Jump to main content.

Revised Baseline Modeling Report

January 2000

This report presents results and findings from the application of mathematical models for PCB physical/chemical transport and fate, as well as PCB bioaccumulation in the Upper Hudson River. The modeling effort for the Hudson River PCBs site Reassessment has been designed to predict future levels of PCBs in Upper Hudson River sediment, water and fish. This report provides predictions under baseline conditions, that is, without remediation of PCB-contaminated sediment in the Upper Hudson River (equivalent to a No Action scenario). The predicted sediment, water and fish PCB concentrations from the models are used as inputs in the Human Health and Ecological Risk Assessments. Subsequently, the models will be used in the Feasibility Study (the Phase 3 Report) to help evaluate and compare the effectiveness of various remedial scenarios.

The Revised Baseline Modeling Report (RBMR or Revised BMR) incorporates changes to the May 1999 Baseline Modeling Report (BMR) based on public comments and additional analyses, and supercedes the May 1999 report. The Revised BMR consists of four books. Books 1 and 2 are on the transport and fate models, with Book 1 containing the report text and Book 2 containing the corresponding tables, figures and plates. Similarly, Books 3 and 4 are on the bioaccumulation models, with Book 3 containing the report text and Book 4 containing the corresponding tables, figures and plates. Predictions of future PCB concentrations in sediment and water from the transport and fate models are used as input values for the bioaccumulation models. The bioaccumulation models forecast PCB concentrations in various fish species based on these inputs.

Modeling Objectives

The overall goal of the modeling is to develop scientifically credible models capable of answering the following principal questions:

The work presented in this Revised BMR provides information relevant to the first and third questions. Forecasts regarding the potential impacts of various remedial scenarios, thus addressing the second question, will be presented in the Feasibility Study (the Phase 3 Report).

Model Development

A large body of information from site-specific field measurements (documented in Hudson River Database Release 4.1), laboratory experiments and the scientific literature was synthesized within the models to develop the PCB transport and fate and the PCB bioaccumulation models. Data from numerous sources were utilized including USEPA, the New York State Department of Environmental Conservation, the National Oceanic and Atmospheric Administration, the US Geological Survey and the General Electric Company.

The proposed modeling approach and preliminary demonstrations of model outputs were made available for public review in the Preliminary Model Calibration Report (PMCR), which was issued in October 1996. The modeling framework of the PMCR was revised based on a peer review and public comment, as well as the incorporation of additional data. The baseline modeling effort and results were documented in the Baseline Modeling Report (BMR) issued in May 1999. USEPA decided to revise the BMR to reflect changes to the models based on public comment and additional analyses that were conducted. The Revised BMR includes model refinements, additional years of data, longer model forecasts, validation to an independent dataset, and additional model sensitivity analyses. This Revised BMR supercedes the May 1999 BMR.

Transport and Fate Models

HUDTOX - The backbone of the modeling effort is the Upper Hudson River Toxic Chemical Model (HUDTOX). HUDTOX was developed to simulate PCB transport and fate for 40 miles of the Upper Hudson River from Fort Edward to Troy, New York. HUDTOX is a transport and fate model, which is based on the principle of conservation of mass. The fate and transport model simulates PCBs in the water column and sediment bed, but not in fish. It balances inputs, outputs and internal sources and sinks for the Upper Hudson River. Mass balances are constructed first for water, then solids and bottom sediment, and finally PCBs. External inputs of water, solids loads and PCB loads, plus values for many internal model coefficients, were specified from field observations. Once inputs are specified, the remaining internal model parameters are calibrated so that concentrations computed by the model agree with field observations. Model calculations of forecasted PCB concentrations in water and sediment from HUDTOX are used as inputs for the forecasts of the bioaccumulation models (as described in Books 3 and 4).

Depth of Scour Model (DOSM) - The Depth of Scour Model was principally developed to provide spatially-refined information on sediment erosion depths in response to high-flow events such as a 100-year peak flow. The DOSM is a two-dimensional, sediment erosion model that was applied to the Thompson Island Pool. The Thompson Island Pool is characterized by high levels of PCBs in the cohesive sediments. DOSM is linked with a hydrodynamic model that predicts the velocity and shear stress (force of the water acting on the sediment surface) during high flows. There is also a linkage between the DOSM and HUDTOX. Relationships between river flow and cohesive sediment resuspension were developed using the DOSM for a range of flows below the 100-year peak flow. These relationships were used in the HUDTOX model for representing flow-dependent resuspension.

Bioaccumulation Models

Three separate bioaccumulation models were developed in a sequential manner, beginning with a simple, data-driven empirical approach (Bivariate BAF Analysis), followed by a probabilistic food chain model, and ending with a time-varying, mechanistic approach (FISHRAND). The three approaches are complementary, with each progressively more complex model building on the results of the preceding, simpler effort. All three bioaccumulation models are presented in the Revised BMR; however, the FISHRAND model is the final bioaccumulation model that is used to predict future fish PCB body burdens.

Bivariate BAF Analysis - The Bivariate BAF (Bioaccumulation Factor) Analysis is a simple empirical approach that draws on the wealth of historical PCB data for the Hudson River to relate PCB levels in water and sediments (two variables, or "bivariate") to observed PCB levels in fish. This analysis is useful in understanding the relative importance of water and sediment sources on particular species of fish. As this empirical approach does not describe causal relationships, the analysis has limited predictive capabilities and accordingly was not used for forecasts.

Empirical Probabilistic Food Chain Model - The Empirical Probabilistic Food Chain Model is a more sophisticated representation of the steady-state relationships between fish body burdens and 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 food chain relationships among larger fish, smaller fish, and invertebrates in the water column and sediments. The Probabilistic Model provides information on the expected range of uncertainty and variability associated with the estimates of average fish body burdens.

(FISHRAND) Mechanistic Time-Varying Model - The FISHRAND model is based on the peer-reviewed uptake model developed by Gobas (1993 and 1995) and provides a mechanistic, process-based, time-varying representation of PCB bioaccumulation. This is the same form of the model that was used to develop criteria under the Great Lakes Initiative (USEPA, 1995). The FISHRAND model incorporates distributions instead of point estimates for input parameters, and calculates distributions of fish body burdens from which particular point estimates can be obtained, for example, the median, average, or 95th percentile. FISHRAND was used to predict the future fish PCB body burdens for the Human Health and Ecological Risk Assessments.

Model Calibration

The principal HUDTOX application was a long-term historical calibration for a 21-year period from 1977 through 1997. Consistent with the Reassessment principal questions, emphasis was placed on calibration of the model to long-term trends in sediment and water column PCB concentrations. However, a short-term hindcast calibration test was also conducted from 1991 to 1997 to establish model performance for certain individual PCB congeners.

Model applications included mass balances for seven different PCB forms: total PCBs, Tri+, and five individual PCB congeners (BZ#4, BZ#28, BZ#52, BZ#[90+101] and BZ#138). Total PCBs represents the sum of all measured PCB congeners and represents the entire PCB mass. Tri+ represents the sum of the trichloro- through decachlorobiphenyl homologue groups. Use of Tri+ as the historical calibration parameter allows for the comparison of data that were analyzed by congener-specific methods with data analyzed by packed-column methods (that did not separate the various PCBs as well and did not measure many of the mono- and dichlorobiphenyls). Therefore, use of the operationally defined Tri+ term allows for a consistent basis for comparison over the entire period for which historical data were available. Tri+ is also a good representation of the PCBs that bioaccumulate in fish.

The five PCB congeners were selected for model calibration based primarily on their physical-chemical properties and frequencies of detection in environmental samples across different media. These individual congener simulations help provide a better understanding of the environmental processes controlling PCB dynamics in the river by testing the model with PCBs with widely varying properties. BZ#4 is a dichloro congener that represents a final product of PCB dechlorination in the sediments. BZ#28 is a trichloro congener that has similar physical-chemical properties to Tri+. BZ#52 is a tetrachloro congener that was selected because of its resistance to degradation and based on its presence in Aroclor 1242, the main Aroclor used by General Electric at the Hudson River capacitor plants. BZ#[90+101] (a pentachloro congener) and BZ#138 (a hexachloro congener) represent higher-chlorinated congeners that strongly partition to solids in the river and bioaccumulate in fish.

The HUDTOX model calibration strategy can be considered minimal and conservative. It is minimal in that external inputs and internal model parameters were determined independently to the fullest extent possible from site-specific data and only a minimal number of parameters were adjusted during model calibration. It is conservative in that parameters determined through model calibration were held spatially and temporally constant unless there was supporting information to the contrary. Consistent with the Reassessment principal questions, emphasis was placed on calibration to long-term trends in sediment and water column PCB concentrations, not short transient changes or localized variations.

The 21-year historical calibration for Tri+ served as the main development vehicle for the PCB fate and transport model used in the Reassessment. This calibration was successful in reproducing observed long-term trends in water and sediment PCB concentrations over the 21-year period. This was primarily demonstrated through comparisons between model results and available data for long-term Tri+ surface sediment concentrations, in-river solids and Tri+ mass transport at low and high flows, and water column solids and Tri+ concentrations. Many different metrics were used collectively in a "weight of evidence" approach to demonstrate model reliability.

The calibration of the FISHRAND model was conducted by a process known as Bayesian updating. This approach optimizes the agreement between predicted distributions of fish concentrations from the FISHRAND model as compared to empirical distributions based on the data by adjusting three input distributions (percent lipid in fish, total organic carbon in sediment, and the octanol-water partition coefficient or Kow). Initial input distributions (referred to as prior distributions) are specified based on site-specific data and values from the published scientific literature. The model is run and calculates the likelihood of obtaining an output distribution that matches observed measurements given the input distribution. The prior input distributions are then adjusted (within constraints of the data) and these adjusted distributions are referred to as posterior distributions. The focus of the calibration was on the wet weight concentrations (as opposed to the lipid-normalized concentrations) because the wet weight concentrations are generally of primary interest to USEPA and other regulators, the lipid content of any given fish is difficult to predict, and the model predicts fish body burdens on a wet weight basis and then lipid-normalizes. It was determined that, overall, the FISHRAND model predicts wet weight Tri+ PCB fish body burdens to within a factor of two, and typically significantly less than that.

Model Validation

Model validation is the comparison of model output to observed data for a dataset that was not included in the calibration of the model. A HUDTOX model validation was conducted to compare predicted and observed water column concentrations for Tri+ using a dataset acquired in 1998 for the Upper Hudson River by General Electric. Results indicated good agreement at both Thompson Island Dam and Schuylerville over an entire year, spanning a range of environmental conditions in the river. The validation was judged successful and it enhances the credibility of the model as a predictive tool.

Several approaches were used to validate the FISHRAND model. One method was to calibrate FISHRAND for one river mile, and then to run the model for a different river mile. Satisfactory agreement for both river miles implied model validity across locations in the Hudson River. In addition, a calibration was conducted using only part of the available dataset, and then the model results were compared with the remaining portion of the dataset. The posterior distributions obtained using only the partial dataset were compared to the posterior distributions obtained using the full dataset. Finally, the partial-data calibrated model was run for the forecast period and these results compared to the full-data calibrated model results. Good agreement across all three metrics implied confidence in the performance of the model.

Model Forecast

In the Revised BMR, the HUDTOX model was run for a 70-year forecast period from 1998 through 2067 for Tri+. The forecast period was lengthened from the 21-year forecast in the May 1999 BMR for two reasons. First, the fish body burdens attained for the 21-year forecast presented risks and hazards above levels of concern as documented in the risk assessments (i.e., the 21-year forecast was too short to predict when PCB concentrations in fish would decrease below levels of concern). Second, the 70-year forecast period was selected in order to provide exposure concentrations that can be used directly in the Monte Carlo analysis in the Human Health Risk Assessment. Tri+ was simulated because it reflects PCB congeners that bioaccumulate in fish and hence are key to the risk assessment.

In order to conduct forecast simulations with the HUDTOX model, it was necessary to specify future conditions in the Upper Hudson River for flows, solids loads, and upstream Tri+ loads. These model inputs are not easily predicted (similar to predicting the future weather), but reasonable estimates were made based on historical observations and current information regarding PCB loading trends.

The baseline forecast simulation was run for an assumed constant Tri+ concentration of 10 ng/L at the model’s upstream boundary at Fort Edward. This level represented the annual average Tri+ concentration that was observed in 1997 and assumes that there will be no future load increases or reductions from upstream sources. In particular, it also assumes that the PCB migration from the GE Hudson Falls Plant site would not increase or decrease and that there would not be any type of event similar to the releases that occurred with the partial failure of the Allen Mill gate structure in 1991. Recognizing the uncertainty in this upstream load, model sensitivity runs were conducted for an assumed Tri+ concentration of zero (0 ng/L) to represent a lower bound on future loads due to the implementation of remedial measures upstream, and for an assumed concentration of 30 ng/L to reflect increased loads similar to observations in 1998.

Results from 70-year forecast simulations contain inherent uncertainty due to uncertainties in estimating future flow and solids loading conditions. Furthermore, various model input assumptions, while less influential in 21-year simulations, can become more important in 70-year forecast simulations. This uncertainty can be assessed and accounted for in USEPA’s decision making by evaluating predictions across a range of alternate scenarios for these inputs. For this reason, model sensitivity runs were also conducted for three additional hydrologic conditions: plus/minus 50 percent changes in future tributary solids loads, a different assumption for the depth of particle mixing in the surface sediments, and different starting concentrations for Tri+ in the sediments.

Risk-based target levels for fish PCB body burdens have not yet been established. In the Feasibility Study, site-specific target levels to be protective of human health and the environment will be developed from the risk assessments. However, it is beneficial at this time to compare forecasted fish PCB levels against example target levels as a matter of perspective. The target levels used for this analysis provide several concentrations spanning two orders-of-magnitude. Again, these are not endorsements of these values for decision making. Appropriate values will be developed in the Feasibility Study for the site.

Major Findings

The primary objective of the modeling effort is to construct a scientifically credible tool to help in the understanding of PCB transport and fate and bioaccumulation in the Upper Hudson River, and to use that tool for making forecasts of what will happen in the future. As such, one of the major findings was that it was possible to construct models that simulate conditions that match the observed data reasonably well. Consequently, the model predictions can be reliably used to evaluate future ecological and human health risks and to assess the relative time it takes for the river to recover under various remedial scenarios.

There are numerous general observations about the river that are apparent from the mass balance exercises. Some important observations that impact the understanding of the system include:

Beyond the general observations above, the model forecasts provide the following findings regarding PCBs in the Upper Hudson River. It should be noted that the findings below are made based on the evaluation of Tri+, and that some of the findings may differ for other mixtures of PCBs, such as total PCBs or individual congeners.

  1. PCB (Tri+) concentrations in the surface sediment are forecasted to decline at annual rates of approximately 7 to 9 percent over the next two decades, consistent with long-term historical trends.
  2. PCB (Tri+) loads from upstream of the model boundary at Fort Edward control the long-term responses of PCB (Tri+) concentrations in the water column and surface sediments, and accordingly, body burdens in fish.
    • For the first two to three decades of the model forecast, depending on location, the in-place PCB (Tri+) reservoir in the sediments and sediment-water transfer processes control responses of surface sediment concentrations.
    • Water column PCB (Tri+) concentrations are increasingly controlled by the upstream boundary at Fort Edward over the long term. The rate at which water column concentrations approach an asymptote depends upon the assumed magnitude of the upstream boundary load and location within the river.
  3. Forecasted surface sediment PCB (Tri+) concentrations in several localized areas in the Stillwater reach and the Thompson Island Pool increase after 40 to 50 years, despite exponential-type decreases up to that time. These computed increases are due to relatively small annual erosion rates that eventually, over an extended length of time, expose PCB concentrations that were previously at depth.
    • The relative magnitudes of computed increases in surface sediment PCB (Tri+) concentrations are small within the context of long-term trends in historical concentrations.
    • The occurrence, magnitude and timing of these computed increases are dependent on forecast assumptions.
    • It is reasonable to assume that localized erosion occurs within the river, but at scales smaller than the spatial scale of the model. Therefore, the model may not accurately reflect the areal extent of such erosion or its timing.
  4. Results of the 100-year peak flow show that<Results of the 100-year peak flow show that a flood of this magnitude would result in only a small additional increase in sediment erosion beyond what might be expected for a reasonable range of annual peak flows.
    • The small sediment scour depths produced by the 100-year peak flow result in only very small increases in surface sediment PCB (Tri+) concentrations. These increases decline to values in the base forecast simulation (without the 100-year peak flow) in approximately four years.
    • Increases in water column PCB (Tri+) concentrations in response to a 100-year peak flow are very short-lived (on the order of weeks) and decline rapidly after occurrence of the event.
    • The 100-year event causes an increase of less than 30 kg (70 lbs) in cumulative PCB (Tri+) mass loading across the Thompson Island Dam by the end of the first year of the forecast. This increase represents approximately 13 percent of the average annual PCB (Tri+) mass loading across Thompson Island Dam during the 1990’s.
  5. The FISHRAND model results for the 70-year forecasts show that predicted wet weight PCB (Tri+) 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 and approximately an order of magnitude higher for the 10 ng/L upstream boundary condition. Under the 30 ng/L upstream boundary condition, the asymptotic value is approximately a factor of five higher than the 10 ng/L result.
  6. FISHRAND model results show that PCB (Tri+) uptake in fish is predominantly attributable to dietary sources, with a smaller contribution from direct water uptake. Analysis of relative sediment and water contributions within the food chain yielded the following results. 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 sensitive to small changes in water concentration.
  7. The time it takes to attain acceptable target levels in fish tissue is greatly dependent upon the target level selected. Target levels will be selected as part of the Feasibility Study for the site.


The modeling effort for the Reassessment has provided USEPA with valuable insights regarding factors that control transport and fate and bioaccumulation of PCBs in the Upper Hudson River. Forecasted responses of water column and surface sediment PCB (Tri+) concentrations in the Upper Hudson River, as calculated by HUDTOX, are sensitive to changes in hydrology, solids loadings, sediment particle mixing depth and sediment initial conditions. Forecasted responses of fish body burdens using the FISHRAND model are sensitive to changes in lipid content of fish, total organic carbon in sediment, and the octanol-water partitioning coefficient (Kow). The models are useful tools for forecasting future sediment, water and fish PCB concentrations. The forecasts can be reliably used to evaluate future ecological and human health risks and to assess the relative time it takes for the river to recover under various remedial scenarios.

Contact: kluesner.dave@epa.gov


Jump to main content.