PHASE 2 REPORT - REVIEW COPY
FURTHER SITE CHARACTERIZATION AND ANALYSIS
VOLUME 2D - REVISED BASELINE MODELING REPORT
HUDSON RIVER PCBs REASSESSMENT RI/FS
JANUARY 2000
For
U.S. Environmental Protection Agency
Region 2
and
U.S. Army Corps of Engineers
Kansas City District
Volume 2D-Book 3 of 4
Bioaccumulation Models
TAMS Consultants, Inc.
Limno-Tech, Inc.
Menzie-Cura & Associates, Inc.
Tetra Tech, Inc.
-------
PHASE 2 REPORT - REVIEW COPY
FURTHER SITE CHARACTERIZATION AND ANALYSIS
VOLUME 2D - REVISED BASELINE MODELING REPORT
HUDSON RIVER PCBs REASSESSMENT RI/FS
JANUARY 2000
K.
^^ f
o
For
U.S. Environmental Protection Agency
Region 2
and
U.S. Army Corps of Engineers
Kansas City District
Volume 2D - Book 3 of 4
Bioaccumulation Models
TAMS Consultants, Inc.
Limno-Tech, Inc.
Menzie-Cura & Associates, Inc
Tetra Tech, Inc.
-------
Table of Contents
-------
PHASE 2 REPORT - REVIEW COPY
FURTHER SITE CHARACTERIZATION AND ANALYSIS
Volume 2D - REVISED BASELINE MODELING REPORT
HUDSON RIVER PCBs REASSESSEMENT RI/FS
BOOK 3 of 4
CONTENTS
Page
LIST OF TABLES iv
LIST OF FIGURES vi
1. INTRODUCTION 1
1.1 BACKGROUND 1
1.2 PURPOSE OF REPORT 1
1.3 REPORT FORMAT AND ORGANIZATION 2
2. GENERAL BACKGROUND ON PCB UPTAKE 3
2.1 PCB COMPOUNDS... 3
2.2 PCB ACCUMULATION ROUTES 3
2.2.1 Direct Uptake from Water 4
2.2.2 Uptake via Food 4
2.2.3 Uptake from Sediments 5
2.3 FOOD WEB MODELS FROM THE LITERATURE AND THEIR SENSITIVITY TO INPUT
PARAMETERS 6
3. MODELING APPROACH: FISH BODY BURDENS 9
3.1 MODELING GOALS AND OBJECTIVES 9
3.2 CONCEPTUAL BASIS FOR HUDSON RIVER BIOACCUMULATION MODELS 11
3.3 BIVARIATE BAF ANALYSIS FOR FISH BODY BURDENS 16
3.3.1 Rationale and Limitations for Bivariate BAF Analysis 16
3.3.2 Theory for Bivariate BAF Analysis of PCB Bioaccumulation 17
3.4 PROBABILISTIC BIOACCUMULATION FOOD CHAIN MODEL 18
3.4.1 Rationale and Limitations 18
3.4.2 Model Structure 19
3.4.3 Spatial Scale for Model Application 21
3.4.4 Temporal Scales for Estimating Exposure to Fish 22
3.4.5 Characterizing Model Compartments 22
3.5 FISHRAND MECHANISTIC MODELING FRAMEWORK 25
3.5.1 Rationale and Limitations 25
3.5.2 Model Structure 25
3.5.3 Spatial Scale for Model Application 28
3.5.4 Temporal Scales for Estimating Exposure to Fish 28
3.5.5 Application Framework 29
3.5.6 FISHRAND Model Validation 36
MCA/TetraTech
-------
PHASE 2 REPORT - REVIEW COPY
FURTHER SITE CHARACTERIZATION AND ANALYSIS
Volume 2D - REVISED BASELINE MODELING REPORT
HUDSON RIVER PCBs REASSESSEMENT RI/FS
BOOK 3 of 4
CONTENTS
Page
4. BIVARIATE BAF ANALYSIS OF FISH BODY BURDENS 37
4.1 DATA USED FOR DEVELOPMENT OF BIVARIATE BAF ANALYSES 37
4.1.1 Fish Data 37
4.1.2 Water Column Data 47
4.1.3 Sediment Data 51
4.1.4 Functional Grouping of Sample Locations for Analysis 53
4.2 RESULTS OF BIVARIATE BAF ANALYSIS 53
4.3 DISCUSSION OF BIVARIATE BAF RESULTS 55
4.3.1 Comparison to Published BAF Values 55
4.3.2 Fit of Bivariate Models to Observations 56
4.3.3 Relative Importance of Sediment and Water Pathways 57
4.4 SUMMARY 59
5. CALIBRATION OF PROBABILISTIC BIOACCUMULATION FOOD CHAIN
MODEL 61
5.1 OVERVIEW OF DATA USED TO DERIVE BAFs 61
5.1.1 Benthic Invertebrates 61
5.1.2 Water Column Invertebrates 61
5.1.3 Fish 62
5.1.4 Literature Values 62
5.2 BENTHIC INVERTEBRATE:SEDIMENT ACCUMULATION FACTORS (BSAF) 63
5.2.1 Sediment Concentrations 63
5.2.2 Approach 63
5.2.3 Calculations of BSAF Values for Benthic In vertebrates 64
5.3 WATER COLUMN INVERTEBRATE:WATER ACCUMULATION FACTORS (BAFs) 65
5.3.1 Approach 65
5.3.2 Calculation of BAFwater for Water Column Invertebrates 67
5.4 FORAGE FISH:DIET ACCUMULATION FACTORS (FFBAFs) 67
5.4.1 Approach 67
5.4.2 Forage Fish Body Burdens Used to Derive FFBAF Values 68
5.4.3 Calculation of FFBAF Values for Forage Fish 69
5.5 PISCIVOROUS FiSH:DiET ACCUMULATION FACTORS (PFBAF): LARGEMOUTH BASS 69
5.5.1 Largemouth Bass to Pumpkinseed BAF for ZTri+ PCBs 70
5.6 DEMERSAL FISH: BROWN BULLHEAD:SEDIMENT ACCUMULATION FACTORS 70
5.7 VALIDATION OF PROBABILISTIC MODEL USING FATE AND TRANSPORT MODEL OUTPUT AS
INPUT 70
5.8 DISCUSSION OF RESULTS 71
MCA/TetraTech
-------
PHASE 2 REPORT - REVIEW COPY
FURTHER SITE CHARACTERIZATION AND ANALYSIS
Volume 2D - REVISED BASELINE MODELING REPORT
HUDSON RIVER PCBs REASSESSEMENT RI/FS
BOOK 3 of 4
CONTENTS
6. FISHRAND: TIME-VARYING MECHANISTIC MODEL BASED ON A GOBAS
APPROACH 73
6.1 OVER VIEW OF CALIBRATION PROCEDURE 73
6.2 SENSITIVITY ANALYSIS TO DETERMINE PARAMETERS FOR UPDATING EM CALIBRATION 73
6.3 MODEL INPUT DATA: USER SPECIFIED PARAMETERS 74
6.3.1 Non Species-Specific Parameters 75
6.3.2 Species-Specific Data 77
6.4 CALIBRATION RESULTS 80
6.5 MODEL VALIDATION: CALIBRATION USING PARTIAL DATASET 82
6.6 RELATIVE CONTRIBUTION OF SEDIMENT AND WATER PATHWAYS 83
7. BIOACCUMULATION MODEL FORECASTS 85
7.1 SEDIMENT AND WATER CONCENTRATION INPUTS 85
7.2 PREDICTED PCB CONCENTRATIONS IN FISH UNDER ZERO UPSTREAM BOUNDARY
CONDITION 85
7.3 PREDICTED PCB CONCENTRATIONS IN FISH UNDER THE 10 NG/L UPSTREAM BOUNDARY
CONDITION 86
7.4 PREDICTED PCB CONCENTRATIONS EM FISH UNDER THE 30 NG/L UPSTREAM BOUNDARY
CONDITION 87
7.5 DISCUSSION OF RESULTS 88
8. DISCUSSION OF UNCERTAINTY 91
8.1 MODEL UNCERTAINTY 91
8.1.1 Model and Parameter Uncertainties in the Fate and Transport Models 91
8.1.2 Model Uncertainties in the Bioaccumulation Models 91
8.2 PARAMETER UNCERTAINTY 94
8.2.1 Sensitivity Analysis 94
9. SUMMARY AND CONCLUSIONS 97
9.1 SUMMARY OF FOOD WEB MODELS 98
9.2 PRINCIPAL REPORT FINDINGS 99
REFERENCES 105
iii MCA/TetraTech
-------
PHASE 2 REPORT - REVIEW COPY
FURTHER SITE CHARACTERIZATION AND ANALYSIS
Volume 2D - REVISED BASELINE MODELING REPORT
HUDSON RIVER PCBs REASSESSEMENT RI/FS
BOOK 3 of 4
LIST OF TABLES
TABLES TITLES
2-1 A Comparison of the BAF Range Predicted by Gobas and Thomann Models
4-1 Count of NYSDEC Hudson River Fish Samples for PCB Aroclor Quantitation Collected
between River Miles 142 and 193 by Laboratory and Year
4-2 Aroclor Standards and NYSDEC Rules for Calculating Total PCBs from Analyses
Reported by Hazleton and Hale Creek for Upper Hudson River Samples
4-3 Packed-Column Peaks Used by NYSDEC Contract Laboratory "Hazleton" and
Associated PCB Congeners for Upper Hudson Fish Sample Aroclor Quantitation
4-4 Weight Percents of Congeners in Packed-Column Peaks Used for "Hazleton" Aroclor
Quantitation Schemes, based on Capillary Column Analyses of Aroclor Standards
4-5 NYSDEC Upper Hudson Fish Concentrations as mg/kg-lipid Converted to Tri+PCBs for
Bivariate BAF Analysis
4-6 Assignment of Water Column Concentrations to Fish Sampling Locations in the Upper
Hudson River
4-7 Summer Average Water Column Concentrations of Tri+ PCBs (ng/1) Used for Bivariate
BAF Analysis
4-8 Annual Average Surface Sediment Tri+PCB Concentrations (|ig/g-OC) used in Bivariate
BAF Analysis
4-9 BAF Models of Mean Tri+ PCB Concentration in NYSDEC Hudson River Fish Samples
(mg/kg-Lipid) Regressed on Water Column Concentration Only
4-10 BAF Models of Mean Tri+ PCB Concentration in NYSDEC Hudson River Fish Samples
(mg/kg-Lipid) Regressed on Sediment Concentration Only
4-11 Bivariate BAF Models of Mean Tri+ PCB Concentration in NYSDEC Upper Hudson
Fish Samples (mg/kg-Lipid) Regressed on Water Column and Sediment Concentration
4-12 Percentage of Variance, Beta Coefficients, and Elasticities for Water and Sediment as
Explanatory Variables for Fish PCB Tri+ Body Burden (mg/kg-Lipid) in the Bivariate
BAF Models
5-1 Coefficient of Variation in Forage Fish Samples by River Mile from USEPA Dataset
5-2 Final Distributions Used in Empirical Probabilistic Model
5-3 Relative Percent Difference Between Predicted and Observed for Empirical Probabilistic
Model
6-1 Initial Empirical Distributions for FISHRAND
6-2 Empirical, Prior, and Posterior Distributions for RM 189 (Thompson Island Pool)
6-3 Empirical, Prior, and Posterior Distributions Defined in FISHRAND for RM 168
(Stillwater)
6-4 Summary of Relative Percent Difference Between Modeled and Observed for
FISHRAND
6-5 Posterior Distributions Defined in FISHRAND for RM 168 (Stillwater) Using Full
Dataset and pre-1990 Only Dataset in Partial Validation
iv MCAyTetraTech
-------
PHASE 2 REPORT - REVIEW COPY
FURTHER SITE CHARACTERIZATION AND ANALYSIS
Volume 2D - REVISED BASELINE MODELING REPORT
HUDSON RIVER PCBs REASSESSEMENT RI/FS
BOOK 3 of 4
LIST OF TABLES
TABLES TITLES
6-6 Difference in Wet Weight ppm Between Forecasts using Partial Dataset Calibration
Results as Compared to Concentrations Obtained Using Full Dataset Calibration Results
6-7 Relative Importance of Sediment vs. Water Pathways from FISHRAND Regression
7-1 Asymptotic Tri+ PCB Concentrations for Standard Fillet Approached by Fish Body
Burden Forecasts
7-2 Year by Which Selected Targets Levels are Achieved Under the 10 ng/L Upstream
Boundary Condition Using FISHRAND
8-1 Results of Sensitivity Analysis for Spearman Rank Correlation - Lipid Normalized
8-2 Results of Sensitivity Analysis for Partial Rank Correlation - Lipid Normalized
8-3 Results of Sensitivity Analysis for Spearman Rank Correlation - Wet Weight
8-4 RESULTS OF SENSITIVITY ANALYSIS FOR PARTIAL RANK CORRELATION - WET WEIGHT
MCA/TetraTech
-------
PHASE 2 REPORT - REVIEW COPY
FURTHER SITE CHARACTERIZATION AND ANALYSIS
Volume 2D - REVISED BASELINE MODELING REPORT
HUDSON RIVER PCBs REASSESSEMENT RI/FS
BOOK 3 of 4
LIST OF FIGURES
FIGURES TITLES
3-1 Conceptual Framework for Empirical Probabilistic Model
3-2 Conceptual Schematic of FISHRAND Model
3-3 Comparison of FISHRAND, FISHPATH, and Gobas Field Measurements for Lake
Ontario
3-4 Comparison of FISHRAND and FISHPATH for Gobas Dynamic Model
3-5 Flow Chart for Bayesian Monte Carlo Simulation Procedure in FISHRAND
3-6 Schematic for Bayesian Updating Procedure
4-1 Comparison of Hazleton PCB Quantitations and Sum of Tri+ Congeners
4-2 Summer Average Water Column Exposure Concentration, Tri+ PCBs
4-3 Scatterplot Matrices for Fish Lipid, Sediment, and Water Tri+ PCB Concentrations in the
Upper Hudson River, 1977-1997
4-4 Relation of Mean Tri+ Concentration in Pumpkinseed to Summer Average Water
Column Concentration
4-5 Observed versus Predicted Concentrations of Tri+ PCBs for Brown Bullhead from
Bivariate BAF Model
4-6 Observed versus Predicted Concentrations of Tri+ PCBs for Largemouth Bass from
Bivariate BAF Model
4-7 Observed versus Predicted Concentrations of Tri+ PCBs for Pumpkinseed from Bivariate
BAF Model
4-8 Comparison of Bivariate BAF Model Predictions and Observations of Mean Summer
Body Burden of Tri+ PCBs in Brown Bullhead
4-9 Comparison of Bivarite BAF Model Predictions and Observations of Mean Summer
Body Burden of Tri+ PCBs in Pumpkinseed
4-10 Comparison of Bivariate BAF Model Predictions and Observations of Mean Summer
Body Burden of Tri+ PCBs in Largemouth Bass
4-11 Comparison of Bivariate BAF Model Predictions and Observations of Mean Summer
Body Burden of Tri+ PCBs for Thompson Island Pool
5-1 TOC-Normalized PCB Concentration in the Hudson River Based on Phase 2 1993 Data
5-2 BSAF Results
5-3 Cumulative Distribution Function for BSAF
5-4 Water Column to Water Column Invertebrate BAF Results
5-5 Forage Fish Concentrations and FFBAF Results
5-6 Summary of Largemouth Bass to Pumpkinseed Ratios
5-7 Summary of Brown Bullhead to Sediment Accumulation Factors
5-8 Whole Water and TOC-Normalized Sediment Concentrations Predicted by HUDTOX
5-9 Comparison to Data for Empirical Probabilistic Model for Largemouth Bass
5-10 Comparison to Data for Empirical Probabilistic Model for Brown Bullhead
5-11 Comparison to Data for Empirical Probabilistic Model for Pumpkinseed
vi MCA/TetraTech
-------
PHASE 2 REPORT - REVIEW COPY
FURTHER SITE CHARACTERIZATION AND ANALYSIS
Volume 2D - REVISED BASELINE MODELING REPORT
HUDSON RIVER PCBs REASSESSEMENT RI/FS
BOOK 3 of 4
LIST OF FIGURES
FIGURES TITLES
6-1 Freely Dissolved Water and Dry Weight Sediment Concentrations Predicted by
HUDTOX for 1977 - 1997
6-2 Lipid Distributions Used in FISHRAND
6-3 Percent Lipid versus Weight for the Fish Species
6-4 Mean Percent Lipid by Year for the Fish Species
6-5 Fish Weight Distributions Used in FISHRAND
6-6 Comparison of FISHRAND Model Results Before and After Calibration Procedure for
Largemouth Bass
6-7 Comparison of FISHRAND Model Results Before and After Calibration Procedure for
Brown Bullhead
6-8 Comparison of FISHRAND Model Results Before and After Calibration Procedure for
Yellow Perch and White Perch
6-9 Comparison of FISHRAND Model Results Before and After Calibration Procedure for
Pumpkinseed
6-10 Predicted vs. Observed Quantiles for River Mile 189
6-11 Predicted vs. Observed Quantiles for River Mile 168
6-12 Predicted vs. Observed Quantiles for River Mile 155
7-1 Freely Dissolved Water and Dry Weight Sediment Concentrations Predicted by
HUDTOX for 1998 - 2067 under Zero Upstream Boundary Condition
7-2 Freely Dissolved Water and Dry Weight Sediment Concentrations Predicted by
HUDTOX for 1998 - 2067 under 10 ng/L Upstream Boundary Condition
7-3 Freely Dissolved Water and Dry Weight Sediment Concentrations Predicted by
HUDTOX for 1998 - 2067 under 30 ng/L Upstream Boundary Condition
7-4 FISHRAND Median (50th Percentile) Predictions for 1998 - 2067 for Largemouth Bass
7-5 FISHRAND Median (50th Percentile ) Predictions for 1998 - 2067 for Brown Bullhead
7-6 FISHRAND Median (50th Percentile) Predictions for 1998 - 2067 for White and Yellow
Perch
7-7 FISHRAND Predictions for 25-50-95 Percentile Under Zero Upstream Boundary
Condition for 1998 - 2067 for Largemouth Bass in ppm Wet Weight
7-8 FISHRAND Predictions for 25-50-95 Percentile Under Zero Upstream Boundary
Condition for 1998 - 2067 for Brown Bullhead in ppm Wet Weight
7-9 FISHRAND Predictions for 25-50-95 Percentiles Under Zero Upstream Boundary
Condition for 1998 - 2067 for Yellow and White Perch in ppm Wet Weight
7-10 FISHRAND Predictions for 25-50-95 Percentiles Under 10 ng/L Upstream Boundary
Condition for 1998 - 2067 for Largemouth Bass in ppm Wet Weight
7-11 FISHRAND Predictions for 25-50-95 Percentiles Under 10 ng/L Upstream Boundary
Condition for 1998 - 2067 for Brown Bullhead in ppm Wet Weight
vii MCA/TetraTech
-------
PHASE 2 REPORT - REVIEW COPY
FURTHER SITE CHARACTERIZATION AND ANALYSIS
Volume 2D - REVISED BASELINE MODELING REPORT
HUDSON RIVER PCBs REASSESSEMENT RI/FS
BOOK 3 of 4
LIST OF FIGURES
FIGURES TITLES
7-12 FISHRAND Predictions for 25-50-95 Percentiles Under 10 ng/L Upstream Boundary
Condition for 1998 - 2067 for Yellow and White Perch in ppm Wet Weight
7-13 FISHRAND Predictions for 25-50-95 Percentiles Under 30 ng/L Upstream Boundary
Condition for 1998 - 2067 for Largemouth Bass in ppm Wet Weight
7-14 FISHRAND Predictions for 25-50-95 Percentiles Under 30 ng/L Upstream Boundary
Condition for 1998 - 2067 for Brown Bullhead in ppm Wet Weight
7-15 FISHRAND Predictions for 25-50-95 Percentiles Under 30 ng/L Upstream Boundary
Condition for 1998 - 2067 for Yellow Perch and White Perch in ppm Wet Weight
8-1 Comparison of Hazleton and Interlaboratory Mean Determinations of Percent Lipid from
1989, 1992, and 1995 Interlaboratory Comparisons
viii MCA/TetraTech
-------
Chapter 1
-------
1. INTRODUCTION
1.1 Background
The Hudson River watershed encompasses an area of 13,390 square miles, principally in
the eastern portion of New York State (Book 2, Figure 1-1). The Hudson River PCB Superfund
Site extends from Hudson Falls, New York, to the Battery in New York Harbor (River Mile 0), a
stretch of almost 200 river miles. The Upper Hudson refers to the approximately 40-mile stretch
of river upstream of Federal Dam at Troy to Hudson Falls (Book 2, Figure 1-2). The Lower
Hudson refers to the portion of the river downstream of Federal Dam to the Battery.
For approximately 30 years, two General Electric (GE) facilities, one in Fort Edward and
the other in Hudson Falls, used polychlorinated biphenyls (PCBs) to make electrical capacitors.
GE discontinued use of PCBs in 1977 when PCBs ceased to be manufactured and sold in the
United States. From 1957 through 1975, between 209,000 and 1.3 million pounds of PCBs were
discharged from these facilities into the Upper Hudson River. Migration of PCBs downstream
was greatly enhanced in 1973 with the removal of Fort Edward Dam and the subsequent release
downstream of PCB-contaminated sediments. A region of special concern is the highly-
contaminated sediments in Thompson Island Pool (TIP) immediately downstream of the old Fort
Edward dam site.
In 1984 the U.S. Environmental Protection Agency (USEPA) completed a Feasibility
Study on the site that investigated remedial alternatives and issued a Record of Decision (ROD)
later that year. The ROD called for: (1) an interim No Action decision concerning river
sediments; (2) in-place capping, containment and monitoring of remnant deposit (formerly
impounded) sediments; and, (3) a treatability study to evaluate the effectiveness of the Waterford
Treatment Plant in removing PCBs from Hudson River water.
1.2 Purpose of Report
In December 1990, USEPA issued a Scope of Work for reassessing the No Action
decision for the Hudson River PCB site. The scope of work identified three phases:
Phase 1 - Interim Characterization and Evaluation
Phase 2 - Further Site Characterization and Analysis
Phase 3 - Feasibility Study.
The Phase 1 Report (USEPA, 1991b) is Volume 1 of the Reassessment documentation
and was issued by USEPA in August 1991. It contains a compendium of background material,
discussion of findings and preliminary assessment of risks.
The Final Phase 2 Work Plan and Sampling Plan (USEPA, 1992) detailed the following
main data collection tasks to be completed during Phase 2:
MCA/TetraTech
-------
High- and low-resolution sediment coring;
Geophysical surveying and confirmatory sampling;
Water column sampling (including transects and flow-averaged composites); and,
Ecological field program.
The Database Report (Volume 2 A in the Phase 2 series of reports; USEPA, 1998b) and
accompanying CD-ROM database re-issued in August 1998 provides the validated data for the
Phase 2 investigation. The Data Evaluation and Interpretation Report (USEPA, 1997) presents
results and findings of water column sampling, high-resolution sediment coring, geophysical
surveying and confirmatory sampling, geostatistical analysis of 1984 sediment data and PCB fate
and transport dynamics.
This Revised Baseline Modeling Report is Volume 2D in the Phase 2 series of reports. It
includes descriptions of the transport and fate mass balance models, and the fish body burden
models that are being used for this PCB Reassessment RI/FS. This report builds upon and
supercedes the Baseline Modeling Report, which was released for public comment in May 1999.
The revisions in this report incorporate changes based on comments received during the public
comment period on the Baseline Modeling Report and on additional analyses.
1.3 Report Format and Organization
Chapter 2 of this report contains background information on the theory of PCB uptake
into fish. Chapter 3 contains a description of the specific approaches taken for each of the fish
body burden models as well as mathematical descriptions of the individual models. Chapter 4
contains the results from the bivariate BAF analyses. Chapter 5 contains calibration and
validation results for the probabilistic empirical model using the hindcasting sediment and water
results from the fate and transport models. Chapter 6 contains calibration and validation results
for the FISHRAND model (a mechanistic time-varying model incorporating probability
distributions and based on a Gobas approach) using the hindcasting sediment and water results
from the fate and transport models. Chapter 7 provides predictive results for 1998 - 2067 based
on inputs from the fate and transport models for the constant upstream boundary condition, and
the zero upstream boundary condition. Chapter 8 contains a discussion of the uncertainties in the
modeling analysis as well as a sensitivity analysis. Chapter 9 presents the summary and
conclusions for Books 3 and 4 of the Baseline Modeling Report.
The material in this report has been divided into four separate books. Book 1 contains the
report text, a list of references, and a glossary of abbreviations and acronyms for the fate and
transport modeling. Book 2 contains all tables, figures, plates and appendices for the fate and
transport modeling discussed in Book 1. Book 3 contains the report text, a list of references, and
a glossary of abbreviations and acronyms for the food chain modeling. Book 4 contains all tables,
figures, plates, and appendices for the food chain modeling discussed in Book 3. Within Book 4,
Appendix A contains ecological profiles for fish species represented in the fish body burden
models and the derivation of feeding preference distributions for the individual fish species.
MCA/TetraTech
-------
Chapter 2
-------
2. GENERAL BACKGROUND ON PCB UPTAKE
2.1 PCB Compounds
This report examines bioaccumulation of PCBs characterized as Aroclors for the
historical datasets and as selected congeners for the Phase 2 dataset. A challenge to developing a
modeling framework for PCB bioaccumulation is that PCBs consist of 209 individual congeners,
which exhibit varying degrees of bioaccumulation potential, depending on the number and
position of chlorine atoms on the molecule. The more highly-chlorinated congeners tend to
accumulate in fish tissues. This effect may be a function not of increased uptake, but rather
decreased elimination efficiency from the fish.
Studies that have measured PCBs as individual congeners have provided insights into the
bioaccumulation processes for watercolumn and sediment-based communities. Several
researchers have noted that even though total PCB levels may or may not increase with higher
position on the food chain, chlorine content of PCB body burdens tends to increase (Smith et al.,
1985; Oliver and Niimi, 1988; Van der Cost et al., 1988; MacDonald et al., 1993). Congener
patterns of caged fathead minnows and feral brown bullhead from the area around Thompson
Island Pool in the Hudson River were generally similar, sharing 60 percent of their 20 most
abundant peaks, but the bullhead had higher concentrations of hexa- and heptachlorobiphenyls
(Jones et al., 1989). The fish contained 17 peaks that were not detectable in water samples. It
has been noted that when young bluefish enter the Hudson River from offshore, heavier, more
chlorinated congeners were accumulated to a greater level than lighter, less chlorinated congeners
(LeBlanc and Brownawell, 1994).
A variety of factors control accumulation of PCB congeners (Shaw and Connell, 1984;
Jones et al., 1989; Kadlec and Bush, 1994; Ankley et al., 1992; LeBlanc and Brownawell, 1994):
1. Individual PCB congener characteristics, including solubility and partition coefficients,
degree of chlorination, and stereochemistry. Shaw and Connell (1984) found that more
planar molecules are more strongly absorbed that those with more typical shapes.
2. Characteristics of the fish, including lipid content of gills, blood, and tissue; cardiac
output; ventilation volume; gill surface area; epithelium layer of gill; aqueous stagnant
layer of gill; ability to biotransform PCBs; and, excretion rates.
3. Environmental factors, including temperature, pH, light, current, suspended particles, and
dissolved organic compounds.
2.2 PCB Accumulation Routes
Fish and other aquatic animals are exposed to PCBs through direct contact with water
(bioconcentration), and sediment, as well as through dietary sources (bioaccumulation). Due to
their hydrophobicity, PCBs tend to accumulate in the lipid portion of organisms. PCBs have also
been found to accumulate in predatory fish tissues at higher concentrations than the
concentrations in the surrounding water would predict (Thomann and Connolly, 1984), a process
MCA/TetraTech
-------
known as biomagnification. Depending upon the position of an aquatic organism within the
aquatic food web, exposure may be intensified through food sources as organisms consume other
organisms that have bioaccumulated PCBs in the lipid portion of their tissues. Because of the
important role of food as an exposure pathway, the feeding ecology of a fish species is a key
aspect in distinguishing between the relative contribution of the water column and sediments to
body burdens of PCBs.
2.2.1 Direct Uptake from Water
For fish, direct uptake of PCBs from water occurs primarily across the gills. No
significant evidence exists for absorption through the epidermis (Shaw and Connell, 1984).
The significance of direct uptake from water of PCBs has been debated. Based upon
laboratory studies, Shaw and Connell (1984) argued that uptake via the gills is the major route
for accumulation of PCBs. Some field studies have indicated that water column uptake could
account for PCB concentrations observed in biota, if PCB concentrations were normalized for
lipid content of the organism (e.g., Clayton et al., 1977).
Other researchers have continued to examine the potential for bioconcentration through
the gills to account for PCB concentrations. Caged rainbow trout that were fed clean,
commercial food appeared to accumulate PCBs directly from contaminated waters of the St.
Lawrence River (Kadlec, 1994; Kadlec and Bush, 1994). Barron (1990) noted that simple
evaluations of uptake directly from the water column have assumed that bioconcentration is
controlled by the hydrophobicity of the compound, as measured by its octanol-water partition
coefficient. He argued that bioconcentration appears to be independent of octanol-water partition
coefficients when the coefficient is small or when the molecule to be accumulated is large. He
summarized other factors that affect bioconcentration: molecular shape, degree to which the
compound is bound to dissolved organic matter, lipid content of the gills, size of the organism,
blood flow, variations in enzyme content and activity, and exposure temperature and ionic
content.
2.2.2 Uptake via Food
Field studies and modeling efforts have indicated that biomagnification through the food
chain is an important component for bioaccumulation. Sloan et al., (1984), for example,
suggested that the presence of higher chlorinated Aroclor mixtures in fish of the Lower Hudson
River might reflect a food chain component to bioaccumulation. Using existing field data,
Thomann (1981, 1989) derived steady-state food chain models, considering uptake of
contaminants from both water and food sources through several trophic levels. The models
indicated that food assimilation, excretion, and net weight gain were important characteristics
that determined bioaccumulation levels. They also demonstrated that for top predators, such as
Hudson River striped bass, almost all the observed PCB body burden could be attributed to a
food source. In Lake Michigan lake trout, only 2 to 3 percent of the PCB accumulation could be
predicted from water column concentrations using an age-dependent model (Thomann and
Connolly, 1984), while transfer through the food chain accounted for up to 99 percent of the
body burden of PCBs in Lake Michigan lake trout.
MCA/TetraTech
-------
Many researchers have tested, refined, or elaborated upon Thomann's food chain models.
One test of the approach examined PCB accumulation in young-of-the-year bluefish which enter
the Hudson River Estuary from relatively uncontaminated offshore waters and grow quickly
(LeBlanc and Brownawell, 1994). Connolly et al., (1985) considered growth rates, respiration
rates, food assimilation efficiency, predator-prey relationships, PCB assimilation efficiency, and
bioconcentration factors for PCBs when they applied a model to existing data from the Hudson
River system. They predicted PCB levels in Hudson River striped bass, assuming various
reductions in concentrations of PCBs in the water column. They also began efforts to incorporate
lipid- and non-lipid components of the striped bass into the model. Pizza and O'Connor (1983)
conducted laboratory experiments to determine rates of PCB accumulation from the gut and
elimination from the body in young-of-the-year striped bass from the Hudson River. An EPA
model, Food and Gill Exchange of Toxic Substances, or FGETS, has been used to predict
average concentrations of contaminants in the food web over time (e.g., Woolfolk et al., 1994).
This model incorporates bioconcentration of contaminants from the water column and
biomagnification in the food chain.
Gobas et al., (1993, 1995, 1999) examined the roles of food digestion, food absorption,
and rates of gill elimination and metabolic transformation upon bioaccumulation. This model has
recently been updated to include exposure from both water and sediment sources, and a
pharmacokinetic module. The mechanistic model presented here (FISHRAND) is based on these
approaches (1993, 1995, 1999).
As part of this modeling effort, Menzie-Cura & Associates have evaluated a number of
fish gut contents from the NYSDEC sampling effort. Similarly, Exponent, Inc. on behalf of
General Electric conducted a study on fish gut contents and identified specific invertebrates
down to the lowest practical taxonomic level in the diets of fish. This information, together with
historical data from the Hudson River power plant studies, have been used to more precisely
define food web relationships in the Hudson. The results of this effort are discussed and
presented in greater detail in Appendix A.
2.2.3 Uptake from Sediments
Equilibrium partitioning has been suggested to be the major factor controlling
bioaccumulation in sediment-based benthic communities. Bierman (1990) used field data from
the Great Lakes to determine that for animals at the lower and middle parts of the food chain,
including oligochaetes, chironomids, amphipods, sculpin, small smelt, and large smelt, predicted
bioconcentration factors based upon equilibrium partitioning coefficients accounted for
concentrations of hydrophobic organic compounds. Comparing laboratory and field data, Ankley
et al., (1992) confirmed that for oligochaetes, concentrations of PCBs in the sediments could be
used to predict concentrations of PCBs in organisms, but that for other species, food or possibly
ingestion of contaminated particles could affect concentrations. Ingestion of contaminated food
also seemed to be a factor in accumulation of PCBs in a freshwater lake (Van der Oost et al.,
1988).
A steady-state food chain model with a benthic invertebrate component was developed to
account for both water column and sediment sources of contaminants (Thomann et al., 1992).
MCA/TetraTech
-------
This model considered four exposure routes for ingestion of paniculate contaminants: sediment
organic carbon, overlying plankton, interstitial water, and overlying water. Applying the model
to an amphipod-sculpin food web in Lake Ontario (Oliver and Niimi, 1988), Thomann and his
co-workers (1992) found that accumulation was based primarily upon a benthic food web rather
than upon direct uptake from the water column. They noted however, that including the
overlying water and phytoplankton as a food source were necessary to explain the field data.
Considering only interstitial water and sediment particles as contaminant sources was not
satisfactory.
2.3 Food Web Models from the Literature and their Sensitivity to Input Parameters
All bioaccumulation models use a set of parameters to predict the body burdens of
organic contaminants in higher organisms. The uncertainty associated with these parameters
contributes to the uncertainty of the risk estimate. Burkhard (1998) compared the sensitivity of
the Gobas (1993) and Thomann (1989) model outputs to changes in their input parameters.
Sensitivity of the models to changes in input parameters was determined by running each model
once with nominal input values, and then changing one input value by 10%, and running the
model with the altered input value. A sensitivity index of 1.0 means that a 10% change in the
input parameter resulted in a 10% change in the model output. In this case, the model output
examined was the bioaccumulation factor, which is equal to the ratio of the lipid-normalized
concentration of chemical in fish to the concentration of freely dissolved chemical in water.
For both models, the input parameters with the largest influences were:
lipid contents of the organisms;
KOW of the chemical;
ratio of the concentration of chemical in sediment organic carbon to the concentration
in overlying water (FIsocw); and,
feeding preferences of the organisms (only for chemicals with log KOW exceeding 6).
The sensitivity index ranged up to about -20 (indicating a decrease in BCF) for the
feeding preference of a benthic invertebrate on phytoplankton in the Thomann model. The
models were less sensitive to changes in organism weight, temperature (input to Gobas model
only) and sediment organic carbon (input to Gobas model only).
The approach described above is limited because it does not take into account uncertainty
in input modeling parameters. For example, an input parameter with low sensitivity (i.e.
sensitivity index is close to 1) adds considerable uncertainty to estimates of model outputs if the
measurement uncertainty distribution of this input parameter is relatively large. Uncertainty
associated with the input parameters may result from analytical errors in the measurement of the
parameter, sampling that is not representative of the population, or lack of sufficient information
about the parameter. Moreover, many input parameters are variable in nature (fish body weight,
lipid content, etc.)
MCA/TetraTech
-------
The dual influences of variability and uncertainty in the input parameters on model
outputs must be considered when evaluating the overall model uncertainty. Monte Carlo
simulations should be performed for each input parameter, using a plausible range of values or
distribution for each input parameter. Burkhard (1998) compared the ratios of the 90th and the
10th percentiles of the model output derived from the simulations among input parameters. For
both models, Ilsocw, KOW, and feeding preferences resulted in the largest range of simulated
output values. Table 2-1 summarizes results from Burkhard (1998).
Note, however, that the findings of Burkhard (1998) are based on the analysis of a Great
Lakes food web in which benthic organisms are an important food source for higher trophic level
organisms. In food webs where the benthic component is less important, the importance of the
sediment-related input parameters on the uncertainties associated with predicted model outputs
may be different.
The model used by lannuzzi et al. (1996) is based on a Monte Carlo version of the
equations developed by Thomann et al. (1992), and Gobas (1993). They developed probabilistic
distributions for several parameters that are typically used in mechanistic bioaccumulation
models to predict the uptake of organic contaminants in aquatic food webs. The ranges, central
tendencies, and distributions of key parameters of the models were derived from a critical
evaluation of the literature on the physiology and ecology of three common estuarine organisms
rather than from site-specific experimental data. Distributions of the physical/chemical
characteristics (i.e. the octanol-water partition coefficient, KOW) for several congeners of PCBs
were also compiled from the literature.
The model used by lannuzzi et al. (1996) was used to estimate the concentrations of five
coplanar PCB congeners in adult mummichog fish, blue crab, and striped bass, using
distributions of available data on PCB and total organic carbon (TOC) concentrations that were
measured in surface sediments from the Passaic River in northern New Jersey. A model
sensitivity analysis was performed to rank input parameters according to their contribution to
model predictions.
Results of the sensitivity analysis suggest that the input parameters that most influence
the model (not listed in order of importance) are:
BSAF (biota-sediment accumulation factor) for infaunal organisms;
lipid content;
chemical concentrations in sediment;
total organic carbon (TOC) content of sediments;
the chemical assimilation efficiency (CAE);
residence time in the river for striped bass; and,
log KOW.
MCA/TetraTech
-------
In summary, both Burkhard (1998) and lanuzzi et al. (1996) concluded that the lipid
content of the exposed organisms and the KOW of the contaminant influence estimates of tissue
concentrations more than other parameters. The ability of organisms to metabolize specific PCB
congeners is also an important factor in the quantitative evaluation of uncertainty.
MCAyTetraTech
-------
Chapter 3
-------
3. MODELING APPROACH: FISH BODY BURDENS
3.1 Modeling Goals and Objectives
The goal of the bioaccumulation component of the modeling effort is to develop a
framework for relating body burdens of PCBs in fish to exposure concentrations in Hudson River
water and sediments. This framework is used to understand historical and current relationships as
well as to predict fish body burdens for future conditions. Estimates of PCB body burdens in fish
are intended to be used in the human health and ecological risk assessments and aid in decision
making regarding options for addressing PCB-contaminated sediments in the Upper Hudson
River.
The objectives of the body burden modeling effort are based on discussions with the
investigators responsible for the human health and ecological risk assessments and with the fate
and transport modeling team. Because PCB analytical protocols have varied over time, the
framework needs to account for historical as well as current data to the extent possible.
Accordingly, the framework is structured to meet the following objectives:
relate historical body burden data (originally reported as PCB Aroclors, Aroclor
totals, and, individual congeners for a limited subset of the historical data) to
exposure concentrations in water and sediments;
relate current and future body burdens (as PCB Aroclors, totals, and individual
congeners) to exposure concentrations in water and sediments;
provide estimates in a form that can be used for human health risk assessments;
provide estimates in a form that can be used for ecological risk assessments; and,
provide a set of modeling tools that can be coupled with the output from the PCB fate
and transport models to evaluate future management goals and the impact of No
Action and/or potential remedial alternatives.
To achieve these objectives, three modeling approaches have been developed to relate
PCB exposure concentrations in water and sediment to body burdens. Each of these approaches
organizes the data in different ways to provide complementary views of PCB uptake. These
approaches are introduced next.
Bivariate BAF Analysis: This analysis uses available time series data to develop statistical
relationships between concentrations in water and sediments and those in fish based on
observations from the historical New York State Department of Environmental Conservation
(NYSDEC) yearly monitoring. This analysis represents an empirical perspective of the statistical
relationship between fish body burdens and sediment and water exposures in a tiered approach to
food chain modeling.
MCA/TetraTech
-------
Empirical Probabilistic Food Chain Model: This model relies on knowledge of feeding
relationships to link body burdens to water and/or sediments through a series of empirical
transfer coefficients using a combination of the historical NYSDEC data, New York State
Department of Health (NYSDOH) data, and the US EPA Phase 2 data. This model provides
ground-truth information on observed relationships between food-web compartments.
FISHRAND: Gobas Time-Varying Mechanistic Model: This mechanistic, time-varying model is
based on the modeling approach presented in Gobas (1993 and 1995). The model relies on
solutions of differential equations to describe the uptake of PCBs over time, and incorporates
both sediment and water sources to predict the uptake of PCBs based on prey consumption and
food web dynamics.
These approaches complement one another and represent a logical progression in the
evaluation of PCB uptake. Both the bivariate analysis and the empirical probabilistic model
utilize derived Bioaccumulation Factors (BAFs) and rely on organizing observed data into
meaningful relationships, while FISHRAND is mechanistic and based on mass-balance of PCBs
rather than direct observations. The agreement between these and the resultant estimates of body
burdens provide a check on the three approaches. The bivariate analysis indicates the relative
importance of water and sediment pathways from a statistical, data-based point of view
irrespective of the underlying biology. The probabilistic bioaccumulation model represents a
slight refinement and limited mechanistic consideration by explicit incorporation of feeding
preference data and uncertainty and variability information. FISHRAND predicts probability
distributions of expected concentrations in fish based on mechanistic mass-balance principles, an
understanding of PCB uptake and elimination, and information on the feeding preferences of the
fish species of interest.
Selection of fish species for modeling body burdens was based on several criteria
including: 1) importance for fishing, 2) abundance, 3) importance in diet of other fish, 4) whether
the selected species is representative of particular habitats or trophic levels, and 5) whether the
selected species is representative of other fish species. Upon discussion with NYSDEC, USEPA,
and NOAA the following species were selected for bioaccumulation modeling:
10 MCA/TetraTech
-------
Fish Species
Spottail Shiner
Pumpkinseed
Brown
Bullhead
Yellow Perch
Largemouth
Bass
White Perch
Characteristics
Forage Fish, feeds on invertebrates in water column and sediments
Forage Fish, feeds on invertebrates in water column (on aquatic plants) and
to a limited degree sediments; popular recreational fish but seldom eaten
Lives in contact with sediment and feeds on a variety of animal life on or in
the sediments; can be fished recreationally and is eaten occasionally
Inhabits water column and feeds on invertebrates and small fish; popular
recreational fish and is commonly eaten
Larger individuals feed primarily on fish but will also eat other vertebrates
and invertebrates; popular recreational fish and is commonly eaten
Feeds on invertebrates and small fish; lives in the tidal portion of the
Hudson; undergoes migrations within the river
Ecological profiles for the selected fish species are provided in Appendix A and are used
to discern behavioral and trophic characteristics that could affect accumulation of PCBs.
The Bivariate BAF Analysis uses pumpkinseed, brown bullhead, largemouth bass, white
perch, and yellow perch. Sufficient historical data were not available for spottail shiner;
however, goldfish were added to the statistical analysis.
In addition to the fish species listed above, the striped bass is included in the evaluation.
However, no new models have been developed for this species. A major confounding factor is
that the striped bass are a migratory species that are resident in the river for only a portion of the
year. As such, it is inappropriate to assume that all PCB exposure occurs within the Hudson
River, and.under the current modeling framework, this is a key assumption. The modeling
program relies upon the work of Thomann to derive estimates for striped bass. It would be
desirable to have a model for the shortnose sturgeon, an endangered fish species in the tidal
portion of the Hudson. However, data are insufficient to develop a model for this species. It is
anticipated that a species-to-species extrapolation will be employed to evaluate the shortnose
sturgeon, based on physiological, feeding and habitat selection characteristics.
3.2 Conceptual Basis for Hudson River Bioaccumulation Models
The food chain models developed here share a common conceptual basis including:
1. PCB body burdens in fish are related ultimately to exposure concentrations in water
and/or sediments;
2. PCBs in the water column and sediments are not necessarily in equilibrium with each
other;
3. Within the water and sediment compartments, an equilibrium or quasi -steady-state
condition exists at temporal scales on the order of a year and spatial scales on the order of
a river segment for the bivariate BAF analysis and the probabilistic empirical model;
11 MCA/TetraTech
-------
4. Fish body burdens are in gwasi'-steady-state with the water and/or sediment at time scales
on the order of one or more years under both the bivariate BAF analysis and the
probabilistic empirical model.
PCB concentrations measured in biota are assumed to be in steady state with PCBs in the
environment for the development of bioaccumulation factors (BAFs), and thus can be related by
linear coefficients or bioaccumulation factors similar to partitioning coefficients. A steady-state
condition is usually considered to hold within a given year; thus the BAF approach represents
temporal changes only annually. The simplest approach considers that biota and all
environmental compartments are in equilibrium with one another, in which case the
concentration in any medium can be predicted from the concentration in any other medium. The
BAF method is readily modified to address situations in which a disequilibrium exists at steady
state between different environmental compartments.
Consider first a completely equilibrated system: Fish may accumulate PCBs through
partitioning from the water column, through ingestion of sediment, or through the food chain,
while organisms at lower trophic levels may also accumulate PCBs from both water column and
sediments. Describing exact accumulation pathways is the task of food web models, but
concentrations in any medium or "compartment" in a fully equilibrated system can be predicted
from those in any other compartment. As PCBs partition strongly to organic matter and have low
solubility, the major environmental reservoir is typically the sediment. "Partitioning" from
sediment to biota is conceptually similar to equilibrium partitioning from sediment and pore
water as well as from sediment to the water column. Thus, for an equilibrated system, dissolved
concentrations in sediment pore water might provide a good index of the bioavailable
component. Typically, analytically resolving truly dissolved and DOC-complexed fractions is a
very difficult task for pore water samples, but, for lipophilic compounds in sediments with
typical organic carbon contents, partition coefficients are such that the mass present in dissolved
and DOC-complexed forms is relatively insignificant compared to the total particulate-sorbed
mass. This implies that the dissolved portion can be quite well predicted from the sediment-
water partition coefficient, regardless of DOC levels. On the other hand, pore water
concentrations vary significantly in response to sediment organic carbon fraction (foe).
Therefore, sediment concentration normalized to foe is the best readily available predictor of
dissolved concentrations in an equilibrated system (Di Toro et al., 1991). This approach is being
used by EPA's Office of Water for establishing sediment quality criteria (USEPA, 199 la).
Of course, PCBs may enter the food chain both through the dissolved phase and ingestion
of particulate matter. As Di Toro et al., state, "biological effects (to invertebrates) appear to
correlate to the interstitial water concentration. This has been interpreted to mean that exposure
is primarily via pore water. However, the data correlate equally well with the organic carbon-
normalized sediment concentration. This suggests that the sediment organic carbon is the route of
exposure. In fact, neither of these conclusions necessarily follow from these data."
The reason for this surprising conclusion is contained in fugacity, or chemical potential
theory, which holds that the biological activity of a contaminant is controlled by its chemical
potential (Mackay, 1979). As discussed by Di Toro et al., if pore water and organic carbon
phases of the contaminant are in equilibrium then the chemical potentials exhibited by the two
12 MCA/TetraTech
-------
phases are equal. "Hence, so long as the sediment is in equilibrium with the pore water, the route
of exposure is immaterial. Equilibrium experiments cannot distinguish between different routes
of exposure." Thus, in the simplified equilibrium case, it is necessary to estimate the chemical
potential in only one phase. The question then becomes determining which phase is easiest to
measure. Where DOC complexing occurs, sediment concentration normalized to foe is clearly
the most directly measurable index of chemical potential.
Fish may accumulate PCBs via pathways which arise in the water column as well as from
the sediment. The simple equilibrium BAF approach works if sediment and water-column
concentrations are in equilibrium with one another, or if all PCB accumulation in fish derives
from pathways commencing in the local sediment. On the other hand, if fish accumulate PCBs
from both water-column and sediment pathways, and water-column concentrations are not in
equilibrium with pore water in the same locale, the full-equilibrium assumptions are not valid. In
the Hudson and other flowing rivers, it is likely that the upper sediment layer and the water
column are generally not in equilibrium with one another for hydrophobic toxicants. Further, the
upper, bioactive sediment zone is typically not in equilibrium with deeper, buried sediments.
However, the sediment-sorbed concentrations and pore-water concentrations within the bioactive
zone should be very close to equilibrium, while, in the water column, the dissolved and sorbed
fractions should also be close to equilibrium, except during transient events.
The equilibrium partitioning/fugacity arguments set forth by Di Toro et al., (1991) state
that the best readily measurable index of chemical potential should be the sediment sorbed
fraction normalized to/
-------
from River Mile 155 and River Mile 170. In summary, below Thompson Island Dam the water
column is not in equilibrium with local sediments. Thus, models for bioaccumulation need to
consider both water and sediment pathways, rather than relying on a BAF based on
concentrations in a single medium.
Very little information is available on how often contaminants in the environment reach
equilibrium among phases. If equilibrium conditions are not reached, time-variant models are
more appropriate for predicting contaminant concentrations. The distributions of contaminant
concentrations might differ from predicted concentrations if the system is not in equilibrium
because there is high temporal variability or because biological processes maintain
disequilibrium conditions. Many ecosystem and physical processes are variable over time. The
input of a contaminant into an estuary, for example, can occur during episodic events, such as
large storms or periodic disposal of dredged sediments.
The FISHRAND model is designed to evaluate the time-varying effects of PCS uptake on
predicted PCB fish tissue concentrations based on sediment and water exposure concentrations
predicted from fate and transport models as inputs. Both this model and the empirical
probabilistic model rely on information regarding feeding preferences of the fish species. To
more precisely define food web dynamics in the Hudson River, Menzie-Cura undertook the
following analysis.
The invertebrate component in the fish diet can consist of invertebrate species that are
themselves exposed to PCBs in surface water, pore water, and through their food. The food items
of invertebrate species may, in turn, be exposed to different levels and types of PCBs.
Understanding this component of the food web is not simple. Food habits of fish species are
described in Appendix A. Invertebrates eaten by Hudson River fish occupy a range of habitats
and eat a range of organic materials. The habitat and feeding preference for individual
invertebrate species influences the extent to which they are exposed directly and indirectly to
PCBs in sediments and in the water column. In our opinion these influences can only be
approximated based on available information and there are uncertainties associated with these
estimates. A qualitative conceptual framework for considering how invertebrates can be exposed
to PCBs in water and sediment is given below. It shows that invertebrate species probably
experience a gradient of exposure conditions ranging from predominantly sediment exposure to
predominantly surface water exposure. However, we believe that there are many species that will
fall between these extremes and which will experience both sediment and water exposures. We
have considered this when ascribing feeding preferences for fish that rely on invertebrates for
food. However, we acknowledge that there is little quantitative information for determining the
extent to which many of invertebrate species - primarily those that live on the surface of
sediments - are influenced by sediment and water exposures.
14 MCA/TetraTech
-------
Conceptual Framework for Considering The Influence
Of Sediment and Water as Exposure Media
For Invertebrates in the Diet of Fish
General
Physical
Habitat
Water column
or Phytophilous
At sediment
surface
Below sediment
surface
Source of Food
Phytoplankton
Periphyton
Zooplankton
Phytophilous
invertebrates
Meroplankton,
Epibenthic
invertebrates
living in-littoral
zone
Surface
Organic
Deposits
Phytophilous
invertebrates
Epibenthic '' -^
invertebrates L '
v,« - '..' V' V
1 ' ?> ,
>.-,:< ;'<; "
;;.:v ->:"- '
Infaunal . /
invertebrates'
Deeper Organic
Deposits
Tfbillilid^ > * i '^1
The simplified conceptual framework indicates how habitat location and food type could
influence the relative degree of influence of water and sediment on PCB exposure for
invertebrates. The increasing influence of sediments is illustrated qualitatively with an increasing
gray scale. Habitat affects the availability of different food types as well as the water exposures
experienced by invertebrates. For example, infaunal invertebrates are exposed primarily to pore
water while zooplankton are exposed primarily to surface water. Epibenthic invertebrates may be
exposed to some mix of pore water and surface water.
Examples of invertebrate species that may occupy the matrix of physical habitat and food
type are given below.
General
Physical
Habitat
Water column
or Phytophilous
At sediment
surface
Below sediment
surface
Source of Food
Phytoplankton
Periphyton
Bosmina
(Cladocera);
Copepods,
Gastropods
Gammarus spp.^
'(Amphipoda),' >
Ostracods .
Surface
Organic
Deposits
Dicrotendipes
spp.
(Chironomidae)
Gastropods; v IX
Ctae£idpt&a* \^
(isopoda) * , ,
^/»m«0^?|f.
KCnntJnomidae):-
w?n<^
Deeper Organic
Deposits
,',wOw> /'' -"*>'
VOTgWhaeteaX '
15
MCA/TetraTech
-------
As the conceptual framework suggests, PCB exposure for invertebrate species can be
complex, involving aspects of their feeding and physical ecology. Some species occur in a variety
of habitat types. Examples include the amphipod Gammarus and the chironomid insect larvae of
the genera Polypedilum and Dicrotendipes. Some invertebrates - planktonic rotifers, copepods,
and cladocerans - are carried with water masses and experience exposures associated with
"parcels" of water that are transported downstream. Other invertebrates live on the surface of
plants and experience water exposures that vary over time as water passes a particular location.
Still others are meroplanktonic (Chaoborus, Gammarus) and may be carried with the currents
diurnally, the remainder of the time spent in the sediments. Therefore, while we simplify the
characterization of food webs for modeling purposes, it should be evident that the system is
complex and that representations of relationships between water, sediment, invertebrates, and
fish should be viewed as uncertain estimates. This uncertainty is represented in the models
through the expression of feeding preferences as distributions in FISHRAND and through
distributions of transfer coefficients as derived in the empirical probabilistic model. However,
our ability to represent this uncertainty is limited by the available knowledge about the system
and the species within it. This uncertainty cannot be easily reduced.
3.3 Bivariate BAF Analysis for Fish Body Burdens
3.3.1 Rationale and Limitations for Bivariate BAF Analysis
The Bivariate BAF Analysis provides an empirical summary of historical data on fish
body burden in the Hudson River. The analysis relies on the available time series of
environmental and fish concentration data in the Upper Hudson to relate observed PCB
concentrations in fish to PCB levels in the water and sediment. If water and sediment
concentrations are not in equilibrium, a single BAF is not adequate; instead bioaccumulation is
controlled by the simultaneous effects of both water and sediment concentrations. Thus, a
statistical model with two independent variables (water and sediment concentrations) is
appropriate.
The development of statistical relationships is enhanced by the availability of extensive
historical monitoring data that enable comparison of PCB levels in fish and the environment over
time. The nature of these data, which consist primarily of Aroclor-equivalent quantitations in the
fish and total PCB estimates by packed-column gas chromatography in the water column,
however, constrains the statistical approach. Although more recent studies by TAMS/Gradient,
NOAA, and GE provide congener-specific PCB measurements in all media, these data are
limited in that they (1) are available only for the 1990s, (2) represent only a small number of
individual samples for a given fish species, and (3) do not provide a time-series perspective on
the relationship between fish body burdens and environmental concentrations.
Statistical relationships do not, of course, prove physical causality. Statistical models that
capture historic conditions are not guaranteed to accurately predict future conditions, particularly
if the characteristics of the PCB source change over time. For this reason, the Bivariate BAF
Analysis has not been used to predict future concentration trends. The Bivariate BAF Analysis,
however, is an important first step for the development of more complex, food web models, for
which the database is limited. By summarizing historical relationships between fish body
16 MC A/TetraTech
-------
burdens and environmental concentrations, the Bivariate BAF Analysis provides important
constraints on the form and parameterization of the food web bioaccumulation model.
3.3.2 Theory for Bivariate BAF Analysis of PCB Bioaccumulation
The general theoretical framework for deriving Bivariate Statistical Models was
introduced in Chapter 3.2. The fact that the water and sediment compartments are not in
equilibrium with each other, but are approximately internally equilibrated, suggests that bivariate
BAFs that relate body burden to both sediment and water-column chemical potential could
account for bioaccumulation pathways from both water and sediment. Correlating fish body
burdens to both water and sediment removes the difficulty of disequilibrium between the
sediment and water compartments.
The Bivariate BAF Analysis is essentially a 'black box1 approach wherein the details of
exposure pathways and physiological processes are not specified but the net effect is captured.
The actual PCB concentration found in a given fish depends on the cumulative effects of
dietary/food chain accumulation, plus direct accumulation from the water (and perhaps
sediment), all balanced by species-specific rates of depuration or metabolism. Net accumulation
in a fish species thus depends on all lower trophic levels. There are, however, only two main
external forcing functions, water and sediment PCB concentrations, which enable a 'black box'
model to be developed through statistical analyses with water and sediment concentrations as
input and fish burden as output.
For steady-state concentrations in the environment, the net result of the unspecified
processes contained within the 'black box' is functionally equivalent to a steady-state food web
model. For instance, the simplified steady-state food web model of Thomann et al., (1992) for
Lake Ontario, which avoids the need for a detailed study of population dynamics through steady-
state assumptions, is externally forced by water and sediment concentrations alone. It is thus
equivalent to a bivariate BAF relating fish body burden to water and sediment concentrations,
where the food web interactions determine the values of the two BAF factors. Therefore, a
bivariate regression relating average PCB body burden in a given species (by location and year)
to concentrations in local water and sediment provides a useful tool for assessing
bioaccumulation of PCBs by fish, and for providing a statistical perspective on the more
sophisticated, biologically-based food chain models.
As discussed in Chapter 3.2, fugacity theory indicates that chemical potential is best
estimated by the sorbed fraction in both sediments and water column, normalized to foe. This
suggests a regression analysis to predict fish PCB burdens from environmental concentrations
through species-specific relationships should take the following form:
fli
BW:
Cs..
(3-1)
17 MCA/TetraTech
-------
in which, for species i:
Cf = PCB concentration in fish (wet-weight basis)
fl = Lipid fraction in fish
Bw = Partial BAF relating fish concentration to water-column
concentration
Csw = PCB concentrations on suspended solids
focw = Organic carbon fraction of suspended solids
Bs = Partial BAF relating fish concentration to upper-zone sediment
concentration
Css = PCB concentration in upper zone sediments (dry-weight basis)
focs = Organic carbon fraction of the sediments.
While this formulation is theoretically optimal, focw is not available in the historic database for
the Hudson River; as a result, Bw must be expressed on a whole-water basis as a matter of
practical necessity.
3.4 Probabilistic Bioaccumulation Food Chain Model
3.4.1 Rationale and Limitations
The Probabilistic Food Chain Models are developed to predict distributions of PCB body
burdens within the selected fish species. These models complement the Bivariate BAF Analyses
that predict single population statistics such as the average values of PCBs. The conceptual
approach is presented in Figure 3-1. The Probabilistic Models have been developed to provide:
1. information on the fractions of the fish populations that are at or above particular PCB
levels; and
2. an empirical framework for constructing biologically-based food chain relationships that
explicitly incorporate variability and uncertainty inherent in the underlying data.
PCB body burdens in Hudson River fish vary among individuals within a species for any
given reach of the river. This intra-species variability in concentrations can be described as a
distribution. The characteristics or shapes of these distributions can be important for evaluating
human health and ecological risks. For example, two distributions may have the same average
value but may differ in spread, one having values distributed closely around the average, the
other including much higher as well as much lower values. The distribution with a greater
fraction of high values may pose a greater risk than the tighter distribution. Probabilistic models
that predict the characteristics of distributions provide risk assessors with the information needed
for making these evaluations. Probabilistic models also provide a tool for quantifying the
uncertainties associated with estimating body burdens of PCBs.
18 MCA/TetraTech
-------
The distribution of concentrations of PCBs within a species reflects a number of factors
that are also variable. These include the composition of PCBs, spatial and temporal exposure
field of PCBs in water and sediments, the uptake and depuration rates of PCBs within and among
trophic levels, and the feeding behavior, lipid content, and history of the fish. Many of these
factors are unknown or poorly known for the selected Hudson River species. The approach taken
in building the Probabilistic Food Chain Models is to combine information from available
measurements for the river with knowledge concerning the ecology of fish species and the
trophic relationships among fish and invertebrates.
The models presume quasi steady-state conditions for which mean seasonal exposure
concentrations in water and surface sediments change slowly relative to the species uptake and
depuration kinetics. The models are constructed by identifying the major pathways linking
individual fish species with sediment and water components. These pathways include direct
exposure as well as trophic relationships. Within the models, each major pathway is represented
by a distribution of transfer or bioaccumulation factors. Using information on species' ecology,
statistical distributions for PCB transfer or bioaccumulation factors are developed among media
and biological components. These factors are derived from measurements of PCB concentrations
in various compartments and do not require assumptions about kinetic processes, although it is
assumed that fish will be in a quasi steady-state with the environment. The transfer and
bioaccumulation factors reflect the sum of the underlying processes and are specific to Hudson
River fish and environmental conditions.
The models are designed to identify the relative contributions of PCBs in Hudson River
sediments and water to body burdens of the six selected fish species. Because exposure to PCBs
may occur via water column and sediments, it is important to distinguish between these two
media. Food is expected to be the primary route of exposure for fish but direct uptake from
water may also be important depending on the specific chemical. In developing the models, the
role of direct water uptake versus food was examined, and quantitatively evaluated using the
mechanistic FISHRAND model.
Because of the important role of food as an exposure pathway, what and where a fish eats
are viewed as key aspects of distinguishing between the relative contribution of the water column
and sediments to a species' body burden of PCBs. Some species feed predominantly on benthic
invertebrates, others on pelagic invertebrates, and still others on forage fish. Some species, such
as the largemouth bass, feed on all three components to varying degrees. As discussed earlier,
identification of the specific life histories of the invertebrates that fish tend to consume plays an
important role in identifying predominant exposure pathways.
3.4.2 Model Structure
The conceptual framework for the probabilistic PCB food chain models is illustrated in
Figure 3-1. A separate model is developed for each fish species reflecting the particular species
biology and available information on PCB BAFs. These models can be developed for individual
congeners, homologue groups, Aroclors, or total PCBs. In this report, the results for ZTri+ PCBs
(the sum of tri- through decachlorinated biphenyls) are discussed. ITri+ is a good representation
19 MCA/TetraTech
-------
of total PCBs in biota. The models are designed to evaluate quasi steady-state conditions on an
annual basis. The features of the models include:
1. Two groups of invertebrates are described: a) invertebrates that live within sediments and
feed primarily on sedimentary material (primarily deposit feeders), and b) invertebrates
that feed primarily on organic paniculate matter transported in the water column
(zooplankton, many epiphytic invertebrates, and some filter feeding invertebrates).
2. Invertebrates in group "a" are presumed to reflect localized sediment concentrations and
to be in steady state with the sediments as described by lipid and organic carbon
normalized BAFs.
3. Invertebrates in group "b" are presumed to reflect PCB concentrations associated with
whole water column concentrations. These invertebrates are presumed to be exposed to
PCBs associated with organic paniculate material in the form of detritus or algae as well
as through direction partitioning of the dissolved phase. In the Hudson, it is presumed
that both forms of organic material will be important in the diets of invertebrates. The
invertebrates that feed in this manner are presumed to be in steady state with temporally
averaged whole water column concentrations of PCBs as described by whole water BAFs.
4. In most cases, the models are designed to estimate body burdens in adult fish. These
larger fish are the ones important for human health risk assessment. In addition, because
the primary population-level risk of PCBs to fish is reproductive impairment, body
burdens in adults can be used in the ecological evaluation. Because young fish of some
species (e.g., pumpkinseed sunfish) are important as forage fish, body burdens are
estimated for these juveniles. Fish fall into one of several types depending on their
foraging strategies. The species-specific models incorporate such information and
recognize the variability that exists among and within species.
5. The lipid normalized BAF factors between invertebrates and fish, and fish and fish are
represented by distributions derived from Phase 1 and 2 studies carried out in the Hudson
and from the literature. Values have been derived for the calibration congeners, Aroclors,
and total PCBs (USEPA, 1996). Results presented here are for STri+. STri+ PCBs
represent total PCBs in biota samples.
6. The food chain models are designed to take as input the water and sediment
concentrations predicted by the fate and transport models described in Books 1 and 2. The
key input parameter for sediments is the PCB concentration normalized to sediment
organic carbon. The key input parameter for the water column is total concentration of
PCBs in the water (including both particulate and dissolved). Since feeding occurs
primarily in the warmer months, the probabilistic model has been developed using
summer averages. The fate and transport model results are averaged to provide summer
water concentrations and annual sediment concentrations.
20 MCA/TetraTech
-------
Based on the above, the following media and biological compartments are identified: 1)
water, 2) sediment, 3) water invertebrates, 4) sediment invertebrates, 5) forage fish, and 6) the
individual fish species.
The food chain models are currently implemented as a Monte Carlo spreadsheet model.
For the Monte Carlo spreadsheet model, the relationships among compartments and the
distributions for BAFs are incorporated into an Excel spreadsheet with a Crystal Ball
software add-in. Excel is a standard spreadsheet and provides the basic computational
framework. Crystal Ball software permits the input data to be represented as distributions
rather than single point values; the software also enables Monte Carlo analyses to be performed.
The species-specific Excel/Crystal Ball spreadsheet incorporates uncertainties in exposure
concentrations, food chain transfers, foraging behavior, and lipid content. Monte Carlo
operations yield cumulative distributions of body burdens on a lipid normalized and whole fish
basis for each species. Key variables in the probabilistic model are represented by a distribution
of values rather than a single point estimate (such as a mean or upper-bound value). Monte Carlo
simulation is a method of sampling from these distributions within a computational framework.
Generally, the greater the number of simulations, the lower the standard error associated with the
mean. In developing the probabilistic model, Monte Carlo simulations were run a minimum of
10,000 trials.
The distributions are representative of variability in the data as described in subsequent
sections. The distributions can also represent uncertainty, for example, by providing a range of
feeding proportions rather than single values. In this case, both variability and uncertainty are
represented in the distributions. For example, observed variability in the relationship between
sediment concentrations and benthic invertebrates is attributable to both true population
heterogeneity (variability) as well as measurement error (uncertainty). It is operationally difficult
to truly separate these two sources. Consequently, the model can be viewed as predicting
population profiles of PCB concentrations rather than the uncertainty associated with predictions
for any given percentile of variability.
3.4.3 Spatial Scale for Model Application
The probabilistic food chain model used the river segmentation developed for the fate and
transport models together with available fish data to assess PCB exposure from the water-column
and sediment.. For most fish species, these model segments are expected to encompass the
exposure zones for fish that may be caught in a particular segment of the river. The primary zone
of exposure for most fish species is presumed to be the summer foraging areas. Fish are
expected to obtain most of their PCB body burden via food. Profiles for the species (Appendix
A) indicate most of the feeding occurs during the warmer periods of the year. On a relative basis,
little feeding occurs in the winter. Therefore, the summer foraging areas are where most of the
fish species' exposure occurs. Because most of the selected fish species exhibit limited spatial
movements during the summer, foraging areas and exposure zones can be highly localized. A
notable exception is the white perch, a semi-anadromous species that migrates over larger
stretches of the river. White perch are found primarily below the Federal Dam in the Lower
Hudson River.
21 MCA/TetraTech
-------
The HUDTOX model provides daily estimates of sediment and water concentrations for
segments in the upper river (see Books 1 and 2). For water concentrations, there are both spatial
and temporal gradients in concentration that are appropriately averaged to provide estimates
representative of how fish integrate exposures. Fish exposures will vary around this mean value.
Calibration results for fish body burdens are presented for two river miles: 189 (Thompson Island
Pool), and 168 (Stillwater). These locations represent the bulk of fish concentration data for the
upper river.
The model covers three river reaches: 189 (TIP), 168 (Stillwater), and 154 (Waterford to
just above the Federal Dam). Each of these encompass a roughly 5 mile interval of exposure.
3.4.4 Temporal Scales for Estimating Exposure to Fish
Exposure concentrations for water are estimated as summer averages (May through
September). This averaging period is coincident with the time that fish are at their summer
foraging areas. Sediment concentrations show very little variation on an annual basis, thus
sediment concentrations are averaged annually.
3.4.5 Characterizing Model Compartments
3.4.5.1 Sediment to Benthic Invertebrate Compartment
This compartment of the model relates the concentrations of PCB in benthic invertebrates
to sediment concentrations of PCB. It assumes that the PCB levels in the invertebrates are
related directly to levels in the surrounding sediments. This relationship is represented by an
empirically-derived biota sediment accumulation factor (BSAF) that reflects the combination of
passive and/or active bioaccumulation mechanisms occurring in the sediments. PCB uptake into
benthic invertebrates appears to be the result of partitioning between the organic carbon of the
sediments and the lipid of the invertebrate species (Bierman, 1990). This relationship is a simple
ratio:
BSAF = -Q^ (3.2)
(^sediment
where,
BSAF = biota - sediment accumulation factor
Cbemhic = the concentration of PCB in an individual organism as u,g/g lipid
Csediment= mean PCB concentration in sediments as |ig/g organic carbon
22 MCATTetraTech
-------
3.4.5.2 Water Column: Water Column Invertebrate Compartment
Individual PCB congeners can be strongly associated with either the truly dissolved phase
in the water column or the paniculate phase. These differences average out to some extent when
evaluating a mixture of PCBs. The Data Evaluation and Interpretation Report (USEPA, 1998)
provides estimated partition coefficients for a number of key congeners. These data show the
fraction of PCB concentrations associated with the particulate phase increases with increasing
chlorination. For the lighter chlorinated congeners, bioaccumulation is driven primarily by direct
uptake from the dissolved phase in the water. For the higher chlorinated congeners, consumption
of particulate matter represents the route of greatest bioaccumulation.
Combining both the dissolved and particulate concentrations in a whole water
concentration, we considered the role of whole water using a BAF approach between water and
fish:
PWBAF= Cinver/Cwater (3-3)
where,
PWBAF = The bioaccumulation factor between water column
invertebrates and ZTri+ water PCB concentrations
Cinvert = mg PCB per Kg lipid in invertebrate tissue
Cwater = nig PCB per L water
3.4.5.3 Forage Fish Compartment
Several of the fish species selected for modeling consume other, smaller forage fish of
which there are numerous species in the Hudson. Rather than quantify PCB concentrations in
individual forage fish species, the model assumes that piscivorous fish will consume any species
less than 10 cm. This assumption is supported by forage fish abundance data for the Hudson
River from the literature as well as piscivorous fish gut analyses (MPI, 1984). A composite
forage fish compartment has been developed that reflects the composition of forage fish in the
Hudson and the feeding habits of these fish. The details of how the forage fish compartment was
derived are presented in Appendix A. The analysis indicated that Hudson River forage fish are
composed of species that feed to varying degrees on invertebrates in the water column and in the
sediments. When the relative abundance and feeding behavior of the species are taken into
account, the composite forage fish diet is comprised of approximately 67% water column
invertebrates and 33% sediment invertebrates. All piscivorous fish that feed on Hudson River
forage fish are assumed to be preying on species that - on average - feed on water column and
sediment invertebrates in these percentages.
23 MCA/TetraTech
-------
The forage fish bioaccumulation factor (FFBAF) is defined as:
FFBAF = S- (3-4)
where,
FFBAF = forage fish bioaccumulation factor
Cff = concentration in individual forage fish (^,g ETri+ per g lipid)
weighted average of diet concentration (jig ZTri+ per g lipid - species-
specific benthic and water column invertebrate fractions)
3.4.5.4 Piscivorous Fish Compartments
Adult piscivorous fish eat a combination of forage fish and invertebrates. Since forage
fish concentrations are derived primarily from water column invertebrate concentrations, it is
assumed that direct ingestion of water column invertebrates by piscivorous fish is encompassed
in this step. In the model, therefore, piscivorous fish PCB body burdens are quantitatively related
(in varying degrees, depending on the fish species) to the benthic invertebrate and forage fish
boxes.
The piscivorous fish under consideration in this model is the largemouth bass. The
piscivorous fish bioaccumulation factor (BAF) is defined as:
- (3-5)
*-diet
where,
BAF = piscivorous fish bioaccumulation factor relative to diet
Cfish = concentration in piscivorous fish (|lg ZTri+ per g lipid)
Cdiet = weighted average of diet concentration (u,g STri+ per g lipid).
The largemouth bass diet consists of 90 percent forage fish and 10 percent benthic
invertebrates.
3.4.5.5 Demersal Fish
The final category of fish to be considered are the demersal or bottom-feeding fish. The
best species to consider for this compartment is the brown bullhead, which feeds primarily from
sediment sources, although it is properly considered an omnivorous fish. Brown bullhead lipid-
normalized concentrations were compared to sediment TOC-normalized concentrations.
24 MCA/TetraTech
-------
The BSAF for brown bullhead is defined as:
BSAF = ^- (3-6)
C*
where,
BSAF = brown bullhead bioaccumulation factor
CBB - concentration in brown bullhead (jxg ZTri+ per g lipid)
CSed = concentration in the sediment (fig STri+ per g carbon).
3.5 FISHRAND Mechanistic Modeling Framework
3.5.1 Rationale and Limitations
FISHRAND incorporates time-varying information on water and sediment concentrations
to mechanistically describe the uptake of PCBs into fish tissue. The model is based on the peer-
reviewed time-varying Gobas model (Gobas, 1993; Gobas et al., 1995; 1999). FISHRAND is
designed to incorporate probability distributions and is programmed in Fortran-90 with a
Microsoft Excel graphical user interface.
Figure 3-2 shows the conceptual model for the Hudson River food web. The numbers
show in the Figure 3-2 represent the mean dietary percentage from particular compartments for
each species. Development of the distributions for each of the parameters described in this
chapter is presented in Chapter 6.
3.5.2 Model Structure
The model consists of a series of compartments as in the empirical probabilistic model.
Pelagic invertebrates are assumed to be in equilibrium with truly dissolved water column
concentrations, and benthic invertebrates are assumed to be in equilibrium with sediment
concentrations. Forage fish feed on these two compartments in accordance with their species-
specific foraging strategies. Piscivorous fish consume some amount from each compartment in
similar proportions as in the empirical probabilistic model, although in this model distributions
are used to reflect feeding preferences.
Biota can gain PCBs via uptake from the water column or through consumption of
contaminated prey (both sediment and water based), and lose PCBs via fecal excretion or
respiration.
The general form of the differential equation describing the change in concentration of
PCBs in biota with respect to time is given by:
- = k\ * C*d + kd * Cdi« -(ki + ke + kn + kg)* Cfish (3-7)
dt
25 MCA/TetraTech
-------
where:
kt = gill uptake rate (L/Kg/d)
CWd = truly dissolved XTri+ PCB concentration in water (ng/L)
kd = dietary uptake rate (d"1)
Cdiet = concentration in the diet (g/g)
k2 = gill elimination rate (d"1)
kg = fecal egestion rate (d"1)
km = metabolic rate (d"1) (assumed to be zero)
kg = growth rate (d"1) (takes the place of explicit age-class consideration)
Cfish = ETri+ PCB concentration in fish (jig ZTri+ per g)
3.5.2.1 Rate Constants
Direct Uptake from Water
The rate at which fish take up chemicals from water depends upon the gill ventilation rate
and the rate of diffusion of the chemical across the gills. The Gobas (1993) model uses
experimental data to derive uptake rates given by:
(3-8)
where:
k| = gill uptake rate (d"1)
KOW = octanol/water partition coefficient
Qw = transport rate in the aqueous phase (L/day)
Qi = transport rate in the lipid phase (L/day)
Vf= fish weight in kg (described by a distribution in FISHRAND)
The transport rates in the aqueous and lipid phases are given by:
(2»- = 88.3*V/06 (3-9)
26 MCA/TetraTech
-------
The gill elimination rate is then given by:
k2= kl (3-11)
Lf * Kow
Uptake from Consumption of Prey Items
The rate at which fish take up chemicals from food depends upon the food ingestion rate,
the rate of diffusion of the chemical across the intestinal wall, and the fecal egestion rate. The
Gobas model (1993) assumes that the efficiency with which chemicals are taken up from food is
related to the transport of chemical across aqueous and lipid phases of the gut:
Ed* Fd ._ ,_.
Kd = - (3-12)
Vf
where:
kd = dietary uptake rate constant (d'1)
Ed = uptake efficiency (unitless)
Fd = food ingestion rate (kg food/day)
Vf = fish weight (kg)
The uptake efficiency, Ed, is given by:
(3-13)
And the food ingestion rate, Fd, in [kg food/day], is given by:
Fd= 0.022 *v/°*5*eOMT (3-14)
where:
Fd = food ingestion rate (kg food/day)
Vf = fish weight (kg) (described by a distribution in FISHRAND)
T = monthly mean water temperature (deg C)
Fecal egestion rate constant
The fecal egestion rate is given by:
(3-15)
= fecal egestion rate (d"1)
= dietary uptake rate constant (d"1)
27 MC/VTetraTech
-------
Growth rate constant
The growth rate constant presented in the original Gobas model is given by the following
equations.
For temperatures greater than 10°C (T>10°C), the growth rate constant, kg, is given by:
fcg=0.01*V/-°2 (3-16)
For temperatures less than or equal to 10°C (T<10°C), the growth rate constant, kg, is given by:
kg= 0.002 *V/-°2 (3-17)
3.5.3 Spatial Scale for Model Application
The initial concentrations are the predicted sediment and water concentrations from the
fate and transport model. The average concentrations across individual sampling grids represent
the integrating effects of fish foraging and habitat strategies. In the Thompson Island Pool (river
mile 189), the nearshore segments just above the Thompson Island Dam and the corresponding
cohesive and noncohesive sediment segments were used to estimate sediment-based exposures.
Sediment concentrations represent a weighted average of cohesive and non-cohesive sediments
based on area and an assumption that fish, on average, spend 75% of their time over cohesive
sediments (the exception may be the white perch, which tend to range throughout the river,
including main channel areas).
The water-column PCB concentrations were adjusted based on flow and upstream
concentration at Fort Edward to better reflect nearshore exposure concentrations. Within the TIP,
strong lateral gradients in PCB concentrations in water have been reported by GE during low
flow conditions, with higher concentrations in the nearshore area. Based on theoretical
considerations and an analysis of the available nearshore and center channel data collected by
GE, it was determined that lateral gradients in concentration are likely to be significant only at
lower flows, approximately less than 4,000 cfs at Fort Edward. Under conditions of flow less
than 4,000 cfs and upstream concentrations of total PCB greater than 15 ng/1, the average ratio of
TED-West to center channel concentrations is 1.14, while for flow less than 4,000 cfs and
upstream concentrations less than 15 ng/1 the average ratio is 1.45. Both ratios are significantly
different from unity. At upstream flows greater than 4,000 cfs, the ratio is not significantly
different from 1.0 and no correction is required.
3.5.4 Temporal Scales for Estimating Exposure to Fish
FISHRAND uses mean monthly dissolved water concentrations, and annual average
sediment concentrations. Sediment concentrations show significant spatial heterogeneity, but
little variation over time. Very little is gained by specifying monthly average sediment
concentrations versus annual averages. Dissolved water concentrations, by contrast, show
significant temporal variability.
28 MCA/TetraTech
-------
The expected value for spatially and temporally averaged exposures is obtained under the
assumption that concentrations follow a lognormal distribution. Under this assumption, the
expected value is given as:
£[x] = eln(W/2 (3-18)
And the variance as:
V[x] = (E[x])2+elaa2-] (3-19)
3.5.5 Application Framework
The FISHRAND model was coded in Fortran-90 with a user interface developed in
Microsoft Excel. It is implemented as a Microsoft Excel add-in program and can be run
interactively. To demonstrate model functionality, the model was run in a steady-state,
deterministic manner to demonstrate and verify concordance with the Gobas (1993) published
results.
Level 1: Using generic model parameters derived from monitoring data and model
constants obtained from the literature without site-specific calibration.
Level 2: Calibration of the model and parameters using available site-specific data.
The generic model application (Level 1) can be used to show general validity of the
modeling approach. The generic model should capture major bioaccumulation processes in
different ecosystems and across sites. General consistency of the generic model predictions with
observed experimental data for a number of different sites can be used to judge model validity.
This report presents the site-specific application of FISHRAND (Level 2). Often
calibration is done by simple model fitting to the observed data without consideration of
associated uncertainties. For non-linear models such as FISHRAND this approach may lead to
unreliable predictions. Our approach to FISHRAND calibration incorporates robust statistical
methods applicable to non-linear models. A sensitivity analysis relying on elasticities is used to
select several of the most important parameters for calibration. Likelihood profiles are used to
select ranges of variation for these parameters and to assign corrected prior distributions for these
parameters. Finally, parameter distribution updating using Bayesian Monte Carlo techniques are
used to incorporate the full range of experimental data to derive posterior distributions for the
model parameters with which future predictions are to be made.
3.5.5.1 Initial Validation: Comparison with Gobas (1993) Lake Ontario Data for the Steady-
State Case
As an intermediate step in the FISHRAND development, a deterministic version of the
FISHRAND model, FISHPATH, was developed and run in a steady-state, deterministic manner
to demonstrate and verify concordance with the Gobas (1993) published results.
29 MCA/TetraTech
-------
The steady-state solution is given by:
' Cdiet
ki + ke + km + kg
(3-20)
Figure 3-3 shows the comparison between FISHRAND, and published data from Gobas
(1993). Note that the figures also include a comparison to a deterministic version of the model,
called FISHPATH, that was used during the development of FISHRAND. EPA has since
discontinued the use of FISHPATH. Pages 1 and 2 of this figure present the variables used in the
model. Page 3 describes bioavailability in the water column and bioaccumulation in
phytoplankton and zooplankton. This page also shows the predicted results from the Gobas
model as published ("Predicted" in the table), observed results from field observations, and the
results from FISHRAND run in steady-state (final column). The final box shows the result from
FISHPATH. Page 3 shows that FISHRAND and the original Gobas predictions show good
agreement.
Page 4 of Figure 3-3 shows the comparison for benthic invertebrates. FISHRAND and
the Gobas model as published show identical predictions. Pages 5 and 6 present the equations
used for fish uptake , while page 7 presents the final comparisons between the Gobas model as
published (1993), field observations, and FISHRAND. FISHRAND predicts virtually identically
to published Gobas results, indicating that the model is performing as published.
3.5.5.2 Initial Validation: Comparison with Gobas (1995) Lake Ontario Data for the Time-
Varying Case
Figure 3-4 shows the comparison between FISHPATH, FISHRAND, and published data
from Gobas (1995). FISHRAND and FISHPATH were run using inputs specified in Gobas
(1995) and compared to results published in that article. Model results showed concordance with
the published data, indicating that the models were correctly coded and ready to be modified for
use in the Hudson River modeling application.
3.5.5.3 Calibration Approach: Sensitivity Analysis
Calibration focused on a few parameters that are (i) considered highly uncertain, and (ii)
important for model performance. To determine the most important parameters, a sensitivity
analysis was conducted using the approximate analytical solution to the Gobas model for small
time intervals t:
C((t) = '2 * [1 - exp(-a3 * 01 + C f (0) * exp(-a3 * t) (3-21)
30 MCA/TetraTech
-------
where
a2=kd*Cd
a3 = (k2 + ke + kM + kg )
In particular, (3-21) provides the steady state Gobas' solution when t > °° and the initial
condition when t > 0 .
The rate coefficients in the Gobas model are functions of 1 1 constants:
C/: the constant equal to 88.3 in equation (3-9);
d\ the power constant equal to 0.6 in equation (3-9);
C$. the constant equal to 100 in equation (3-10);
4. the constant equal to 5.3*10"8 in equation (3-13);
Cs\ the constant equal to 2.3 in equation (3-13);
Q: the constant equal to 0.022 in equation (3-14);
7'. 0.85 in the same equation (3-14);
C«: the constant 0.06 in equation (3-14);
C9: 0.2 factor in equation (3-15);
CIQ. 0.01 constant in equation (3-16);
: the 0.2'power in the last equation for T > 25°C) (3-16).
As there are six individual fish species, there are 66 additional variables are considered as
unknown in addition to the environmental variables (e.g., fish weight, diet, etc.). It is not
mathematically feasible to obtain best estimates of all 66 variables.
The sensitivity analysis focuses on the relationship between predicted fish body burden
and these 1 1 constants. After obtaining the partial derivatives, elasticities were estimated.
Elasticities interpret the effect of a percentage change in the independent variable on the
dependent variable. The elasticity for a parameter is calculated at the point of the means of each
of the independent variables as:
(3-22)
31 MCA/TetraTech
-------
where:
Y = ZTri+ PCB concentration in fish; and,
X = each of the constants C/ - C/y above, as well as the user-specified parameters.
The analytical expressions for the elasticities were obtained using the Maple software
package. The expressions of the derivatives were coded in FORTRAN and the elasticities were
calculated simultaneously with the calculations of the concentrations themselves. Elasticities
averaged over time and environmental parameters were considered in selecting the parameters to
evaluate using likelihood profile methods and then for updating in the Bayesian Monte Carlo
procedure.
3.5.5.4 Calibration Approach: Likelihood Profiling
Likelihood Profiling is a powerful technique to determine confidence limits for
model parameters. Likelihood profiling was implemented in a simplified FISHRAND model
(fish diets were not randomized, and the simulation scheme was modified to eliminate random
fluctuations of the calculated concentrations for a fixed value of the model parameter selected for
the profiling) to derive corrected prior distributions. Bayesian updating (see chapter 3.5.5.5) was
then used to derive posterior parameter distributions in the complete FISHRAND model.
Likelihood profiling is used before formal Bayesian updating for three reasons: a) To obtain best
estimates and probability for the parameters for which empirical distributions were unavailable
(e.g., Gobas model constants such as growth rate, assimilation efficiency etc.); b) To reduce
uncertainty and assign narrower distributions for the parameters for which empirical
measurements resulted in very broad and/or highly uncertain distributions (both methods are
based on the same likelihood function, so if the likelihood function has a higher value in the
profiling, it will correspondingly have a larger probability following Bayesian updating); and c)
to minimize the intensive computation required by the Bayesian updating calibration technique
(both methods are based on the same likelihood function). Often the empirically derived prior
distributions do not adequately reflect values in the tails, and this method allows for a better
representation of the full distribution.
The idea of extending likelihood profiling to compute parameter distributions has been
suggested by Bates and Watts (1988) but it was only recently that appropriate computational
methods have been developed (Quinn et al., 1999).
The likelihood function reflects consistency between experimental and modeled data,
The following form of likelihood function is used in our calculation assuming lognormal
distributions of measured body burdens:
l
y(x)) = -]= exp- (\n(yj-\n(y(x)))2 - (3-23)
i
L(y
where:
ym = measured concentration in fish,
y = calculated concentration from the model,
32 MCA/TetraTech
-------
aym = measured standard deviation for fish concentration, and,
x = vector of model parameters.
Equation 3-23 provides the likelihood function for one measurement. The method uses the
product of likelihood functions for all available measurements.
The likelihood function 3-23 depends on model parameters. The values of x for which
likelihood function is maximized are called the maximum likelihood estimates (MLE) for the
model parameters. The likelihood ratio is defined as the ratio of the likelihood function for
specified values of x and the likelihood function for the MLE of x.
(3.24)
The statistical inference about parameter values is based on the fact that -2Ln(LR) is
asymptotically distributed as %2 (r) where %2 is the chi-squared distribution and r is the number of
degrees of freedom equal to the dimension of vector of parameters x.
The likelihood ratio Ln(LR) is plotted as a function of only one model parameter x in
small increments from its maximum likelihood estimate. For each fixed value of the selected
parameter x in equation 3.24, the likelihood function in the numerator is maximized with respect
to all other model parameters. This maximum is calculated using numerical simplex methods.
The tables of chi-square distributions with one degree of freedom are used to build confidence
intervals for the parameter. These intervals were used to approximate either normal, lognormal,
or triangular parameter distributions.
The likelihood profile can be utilized only for a simplified model. Therefore Bayesian
updating is used for calibration of the complete model.
3.5.5.5 Calibration Approach: Formal Bayesian Updating Procedure
FISHRAND implements a "multi-dimensional" Monte Carlo approach in which all model
parameters are categorized as uncertain or variable. Variable parameters are those reflecting
population heterogeneity, while uncertain parameters reflect lack of knowledge. This Bayesian
approach reduces uncertainties in our knowledge about the simulation of uptake of PCBs, and
improves our knowledge about natural variability in the system. The user can assign
"uncertainty" and "variability" attributes to all model parameters interactively. The general
scheme of random sampling implemented in FISHRAND is presented in Figure 3-5.
The FISHRAND model incorporates distributions instead of point estimates for model
parameters. The calibration approach takes advantage of this feature by incorporating Bayes
Rule. The procedure is as follows. Using the distributions specified in Level 1 (generic model
33 MCA/TetraTech
-------
constants together with site-specific distributions of lipid content, sediment and water
concentrations generated from the HUDTOX model, fish weight, KOW, and dietary preferences),
the model generates distributions of fish body burdens for each species, location, and year.
These simulated values are compared to available NYSDEC monitoring data. The model output
and the observation are reconciled using Bayes rule to determine a posterior mass function for
the model output, that is, the distribution that leads to a best fit between model output and
observations. The algorithm proceeds as follows:
1. Define a prior density p(xj)on all model parameters (Xj e Jc) (obtained from site-
specific data for lipid, weight, TOC, etc.)
2. Sample from this distribution n times using a Monte Carlo scheme and generate
sample inputs jt, (i = 1,«) for each (Xj e x)
3. Run the model for each set of input samples to determine the sampled output
yio (i = l,n) for model output of interest y0 (y0 e y).
4. Evaluate the likelihood function L(ym y,,0)for each sample y,0of model output
yg using the error structure of the observed data/measurement.
5. Reconcile the model output and observation using Bayes Ruleobtain posterior mass
function for inputs and outputs.
Figure 3-6 provides a schematic of the Bayesian updating procedure. Using Monte Carlo
simulation, a representative sample jt(. . (i = 1, n) for each input Xj is generated from the initially
specified (prior) distributionp(x.). A probability mass function p(xt ,) = is associated with
n
the ith sample x,. . (i = 1, n) for each input jc;. The model is then run iteratively n times for each
vector of sampled inputs. This results in n sample values, yio(i = l,n) for model output y0 each
with probability mass function equal to -. If the log-error (e) on ym is normally distributed
n
(assuming a lognormal distribution for XTri+ PCB concentration in fish) with a mean of zero and
standard deviation oym, then the likelihood function L(ym y/0) is expressed as:
11 1
1 1 / \2
y,- 0;= p= CXP r "n^".) ~ m^.«) J (3~25)
'0/ - r^- ic
vm \m
The posterior mass function for each sample yia is determined from Bayes rule as:
34 MCATTetraTech
-------
v->)
v/,0)
(3-26)
Since p(yf ) = , equation (3-26) readily reduces to
n
v
V ) =
J m /
y,.0)
(3-27)
It is critical to recognize that for each simulated replication, the same posterior probability mass
function is associated with inputs and outputs. Hence, the posterior probability mass function for
input samples xLj is also given by:
(3-28)
y
I st.o
The sample values xt and the associated posterior probability mass functions P(xitj y\
characterize the posterior density function for model input x}.
Systematic and unrecognized errors between models and experimental data were found to
result in false precision in Bayesian updating, i.e. posterior parameter distributions are estimated
to be narrow while in fact they are much broader (Small and Fischbeck, 1999). This false
precision does not significantly affect central estimates for the parameter distribution. One way
to address this issue is to implement Markov Chain Monte-Carlo sampling (MCMC). Given the
complexity of the FISHRAND model, implementation of MCMC would result in significant
computational difficulties. Therefore we implemented multiple regression analysis in which
parameter variance as derived from the likelihood profiling and Bayesian updating is matched
with experimentally-observed data by means of the least squares method (variance correction
procedure).
The FISHRAND calibration focused on optimizing wet weight concentrations. This was
done for three reasons. First, the model predicts a wet weight concentration in fish, and provides
lipid normalized results by dividing the predicted wet weight concentration by a percent lipid.
35
MCA/TetraTech
-------
Second, the lipid content of any given fish is difficult to predict from first principles alone.
Finally, potential target levels in fish are typically described as wet weight concentrations.
Optimizing the model for wet weight concentrations provides a more sound basis upon
which to make future predictions. In addition to predicting fish responses to changes in sediment
and water concentrations, it is also necessary to predict lipid content. Although it is possible to
obtain close to perfect agreement between model predictions and observed body burdens by
inputting the observed lipid concentrations for each year for which measurements are available,
this approach limits the ability of the model with respect to forecasts of fish tissue PCB
concentrations. The FISHRAND model predicts wet weight concentrations by relying on a
distribution of lipid values in each fish species that is representative of the observed variability in
lipid content. This provides a more robust basis upon which to make predictions.
3.5.6 FISHRAND Model Validation
To validate the model, several approaches were followed.
First, the calibrated model for one river mile was run for another river mile and predicted
body burdens compared to measured body burdens at this location. Satisfactory agreement for
both river miles implies model validity across locations in the Hudson River.
A second approach involves model calibration using only part of the available dataset and
comparison of model predictability of the remaining portion of the dataset. A good concordance
of the model prediction with observed data implies model validity within the timeframe of
available measurements and therefore model usability for the future predictions.
Finally, model predictions for the policy-relevant endpoints (such as concentrations at
some point in the future) can be compared for the model calibrated using all available
experimental data and then only a portion of the data. Closeness of the model predictions shows
robustness of the model.
36 MCA/TetraTech
-------
Chapter 4
-------
4. BIVARIATE BAF ANALYSIS OF FISH BODY BURDENS
4.1 Data Used for Development of Bivariate BAF Analyses
Equation (3-1) presents an idealized formulation for developing bivariate BAFs. Actual
implementation is constrained by data availability. Among other issues, quantitation methods
used for fish are not directly equivalent to those used for water, and quantitation methods have
changed over time. Establishing the spatial/temporal history of sediment concentrations also
presents difficulties.
Initial attempts to develop bivariate BAFs for the Hudson River were presented in the
PMCR (EPA, 1996), using data through 1992. Since that time, additional fish, water column,
and sediment data have become available, running through 1997. Additional evidence has also
been developed on the proper interpretation of historical Aroclor PCB quantitations. Finally, the
approach used for bivariate BAFs has been refined based on comments generated in EPA's Peer
Review of the PMCR and initial draft of the BMR. Data and methods used for development of
the bivariate BAF analysis are described below.
4.1.1 Fish Data
4.1.1.1 Locations and Species Analyzed
Statistical development of a bivariate BAF requires a sufficiently large range of data (over
differing environmental conditions in space and/or time) to distinguish accumulation originating
from water column and sediment pathways. As in the PMCR, the bivariate BAF analysis is
based on NYSDEC fish data from the Upper Hudson River below Fort Edward coupled with
NYSDEC data from the uppermost part of the Lower Hudson River (above River Mile 142).
Samples collected between River Mile 142 and 153 are from the freshwater portion of the Lower
Hudson. The species collected in this area are largely the same as those collected in the Upper
Hudson, and PCBs in this reach are derived primarily from the Upper Hudson. It is therefore
appropriate to include samples between River Mile 142 and 153 (if adjustment is made for the
lower exposure concentrations expected in this reach), thus providing a larger database for
analysis. Samples collected further downstream within the freshwater portions of the Hudson
were not included due to lack of contemporaneous measurements or estimates of water column
and sediment concentrations.
The longest-running and most extensive sample data in the Upper Hudson come from
NYSDEC collections at River Miles 168-176 (near Stillwater) and at River Miles 142 and 152
(below Federal Dam). A good representation over time is also available for River Miles 189-190
(lower Thompson Island Pool), and smaller amounts of data are available at River Mile 160
(Waterford, above Federal Dam). The species for which the most data are available are
pumpkinseed (Lepomis gibbosus), largemouth bass (Micropterus salmoides), and brown bullhead
(Ictalurus nebulosus). Lesser, but still extensive, data are available for goldfish (Carassius
auratus), white perch (Morone americana), and yellow perch (Percaflavescens).
37 MCA/TetraTech
-------
These species represent a range of trophic levels, habitat preference, and foraging
behavior: Largemouth bass are piscivorous, with adults occupying the top of the aquatic food
chain. Yellow perch represent an intermediate trophic level, foraging on invertebrates and small
fish. Unlike largemouth bass, yellow perch are migratory within the river. Adult white perch are
benthic predators, with older white perch becoming increasing piscivorous, and utilize both
shallow areas and the main channel bottom. The species is semi-anadromous, with spawning
occurring in the upper reaches of the Lower Hudson River and winter movement down river.
They are also found in the lower two lock pools of the Upper Hudson River. Pumpkinseed
occupy a lower trophic level, feed primarily on invertebrates, and are an important food source
for larger fish. Goldfish also occupy a lower trophic level, feed primarily on invertebrates in the
water column, and consume detrital algae. Brown bullhead are omnivorous bottom feeders, with
diet including offal, waste, small fish, mollusks, invertebrates, and plants. Feeding preferences
may vary with the age and size of the individual. Thus, a range of trophic positions and forage
preferences is available for analysis in the historic data. Appendix A provides more detailed
information on the foraging strategies of each of these species (except goldfish).
Data summaries for the NYSDEC fish analyses through 1988 were provided in the Phase
1 report, while the PMCR provided a summary through the 1992 sampling, with a total of 10,311
fish analyses available, of which 3,412 were collected between River Miles 142 and 194.
Additional data are now available for 1993 through 1997, including 994 NYSDEC samples
collected between River Miles 142 and 194, and some corrections have been made to the
database supplied by NYSDEC. Analyses presented in this chapter are based on a release of the
NYSDEC database provided on November 17, 1998, which contains some minor additions and
updates subsequent to the release of TAMS/Gradient Database Release 4.1.
4.1.1.2 Lipid Normalization
As described in Chapter 3, PCBs accumulate primarily in fish lipid tissue, and it is
appropriate to normalize fish body burdens to concentration on a lipid basis. This helps remove
variability in concentrations due to variability in individual lipid content. Nearly all the
NYSDEC fish analyses report percent lipid, so lipid-normalized concentrations are readily
calculated. It should be noted, however, that extraction and determination of lipid content is also
subject to uncertainty. This does not, however, present a major problem. Laboratory analyses
for PCBs are based on a lipid extract; thus the lipid-normalized concentration should be
consistent (except for round-off error) as long as the extraction procedures used for PCB and
lipid analysis are consistent, even though results are reported on a wet-weight basis. Error in
lipid determination primarily introduces error into reported wet-weight concentrations, which are
not used in the BAF analysis.
4.1.1.3 Season, Age, and Sex
PCB body burdens in fish may vary in accordance with seasonal growth and spawning
cycles. These bioenergetic factors are not included in the simple BAF approach; however, their
importance as potential confounding factors should be recognized. To help minimize these
38 MCA/TetraTech
-------
effects, only data from summer collections (May to September) were used. Within this time
period, collections for individual species have tended to be even more focused. Most summer
samples are in the May-June period for brown bullhead (95%), goldfish (100%), largemouth bass
(97%), white perch (100%), and yellow perch (100%). Pumpkinseed samples are predominantly
from August-September (90%). The empirical models, which result, will be specific to these
collection times.
Age of individuals also affects PCB body burden, as various PCB congeners tend to
bioaccumulate over time, feeding preference often shifts to higher trophic levels with increasing
size, and growth dilution effects change with age. Sex differences in PCB concentrations have
also been noted in the Hudson and elsewhere, perhaps due in part to loss of PCBs from females
when eggs are expelled (see Sloan et al. 1995). Within the historical database, age is usually not
given, and weight and length are uncertain surrogates. Sex determination is also missing for
many samples. Therefore, the BAF analysis has not accounted for age and sex effects, although
these undoubtedly contribute to the variability among individual samples.
4.1.1.4 Laboratories and Methods for PCB Analysis
An important conclusion of the PMCR (see also Butcher et al., 1997) is that valid
interpretation of historical trends in PCB concentrations cannot be made without consideration of
the changes in analytical methods which have occurred over time. That is, a comparison is valid
only when there is consistency in what is being measured. The most dramatic change in
analytical methods is that between the Phase 2 TAMS/Gradient data, using state-of-the-art,
capillary-column, PCB congener analyses, and older analyses based on packed-column
quantitation of Aroclor equivalents. The historical fish analyses in the NYSDEC database
primarily consist of packed-column Aroclor quantitations. Because an Aroclor is a complex
mixture of many individual congeners, interpretation of the historic Aroclor data raises difficult
technical issues. In addition, Aroclor quantitation methods have changed over time, and these
changes have significant implications for the interpretation of historical trends in the data and the
development of valid statistical relationships.
Shifts in laboratories may also influence results. A summary of samples between River
Miles 142 and 193 by laboratory and year is provided in Table 4-1. As will be seen from this
table, a majority of the Upper Hudson samples from 1977 on were analyzed by the same contract
laboratory (referred to for convenience as "Hazleton"), although this laboratory has undergone a
number of changes in name and/or ownership (see also Sloan et al., 1985). The major
exceptions are samples from 1991 to 1992, analyzed by NYSDEC's Hale Creek Field Station
("Hale Creek"). As described below, it has been possible to develop analyses of what was
actually measured (in terms of PCB congeners) by the various Aroclor quantitation methods used
by Hazleton and Hale Creek. This has not been possible for the six laboratories represented in
the "Other" category. Therefore, the analysis has been restricted to Hazleton and Hale Creek
results, 1977 to 1997.
Aroclor standards used by these two laboratories for quantitation, and NYSDEC
conventions for estimating total PCBs from Aroclor data, are summarized in Table 4-2.
39 MCA/TetraTech
-------
Quantitations by Hazleton for 1977 through 1990 are consistently based on analysis against
Aroclor 1016 and Aroclor 1254 standards on packed column GC; an Aroclor 1221 standard was
used on most Hazleton analyses through early 1993, but not thereafter. Reported detection limits
range from 0.01 to 1.0 ppm wet weight for each Aroclor, with detection limits for most samples
at 0.1 ppm, and the vast majority of samples collected between River Miles 142 and 193 were
reported with values above quantitation limits for both Aroclor 1016 and Aroclor 1254. Total
PCB concentrations in fish through 1990 were calculated by NYSDEC as the sum of Aroclor
1016 plus Aroclor 1254, because (1) 68 percent of the total Aroclor 1221 results, and 55 percent
of those between River Mile 142 and 196 are reported as nondetects (versus less than 1 percent
nondetects for Aroclor 1016 and Aroclor 1254 in this portion of the river); (2) Aroclor 1221
quantitations are not available for later data; and (3) when Aroclor 1221 is detected, substantial
double-counting may occur between quantitations to Aroclor 1016 and Aroclor 1221 standards.
Hazleton analyses through 1990 are discussed in detail in the PMCR and in Butcher et al.
(1997). These analyses against Aroclor standards on an OV-1 stationary phase were based on
only a few packed-column peaks, and are sensitive to the quantitation method used, which has
changed over time. Estimating an Aroclor concentration from a few peaks can introduce
significant error in estimates if the environmental distribution of PCB congeners differs from that
of the unaltered Aroclor standard. After commencing in 1977, quantitation peaks were changed
in 1979 and in 1983; the 1983 quantitation scheme was used consistently through 1990 (see
Sloan and Jock, 1990; Armstrong and Sloan, 1988). Hazleton analyses from 1992 on substituted
an Aroclor 1248 or 1242 standard for Aroclor 1016, and added Aroclor 1260. Quantitation
peaks for the 1992 to present Aroclor 1248 method were tentatively identified from area reports
and sample calculation sheets provided by EnChem, successor to Hazleton, coupled with
interpretation of sample chromatograms to identify peaks identified on absolute retention time
(RT) in terms of retention time relative to p,p'-DDE (RRT), as used by Webb and McCall (1973)
and others. Packed-column GC peaks and associated congeners are summarized in Table 4-3.
For 1991-1993, the database contains many fish analyses for Aroclors performed using
capillary column GC at NYSDEC's Hale Creek field station. The analytical approach is
documented in "Analytical and Laboratory Procedures at Hale Creek Field Station", which
contains the method documentation for "OC 1.103. Organochlorine Residues", dated 9/27/1990.
The Hale Creek analyses were performed on a Perkin-Elmer Sigma 115 with SPB-1 methyl
silicone bonded phase capillary column. The Control inputs attached to this method appear to
show that Aroclor 1016 was analyzed via 7 capillary column peaks (with retention times relative
to p,p'-DDE ranging from 0.73 to 0.87), and Aroclor 1254-1260 (combined) by 14 peaks (with
retention times relative to p,p'-DDE ranging from 0.96 to 1.31). A specific identification of
congeners associated with these SPB-1 peaks has not been made.
4,1.1.5 Standardization of PCB Analytical Results
The "Hazleton" and Hale Creek results in the NYSDEC database include Aroclor
quantitations by five different sets of methods/quantitation peaks. As demonstrated in Butcher et
al. (1997), these shifts in quantitation can introduce spurious apparent changes in reported
Aroclor and total PCB concentrations in fish. For instance, the change in quantitation peaks
40 MCA/TetraTech
-------
between 1977 and 1979 is estimated to result in an apparent decline in Aroclor 1016
concentration of approximately 40 percent, regardless of actual environmental trends.
It is thus essential to establish a consistent quantitation basis, or "translation" procedure,
to develop an empirical analysis of trends in fish concentrations and correlations between fish
body burdens and environmental concentrations. Development of translations for historical data
has relied on a weight of evidence approach. Three separate lines of evidence have been
pursued:
Split Sample Analyses, in which one sample is split and analyzed by different
methods. This is the most direct approach, but is available for only a limited number
of methods and samples.
Interlaboratory Comparisons, designed to evaluate contract laboratory performance.
The interlaboratory comparisons are similar to split samples, in that they provide
direct comparison between methods, but do not provide detailed documentation on
methods used.
Theoretical "What If?" Analyses, in which the performance of historical Aroclor
quantitation methods is evaluated in terms of PCB congeners, based on interpretation
of congener data "as if analyzed by the historical methods.
The baseline or reference condition for the development of translation procedures is taken
as the sum of PCB congeners as quantitated by Aquatec for the TAMS/Gradient Phase 2
sampling. Translations have been developed for two targets: total PCBs (i.e., sum of quantitated
congeners, consisting of 90 target and 36 non-target congeners and representing more than 90
percent of the total concentration of Aroclors 1016, 1242, and 1254, as described in the DEIR,
Appendix A), and the sum of trichloro- through decachlorobiphenyls (denoted STri-f). The latter
target was selected for the BAF analysis because most of the historical monitoring of PCB
concentrations in water and sediment is most readily interpreted in terms of ZTri+, as described
in Volume 1, Chapter 2.6 of this report. Because fish tend not to accumulate significant amounts
of mono- and dichlorobiphenyls, translations of historical quantitations to either total PCBs or
ZTri+ are expected to be similar.
4.1.1.6 Theoretical "What If?"Analyses
The theoretical analysis is presented first, because it can be developed for all the
"Hazleton" methods and provides some insights for interpreting the limited data available from
split samples and interlaboratory comparisons.
An interpretation of what was actually measured in historical packed-column analyses can
be made by converting the TAMS/Gradient Phase 2 fish congener data to equivalent Aroclor
measurements as //"analyzed by NYSDEC methods. According to Sloan et al. (1984):
41 MCA/TetraTech
-------
Quantitation was done by comparing several peak heights or areas to those
produced by the respective Aroclors. The principal peaks used for quantitation
include a single one for Aroclor 1221 representing a monochlorobiphenyl; two for
Aroclor 1016 reflecting mixtures of trichlorobiphenyl; and three peaks for Aroclor
1254 primarily composed of tetra-, penta- and hexachlorobiphenyl congeners.
While the NYSDEC method employs several peaks for Aroclor quantitation, these are
evaluated via a single composite response factor. Given selection of m packed-column peaks for
quantitation, the reported Aroclor value is obtained as
I I
[Aroclor] = £ area.} RFS (4-1)
where
area; = the area associated with packed-column peak j, and
RFS = a composite or net response factor defined as the concentration of standard
Aroclor injected divided by the sum of the peak areas of the selected packed-
column peaks.
The area within the selected packed-column peak is related to the sum of the
concentrations of individual PCB congeners associated with those peaks by congener peak
response factors:
^, A [congene^ ]
2rfareaj = 2-j^ l>4'2)
where
n = number of congeners associated with selected packed
column peaks,
[congenen] = concentration of an individual PCB congener i associated
with the selected packed column peaks, and
RFCj = the response factor for congener i, defined as the
concentration of congener i in the Aroclor standard
divided by the peak area contributed by this congener.
If the congener response factors within the individual peaks are relatively consistent, this
may also be approximated as
42 MCA/TetraTech
-------
congene^ ]
(4-3)
where
RF = area-weighted mean response factor for the selected packed
column peaks or their constituent congeners in a capillary
column analysis. RF is defined as the concentration of the
Aroclor standard times the weight percent of PCB congeners
contained in the selected peaks divided by the peak area, or:
2, wt % Peak j X wt % congene^
RFp = [ Aroclorstd ] - ^ - = [ Aroclors(d ] **. -
k=l
Substituting Equation (4-3) into Equation (4-1) yields
n D
[Aroclor] « \ [congener; ] -- - (4-4)
Because the ratio of the response factors on the right-hand side of this equation is
equivalent to the inverse of the weight percent of total PCBs contained in the selected packed
column peaks, this simplifies to:
n
^ [congener; ]
[Aroclor] « -^ - (4-5)
where the denominator represents the total weight percent of the Aroclor contained in the
congeners making up the packed column peaks used for quantitation. The relationship is only
approximate, because the response factors of individual congeners are not equal. Calibrated
response factors for the congeners that are (1) included within peaks used for quantitation of a
specific Aroclor and (2) regularly detected in Hudson River biota were, however, found to vary
over a small range, and, in most cases, estimated response factors relative to BZ #52 for these
congeners are within 15% of unity. Thus, the simple approximation of Equation (4-5) is judged
to provide an adequate basis for comparing historical packed-column GC analyses with more
recent capillary column results.
As indicated by Equation (4-5), translating between congener data and historical Aroclor
quantitations also requires the total weight percent of the quantitated peaks in the Aroclor
standards. These values were obtained by summing the weight percentages of congeners
43 MCA/TetraTech
-------
associated with packed column peaks in Aroclor standards (see Table 4-3) as developed from
analyses of Aroclor standards in the Phase 2 laboratory effort. The weight percentages are given
in Table 4-4. It should be noted that weight percentages reported for individual congeners in
Aroclor standards vary considerably (e.g., Albro and Parker, 1979; Schulz et al., 1989; Draper et
al., 1989 for Aroclor 1016). Some of this variability is likely due to batch differences in Aroclor
standards, and some to analytical methods. For purposes of this study, it is most important to use
consistent results for Aroclor standards analyzed by the same methods and laboratory as the
reference biological data.
Estimates of ZTri+ obtained from the Phase 2 congener data may be regressed against
total PCB estimates by Aroclor quantitation "as if calculated by Hazleton methods to yield a
translator. Regression results are summarized below and in Figure 4-1. Standard errors for the
dependent variable estimates and for each coefficient are shown in parentheses below the
equation.
ZTri+
(862.7)
nri+
(881.6)
ITri+
(961.8)
ITri+
(1762)
-200.7
(97.2)
-62.5
(98.7)
-216.5
(108.4)
-111.0
(198.4)
+ 0.08720 x 1977 Sum (1016+1254)
(0.0065)
+ 1.224 x 1979 Sum (1016+1254)
(0.0093)
+ 1.320 x 1983 Sum (1016+1254)
(0.0109)
R2 = 99.4%
R2 = 99.3%
R2 = 99.2%
+ 0.8798 x 1992 Sum (1248+1254+1260) R2 = 97.3%
(0.0135)
4.1.1.7 Split Sample Comparisons
The NYSDEC database (11/17/98 update) contains a limited number of fish samples
analyzed for PCBs by multiple laboratories. Most relevant for the "Hazleton" analyses are splits
of 1995 samples from the Hudson analyzed by both Hazleton (using the 1992 method) and
NOAA (using capillary column GC analysis comparable to the Aquatec results). There are two
other series of splits between Hazleton and Hale Creek (1987 Smith Pond, 1996 Queensberry
area), but for these samples Hazleton reports against Aroclor 1016 and 1254/60 standards.
Hazleton thus apparently used a version of the Hale Creek method, and not their own "1992"
method for these analyses. There are also 1993 split samples between Aquatec and Hale Creek
for pumpkinseed in the Upper Hudson. These samples may be matched on tag number to
identify true split samples.
The 1995 Hazleton-NOAA splits consist of 20 largemouth bass (collected between river
miles 113 and 189) and 35 striped bass (collected between river miles 27 and 152) quantified for
107 target congeners. In 54 out of the 55 samples the total calculated by Hazleton was greater
than the total calculated by NOAA (the one exception is the most highly contaminated sample).
The slope of a regression of the NOAA results against the Hazleton results is 0.87, and is not
44
MCA/TetraTech
-------
significantly different from the theoretical relationship obtained between sum of congeners and
the Hazleton 1992 method using the "What if?" analysis presented above. The split samples
thus appear to confirm the theoretical analysis.
The 1993 Hale Creek-Aquatec splits consist of 15 pumpkinseed samples, including three
highly contaminated specimens from Griffin Island. For. 13 of the 15 samples, the total reported
by Aquatec using capillary column GC is higher than the Hale Creek Aroclor sum. The two
exceptions are very lightly contaminated specimens. The slope of a regression of the Aquatec
results against the Hale Creek results is 1.46, with an R2 of 94%. This result is consistent with an
interpretation that Hale Creek analyses are approximately equivalent to Hazleton analyses by the
1983 method.
The results of 1997 split samples between EnChem (successor to Hazleton) and GE's
contractor NBA (identified to peak/congener basis by capillary column GC) are not yet ready to
be released or reported in detail, but results of 56 samples were made available for preliminary
inspection by NYSDEC. The theoretical "What if?" analysis suggested that the 1992-1997
Hazleton/EnChem Aroclor method should result in substantially higher results than the 1983
Hazleton method, and should yield a slight overprediction of the sum of congeners, with a slope
of about 0.90 for congener sum versus Hazleton Aroclor sum. The provisional data suggest that
this is indeed the case, as the EnChem Aroclor sum appears to be consistently higher than the
NBA sum of congeners. The average ratio between NEA and EnChem results is approximately
equal to the theoretical slope of 0.90. Regression analysis suggests that the over-prediction could
be even greater. However, it should be noted that the NEA congener analysis is not necessarily
fully equivalent to the Aquatec congener analysis that serves as a baseline for our comparison.
Thus, the provisional 1997 data also appear to confirm the theoretical analysis.
4.1.1.8 Interlaboratory Comparisons
NYSDEC has conducted several rounds of interlaboratory comparison for contract
laboratory evaluation. Results for 1989, 1992, and 1995 comparisons were provided by
NYSDEC. For the 1989 study, eight laboratories participated, analyzing four samples. These
samples are not identified, but three of the four appear to have had significant PCB
contamination. The 1992 study included twelve laboratories and analysis of five samples (two
Lake Ontario coho salmon, clean largemouth bass composite, Hudson River striped bass, and
great horned owl tissue). The 1995 study involved four laboratories and three samples. One of
the samples was a composite of previously analyzed fish with no detectable PCBs. Samples 2
and 3 were splits of the same sample, which was a composite of striped bass fillets collected
from New York City Harbor with less than 1 ppm PCBs. Hazleton and Hale Creek participated
in each of these interlaboratory comparisons. The quantitations were to Aroclor standards of the
individual laboratory's choosing, and separate reference analyses for PCB congeners by capillary
column were not included.
No clear trend among laboratories is evident in the 1989 comparisons. Comparison of
Hazleton results is difficult, however, because Hazleton used Aroclor 1248, 1254, and 1260
standards, while Hale Creek results, using Aroclor 1016 and 1254/60 standards, predate their
45 MCA/TetraTech
-------
1990 methods documentation. Hazleton results were lower than Hale Creek on the two more
contaminated samples (total PCB concentration of about 10 ppm), and higher than Hale Creek on
the two lightly contaminated samples (less than 1 ppm). Comparison is also hampered by not
knowing which (if any) samples are Hudson River fish. Samples that represent congener/Aroclor
mixtures significantly different from those found in the Hudson River would likely provide
different results on a comparison of Hazleton and other methods.
In the 1992 interlaboratory comparisons, Hazleton Environmental Services (HES) used
Aroclor 1242, 1254, and 1260 standards, which approach differs from the methods used by
Hazleton for Upper Hudson River fish samples in the 1990s. 1992 Hale Creek analyses were
apparently done using their capillary column method OC 1.103, as discussed above. Hazleton
and Hale Creek were in relatively close agreement for four of the five samples, including all the
fish samples. The major discrepancy is in the analysis of the owl tissue, for which Hazleton
reported 4.5 ppm total PCBs, versus 1.5 for Hale Creek. One reason for the discrepancy is that
Hazleton quantitated this sample as Aroclor 1260 only. Hazleton's "1992" method for Aroclor
1260 uses only three peaks, which represent the more chlorinated end of the 1260 spectrum,
accounting for only about 8 percent of the total mass of Aroclor 1260. Scaling up to total PCBs
from a few peaks at one end of the spectrum is likely to result in significant potential for mis-
estimation. In all the fish samples, Hazleton's results were somewhat less than those reported by
Hale Creek, with an average difference of -13%. The discrepancy is greatest (-21%) for the
Hudson River striped bass sample.
In the 1995 interlaboratory comparisons, Hazleton used their standard "1992" approach of
quantitating to Aroclor 1248, 1254 and 1260 standards. For the two contaminated 1995 samples,
results from Hazleton were approximately 1.4 times those from Hale Creek. The report
transmitting the 1995 results (memorandum from Larry Skinner to Robert Bauer, January 17,
1996, Comparison Study of Contract Labs for Total PCB and % Lipids) states: "All laboratories
were in the acceptance limits of ±3 standard deviations of the mean, with laboratory 2 [Hazleton]
being consistently higher than the rest." The ratio of Hazleton to Hale Creek in 1995 is
consistent with predictions from the theoretical analysis of 'Hazleton' methods, assuming that
Hale Creek results are similar to Hazleton 1983 method results."
4.1.1.9 Translation Methods
The available evidence suggests that the "What if?" analyses provide a reasonable basis
for translating "Hazleton" Aroclor results to a basis consistent with congener analyses.
Approximate translation of the Hale Creek Aroclor data can be based on the analyses of split
samples described above.
Regression relationships between Aroclor sum and congener total can be performed with
or without a constant. In most cases, it was found that the constant was not significantly different
from zero. In addition, a zero-intercept regression is attractive because (1) samples detected as
near-clean by packed column are best interpreted as likely to be near-clean on capillary column
analysis as well, and (2) a zero-intercept regression will prevent prediction of any negative
concentrations on transformation. Therefore, zero-intercept results are presented below.
46 MCA/TetraTech
-------
Resulting zero-intercept translation methods for the state variable ITri+ are presented
below. Applicable laboratory codes from the database are also indicated. Note that the proposed
translation factors are only applicable to the laboratories for which they were developed.
Time
Period
1977-
1978
1979-
1982
1983-
1990
1990-
1993
1992-
1997
Equation
0.8642 (Aro 1016 + Aro 1254)
1.2210 (Aro 1016 + Aro 1254)
1.3070 (Aro 1016 + Aro 1254)
1.4157 (Aro 1016+ Aro 1254/60)
0.8754 (Aro 1248 + Aro 1254 + Aro 1260)
Applicable
Laboratory Codes
WI, RAL
RAL, HAZ
HAZ, RAL, HES
HC
HAZ, HES, EC
The annual averages of £Tri+ PCB concentrations (as mg/kg-lipid) for summer-collected
fish samples, arranged by species and a "group" designating location, are shown in Table 4-5.
The original NYSDEC data, contained in the TAMS/Gradient database, have been corrected to a
consistent ZTri+ basis using the relationships described above.
4.1.2 Water Column Data
As noted in the PMCR (USEPA, 1996) and earlier by Brown et al. (1985), a good
predictor of annual average PCB body burden in many fish species appears to be the summer
average water column concentration. Therefore, the BAF analyses use summer averages of water
column data, based on observations for May through September for consistency with the
averaging period used for fish. For fish collected in May or June this means that the water
column average includes samples from after the time of fish collection. Given the relative
sparseness of water column observations, however, it appears likely that including all water
column data for May through September will provide a better statistical estimate of
concentrations in a given season than restricting the estimate to May-June observations only.
For most of the period of fish sampling, the only data available on water-column
concentrations are the USGS monitoring. These data commence in 1977 for most locations in
the Upper Hudson, with 6 to 58 samples per station per year. Sampling locations and
methodology were described in detail in the Phase I Report (USEPA, 1991b). For the Phase 2
analysis, USGS data have been obtained through the end of Water Year 1997. Significant
47
MCA/TetraTech
-------
corrections and updates to the USGS data have occurred since the release of the PMCR, and are
reflected in Database Release 4.1.
There are three major sources available for the USGS water column PCB data:
WATSTORE, USGS/Albany NWIS database, and printed USGS Water Resources Data, New
York. For some years there are significant discrepancies between these data sources, requiring a
retrospective reconciliation. Data used in the PMCR were obtained primarily from
WATSTORE, but WATSTORE is a secondary source, which is periodically updated from the
USGS/New York NWIS electronic database system. Where discrepancies exist, WATSTORE is
less reliable than the other two sources. We noted major differences between these sources for
the period prior to October 1986, primarily related to (1) failure to reflect actual PCB detection
limit of 0.01 fig/1 for many observations, which was lower than the default detection limit of 0.1
|ig/l expected by WATSTORE for the relevant parameter codes, and (2) failure to report
identified Aroclors shown in the printed reports. Almost all USGS PCB data from the Hudson
from October 1983 on was quantitated at an 0.01 \ig l\ detection limit, but WATSTORE
generally does not show this until 10/86. In addition, a significant fraction of the data prior to
October 1983 was also quantitated at the 0.01 jig /I detection limit.
USGS PCB data were revised using both NWIS and the printed Water Resources Data.
For October 1983 through September 1986, data at the lower detection limit of 0.01 }ig/l are
primarily given only in the printed data, which is also the source for Aroclor identification. For
1978-1982, the printed data show total PCBs at a detection limit of 0.1 u,g/l and do not report
identified Aroclors; however, NWIS for these years shows that some samples were quantitated at
the 0.01 (ig/1 level and does show Aroclors.
USGS analyses prior to 1986 were obtained using packed-column GC; those from 1988
on used a capillary column methodology (personal communication from Ken Pearsall,
USGS/Troy, to Jonathan Butcher, Tetra Tech, based on letter received from Brooke Connor in
USGS Denver laboratory). It was previously believed that all analyses prior to November 1987
used packed column GC; however, QEA has obtained original chromatograms and sample
analysis sheets indicating use of a capillary column method as early as fall of 1986 (personal
communication from Jim Rhea, QEA, to Jonathan Butcher, Tetra Tech, 10/30/1998).
The USGS packed column methodology is described in general in Wershaw et al. (1983).
A clearer description of exactly what was done is given in Schroeder and Barnes (1983). The
analysis was a two-step procedure: (1) Determine an appropriate Aroclor standard, based on
requirements that at least 60 percent of the peaks in the standard are present in the sample and
"both relative peak ratios and column detention time must match." If a single Aroclor standard
cannot be found which matches these criteria, use a standard containing a mixture of two or more
Aroclors. (2) Calculate concentrations by dividing the area of a sample's identified PCB peaks by
the area of all peaks for an Aroclor standard, then multiplying this ratio by the concentration of
the Aroclor standard.
Step 2 indicates that this is not a Webb and McCall (1973) procedure with peak-by-peak
quantitation. Instead, the observed peaks in a sample are scaled-up to estimate a complete
Aroclor concentration. No compensation is made for differing response factors, only the sum of
48 MCA/TetraTech
-------
peak areas is used. It is not certain exactly which packed-column peaks were observed by USGS,
although it appears likely that the mono- and dichlorobiphenyls were not represented. The first
peak used is thought to be either RRT .21 or RRT .28. For quantitations against an Aroclor 1221
or 1232 standard (where there is substantial unobserved concentration in peaks below RRT .21)
this approach is equivalent to assuming that the early-eluting (unobserved) congeners in the
sample are present in the same fraction as in the Aroclor standard. In reality, concentrations of
these congeners (e.g., BZ#4) are likely to be higher in the environment due to dechlorination. In
addition, USGS used a dual column method, and always selected the lower of the two values
obtained. Finally, no corrections were made for incomplete extraction. Extraction efficiency, it
is estimated, probably exceeds 80 percent in nearly all samples.
Because of these factors, and the fact that the original chromatograms are not available, it
is difficult to predict exactly what was measured in USGS packed column analyses. For GE,
NBA conducted split sample experiments to compare the USGS packed column method (based
on the description in Schroeder and Barnes) to capillary column analyses, using individual or
mixed standards composed of Aroclor 1242, 1254, and 1221 (O'Brien & Gere, 1993). Updated
results of these analyses are contained in TAMS/Gradient Database Release 4.1 (1998).
Regression analysis of the split samples reveals that a linear relationship exists between USGS-
method total PCBs and capillary column ZTri+, with an intercept not significantly different from
zero and a slope not significantly different from one. Thus, the USGS packed-column data can
be used as a direct measure of ZTri+.
USGS capillary column methods (in use after September 1986) are capable of detecting
responses to a wider range of PCB congeners; however, quantitations were still reported based on
composite response factors derived from manufactured Aroclor standards. This technique results
in potential biases in calculating either total PCBs or ZTri+, because the relative weight
percentages of congeners in the environment generally differs from those found in the Aroclor
standards. QEA (Rhea and Werth, 1999) investigated the potential biases in this method by
reanalyzing the original chromatograms from USGS 1987 samples from the Hudson River. This
reanalysis indicated that USGS capillary column quantitations for Aroclor 1242 provide an
approximately unbiased estimated of ZTri+, while the sum of Aroclor 1242 and Aroclor 1254 (as
used in earlier versions of this report) over-estimates ZTri+. Based on the results of this study,
the following conventions were applied to the USGS capillary column data:
1. When USGS capillary column direct quantitation for Aroclor 1242 is available, use
this number as an estimate of £Tri+.
2. When capillary column direct quantitation for Aroclor 1248 is present, but
quantitation for Aroclor 1242 is not, use Aroclor 1248 as an estimate of ZTri+.
3. When USGS capillary column data reports only total PCBs, not Aroclors, estimate
ZTri+ as 75 percent of the reported total PCB concentration.
Most of the historical USGS results are available only as whole water quantitations. Few
USGS samples distinguish dissolved and particulate PCB fractions, and almost no organic
carbon data were collected. Therefore, the preferred formulation of normalizing the particulate
49 MCA/TetraTech
-------
fraction corrected to an organic carbon basis cannot be employed. Instead, all regressions were
based on whole water, unfiltered PCBs. The BAFs for fish concentrations are thus relative to
whole water rather than organic carbon-normalized paniculate PCBs.
Starting in 1991, capillary-column determinations of PCBs in the water column are
available on a homologue and congener basis from GE. These high-resolution data are presumed
more accurate than USGS results, and may be used to directly estimate ZTri+. The same may be
done with TAMS/Gradient Phase 2 water column results from 1993.
Summer average water column concentrations were estimated at four locations,
corresponding to reaches with available fish sampling. Assignment of sources for water column
concentrations is shown in Table 4-6. For the period from 1991 on, capillary column PCB
analysis by EPA and GE is used where available. 1993 concentrations below Thompson Island
Dam are estimated from TAMS/Gradient Phase 2 monitoring. Flow-averaged samples are
available at Waterford, while instantaneous transect samples are used at Stillwater and Green
Island. During 1994-1996 neither EPA nor GE sampled below Thompson Island Dam, so USGS
data are used.
For 1991 on, GE Thompson Island Dam-West (TID-West) data are used to represent
water column concentrations in the lower Thompson Island Pool. Within the TIP, GE has
reported strong lateral gradients in PCB concentrations in water during low flow conditions, with
higher concentrations in the nearshore area. Nearshore concentrations are, however, theorized to
be the more relevant measure of exposure concentrations for fish and their food webs, which are
believed to rely to a much greater extent on the nearshore habitat than the channel habitat. Thus,
no bias correction factor is applied to the TID-West observations for use in the BAF analysis.
For the Thompson Island Pool prior to 1991, direct measurements are not available and
upstream USGS data at Rt. 197, Fort Edward is judged of limited value for determining exposure
concentrations, due to the gain in PCB concentrations within the pool. Therefore, Thompson
Island Pool concentrations are estimated from downstream measurements, scaled by a drainage
ratio where appropriate. Prior to 1987, scaled USGS Stillwater data have been used in preference
to Schuylerville data to estimate Thompson Island Pool concentrations because averages at the
two stations are generally similar, but greater sampling density is available at Stillwater. USGS
Fort Miller data, commencing in 1987, are assumed representative of outflow from the
Thompson Island Pool for 1987-1990.
Use of downstream data to estimate TIP concentrations prior to 1991 introduces a
potential inconsistency, as the downstream results will be more similar to center channel than
nearshore concentrations in the TIP when a lateral concentration gradient exists. Therefore, a
correction is applied to these data to approximate TIP nearshore concentrations. An analysis of
ZTri+ in 53 GE samples pairing TID-West observations to observations in the center channel or
immediately below Thompson Island Dam from Sept. 18, 1996 through September 15, 1998
suggests that a consistent bias in nearshore samples is only present when flow at Fort Edward is
less than about 4,000 cfs (so that lateral mixing is reduced). The relative bias is also smaller
when upstream concentrations at Fort Edward are higher, which increases center channel
concentrations. Under conditions of flow less than 4,000 cfs and upstream concentrations of
50 MCATTetraTech
-------
total PCB greater than 15 ng/1, the average ratio of TDD-West to center channel concentrations is
1.14, while for flow less than 4,000 cfs and upstream concentrations less than 15 ng/1 the average
ratio is 1.45. Both ratios are significantly different from unity. At upstream flows greater than
4,000 cfs, the ratio is not significantly different from 1.0 and no correction is required. Because
USGS sampling does not reliably track a parcel of water from Fort Edward to downstream
stations, and detection limits were often high, the upstream concentration criterion is difficult to
apply. Therefore, the estimated correction factor of 1.14 was used to correct all downstream-
inferred concentrations with flow at Fort Edward less than 4,000 cfs to approximate TIP
nearshore conditions prior to averaging.
For years other than 1993, direct water column monitoring results are not available below
Federal Dam (except for a limited number of early USGS data, all non-detects). Concentrations
in this reach are therefore estimated by drainage area scaling from Waterford or other upstream
stations. This scaling is equivalent to assuming that incremental flow from the Mohawk River
contributes insignificant PCB concentration. Summer average concentrations used for BAF
estimation are summarized in Table 4-7 and Figure 4-2.
4.1.3 Sediment Data
The second forcing function for the bivariate BAFs is sediment concentration. Fish may
accumulate PCBs from the sediment directly through the consumption of benthic organisms or
direct ingestion in the case of deposit feeders, or indirectly through the consumption of other
organisms which consume benthos. Surface sediment concentrations are anticipated to be
correlated to water column concentrations; however, full equilibrium with the water column is
likely to exist only at the interface, and not through the entire bioactive depth. In depositional
areas, sediment concentrations will resemble water column concentrations, but with a "memory"
integrating across several years. Further, because most of the movement of sediment occurs
during spring floods, sediment concentrations should be more closely tied to spring high flow
concentrations than to summer low flow concentrations. Thus, sediment concentration data
provides a separate, semi-independent exposure data series to the bivariate BAF. The Pearson
correlation coefficient between average water column and sediment concentrations used in this
analysis is 0.32.
Areally-averaged annual observations of sediment concentrations for reaches in which
fish collections occurred do not exist. Indeed, the sediment database covers only a few points in
time, including the 1976/78 NYSDEC survey of the Upper Hudson, the 1984 NYSDEC survey
of the Thompson Island Pool, the 1991 GE survey of the Upper Hudson, and targeted sampling
of hotspot locations in the 1994 EPA Low Resolution Sediment Coring program. As with the
fish data, there are significant analytical differences between these sampling campaigns. Finally,
sediment concentrations in the Hudson are known to exhibit a high degree of spatial
heterogeneity, so that inference from small samples may not be representative of a reach-average
exposure concentration.
Because of these limitations, observed sediment data are not used directly in the Bivariate
BAF analysis. Instead, predicted sediment concentrations, averaged over 0 to 4 cm depth, from
the HUDTOX model were used. For the HUDTOX hindcast run, all the available sediment data
51 MCA/TetraTech
-------
were processed to provide a consistent estimate of ZTri+ PCBs and the model was calibrated to
provide a reasonable fit to available observations in time and space. The HUDTOX predictions
thus provide a best-estimate, process-based interpolation of the available sediment data.
HUDTOX results are a smoothed estimate of observed data in space and time, which helps
minimize the effects of sparse data and analytical uncertainty on BAF estimates which depend on
spatially averaged exposure concentrations.
The calibrated HUDTOX model provides reach-by-reach estimates of ITri+ for the
Hudson River between Fort Edward and Federal Dam, with separate estimates for cohesive and
non-cohesive sediments. It is assumed that cohesive (fine-grained) sediment concentrations are
most relevant to fish exposure pathways from sediment independent of water column
concentrations. Accordingly, organic-carbon normalized concentrations of ZTri+ in cohesive
sediment are used for all reaches in which the model includes a cohesive sediment segment. For
the reach immediately above Federal Dam, the model does not include a cohesive sediment
segment; organic-carbon normalized concentrations of ETri+ in non-cohesive sediment were
used for this reach. The model provides logarithmic predictions of concentration by reach, which
are converted to arithmetic estimates of sediment exposure concentration as
arithmetic mean = exp((x^ + a2, / 2j (4-6)
where u,x is the average logarithm of concentration in the reach, and
-------
4.1.4 Functional Grouping of Sample Locations for Analysis
Four functional groupings of available data were formed for the purposes of analysis.
These represent the major fish sampling locations and associated environmental data. The
groups are:
Group 1: River Mile 188 to 193, the lower Thompson Island Pool from Griffin Island to
Thompson Island Dam.
Group 2: River Mile 168 to 176, the NYSDEC fish collection station near Stillwater.
Prior to 1997, samples are from River Mile 168.
Group 3: River Mile 155 to 157, Waterford area above Federal Dam (limited NYSDEC
sample collection only). Most of these samples are from River Mile 157, several miles above the
confluence with the Mohawk River.
Group 4: River Mile 142 to 152, the upper part of the Lower Hudson, below Federal
Dam. These stations are influenced by dilution from the Mohawk River. Most samples are from
River Mile 142 (Albany Turning Basin) and River Mile 152 (Green Island).
4.2 Results of Bivariate BAF Analysis
For a given location and year, the PCB analyses of individual samples for a given species
exhibit a high degree of variability, reflecting individual characteristics (e.g., age, weight,
condition, and life history) and intra-year environmental effects that cannot be addressed in the
simple regression approach described here. In contrast, the central tendency or mean of species-
location-year observations shows much less variability. Analysis of means used a weighted
regression, with weights given as the inverse of the standard error of the mean (Theil, 1971),
giving relatively less weight to smaller or less consistent samples. As expected, models on
means have much stronger predictive ability than models on individual observations. As the
intention of the bivariate BAF analysis is to provide initial information on the central tendency of
fish body burden response, models on the means are reported here.
In contrast to the PMCR (USEPA, 1996), all analyses presented here are in terms of
ITri+ PCBs. Quantitations of individual Aroclors potentially provide information on
bioaccumulation of lighter versus heavier Aroclors, as presented in the PMCR. However, the
changes in quantitation methods for fish (Chapter 4.1.1) make it difficult to draw inferences
regarding individual Aroclor quantitations over time.
Regression models were created by species for the four individual sample location groups
described above and across all groups based on (1) a standard BAF approach with univariate
regression on water-column concentration only, (2) univariate regression on sediment
concentration only, and (3) bivariate BAF regression on water column and sediment
concentrations. Results were generally consistent among location groups, implying that cross-
sectional models across groups are appropriate, so these are reported here. Results vary strongly
between species, as expected.
53 MCA/TetraTech
-------
For a given species, plots of mean fish body burden versus water column concentration
show a general positive correlation, but with variability which appears to increase with water
column concentration. Figure 4-3 displays scatterplot matrices for lipid-normalized fish
concentration versus water and organic carbon-normalized sediment concentrations for all six
fish species under consideration. The scatterplots include a 68.3 percent bivariate confidence
ellipse about the sample means, which helps visualize the strength of correlation. In all species,
except perhaps goldfish, there appears to be a strong positive correlation between fish body
burden and both water and/or sediment concentrations. However, the strength of the
relationships varies by species. For instance, brown bullhead have a stronger linear relationship
to sediment, while pumpkinseed have a stronger linear relationship to water concentrations. For
each species, regressions were conducted against water concentration only (standard univariate
BAF approach), against sediment concentration only, and against water and sediment
concentrations simultaneously (bivariate BAF approach). Table 4-9 shows results of regression
analysis of arithmetic average fish concentrations versus water concentrations. The percentage of
total variability explained by the regressions is fairly low (adjusted multiple R2, which adjusts the
standard R2 estimate of the percent of variability explained by the regression downward to
account for model improvement due solely to adding an extra variable, ranging from 34 to 71
percent); however, the coefficient on water column concentration is in all cases statistically
significant at the 95 percent confidence level. Models for all species except goldfish include all
NYSDEC fish data selected in Chapter 4.1.1. For goldfish, regression diagnostics suggested that
arithmetic average lipid-normalized concentrations for 1977 and 1978 at Stillwater (Group 2)
were high outliers. In each of these years, the average is strongly influenced by one extremely
high value, which raises the average about 50 percent. These samples might represent inaccurate
quantitations of either PCB or lipid content. The averages for these two years were recalculated
with the high outlier value eliminated before calculating the regression models shown in Tables
4-9 through 4-11.
Figure 4-4 shows a plot of ZTri+ lipid concentration in pumpkinseed versus summer
average water concentration, with labels indicating location group. A strong positive correlation
is evident, although the quality of fit is degraded by a few samples, particularly one from Group 1
that combines a high fish tissue concentration and low estimated water column concentration.
This may reflect a poor estimate of the water column exposure concentration in this year. The
scatterplot does not reveal strong evidence for scale-dependent variance (heteroscedasticity).
Table 4-10 presents the complementary regressions against sediment only. Although
there is an increase in adjusted multiple R2 for brown bullhead, the quality of the fits generally
remain weak.
Table 4-11 shows a bivariate regression on arithmetic average water concentrations and
organic-carbon normalized sediment concentrations. The bivariate approach increases the
adjusted multiple R2, relative to regression on water column concentrations alone, for all species
except white perch, with species other than goldfish and white perch having adjusted multiple R2
values greater than 70 percent. Large improvements relative to the water-only models, however,
are seen only for brown bullhead, goldfish, and largemouth bass; species that presumably have a
significant sediment-originated food chain pathway of PCB bioaccumulation.
54 MCA/TetraTech
-------
Figures 4-5 through 4-7 show observed versus predicted average concentrations from the
bivariate BAF model for brown bullhead, largemouth bass, and pumpkinseed. In each case a
strong positive, and approximately linear, correlation is evident, although there is also clearly
variability which is unexplained by the simple BAF model. Significant outliers are labeled in the
plots. For largemouth bass, the 1977 and 1978 observations from Group 2 are much higher than
predicted. This could perhaps reflect carry-over body burden from years prior to 1977 in this
relatively long-lived species. For pumpkinseed, the major under-prediction is for Group 1 in
1989. This suggests that water column exposure in 1989 may have been higher than is estimated
from sparse USGS samples below Thompson Island Dam in this year
4.3 Discussion of Bivariate BAF Results
4.3.1 Comparison to Published BAF Values
For comparison to published BAF results, Tables 4-9 and 4-11 contain estimates of a
univariate logic BAF for total PCBs in units of liters of water per kg of fish lipid. The BAF may
be obtained directly from the coefficient on water concentration (with appropriate units
correction) from the arithmetic univariate model. A BAF estimate may also be obtained from the
coefficient on water in the bivariate model, but the result may not be fully comparable to a
univariate BAF.
The calculated Iogi0 BAFs for the univariate models range from 6.21 for goldfish to 6.62
for largemouth bass on a L/kg basis. Estimates are somewhat lower for the bivariate models.
The univariate BAFs, relating lipid-normalized body burden in fish to total PCB concentrations
in water, are sometimes denoted as BAFj1 (U.S. EPA, 1994). BAFs are also frequently reported
on the basis of the freely-dissolved fraction of a chemical in the water column, BAF^d. The two
forms of the univariate BAF can be related as
(4-7)
fd
where fd is the freely dissolved fraction of the chemical. Under average conditions in the Upper
Hudson, the freely dissolved fraction of £Tri+ is estimated, based on analysis of three-phase
partitioning in the DEIR for representative congeners, to be about 50 percent for ZTri+ PCBs.
Using Equation (4-7), base- 10 logarithms of BAFj^s would thus be equal to the calculated
BAF^s plus about 0.3 log units.
U.S. EPA (1994) summarizes estimated BAF^s for PCB congeners by trophic level
based on the food-web/fugacity model of Gobas (1993) for conditions in Lake Ontario. Results
calculated here compare favorably to results presented by U.S. EPA (1994) for BZ #28 and BZ
#31. These congeners are both included in the quantitation scheme used by NYSDEC for
Aroclor 1016, and constitute about 14 percent of the total weight of raw Aroclor 1242. For
BZ#28 and BZ#31, the Gobas model predicts a BAFifd of 6.51 for alewives. Similar to
55 MCA/TetraTech
-------
pumpkinseed, this species feeds on invertebrates that accumulate PCBs from the water column
(assumed alewife diet of 60 percent zooplankton and 40 percent Dlporeia spp.) The Gobas
model estimate compares well to the estimate of 6.23 to 6.27 + 0.3 presented here for
pumpkinseed BAFj^. The Gobas model prediction for BZ#28 and BZ#31 in piscivorous fish is
6.68, which compares well with the Hudson River largemouth bass estimate of BAFjfd of 6.47 to
6.62 + 0.3.
4.3.2 Fit of Bivariate Models to Observations
A bivariate BAF approach, including both water and sediment as independent variables,
generally improves on the ability of a simple univariate BAF approach to fit observations of fish
body burdens of ZTri+ PCBs. While the overall model fit is reasonable, the bivariate model does
not accurately predict a number of the individual data points. Performance of the model can best
be visualized by examining long runs of data at specific locations. The most extensive fish time-
series data are for brown bullhead, pumpkinseed, and largemouth bass in Group 2 (River Miles
168-176), and for pumpkinseed and brown bullhead in Group 4 (River Miles 142-152).
Observations and model predictions for these series are shown in Figures 4-8 through 4-10. In
examining these figures, it should be recalled that individual observations have been weighted by
the inverse of their standard error. Thus, some apparent outliers represent small sample sizes
with high uncertainty.
For brown bullhead (Figure 4-8), the model does a reasonable job of capturing trends in
concentration in Group 2 (although underestimating a number of observations), while in Group 4
the model provides a closer fit to most observations. The model underpredicts concentrations in
brown bullhead in Group 4 from 1993 on, perhaps reflecting an error in sediment concentrations,
which are based on high resolution core data through 1992, but estimated thereafter.
For pumpkinseed (Figure 4-9), model fit is quite close in Group 4, with the exception of a
few early years. This species is less sensitive to sediment concentrations than brown bullhead (as
described in Chapter 4.3.3), and predictions are apparently unaffected by estimated sediment
concentrations after 1992. In Group 2, the general trend in PCB body burden is captured, but
some individual observations lie well off the regression line. For instance, high body burdens in
1989 and 1992 are not captured by the model. This is a period in which the upstream Bakers
Falls source was active, and exposure concentrations may have been higher than captured in
limited water column monitoring.
Finally, for largemouth bass (Figure 4-10), the model does an adequate job of capturing
trends over time, except that average body burdens in small samples in the earliest years are
under-estimated.
Observations for all three species are also available since the mid-1980's in Thompson
Island Pool. Within the TIP, water and sediment concentrations are better characterized by
frequent sampling than downstream; however, proximity to the upstream and TIP sediment
sources also likely increases intra-year and spatial variability of exposure concentrations.
Examination of model performance against TIP samples is thus a good indicator of model
56 ' MCA/TetraTech
-------
robustness. Results for the three species are compared in Figure 4-11. From this figure it will be
noted that (1) the bivariate BAF model represents the general trend in concentration in each
species, and (2) the model does a good job of replicating the relative difference in lipid-based
concentrations between species. For pumpkinseed, the fit is close except for the 1989
observation noted previously. Brown bullhead and largemouth bass were not sampled in 1989,
but are under-estimated by the model in 1990, suggesting that the available downstream USGS
data may under-estimate water column exposure in the TIP in this period. For brown bullhead,
observations start higher and end lower than the model predictions. One potential cause could be
HUDTOX misrepresentation of the rate of decline of surface sediment concentrations. This
would also impact the largemouth bass predictions, as both species exhibit substantial correlation
between body burden and sediment concentrations. Largemouth bass display the greatest
discrepancies between predictions and observations. In part, this may reflect the fact that adult
largemouth bass concentrations are likely to integrate over several years of exposure.
In sum, variability in observations that is unexplained by the bivariate BAF may have a
number of sources. These can generally be divided into data uncertainty and model uncertainty.
Data uncertainty addresses the fact that exposure concentrations in water and sediment are not
precisely known. Water column concentrations are in many cases estimated from only a few
samples, and the estimates have considerable uncertainty relative to actual summer average
concentrations. Sediment concentrations are derived from the output of the HUDTOX model,
which has been calibrated to sediment observations at a limited number of points in time. As
with water, sediment concentration estimates may misrepresent actual exposure concentrations in
a given year. Data uncertainty has two effects: it may cause individual observations to be mis-
estimated, and it may bias the regression coefficients. Use of the full data set, including
observations over 21 years at multiple sample locations, provides a robust model that should
minimize biases in the regression coefficients. The major source of unrepresented variability is
likely to be uncertainty in the estimates of water column exposure concentrations.
The second component of unexplained variation, model uncertainty, reflects the fact that
the simple bivariate BAF model does not provide a complete representation of the factors
controlling PCB bioaccumulation in fish. Most notably, the BAF model does not take into
account age, weight, size-related foraging strategies, and sex of individuals, all of which may be
important to PCB bioaccumulation and could result in systematic differences between individual
samples. The simple BAF approach also does not take into account the differences in PCB
congener patterns present in water, sediment, and biota, or differences in congener patterns
among locations. Unlike data uncertainty, model uncertainty can be addressed through use of
more sophisticated models, such as those presented in Chapters 5 and 6.
4.3.3 Relative Importance of Sediment and Water Pathways
As discussed in Chapter 3, PCBs may enter the food chain from environmental
concentrations in either water or sediment. The relative importance of these two environmental
sources will depend on food preferences and behavior of a given species, among other factors.
The bivariate model gives a qualitative indication of the importance of water versus sediment
that is useful in developing more complex bioaccumulation models. The two sources cannot be
57 MCA/TetraTech
-------
fully separated by statistical analysis, however, as water and sediment concentrations are
correlated, as are coefficient estimates in the bivariate model.
Three methods can be used to make statements about the apparent relative importance of
the independent variables in a multiple regression model: partial correlation coefficients,
normalized beta coefficients, and elasticities (Pindyck and Rubinfeld, 1981). Note that the
measures of relative importance are not direct measures of whether PCBs in fish derive
ultimately from water or sediment-mediated pathways, as exposure concentrations at the
sediment-water interface will tend toward equilibrium between the two media. Instead, these
measures will tend to distinguish the relative importance of contributions from near-surface
sediments that are not in direct contact with the water column.
A partial correlation coefficient is a measure of the correlation of one independent
variable with the dependent variable when other independent variables are held constant. The
square of the partial correlation coefficient may be interpreted as the percentage of variance in
the dependent variable which is accounted for by the part of the independent variable in question
which is uncorrelated with the other independent variable(s) (Pindyck and Rubinfeld, 1981).
Normalized beta coefficients are the coefficients obtained from a linear regression in
which each variable is normalized by subtracting its mean and dividing by its standard deviation.
For two independent variables, Xi and X2, the normalized regression model has the following
form:
(4-8)
where the s values indicate standard deviations and an overbar indicates the mean value. The
normalization corrects for scale differences among the independent and dependent variables. A
normalized beta coefficient of 0.7 can be interpreted to mean that a 1 standard deviation change
in the independent variable will lead to an. 0.7 standard deviation change in the dependent
variable.
Elasticities interpret the effect of a percentage change in the independent variable on the
dependent variable, and also represent a normalization of the regression. The elasticity for a
coefficient; is calculated at the point of the means of each of the independent variables as
(4-9)
Estimated percent contributions, normalized beta coefficients and elasticities for the
bivariate arithmetic model are given in Table 4-12. For pumpkinseed, which forage primarily in
the water column, and for white and yellow perch, water column concentrations appear to be the
most important variable in determining body burden of ZTri+ PCBs. In contrast, brown
bullhead, resident fish which forage on the bottom, are more sensitive to sediment
58 MCA/TetraTech
-------
concentrations. At the highest trophic level, lipid-based concentrations in largemouth bass,
which are primarily piscivorous, are correlated with about equal strength to water and sediment
exposure fields.
4.4 Summary
A bivariate BAF analysis, relating lipid-based ZTri+ PCB concentrations in fish to PCB
concentrations in both the water column and sediment, provides good explanatory power in
predicting annual mean body burden in six fish species throughout the Upper Hudson River,
based on analysis of NYSDEC monitoring data for 1975 through 1997. Water-column and
sediment PCB concentrations are clearly not in complete equilibrium in most of the Upper
Hudson, and inclusion of sediment concentration as an independent variable results in a
significant increase in explanatory power for most species.
The increase in explanatory power provided by the bivariate approach is greatest for those
species that have a larger sediment-derived component of food-chain pathways. PCBs in brown
bullhead appear to be most strongly determined by sediment concentrations, while PCBs in
pumpkinseed and white and yellow perch are more strongly related to water column
concentrations. Largemouth bass tissue concentrations are correlated with both sediment and
water exposure concentrations.
The BAF analysis summarizes the historic data on PCB concentrations in fish, water, and
sediment. It is not intended to be a quantitative tool for prediction of future fish body burdens, as
the coefficients which have been derived are potentially biased by uncertainty in exposure
concentration data, and the simple BAF representation makes no attempt to account for causal
relationships between exposure and body burden. While the BAF approach appears adequate to
estimate annual average concentrations, it does not represent individual and within-year
variability expected to result from age and variations in foraging with size, nor seasonal patterns
related to temperature and the spawning cycle. The bivariate BAF analysis does, however,
provide an important foundation for more sophisticated analyses, as presented in Chapters 5 and
6 of this report.
59 MCA/TetraTech
-------
THIS PAGE INTENTIONALLY LEFT BLANK
60 MCA/TetraTech
-------
Chapter 5
-------
5. CALIBRATION OF PROBABILISTIC BIOACCUMULATION FOOD
CHAIN MODEL
The components of the food chain model and general model structure are described in
Chapter 3.5. The model takes as exposure concentrations the summer-averaged whole ZTri+
water concentration for PCBs and the annual average sediment concentration for PCBs
normalized to fraction organic carbon. As discussed in Chapter 3.5, these exposure
concentrations are converted to body burdens of PCBs through a number of bioaccumulation
factors (BAFs) that link media and food chain components. These BAF values and the
uncertainty or variability around them are derived from the available data for the Hudson and
from data for other systems. The derivation of the BAFs is presented in the Preliminary Model
Calibration Report (1996). Analyses presented here are based on Release 4.1b of the
TAMS/Gradient database. The original NYSDEC data, contained in the TAMS/Gradient
database, have been corrected to a consistent ZTri+ basis using the relationships described in
Chapter 4.1.1.9.
Each compartment in the model is briefly described. The relationship between each of
the compartments is described by a distribution of accumulation factors for total PCBs expressed
as £Tri+ based on field data. These BAFs relate the body burden of one compartment to the
expected dietary exposure of that compartment. The dietary exposure is assumed to implicitly
incorporate actual exposures from all sources (i.e., direct water uptake). Distributions presented
in the Preliminary Model Calibration Report (USEPA, 1996) report are derived for the
calibration congeners (BZ#4, BZ#28, BZ#52, BZ#101+90, and BZ#138), Aroclors 1016 and
1254, and for ZTri+ PCBs to describe the range of expected bioaccumulation factors between
two compartments.
5.1 Overview of Data Used to Derive BAFs
5.1.1 Benthic Invertebrates
The EPA team collected 20 (including background) colocated benthic invertebrate and
sediment samples during the Phase 2 field collection program. Five sediment samples and three
to five benthic invertebrate samples were taken at each location. Benthic invertebrates were
identified to the taxonomic group level for PCB analyses. PCB results were provided for
individual congeners, homologue sums, total PCBs, and Aroclor equivalents. In addition,
percent lipid data are also provided. These data were used to characterize the relationship
between sediment PCB concentrations and resulting benthic invertebrate body burdens.
5.1.2 Water Column Invertebrates
Phase 2 activities did not include data collection related to water column invertebrates.
The data on water column invertebrates is obtained from the NYSDOH studies done as part of
the Hudson River PCB Reclamation Demonstration Project (Simpson et ah, 1986). NYSDOH
conducted long-and short-term biomonitoring studies from 1976 to 1985 using caddisfly larvae,
multiplate samples, and chironomid larvae. NYSDOH placed artificial substrate samplers
(multiplates) along 17 sites for five weeks in the Hudson river from Hudson Falls to Nyack, New
61 MCA/TetraTech
-------
York (Novak et al., 1988). Samplers remained in place for five weeks during July through
September collecting a composite of sediment, algae, plankton and various macroinvertebrates.
After collection, the samplers were analyzed for Aroclors 1016 and 1254. Total PCB values are
obtained by summing the individual values for Aroclors 1016 and 1254. Percent lipid values are
also provided. These data, combined with information from the Phase 2 dataset, provide an
indication of the relationship between water column invertebrates and water column sources.
The short-term biomonitoring study conducted by NYSDOH involved the chironomid
larvae, Chironomus tentans. Twenty-five laboratory-raised chironomid larvae in nylon mesh
packets were placed, in groups of ten, in steel mesh baskets at four Hudson River locations (one
at Bakers Falls, two at Thompson Island Pool, and one at Fish Creek). One set of packets was
exposed to the sediment at a collection site on the eastern shore of Thompson Island Pool. The
remainder were placed in the water column. These short-term data are available for selected
congeners and provide some information related to the time-frame and magnitude of the short-
term relationship between water column invertebrates and water column sources.
5.1.3 Fish
The EPA team collected fish data from the same 20 benthic invertebrate and sediment
locations. Between three to five of the selected fish species were collected at each location (i.e.,
not all species were collected from all locations. For further detail, refer to the TAMS/Gradient
SAP/QAPP, 1992). Data are provided for individual congeners, homologue sums, total PCBs,
and Aroclor equivalents. Percent lipid, length and weights of individual fish as well as
composited samples are also provided.
NYSDEC has been collecting fish data for over 30 species in the Upper Hudson since
1975. From 1975 to 1988, fish data were collected every year. In 1988, fish sampling frequency
changed from yearly to every other year. The bulk of the sampling (75 percent) has been
conducted for largemouth bass, brown bullhead, and pumpkinseed.
For the NYSDEC samples, chemical analyses for Aroclors 1016, 1254 and in some years,
1221 and 1242, are provided in the database as well as weight, length, percent lipid, and, for
some years, sex and age. Generally, 30 fish were collected for each species at several locations.
5.1.4 Literature Values
There are studies from the literature which provide additional information on the
relationship between sediment, benthic invertebrates, water and water column invertebrates.
(e.g. Whittle et al., 1983; Bierman, 1990; Bierman, 1994; Wood et al, 1987; Larsson, 1984;
Lake et al., 1990; Oliver, 1987; Oliver & Niimi, 1988; Thomann, 1981; van der Cost et al., 1988;
Thomann, 1989; Thomann & Connolly, 1984; Bush et a/., 1994; Thomann et al., 1992; Harkey et
al., 1994; Endicott et al., 1994; and others). These studies are primarily useful for comparative
purposes, as they refer to systems that may experience conditions unlike those in the Hudson
River.
62 MCA/TetraTech
-------
5.2 Benthic InvertebraterSediment Accumulation Factors (BSAF)
Distributions of BSAFs between sediment concentrations and benthic invertebrate
concentrations were derived by:
1. Evaluating the sediment data to determine which river miles display significant
heterogeneity and variability in concentrations;
2. Calculating the BSAF by dividing a measured individual benthic invertebrate
concentration by the mean sediment concentration at a sampling location; and,
3. Using the final distribution representative of the relationship between benthic
invertebrates and sediment within the overall model to predict the historical fish data in a
validation exercise.
5.2.1 Sediment Concentrations
An assessment of the range of sediment concentrations by river mile and congener
provides information on the variability inherent in these data. Figure 5-1 shows mean TOC-
normalized sediment concentrations (|ig/g) and associated 95% confidence intervals for the upper
and lower portions of the Hudson River. This figure shows that sediment concentrations, even
normalized, show significant spatial variability.
5.2.2 Approach
BSAF for benthic invertebrates were calculated from the Phase 2 dataset using colocated
sediment and benthic samples. The sampling rationale will be presented as part of the ecological
risk assessment (work in progress). PCB concentration and lipid data were available for
Amphipods, Bivalves, Chironomid, Gastropods, Isopods, Odonata, Oligochaetes, Unsorted Total
(everything in a sample), Sorted Total (unidentified remaining after sorting), and Epibenthic
species.
The ideal data pairs to calculate BSAF are individually collected samples of sediment and
benthic invertebrates. In the absence of this ideal condition, we used individual benthic
invertebrate samples and mean sediment concentrations for a given co-located sampling location.
However, in the areas that display highly variable PCB concentrations in sediments, it may be
that the mean does not adequately represent the exposure level for benthic invertebrates. The
heterogeneity in sediment concentrations over small spatial scales contributes to higher
variability in the BSAF calculated from data collected in these areas. Thompson Island Pool is
an area in which such variability in calculated BSAF occurs. Matching individual invertebrate
concentrations to the mean sediment exposure in this area results in more variable ratios. Also,
the ratios for Thompson Island Pool are higher in magnitude than for the upper river generally
and significantly higher than the lower river.
Species identified as epibenthic showed BSAF that were not significantly different from
species identified as benthic based on t-tests. In addition, the sampling program did not
specifically sample for epibenthic species and were only identified as such as a function of
63 MCAATetraTech
-------
sampling rather than species identification. The BSAF calculated for each river mile were
combined to represent the range of accumulation factors in the river generally. The implications
for the food chain model are that this distribution of BSAF represent the range among the prey
species of fish feeding off the bottom. This is a reasonable approximation if the fish feed on
benthic invertebrates indiscriminately such that the probability of preying on a particular species
is proportional to that species' abundance.
For those sampling locations at which there were enough data to run normality tests, it
was determined that the benthic invertebrate data follow a lognormal distribution. This was
verified by log-transforming benthic invertebrate PCB concentrations and running standard
normality tests. The final BSAF distribution is characterized by a geometric mean and geometric
standard deviation. The variability in the sediment and benthic invertebrate concentrations has a
significant impact on calculated BSAF, because widely divergent individual benthic invertebrate
concentrations are normalized to one sediment concentration considered to be indicative of
exposures.
The BSAF by river mile charts were developed using the data for the combined benthic
species as reported in database release 4.1 and shown in Figure 5-2. The charts for BSAF by river
mile and the BSAF by species show the mean BSAF and the associated 95% confidence interval.
These plots provide information on the variability of BSAF by river mile, and the species that
contribute most to the observed variability. Those species showing the highest variability also
have the lowest number of samples, indicating the sensitivity of statistical analyses to artifacts of
undersampling.
5.2.3 Calculations of BSAF Values for Benthic Invertebrates
Figure 5-2 shows the BSAF for ZTri+ PCBs (all species combined) by river mile.
Typically, the calculated BSAF values are around one, with the exception of river mile 189,
which is at approximately 3. Error bars for river mile 100 are very wide, with an upper bound
comparable to the error bar for river mile 189.
Figure 5-2 also shows the BSAF ZTri+ PCBs (all river miles combined) by species. The
BSAF for chironomids, about 2, is higher and has wider error bars than the other river miles.
However, this is based on only three samples. The BSAF for sorted and unsorted totals, which
represent the diversity of species found at any given location, show a mean of approximately one
with narrow error bars. Odonata and bivalves show the lowest BSAF.
Differences in BSAF values by location and/or species may be attributable to:
True sediment exposure concentrations may be higher or lower than those estimated (the
BSAF procedure involves dividing an individual measured invertebrate concentration by an
average sediment concentration from the same sampling location. For the highly variable
sediment concentrations, there are both high and low individual sediment values in the
average. Thus, it may be that the true sediment concentration corresponding to the individual
measured invertebrate concentration is higher or lower than the average.)
64 MCA/TetraTech
-------
Exposure for certain species may be derived from water column sources, particularly for
those invertebrates which are surface scramblers and more like invertebrates that might be
found on the vegetation.
The model was run by applying the distribution derived above to each mean sediment
concentration by river mile. The 10th, 25th, 50th, 75th, 90th percentiles and maximum were
calculated. These percentiles were compared to the output from the frequency analysis on the
benthic invertebrate data using the SPSS software package. After log-transforming the results,
the observed benthic invertebrate concentrations were plotted against the percentiles predicted
from the model. The results of this exercise were presented in the Preliminary Model Calibration
Report (EPA, 1996). Figure 5-3 presents the cumulative distribution for BSAF estimated for
ZTri+ PCBs.
The modeled XTri+ PCB distributions in benthic invertebrates compared favorably to the
observed distributions of £Tri+ PCB concentrations as presented in the PMCR (EPA, 1996).
The BSAF model for benthic invertebrates captures the observed variability in the underlying
data. In areas where the sediment concentrations display heterogeneity (such as Thompson
Island Pool), the model accurately captures maximum observed concentrations.
5.3 Water Column Invertebrate:Water Accumulation Factors (BAFs)
5.3.1 Approach
Water column invertebrates are defined as those that receive most of their exposure to
PCBs via the water column. As defined, this group includes zooplankton as well as invertebrates
living on substrates such as plants or rock surfaces but are not in direct contact with the
sediments. The approach presented in the Preliminary Model Calibration Report (1996) was
based on relating body burdens in water column invertebrates (on a lipid-normalized basis) to
water concentrations (normalized to particulate organic carbon). This was done for the following
reasons:
1. It is assumed that PCBs in the particulate phase in the water column and PCBs
in the dissolved phase in the water column are in quasi steady-state over time
scales of months during the Summer as discussed in Chapter 8. Thus by
establishing relationships between invertebrates and a particular phase
(particulate organic carbon in this case), overall accumulation from the water
column will be taken into account.
2. The relationship to PCBs normalized to particulate organic carbon was
selected because, while water column invertebrates will accumulate PCBs
directly from the dissolved phase, the higher chlorinated congeners are
predominantly associated with the particulate phase which form the food base
for the invertebrates. Partition coefficients derived in the Data Evaluation and
Interpretation Report (USEPA, 1998) show that as much as 60 percent of
PCBs in the water column are associated with the particulate phase for tetra-
and higher chlorinated congeners.
65 MCA/TetraTech
-------
This report presents an alternative approach which also relates water concentrations to
observed water-column macroinvertebrate concentrations using a BAF approach, but rather than
incorporating the POC-normalized water column concentration, this approach relies on a whole
water concentration (i.e., uptake from both the dissolved and paniculate phases). This alternative
approach was explored because the historical data only measured PCBs in whole water. In the
PMCR (EPA, 1996), assumptions were made about the relationship of total suspended solids
(measured by the USGS) and total water concentrations based on observed relationships from the
Phase 2 dataset. To estimate paniculate organic carbon from a whole water concentration, it was
necessary to assume a fraction organic carbon of the total suspended sediments. The BAF
approach presented here was chosen to avoid making these assumptions.
These BAF derivations rely upon historical data from the New York State Department of
Health studies for the Hudson River PCB Reclamation Demonstration Project (Simpson et al.,
1986). NYSDOH conducted long- and short-term biomonitoring studies from 1976 to 1985
using caddisfly larvae, multiplate samples and chironomid larvae.
NYSDOH placed artificial substrate samplers (multiplates) along 17 sites for five weeks
in the Hudson river from Hudson Falls to Nyack, New York (Novak et al., 1988). Samplers
remained in place for five weeks during July through September collecting a composite of
sediment, algae, plankton and various macroinvertebrates. After collection, the samplers were
analyzed for Aroclors 1016 and 1254. Invertebrates collected on the samplers included:
Chironomidae, Oligochaetes, Trichoptera, Ephemeroptera, Amphipoda and Elirnidae.
Chironomid larvae and pupae were the most abundant invertebrate component from Fort Edward
to Saugerties. In addition, caddisfly larvae were hand-picked from rocks at five designated sites:
Hudson Falls, Fort Edward, Fort Miller, Stillwater and Waterford.
The short-term biomonitoring study conducted by NYSDOH involved the chironomid
larvae, Chironomus tentans. Twenty-five laboratory-raised chironomid larvae in nylon mesh
packets were placed, in groups of ten, in steel mesh baskets at four Hudson River locations (one
at Bakers Falls, two at Thompson Island Pool, and one at Fish Creek). One set of packets was
exposed to the sediment at a collection site on the eastern shore of the Thompson Island Pool.
The remainder were placed in the water column.
The study found that the congener pattern of PCBs in C. tentans differed substantially
from that in the water. Specifically, the whole water column concentrations were dominated by 2
or 3-dichlorinated congeners, contributing nearly 50% of the total concentration. The C. tentans
samples were characterized by a greater number of congeners, with each congener contributing a
much lesser proportion to the overall total (i.e., no single congener contributed greater than 10%
to the total body burden), and higher chlorinated congeners dominated. For the 26 congeners
evaluated, most congeners reached 90% equilibrium in under eight days. The September results
showed even higher C. tentans concentrations corresponding to lower water concentrations.
However, the September results are considered suspect in the article due to suspected analytical
error.
The chironomid species (C. tentans) were raised in the laboratory and only experienced
water-based exposures in this study. They were, however, allowed to come into contact with
66 MCA/TetraTech
-------
detrital matter and the like in the water column. C. tentans is primarily a filter feeder or surface
deposit feeder (Swindell and Applehans; 1987; Wood et al., 1987).
The NYSDOH multiplate samples represent the only Hudson River specific information
available on the potential relationships between water column invertebrates and water column
concentrations. The short-term studies address uptake of specific congeners, but cannot be used
in this analysis, as they reflect uptake responses on the order of 48-96 hours, rather than quasi-
steady state conditions.
In this approach, total water column concentrations are related to macroinvertebrates by:
13 Ar water = Mnvert'^water W~U
where,
BAFwater = The bioaccumulation factor between water column invertebrates
and paniculate bound PCB in mg/Kg / mg/L
Qnvert = mg PCB per Kg lipid in invertebrate tissue
Cwater = mg PCB per L total water
5.3.2 Calculation of BAFwater for Water Column Invertebrates
Figure 5-4 presents the results of BAF calculations for water column invertebrates.
Values shown are the mean with 95% confidence intervals. The mean log-transformed BAF is
approximately 6.1. The bottom portion of Figure 5-4 shows the cumulative distribution function
for whole water to water column invertebrates.
5.4 Forage Fish:Diet Accumulation Factors (FFBAFs)
As a group, forage fish are expected to have a diet that varies depending on the data
available for that given river mile. Individual forage fish will vary from this percentage. For
example, spottail shiners are expected to feed evenly on water column and benthic invertebrates,
while pumpkinseed favor water column food sources. An appropriate weighted mean was used
in the model depending on the specific species caught at a sampling location in order to develop
the accumulation factors. The approach used to develop FFBAF for forage fish is described
below.
5.4.1 Approach
Forage fish consume both water column and benthic invertebrates. As a result, their
dietary exposure to PCBs is represented as a weighted average of the PCB concentration in the
diet. Distributions in the FFBAF are derived from measured concentrations of PCBs in forage
fish at a river mile divided by the estimated concentrations in their diet. Measured benthic
invertebrate concentrations were used to estimate the benthic component combined with water
column invertebrate concentrations estimated from the water column BAF discussed previously.
67 MCA/TetraTech
-------
FFBAF values were derived by:
1. Evaluating the available data for forage fish <10 cm for each river mile. The dietary
concentration was estimated based on life history and foraging information (see Appendix
A).
2. Plotting concentrations to identify a) which species contribute most to data variability and
b) which river miles show the greatest uncertainty and variability in observed
concentrations.
3. Estimating the expected PCB concentrations in water column invertebrates for total PCBs
using the distribution described earlier in this chapter and combining these estimates with
measured benthic invertebrate concentrations.
4. Deriving a river-wide distribution of FFBAF by taking the ratio of a measured individual
forage fish concentration to the arithmetic mean dietary concentration. The mean diet is
represented by the weighted average of the benthic invertebrate (measured) and water
column invertebrate (estimated) compartments.
The method provides a basis for deriving FFBAF values for forage fish as a group as well
as for the selected fish species, spottail shiner and adult pumpkinseed sunfish. The Phase 2 data
were not adequate for estimating FFBAF values specifically for small pumpkinseed sunfish that
may be eaten by other fish species. Other approaches for pumpkinseed are discussed in
subsequent chapters.
5.4.2 Forage Fish Body Burdens Used to Derive FFBAF Values
Bar charts were developed to show lipid-normalized concentrations in forage fish by river
mile. Mean concentrations and 95% confidence intervals are shown for the upper and lower
Hudson River for STri+ PCBs in Figure 5-5.
In general, concentrations show far less variability in the lower river than in the upper
river. As a trend, concentrations decline relatively steadily from river mile 169.5 down to 88.9.
At river mile 58.7, a slight increase is seen. Within the upper river, concentrations are highest at
river mile 189.5. River mile 191.5 shows lower concentrations than river miles 194.1 or 189.5,
probably as a result of the specific location chosen for sampling. However, these data show that
PCB body burdens in forage fish are highly variable in the Thompson Island Pool area and areas
close to sources of PCBs. Forage fish body burdens may also reflect the sediment type of the
habitat (i.e. fine-grain sediments tend to accumulate higher levels of PCBs).
Table 5-1 shows the coefficient of variation for the forage fish from the EPA/NOAA
Phase 2 dataset sorted in order of increasing coefficient of variation for wet weight and lipid
normalized PCB results. The numbers in parentheses refer to the number of samples in each
calculation. This table shows that the wet weight coefficient of variation is attributable to
absolute differences in PCB concentration while the lipid-normalized values are attributable to
lipid content.
68 MC/VTetraTech
-------
Figure 5-5 shows that mean concentrations are similar for river miles 189.5 and 194.1, are
significantly higher at these locations than elsewhere in the river. This figure shows that forage
fish ITri+ PCB concentrations at most of the river miles ranged from just above 0 to about 300
|ig/g. River miles 189.5, 191.5, and 194.5 show significantly higher concentrations than at other
locations in the river. Concentrations are highest at 189.5, lower but still much higher than river-
wide averages at 191.5, and then increasing again at 194.1 to nearly the level at 189.5.
5.4.3 Calculation of FFBAF Values for Forage Fish
The body burden data provide important information on the expected variability in forage
fish concentrations. The data show that the greatest variability in fish concentrations exists
within the Thompson Island Pool and areas closest to the source of PCBs. This is also the area
showing greatest sediment concentration heterogeneity, and an analysis of the water column data
show that water column concentrations vary significantly depending on the time of year. Fish in
this area experience transient exposures and integrate both "hot spots" and less contaminated area
exposures.
The forage fish model was run for ZTri+ PCBs to evaluate the goodness-of-fit between
observed and modeled fish body burdens. As described in Appendix A, the expected
contribution of benthic and water column invertebrates was estimated based on the forage fish
data available for each river mile. For example, there are a number of river miles for which
forage fish concentrations are represented by spottail shiners. Data show that spottail shiners
consume relatively equal amounts of benthic and water column invertebrates. Other river miles
have a number of forage fish species represented, and accordingly a weighted mean was used to
estimate an overall feeding preference by river mile.
The model calculated 10th, 25th, 50th, 75th, and 90th percentiles and the maximum.
Percentiles were calculated from the observed forage fish body burden distribution at each river
mile using the SPSS software package. The modeled concentrations of PCBs in forage fish
follow a lognormal distribution, characterized by long right tails. After log-transforming the fish
concentration percentiles (both observed and modeled), the observed percentiles were plotted
against the model-generated percentiles. These results were presented in the Preliminary Model
Calibration Report (EPA, 1996). The lower portion of Figure 5-5 shows the cumulative
distribution function for STri+ PCB forage fishidiet accumulation factor.
5.5 Piscivorous Fish:Diet Accumulation Factors (PFBAF): Largemouth Bass
The Phase 2 dataset imposes limitations on these analyses. In the TAMS/Gradient Phase
2 dataset, there were no data available for largemouth bass of the correct size (all samples were
for largemouth bass less than 16 cm). Largemouth bass do not become piscivorous until at least
20 cm. At the small sizes of the largemouth bass in the Phase 2 dataset, the largemouth bass
display feeding patterns equivalent to a typical forage fish, such as pumpkinseed. Therefore,
analysis for largemouth bass has to rely on the data from the Phase I NYSDEC dataset. In the
absence of suitable Phase 2 data, an analysis was made relating largemouth bass lipid-normalized
69 MCA/TetraTech
-------
concentrations to pumpkinseed lipid-normalized concentrations for measurements reported as
Aroclors 1016 and 1254 (representative of ZTri+, which, in turn, is representative of total PCBs).
5.5.1 Largemouth Bass to Pumpkinseed BAF for ZTri+ PCBs
Figure 5-6 shows the ratio of largemouth bass greater than 25 cm to pumpkinseed less
than 10 cm for ZTri+ PCBs by river mile and year. The lower portion of this figure shows the
cumulative distribution function for largemouth bass to pumpkinseed ratios. The largemouth
bass samples were collected in the spring, and the pumpkinseed samples in the fall. The
following spring individual largemouth bass concentrations were divided by the arithmetic mean
pumpkinseed concentration for the previous fall.
These BAF values implicitly incorporate seasonal variation. Insofar as these ratios are
consistently constructed (that is, always a spring-caught piscivorous fish over the forage fish
average from the previous fall), their application is valid. However, these ratios may not only
capture trophic level differences, but seasonal differences as well.
5.6 Demersal Fish: Brown Bullhead:Sediment Accumulation Factors
Data are available for brown bullhead from the NYSDEC dataset for river miles 189 and
168 for intermittent years since 1977. The approach taken to develop brown bullhead:sediment
accumulation factors was to divide individual observed brown bullhead lipid-normalized ETri+
body burdens by the average TOC-normalized ZTri+ annualized sediment concentrations
predicted by the HUDTOX model for each reach. These BSAF were developed for 1977 - 1990,
and then the resulting distributions used to predict 1991 - 1996 concentrations for each reach to
validate the distributions.
Table 5-2 and Figure 5-7 provide the parameters of the final distributions developed for
brown bullhead BSAF. Distributions are presented separately for river mile 189, river mile 168,
and combined. Statistical tests (t-test assuming equal and unequal variance) were significant at
p< 0.005 between the two locations, suggesting that given the available data, the relationship
between sediment and brown bullhead concentrations is different between the two locations. The
mean accumulation factors derived for each location are below one, but higher at river mile 189
than at river mile 168. Several factors could account for this difference:
Inconsistencies or incorrect averaging of the HUDTOX sediment concentrations do
not accurately reflect true exposure concentrations to the brown bullhead; and,
The BSAF do not account for water-column based exposures (across the gill, diet,
etc.) that may be occurring.
5.7 Validation of Probabilistic Model Using Fate and Transport Model Output as Input
Table 5-2 presents the final distributions used in the empirical probabilistic model. Full
details on distribution development were presented in the Preliminary Model Calibration Report
(1996). The sediment and water concentrations used to generate pumpkinseed and largemouth
70 MCA/TetraTech
-------
bass concentrations were obtained from the hindcasting results from the fate and transport model
(see Books 1 and 2). Figure 5-8 shows the TOC-normalized sediment concentrations and whole
water summer concentrations used in the empirical probabilistic model.
The model was run for river miles 168 (Stillwater), 189 (TIP), and 155 (Waterford-
Federal Dam region). Figure 5-9 presents the results of the calibration for largemouth bass and
pumpkinseed. Wet weight results were calculated by multiplying the lipid-normalized results by
the average observed lipid for that location and species (across all years).
5.8 Discussion of Results
Table 5-3 presents the relative percent difference estimated between predicted and
observed body burdens on a lipid-normalized basis. Corresponding wet weight concentrations
are obtained by multiplying the lipid-normalized results by an appropriate value for lipid. Wet
weight results for largmeouth bass at river mile 189 show fairly good agreement with the data,
although the median predicted body burden tends to be underpredicted for recent years. The
model appears to underpredict on a lipid-normalized basis. For pumpkinseed at river mile 189,
both wet weight and lipid-normalized concentrations show roughly the same relationship to the
data. The model performs better at river mile 168. For this location, both wet weight and lipid
normalized results show good agreement with the data for largemouth bass and pumpkinseed. At
river mile 155, data were only available for the largemouth bass. At this location, the model
performs well, particularly as there is little fluctuation in the mean observed PCB content of
largemouth bass from year to year.
Figure 5-10 presents the results for brown bullhead, and Figure 5-11 presents the results
for the pumpkinseed. Brown bullhead shows good agreement between lipid normalized
predictions and observed data for river mile 189, but significantly overpredicts at river mile 168.
However, applying an average lipid percent (1.2) results in wet weight predictions that show
good agreement at both river miles 189 and 168. Pumpkinseed concentrations are fairly well
captured, although significant increases and/or decreases are not as well captured.
The predicted 95th percentile typically captures maximum observed concentrations,
suggesting that predicted 95th percentile concentrations are protective of the population at this
level.
As an empirical model, this model represents quasi-steady state conditions. To the extent
that the BAF relationships constructed between compartments represent a variety of conditions in
the river, these will be represented in the output. The model is not designed to predict short-term
fluctuations in concentrations, or short-term responses in the system.
71 MCA/TetraTech
-------
THIS PAGE INTENTIONALLY LEFT BLANK
72 MCA/TetraTech
-------
Chapter 6
-------
6. FISHRAND: TIME-VARYING MECHANISTIC MODEL BASED ON A
GOBAS APPROACH
6.1 Overview of Calibration Procedure
The calibration procedure began by estimating the elasticity of user-specified parameters
and model constants to determine the sensitivity of model results on input assumptions. The next
step was to evaluate the literature and site-specific information to obtain best estimates of central
tendency values and distributions for the input parameters in the FISHRAND model. Prior to
implementing the formal Bayesian updating calibration procedure, these empirical distributions
were refined using likelihood profiling techniques in the simplified FISHRAND model. The full
FISHRAND model was then formally calibrated starting with the prior distributions obtained
through the likelihood profiling method and applying the Bayesian updating procedure to obtain
posterior estimates of distributions. Both the sets of model results are compared to data for the
state variable ITri+ PCB in fish tissue on a wet weight and lipid normalized basis.
The model was calibrated first for Stillwater (river mile 168) and then applied to the other
two locations. For river mile 154, the calibrated model for river mile 168 was run without
further calibration or adjustment in the distributions. The calibration was refined for river mile
189 as compared to 168 since environmental parameters (e.g. TOC) differ between these two
locations.
6.2 Sensitivity Analysis to Determine Parameters for Updating in Calibration
The FISHRAND calibration procedure focused on optimizing wet weight concentrations.
This was done for a number of reasons. First, the model is designed to predict a wet weight
concentration in fish, and lipid normalized results are calculated by dividing the predicted wet
weight concentration by the percent lipid. Second, the lipid content of any given fish is difficult
to predict from first principles alone, and lipid content is a highly significant parameter in
predicting body burdens (see Chapter 8). Finally, potential target levels in fish are typically
described on a wet weight basis.
To determine the most important parameters, a sensitivity analysis was conducted using
the analytical solution to the Gobas model. The sensitivity analysis focuses on the relationship
between predicted fish body burden and the 11 constants plus environmental parameters
described in chapter 3. After obtaining the partial derivatives, elasticities were estimated.
Elasticities interpret the effect of a percentage change in the independent variable on the
dependent variable based on equation 3-22. The results of this exercise are shown in the
following table:
73 MCA/TetraTech
-------
Parameter
Q
Cj related
parameters
Q
Cw
L
TOC
Vf
Sign of
derivative
-
+
+
-
+
+-
+-
Comment
Uptake efficiency
KOW, plankton lipid concentrations
Food ingestion rate
Growth rate
Percent lipid in fish
Total organic carbon
Fish weight
The model was found to be insensitive to the other parameters. In addition to sensitivity
to parameters, correlation between variables was also evaluated in the selection of calibration
parameters. For example, KOW affects uptake efficiency, PCB partitioning at the base of the food
web, and excretion rate. Thus, rather than select all three parameters, only KOW was selected.
The final parameters selected for calibration include: TOC, KOW, growth rate coefficient, and
percent lipid in fish.
The model was calibrated first for Stillwater (river mile 168) and then applied to the other
two locations. For river mile 154, the calibrated model for river mile 168 was used without
further calibration. Since experimental data show that TOC is significantly different at river mile
189 as compared to the remainder of the river, a different distribution was used for this river
mile. Percent lipid is also different for river mile 189. Each of the model inputs is discussed
next.
6.3 Model Input Data: User Specified Parameters
Both the historical NYSDEC and EPA Phase 2 datasets were used in the development
and validation of the FISHRAND model. Distributions of species-specific fish weight, lipid
content (expressed as a percentage), organic carbon content of sediment (expressed as a
percentage), and feeding range preferences for the individual fish species were developed for use
in FISHRAND. Sediment and truly dissolved water concentrations from the 21-year hindcasting
of the fate and transport model were used to generate fish body burdens to compare to the
historically observed NYSDEC data set. Further distributions incorporated include a distribution
for KOW and for starting sediment and water concentrations as predicted by the fate and transport
models. Analyses presented here are based on Release 4.1b of the TAMS/Gradient database. The
original NYSDEC data, contained in the TAMS/Gradient database, have been corrected to a
consistent XTri+ basis using the relationships described in Chapter 4.1.1.9.
74
MCA/TetraTech
-------
There are two kinds of model parameters:
Non species-specific parameters that apply either to the location being modeled or the
form of PCB being modeled, and,
Species-specific parameters (e.g., lipid content, weight, etc.).
Table 6-1 provides the empirical distributions derived for each of the user-specified input
parameters based on site-specific data, except for sediment and water concentrations. These are
provided in Figure 6-1.
6.3.1 Non Species-Specific Parameters
A number of environmental parameters specific to either the location or form of PCBs
being modeled were described by distributions, including:
Annual sediment concentrations (location specific);
Monthly water concentrations (location specific);
Monthly temperature (location specific);
Log octanol-water partition coefficient (KoW) (£Tri+); and,
Total organic carbon in sediment (TOC).
Prior distributions for K<,w and TOC were obtained using the likelihood profiling method.
Sediment and water distributions were obtained directly from HUDTOX and were not adjusted in
any way. Temperature was obtained empirically and not adjusted.
6.3.1.1 Sediment and Water Concentrations
The sediment and water concentrations used in calibrating and validating the FISHRAND
model were generated from the fate and transport model (Books 1 and 2). Figure 6-1 presents the
dry weight sediment concentrations and dissolved water concentrations predicted by the
hindcasting calibration for £Tri+. The probabilistic empirical model uses TOC-normalized
sediment concentrations and whole water concentrations, while FISHRAND relies on freely
dissolved water concentrations and dry weight sediment concentrations (|ig PCB / g solid).
The model requires monthly dissolved water column concentrations and annual sediment
concentrations (sediment concentrations vary only slightly within a given year, allowing for the
use of an annual concentration). HUDTOX generates daily water column and sediment
concentrations for the hindcasting period and every other day for the prediction period. These
results are averaged by month for water and by year for sediment, characterized by a mean and
standard deviation (equations 3-19 and 3-20). Sediment concentrations represent an area-
75 MCA/TetraTech
-------
weighted average of cohesive and non-cohesive sediments and assume that fish preferentially
spend 75% of their time in cohesive sediment areas.
Water column concentrations were weighted toward nearshore areas in the TIP and
averaged across the river for downstream locations. Lateral gradients are of most importance in
the lower TIP and less important downstream because (1) downstream dams have generally
smaller, narrower pools plus higher flows, so lateral mixing should be better, (2) the lateral
gradient in the TIP is only strong when flows are low AND the upstream concentration at Ft.
Edward is less than 15 ng/L on a ZTri+ basis. (Water downstream of the TIP is almost always in
excess of 15 ng/L ETri+); (3) the density of hot spots and surface sediment concentrations are
generally lower downstream, thus the lateral gradient should be less; (4) lateral gradients in the
TIP are likely enhanced by shallow macrophyte beds, and there are fewer of these in the
downstream pools. Lateral gradients are enhanced by shallow macrophyte beds due to (1)
structure decreasing flow and making the flow field more heterogeneous; (2) increased sediment
trapping and deposition; and (3) enhanced and more varied biological activity.
6.3.1.2 Temperature
Growth rate is modeled as a temperature dependent relationship, thus, monthly average
temperature is required for FISHRAND. Temperature data for all upper Hudson river locations
was compiled from the General Electric and EPA datasets. Together, these datasets provided
nearly 2,200 datapoints over the course of several years. Temperature data were grouped by
month and year of collection and river mile and statistically evaluated across locations. The
mode of the distribution for any location is the same as the average value used in the HUDTOX
model for that segment.
During the summer months, when temperatures are highest and fish are consuming the
most dietary items, some fish species are likely to spend proportionally more of their time in
shallower, nearshore areas which may not have been captured in the monitoring program.
However, temperature is also required by the HUDTOX model to estimate partitioning behavior,
and to use a very different temperature from that used in HUDTOX (and shown by the data)
would result in an inconsistency between the two models. A sensitivity analysis in which the
temperature was adjusted upward by 20% for the summer months was conducted for the
FISHRAND model, and the resulting body burdens changed by less than 5%. Consequently, the
same observed temperatures as were used in HUDTOX were also used in FISHRAND.
6,3.1.3 Total Organic Carbon in Sediment
TOC is most important in the estimation of ZTri+ PCB concentrations in benthic
invertebrates (the FISHRAND equation takes the same form as equation 3-6), which are
consumed by upper trophic level fish. From a calibration perspective, it does not matter whether
benthic lipid or TOC is selected as a calibration parameter as the net effect is the same. Benthic
invertebrate data are only available for one year (1993), thus data for lipid in invertebrates are
only available for that year.
76 MCA/TetraTech
-------
TOC was selected as a calibration parameter because:
It has an analagous relationship in the sediment-based pathway as KOW does in the
water-based pathway;
The sensitivity analysis showed that KOW and TOC are dependent in the model (which
may be a reflection of indirect dependence due to model structure and data
imperfection); and,
There are more data available for TOC than benthic lipid (although note that TOC as
reflected in fish diet as compared to composition in bottom sediment may be
different).
Mathematically, TOC in the FISHRAND model is in the form of 1/TOC, thus, small
values of TOC will lead to large changes in results while 1/TOC approaches a constant value for
larger values of TOC.
6.3.1.4 Log Octanol-Water Partition Coefficient (Kow)
The KOW used in this analysis is representative of the distribution of KOWS that might be
expected in the ZTri+ PCB mixture. Several approaches for characterizing KOW were evaluated.
Individual PCB congeners contained in the STri+ mixture will be taken up by fish to varying
degrees as expressed by the KOW One approach was to evaluate an average congener profile in
water and fish in the upper Hudson and weight the KOW values according to the weighting of that
particular congener in the mixture. This approach proved infeasible, however, and another
approach was taken.
In the approach taken, KOW is described by a triangular distribution according to the
cumulative distribution of KOWS in the mixture. This distribution ranges from 5.12 to 8.3 with a
mode of 6.6. Individual KOW values were obtained from the Great Lakes Initiative Technical
Support Document for the Procedure to Determine Bioaccumulation Factors (EPA, 1994).
6.3.2 Species-Specific Data
Data from the historical NYSDEC fish monitoring results, EPA Phase 2 data and the
NYS DOH macroinvertebrate data collection effort were used to develop species-specific
distributions for:
Lipid content for fish, benthic invertebrates, water column invertebrates, and
phytoplankton;
Fish weight; and,
Dietary composition of the fish diet.
77 MCA/TetraTech
-------
These distributions represent typical values found in the population of interest based on
observed data. Using distributions for particular parameters instead of point estimates in effect
follows a population over time in which fish enter and leave the compartment in equal rates.
Triangular distributions were derived for the dietary composition for each fish species based on
the proportion of the diet represented by benthic invertebrates, water column invertebrates,
phytoplankton, and/or forage fish based on the indicator species gut contents analysis presented
in Appendix A. Table 6-1 presents a summary of the distributions used in this analysis.
6.3.2,1 Lipid Content
Lipid Content for Fish
Figure 6-2 presents the cumulative distribution functions for lipid content in each of the
fish species. Lipid data were combined across years and locations based on a series of analyses
described next. Only those lipid data were used for the fish of appropriate size (i.e., only
largemouth bass > 25 cm; pumpkinseed < 10 cm; white perch > 17 cm; yellow perch > 15 cm).
This resulted in keeping all of the historical NYSDEC largemouth bass data (no exclusions as all
fish were greater than 25 cm) and none of the USEPAEPA Phase 2 data (fish were all very
small). The Phase 2 data was also not suitable for pumpkinseed, which were all very large fish
(larger than the largemouth bass). White perch, yellow perch, and brown bullhead lipid were
obtained from the historical NYSDEC dataset and the Phase 2 USEPA. None of the data points
were excluded for brown bullhead and approximately 100 small fish were excluded for white
perch.
Individual percent lipid measurements were regressed against both weight and length for
each species and location to determine if there was a correlation between lipid content and either
weight or length which should be accounted for in the model. In a few cases, this analysis
showed a weak correlation but overall there was no relationship between lipid and weight or
length. Thus, the model assumes no correlation between the two but rather samples randomly
from the assigned lipid distribution for each species. Figure 6-3 shows the combined results of
weight-lipid relationships for each of the species, although the analysis was originally conducted
for each individual location, year, and species.
Lipid content in fish will depend on a number of factors, including temperature, prey
availability, and foraging success. Year-to-year differences in lipid content are difficult to
predict from first principles, so the ideal situation is one in which species-specific lipid
distributions can be developed irrespective of location or time. The first step in developing
species-specific lipid distributions was to statistically evaluate lipid data across years and
locations to determine if there were clear differences. Comparisons of means (using the
Bonferroni correction to account for multiple comparisons) was carried out to determine
significant differences.
There was no pattern to differences in lipid content within a species by location or year.
Typically, differences were observed across years and locations, for example, between river mile
168 in 1993 and river mile 189 in 1995. There were no observable consistent differences such
78 MCA/TetraTech
-------
as, for example, 1995 lipid content was lower at all locations, or river mile 189 was consistently
lower than 168. As there were no observable patterns to differences in lipid, and no clear basis
upon which to predict a lipid distribution for any given year, lipid data across all years and
locations were combined within a species. Figure 6-4 shows the results of the mean lipid content
for each fish species by year for each location.
All derived lipid distributions were compared to the literature (EPA, 1994 and 1995) to
determine whether they were within the range observed for these species in other systems. These
values all proved consistent.
Lipid Content for Benthic and Water Column Invertebrates
The US EPA Phase 2 data were used to develop a lipid distribution for benthic
invertebrates presented in Table 6-1. The NYS DOH dataset was used to develop a lipid
distribution for water column invertebrates from the multiplate sampling effort. These
distributions were compared to literature values. The water column invertebrate lipid
distribution was used as an updating parameter in the Bayesian procedure.
Literature values were used to construct a percent lipid distribution in phytoplankton
(Gobas, 1993). Note, however, that only the spottail shiner consumes a small amount of
phytoplankton (5% or less of the diet).
6.3.2.2 Fish Weight
Figure 6-5 presents the cumulative distribution functions for fish weight for each of the
fish species. As described previously, no observable relationships between weight and lipid
content were discovered which should be accounted for in the model structure. The same data
were used to develop both the lipid content and weight distributions.
6.3.2.3 Dietary Composition
Dietary composition is based on the results of the analysis presented in Appendix A for
each individual fish species and summarized in Table 6-1. As noted in Chapter 3, it is very
difficult to quantitatively describe feeding preferences based on snapshots of information.
Further, despite the extensive gut content analyses that have been conducted by Menzie-Cura and
Associates, Inc. and Exponent, Inc., soft-bodied organisms that may have been consumed
typically will have been digested, thus, it is virtually impossible to specifically identify all the
prey organisms in the diet of fish. The results presented in Table 6-1 represent professional
judgment and a careful analysis of all the available data.
PCB concentrations in the diet are described as a "random walk" in monthly time
intervals in which it is assumed that fish and prey meet randomly from month to month. The
concentration in the fish diet assumes that distributions are fixed on monthly intervals, but
79 MCA/TetraTech
-------
concentrations in the diet can change from month to month while still relying on the same
feeding preference distribution for each species.
6.4 Calibration Results
Tables 6-2 and 6-3 provide the empirical, prior, and posterior distributions obtained from
the calibration procedure. Following the sensitivity analysis described in chapters 3 and 6.1,
likelihood profiling methods were used to determine the best prior distributions for the Bayesian
updating procedure. The posterior distributions were obtained by applying the Bayesian Monte
Carlo updating procedure described in chapter 3.5.5.5.
Figures 6-6 through 6-9 present the results of the calibration procedure. Two sets of
results are presented: the first set of results rely on the "generic" model constants as described in
the literature by Gobas (1993 and 1995) together with the "prior" site-specific and species-
specific distributions as described previously. These are the results of the FISHRAND model
prior to updating any of the distributions. The next set of results incorporates a formal
calibration procedure in which the prior distributions are updated based on a comparison of the
model output to observed data.
As described previously, the calibration procedure emphasizes a close fit between
predicted and observed body burdens on a wet weight basis, sometimes at the expense of lipid-
normalized results. Since the model is very sensitive to lipid concentrations, it is possible to
obtain nearly perfect agreement between predicted and observed data by incorporating observed
lipid concentrations. Direct incorporation of these observed temporal changes in lipid
concentrations is not useful for forecast purposes (there is no basis upon which to predict future
lipids), and the approach taken here was to describe lipid content as an empirical distribution
based on the available data (as described above).
Using the predicted hindcasting for sediment and water from the fate and transport
models, Figure 6-6 shows the results of the comparison between the initial model runs prior to
updating and the updated model runs for largemouth bass, Figure 6-7 for brown bullhead, Figure
6-8 for yellow and white perch, and Figure 6-9 for pumpkinseed. The calibration procedure
focused on the subset of parameters that most influence predicted fish concentrations. The
sensitivity analysis described previously was used to determine which model constants have the
most influence on predicted body burdens. This figure shows a comparison of the predicted 50lh
percentile (median) as compared to the median from the data. The bars represent the 95%
confidence interval on the median from the NYSDEC data.
The model predicts a monthly fish body burden, which can be further averaged to
represent a seasonal or annual concentration. The results shown in Figures 6-6 through 6-9 are
results obtained for the same month during which the samples were taken (e.g., typically May -
June samples). Table 6-4 presents the relative percent difference between predicted and observed
using the monthly results. Slightly different results are obtained when comparing the annualized
output with observed concentrations. Observed concentrations are more likely to represent a
seasonal concentration rather than an annualized concentration as demonstrated by limited same
80 MCA/TetraTech
-------
year seasonal data available for white perch and yellow perch from below Federal Dam collected
during 1995 by NOAA.
As mentioned previously, the calibration focused on optimizing results on a wet weight
basis. The updating procedure significantly improved wet weight fits while often changing lipid
normalized results only slightly or not at all. The most significant differences occurred for
largemouth bass at 189 and 168. Figure 6-6 shows that prior to updating, lipid normalized
concentrations were very close, but wet weight concentrations showed a positive high bias. This
high bias was eliminated through the updating procedure. Wet weight concentrations typically
fall within the error bars of the data following updating, and lipid normalized concentrations
show roughly the same relationship to the data after updating as prior to updating.
Table 6-4 provides a summary of the relative percent difference between modeled and
observed. The values in this table were calculated on a median basis by taking the observed
concentration minus the predicted concentration and dividing by the observed concentration. On
a wet weight basis for river mile 189, largemouth bass results (first page) show that the highest
difference is 100% in 1991 and the next highest relative percent difference is 48% in 1985.
Typically, the model predicts within 16% or less, and Figure 6-6 shows that the predicted model
results are within the error bars for the observed median. In general, the model captures the
trends in the data, decreasing in 1991 although not as much as the data suggest. However, in
absolute concentrations, the difference between predicted and observed in 1991 is approximately
1 ppm.
For brown bullhead at river mile 189, the model shows excellent agreement on both a wet
weight and lipid normalized basis for all years. Predicted brown bullhead body burdens follow
the trend in the data, and are within the error bars of the median for all years except 1991.
Relative percent differences (shown in Table 6-4) are within 12% or less for seven of the nine
years for which data are available, and within 38% - 41% for the remaining two years.
Data are available for yellow perch at river mile 189 for three years as shown in Figure 6-
8. On a wet weight basis, the model predicts within the error bars of the median for all three
years, and overpredicts the median by 1% - 32% as shown in Table 6-4.
Predicted pumpkinseed concentrations at river mile 189 follow the trend in the data for
both wet weight and lipid normalized results as shown in Figure 6-9. Predicted concentrations
are within the error bars on the median for all but two years, and fall within 22% or less for five
of seven years for which data are available, and within 53 - 60% for the remaining two years.
For river mile 168, again largemouth bass concentrations typically capture the trend but
overpredict from 1989 - 1991 and underpredict slightly for 1992. This may reflect an inaccurate
representation of the true exposure concentrations, or changes in the food web structure during
those years (i.e., largemouth bass diet shifted significantly from the specified distributions. In
general, however, the model predicts wet weight body burdens at river mile 168 that are within
the error bars, except for a few years.
81 MCA/TetraTech
-------
Observed brown bullhead concentrations at river mile 168 are much more variable than at
river mile 189. At river mile 168, brown bullhead concentrations do not appear to follow
predicted sediment concentrations as smoothly as at river mile 189. The model generally
captures trends at this location, but overpredicts during the late 1970's, and underpredicts in 1990
and 1992. Of the sixteen years for which comparisons are available, the relative percent
difference between predicted and observed is less than 12% for ten of these years, within 30 -
35% for five of the years, and within 56% for one year (1977).
Results for yellow perch at river mile 168 are shown in Figure 6-8. The model predicts
within the error bars on the median on a wet weight basis for all years except 1980 and 1992.
The absolute difference in concentration is within 2 ppm for 1992. Relative percent differences
shown in Table 6-4 are within 60% for all years.
Pumpkinseed concentrations follow the trend in the data for river mile 168 as shown in
Figure 6-9. Error bounds on the observed medians are very tight for this location. On an
absolute basis, predicted pumpkinseed median concentrations fall within less than 1 ppm of
observed medians, and within 35% or less expressed as a relative percent difference, shown in
Table 6-4.
The calibrated model for river mile 168 was run without any further updating for river
mile 155 in a quasi-validation exercise. Figure 6-6 presents the results for largemouth bass. On
a wet weight basis, of the eight years, the model predicts within 10% for four years, within 50%
for three years, and within 100% for one year. These values (except for 1991 - 100%) are within
the error bars of the median and in absolute concentrations within 1 ppm of the observed median.
There is only one year of data available for yellow perch, and for this one year the predicted
median was within 38% of the observed median, and within the error bars on the median.
Typically, calibration results are within a factor of two or less of the median and fall
within the error bars of the median.
Figures 6-10 through 6-12 show quantile-quantile plots for river miles 189, 168, and 155,
respectively. These plots provide a measure of the goodness of fit of the variability of predicted
fish body burdens as compared to the observed variability in fish body burdens.
6.5 Model Validation: Calibration Using Partial Dataset
To validate the model, several approaches were followed. First, the calibrated model for
river mile 168 was run for river mile 155 and predicted body burdens compared to measured
body burdens at this location. Figure 6-6 presents the results for largemouth bass. On a wet
weight basis, of the eight years, the model predicts within 10% for four years, within 50% for
three years, and within 100% for one year. These values (except for 1991 - 100%) are within the
error bars of the median and in absolute concentrations within 1 ppm of the observed median.
There is only one year of data available for yellow perch, and for this one year the predicted
median was within 38% of the observed median, and within the error bars on the median.
82 MCA/TetraTech
-------
A second approach involved recalibrating the model using only pre-1990 data, then
running the model for 1991 - 2067 and comparing the results. A number of comparisons were
evaluated:
Comparison of previously obtained posterior distributions with posterior distributions
obtained using pre-1990 data only;
Comparison of predicted versus observed body burdens for 1991 - 1996 (these data
were not used in the pre-1990 calibration); and,
Comparison of predicted results for 1998 - 2067 from full calibration to pre-1990
only calibration results.
The results showed that the posterior distributions obtained from the pre-1990 only
calibration are close to the results obtained from the full calibration. Most importantly, the
relative proportion of change between KOW and TOC remained the same although the absolute
values changed somewhat. These results are shown in Table 6-5.
The relative percent differences using pre-1990 only data are within 30% of the values
obtained using the full dataset. Forecast results are also similar. Table 6-6 shows the ppm wet
weight difference between the annualized forecast results obtained for largemouth bass, brown
bullhead, and yellow perch using the posterior distributions from the pre-1990 only data as
compared to the full dataset. Largemouth bass concentrations are 0.2 ppm higher in the long term
than predicted using the full dataset, while brown bullhead and yellow perch show a difference of
less than 0.08 ppm wet weight.
6.6 Relative Contribution of Sediment and Water Pathways
The relative contribution of the different pathways were evaluated several different ways.
Using the results from FISHRAND directly, the following contribution of direct water uptake
across the gill versus diet was determined:
Species
Spottail shiner
Pumpkinseed
Yellow Perch
White Perch
Brown Bullhead
Largemouth Bass
<- River Mile ->
168-154
189
Direct Water Uptake
15%
13%
6%
3%
2%
4%
15%
12%
6%
4%
5%
4%
168-154
189
Diet
85%
87%
94%
97%
98%
96%
85%
88%
94%
96%
95%
96%
The second approach was to run the model in steady-state mode to obtain average
estimates of wet weight fish body burdens and regress the predicted fish concentrations against
sediment (dry weight ppm) and whole water (ng/L) concentrations. Although the FISHRAND
83
MC/VTetraTech
-------
model is nonlinear in specific parameters, the best fit between sediment, water, and fish was
linear. From these results, it is possible to obtain percent contribution of sediment and water to
the overall variance, normalized beta coefficients, and elasticities. These results are presented in
Table 6-7. These results can be compared to the results from the bivariate statistical model,
although note that the bivariate model regresses lipid-normalized fish body burdens against
whole water and TOC normalized sediment concentrations, which is not directly comparable to
the FISHRAND approach, which regresses wet weight fish body burdens against dry weight
sediment and dissolved water concentrations.
This table shows that predicted fish body burdens are more sensitive to changes in
sediment than they are to changes in the dissolved water concentrations, given the assumptions
inherent in the regression. These results should be interpreted as indicative of the relative role of
sediment versus water rather than a strictly quantitative absolute relationship. The FISHRAND
model is designed to provide information on the ultimate origin of PCBs (water or sediment) as it
is a food web model - although this is to some extent predefined by model assumptions. The
Bivariate BAF model cannot do this: instead, the Bivariate BAF model assesses the correlation of
fish concentration with the part of the sediment time series that is not correlated with water
concentrations. This might be taken to reference deeper (i.e., non-interface) sediment pathways.
However, the Bivariate BAF model combines observed water with modeled sediment
concentrations. This means that the water component also has attributed to it all the parts of the
exposure time series which were not captured by HUDTOX (but are reflected in the observed
summer water concentrations).
84 MCA/TetraTech
-------
Chapter 7
-------
7. BIOACCUMULATION MODEL FORECASTS
This chapter describes the initial modeling results from the FISHRAND model.
Sediment and water concentration inputs are taken from the fate and transport model (Books 1
and 2). Three modeling forecasts were provided from the HUDTOX model: a zero upstream
boundary condition (cessation of the source at Ft. Edward), a 10 ng/L constant upstream
boundary condition (assuming a small but constant upstream source at 10 ng/L), and a 30 ng/L
constant upstream boundary condition (assuming a larger but still constant upstream source at 30
ng/L).
The FISHRAND model requires freely dissolved water concentrations averaged monthly
and annual average sediment concentrations as inputs. The model mechanistically describes PCB
uptake over time and results are presented here for largemouth bass, yellow perch, pumpkinseed,
brown bullhead and white perch under the three scenarios. All uptake parameters are described
by distributions which reflect the variability in fish responses to changes in sediment and water
concentrations. The sediment and water concentrations themselves are also described as
distributions from the daily HUDTOX output.
It is difficult to quantify the uncertainty in every single model parameter. Because lipid
content, fish weight, and other important variables in the model reflect population heterogeneity
more than they reflect uncertainty, these were described as variable. Approximate uncertainty is
estimated by applying the maximum under and overpredictions from the relative percent
differences presented in Table 6-4 from the hindcast calibration.
7.1 Sediment and Water Concentration Inputs
Figure 7-1 shows the sediment and water concentrations used for the zero upstream
boundary condition; Figure 7-2 presents the exposure sediment and water concentrations
predicted from the fate and transport model under the 10 ng/L upstream boundary condition; and
Figure 7-3 provides the predicted exposure concentrations under the 30 ng/L upstream boundary
condition. These figures show that sediment concentrations decline exponentially between 1998
and 2067 under all scenarios.
7.2 Predicted PCB Concentrations in Fish under Zero Upstream Boundary Condition
Figure 7-4 presents the results of predicted fish body burdens on a wet weight basis for
largemouth bass predicted median concentrations under the three upstream boundary conditions.
Figure 7-5 shows the same results for brown bullhead, and Figure 7-6 shows these results for
yellow perch and white perch. For all species, concentrations decline roughly exponentially and
approach different asymptotes depending on the species and upstream boundary condition. The
median and 95th percentile asymptotes approached by each species are found in Table 7-1. The
values in parentheses are approximate uncertainty bounds on the predicted values based on the
maximum difference between predicted and observed from the hindcast calibration. Table 7-2
provides a comparison of example target levels (note that target levels will be determined during
the feasibility study. The values shown in Table 7-2 are to provide a benchmark for when order-
85 MCA/TetraTech
-------
of-magnitude concentrations will be achieved) based on the 10 ng/L upstream boundary
condition.
Figure 7-4 shows that concentrations in largemouth bass at river mile 189 decline roughly
exponentially. The lowest achieved concentration is approximately 0.05 ppm wet weight, with a
margin of error of approximately 0.03 ppm on either side on a median basis. This is interpreted
as 50% of fish are expected to show concentrations above this value and below this value.
Figure 7-7 shows the results for the 25th, 50th and 95th percentiles for each of the locations. The
95th percentile concentration is interpreted as the expected body burden for 95% of the
population. That is, 95% of the population would be expected to experience the shown
concentration or less. For 95% of the population at river mile 189, the lowest concentration
achieved is roughly 0.1 on an annualized basis. At river mile 168, largemouth bass
concentrations decline to approximately 0.005 to 0.06 ppm on a median basis (best estimate of
0.02), and at river mile 154, concentrations are predicted to decline to approximately 0.007 to
0.02 ppm.
Brown bullhead concentrations at river mile 189 are predicted to be somewhat higher
than predicted largemouth bass concentrations. By the end of the forecast period, the forecast
median is approximately 0.1 ppm, and the 95th percentile is predicted to fall at approximately 0.1
to 0.24 ppm. For river mile 168, predicted brown bullhead median concentrations achieve 0.01
to 0.04 ppm, and 0.005 to 0.02 ppm at river mile 154. The corresponding 95th percentile values
are 0.015 to 0.06 ppm and 0.01 to 0.04 ppm, for river miles 168 and 154, respectively. The 25th,
50th, and 95th percentiles predicted under the zero upstream boundary condition are presented in
Figure 7-8.
Yellow perch concentrations at river mile 189 are predicted to fall to 0.03 to 0.06 ppm on
a median basis, and to 0.05 to 0.11 ppm on a 95th percentile basis. Predicted concentrations fall
to 0.005 to 0.02 ppm on a median basis at river mile 168, while the 95th percentile is expected to
reach 0.01 to 0.04 ppm. At river mile 154, median concentrations fall to approximately 0.004
ppm, and the 95th percentile to 0.005 to 0.008 ppm. Median concentrations for all three boundary
conditions are shown in Figure 7-6, and the percentile concentrations under the zero upstream
boundary condition in Figure 7-9.
White perch concentrations are predicted to fall to 0.005 - 0.02 ppm on a median basis at
river mile 154, and to approximately 0.01 - 0.04 ppm on a 95th percentile basis.
7.3 Predicted PCB Concentrations in Fish under the 10 ng/L Upstream Boundary
Condition
Figure 7-4 presents the results of predicted fish body burdens on a wet weight basis for
largemouth bass predicted median concentrations under the three upstream boundary conditions.
Figure 7-5 shows the same results for brown bullhead, and Figure 7-6 shows these results for
yellow perch and white perch. For all species, concentrations decline roughly exponentially and
approach different asymptotes depending on the species and upstream boundary condition. The
median and 95th percentile asymptotes approached by each species are found in Table 7-1. The
values in parentheses are approximate uncertainty bounds on the predicted values based on the
86 MCA/TetraTech
-------
maximum difference between predicted and observed from the hindcast calibration. Table 7-2
provides a comparison of example target levels (note that target levels will be determined during
the feasibility study. The values shown in Table 7-2 are to provide a benchmark for when order-
of-magnitude concentrations will be achieved) based on the 10 ng/L upstream boundary
condition.
Figure 7-4 shows that concentrations in largemouth bass at river mile 189 decline roughly
exponentially. The lowest achieved concentration is approximately 1.5 ppm wet weight, with a
margin of error of approximately 0.8 ppm on either side on a median basis. This is interpreted as
50% of fish are expected to show concentrations above this value and below this value. Figure
7-10 shows the results for the 25th, 50th and 95th percentiles for each of the locations. The 95th
percentile concentration is interpreted as the expected body burden for 95% of the population.
That is, 95% of the population would be expected to experience the shown concentration or less.
For 95% of the population at river mile 189, the lowest concentration achieved is roughly 3.4 on
an annualized basis. At river mile 168, largemouth bass concentrations decline to approximately
0.08 to 0.9 ppm on a median basis (best estimate of 0.3), and at river mile 154, concentrations are
predicted to decline to approximately 0.07 to 0.2 ppm.
Brown bullhead concentrations at river mile 189 are predicted to be somewhat higher
than predicted largemouth bass concentrations. By the end of the forecast period, the forecast
median is approximately 0.7 ppm, and the 95th percentile is predicted to fall at approximately 0.6
to 1.3 ppm. For river mile 168, predicted brown bullhead median concentrations achieve 0.3 to
1.2 ppm, and 0.1 to 0.4 ppm at river mile 154. The corresponding 95th percentile values are 0.5
to 1.8 ppm and 0.015 to 0.06 ppm, for river miles 168 and 154, respectively (Figure 7-11).
Yellow perch concentrations at river mile 189 are predicted to fall to 0.7 to 1.5 ppm on a
median basis, and to 1.8 to 3.9 ppm on a 95th percentile basis. Predicted concentrations fall to 0.1
to 0.4 ppm on a median basis at river mile 168, while the 95th percentile is expected to reach
0.015 to 0.06 ppm. At river mile 154, median concentrations fall to approximately 0.1 ppm, and
the 95th percentile to 0.15 to 0.4 ppm. Median concentrations for all three boundary conditions
are shown in Figure 7-6, and the percentile concentrations under the 10 ng/L upstream boundary
condition in Figure 7-12.
White perch concentrations are predicted to fall to 0.1 - 0.4 ppm on a median basis at
river mile 154, and to approximately 0.2 - 0.8 ppm on a 95th percentile basis.
7.4 Predicted PCB Concentrations in Fish under the 30 ng/L Upstream Boundary
Condition
Figure 7-4 presents the results of predicted fish body burdens on a wet weight basis for
largemouth bass predicted median concentrations under the three upstream boundary conditions.
Figure 7-5 shows the same results for brown bullhead, and Figure 7-6 shows these results for
yellow perch and white perch. For all species, concentrations decline roughly exponentially and
approach different asymptotes depending on the species and upstream boundary condition. The
median and 95th percentile asymptotes approached by each species are found in Table 7-1. The
87 MCA/TetraTech
-------
values in parentheses are approximate uncertainty bounds on the predicted values based, on the
maximum difference between predicted and observed from the hindcast calibration.
Figure 7-4 shows that concentrations in largemouth bass at river mile 189 decline roughly
exponentially. The lowest achieved concentration is approximately 3.5 ppm wet weight, with a
margin of error of approximately 1.8 ppm on either side on a median basis. This is interpreted as
50% of fish are expected to show concentrations above this value and below this value. Figure
7-13 shows the results for the 25th, 50th and 95th percentiles for each of the locations. The 95th
percentile concentration is interpreted as the expected body burden for 95% of the population.
That is, 95% of the population would be expected to experience the shown concentration or less.
For 95% of the population at river mile 189, the lowest concentration achieved is roughly 8.1 on
an annualized basis. At river mile 168, largemouth bass concentrations decline to approximately
0.3 to 3.0 ppm on a median basis (best estimate of 1.0), and at river mile 154, concentrations are
predicted to decline to approximately 0.3 to 0.08 ppm.
Brown bullhead concentrations at river mile 189 are predicted to be somewhat lower than
predicted largemouth bass concentrations. By the end of the forecast period, the forecast median
is approximately 1.8 ppm (1.0 - 2.2 error bounds), and the 95th percentile is predicted to fall at
approximately 1.4 to 3.1 ppm (best estimate of 2.6 ppm). For river mile 168, predicted brown
bullhead median concentrations achieve 0.8 to 3.0 ppm, and 0.3 to 1.2 ppm at river mile 154.
The corresponding 95th percentile values are 1.4 to 5.2 ppm and 0.5 to 1.8 ppm, for river miles
168 and 154, respectively. The 25th, 50th, and 95th predicted percentiles are shown in Figure 7-
14.
Yellow perch concentrations at river mile 189 are predicted to fall to 1.9 to 4.2 ppm on a
median basis, and to 3.1 to 6.7 ppm on a 95th percentile basis. Predicted concentrations fall to 0.4
to 1.4 ppm on a median basis at river mile 168, while the 95th percentile is expected to reach 0.8
to 3.0 ppm. At river mile 154, median concentrations fall to approximately 0.3 ppm, and the 95lh
percentile to 0.4 to 0.7 ppm. Median concentrations for all three boundary conditions are shown
in Figure 7-6, and the percentile concentrations under the 30 ng/L upstream boundary condition
in Figure 7-15.
White perch concentrations are predicted to fall to 0.3 - 1.2 ppm on a median basis at
river mile 154, and to approximately 0.6 - 2.4 ppm on a 95th percentile basis.
7.5 Discussion of Results
The models were designed to predict the observed variability in fish tissue measurements
taken since 1977. Some of the variability that has been observed over time is attributable to
uncertainty, but this is likely to be small relative to the actual population heterogeneity in the
environment. The parameter-specific distributions developed here were designed to capture
variability rather than uncertainty. It can be argued that the dietary composition distributions, for
example, represent uncertainty, but in fact they were derived based on observations of what fish
have consumed in the environment. Similarly the lipid distribution, which contains measurement
error, is primarily a distribution reflecting the differences in lipid content among individual fish.
MCA/TetraTech
-------
Presenting predicted fish body burdens probabilistically provides important information
for decisionmakers and for other aspects of the analysis. The ecological and human health risk
assessments require predicted body burdens to evaluate the potential risk from PCB exposure
under specific conditions. These results characterize exposure concentrations in fish as
distributions rather than single point estimates.
The modeling results can be used directly in the context of specific numerical target
levels. It is straightforward to obtain specific modeling results, that is, if risk managers
determine a particular percentile of population should achieve a target level (say, the 75th or 90th
need to achieve 0.1 ppm wet weight, or 2.0 ppm wet weight), these results can be explicitly
predicted. Variability in the population response to sediment and water concentrations are
reflected in the individual fractiles. Uncertainty for any given fractile based on the uncertainty in
sediment and water concentrations can also be modeled.
Figures 7-4 through 7-6 present the FISHRAND forecast results on a median basis for the
three upstream boundary conditions. These figures show the effect the difference in the upstream
boundary condition has on the asymptotic concentration that predicted fish body burdens
approach. Figures 7-7 through 7-9 provide the 25th, 50th, and 95th percentiles for largemouth
bass, brown bullhead, and white and yellow and perch, respectively, for each of the upstream
boundary conditions.
89 MCA/TetraTech
-------
THIS PAGE INTENTIONALLY LEFT BLANK
90 MCA/TetraTech
-------
Chapter 8
-------
8. DISCUSSION OF UNCERTAINTY
This chapter provides a discussion of uncertainties in the bioaccumulation model
approach and assumptions. These uncertainties can be broadly categorized as model uncertainty
and parameter uncertainty. Model uncertainty is the error associated with how well a model
approximates the true relationships between environmental components. For example, these
would include terms representing functional aspects of the environment that were not included in
the analysis. Model error includes: inappropriate selection or aggregation of variables, incorrect
functional forms, and incorrect boundaries. Parameter uncertainty refers to the uncertainty in
estimating specific values of parameters and forcing functions in the models (e.g., sediment and
water concentrations, etc.) as well as inherent variability (e.g., lipid content, fish weight). Most
modeling parameters will exhibit both variability and uncertainty. Variability, which typically
cannot be reduced but can be better characterized by collecting additional data, represents known
variations in parameters based on observed heterogeneity in the environment. True uncertainty
in parameter estimates could be reduced by collecting more data.
8.1 Model Uncertainty
8.1.1 Model and Parameter Uncertainties in the Fate and Transport Models
Since the bioaccumulation models rely on the sediment and water concentrations from the
fate and transport models, it is important to identify potential sources of uncertainty in these
models to be able to understand the effect on predicted fish body burdens. By necessity, the fate
and transport models are not able to capture every single mechanism contributing to transport
processes. The uncertainty associated with water and sediment concentrations resulting from
potential changes in most sensitive parameters have been selected for explicit modeling, based on
professional judgment, prior experience and existing models. See Book 1 Chapter 7.5 for a
further discussion of uncertainties in the fate and transport models.
8.1.1.1 Sediment and Water A veraging
To forecast future Tri+ body burdens in fish, the same 75% cohesive and 25%
noncohesive averaging was conducted to maintain consistency with the hindcast calibration.
There is uncertainty in these estimates. The true exposure concentration that fish experience
relative to sediment is unknown.
8.1.2 Model Uncertainties in the Bioaccumulation Models
The bioaccumulation models contain a number of simplifications in uptake processes. In
addition, the two statistical approaches presented here contain inherent limitations as compared
to the mechanistic approach. These two aspects of model uncertainty in the bioaccumulation
models are discussed next.
91 MCA/TetraTech
-------
8.1,2.1 Probabilistic Empirical Model and Bivariate BAF Model
These two models use observed data to construct relationships between compartments.
One limitation of these kinds of statistical approaches lies in their predictive power. Models of
this sort cannot reliably be used in terms of prediction as they do not necessarily capture the
mechanistic basis for responses to changes in the system. They can be used to extrapolate
beyond the range of observed data to evaluate trends based on current conditions, but they cannot:
be used to evaluate changes in the system and expected responses to those changes.
8.L2.2 FISHRAND
FISHRAND is based on the modeling approach developed by Gobas (1993). This
approach has been used in the Great Lakes as well as in a number of other modeling contexts.
Further refinements on the original model have been presented in the literature (Gobas et al.,
1995; Morrison et al., 1997). These later approaches involve the following modifications:
Explicit consideration of benthic invertebrate feeding preferences (e.g., burrowers
versus epibenthic species etc.) resulting in a biomagnification mechanism rather than
the equilibrium partitioning (BSAF) approach taken here;
An age-class model for each year of a fish's life rather than the growth dilution
approach presented here; and,
An explicit pharmacokinetic model to consider the role of metabolism.
Benthic feeding: FISHRAND does not explicitly consider benthic feeding strategies but
rather relies on the original equilibrium partitioning approach for several reasons. First,
distributions are used in FISHRAND for a) sediment concentrations, b) total organic carbon in
sediment, and c) benthic invertebrate percent lipid. The sediment concentration distributions are
described as lognormal, while the TOC and lipid distributions are described as triangular. Given
these distributional shapes and the nature of the relationship between sediment concentrations
and invertebrate concentrations, the use of these distributions in the BSAF equation adequately
describe the observed variability in benthic invertebrate concentrations as compared to empirical
data. This observed variability may be attributable to biomagnification but insofar as the model
adequately describes observed data and the equilibrium partitioning equation has been widely
used and accepted, it was decided to take this approach for FISHRAND.
As shown in Figure 5-2, observed biota:sediment accumulation factors from the EPA
Phase 2 database average one, exactly what equilibrium partitioning would predict. The species
categorized as benthic versus epibenthic from the Phase 2 dataset did not show statistically
significant different BSAFs (t-tests).
Age-Class Modeling: The body weight, lipid content and dietary preferences change
significantly over the lifespan of individual fish and the latest Gobas model is developed for
92 MCA/TetraTech
-------
individual generations of age classes of organisms (Gobas et al., 1995). In this study, we have
categorized fish into species-specific age classes. For example, in the case of largemouth bass,
yellow perch, white perch and brown bullhead, the adults in the population are of primary
concern. It is the adult fish in the population that will be consumed by humans and some
ecological receptors. Forage fish (pumpkinseed and spottail shiner) serve as primary prey base
for the larger fish (that are piscivorous) and also other ecological receptors (such as mink and
kingfisher, as examples). Juvenile fish of all species are assumed to have feeding habits more
similar to the forage fish. Two classes of forage fish are considered: one that obtains its
predominant food source from the water column (pumpkinseed) and the other equally from water
and sediment (spottail shiner). These two categories are representative of the kinds of feeding
strategies forage fish and juvenile fish will utilize.
These discreet fish populations are represented by distributions for fish weight and lipid
concentrations. Each individual fish in the population is assumed to grow, i.e. to increase its
individual volume and weight. Such volume increase can lead to decrease in concentration in
this fish if uptake is too slow to compensate for the reduction in chemical mass per volume. The
volume of the population is assumed to be equilibrated by the processes of fish death and
reaching the minimal size to be included in the population.
The approach taken in this report was chosen to maximize the utility of the existing
database and to minimize the number of assumptions required for modeling. Virtually all the
data available for the Hudson River are for fish falling within a particular grouping of age-
classes. Within these age-classes, feeding preferences are consistent and key parameters (e.g.,
weight, lipid, etc.) are represented by distributions. This approach minimized the number of
assumption that had to be made since there are not enough site-specific data available to support
explicit age-class considerations for the larval and juvenile largemouth bass, brown bullhead,
yellow perch, and white perch.
Pharmacokinetics: The metabolism of PCBs likely plays an important role in the ability
of fish to retain PCBs (Niimi, 1997; Gobas, 1999). Experimental data suggest that PCBs can
biomagnify in the food chain due to pharmacokinetic processes in fish (Gobas, 1999, Connolly,
1988, Gobas, 1993). Specifically, food digestion and absorption in gastrointestinal tract is
hypothesized to increase PCB fugacity. Even though these processes have been recently
incorporated in the fish bioaccumulation model by Gobas (Gobas et al., 1999) we believe that the
experimental database and theoretical foundation of this model have to be developed further to
provide better estimates for the required parameters and associated uncertainties. In addition, it
would be best if the Gobas et al., 1999 model were validated for a number of sites before using it
for regulatory decisions. Therefore, the FISHRAND model does not directly account for these
processes and uses as the prototype an earlier version of the Gobas model that was tested and
applied for several sites and in different environmental settings (Morrison et al., 1997, Buckhard,
1998).
Table 6-2 shows that the relative percent difference between predicted and observed for
FISHRAND is typically within 25-40%, and significantly less than that for many individual
years, species, and locations.
93 MCA/TetraTech
-------
8.2 Parameter Uncertainty
All of the parameters used in FISHRAND have some uncertainty associated with them.
For example, even though there is an extensive database of percent lipid for specific fish species
across locations and times, there is laboratory uncertainty associated with these measurements.
The full extent of that uncertainty is not known. Fish feeding preferences are highly uncertain.
Stomach content analyses provide only limited information as the soft-bodied organisms are the
first to be digested and cannot typically be observed, even if a fish is caught immediately after
consuming such organisms. Biomass data, which are required to translate numbers of organisms
observed in the stomach contents to meaningful percent mass or volume estimates, are often
unavailable. Further it is typically not known whether a fish will selectively feed on particular
organisms or whether the fish is strictly an opportunistic feeder, in which case feeding will in
large measure depend on the biomass of prey items in the environment.
8.2.1 Sensitivity Analysis
Our literature review and experimental data collected for the Hudson River has shown
that: 1) river ecosystem characteristics vary significantly from one location to another depending
on flow rate, depth, sediment structure, etc.; and 2) certain parameters in the model (such as
feeding preferences) are only imprecisely known. Moreover, most of the measurements are not
easily related to the FISHRAND generic input parameters because, by their own nature,
experimental measurements are taken at a specific time and space while the FISHRAND model
parameters are, in contrast, values corresponding to averages over time, space and species.
The effect of variation of all input parameters on all model outputs were evaluated in a
sensitivity analysis using the Monte Carlo methodology. In this method, combinations of values
for the input parameters are generated randomly. Each parameter appears with the frequency
suggested by its probability distribution. For each combination of input parameters, the output of
the model is recorded. The combination of all possible outputs generated in this manner is used
to construct the distribution of model outputs, which reflect the influence of the undetermined
parameters on the output values.
The partial rank and Spearman rank regression techniques (Morgan and Henrion, 1990)
are used as a formal method to find the most important parameters for the model performance. If
the Spearman or partial rank regression coefficient (PRRC or SRRC) is close to 1 or -1 for a
specific input model parameter, this parameter significantly influences model output. Table 8-1
shows that the correlation coefficients estimated for the percent lipid in water column
invertebrates are above 0.5 for most species and location for the lipid normalized results. The
percent lipid in fish is strongly negatively correlated with PCB body burden expressed on a lipid-
normalized basis. This is because increases in lipid increase the PCB storage capacity of the fish,
reducing the apparent concentration. As expected, the percent lipid in fish is positively
associated for the wet weight results, but less so. This confirms that particularly on a lipid-
normalized basis, the percent lipid distribution is very important. KOW and benthic percent lipid
are also important for some species on a wet weight basis. Feeding preferences are only weakly
correlated with body burdens in terms of sensitivity to this parameter. Tables 8-2 through 8-4
94 MCA/TetraTech
-------
present the correlation coefficients for weight and lipid normalized results, expressed as partial
rank correlation coefficients, and Spearman rank correlation coefficients.
As described in Chapters 3 and 6, sensitivity to model constants was evaluated by
approximating an analytical solution and then taking partial derivatives of all the model constants
with respect to fish concentration. Derivatives of the model constants were evaluated across the
full ranges of all parameters to determine the sign and magnitude of each of the derivatives. The
assimilation efficiency and growth rate were determined to be the most important parameters in
terms of effect on predicted fish concentration. This procedure was described in the approach to
calibration in Chapters 3 and 6.
8.2.1.1 Lipid Content
Lipid content of organisms plays an important role in the model. Uncertainty in the
interpretation of observed data is attributable to differences in laboratory determination of lipid
content of fish tissue. PCBs are lipophilic, stored primarily in fatty tissue, and it is generally
agreed that lipid normalization (i.e., expressing PCB body burden on a lipid basis) provides a
more consistent basis for evaluating bioaccumulation than wet-weight PCB concentrations.
Lipid-normalized PCB body burden is calculated as the reported wet-weight PCB concentration
divided by the lipid concentration. FISHRAND first estimates a wet-weight concentration and
then lipid-normalizes these results. Unfortunately, any imprecision in the determination of lipid
concentration will also result in imprecision in the calculation of lipid-normalized PCB body
burden. Further, the propagation of uncertainty will be non-linear, as the lipid-normalized
concentration involves division by the lipid content. Therefore, estimation of the uncertainty in
lipid-based PCB concentrations must also include an analysis of the uncertainty in determination
of lipid concentration. Inter-laboratory comparisons conducted by NYSDEC in September 1992
showed an average variability between laboratories of ten percent in determining lipid content of
biological specimens, with results from some pairs of laboratories showing a consistent relative
bias.
Information on the precision of lipid determinations in Hudson River fish data is provided
by three sets of interlaboratory comparisons performed for NYSDEC in 1989, 1992, and 1995.
The 1989 comparisons involved 4 samples and 8 laboratories, the 1992 comparisons involved 5
samples and 12 laboratories, and the 1995 comparisons involved 3 samples and 4 laboratories.
The two laboratories responsible for the majority of NYSDEC fish analyses (Hazleton and
successors, and Hale Creek) participated in each of the interlaboratory comparisons.
Over the 12 samples, standard deviations between laboratories on percent lipid
determinations ranged from 0.052 to 0.52. The standard deviation is scale dependent, however,
and it is more informative to examine the coefficient of variation (standard deviation divided by
the mean). Coefficients of variation on percent lipid ranged from 0.023 to 0.38, with an average
of 0.099, indicating a relatively high degree of precision in lipid determinations.
Results reported by Hazleton appear to show consistent deviations relative to the mean
across all laboratories. For the 1989 results, all Hazleton lipid determinations were less than the
mean, with the discrepancy ranging from -0.75 to -3.88 standard deviation units on the
95 MCA/TetraTech
-------
percentage value, with an average of -2.55 standard deviations. For both 1992 and 1995, all
Hazleton results were greater than the interlaboratory mean, with an average discrepancy of 4.47
standard deviations. The discrepancies appear relatively large because the standard errors are
small.
The interlaboratory mean depends on the characteristics of the laboratories that
participated in a given year. When Hazleton is compared to Hale Creek, however, the same
pattern emerges: Hazleton results are consistently lower than Hale Creek in 1989, and
consistently higher in 1992 and 1995. The Hale Creek lipid determinations do not show any
consistent bias with time relative to the interlaboratory mean. Across all samples, the
discrepancies for Hale Creek versus the mean range from -1.0 to +0.99 standard deviation units,
with an average of -0.29 standard deviations.
Hazleton results are compared to the interlaboratory means in Figure 8-1. While the
discrepancies are sometimes large in terms of standard deviation units, the average absolute
difference between Hazleton and the interlaboratory mean is only 0.65 percentage points.
Based on the results of NOAA's mussel method detection limit (MDL) study (see
USEPA, February 1993a for details), the percent lipid determination for benthic invertebrates
was considered as estimated. Therefore, the percent lipid of benthic invertebrates was based on
the mean of all invertebrates analyzed in the Phase 2 study. The variability seen in the percent
lipid composition was associated with the small sample mass associated with some of the
samples (1 gram wet weight). The confidence of percent lipids was higher for fish samples,
which had more material available for analysis.
O.2.1. 2 K.QW
The optimal posterior distribution for KOW was determined through the Bayesian
calibration procedure. However, there is uncertainty as to whether this optimized distribution
represents the true distribution in the future. It is likely that the congener composition of the
PCB mixture in the environment may change over time, and there is uncertainty as to whether the
optimized distribution obtained through calibration to historical data remains valid for future
forecasts. However, the for which the Kow distribution was optimized represent data obtained
over a 21 -year period, and for the most part, direct source contributions (as opposed to sediment
or in-river PCB contributions) have declined. There is greater confidence having used data over
a longer time period than simply one or two years.
The optimized KOW distribution is quite different between river miles 189 and 168 (the
same distribution was used at 154 as 168). This suggests that the congener distribution may
differ between river miles (as has been suggested by other analyses, e.g., USEPA, 1999b; NOAA,
1998). Also, the river behaves quite differently between the Thompson Island Pool (river mile
189) and the remainder of the river.
96 MCA/TetraTech
-------
Chapter 9
-------
9. SUMMARY AND CONCLUSIONS
Three food chain models were developed to describe the uptake of PCBs, expressed as
ZTri+, which is representative of total PCBs in fish tissues. These models are:
Bivariate BAF Analysis
The Bivariate BAF Analysis relates measured PCB levels in water and sediments (two
variables, or "bivariate") to measured PCB levels in fish. This analysis was applied to the Upper
Hudson River and to a segment of the Lower Hudson River near Albany. The Bivariate BAF
Analysis was developed using the historical PCB Aroclor database. Results presented in this
report build upon the earlier analysis presented in the Preliminary Model Calibration Report
(1996).
Empirical Probabilistic Food Chain Model
The Empirical Probabilistic Food Chain Model is contructed by linking fish body burdens
to PCB exposure concentrations in water and sediments. The model combines information from
available PCB exposure measurements with knowledge about the ecology of different fish
species and the relationships among larger fish, smaller fish, and invertebrates in the water
column and sediments. The Probabilistic Model was developed using both historical and 1993
field data, and was applied to the Upper Hudson River down the Federal Dam at Troy. In
contrast to the Bivariate BAF Analysis, which provides average body burden estimates, the
Probabilistic Model provides information on the expected range of uncertainty and variability
around these average estimates.
Mechanistic Time-Varying Model (FISHRAND) Based on Gobas (1993)
The FISHRAND model is based on the peer-reviewed uptake model developed by Gobas
(1993 and 1995). This is the same form of the model that was used to develop criteria under the
Great Lakes Initiative (EPA, 1995). This probabilistic model was programmed in Fortran-90
using the LSODE (the Livermore Solver for Ordinary Differential Equations (Radhakrishnan and
Hindmarsh, 1993)) and incorporating a Microsoft Excel graphical user interface.
Food Web Biology
As part of the development of the food web models, species-specific profiles (i.e., descriptions of
feeding behavior, habitat preferences, range and movement) were developed for yellow perch,
largemouth bass, pumpkinseed sunfish, brown bullhead, white perch, spottail shiner, shortnose
sturgeon and striped bass. These profiles include: information on species-specific characteristics
influencing bioaccumulation potential of PCBs; as well as the details of specific gut analyses
conducted by Menzie-Cura & Associates, Inc., Exponent, Inc.; and information in the literature
from the Hudson River power plant studies. These profiles helped develop dietary composition
distributions for each of the fish species.
97 MCA/TetraTech
-------
Target Levels for Fish Body Burdens
Appropriate fish body burden target levels for the protection of ecological receptors and
human health have not yet been established for the Hudson River PCBs Site Reassessment. To
provide perspective on the range of concentrations predicted for each fish species, four different
values have been selected. These values do not represent particular target levels and should not
be interpreted as potential target levels for this site. These values were selected strictly for
comparative purposes. The concentrations selected represent a range of concentrations and
orders of magnitude.
9.1 Summary of Food Web Models
The Bivariate BAF Analysis represents PCBs in terms of the sum of trichloro- through
decachlorbiphenyls (denoted ZTri+). Historical Aroclor quantitation schemes are not
consistent with one another, but can be translated to a consistent estimate of 2Tri+.
Information on mono- and dichlorobiphenyl concentrations is not available in most of the
historical PCB monitoring data. The Probabilistic Bioaccumulation Food Chain Model
and FISHRAND also represent ZTri+ (approximately equivalent to total PCBs in fish
tissue).
The Bivariate BAF Analysis for fish body burden in a given species is based on the
historical dataset of Aroclor measurements, with corrections for changing quantitation
methods. It is designed to provide a statistical perspective on the empirical relationships
between water, sediment, and fish body burdens. The statistical model relies on a
bivariate regression approach which relates fish body burdens to concentrations in both
water and sediment. This allows for the possibility that water and sediment
concentrations are not in equilibrium, as is frequently observed in the Upper Hudson
River.
The Probabilistic Bioaccumulation Food Chain Model consists of the following biotic
compartments: (a) benthic invertebrates; (b) water column invertebrates; (c) forage fish;
(d) piscivorous fish; and (e) demersal fish. PCB concentrations are expressed as lipid-
normalized in biota, total organic carbon normalized in sediments and whole water in the
water column. Relationships among compartments are expressed as bioaccumulation
factors between the concentration in a given compartment and the expected dietary
exposure for that compartment. The dietary exposure is based on a weighted
concentration in the diet.
Statistical distributions of bioaccumulation factors have been derived for:
sediments to benthic invertebrates;
whole water PCB concentrations to water column invertebrates;
expected dietary concentrations to composite forage fish; and
pumpkinseed to largemouth bass.
98 MCA/TetraTech
-------
FISHRAND was developed based on Gobas (1993) and compared to published modeling
results for Lake Ontario to verify model functionality. This model was then modified for
the Hudson River by eliminating Lake Ontario species and including Hudson River
species along with site-specific and fish specific parameters.
Species-specific profiles are presented for yellow perch (Perca flavescens), largemouth
bass (Micropterus salmoides), pumpkinseed (Lepomis gibbosus), brown bullhead
(Ictalurus nebulosus), white perch (Morone americana), spottail shiner (Notropis
hudsonius), shortnose sturgeon (Acipenser brevirostrum) and striped bass (Morone
saxatilis). These profiles describe foraging strategies, home-ranges, habitat preferences
and information on reproduction for each of these species.
The foraging strategies of the invertebrate prey base for the fish species is viewed as a key
component to evaluating relative sediment versus water influences on fish body burdens.
An analysis is presented here that uses an indicator species approach based on identified
macroinvertebrates from the gut contents of Hudson River fish in order to differentiate
sediment versus water exposure pathways via the food chain.
FISHRAND predicts expected body burdens in fish on a population-level basis. The
model assumes a cycling of the population in which older fish are replaced by younger
fish within a particular size range. For this modeling application, the age-class of interest
includes the adult of the species for piscivorous, semi-piscivorous and omnivorous fish
while for the forage fish the age-class of interest is the young-of-year (or yearlings).
The FISHRAND model calibration procedure focused on achieving wet weight
concentrations rather than lipid normalized concentrations. This is because the model
predicts a wet weight concentration and the method provides for more robust predictions
within the decisionmaking context for this site.
Both the probabilistic and mechanistic models were run using predicted hindcasting water
and sediment concentration results from the fate and transport models as inputs in a
validation exercise. The models were used to predict observed fish concentrations (from
NYSDEC) for the period 1977 - 1997 for several locations above the Federal Dam at
Troy.
The FISHRAND model was run for 70-year forecasts (1998 - 2067) using sediment and
water concentrations from the HUDTOX model. Three scenarios were run, assuming: a) a
zero upstream boundary condition, and b) a 10 ng/L constant upstream boundary
condition, and c) a 30 ng/L constant upstream boundary condition (see Books 1 and 2).
9.2 Principal Report Findings
The following conclusions have been drawn based on the work presented in this Revised
Baseline Modeling Report:
The Bivariate BAF Analysis for fish body burdens explains about 80 percent of the
observed variability in summer average concentrations of tri- through deca-chlorinated
PCBs in fish from the freshwater portion of the Hudson River. Much of the remaining,
99 MCA/TetraTech
-------
unexplained variability is due to uncertainty in historic water column concentrations. The
BAF analysis suggests a need to consider both the water column and local sediments as
sources for bioaccumulation of PCBs in Upper Hudson River fish. The relative
importance of water and sediment sources determined in the Bivariate BAF Analysis is
consistent with species feeding behavior: for species that feed in the water column, the
water column pathway tends to dominate, while for bottom-feeders, the sediment
pathway tends to be dominant. Fish-eating species at higher levels in the food chain
appear to accumulate PCBs from both water column and sediment pathways.
Using the hindcast calibration results from the fate and transport models, the probabilistic
empirical model reasonably captures observed historical PCB concentrations in fish.
Comparisons are available for largemouth bass, brown bullhead, and pumpkinseed at
river miles 168 and 189, and for largemouth bass at river mile 154.
Using the hindcast calibration results from the fate and transport models, the FISHRAND
model captures observed historical PCB concentrations in fish to within a factor of two
for most locations and species, and typically significantly better than that. Largemouth
bass concentrations are captured within a factor of 1.3 for 1990 - 1997. Comparisons are
available for largemouth bass, pumpkinseed, yellow perch and brown bullhead at river
miles 189, 168, and 154. White perch comparisons are available at river mile 154.
Predictions from the probabilistic empirical model for largemouth bass compare
favorably to the results for FISHRAND on a median basis. On a 95th percentile basis, the
probabilistic model typically predicts approximately a factor of two higher than the
FISHRAND model.
Within year variability predicted by the FISHRAND model is approximately a factor of
two. Month to month comparisons of model output to data (that is, comparing model
results for the month corresponding to the month of sample collection) showed the lowest
relative percent difference. However, comparisons to data for annualized FISHRAND
predictions are similar although individual relative percent differences are slightly larger
as the annualized results average out this seasonal variation.
The FISHRAND 70-year forecasts show that predicted wet weight STri+ PCB fish body
burdens asymptotically approach steady-state concentrations. These concentrations are
species-specific, depending on the relative influence of sediment versus water sources,
and reflect the upstream boundary assumption. That is, the asymptotic value is lowest for
the 0 ng/L upstream boundary condition, approximately an order of magnitude higher for
the 10 ng/L upstream boundary condition, and approximately a factor of five higher under
the 30 ng/L upstream boundary condition.
At the end of the 70-year forecast period, the lowest achieved concentration for
largemouth bass at river mile 189 under the zero upstream boundary condition is
approximately 0.1 ppm wet weight, with an error of approximately a factor of two on
either side on a median basis. This asymptote is approached in roughly 2039 and median
predicted concentrations remain approximately at that level from then on. For 95% of the
100 MCA/TetraTech
-------
population at river mile 189, the lowest concentration achieved is roughly 0.3 on an
annualized basis, again in approximately 2039. At river mile 168, largemouth bass
concentrations decline to approximately 0.005 to 0.06 ppm on a median basis (best
estimate 0.02 ppm), and at river mile 154, concentrations are predicted to decline to
approximately 0.007 to 0.02 ppm (best estimate 0.01). These values all occur at roughly
2039.
Under the zero upstream boundary condition, brown bullhead concentrations at river mile
189 are predicted to be somewhat higher than predicted largemouth bass concentrations.
By the end of the forecast period, the median is predicted to be approximately 0.1 ppm,
and the 95th percentile is predicted to fall at approximately 0.1 to 0.24 ppm. These values
are first achieved in approximately 2039. Concentrations increase briefly from 2048 -
2052 (to slightly above 0.4 ppm), and then decrease again by 2059. For river mile 168,
predicted brown bullhead median concentrations achieve 0.01 to 0.04 ppm, and 0.005 to
0.02 ppm at river mile 154. The corresponding 95th percentile values are 0.015 to 0.06
ppm and 0.01 to 0.04 ppm, for river miles 168 and 154, respectively.
Under the zero upstream boundary condition, yellow perch concentrations at river mile
189 are predicted to fall to 0.03 to 0.06 ppm on a median basis, and to 0.05 to 0.11 ppm
on a 95th percentile basis. Predicted concentrations fall to 0.005 to 0.02 ppm on a median
basis at river mile 168, while the 95th percentile is expected to reach 0.01 to 0.04 ppm. At
river mile 154, median concentrations fall to approximately 0.004 ppm, and the 95th
percentile to 0.005 to 0.008 ppm..
Under the zero upstream boundary condition, white perch concentrations are predicted to
fall to 0.005 - 0.02 ppm on a median basis at river mile 154, and to approximately 0.01 to
0.04 ppm on a 95th percentile basis.
At the end of the 70-year forecast period, the lowest achieved concentration for
largemouth bass at river mile 189 under the 10 ng/L upstream boundary condition is
approximately 1.5 ppm wet weight, with an error of approximately a factor of 1.4 on
either side on a median basis. For 95% of the population at river mile 189, the lowest
concentration achieved is roughly 3.4 on an annualized basis. At river mile 168,
largemouth bass concentrations decline to approximately 0.08 to 0.9 ppm on a median
basis (best estimate of 0.3 ppm), and at river mile 154, concentrations are predicted to
decline to approximately 0.07 to 0.2 ppm (best estimate of 0.1).
Under the 10 ng/L upstream boundary condition, brown bullhead concentrations at river
mile 189 are predicted to be somewhat higher than predicted largemouth bass
concentrations at river miles 168 and 154, and somewhat lower at river mile 189. By the
end of the forecast period, the median is predicted to be approximately 0.7 ppm, and the
95lh percentile is predicted to fall at approximately 0.6 to 1.3 ppm. For river mile 168,
predicted brown bullhead median concentrations achieve 0.3 to 1.2 ppm, and 0.1 to 0.4
ppm at river mile 154. The corresponding 95th percentile values are 0.5 to 1.8 ppm (best
estimate of 0.9 ppm) and 0.15 to 0.6 ppm (best estimate ofO.3), for river miles 168 and
154, respectively.
101 MCA/TetraTech
-------
Under the 10 ng/L upstream boundary condition, yellow perch concentrations at river
mile 189 are predicted to fall to 0.7 to 1.5 ppm on a median basis (best estimate of 1.4
ppm), and to 1.8 to 3.9 ppm (best estimate of 3.5) on a 95th percentile basis. Predicted
concentrations fall to 0.1 to 0.4 ppm on a median basis at river mile 168, while the 95th
percentile is expected to reach 0.15 to 0.6 ppm. At river mile 154, median concentrations
fall to approximately 0.1 ppm, and the 95th percentile to 0.2 ppm.
Under the 10 ng/L upstream boundary condition, white perch concentrations are predicted
to fall to 0.1 - 0.4 ppm on a median basis at river mile 154, and to approximately 0.4 ppm
on a 95th percentile basis.
At the end of the 70-year forecast period, the lowest achieved concentration for
largemouth bass at river mile 189 under the 30 ng/L upstream boundary condition is
approximately 3.5 ppm wet weight, with an error of approximately a factor of 1.4 on
either side on a median basis. For 95% of the population at river mile 189, the lowest
concentration achieved is roughly 8.4 on an annualized basis. At river mile 168,
largemouth bass concentrations decline to approximately 0.3 to 3.0 ppm on a median
basis (best estimate of 1.0 ppm), and at river mile 154, concentrations are predicted to
decline to approximately 0.3 to 0.8 ppm (best estimate of 0.4).
Under the 30 ng/L upstream boundary condition, brown bullhead concentrations at river
mile 189 are predicted to be somewhat higher than predicted largemouth bass
concentrations at river miles 168 and 154, and somewhat lower at river mile 189. By the
end of the forecast period, the median is predicted to be approximately 1.8 ppm, and the
95th percentile is predicted to fall at approximately 1.4 to 3.1 ppm at river mile 189. For
river mile 168, predicted brown bullhead median concentrations achieve 0.8 to 3.0 ppm,
and 0.3 to 1.2 ppm at river mile 154. The corresponding 95th percentile values are 1.4 to
5.2.ppm (best estimate of 2.6 ppm) and 0.5 to 1.8 ppm (best estimate of 0.9), for river
miles 168 and 154, respectively.
Under the 30 ng/L upstream boundary condition, yellow perch concentrations at river
mile 189 are predicted to fall to 1.9 to 4.2 ppm on a median basis (best estimate of 3.8
ppm), and to 3.1 to 6.7 ppm (best estimate of 6.1) on a 95th percentile basis. Predicted
concentrations fall to 0.4 to 1.4 ppm on a median basis at river mile 168, while the 95th
percentile is expected to reach 0.8 to 3.0 ppm. At river mile 154, median concentrations
fall to approximately 0.3 ppm, and the 95th percentile to 0.5 ppm.
Under the 30 ng/L upstream boundary condition, white perch concentrations are predicted
to fall to 0.3 - 1.2 ppm on a median basis at river mile 154, and to approximately 0.6 ppm
on a 95th percentile basis.
The results of the FISHRAND model show that between 4 and 15% of the !Tri+ PCB
uptake in fish is attributable to direct water uptake, and the remainder to dietary sources.
The forage fish (pumpkinseed and spottail shiner) are at the high end of this range; the
remaining fish at the low end. It is difficult to analytically separate water and sediment
sources in the dietary pathway, so the relative influence of water and sediment was
evaluated using a steady-state solution to the dynamic model. Sediment (mg/kg dry
102 MCA/TetraTech
-------
weight) and dissolved water (ng/L) were regressed against predicted fish concentration
(mg/kg wet weight) to evaluate the effect of changes in sediment and water
concentrations on predicted fish body burdens. This analysis showed that brown bullhead
are most sensitive to changes in sediment concentration and not very sensitive to changes
in water concentration; largemouth bass are more sensitive to sediment concentrations
than to water concentrations but water plays a larger role than for brown bullhead; yellow
perch are driven primarily by the water; white perch show greater sensitivity to sediment;
and pumpkinseed and spottail shiner are more sensitive to small changes in water
concentration.
103 MCAyTetraTech
-------
THIS PAGE INTENTIONALLY LEFT BLANK
104 MCA/TetraTech
-------
References
-------
REFERENCES
Albro, P.W. and C.E. Parker. 1979. Comparison of the compositions of Aroclor 1242 and
Aroclor 1016. Journal of Chromatography, 169: 161-166.
Armstrong, R.W. and R.J. Sloan. 1988. PCB Patterns in Hudson River Fish: 1. Resident
Freshwater Species. Pp. 304-324 in C.L. Smith, Ed., Fisheries Research in the Hudson River.
State University of New York Press, Albany, NY.
Anderson, R.C. and D. Brazo. 1978. Abundance, feeding habits and degree of segregation of the
spottail shiner (Notropis hudsonius) and longnose dace (Rhinichthys cataractae) in a Lake
Michigan surge zone near Ludington, Michigan. Mich. Acad. 10(3):337-346.
Ankley, G.T., P.M. Cook, A.R. Carlson, D.J. Call, J.A. Swenson, H.F. Corcoran, and R. A Hoke.
1992. Bioaccumulation of PCBs from sediments by oligochaetes and fishes: comparison of
laboratory and field studies. Can. J. Fish. Aquat. Sci. 49:2080-2085.
Barren, M. 1990. Bioconcentration. Environ. Sci. Technol 24:1612-1618.
Bates, D. and D. Watts. 1988. Nonlinear Regression Analysis and its Applications. New York:
John Wiley & Sons.
Bath, D.W. and J.M. O'Connor. 1982. The biology of the white perch (Morone Americana) in
the Hudson River estuary. Fish Bull. U.S. 80:599-610.
Beebe, C. A., and I.R. Savidge. 1988. Historical perspective on fish species composition and
distribution in the Hudson River estuary. American Fisheries Society Monograph. 4:25-36.
Bierman, V. J. Jr., et al. 1992. Development and Validation of an Integrated Exposure Model
for Toxic Chemicals in Green Bay, Lake Michigan. U.S. EPA Cooperative Agrement CR-
814885.
Bierman, V.J., Jr. 1990. Equilibrium partitioning and biomagnification of organic chemicals in
benthic animals. Environ. Sci. Technol. 24:1407-1412.
Bierman. V.J., Jr. 1994. Partitioning of organic chemicals in sediments: Estimation of interstitial
concentrations using organism body burdens. Chapter 9 in: Transport and Transformation of
Contaminants Near the Sediment-Water Interface. (Ed. J.V. DePinto, W. Lick and J.F. Paul)
CRC Press, pp. 153-175.
Biscayne, P.E. and C.R. Olsen. 1976. Suspended particulate concentrations and compositions in
the New York bight; proceedings of the symposium, special symposia Vol. 2, Middle Atlantic
Continental Shelf and the New York Bight. American Soc. of Limnology and Oceanography; pp.
124-137.
105 MCA/Tetra Tech
-------
Bopp, R.F. and H.J. Simpson, 1984. Persistent chlorinated hydrocarbon contaminants in the New
York Harbor complex, 1st year final report for Contract HUD 1183-A38, by Lamont-Doherty
Geological Observatory of Columbia Univ., submitted to Hudson River Foundation; 6 pp. +
extensive tables, figures, app.
Bopp, R.F. 1979. The geochemistry of polychlorinated biphenyls in the Hudson River. Ph.D.
Dissertation, Columbia University. Univ. Micro. Intern. 8008704, 191 pp.
Boreman, J. 1981. Life histories of seven fish species that inhabit the Hudson River Estuary.
National Marine Fisheries Service, Woods Hole Laboratory, No. 81-34, October.
Brown, M.P., M.B. Werner, R.J. Sloan, and K.W. Simpson. 1985. Polychlorinated biphenyls in
the Hudson River, recent trends in the distribution of PCBs in water, sediment and fish.
Environmental Science and Technology 19(8):656-661.
Buckley, J. and B. Kynard. 1981. Spawning and rearing of shortnose sturgeon from the
Connecticut River. Prog. Fish-Cult. 43:74-76.
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.
Burmistrov, D., Linkov, I. (1999). "Reducing Uncertainty in The Radionuclide Transport
Modeling for the Techa River Using Bayesian Updating." In: "Proceedings of the IUR
Workshop 1-5 June 1998" E. Van den Strict, ed. IUR, 1999 (in press)
Bush, B., S. Dzurica, L. Wood, and E.G. Madrigal. 1994. Sampling the Hudson River Estuary for
PCBs using multiplate artificial substrate samplers and congener-specific gas chromatography in
1991. Environmental Toxicology and Chemistry 13(8): 1259-1272.
Butcher, J.B., T.D. Gauthier, and E.A. Garvey. 1997. Use of historical PCB Aroclor
measurements: Hudson River fish data. Environmental Toxicology and Chemistry, 16(8): 1618-
1623.
Carlson, D.M. 1986. Fish and their habitats in the Upper Hudson Estuary. Region 4 Fisheries,
Stamford, New York. November.
Carlson, D.M. 1992. Importance of wintering refugia to the largemouth bass fishery in the
Hudson River Estuary. Journal of Freshwater Ecology 7:173-180.
Chabot, F. and E. J. Maly. 1986. Variation in diet of yellow perch (Perca flavescens) in a
Quebec reservoir. Hydrobiologia 137:117-124.
Chiou, C.T. 1985. Partition coefficients of organic compounds in lipid-water systems and
correlations with fish bioconcentration. Environ. Sci. Technol. 19:57-62.
106 MCA/Tetra Tech
-------
Clayton, J.R., S.P. Pavlou, and N.F. Brietner. 1977. Polychlorinated biphenyls in coastal marine
zooplankton: bioaccumulation by equilibrium partitioning. Environ. Sci. Technol. 11:676
Cole, J.J., N.F. Caraco and B.L. Peierls. 1992. Can phytoplankton maintain a positive carbon
balance in a turbid, freshwater, tidal estuary?. Limnology and Oceanography 37(8): 1608-1617.
Connolly, J.P. and R. Tonelli, 1985. Modelling Kepone in the striped bass food chain of the
James River estuary. Estuarine, Coastal and Shelf Science 20:349-366.
Connolly, J.P., R.P. Winfield, and J.R. Blasland. NO DATE. Application of a food chain model
of PCB accumulation to the striped bass of the Hudson Estuary. Report to the Hudson River
Foundation. 35 pp.
Crance, J.H. 1986. Habitat suitability index models and instream flow suitability curves:
shortnose sturgeon. U.S. Fish Wildl. Serv. Biol. Rep. 82(10.129) 31 pp.
Di Toro, D.M., C.S. Zarba, D.J. Hansen, W.J. Berry, R.C. Swartz, C.E. Cowan, S.P. Pavlou,
H.E. Allen, N.A. Thomas, and P.R. Paquin. 1991. Technical basis for establishing sediment
quality criteria for nonionic organic chemicals using equilibrium partitioning. Environmental
Toxicology and Chemistry 10(12): 1541-1583.
DiToro, D.M. 19085. A particle interaction model of reversible organic chemical sorption.
Chemosphere 14(10): 1503-1138.
Domermuth, R.B. and R.J. Reed. 1980. Food of juvenile American shad (Alosa sapidissima),
juvenile blueback herring (Alosa aestevalis) and pumpkinseed (Lepomis gibbosus) in the
Connecticut River below Holyoke Dam, Massachusetts. Estuaries 3:65-68.
Dovel, W.L. 1992. Movements of immature striped bass in the Hudson Estuary. In Estuarine
Research in the 1980s: The Hudson River Environmental Society Seventh Symposium on Hudson
River Ecology (Ed: C.L. Smith) State University of New York Press, pp. 276-300.
Dovel, W.L., A.W. Pekovitch, and T.J. Bergren. 1992. Biology of the shortnose sturgeon
(Acipenser brevirostrum Lesueur, 1818) in the Hudson River Estuary, New York. In Estuarine
Research in the 1980s: The Hudson River Environmental Society Seventh Symposium on Hudson
River Ecology (Ed: C.L. Smith) State University of New York Press, pp. 187-216.
Draper, W.M., D. Wijekoon, and R.D. Stephens. 1989. Speciation and quantitation of Aroclors
in hazardous wastes based on PCB congener data. Chemosphere, 22: 147-163.
Endicott, D., R. Kreis, D. Griesmer, and L. Mackelburg. 1994. PCB partitioning and
bioaccumulation in Green Bay, Lake Michigan. Presented as a poster at the 15th Annual Meeting
of the Society of Environmental Toxicology and Chemistry, October 30 - November 3, 1994.
Denver, Colorado.
107 MCA/Tetra Tech
-------
Exponent, Inc. 1998. Data documentation and interpretation report: Submerged aquatic
vegetation and fish community analysis. Prepared for General Electric Company, Albany, NY,
April.
Exponent, Inc. 1998. Volume 1: Data report - macroinvertebrate communities and diets of
selected fish species in the upper Hudson River. Fall and Spring. Prepared for General Electric
Company, Albany, NY. May and September.
Ewald, G., and P. Larsson. 1994. Partitioning of 14C-labelled 2,2',4,4'-tetrachlorobiphenyl
between water and fish lipids. Environmental Toxicology and Chemistry 13(10): 1577-1580.
Feldman, R.S. 1992. PCB accumulation in Hudson River pumpkinseed sunflsh and bullhead:
influences of invertebrate prey. Ph.D. Dissertation, State University of New York.
Fish, P.A. and J. Savitz. 1983. Variations in home ranges of largemouth bass, yellow perch,
bluegills, and pumpkinseeds in an Illinois lake. Trans. Am. Fish. Soc. 112:147-153.
Gauthier, T.D. 1994. Aroclor Translation Procedures. Internal memorandum, July 7, 1994.
Gradient Corporation, Cambridge, MA.
General Electric. 1993a. Data Summary Report, Hudson River Project Sampling and Analysis
Program, 1991 Hydrographic Survey of the Upper Hudson River. Prepared by O'Brien and Gere
Engineers, Inc., Syracuse, NY. Prepared for General Electric Company, Corporate
Environmental Programs, Albany, NY.
General Electric. 1993b. Data Summary Report, Hudson River Project Sampling and Analysis
Program, 1991 Sediment Sampling and Analysis Program. Prepared by O'Brien and Gere
Engineers, Inc., Syracuse, NY. Prepared for General Electric Company, Corporate
Environmental Programs, Albany, NY.
Geoghegan, P., M.T. Mattson, and R.G. Keppel. 1992. Distribution of the shortnose sturgeon in the
Hudson River Estuary, 1984-1988. In Estuarine Research in the 1980s: The Hudson River
Environmental Society Seventh Symposium on Hudson River Ecology (Ed: C.L. Smith) State
University of New York Press, pp. 217-227.
Gerking, S.D. 1958. The restricted movement offish populations. Biol. Rev. 34:221-242.
Gilbert, C.R. 1989. Species profiles: life histories and environmental requirements of coastal fishes
and invertebrates: Atlantic and shortnose sturgeons. U.S. Fish Wildl. Serv. Biol. Rep. 82(11.122).
U.S. Army Corps of Engineers. TR EL-82-4. 28 pp.
Gladden, J.B., F.R. Cantelmo, J.M. Groom and R. Shapot. 1988. Evaluation of the Hudson River
ecosystem in relation to the dynamics of fish populations. American Fisheries Society Monograph
4:37-52.
Gobas, F.A.P.C., J.B. Wilcockson, R.W. Russell, and G.D. Haffner. 1999. Mechanism of
biomagnification in fish under laboratory and field conditions. Environ. Sci. Technol. 33:133-141.
108 MCA/TetraTech
-------
Gobas, F.A.P.C., M.N. Z'Graggen and X. Zhang. 1995. Time response of the Lake Ontario
ecosystem to virtual elimination of PCBs. Env. Science TechnoL 29(8):2038-2046.
Gobas, F.A.P.C. 1993. A model for predicting the bioaccumulation of hydrophobic organic
chemicals in aquatic food-webs: Application to Lake Ontario. Ecological Modelling 69:1-17.
Gobas, F.A.P.C., X. Zhang, and R. Wells. 1993. Gastrointestinal magnification: the mechanism
of biomagnification and food chain accumulation of organic chemicals. Environ. Sci. TechnoL
27:2855-2863.
Gobas, F.A.P.C., D.C.G. Muir, and D. Mackay. 1988. Dynamics of dietary bioaccumulation
and fecal elimination of hydrophobic organic chemicals in fish. Chemosphere 17:943-962.
Gunn, J.M., S.U. Quadri and D.C. Mortimer. 1977. Filamentous algae as a food source for the
brown bullhead (Ictalurus nebulosus). J. Fish Res. Board Can. 34:396-401.
Harkey, G.A., M.J. Lydy, J. Kukkonen, and P.G. Landrum. 1994. Feeding selectivity and
assimilation of PAH and PCB in Diporeia spp. Environmental Toxicology and Chemistry
13(9): 1445-1455.
Hasler, A.D. and W.J. Wisby. 1958. The return of displaced largemouth bass and green sunfish to a
'home1 area. Ecology 39:289-293.
Hjorth, D.A. 1988. Feeding selectivity of larval striped bass and white perch in the Hudson River
Estuary. In Fisheries Research in the Hudson River (Ed: C.L. Smith) State University of New York
Press, Albany, pp. 134-147.
Hydroscience, Inc. 1975. Development of a steady state water quality model for New York
Harbor. Volume I, for ISC; 302 pp.
Hydroscience, Inc. 1978a. Estimation of PCB reduction by remedial action on the Hudson River
ecosystem. For NYS DEC; 107 pp.
Hydroscience, Inc. 1987b. Seasonal steady state modeling. NYC 208 Task Report, Task 314 for
Hazen & Sawyer Engineers, and NYC DWR; 697 pp.
lannuzzi, T.J., Harrington, N.W., Shear, N.M., Curry, C.L., Carlson-Lynch, H., Henning, M.H.,
Su, S.H., and Rabbe, D.E. (1996). "Distributions of key exposure factors controlling the uptake
of xenobiotic chemicals in an estuarine food web." Environ. Toxicol. Chem. 15: 1979-1992.
Johnson, J.H. 1983. Summer diet of juvenile fish in the St. Lawrence River. New York Fish and
Game Journal 30(1).
Jones, P.A., R.J. Sloan, and M.P. Brown. 1989. PCB congeners to monitor with caged juvenile
fish in the Upper Hudson River. Environ. Toxicol. Chem. 8:793-803.
109 MCA/Tetra Tech
-------
Kadlec, M.J. 1994. Bioconcentration of Congener Specific Poly chlorinated Biphenyl (PCB) in
Rainbow Trout (Oncorynchus mykissj Exposed to the Water Column of the St. Lawrence River.
M.S. Thesis, State University of New York at Albany.
Kadlec, M.J. and B. Bush. 1994. Bioconcentration of congener specific polychlorinated biphenyl
(PCB) in rainbow trout (Oncorynchus mykiss) exposed to the water column of the General
Motors superfund site, Massena, N.Y. Presented at the Superfund XIV Conference, Hazardous
Materials Research Institute, Washington, DC.
Klauda, R.J., J.B. McLaren, R.E. Schmitt and W.P. Dey. 1988. Life history of white perch in the
Hudson River estuary. American Fisheries Society Monograph 4:69-88. Also in Science, Law and
Hudson River Power Plants: A Case Study in Environmental Impact Assessment.
Kramer, R.H. and L.L. Smith, Jr. 1960. Utilization of nests of largemouth bass, Micropterus
salmoides, by golden shiners, Notemigonus crysoleucas. Copeia (l):73-74.
Lake, J.L, N.I. Rubinstein, H. Lee H, C.A. Lake, J. Heltsche and S. Pavignano. 1990. Equilibrium
partitioning and bioaccumulation of sediment-associated contaminants by infaunal organisms.
Environ. Tox. Chem. 9:1095-1106.
Larsson, P. 1984. Transport of PCBs from aquatic to terrestrial environments by emerging
chironomids. Environmental Pollution (Series A), 34:283-289.
Lawler, Matusky & Skelly Engineers. 1992. 1990 Year Class Report of the Hudson River Estuary
Monitoring Program. Report to Consolidated Edison Company of New York, Inc.
Lawler, Matusky & Skelly Engineers. 1974. 1973 Hudson River aquatic ecology studies at
Roseton and Danskammer Point. Volume III: Fish. Prepared for Central Hudson Gas & Electric
Corporation, October.
LeBlanc, L. and B. Brownawell. 1994. Tests of biaccumulation models for PCBS: a study of
young-of-the-year bluefish in the Hudson River Estuary. Report to the 1993 Polgar Fellowship
Program, pp. VIM- VII-42.
Linkov, I, Burmistrov, D. Kandlikar, M. and schell, W.R. (1999). "Reducing Uncertainty In The
Radionuclide Transport Modeling for Chernobyl Forests Using Bayesian Updating." In:
"Contaminated Forests: Recent Developments In Risk Identification and Future Perspectives, "
Linkov, I., Schell, W.R., eds. Kluewer, Amsterdam (in press)
Lutz, C.H. and V.A. McFarland. 1994. PCB accumulation factors indicate steady-state
bioaccumulation in impounded black bullheads, Ameiurus melas. Presented at SETAC.
November 9, 1994.
MacDonald, C.R., C.D. Metcalfe, G.C. Balch and T.L. Metcalfe. 1993. Distribution of PCB
congeners in seven lake systems: interactions between sediment and food-web transport.
Environ. Toxicol. Chem. 12:1991-2003.
10 MCA/Tetra Tech
-------
Mackay, D. 1979. Finding fugacity feasible. Environ. Sci. Technol. 13:1218.
Makarewicz, J.C. 1983. Champlain canal fisheries survey: New York state barge canal. Report to
Malcolm-Pirnie, Inc.
Makarewicz, J.C. 1987. Hudson River fisheries study. Report to Halfmoon Electric Project for
Interpower of New York, Inc.
Mansueti, R.J. 1957. Movements, reproduction and mortality of the white perch in the Patuxent
River estuary, Maryland. Diss. for D.SC. Johns Hopkins Univ.
McBride, N.D. 1985. Distribution and relative abundance of fish in the Lower Mohawk River. New
York State Department of Environmental Conservation. Stony Brook, New York.
McFadden, J.T. Texas Instruments Incorporated and Lawler, Matusky & Skelly Engineers. 1978.
Influence of the proposed Cornwall pumped storage project and steam electric generating plants on
the Hudson River Estuary with emphasis on striped bass and other fish populations. Revised. Report
to Consolidated Edison Company of New York, Inc.
McLaren, J.B., J.C. Cooper, T.B. Hoff and V. Lander. 1981. Movements of Hudson River striped
bass. Trans. Am. Fish. Soc. 110:158-167.
Menzie, C.A. 1980. The chironomid (Insecta:Diptera) and other fauna of a Myrophyllum spicatum
L. plant bed in the lower Hudson River. Estuaries, 3(l):38-54.
Merritt, R.W. and K.W. Cummins. 1978. An Introduction to the Aquatic Insects of North America.
(Iowa: Kendall/Hunt Publishing Company).
Mesing, C.L. and A.M. Wicker. 1986. Home range, spawning migrations, and homing of radio-
tagged Florida largemouth bass in two central Florida lakes. Trans. Am. Fish. Soc. 115:286-295.
Morgan, M.G., and M. Henrion. 1990. Uncertainty: A guide to dealing with uncertainty in
quantitative risk and policy analysis. (Cambridge, UK: Cambridge University Press).
Morrison, H., F.A.P.C. Gobas, R. Lazar, D.M. Whittle, and G.D. Haffner. 1997. Development and
verification of a benthic/pelagic food web bioaccumulation model for PCB congeners in western
Lake Erie. Environ. Sci. Technol. 31:3267-3273.
MPI. 1984. New York State barge canal environmental report maintenance dredging program,
1985-1995. Report to New York State Department of Transportation.
Nack, S. and W. Cook. 1986. Characterization of spawning and nursery habitats of largemouth bass
(Micropterus salmoides) in the Stockport component of the Hudson River National Estuarine
Research Reserve. In Polgar Fellowship Reports of the Hudson River National Estuarine Research
111 MCAyTetra Tech
-------
Reserve Program (Eds: E.A. Blair and J.C. Cooper) New York State Department of Environmental
Conservation, The Hudson River Foundation, and The U.S. Department of Commerce.
National Oceanic and Atmospheric Administration (NOAA). 1984. Emergency striped bass study,
Study V: biotic factors affecting juvenile striped bass survival in the Hudson.Estuary. U.S.
Department of Commerce. National Marine Fisheries Service. September.
Niimi, AJ. 1997. Evaluation of PCBs and PCDD/Fs retention by aquatic organisms. The Science
of the Total Environment 192(2): 123-150.
Novak, M.A., A.A. Reilly, B. Bush and L. Shane. 1990. In situ determination of PCB congener-
specific first order absorption/desorption rate constants using chironomus tentans larvae (insecta:
diptera: chironomidae). Wat. Res. 24(3):321-327.
Novak, M.A., A.A. Reilly, and SJ. Jackling. 1988. Long-term monitoring of polychlorinated
biphenyls in the Hudson River (New York) using caddisfly larvae and other macroinvertebrates.
Arch. Environ. Contam. Toxicol. 17:699-710.
NUS. April 1984. Volume 1, Feasibility Study, Hudson River PCBs Site, New York, EPA
Contract No. 68-01-6699. Table 4-2, pp 4-13.
O'Brien & Gere. 1993. Data Summary Report, Hudson River Project, 1991-1992 Sampling and
Analysis Program, Temporal Water Column Monitoring Program. Report to General Electric
Company, Corporate Environmental Programs. O'Brien & Gere Engineers, Inc., Syracuse, NY.
O'Brien & Gere Engineers, Inc. 1993. 1991 hydrographic survey of the Upper Hudson River
Hudson River project sampling and analysis program. General Electric Company, Corporate
Environmental Programs. Albany, New York.
O'Connor, D.J. 1985. Modeling Frameworks. Toxic Substances Notes, Manhattan College
Summer Institute in Water Pollution Control, Manhattan College, Bronx, New York.
Oliver, E.G. 1987. Biouptake of chlorinated hydrocarbons from laboratory-spiked and field
sediments by oligochaete worms. Environ. Sci. Technol. 21:785-790.
Oliver, B.G., and A. J. Niimi. 1988. Trophodynamic analysis of polychlorinated biphenyl
congeners and other chlorinated hydrocarbons in the Lake Ontario ecosystem. Environ. Sci.
Tech. 22:388-397.
Piavis, P.G. 1991. Yellow perch. Habitat requirements for Chesapeake Bay living resources.
Chesapeake Research Consortium, Inc. Solomons, Maryland. 2nd Edition, pp 14-15.
Pindyck, R.S. and D.L. Rubinfeld. 1981. Econometric Models and Economic Forecasts.
McGraw-Hill Book Company, New York.
12 MCA/Tetra Tech
-------
Pizza, J.C. and J.M. O'Connor. 1983. PCB dynamics in Hudson River striped bass. II.
Accumulation from dietary sources. Aquatic Toxicology 3:313-327.
Quinn, S.L, D.W. Bacon and T.J. Harris. 1999. Assessing the precision of model predictions
and other functions of model parameters. The Canadian Journal of Chemical Engineering
77:723-737, August.
Raney, E.G. 1952. The life history of the striped bass, Roccus saxatilis (Walbaum). Bull. Bingham
Oceanog. Coll 14:5-110.
Raney, E.G. 1967. Some catfishes of New York. Conservationist 21(6):20-25.
Raney, E.G. and D.A. Webster. 1940. The food and growth of the young common bullhead,
Americurus nebulosus (LeSueur) in Cayuga Lake, New York. Trans. Am. Fish. Soc. 69:205-209.
Rhea, J. and M. Werth. 1999. Technical Memorandum Re: Phase 2 Evaluation of Analytical
Biases in the USGS Water Column Database. Memorandum to John Haggard, GE Corporate
Environmental Programs, March 22, 19899. Quantitative Environmental Analysis, LLC, Montvale,
NJ.
Sadzikowski and Wallace. 1976. A comparison of food habits of size classes of three sunfishes
(Lepotnis macrochirus [Rafinesque], L. gibbosus [Linnaeus] and L. cyanellus [Rafinesque]). Am.
Midi. Nat. 95:220-225.
Schaefer, R.H. 1970. Feeding habits of striped bass from the surface waters of Long Island, NY. NY
Fish and Game Journal, 17:1-17.
Schroeder, R.A. and C.R. Barnes. 1983. Trends in Polychlorinated Biphenyl Concentration in
Hudson River Water Five Years after Elimination of Point Sources. Water-Resources
Investigations Report 83-4206. USGS, Albany, NY.
Schulz, D.E., G. Petrick, and J.C. Duinker. 1989. Complete characterization of polychlorinated
biphenyl congeners in commercial Aroclor and Clophen mixtures by multidimensional gas
chromatography-electron capture detection. Environmental Science and Technology, 23: 852-
859.
Scott, W.B. and E.J. Grossman. 1973. Freshwater Fishes of Canada. Bulletin 184. Fisheries Board
of Canada, Ottawa.
Setzler, E.M., W.R. Boynton, K.V. Wood, H.H. Zion, L. Lubbers, N.K. Mountford, P. Frere, L.
Tucker, and J.A. Mihursky. 1980. Synopsis of biological data on striped bass, Morone saxatilis
(Walbaum). NOAA Technical Report NMFS Circular 433, FAO Synopsis No. 121, U.S. Dept. of
Commerce, Rockville, MD. 69 pp.
Setzler-Hamilton, E.M. 1991. White perch. Habitat requirements for Chesapeake Bay living
resources. Chesapeake Research Consortium, Inc. Solomons, Maryland. 2nd Edition, pp 12-20.
113 MCA/Tetra Tech
-------
Shaw, G.R. and D.W. Connell. 1984. Factors controlling bioaccumulation of PCBs. In PCBs and
the Environment Volume 1 National Technical Information Service, pp. 501-516.
Shoemaker, H.H. 1952. Fish home areas of Lake Myosotis, New York. Copela 2:83-87.
Simpson, K.W. and R.W. Bode. 1980. Common Larvae of Chironomidae (Diptera) from New
York State Steams and Rivers. Bulletin No. 439, New York State Museum, The University of the
State of New York
Sloan, R., M. Brown, R. Brandt, and C. Barnes. 1985. Hudson River PCB relationships between
resident fish, water and sediment. Northeastern Environmental Science 3:137-151.
Sloan, R.J., B. Young, and K. Hattala. 1995. PCB Paradigms for Striped Bass in New York
State. Technical Report 95-1 (BEP). New York State Department of Environmental
Conservation, Albany, NY.
Small, M. and Fischbeck. 1999. False prediction in Bayesian updating with incomplete models.
Human and Ecological Risk Assessment 5(2):291-304.
Smith, V.E., J.M. Spurr, J.C. Filkins, and J.J. Jones. 1985. Organocholorine contaminants of
wintering ducks foraging on Detroit River sediments. J. Great Lakes Res. 11:231-246.
Smith. S. 1987. Trophic status of the spottail shiner, (Notropis hudsonius) in Tivoli North Bay's
Hudson River fresh tidal marsh. Polgar Fellowship Reports of the Hudson River National Estuarine
Research Reserve Program.
Steinberg, L.J., Reckhow, K.H., and R.L. Wolpert. 1997. Characterization of parameters in
mechanistic models: a case study of a PCB fate and transport model. Ecological Modelling
97(l-2):35-46, April.
Steinberg, L.J., Reckhow, K.H., and R.L. Wolpert. 1996. Bayesian model for fate and transport
of polychlorinated biphenyl in upper Hudson River. Journal of Environmental Engineering
122(5):341-349.
Swindoll, C.M., and Applehans, P.M. 1987. Factors influencing the accumulation of sediment-
sorbed hexachlorobiphenyl by midge larvae. Bull. Environ. Contam. Toxicol. 39:1055-1062.
Texas Instruments. 1976. Hudson River ecological study in the area of Indian Point. Thermal effects
report. Dallas, Texas.
Texas Instruments. 1980. 1978 year class report for the multiplant impact study: Hudson River
Estuary. Report to Consolidated Edison Company of New York, Inc., September.
Theil, H. 1971. Principles of Econometrics. John Wiley & Sons, New York.
114 MCA/Tetra Tech
-------
Thomann, R.V. 1981. Equilibrium model of fate of microcontaminants in diverse aquatic food
chains. Can. J. Fish. Aquat. Sci. 38:280-296.
Thomann, R.V. 1989. Bioaccumulation model of organic chemical distribution in aquatic food
chains. Environ. Sci. Technol. 23:699-707.
Thomann, R.V. and J.P. Connolly. 1984. Model of PCB in Lake Michigan lake trout food chain.
Environ. Sci. Technol. 18:65-71.
Thomann, R.V., Mueller, J.A., Winfield, R.P. and Huang, C.R. (1991). Model of fate and
accumulation of PCB homologues in Hudson Estuary. Journal of Environmental Engineering,
Vol. 117,No.2.
Thomann, R.V. 1989. Bioaccumulation model of organic chemical distribution in aquatic food
chains. Accepted for publication, Environ. Sci. Technol.
Thomann, R.V., J.P. Connolly, and T.F. Parkerton. 1992. An equilibrium model of organic
chemical accumulation in aquatic food webs with sediment interaction. Environmental
Toxicology and Chemistry 11:615-629.
Thomann, R.V., Mueller, J.A., Winfield, R.P., and Huang, C.R. (1989). Mathematical model of
the long-term behavior of PCBs in the Hudson River estuary. The Hudson River Foundation,
New York, NY.
Thomas, W. A. and McAnally, W.H., Jr. 1990, User's manual for the generalized computer
program system: open-channel flow and sedimentation, TABS-2. US Army Engineer Waterways
Experiment Station, Vicksburg, Mississippi.
Thurston, Kerry A., 1998. "Database Transmittal 10/13/98 (0612.246)." E-mail from Kerry A.
Thurston, O'Brien & Gere Engineers, Inc. to John Haggard, General Electric, regarding GE
database update as of October 12, 1998.
United States Environmental Protection Agency (USEPA). 199la. Proposed technical basis for
establishing sediment quality criteria for nonionic organic chemicals using equilibrium
partitioning. August 1991. Office of Science and Technology, Health and Ecological Criteria
Division, Washington, DC.
United States Environmental Protection Agency (USEPA). 199Ib. Phase 1 Report - Review
Copy Interim Characterization and Evaluation, Hudson River PCB Reassessment RI/FS. USEPA
Work Assignment No. 013-2N84. Prepared by TAMS Consultants, Inc. and Gradient
Corporation. August, 1991
United States Environmental Protection Agency (USEPA). 1992. Final Phase 2 Work Plan and
Sampling Plan, Hudson River PCB Reassessment RI/FS. USEPA Region II, New York. Prepared
by TAMS/Gradient. September, 1992.
115 MCA/Tetra Tech
-------
United States Environmental Protection Agency (USEPA). 1993a. Phase 2B Sampling and
Analysis/Quality Assurance Project Plan, Volume 2 Benthic Invertebrate and Sediment Grab
Sampling, Hudson River PCB Reassessment RI/FS, February 18, 1993. USEPA, Region II, New
York. Prepared by TAMS/Gradient. February, 1993.
United States Environmental Protection Agency (USEPA). 1993b. Great Lakes water quality
initiative technical support document for the procedure to determine bioaccumulation factors. Office
of Water. EPA-822-R-93-009, August.
United States Environmental Protection Agency (USEPA). 1994. Great Lakes Water Quality
Initiative Technical Support Document for the Procedure to Determine Bioaccumulation Factors.
EPA-822-R-94-002. Office of Water, Office of Science and Technology, U.S. Environmental
Protection Agency, Washington, DC.
United States Environmental Protection Agency (USEPA). 1995. Great Lakes Water Quality
Initiative Technical Support Document for the Procedure to Determine Bioaccumulation Factors.
EPA-820-B-95-005. Office of Water, Office of Science and Technology, U.S. Environmental
Protection Agency, Washington, DC.
United States Environmental Protection Agency (USEPA). 1996. Phase 2 Report, Further Site
Characterization and Analysis, Volume 2B - Preliminary Model Calibration Report, Hudson
River PCBs Reassessment RI/FS. Prepared for U.S. Environmental Protection Agency, Region 2,
New York, NY, by Limno Tech, Inc., Menzie-Cura & Associates, Inc., and the Cadmus Group,
Inc.
United States Environmental Protection Agency (USEPA). 1997. Phase 2 Report, Further Site
Characterization and Analysis, Volume 2C - Data Evaluation and Interpretation Report, Hudson
River PCBs Reassessment RI/FS. Prepared for USEPA, Region II, New York. Prepared by
TAMS etal. February, 1997.
United States Environmental Protection Agency (USEPA). 1998a. Phase 2 Report- Further Site
Characterization and Analysis. Volume 2C-A Low Resolution Sediment Coring Report.
Addendum to the Data Evaluation and Interpretation Report, Hudson River PCBs Reassessment
RI/FS. USEPA, Region II, New York. Prepared by TAMS/GradienuTetra Tech. July, 1998.
United States Environmental Protection Agency (USEPA). 1998b. Database for the Hudson
River PCBs Reassessment RI/FS. Release 4.1b (Compact Disk). Prepared for USEPA, Region II
and the US Army Corps of Engineers, Kansas City District. Prepared by TAMS Consultants, Inc.
August, 1998.
United States Environmental Protection Agency (USEPA). 1999a. Phase 2 Report - Further Site
Characterization and Analysis, Volume 2D - Baseline Modeling Report Hudson River PCBs
Reassessment RI/FS. Prepared for USEPA Region 2 and US Army Corps of Engineers, Kansas
City District. Prepared by Limno-Tech, Inc. (LTI), Menzie-Cura & Associates, Inc. (MCA), and
Tetra-Tech, Inc. May, 1999.
16 MCA/Tetra Tech
-------
United States Environmental Protection Agency (USEPA). 1999b. Phase 2 Report - Further Site
Characterization and Analysis, Volume 2E - Baseline Ecological Risk Assessment Hudson River
PCBs Reassessment RI/FS. Prepared for USEPA, Region 2 and the US Army Corps of
Engineers, Kansas City District. Prepared by TAMS/MCA. August, 1999.
Van der Oost, R., H. Heida, and A. Oppenhuizen. 1988. Polychlorinated biphenyl congeners in
sediments, plankton, molluscs, crustaceans, and eel in a freshwater lake: Implications of using
reference chemicals and indicator organisms in bioaccumulation studies. Arch. Environ. Contam.
Toxicol. 17:721-729.
Velleux, M. and D. Endicott. 1994. Development of a mass balance model for estimating PCB
export from the lower Fox River to Green Bay. Journal of Great Lakes Research, Vol. 20, No. 2.
Waldman, J.R. 1988a. 1986 Hudson River striped bass tag recovery program. The Hudson River
Foundation, New York, NY 48 pp + Append.
Waldman, J.R. 1988b. Private Communication
Webb, R.G. and A.G. McCall. 1973. Quantitative PCB standards for electron capture gas
chromatography. Journal of Chromatographic Science, 11: 366-373.
Wells, A.W., J.A. Matousek, and J.B. Hutchinson. 1992. In Estuarine Research in the 1980s: The
Hudson River Environmental Society Seventh Symposium on Hudson River Ecology (Ed: C.L.
Smith) State University of New York Press, pp. 242-264.
Wershaw, R.L., M.J. Fishman, R.R. Grabbe, and L.E. Lowe, ed.s. 1983. Methods for the
Determination of Organic Substances in Water and Fluvial Sediments. USGS Techniques of
Water-Resources Investigations, Book 5, Laboratory Analysis, Chapter A3. Open-File Report
82-1004. USGS, Denver, CO.
Whittle, D.M. and J.D. Fitzsimmons. 1983. The influence of the Niagara River on contaminant
burdens of Lake Ontario biota. /. Great Lakes Res. 9(2): 295-302.
Wood, L.W. G.Y. Rhee, B. Bush and E. Barnard. 1987. Sediment desorption of PCB congeners
and their bio-uptake by dipteran larvae. Water Res. 21:875-884
Woolfolk, M., M. Barta, and G. Drendel. 1994. Modeling the accumulation of PCBs in
largemouth bass from Lake Hartwell, South Carolina. Presented at SETAC.
Young. 1988. A report on striped bass in New York marine water. NYS Marine Fisheries, Stony
Brook, NY.
117 MC/VTetra Tech
-------
THIS PAGE INTENTIONALLY LEFT BALNK
118 MCA/Tetra Tech
------- |