, NY 12210
518-463-9760
Fax: 518-434-0392
John Connolly
President
Quantitative Environmental Analysis, LLC
305 West Grand Avenue
Montvale, NJ 07645
201-930-9890
Fax: 201-930-9805
David Glaser
Quantitative Environmental Analysis, LLC
305 West Grand Avenue
Montvale, NJ 07645
201-930-9890
Fax: 201-930-9805

-------
John Haggard
Technical Program Manager
Corporate Environmental Programs
General Electric Company
1	Computer Drive South
Albany, NY 12205
518-458-6619
Fax: 518-458-1014
E-mail: john.haggard@corporate.ge.com
Robert Henshaw
President
Hudson River Environmental Society
91 Louis Drive
West Sand Lake, NY 12196-3011
518-283-0415
Fax: 518-283-0415
E-mail: bobandnancy@worldnet.att.net
George Hodgson
Director
Saratoga County Environmental Management
50 West Hight Street
Ballston Spa, NY 12020
518-884-4778
Fax: 518-885-2220
Damien Hughes
Remedial Project Manager
New York Remediation Branch
Emergency and Remedial Response Division
U.S. Environmental Protection Agency
290 Broadway
New York, NY 10007
212-637-3957
Fax: 212-637-4284
E-mail: hughes.damien@epamail.epa.gov
Ridenour James
Research Scientist
New York State Department of Health
2	University Place, Room 240
Albany, NY 12203-3313
518-458-6409
Fax: 518-458-6372
E-mail: jar05@health.state.ny.us
Cara Lee
Environmental Director
Scenic Hudson
9 Vassar Street
Poughkeepsie, NY 12601
914-473-4440
Fax: 914-473-2648
Wilbert Lick
Professor
Department of Mechanical & Environmental
Engineering
University of California
Santa Barbara, CA 93106
805-893-4295
Angus Macbeth
Attorney
Sidley & Austin
1722 Eye Street, NW
Washington, DC 20006
202-736-8271
Fax: 202-736-8711
William McCabe
Deputy Division Director
Emergency and Remedial Response Division
U.S. Environmental Protection Agency
290 Broadway
New York, NY 10007
212-637-4405
Fax: 212-637-4439
E-mail: mccabe.bill@epamail.epa.gov
Robert Montione
Public Health Specialist
Bureau of Environmental Exposure Investigation
Division of Environmental Health Assessment
New York State Department of Health
11 University Place
Albany, NY 12203
518-458-6316
Fax: 518-458-6372
E-mail: rim04@health.state.ny.us
Kathleen O'Connor
Associate Engineer
Fluor Daniel GTI, Inc.
1245 Kinas Road
Schenectady, NY 12303
518-370-5631
Fax: 518-370-5864
E-mail: ko'connor@gtionline.com
William Ports, P.E.
Senior Engineer
Bureau of Central Remedial Action
Environmental Remediation
New York State Environmental Conservation
50 Wolf Road
Albany, NY 12233-7010
518-457-5637
Fax: 518-457-7925

-------
Thorn Randall
Reporter
The Post - Star
Lawrence and Cooper Streets
Glens Falls, NY 12801
518-792-3131 3271
Fax: 518-761-1255
E-mail: randall@poststar.com
James Rhea
Quantitative Environmental Analysis, LLC
290 Elwood Davis Road
Liverpool, NY 13088
315-453-9009
Fax: 315-453-9010
Ann Rychlenski
Communications Division
U.S. Environmental Protection Agency
290 Broadway
New York, NY 10007
212-637-3672
E-mail: rychlenski.ann@epamail.epa.gov
John Santacrosse
Chair of the Environmental Liaison Committee
New York Audubon Society
P.O. Box 3705
Albany, NY 12203
518-489-9945
Rich Schiafo
Environmental Associate
Scenic Hudson
9 Vassar Street
Poughkeepsie, NY 12601
914-473-4440
Fax: 914-473-2648
Melvin Schweiger
Manager Hudson River Project
Corporate Environmental Programs
General Electric Company
1 Computer Drive South
Albany, NY 12205
518-458-6648
Fax: 518458-1014
E-mail: melvin.schweiger@corporate.ge.com
Douglas Tomchuk
Remedial Project Manager
Emergency and Remedial Response Division
U.S. Environmental Protection Agency
290 Broadway
New York, NY 10007
212-637-3956
Fax: 212-637-4284
E-mail: tomchuk.doug@epamail.epa.gov
Stephen Wilson
Executive Director
Hudson River Environmental Society
6626 Stitt Road
Altamont, NY 120094523
518-861-8020
Fax: 518-861-8020
E-mail: stephenwilson1@compuserve.com
C. Kirk Ziegler
Vice President
Quantitative Environmental Analysis, LLC
305 West Grand Avenue
Montvale, NJ 07645
201-930-9890
Fax: 201-930-9805

-------
APPENDIX E
AGENDA FOR THE PEER REVIEW MEETING

-------
^	United States
Environmental Protection Agency
W 1^1	Region2
Hudson River PCBs Site
Modeling Approach Peer Review
Sheraton Saratoga Springs Hotel and Conference Center
Saratoga Springs, NY
September 9-10, 1998
Agenda
Meeting Chair:	Alan Maki, Exxon
Meeting Facilitator:	Jan Connery, Eastern Research Group, Inc. (ERG)
WEDNESDAY,	SEPTEMBER 9 ,1998
8:00AM	Registration/Check-In
9:00AM Welcome Remarks and Panel Introduction (Disclosures)
Jan Connery, ERG, and Alan Maki, Exxon
9:20AM Overview, Background, Meeting Structure, and Objectives
Jan Connery
9:30AM Presentation on the Development of the Model
Doug Tomchuk, EPA Region 2
10:00AM
BREAK
10:15AM Charge to the Panel/Highlights of Premeeting Comments
Alan Maki
10:50AM
11:40AM
I2:OONOON
1:00PM
1:45PM
Discussion on Question A: Is EPA Using Appropriate Models, Datasets, and
Assumptions on Which to Base a Scientifically Credible Decision?
Discussion facilitated by Jan Connery and Alan Maki
Summary of Discussions on Question A
LUNCH
Discussion of Question B: Will the Models be Able to Answer the Principal
Study Questions as Stated in the PMR?
Summary of Discussions on Question B
» Printed on Recycled Paper

-------
WEDNESDAY, SEPTEMBER 9, 1998 (Continued)
2:00PM Discussion on Question C1, C2, and C3: Are the Modeling Approaches For
Developing Quantitative Relationships Appropriate...
3:15PM	BREAK
3:30PM	Discussion continues on Questions C1, C2, and C3
4:15PM	Summary of Discussions on Questions C1, C2, and C3
4:35PM	Observer Comments
5:15PM Review of Charge for Day Two
Jan Connery and Alan Maki
5:30PM ADJOURN
THURSDAY, SEPTEMBER 10, 1998
8:30AM Planning and Logistics
Jan Connery
8:40AM Discussion on Questions C4 and C5
9:40AM Summary of Discussions on Questions C4 and C5
10:15AM BREAK
10:45AM Discussion on Question D: Are There Any Changes in the Work Effort That
Would Significantly Improve the Outcome?
11:15AM	Summary of Discussions on Question D
11:45PM	LUNCH
1:00PM	Discussion on Question E
2:00PM	Summary of Discussions on Question E
2:30PM	Observer Comments
3:15PM Recommendations and Chair's Summary
Jan Connery and Alan Maki
4:15PM Closing Remarks
4:30PM ADJOURN

-------
APPENDIX F
SUMMARIES OF OBSERVERS' COMMENTS

-------
List of Observers Making Comments
Day #1 (September 9,1998):
David Adams, Saratoga County Environmental Management Council
Leigh Foster, Arbor Hill Environmental Justice Corporation
Wilbert Lick, University of California at Santa Barbara
Kirk Ziegler, Quantitative Environmental Analysis
David Glaser, Quantitative Environmental Analysis
Tim Rhea, Quantitative Environmental Analysis
John Connolly, Quantitative Environmental Analysis
Day #2 (September 10,1998):
George Hodgson, Saratoga County Environmental Management Council
Wilbert Lick, University of California at Santa Barbara
Kirk Ziegler, Quantitative Environmental Analysis
David Glaser, Quantitative Environmental Analysis
John Connolly, Quantitative Environmental Analysis
Robert Henshaw, Hudson River Environmental Society
Leigh Foster, Arbor Hill Environmental Justice Corporation
The remainder of this Appendix summarizes the comments made by the observers listed above.
Comments are summarized in the order in which they were presented

-------
Appendix F—Summaries of Observers' Comments
Day #1, Comments from David Adams, Saratoga County Environmental Management Council
Mr. Adams' comments addressed both the peer review process and technical aspects of
the EPA modeling approach. Regarding the peer review process, Mr. Adams noted that
independent peer review is important to help concerned citizens of the Upper Hudson River area
understand why EPA and GE may make different interpretations on technical issues for this site.
Mr. Adams thanked the peer reviewers for their input, which ultimately may help reconcile the
different interpretations. Mr. Adams then offered several recommendations for implementing
future peer reviews. First, for the sake of improving the peer review process, Mr. Adams
suggested that the peer reviewers be given more time and information to conduct thorough
reviews. He also recommended that the peer review report document this recommendation.
Second, noting that some peer reviewers thought it would have been more useful to hear EPA's
overview of the site history at the beginning of the peer review instead of at the peer review
meeting, Mr. Adams suggested that future peer reviews include "orientation" presentations to
give peer reviewers a better basis on which they can conduct their reviews. Third, Mr. Adams
recommended that EPA publish a response to the comments raised in the current peer review,
even though a project schedule distributed at the meeting did not explicitly state that a response
would be prepared. Finally, recognizing the experience gained by the peer reviewers during the
current meeting, Mr. Adams recommended that ERG consider selecting some of the same peer
reviewers for future peer reviews.
Regarding the technical aspects of the EPA modeling approach, Mr. Adams was
concerned that the models may not be able to simulate river conditions during the Allen Mill
incident of 1991, which Mr. Adams called a "discontinuity" in the source of PCBs to the Upper
Hudson River. More specifically, Mr. Adams thought it was not clear whether EPA's models can
forecast through this "discontinuity" of PCB sources to the river. Mr. Adams thought the
modeling approach should attempt to forecast river conditions before the 1991 incident separately
from forecasting river conditions following the 1991 incident. He thought the peer review report
should address this specific concern regarding the modeling approach. Finally, Mr. Adams noted
that the peer reviewers had addressed most of his other questions and comments, such as the need
for modeling individual PCB congeners in addition to modeling total PCBs.
Day #1, Comments from Leigh Foster, Arbor Hill Environmental Justice Corporation
Mr. Foster's comments addressed the peer review process and the relevance of the
ongoing reassessment studies to potentially exposed populations, particularly communities of
color in Albany and in other parts of the Hudson Valley. Noting that these communities are
already affected by lead exposure, Mr. Foster indicated that human health risks from PCBs
presents "yet another obstacle" that communities of color in this area (especially subsistence
fishers and residential communities along the river) must overcome.

-------
Appendix F—Summaries of Observers' Comments
In reference to the peer review process, Mr. Foster suggested that, by assembling peer
review teams that are just as dynamic and diverse as the Hudson River community, ERG and EPA
can ensure that the perspectives and opinions provided by the reviewers would be equally dynamic
and diverse. Mr. Foster believed that these diverse opinions would ultimately help the modeling
efforts. More specifically, Mr. Foster recommended that the review teams should have better
community representation by including community experts and community "systems dynamics
experts." Mr. Foster noted that many of these experts work in academia (including at the
University of California at Berkeley and at the Massachusetts Institute of Technology) in the fields
of human health risk assessment, human health risk monitoring, and modeling. He indicated that
some of these experts have already worked with the Arbor Hill community. Finally, citing an
analogy to his experiences with designing paries, Mr. Foster emphasized the importance of
assembling diverse teams for projects of local interest, including people with "native knowledge"
of the resource being considered (i.e., the Hudson River). Mr. Foster mentioned that he was
pleased that one of the reviewers was from Cornell and therefore might have "native knowledge"
of the Hudson River system.
Regarding the ultimate impact of the site reassessment on potentially exposed populations,
Mr. Foster made a series of comments, observations, and suggestions. For example, he expressed
concern that the reviewers did not discuss whether human health studies would be incorporated in
the modeling, especially considering that an ultimate goal of the modeling efforts is to evaluate
human health concerns. Mr. Foster mentioned that he has recently attended public health
meetings at which other agencies have been trying to determine both short-term and long-term
human health impacts. He indicated that it would be useful if the current modeling could
determine and document the cycles over which human health impacts might occur. Further, he
thought that the ongoing site reassessment should include "short-term human health risk
mitigation" in addition to the current modeling effort, which focuses on the long-term impacts of
remediation strategies, such as "re-engineering" the Hudson River. Mr. Foster emphasized as a
key issue that risk mitigation is cost-effective and could increase the quality of life for residents
who live along the Hudson River right now, while the modeling efforts only address future
scenarios. Mr. Foster did acknowledge that EPA and the State of New York have issued human
health risk advisories for some river communities, but he noted that these advisories have "failed
miserably." The failure of these advisories, according to Mr. Foster, has had the greatest impact
on communities in dire need of information: communities of color, communities of non-native
English speakers, and communities of subsistence fishers who depend on the Hudson River as a
source of nutrition. Noting that it is his job to communicate and articulate health risks and
technical information to these communities, Mr. Foster indicated that the community members
will be best served right now if the peer reviewers communicate their ideas, findings, reservations,
and enthusiasm regarding human health issues.
Mr. Foster closed his comments by mentioning specific concerns expressed by community
members. First, Mr. Foster informed the peer reviewers that residents have expressed great
concerns over how flooding events might affect their health, especially since floods have occurred

-------
Appendix F—Summaries of Observers' Comments
in the past. More specifically, Mr. Foster wondered what effects might result when basements are
flooded with water containing sequestered PCBs. Mr. Foster noted that residents may already
understand dangers posed by radon in their homes, but they generally do not understand dangers
posed by PCBs. Mr. Foster also mentioned that he had been asked if there was some way to
incorporate more residential studies in the ongoing study of the river system; he said these might,
for example, consider the residential areas that will "experience sequestration of PCBs" during
flood, or evaluate the impacts of PCBs on older housing in the city of Troy. Mr. Foster expressed
gratitude that the reviewers completed such a thorough evaluation of the modeling approach, but
noted that a lot of the peer reviewers' comments were critical. He hoped that EPA will ultimately
use these critical comments to improve the modeling approach. Finally, Mr. Foster indicated that
he felt "informed and somewhat empowered" by the information the peer reviewers had provided.
Day #1, Comments from Wilbert Lick, University of California at Santa Barbara
Mr. Lick's comments addressed technical aspects of EPA's plan to model sediment
resuspension and PCB flux from the sediment to the water. Before discussing these topics, Mr.
Lick first mentioned that he found the peer review discussions both informative and interesting.
Mr. Lick then indicated that, to predict PCB transport and fate, one must first understand
sediment transport, PCB partitioning, and flux of PCBs from sediment to water. Mr. Lick further
noted that resuspension, deposition, bioturbation, and diffusion all affect this flux of PCBs.
Regarding how to model sediment resuspension, Mr. Lick indicated that sediment
resuspension properties (which are dependent on sediment particle size, mineralogy, sediment
bulk density, and gas bubbles resulting from decay of organic material) vary by orders of
magnitude through river systems. Mr. Lick said he did not know how to predict these
parameters, but he noted that sediment erosion rates can be measured at the surface, as a function
of depth, and also as a function of shear stress. By coupling measured erosion rates with a
"reasonably decent" hydrodynamic model and sediment transport model, Mr. Lick indicated that a
modeling approach can predict sediment fete and transport well both during 100-year flood events
and during low-flow periods. Mr. Lick suggested that the models described in EPA's PMCR are
currently insufficient for characterizing both types of flow events. To put sediment resuspension
into context, Mr. Lick noted that significant transport of PCBs and sediments occurs during high-
flow events in almost every other river he knows of. As an example, Mr. Lick mentioned that
data for the Fox River, another river with locks and dams, indicate that over 80 percent of
sediment transport "in a moderate year" occurs during less than 20 percent of that time.
Regarding how to model the flux of PCBs from sediment to water, Mr. Lick noted that
EPA's models approximate this flux using the concept of an "active layer" or a "mixed layer" of
sediments, with a thickness that EPA arbitrarily set at 5 centimeters (cm). Mr. Lick indicated that
the sediment transport algorithms in EPA's models are quite sensitive to the thickness of the
active layer. For example, Mr. Lick noted the model would predict that decay of PCBs in the

-------
Appendix F—Summaries of Observers' Comments
Hudson River would take twice as long if the active layer thickness were assumed to be 10 cm
(instead of S cm); similarly, he noted the model would predict that decay of PCBs would take half
as long if the active layer thickness were 2.5 cm (again, instead of 5 cm). Based on these
observations, Mr. Lick indicated that the active layer thickness parameter is "absolutely crucial"
for long-term prediction of the models. Mr. Lick also commented that EPA's report did not
really explain how the active layer thickness of 5 an was selected, except "on the basis of some
judgment." Mr. Lick recommended developing and using a model that better characterizes the
flux of PCBs from the sediment to the water by considering sediment resuspension and
deposition, bioturbation, and diffusion as separate physical processes. Mr. Lick then noted that
such an improved model could better answer questions regarding the effects of the Allen Mill
discharge. More specifically, he estimated that the discharge probably sent millimeters of
sediments throughout the Thompson Island Pool, and he suggested that EPA's models would not
be able to characterize this discharge by assuming a sediment active layer thickness of 5 cm. Mr.
Lick indicated that "a more detailed sediment-water flux calculation" would be needed to model
this event. Finally, Mr. Lick commented that the models need to "get the partitioning right" and
he noted that partitioning kinetics (the time over which absorption and desorption occurs) are an
important factor for modeling the flux of PCBs. Mr. Lick indicated some of his experiments have
found that highly chlorinated PCBs have partitioning rates on the scale of months or years.
Day #/, Comments from Kirk Ziegler, Quantitative Environmental Analysis (QEA)
Mr. Ziegler's comments focused on how he thought EPA's modeling approach,
specifically the HUDTOX model, should treat sediment resuspension and deposition processes.
Mr. Ziegler began his comments by indicating that he thought the best approach to modeling these
processes in the Hudson River, and the approach that EPA should adopt, is to use mechanistic
models. Mr. Ziegler then noted that EPA chose to use an empirical model (i.e., HUDTOX) to
describe the sediment resuspension and deposition processes, an approach he does not
recommend. Mr. Ziegler mentioned that "several important, critical processes" that affect
sediment resuspension and deposition need to be incorporated into the empirical model, in the
event that EPA chooses not to adopt a mechanistic approach.
After reiterating that using mechanistic models is the best approach to characterize
sediment resuspension and deposition processes in the Hudson River, Mr. Ziegler clarified that a
mechanistic model would include formulations that describe sediment resuspension and deposition
that are based on experiments conducted both in the lab and in the field. Mr. Ziegler indicated
that site-specific data should then be used to specify parameters used in the mechanistic
formulations. Mr. Ziegler noted that the data, science, and models are available for developing
and applying mechanistic models to the Upper Hudson River, just as mechanistic models have
been developed for other rivers. Mr. Ziegler acknowledged that EPA has attempted to
incorporate some mechanistic formulations into its proposed 100-year flood model, but he noted
that EPA proposes using an empirical approach to modeling sediment transport in the HUDTOX

-------
Appendix F—Summaries of Observers' Comments
model—the model that Mr. Ziegler thought was "key" to predicting long-term fete of PCBs in the
river system. As he understands it, Mr. Ziegler said, EPA will (1) develop empirical functions
relating sediment resuspension and deposition rates to local shear stress and other properties, (2)
use data collected during a 1994 spring flood to calibrate the empirical functions, and (3) use data
collected from other flood events to validate the empirical functions. Mr. Ziegler found such a
modeling approach to be "very problematical" and thought it should not be used.
Mr. Ziegler recommended that the modeling approach at least recognize and incorporate
four different processes when calculating sediment resuspension and deposition velocities, even if
EPA cannot avoid using its proposed empirical modeling approach. First, Mr. Ziegler thought the
modeling should account for "bed armoring," which he defined as the process by which the
amount of sediment that can be eroded is limited at a particular shear stress. Mr. Ziegler indicated
that various bed properties, such as particle size heterogeneity and the cohesiveness of sediments,
affect the extent of bed armoring. Mr. Ziegler noted that bed armoring is important because it can
create discontinuities in the sediment resuspension velocity functions, which, if not accounted for
by a model, can lead to incorrect predictions of sediment erosion during a flood. Second, Mr.
Ziegler thought EPA's models should better address sediment resuspension during periods of low
flow. More specifically, he noted that his previous modeling analyses, laboratory work, and field
work have all found that resuspension from sediment beds in the Hudson River is negligible
during periods of low to moderate flow, or during periods with flow rates less than 4,000 cubic
feet per second. Based on these findings, Mr. Ziegler suggested that PCB fete and transport
modeling does not need to consider resuspension during periods of low flow and that EPA's
calculated sediment resuspension velocities should reflect the results of his low-flow studies.
Third, Mr. Ziegler recommended that EPA's modeling approach should address the time history
of bed properties. If models do not consider temporal variations in "erosional characteristics" of
sediment beds, Mr. Ziegler noted that the calculated resuspension velocities, and the modeling
results in general, would have "tremendous amounts of uncertainty." Finally, Mr. Ziegler
expressed concern over the issue of spatial and temporal variability in the composition of clay, silt,
and sand in suspended sediments in the water column. Mr. Ziegler indicated that these "column
composition effects" cause complex changes in sediment deposition rates at different locations in
the Hudson River. Further, he noted that modeling approaches that neglect these effects, such as
EPA's proposed modeling approach, will create an "unrealistic and inaccurate" representation of
sediment deposition in river systems.
Day #1, Comments from David Glaser, Quantitative Environmental Analysis (QEA)
Mr. Glaser's comments addressed the bioaccumulation models that EPA developed. Prior
to critiquing these models, Mr. Glaser first distributed a written summary of his comments to the
peer reviewers. Mr. Glaser then identified the two functions of the bioaccumulation models: (1)
"to estimate the average concentration of PCBs" in fish and (2) "to estimate the variability" in the
average concentrations. Before addressing these functions in detail, Mr. Glaser presented his

-------
Appendix F—Summaries of Observers' Comments
major findings. First, concerning average concentrations, Mr. Glaser strongly agreed with the
sentiments of the reviewers that a "mechanistic, time-variable model" needs to be used to predict
levels of PCBs in fish. He further noted that QEA has developed such a model for GE and has
made the model available to EPA. Regarding estimates of variability, Mr. Glaser found "severe
problems" with the probabilistic food chain model that EPA has developed. Namely, Mr. Glaser
noted that EPA's model is static, is based upon the variability of the data, and is validated against
this same data. As an alternative, Mr. CHaser suggested that EPA simply use the available data to
estimate the variability ofPCB concentrations in fish. Mr. CHaser indicated that there are plenty
of data available for several species of fish to estimate variability in the concentrations and to
estimate the uncertainty associated with this variability. For three particular species of fish
(pumpkinseed, brown bullhead, and largemouth bass), Mr. Glaser noted that fish concentration
data are available for three monitoring locations over a 20-year period, with many of the sampling
events measuring concentrations in roughly 20 fish.
Commenting in detail on the use of bioaccumulation models to estimate average
concentrations of PCBs in fish, Mr. Glaser first mentioned that he would not describe why he
thought a mechanistic, time-variable model is needed, largely because he sensed "general
agreement" among the peer reviewers on this recommendation. Mr. Glaser instead described in
detail the bioaccumulation model developed by QEA, which he called a "full life cycle, full food
web, bioenergetic based, toxicokinetic model." Mr. CHaser noted that this model incorporates
time-dependent lipid contents—a factor that one of the reviewers thought should be included in
such models. He further noted that considering time-varying lipid contents was a "key
component" in his modeling analysis and that models assuming constant lipid contents find very
different results than models allowing lipid contents to vary with time. As an example of the
importance of considering time-dependent lipid contents, Mr. Glaser mentioned that the lipid
content of largemouth bass can vary by an order of magnitude from year to year. On the issue of
interfacing the bioaccumulation model with other models, Mr. Glaser indicated that the QEA
model links directly with the sediment and water concentrations output by a time-variable fete and
transport model. Mr. Glaser then noted that the QEA bioaccumulation model accounts for
several other factors that the peer reviewers found important. As an example, Mr. Glaser
indicated that the model accounts for feeding rates that are seasonal (or temperature-dependent).
Referring to a comment made earlier by EPA concerning the lack of available data, Mr. Glaser
mentioned that "a tremendous amount of information" (e.g., species-specific bioenergetic
information; site-specific, species-specific growth rates; and 20 years of toxicokinetic
experimentation) are available from the scientific literature to parameterize bioaccumulation
models. Mr. CHaser then indicated that QEA's time-varying, mechanistic model has been used to
identify the key uncertainties in the Hudson River system, such as the incomplete understanding of
the food web in the Thompson Island Pool. Mr. Glaser noted that his modeling analyses have
already prompted a field study to characterize the food web in this area. Noting other aspects of
QEA's model, Mr. CHaser mentioned that QEA's bioaccumulation model has been extended to
account for different PCB homologues—an improvement that is "just one step away" from
modeling individual PCB congeners. Further, Mr. Glaser noted that the model has helped QEA

-------
Appendix F—Summaries of Observers* Comments
better understand the role of the "active layer" or "the bioavailable layer" within sediments.
Finally, Mr. Glaser said QEA's bioaccumulation model, including all information used in the
model's development, is available to EPA for review and use.
Before discussing his second major topic (using models to predict variability), Mr. Glaser
clarified a comment made by a peer reviewer during the meeting regarding the distribution of PCB
homologies, particularly BZ4, in the Hudson River sediments. Mr. Glaser explained that the
elevated concentrations of BZ4 in some of the sediment samples likely result from the
dechlorination of PCBs that is known to occur in sediments, particularly at depth. Thus, Mr.
Glaser believed that reports of relatively high concentrations of BZ4 (which he noted was a
terminal dechlorination product) are probably not the result of sampling or analytical errors.
Further elaborating, Mr. Glaser mentioned that BZ4 bioaccumulates "very little," thus causing
"very little" BZ4 to be found in fish. Referring the peer reviewers to the package of comments he
distributed at the beginning of his presentation, Mr. Glaser noted that he has used observations of
dechlorination to understand the "bioavailable depth" of sediments. For example, knowing that
dechlorinated PCBs are most prevalent in sediments at depth and "relatively fresh" PCBs are
more prevalent in surface sediments, Mr. Glaser mentioned that he could use his model "in a
diagnostic fashion" to understand to which sources of PCBs fish are most likely exposed.
Summarizing the results of his modeling analyses, Mr. Glaser reported that fish appear to be
exposed to a "relatively un-dechlorinated source," which would not be the buried PCBs.
Regarding estimates of variability in the modeled concentrations of PCBs in fish, Mr.
Glaser first stated that EPA's probabilistic food chain model confuses variability and uncertainty.
Mr. Glaser defined variability as "the true variance in the population concentration of PCBs" and
uncertainty as "what we know about the concentration of PCBs in fish." To give evidence of his
claim, Mr. Glaser noted that EPA's model uses measured concentrations of PCBs in invertebrates
to estimate the relationship between (1) PCB concentrations in the sediment and (2) PCB
concentrations in invertebrates. Mr. Glaser disapproved of this method of deriving
bioaccumulation factors because the available data include "several important sources of
uncertainty," in addition to data variability. Most importantly, Mr. Glaser indicated that sediment
concentrations of PCBs vary by orders of magnitude, even over short distances. Without
knowing exactly what sediments invertebrates eat, Mr. Glaser contended that one cannot
determine the right bioaccumulation factor, thus introducing a "very big" source of uncertainty in
EPA's bioaccumulation model. (Mr. Glaser acknowledged that it is "nearly impossible" to collect
and analyze the exact sediments that invertebrates eat.) With this uncertainty in the estimated
bioaccumulation factors, Mr. Glaser noted that EPA's model, which is designed to calculate
variability, actually calculates a mixture of uncertainty and variability. Mr. Glaser indicated that
this "mixture" of terms has no physical meaning. As a second issue, Mr. Glaser said EPA "used
the answer to solve the problem." As an example, Mr. Glaser noted that EPA used variability in
the predator data to estimate variability in the trophic transfer from forage fish to predators. Mr.
Glaser claimed that this approach was "using the answer to develop the model": variability in the
predator data is supposed to be an output of the model, yet EPA used these variability data

-------
Appendix F—Summaries of Observers' Comments
essentially as an input when parameterizing the model. Finally, agreeing with a comment raised
by a peer reviewer, Mr. Glaser mentioned that EPA's proposed model is static and therefore
provides no "predictive value." Based on his list of inadequacies for the EPA bioaccumulation
model, Mr. Glaser recommended that EPA estimate variability in the concentrations of PCBs in
the fish simply by using the large set of available data, rather than by attempting to model the
uncertainty with the probabilistic bioaccumulation model.
Day #1, Comments from Jim Rhea, Quantitative Environmental Analysis (QEA)
Mr. Rhea began by commending the peer reviewers for their work, especially considering
the limited time they had to review the PMCR. Mr. Rhea noted that many scientists at the
meeting have spent nearly 10 years working on the Upper Hudson River project. Mr. Rhea also
acknowledged that the reviewers were "handcuffed" by reviewing a 2-year old report, particularly
because he thought many advances have since been made in understanding the Hudson River data.
Mr. Rhea then described four areas in which understanding of the river system has evolved since
the PMCR was published.
First, Mr. Rhea responded to concerns raised by the peer reviewers regarding how to
interpret historical data collected in the Hudson River. Noting that the river models "are only as
good as the data upon which" the models are calibrated, Mr. Rhea mentioned that GE and EPA
have both emphasized in their work the importance of understanding what the available data
actually mean. More specifically, Mr. Rhea noted that both GE and EPA have carefully analyzed
and interpreted nearly 20 years of fish, sediment, and water column data that were collected by
different organizations using different analytical techniques. As a result, Mr. Rhea claimed that
scientists from GE and EPA now "have a firm handle" on the significance of the different
analytical techniques Otherwise stated, Mr. Rhea indicated that both GE and EPA have a much
better understanding of exactly what historical monitoring data actually represent.
Second, Mr. Rhea addressed the "excess load" of PCBs in the Thompson Island Pool,
which he described as a "key issue" of the PMCR and which also was an issue of discussion
among the peer reviewers. Mr. Rhea mentioned that GE conducted an extensive study in 1997 to
understand why the models predicted, yet could not explain, the excess load in this section of the
Hudson River. Mr. Rhea indicated that QEA used its models in "a diagnostic fashion" to resolve
this issue. Mr. Rhea reported that the modeling analyses found a bias at a downstream sampling
station accounted for "a majority of the excessive load." Mr. Rhea further noted that the PCB
loading leaving the Thompson Island Dam, from what he called an "un-biased station," is
consistent with diffusive transport mechanisms and a sediment-water exchange rate of
approximately 2 centimeters per day. Mr. Rhea indicated that this exchange rate is consistent
with, though possibly a little higher than, exchange rates observed in other river systems.

-------
Appendix F—Summaries of Observers' Comments
Third, addressing reviewers' discussions of groundwater advection, Mr. Rhea described
the findings of a recent field study that addressed this issue. He noted that "seepage meter
measurements" and piezometer measurements were conducted to measure seepage rates and
hydraulic gradients during the spring of 1997. He further indicated that this field study found that
groundwater advection was an "insignificant process" that "was not contributing much PCBs to
the water column" during the spring flow conditions.
Fourth, Mr. Rhea commented on EPA's use of a limited data set to calibrate the
HUDTOX model. More specifically, Mr. Rhea noted that EPA used data collected from a
9-month period in 1993 to calibrate the model—a time that followed a 2-year period of elevated
loadings of PCBs to the Hudson River system in the form of dense non-aqueous phase liquids.
Because he thought these elevated loadings were not representative of the Hudson River at
"steady state," Mr. Rhea cautioned against calibrating the HUDTOX model data from this limited
time period and then projecting these limited calibration results to longer time periods. As an
alternative, Mr. Rhea recommended that EPA use more historical data, "from an extended period
of time," to calibrate its models in order to avoid having the model results biased by a limited
calibration period.
Day Ml, Comments from John Connolly, Quantitative Environmental Analysis (QEA)
Mr. Connolly's comments primarily addressed two issues: (1) the calibration and
validation of EPA's models and (2) the location of the upstream boundary for the HUDTOX
model. Regarding the calibration and validation of the models, Mr. Connolly presented numerous
comments and suggestions. Mr. Connolly first noted that "a wealth of data" are available for
calibrating the Hudson River models: water column data for 20 years; fish data for 20 years;
"major collections" of sediment data from the late 1970s, the mid-1980s, the early 1990s, and the
mid-1990s. Mr. Connolly then presented what he considered to be "key comparisons" that he
thought should be made between the available data and the Hudson River models. Referring to an
observer's comments made earlier, Mr. Connolly noted that the validity of EPA's modeling results
will depend largely on the successful calibration of sediment transport mechanisms. Mr. Connolly
mentioned that EPA currently proposes to calibrate its empirical fate and transport model
(HUDTOX) by "essentially fitting a sediment transport model to suspended solids data in the
water column." As summarized below, Mr. Connolly then described in detail how he proposes
EPA should calibrate its models.
In modeling the fate and transport of suspended solids, Mr. Connolly indicated the only
parameter "that matters" is the net transport of solids between the sediment and the water. More
specifically, in terms of suspended solids, Mr. Connolly suggested that absolute sediment
resuspension rates and absolute sediment deposition rates "are irrelevant," provided the difference
between these rates (i.e., the net transport rate) is accurate. If the estimated net transport rate of
sediments is indeed accurate, he indicated the models will predict "the right amount of material"

-------
Appendix F—Summaries of Observers' Comments
transferring between the two phases and will predict the correct total suspended solid
concentrations. Noting that suspended solids can originate both from the river bed and from the
watershed, Mr. Connolly also mentioned that the "amount of solids loading" in the water can
affect model calibration of suspended solids data: if the model is not calibrated using correct
loadings from the watershed, then the model will not predict the correct net resuspension rates.
Mr. Connolly reiterated that, even assuming the models have valid solids loading data from the
watershed, model calibration will only determine the "net flux" of solids from the sediments. Mr.
Connolly then summarized that calibration of the model during "the 1994 flood event" will only
generate the "net transport of sediment" from the river bottom into the water. Mr. Connolly
emphasized that knowledge of the net transport of sediment is sufficient for modeling the
transport of solids, but is insufficient for modeling the transport of PCBs.
Mr. Connolly stressed that, when modeling transport of PCBs, the absolute sediment
resuspension rates and the absolute sediment deposition rates are critical parameters, because
PCBs enter the water through resuspension of contaminated sediments. Mr. Connolly then
suggested that one can determine whether the estimated absolute transfer rates are correct by
computing the amounts of PCBs in the water column during high-flow events. He explained that
high-flow events are characterized by "clean solids" entering the modeling region from the
watershed and "contaminated solids" entering the modeling region through sediment
resuspension. By combining these two factors, Mr. Connolly noted that the models can predict
the amounts of PCBs in the water column. Mr. Connolly then suggested that the Hudson River
database includes observations from "a multitude of high-flow events" over the last 20 years.
More specifically, Mr. Connolly cited events with flows peaking at 34,000 cubic feet per second
(e.g., for high flows in 1983 and in 1998) and other events with flows peaking at slightly lower
levels. Noting that PCB and total suspended solid data exist for several of these high-flow events,
Mr. Connolly emphasized the importance of challenging EPA's models with the data sets that
characterize these periods of high flow.
Citing another source of data that can be used to validate the models, Mr. Connolly noted
that EPA's 1992 high resolution coring samples with cesium dating could be used to assess
sediment resuspension and deposition. He explained that these cesium-dated samples provide
estimates of long-term deposition rates, or "how many centimeters per year of sediment are
accumulating." He then suggested that EPA compare what its models predict for long-term
deposition rates at selected locations to what the high resolution coring samples suggest as long-
term deposition rates. Mr. Connolly thought it was important that such comparisons be made.
For his final point related to data calibration and validation, Mr. Connolly addressed
sediment transport during low-flow conditions. He first described the mechanisms that affect
sediment transport during low flow, such as "diffusion out of the pore water," "bioturbation
effects," and "advection between the sediment and the water column." Mr. Connolly suggested
that EPA's fate and transport models would not be able to treat these mechanisms separately, but
would rather combine these mechanisms into a single term that would be quantified by model

-------
Appendix F—Summaries of Observers' Comments
calibration. Mr. Connolly referred to this single term, which characterizes the amounts of PCBs
leaving the sediment during low-flow conditions, as "a free knob" in EPA's models. Mr.
Connolly then noted that upstream and downstream weekly water column monitoring data are
available from 1991 to the present, and additional (but less frequent) monitoring data are available
dating back to the 1970s. Mr. Connolly considered it important that EPA calibrate its models
against the available water column monitoring data during low-flow conditions, including those
from different seasons. Mr. Connolly suggested that this use of the historical water column data
is "the only way" that EPA can validate both the "flux rates from sediment to the water" and the
seasonal variations in these flux rates.
Addressing his second major issue (i.e., the upstream boundary for EPA's modeling
approach), Mr. Connolly asked the peer reviewers to reconsider their recommendation that EPA
move the upstream boundary of its models from Rogers Island to a location "above GE's Hudson
Falls plant." Noting the extensive monitoring data available at Rogers Island from 1977 to 1998,
Mr. Connolly suggested that Rogers Island is a "convenient place" for the upstream boundary of
EPA's model because the monitoring data adequately characterize the flux of PCBs into the
Thompson Island Pool. Mr. Connolly then indicated that moving the modeling boundary further
upstream would necessitate knowing the load of PCBs entering the Hudson River from GE's
plant. Without knowing exactly what this load is, Mr. Connolly suspected that modelers would
have to "twist knobs" so that the model would reproduce the large set of monitoring data
available for Rogers Island. This approach, according to Mr. Connolly, would not provide any
greater insight into the system than would leaving the upstream boundary of the model at Rogers
Island. Mr. Connolly further noted that it would be especially difficult to model future conditions
with an upstream boundary above the Hudson Falls plant, particularly because no one knows the
current PCB loadings from the plant site or how these loadings might change in the future. Mr.
Connolly therefore recommended leaving the upstream boundary of the model unchanged and
modeling different scenarios of future PCB loadings into the river. He provided several examples
of loading scenarios: the PCB loading remains constant at its present value, the PCB loading
steadily decreases at the rate it has been decreasing, or "GE successfully eliminates all loading" so
that loading decreases to atmospheric levels. In conclusion, Mr. Connolly suggested that "we do
not gain much from moving the model upstream," and he recommended again that the peer
reviewers reconsider this issue.
Day #2, Comments from George Hodgson, Saratoga County Environmental Management
Council
Mr. Hodgson began by introducing himself as the Director of the Saratoga County
Environmental Management Council (EMC). Mr. Hodgson then explained that EMCs are
"environmental citizen advisory groups," and he noted that the EMC he directs was established by
the Saratoga County Board of Supervisors. Citing the level of concern among residents regarding
the water quality of the Hudson River, which he noted forms two borders of Saratoga County,

-------
Appendix F—Summaries of Observers' Comments
Mr. Hodgson mentioned that the Saratoga County EMC has been, and continues to be, "actively
involved in commenting on" EPA's assessments of PCBs in the river. Mr. Hodgson noted that
David Adams, who presented comments on behalf of the Saratoga County EMC on the first day
of the peer review meeting, could not attend the second day of the meeting. Mr. Hodgson then
presented three observations or comments regarding the peer review.
First, Mr. Hodgson noted that the Saratoga County EMC has expressed concern "several
times in the past" that EPA conducted its data acquisition program before developing its models.
Explaining his concerns, Mr. Hodgson wondered if the data that EPA collected is adequate for
running the models and if EPA developed its models simply to "fit the database." Second,
echoing concerns raised by some of the peer reviewers, Mr. Hodgson thought "a lack of adequate
information" is currently available to make judgments. Mr. Hodgson noted that the Saratoga
County EMC stresses that all "pertinent information" should be made available so that "a
comprehensive peer review" can be conducted for this she. Mr. Hodgson indicated that the EMC
has recommended a comprehensive peer review in the past. He explained that a comprehensive
peer review would consider public comments, data that GE has generated, and information
provided by EPA Third, Mr. Hodgson had several comments regarding the notes the meeting
chair wrote on an easel located at the front of the meeting room. Mr. Hodgson found the notes
difficult to read, especially from the back of the meeting room, and he thought ERG should make
these notes available to the public "in some form." Mr. Hodgson ended his comments by
commending the peer reviewers on their efforts, especially considering that the Saratoga County
EMC had "felt very strongly" about conducting periodic peer reviews of the site reassessment.
Mr. Hodgson concluded by indicating that he was hopeful that the current peer review was
conducted in part to address the previous recommendations made by the Saratoga County EMC.
Day #2, Comments from Wilbert Lick, University of California at Santa Barbara
Mr. Lick's comments addressed the adequacy of EPA's proposed models and whether any
of the models are "fatally flawed." Mr. Lick began by mentioning that one should consider the
ultimate use of models when evaluating their adequacy. He then noted that the models for the
Hudson River project are intended to address general questions regarding concentrations of PCBs
in fish. Moreover, he commented, the models have to provide insight into the impacts of
performing different remedial actions: how would concentrations of PCBs in fish change if no
remedial actions are performed? if depositional areas are capped? if erosional areas are capped
with heavy material? if erosional areas are dredged? Mr. Lick then noted that rivers generally
have some depositional areas, some erosional areas, and some areas that are depositional during
low flows and erosional during high flows. To predict fate and transport in such rivers, Mr. Lick
thought, models need to identify and characterize depositional and erosional areas as functions of
space, time, and flow. After commenting that models that cannot characterize sediment
deposition to such an extent are inadequate, Mr. Lick claimed EPA's sediment transport models,

-------
Appendix F—Summaries of Observers' Comments
as described in the PMCR reports, are not adequate and cannot "help in deciding appropriate
remedial action."
Mr. Lick noted that many alternative models are available that are adequate by his
standards. He noted further that QEA and some of his students have developed models—some of
which were developed for other rivers—that adequately address sediment transport. He also
indicated that EPA funded the development of some of these models. Finally, Mr. Lick
mentioned that "it is no secret" that adequate models are available and that they can help users
decide how to take appropriate remedial actions. Mr. Lick concluded his comments by
recommending that EPA use these more adequate models in its reassessment of PCBs in the
Hudson River.
Day #2, Comments from Kirk Ziegler, Quantitative Environmental Analysis (QEA)
Mr. Ziegler's comments focused on sediment transport modeling. Mr. Ziegler began by
agreeing with a comment raised by a peer reviewer regarding resuspension and deposition of
non-cohesive sediments. In particular, Mr. Ziegler indicated that non-cohesive sediment transport
needs to be included in fate and transport modeling, particularly for accurately modeling
sediments in the Thompson Island Pool. He then offered several recommendations to EPA for
improving its sediment transport models. First, Mr. Ziegler thought a mechanistic model is
needed to "more realistically describe resuspension and deposition processes." Mr. Ziegler
acknowledged that EPA has incorporated mechanistic algorithms in its 100-year flood model, but
he thought the agency should also "include a more mechanistic approach" in the HUDTOX
model. As part of adopting this mechanistic approach, Mr. Ziegler recommended developing a
"fine-resolution, two-dimensional sediment transport model" that accounts for how shear stresses,
bed properties, and bed types (i.e., non-cohesive and cohesive sediments) vary from one location
in the river to another. Mr. Ziegler indicated that such a fine-grid model should be used to
calculate time-dependent sediment resuspension and deposition fluxes, which can then be
aggregated to the scale of a coarse-grid, one-dimensional fate and transport model, like
HUDTOX. In review, Mr. Ziegler stated that outputs from a fine-grid sediment transport model
can and should be used as inputs to a coarse grid fate and transport model. Mr. Ziegler then
emphasized that developing the modeling approach he just described is not "an insurmountable
technical issue," and he noted that QEA has already incorporated his recommendations in a
Hudson River sediment transport model for GE. Mr. Ziegler concluded by indicating that QEA's
two-dimensional sediment transport model, coupled with its one-dimensional fate and transport
model, has enabled him to conduct "relatively long-term simulations of PCB fate and transport in
the river" with increased confidence.

-------
Appendix F—Summaries of Observers' Comments
Day #2, Comments from David Glaser, Quantitative Environmental Analysis (QEA)
Mr. Glaser' s offered brief comments and recommendations regarding bioaccumulation
models. First, Mr. Glaser 'Very strongly" suggested that EPA use a time-variable, mechanistic
bioaccumulation model. Second, Mr. Glaser noted that QEA has already developed such a
model. Finally, knowing the amount of effort needed to develop this kind of model, Mr. Glaser
suggested that EPA not start to develop its own bioaccumulation model, which he suggested
would be "re-inventing the wheel to a certain degree." Rather, Mr. Glaser recommended that
EPA perform "a critical, in-depth review" of QEA's model with the ultimate intent of EPA using
the model to evaluate bioaccumulation in the Hudson River.
Day #2, Comments from John Connolly, Quantitative Environmental Analysis (QEA)
Mr. Connolly began by reiterating the main recommendation proposed earlier in the
observer comment period by his colleagues: that EPA develop a mechanistic approach to model
both sediment transport and bioaccumulation. If EPA incorporates these changes, Mr. Connolly
noted, the Agency's "modeling framework" will be sufficient to answer questions regarding
remedial actions. However, regardless of the validity of the algorithms programmed into the
modeling framework, Mr. Connolly warned, the model may not be accurate if it is not correctly
parameterized. Mr. Connolly suggested the correct parameterization ultimately depends on the
extent to which the models are calibrated and validated. Finding it unfortunate that the peer
reviewers did not have the opportunity to appreciate "the full scope of information" available for
model calibration, Mr. Connolly indicated that he would provide an overview of the available data
and that he would also provide his recommendations to EPA for validating its models.
Regarding the available data, Mr. Connolly referred the peer reviewers to a map of the
water quality sampling locations for the Upper Hudson River. (Mr. Connolly noted that this map
was included with the written comments that David Glaser of QEA distributed on the first day of
the meeting.) In reference to the map, Mr. Connolly specifically noted the locations of the
following monitoring stations: (1) the Bakers Falls station, upstream from GE's Hudson Falls
plant at river mile 197; (2) the Rogers Island station, upstream of the Thompson Island Pool at
river mile 194; (3) the Schuylerville station; (4) the Stillwater station; and (5) the Waterford
station. Mr. Connolly also indicated that "more recent data" are available from the Thompson
Island Dam. Overall, Mr. Connolly pointed out that data from these stations characterize water
quality from locations just upstream of GE to Waterford, which he called "the interface" between
the Upper Hudson River and the Lower Hudson River. The placement of these stations,
according to Mr. Connolly, provides information on "what goes from the upper river to the lower
river." Mr. Connolly commented that such information is relevant to questions the peer reviewers
had about how EPA should couple its two fate and transport models.

-------
Appendix F—Summaries of Observers' Comments
Given these data sources (including the sources he described during his comments on the
first day of the peer review meeting), Mr. Connolly offered several recommendations to EPA for
calibrating and validating its models. First, noting that historical sediment data characterize "long-
term changes in surface sediment PCBs," Mr. Connolly thought it was "critical" that EPA
compare modeling results to measured concentrations from 1977 to the present. Second, he
recommended that EPA compare its modeling results over time to the amounts of PCBs that flow
from the Upper Hudson River to the Lower Hudson River, as gauged by the monitoring station at
Waterford and as estimated by PCB flux calculations conducted by 'Various organizations."
Third, Mr. Connolly recommended that EPA perform "upstream to downstream" comparisons of
the modeling results to the monitoring data for the sampling locations he mentioned earlier in his
presentation. Mr. Connolly thought these upstream and downstream comparisons should be made
during high-flow and low-flow conditions and for both PCBs and total suspended solids. As part
of this data validation of spatial variations, Mr. Connolly also suggested, the models should be
able to predict the "monstrous" changes in solids loadings observed in the Upper Hudson River.
Mr. Connolly mentioned that he would not discuss how important it is that the models be able to
predict levels of PCBs and total suspended solids during floods, but said that he had addressed
that topic during his comments on the first day of the peer review and would not discuss it
further. Fourth, regarding concentrations of PCBs in fish, Mr. Connolly suggested that EPA's
models should be compared to two sets of available data: (1) long-term monitoring data for
"almost 20 years in two locations" and (2) short-term data from the time following the Allen Mill
event. Elaborating on the Allen Mill event, Mr. Connolly indicated that the "pulse" release of
PCBs during this event "increased the concentrations in the water column of the system by orders
of magnitude" in 1991, and to a lesser extent in 1992. Mr. Connolly suggested that the response
of the food web to this "pulse loading" of PCBs is "a real test" for EPA's models. He reiterated
this point by claiming that the ability of the models to reproduce the actual response of the food
web to the Allen Mill event "is another important challenge" for EPA's models. Concluding his
remarks, Mr. Connolly again recommended that EPA conduct the model data comparisons listed
above to evaluate how well the models can predict future conditions of the Hudson River.
Day #2, Comments from Robert Henshaw, Hudson River Environmental Society
Mr. Henshaw first indicated that he would rather ask a series of questions than make
technical comments. Mr. Henshaw began by synthesizing his understanding of the peer reviewers'
discussions: he thought the peer reviewers gave "grudging agreement" that EPA's models are
appropriate, but he also noted that the meeting chair wrote "virtually entirely negative comments"
and problems on the easel located in the front of the meeting room. Based on these observations,
Mr. Henshaw asked the reviewers the following: (1) "Do you think you were invited into the
process early enough so that you could be helpful to EPA?" (2) "Did you have enough
information available to you?" (3) "Was all of the information that you had adequately up to date
so that your judgments will be maximally valuable to EPA?" and (4) "Overall, how good do you
think these models will be?" Regarding his last question, Mr. Henshaw noted that the peer

-------
Appendix F—Summaries of Observers' Comments
reviewers had provided "generally negative" comments on the models, and he asked the peer
reviewers to comment on how valuable the models are to EPA or "how optimistically" the
reviewers can recommend that EPA use the models in its reassessment.
Day #2, Comments from Leigh Foster, Arbor Hill Environmental Justice Corporation
Mr. Foster provided several brief comments regarding the content and the format of the
peer review. Mr. Foster first commended the peer reviewers "for taking so much time and care"
in making recommendations to EPA Referring to the comments that the meeting chair wrote on
an easel and lata* displayed on the walls of the meeting room, Mr. Foster requested that ERG
make the comments publicly available. Mr. Foster noted that having copies of the comments
would help him communicate the findings of the peer review to the communities he represents.
Regarding the reviewers' findings, Mr. Foster requested that the peer reviewers provide some
"comments and recommendations" relevant to issues of concern to the Hudson River
communities, even though, he acknowledged, the purpose of the peer review was to critique
EPA's specific modeling approach Moreover, Mr. Foster requested that the peer reviewers
assign priorities to their recommendations with the ultimate understanding that the models are to
be used to support a human health risk assessment.
Regarding the format of the peer review, Mr. Foster again expressed concern about the
composition of the peer review committee. He recommended that EPA review and revise the
peer review guidance and procedures that the agency gives to ERG. Mr. Foster was thankful for
the enthusiasm, energy, and experience that the peer reviewers brought to this meeting; however,
he mentioned that these meetings need to have groups of reviewers that "mirror or reflect" the
dynamic nature of the site of concern, which, in this case, is the Hudson River community. Mr.
Foster noted that this recommendation applies both to the composition of future peer review
panels and to the composition of the much largo* "Scientific Committee." Mr. Foster reiterated
that the peer review committee clearly does not include representatives of the Hudson River
community or of the stakeholders for the Hudson River PCBs site. In summary, Mr. Foster asked
that EPA consider his suggestions regarding the peer review process, and he also asked the peer
reviewers to make recommendations that reflect his concerns.

-------