Lake Michigan Mass Balance

LAKE MICHIGAN MASS BUDGET/
MASS BALANCE

WORK PLAN

U.S. Environmental Protection Agency
Great Lakes National Program Office

MODELING

BACKGROUND

The USEPA Great Lakes National Program Office has proposed a mass balance approach
to provide a coherent, ecosystem-based evaluation of toxics in Lake Michigan (USEPA,
1993). The Lake Michigan Mass Balance Study (LMMBS) will also study hazardous air
pollutants for the Clean Air Act Amendments’ Great Waters Program. The mass balance
approach, demonstrated in the Green Bay Mass Balance Study (GBMBS), provides a
consistent framework for integrating load estimates, ambient monitoring data, process
research efforts, and modeling, leading to the development of scientifically credible,
predictive cause?effect tools. The primary goal of the mass balance study is to develop a
sound, scientific base of information to guide future toxics load reduction efforts for Lake
Michigan at the State and Federal levels. From this goal, a number of specific objectives
have been identified. Several of the plan's objectives call for identifying and quantifying the
sources of toxics to Lake Michigan, as well as establishing cause-effect relationships and
developing forecasting tools:

1. Determine loading rates for critical pollutants from major source categories (tributaries, atmospheric deposition, contaminated sediments) to establish a baseline loading estimate to gauge future progress, and to better target future load reduction efforts.
2. Predict the environmental benefits (in terms of reducing concentrations) of specific load reduction alternatives for toxic substances, including the time required to realize the benefits.
3. Evaluate the environmental benefits of load reductions for toxic substances expected under existing statutes and regulations and, thereby, determine if there is a need for more stringent, future regulations to realize further benefits.
4. Improve our understanding of how key environmental processes govern the transport, fate, and bioavailability of toxic substances in the ecosystem.

The mass balance project will be based upon the Enhanced Monitoring Program (EMP), a
comprehensive, 1.6-year synoptic survey for selected toxic chemicals in the Lake Michigan
ecosystem. In support of the mass balance study, the Environmental Research
Laboratory-Duluth (ERL-D) Large Lakes Research Station in cooperation with the
Atmospheric Research and Exposure Assessment Laboratory (AREAL), the NOAA Great
Lakes Environmental Research Laboratory (GLERL), and other cooperators, will develop a
suite of integrated mass balance models to simulate the transport, fate and
bioaccumulation of toxic chemicals in Lake Michigan. This work plan describes these
models, the manner in which they will be integrated, the relationship between their
development and the EMP data, and their intended application.

MODELING PURPOSE AND OBJECTIVES: MASS BALANCE APPROACH

Development of effective strategies for toxics management requires a quantitative
understanding of the relationships between sources, inventories, concentrations, and
effects of contaminants in the ecosystem. A mass balance modeling approach is
proposed in this work plan, to address the relationship between sources of toxic chemicals
and concentrations in air, water, sediment, and biota. This approach integrates load
estimation, ambient monitoring and research efforts within a modeling framework that is
compatible with both scientific as well as ecosystem management objectives. The mass
balance approach estimates the magnitude of mass fluxes that constitute the pathways for
toxics transport into and out of the lake, that distribute toxics within the lake water column
and sediment, and that lead to bioaccumulation of the aquatic food web. Based upon these
estimates, the mass balance can determine the rate of change in concentrations and
inventories of toxics as inputs such as atmospheric and tributary loadings are changed, or
other aspects of the system are perturbed. Thus, the mass balance can serve as a useful
tool to estimate or predict the outcome of alternatives under consideration for toxics
management.

More specifically, the modeling efforts associated with the Lake Michigan mass balance
project will meet the following objectives:

1. Provide a consistent framework for integrating load estimates, ambient
   monitoring data, process research efforts, and prior modeling efforts, leading
   to a better understanding of toxic chemical sources by media, transport, fate
   and bioaccumulation in Lake Michigan.
2. Estimate the loading of priority toxics, solids, and nutrients from major
   tributaries to Lake Michigan for the duration of the EMP study.
3. Estimate the atmospheric deposition and air-water exchange of priority
   toxics, including spatial and temporal variability over Lake Michigan.
4. Calibrate and confirm mass balance models for priority toxics using EMP
   data, based upon models for hydrodynamic and sediment transport,
   eutrophication/organic carbon dynamics, toxics transport and fate, and food
   web bioaccumulation.
5. Based upon the mass balance models, evaluate the magnitude and variability
   of toxic chemical fluxes within and between lake compartments, especially
   between the sediment and water column and between the water column and
   the atmosphere.
6. Apply the mass balance models to forecast contaminant concentrations in
   water and sediment throughout Lake Michigan, based upon meteorological
   forcing functions and future loadings based upon load reduction alternatives.
7. Predict the bioaccumulation of persistent toxic chemicals through the food
   web leading to top predator fish (lake trout and coho salmon) for specific fish
   populations in the lake, in order to relate mass balance predictions of water
   and sediment exposure to this significant impaired use.
8. Estimate (quantify) the uncertainty associated with estimates of tributary and
   atmospheric loads of priority toxics, and model predictions of contaminant
   concentrations.
9. Identify and prioritize further monitoring, modeling, and research efforts to (1)
   address additional toxic substances, (2) further reduce uncertainty of
   predictions, (3) establish additional cause-effect linkages, such as
   ecological risk endpoints and feedbacks, and (4) evaluate additional source
   categories, such as non-point sources in the watershed.

The purpose of modeling will be to simulate the transport, fate and bioaccumulation of
four priority toxics in Lake Michigan: PCB congeners, trans?nonachlor (TNC), atrazine,
and total mercury. These toxics are collectively referred to as “contaminants” in this work plan.

MODELING FRAMEWORK

The model design for the Lake Michigan Mass Balance Project is based upon the linked
sub-model approach used in the Green Bay Mass Balance Study, and retains the same
basic models: hydrodynamics, sediment transport, sediment bed dynamics,
eutrophication/ sorbent dynamics, contaminant transport and fate, and food web
bioaccumulation. A schematic representation of the overall mass balance design is
shown in Figure 1. The Lake Michigan submodels will be applied at several different
levels of spatial resolution, and will incorporate predictive hydrodynamic and sediment
transport simulations as the modeling “foundation”. This approach is consistent with
other state-of-the-art ecosystem modeling exercises, such as the Chesapeake Bay
Watershed Model (Linker et al., 1993), which emphasize increasing computational
effort, complexity, and predictive resolution. As discussed below, linkages will also be
established with atmospheric transport and watershed delivery models, to allow
simulation of multimedia toxics transport as well as loads and boundary conditions to the
lake. Ultimately, such linkages will be essential to relate watershed and “airshed”
management to water quality. Descriptions of the lake process, atmospheric and
watershed delivery model frameworks follow.

Figure 1
Overall Mass Balance Model

Lake Process Models

The mass balance for toxics in Lake Michigan will be comprised of linked
hydrodynamic, eutrophication/sorbent dynamics, particle transport, contaminant
transport and transformation, and bioaccumulation simulations. Each of these models
represents significant processes affecting the mass balance for toxic chemicals. The
hydrodynamic model predicts water movements necessary to describe the
3-dimensional transport of dissolved and particulate constituents in the water column.
The eutrophication model describes the production, respiration, grazing and
decomposition of planktonic biomass within the lake. The particle transport model
describes the resuspension, transport and deposition of particulate materials including
sorbent phases necessary to describe the movement of particle-associated
contaminants. The contaminant transport and fate model describes contaminant
partitioning between dissolved and sorbed phases, transfer between media (air, water,
sediment), and biogeochemical transformations. The bioaccumulation model simulates
contaminant accumulation from water and sediments to predator fish via direct exposure
and trophic transfer through benthic and pelagic food webs. Together, these
submodels form an integrated description of toxic chemical cycling in the aquatic
ecosystem, with which to predict the relationship between loadings and concentrations
for contaminants of interest.

Hydrodynamics

The Princeton Ocean Model (POM; Blumberg and Mellor, 1980 and 1987) will be used
to compute three dimensional current fields in the lake. The POM will simulate large-
and medium(km)-scale circulation patterns, vertical stratification and velocity distribution,
seiche, and surface waves. This model will also be used to simulate a thermal balance
for the lake, and will generate turbulent shear stresses for the sediment transport model.
The POM is a primitive equation, numerical hydrodynamic circulation model that predicts
three dimensional water column transport in response to wind stress, temperature,
barometric pressure, and coriolis force. The POM has been demonstrated to accurately
simulate the predominant physics of large water bodies (Blumberg and Mellor, 1983
and 1985; Blumberg and Goodrich, 1990). This model will be used to develop
year-long simulations on a 5-km horizontal grid, with 15 sigma-coordinate vertical levels,
at one-hour intervals for Lake Michigan. Observed and simulated meteorological data
will be used to define model forcing functions. Extensive measurements of
temperature, transmissivity, and current distributions collected in Lake Michigan during
1982-83 will provide the necessary data for model confirmation; measurements of daily
surface temperature (from satellite) and temperature, transmissivity, and current
distributions will also be used to confirm hydrodynamic simulations for 1994-95.

The hydrodynamic model is the appropriate transport foundation for an accurate lake
mass balance model, for a number of reasons. A confirmed hydrodynamic model offers
a credible basis for extrapolating transport, in terms of forecasting the response to
expected and extreme meteorological forcing functions, that is desirable for a mass
balance simulation. The hydrodynamic model results are scaleable to provide transport
predictions at the desired spatial and temporal resolution. This is useful when
considering that the various processes incorporated in the mass balance are not
necessarily modeled at the same scale or resolution, yet all depend upon a consistent
transport simulation. In particular, the sediment and contaminant transport model
described below, requires high resolution simulations of current- and wave-induced
shear stress to predict sediment transport. Hydrodynamic models are also
transportable, with little system-specific parameterization in comparison to traditional
water quality models. A mass balance design based upon hydrodynamic transport is
advantageous, for instance, when considering applying the mass balance model for
Lake Michigan to the other Great Lakes.

Sediment and Contaminant Transport

A 3-D version of the sediment transport model, such as SEDZL, will be used to
simulate the movement of sediment particles in both the water column and sediment
bed, including settling, resuspension, flocculation, transport and deposition. SEDZL will
simulate the significant short- and long-term processes which transport sediment
particles and particle-associated contaminants in the lake. SEDZL will be linked to
hydrodynamic output from the POM, and will be based upon the same 3-D water column
grid. State variables will include 3 particle classes (plankton/biotic solids, cohesive
fine-grained sediment/detritus, and coarse-grained solids) and PCBs. SEDZL will
simulate the 1982-83 and 1994-95 periods for which hydrodynamic forecasts will be
available, as well as intensive confirmation data provided by sediment trap and
radionuclide monitoring. Further confirmation data for 1994-95 will be provided by
remote sensing, transmissometer arrays, and water intake monitoring. Sediment bed
properties, particle resuspension rate parameters, flocculation parameters and settling
properties necessary for the model will be determined by field measurements to be
performed on Lake Michigan sediments, and by results of experiments conducted with
other sediments from the Great Lakes. Allochthonous sediment loadings will be
estimated for tributary export, shoreline erosion, and atmospheric particle deposition.
Autochthonous production will be provided from the eutrophication/sorbent dynamics
model, and input as loadings to the sediment transport model.

The sediment transport model is applied to predict the transport of particles in the lake,
which predominantly carry hydrophobic contaminants from near-shore locations such as
tributary mouths, to deposition zones usually in deep water. The transport of sediment
and associated contaminants is a complex interaction of the properties of sediment
particles and the sediment bed, circulation, bathymetry, and turbulent shear stresses
applied by waves and current. Moving from shore to deep water, regimes of sediment
transport are encountered, resulting in distinct distributions of grain size, bed thickness,
sedimentation rate, and contaminant concentrations in the lake sediments.
Contaminants move along this gradient associated primarily with the fine-grained
sediments, yet their transport is influenced by the entire particle assemblage. In terms
of resuspension and deposition, most sediment transport is associated with the
sequence of short, infrequent events such as storms. SEDZL simulates the interactions
and dynamics of sediment transport, and offers predictive capabilities beyond that
obtainable by a calibrated-transport approach. Advantages include compatibility with the
hydrodynamic simulation, high spatial resolution consistent with the spatial variability of
the resuspension process, and verified process descriptions for the dynamics of
sediment resuspension and deposition under event conditions which are the most
difficult to model. SEDZL predictions have been confirmed mostly in tributary systems;
in large water bodies simulations have been conducted for events, with only limited
confirmation. Thus, significant development is still required for credible application of
SEDZL in the Lake Michigan mass balance model. Sediment and contaminant transport
model predictions will require extensive confirmation against EMP data to ensure model
credibility.

The alternative approach to treating sediment transport is descriptive, where direct
calibration of Total Suspended Solids (TSS) and associated particle tracers is used to
specify settling and resuspension fluxes. The descriptive approach ensures a model
calibration that is consistent with available observations. However, the spatial complexity
and event-responsive nature of sediment transport described above introduce too many
degrees of freedom to allow model calibration to the data being generated by the EMP.
This approach relies entirely upon fitting suspended constituent data, which will be too
sparse (both in space and time) to allow accurate description of sediment transport
fluxes. The second major disadvantage of descriptive transport, is that the resulting
model has no forecasting basis other than replaying the calibration. Attempts to go
beyond the calibration are, in general, weak emulations of predictive transport
approaches.

Eutrophication/Sorbent Dynamics

The eutrophication/sorbent dynamics (ESD) model predicts the production,
transformation and decay of plankton biomass in response to seasonal dynamics of
temperature, light, and nutrient concentrations. In the open lake, living and dead
plankton comprise the majority of suspended particles and generate significant
autochthonous loads of particulate and dissolved organic carbon (POC and DOC) to
which PCBs and other contaminants preferentially partition (Richardson et al., 1983;
DePinto et al., 1993). The ESD model simulates the non-conservative,
seasonally-variable dynamics of the biotic organic carbon pool, which has a significant
influence upon partitioning of HOCs (Dean et al., 1993). Such a model was applied to
simulate the dynamics of organic carbon states in Green Bay as part of the GBMBS
(DePinto et al., 1993). However, a more resolute, multi-class eutrophication model
(Bierman and McIlroy, 1986) will be applied to Lake Michigan, and the linkage between
plankton and organic carbon states will be refined. Model outputs include
autochthonous solids loads (primary production), and transformation and decay rates,
that will be used as inputs for the sediment transport and the contaminant transport and
fate models. The biomass growth rates may also be linked to the plankton
bioconcentration submodel of the food web bioaccumulation model.

The eutrophication/sorbent dynamics model is an important component of the mass
balance model for hydrophobic contaminants, because it simulates the dynamics of a
significant sorbent particle class (phytoplankton) in the water column. The dynamics of
phytoplankton production and loss cannot be adequately described by seasonal EMP
limnological monitoring, which will occur too infrequently to observe major events such
as blooms, assemblage shifts, and die-offs. Furthermore, the ESD model component
will allow forecasting for integrated toxics and nutrient management options, because
mass balances for toxics and nutrients are coupled via eutrophication/sorbent dynamics
processes. Finally, the ESD model is the appropriate framework for inclusion of zebra
mussels in the mass balance model. Zebra mussels, which at high density can impact
the lower food web and alter sediment and contaminant transport, are currently (1994)
infesting Lake Michigan and are reaching high densities in areas of suitable habitat such
as Green Bay.

Contaminant Transport and Fate

The mass balance for toxic chemicals in the lake will be computed in a contaminant
transport and fate (CTF) model which describes contaminant transport, intermedia
exchange, phase distribution, and biogeochemical transformations, in both the water
column and sediments. The CTF model will be calibrated and confirmed for each of the
priority toxics: atrazine, mercury, selected individual and sum of PCB congeners, and
TNC. Mass balance analyses will be performed for each contaminant, to evaluate the
significant source, transport, and loss pathways. Effectiveness of alternative load
reduction scenarios upon reducing toxic chemical concentrations, will also be forecast.
Although calibration and confirmation will be limited to the period of available EMP data,
the CTF model will be required to forecast contaminant concentrations for substantially
longer periods: on the order of 20-50 years. Long simulations are necessary because
of the substantial lag time associated with the chemical concentration response in the
lake to changing loads. The lag time is associated with the residence time of
contaminants in the surficial sediments, which is constrained by confirmation of CTF
model hindcasts for cesium-137 and/or plutonium-239/240. These particle?associated
radionuclides have been demonstrated as important tracers for the long?term transport
of sediments and contaminants in Lake Michigan and the Great Lakes. Because their
loading histories are known with relative certainty, available water and sediment data for
these contaminants are directly useful for model confirmation. Such data are critical to
develop of a model intended to make long?term forecasts, especially since EMP
monitoring will be only 2 years in duration. Intensive sediment trap data collected in
1982-83 (Robbins and Eadie, 1991) and water column measurements from the same
period, will provide further measurements for confirmation of particle transport fluxes.

A schematic diagram of the CTF model as applied for PCBs in Lake Michigan is
presented in Figure 2. Chemical fluxes between model compartments are computed
from advective and dispersive transport of aqueous and particulate contaminant
fractions. The model will describe chemical partitioning between dissolved and
particulate sorbent compartments, including multiple particle types, using an organic
carbon-based equilibrium assumption. Both local equilibrium and first-order kinetic
partitioning process descriptions will be tested. Chemical transformations such as
hydrolysis and biodegradation are modeled as first-order or pseudo first-order
reactions, with daughter chemicals retained in the mass balance as additional state
variables (for atrazine, these include desethylatrazine and deisopropylatrazine). For
mercury, a two-state (organic and inorganic) multiple-sorbent class framework proposed
by Thomann (1993) will be applied.

The CTF model incorporates simulations of other submodels by the following linkages:

Table 1.
Contaminant Transport and Fate Model Linkages 

The CTF model will be linked to hydrodynamic and sediment transport simulations, by
appropriate filtering and averaging of transport fields (Hamrick, 1987; Hall, 1989; Dortch
et al., 1992). Total suspended solids (TSS) and SPCB (sum of congeners) simulations
will be reproduced in both SEDZL and CTF models, providing computational “tracers”
to validate the transport linkages.

The CTF model will be applied at an intermediate (Level 2) scale. In the water column,
segment resolution is defined at a scale compatible with the definition of food web
zones (approximately 20x40 km), with 2-5 vertical layers. In sediments, segmentation
will be based upon deposition regime and contaminant distribution, with 1-cm vertical
resolution. Fine-scale simulations are necessary for accurate predictions of
hydrodynamic and cohesive particle transport as well as accurate simulation of
short-duration event processes. However, the computational cost of fine-scale models
is high and makes long-term (20 to 30 year) simulations infeasible, especially with the
significant number of state variables required for multiple contaminants, sorbent
phases, etc. Resolution at the scale of POM and SEDZL is also not appropriate for the
mass balance objectives of this project. Intermediate scale models have substantially
lower computational cost and have been demonstrated for contaminant transport and
transformation over temporal and spatial scales appropriate for toxics exposure
prediction and linkage to bioaccumulation models (DePinto et al., 1993; Connolly et al.,
1992).

Figure 2
Schematic of Contaminant Transport and Fate Model

Although CTF model compartments are generally well-defined, no single framework
presently available has the capacity to accurately predict all components of CTF
while retaining the aggregate behavior of hydrodynamic and sediment transport
simulations. To develop an appropriate framework for the LMMBS and future lake-wide
analysis and management projects, existing and developmental mass balance water
quality modeling frameworks such as those used for Chesapeake Bay (Cerco and Cole,
1993), Green Bay (Bierman et al., 1992; Velleux et al., 1994), and other projects
(Richards et al., 1993; Katopodes, 1994) will be reviewed. Appropriate features of
these models will be synthesized into a single framework and extended to meet the
requirements of the LMMBS.

Food Web Bioaccumulation

A bioaccumulation model simulates chemical accumulation in the food web in response
to chemical exposure, based upon chemical mass balances for aquatic biota. The
general form of the bioaccumulation equation is well defined, and equates the rate of
change in chemical concentration within a fish (or other aquatic organism) to the sum of
chemical fluxes into and out of the animal. These fluxes include direct uptake of
chemical from water, the flux of chemical into the animal through feeding, and the loss
of chemical due to elimination (desorption and excretion) and dilution due to growth. To
predict bioaccumulation for top predator fish (the modeling objective here), the
bioaccumulation mass balance is repeatedly applied to animals at each trophic level to
simulate chemical biomagnification from primary and secondary producers, through
forage species to top predators. Food web bioaccumulation models have been
successfully applied for PCBs and other HOCs in several large-scale aquatic
ecosystems (Thomann and Connolly, 1984; Connolly and Tonelli, 1985) and, most
recently, for the GBMBS (Connolly et al.,1992). The model developed for that project,
FDCHN, will be adapted for use in Lake Michigan. FDCHN is a time-variable,
population-based age class model, incorporating realistic descriptions of bioenergetic,
trophodynamic, and toxicokinetic processes. The general features of FDCHN are
well-suited to a modeling application such as the Lake Michigan mass balance project.

For Lake Michigan, bioaccumulation of PCB congeners and TNC will be modeled for
lake trout and coho salmon food webs. Food web bioaccumulation will be simulated for
sub-populations of lake trout in three distinct biotic zones. The general structure of the
lake trout food web in Lake Michigan is shown in Figure 3. In each zone, different
food webs support lake trout, including benthic and pelagic food web linkages. Biotic
zones are defined by the approximately 50-mile range of movement of lake trout. The
coho salmon, in comparison, is strictly pelagic. Although the coho food web is simpler,
the bioaccumulation simulation must account for significant migration over the two year
lifetime of this stocked salmonid in Lake Michigan.

It should be recognized that FDCHN, and in fact all current food web bioaccumulation
models, is not predictive in terms of the dynamics of the food web itself. In other words,
the food web structure is described as model input. FDCHN does not predict changing
forage composition, trophic status in response to nutrients, exotic species invasion, or
fisheries management. Yet such factors have been demonstrated to alter food web
structures in the Great Lakes, and these changes have been suggested to affect
bioaccumulation in top predators including salmonids. To address the sensitivity of
bioaccumulation predictions to food web dynamics, the SIMPLE model (Jones,
Koonce, and O’Gorman; 1993), a bioenergetic model for fish population dynamics in
the Great Lakes, will be used to construct scenarios for food web change that will then
be tested in FDCHN. While less satisfactory than an integrated population dynamics
simulation, such testing will demonstrate the sensitivity of bioaccumulation predictions to
food web dynamics in comparison to changes in contaminant concentrations in fish due
to reducing exposure concentrations.

Atrazine bioaccumulation will not be modeled, because it is not expected to accumulate
in biota due to its low hydrophobicity. It is not presently feasible to model
bioaccumulation of mercury because a mass balance for the bioaccumulative fraction
(the methyl species) is beyond present analytical and modeling capabilities. As
identified in Mercury in the Great Lakes: Management and Strategy (Rossmann and
Endicott, 1992), the development of such capabilities must initially take place on small,
constrained ecosystems as opposed to the Great Lakes. This is consistent with the
research approach of Porcella et al. (1992) in developing the EPRI Mercury Cycling
Model, which was based upon data gathered from Little Rock Lake and other bog
seepage lakes in Wisconsin.

A number of FDCHN enhancements will be considered in the Lake Michigan
application. These include incorporating specialized sub-models for phytoplankton
(Swackhamer and Skoglund, 1993) and Diporeia (Landrum et al., 1992), the organisms
at the base of the pelagic and benthic food webs. The bioaccumulation process
formulations of Gobas (1993), Barber et al. (1991), and Sijm et al. (1992) will be
reviewed for possible updating of FDCHN toxicokinetic descriptions. The detailed
bioenergetics model of Hewett and Johnson (1987, 1989), which is currently employed
in simplified form in FDCHN, may also be more fully incorporated in the model.

Figure 3
Simplified Lake Michigan Lake Trout Food Web

Current estimates suggest that atmospheric deposition is the major source of several
contaminants to Lake Michigan, including PCBs (Pearson et. al., 1994), and mercury
(Rossmann, 1994). In addition, net volatilization to the atmosphere may be the
predominant loss mechanism for semi-volatile contaminants such as PCBs from Lake
Michigan (Endicott and Kandt, 1993) as well as Lake Superior (Jeremiason et al.,
1994). Due to the importance of the deposition and exchange of toxics between Lake
Michigan and the atmosphere, air?water fluxes of contaminants must be accurately
predicted. This will be accomplished initially by observation-based
interpolation/extrapolation of atmospheric monitoring data. A longer-term objective will
be to model the deposition and exchange of contaminants by linkage and coupling
between the CTF model and a compatible atmospheric transport model. The Regional
Acid Deposition Model (RADM) will be adapted by the EPA Atmospheric Research and
Exposure Assessment Lab (AREAL) for this application.

Observation-Based

Observation-based interpolation/extrapolation of atmospheric monitoring data will be
used to estimate over-lake wet deposition, dry deposition, and vapor phase contaminant
concentration distributions. These estimates will be based upon: (1) routine monitoring
at 9 land-based sites, (2) ship-board sampling in conjunction with open water
monitoring, and (3) 3 intensive studies focusing on Chicago as an urban source of air
toxics.

Measurements from the Integrated Atmospheric Deposition Network (IADN) and the
Lake Michigan Enhanced Monitoring Project (EMP) will be used to drive the CTF
model. An overview of the procedures to be used for deriving atmospheric loadings
from monitoring data is provided in the Atmospheric Monitoring Overview and Appendix
3 of the Mass Balance Project Work Plan. The Lake Michigan Atmospheric Technical
Workgroup will be responsible for calculating atmospheric loadings. This effort must be
coordinated with the Modeling Workgroup to ensure compatibility with regard to
contaminants of interest, simulation time periods, and spatial scales.

The primary use of observed atmospheric loadings will be to calibrate the CTF model
using the best available information to characterize present conditions. Ambient gas
phase observations above the water surface will be used in the air/water surface
exchange calculations performed by the CTF model.

Atmospheric Transport and Deposition Model

A version of RADM adapted for toxics (the Linear Chemistry Model, LCM) will simulate
transport above the watershed and lake, the partitioning and transformations of
contaminants in the atmosphere, and the significant deposition and exchange
processes with the watershed and lake. Atmospheric transport in RADM is in turn
driven by a meteorological model, which generates prognostic simulations of wind,
temperature, insolation, etc. The atmospheric model will also generalize measurements
of atmospheric deposition and vapor concentrations into fluxes on an appropriate
spatial and temporal resolution. The volatile flux may be a significant mass balance
component for contaminants in both the lake and regional atmosphere. Because
volatile flux is driven by the local concentration (fugacity) gradient between water and air,
contaminant transport and fate models for lake and atmosphere must eventually
approximate or achieve coupled simulations. The LCM will be used to predict the air
component of contaminant transport and fate. This model will be linked and eventually
coupled to the CTF model. LCM will compute transport, dispersion, gas-particle phase
distribution, and chemical transformation of airborne contaminants. Meteorological
model output is used to define wind and temperature fields for transport. Emission
inventory data are used to define contaminant source inputs, although specified
boundary condition data may be used to augment emission inventories. This model
predicts wet deposition, dry deposition, and vertical air phase contaminant concentration
distributions.

The diagnostic and analytic capabilities provided through atmospheric modeling can
complement observation based loading calculations by providing enhanced temporal
and spatial resolution of deposition during time periods consistent with observations.
Although this potential for enhancing resolution of the observed input field is important,
atmospheric modeling provides an objective method of linking atmospheric sources
directly to watershed/water body impacts. Consequently, the atmospheric model
should be a valuable tool in the regulatory decision-making process for assessing the
aquatic impacts due to modifying emission releases in future or past scenarios. The
role of atmospheric modeling and plans for model deployment are discussed further in
Section 13.

Air/Water Linkage and Coupling

The first stage of air process model development for the LMMBS is to link the RADM to
the CTF model. The linkage outputs are wet and dry deposition contaminant fluxes and
near surface atmospheric concentrations. The output fluxes and concentrations will be
used to define input atmospheric loads and the gradient for gas exchange for the CTF
model. Linkage can also occur in the other direction, where volatilization is treated as a
source of contaminants to RADM.
Initially, the models will be linked, with one- and two-way transfer of flux output between
RADM and the lake process models. The final goal is model coupling; the models will
run simultaneously to simulate the bi-directional transfer and feedback of contaminant
mass balances for air and water. Coupling is a dynamic, two-way process between the
atmosphere and water surface. In this case, volatile exchange (volatilization or
absorption) is computed based on conditions in both the atmosphere and water column.
For both linkage and coupling, atmospheric and lake process inputs/outputs will be
defined on compatible spatial and temporal scales.

Watershed Delivery

Transport and fate frameworks may be applied to predict the multimedia delivery of
toxics from the watershed to the lake. While contaminant loadings from major tributaries
are being monitored as part of the LMMBS, these data alone may not be sufficient to
accurately define contaminant inputs from the watersheds, tributaries, and harbors that
adjoin the lake. Furthermore, quantifying tributary loads based upon monitoring at the
river mouth does not identify sources of toxic chemicals. For instance, atmospheric
deposition to the watershed will indirectly contribute to tributary loading. Depending
upon the actual source, toxics loading from the watershed may or may not decline over
time without action, respond to meteorology, hydrology, or land use change. Modeling
these significant loads would produce more complete and accurate load estimates and
allow more realistic long-term forecasting ability.

While such modeling capability is important for forecasting purposes, this development
should be addressed separately due to the difficulty of managing such efforts within a
project of this scope and duration. Development of watershed delivery models is
distinct from the lake mass balance model development, because these models
simulate toxics transport and fate at fundamentally different scales and have unique data
requirements. Furthermore, it is not clear that watershed simulation on this scale is
feasible at this time. Results of the LMMBS will be useful for identifying specific toxics
and watersheds to prioritize for watershed delivery modeling, based upon the
magnitude of tributary loading estimates.

MODEL RESOLUTION

Model resolution is the spatial and temporal scale of predictions, as well as the
definitions of model state variables. While factors such as data availability, model
sophistication, and computer resources constrain resolution to a degree, different levels
of model resolution are possible and, are in fact, necessary. Three "levels" of spatial
resolution, indicated by the segmentation grid of the lake surface, are illustrated in
Figure 4. Level 1 is resolved at the scale of lake basins (characteristic length, L= 150
km), with an associated seasonal temporal resolution. This is a screening?level model
resolution used in MICHTOX. Level 2 is resolved at a regional scale defined by food
webs (L= 40 km) including gross resolution of the nearshore and offshore regions;
temporal resolution is weekly-to-monthly. This resolution is roughly comparable to that
achieved by models developed in the Green Bay Mass Balance study. Level 3 is a
hydrodynamic scale resolution (L= 5 km), with associated daily temporal resolution.
Level 3 is scaled to resolve and predict particle transport processes as well as
hydrodynamic transport.

Although LaMP and Great Waters Program objectives are "lake?wide", these
emphasize biotic impairments occurring primarily in localized, nearshore regions. LaMP
objectives also require that the transport of contaminants from tributaries and other
near-shore sources to the open lake be resolved. Therefore, the Level 1 model is not
adequate for the study objectives. Level 2 resolution is adequate for most modeling
objectives, but not for resolution of significant hydrodynamic and sediment transport
events. Level 3 resolution is required for accurate hydrodynamic and sediment
transport modeling and is desirable for predicting nearshore gradients, especially those
formed by transients such as thermal bars, upwelling, and storm?induced resuspension,
as well as more persistent features such as tributary plumes, thermal stratification, and
the benthic nepheloid layer. Level 3 transport resolution would also be valuable in
relating toxics loading from the 10 AOCs adjoining Lake Michigan, which must be
addressed by the Remedial Action Plan (RAP) process, to the LaMP via the LMMBS.

The modeling design for the LMMBS will be based upon the development of several
submodels, at two levels of resolution. The CTF model will be resolved at a level
comparable to Level 2; the eutrophication model will be resolved at the same level.
Because the CTF will be linked to atmospheric fate and transport model predictions, the
two will share the Level 2 resolution at the Lake Michigan surface. The POM and SEDZL
models will be Level 3 resolution. Results of these transport models will be spatially and
temporally averaged prior to coupling to the CTF model. The rationale for specifying
different resolutions is that hydrodynamic and predictive sediment transport models
demand a Level 3 resolution, and these models offer the best capability for transport
simulation and forecasting. A lower resolution is specified for CTF and ESD because
these models have been demonstrated at this resolution, and the need for Level 3 toxics
resolution is not clear.

Figure 4
Spatial Segmentation of Mass Balance Model

GOALS FOR ACCURACY

The stated goal for model accuracy is prediction of lakewide average concentrations of
toxics in water (volume-weighted average), surficial sediment (spatial average), and top
predator fish (average fish in each biota zone) within a factor of two of the average
concentrations based upon monitoring data. To achieve this model accuracy, loadings
and contaminant mass in each compartment must be determined to within 25% of the
actual lakewide, annual average value. Approximately 20% of the samples for toxics
analyses should be replicates, as a basis for estimating measurement variability. (In
this context, replication refers to multiple observations per model segment and
sampling interval). In addition, 75% of loading and ambient samples in all compartments
must be quantified for each contaminant (completeness). These data quality objectives
are based upon expert opinion, and experience gained in the GBMBS. Failure of the
EMP to achieve these goals will degrade the accuracy of the mass balance and model
predictions.

ANALYSIS OF UNCERTAINTY

It should be recognized that model accuracy refers to a comparison of model
predictions to data collected during the EMP. In a forecasting application, the accuracy
of model predictions will degrade over time. In either case, parameterization error is a
significant source of model prediction uncertainty. To evaluate and quantify the effects
of parameterization error, uncertainty analysis will be performed for selected model
simulations. The parameter variance-covariance estimation procedure of Di Toro and
Parkerton (1993) will be applied to estimate data, parameter, and model error
components. With these estimates, confidence intervals for model predictions will be
generated using Monte Carlo/Latin Hypercube simulation. Uncertainty analysis will also
provide a check on the quality of model parameterization and calibration, via the
estimation of parameter errors, which will be applied periodically during model
development.

LONG TERM SIMULATIONS

Long term simulations will include both hindcast and forecast applications. CTF
forecasts will be performed to determine time to steady state, for both continuing and
discontinued loads. Forecasts will also be run to evaluate reductions in exposure
concentrations resulting from elimination of tributary and/or atmospheric loading. These
forecasts will be propagated through the food web bioaccumulation model for PCBs
and TNC, to estimate time for sport fish contaminant concentrations to decline below
criteria limits. As described previously, SIMPLE model scenarios will be used to test
the sensitivity of long-term bioaccumulation predictions to food web dynamics. Based
upon the results of long term simulations, graphs will be developed to illustrate the
fundamental loading-concentration relationships, for both transient and steady state
conditions.

SCHEDULE

A two year project period is proposed for modeling, with model development coincident
with data collection. However, the schedule for completion of model development and
applications must be contingent upon availability of data from the Mass Balance study,
because many aspects of model development cannot proceed without data. In other
words, model final reports will be completed two years after receipt of all data identified
above. Delays in data analyses and reporting will cause equal delays in modeling.

MODEL COMPONENTS AND WORK ELEMENT DESCRIPTIONS

INPUTS

TRIBUTARY LOADS

Background

Tributaries discharging to Lake Michigan are a major source of nutrients, conservative
ions (IJC, 1987), and PCBs (Marti and Armstrong, 1990). Therefore, estimates of
contaminant loads from the tributaries will be an important component of the mass
balance model. Tributary load estimates of critical pollutants that are not part of the
mass balance modeling effort will be measured along with mass balance model
parameters. These critical pollutant loads will provide Lake Michigan environmental
managers with information necessary to set priorities for load reduction activities.

The objectives of the tributary monitoring are:

1. to identify relative loading rates of critical pollutants from major tributaries to
   the Lake Michigan basin in order to better target future load reduction and
   remedial efforts; and
2. to compare tributary loading rates to other media (atmospheric deposition and
   contaminated sediments) in order to better target future load reduction
   efforts and to establish a baseline loading estimate to gauge future
   progress.

Pollutant loads from tributaries must be accurately and precisely determined in order to:
1) quantify the contaminant loads from each tributary; 2) prioritize tributaries for potential
remediation based on contaminant load, and; 3) provide an estimate of the total
contaminant load from tributaries for comparison with loads from atmosphere and open
lake sediments. In order to address the study objectives, the tributary monitoring plan
has been designed to obtain load estimates of target compounds to within +/- 25 to 30
percent of the actual loads.

The tributary monitoring program is intended to assess the contribution of a number of
critical pollutants to Lake Michigan from the major tributaries. The critical pollutants were
identified in the draft Lake Michigan Lakewide Management Plan for toxic pollutants
(LaMP) and are listed in Appendix 2 of the Mass Balance Work Plan. Achieving the
objectives of the tributary monitoring plan will address the needs of the mass balance
model and the Lake Michigan LaMP as driven by the federal Clean Water Act and the
Federal Clean Air Act Amendments.

This study will not provide data on the specific sources (pipes, nonpoint, sediment, etc.)
which contribute to a tributary's load: attempts to answer that question are beyond its
scope. However, additional source identification work will occur through the Lake
Michigan LaMP process. Any additional source identification work within the tributaries
could build upon the Mass Balance Model database.

Sampling Design

Detailed Quality Assurance Project Plans (QAPjP) outlining sampling and analytical
procedures have been developed and approved. These QAPjPs are available upon
request. However, a brief overview of the sample design and sampling methods is
provided below.

With the exception of a study by Marti and Armstrong (1990) for PCBs in the early
1980's, very little work has been done to estimate organics and metals loads from Lake
Michigan tributaries. It is, therefore, necessary to use data from other media (e.g.
contaminants in resident fish) to determine which tributaries are potential sources of the
target contaminants. In addition, the use surrogate parameters such as suspended
solids and flow is necessary to develop a sampling scheme necessary to meet the
objective of monitoring the loadings with an accuracy of +/- 25 to 30 percent.

The tributaries in Table 2, with the exception of the Pere Marquette, were selected
because of elevated concentrations of one or more of the target contaminants in
resident fish collected in 1981-82 (De Vault, 1985; USEPA unpublished data). The
Pere Marquette River was selected because it has a fairly large and pristine watershed.
Samples collected from the Pere Marquette River could be used to estimate loads from
significant portions of the Lake Michigan watershed that will not be monitored.

The tributary samples will be collected by three crews lead by the United States
Geological Survey (U.S.G.S). One crew will be based in Madison, Wisconsin and
collect samples from the Milwaukee, Sheboygan, Fox and Menominee Rivers. A
second crew will be based in Grayling, Michigan and collect samples from the
Muskegon, Pere Marquette and Manistique Rivers. The third crew will be based in
Lansing, Michigan and collect samples from the Grand, Kalamazoo, St. Joseph and
Grand Calumet Rivers.

Sampling sites will be located as far downstream as is practical to monitor the
accumulated point and nonpoint source loads (Figure 5). Flow will be monitored
continuously at each of the sites. Acoustic velocity meters (AVMs) will be used to
monitor flow reversals at sites that are impacted by seiches. Continuous turbidity
monitoring and automated suspended solids sampling will be employed to assess
particulate loads from each tributary.

Each sample collected for analysis of organic pollutants will consist of separate
samples for dissolved (<0.7 microns) and particulate (>0.7 microns) organics. Analysis
of non-polar organic samples collected during pilot work at four tributaries has indicated
that quarter point sampling of the tributaries is appropriate. Quarter point sampling
includes preparing composite samples of subsamples collected at 0.2 and 0.8 of the
river depth at three locations in a cross sectional transect. The three points on the
transect will be located at 0.25, 0.5 and 0.75 the length of the transect. Non-polar
organic samples will be filtered through Whatman GF/F filters in a pentaplate filter. The
filters will be used to analyze contaminants in the particulate phase. Filtered water will
be passed through XAD2 resin columns to extract the dissolved contaminant fraction.
Total sample volume will range from 80 to 160 liters, depending on expected
contaminant concentrations at the sites and logistical constraints faced by field crews.

Atrazine samples will be collected using the same quarter point sampling methods.
Samples will be collected using carbopak resin cartridges. Atrazine samples will be
collected between April 1 and October 31, 1995, coinciding with the normal atrazine
application period.

Metals sampling will include collection of a sample for total metal analysis and a filtered
sample for analysis of dissolved (<0.45 microns) metals. Analysis of samples collected
during pilot monitoring at four tributaries indicated that sampling at two depths at the
centroid of the tributary is appropriate. Samples will be collected at 0.2 and 0.8 the
depth of the centroid.

Due to the hydrophobic nature of the nonpolar organic critical pollutants, we assume that
they will behave similarly to suspended sediment and be event responsive. Therefore,
the tributary monitoring plan was designed to focus sampling effort on high flow events.
Polar organics such as atrazine may also respond to high flow events, mainly as runoff
from agricultural lands, directly entering tributaries. Loads of herbicides calculated for
tributaries to Lake Erie indicate event responsiveness and seasonal dependence
(Baker and Richards 1989). At present, the behavior of mercury and other metals on
the critical pollutant list during a precipitation event is unknown.

The flow variability of each tributary was used to predict the level of sampling required to
achieve a load estimate with the given level of accuracy and precision. The Lake
Michigan tributaries targeted for sampling fall into three categories of flow variability as
described by Richards (1990): super stable, stable, and variable. The level of sampling
required increases from super stable to variable. Table 2 indicates the classification of
selected tributaries by Richards. Work done by Dolan (1981) for phosphorus and Day
(1989) for several parameters indicates that the number of samples required from most
Michigan tributaries (Grand, Pere Marquette, St. Joseph, Muskegon) to determine loads
with 95 percent confidence levels +/- 20% to 30% would be 20 to 30 per year. Based
on the suspended solids and nutrient loading work the estimated sample sizes
necessary to calculate critical pollutant load with the required precision and accuracy
range from 16 to 45 (Table 2). Super stable tributaries will be sampled 16 times, 26
samples will be collected at the stable tributaries and 45 samples will be collected at the
variable tributaries. The only exception to this sampling strategy is the Grand River,
where 36 samples will be taken. The potentially large load of contaminants (due to high
flow volume) from the Grand River warrants the additional effort.

The two tributaries to Green Bay, which deliver the largest load of contaminants to the
Bay, namely the Fox and Menominee Rivers, will be monitored at a frequency of 26
samples per year. However, sedimentation, volatilization, and other processes may
prevent pollutant loads from Green Bay tributaries from reaching Lake Michigan. The
Green Bay Mass Balance Model will be used, along with monitored boundary conditions
to estimate the pollutant load from Green Bay to Lake Michigan.

Approximately two-thirds of the samples will be collected during high flow events. High
flow events have been defined in advance to include any event that exceeds the upper
twentieth percentile of flow based on historical flow records maintained by the U.S.G.S.
The high flow monitoring frequency has been predicted based on the expected number
of high flow days in an average year and the estimated number of samples for each
tributary. The estimated high flow sampling frequency will range from one sample every
6.5 days for the super stable Pere Marquette River to one high flow sample every 2.5
days from the variable Milwaukee and Sheboygan Rivers (Table 3). These high flow
sampling frequencies were estimated to provide guidance to the field crews charged
with collecting samples. However, the field crews have the discretion to temporarily
alter sampling frequencies in order to respond to any unique situations that may occur.
The estimated total number of samples to be collected is 314. An additional 10 percent
for quality assurance will bring the total number of samples to 345. However, the
sampling guidance outlined above will allow crews flexibility to collect more samples if
the project period is unusually wet and fewer samples if the project period is unusually
dry. Low flow samples will be scheduled and collected during base flow periods after
the sampling crews have determined that the sampling locations are not being
influenced by seiches.

Sample collection on the Grand Calumet River will be scheduled in advance and not
based on flow conditions in the river. Industrial discharges contribute the majority of
flow to the Grand Calumet River and effectively stabilize the flow hydrograph at the
mouth. Scheduled sampling runs are preferred at the Grand Calumet River since the
flow is stable and scheduled sampling runs are logistically easier to plan and implement
than event monitoring strategies.

These sample numbers are estimates, based on optimizing crew availability and
logistics, weather conditions, and government funding and quality assurance review.
However, several of these factors combined during water year 1994 to delay and
hamper the collection of the expected samples. In order to provide accurate and
precise load estimates for a complete year, tributary sampling has been extended
through October 1995. The sampling intensity for the one year period ending in 1995
will be the same as our initial estimate for 1994. That is, the sample numbers for that
year will be those listed in the following tables. The samples collected up to October,
1994 will be analyzed and will provide less precise load estimates for that period.

Table 3
Estimated sample volumes, sample sizes and sampling frequency.

Figure 5
Tributary Sampling Locations for Loading Estimates

Load calculation methods are presented in detail in the QAPjP. However, a general
overview is presented below.

The load will be calculated using short term averages (5 to 15 minutes) of flow volume
and direction for dissolved phase contaminants. Along with flow measures, short term
averages of turbidity will be utilized for particulate load calculations. In order to
determine the relationship between turbidity and suspended solids concentration, an
automated, ISCO sampler will be programmed to take three water samples per day.
These water samples will be analyzed for suspended solids concentration, and linear
regressions of suspended solids versus turbidity measurements made at sample
collection times will be developed, stratified by season and flow. Continuous turbidity
monitoring and suspended solids sampling will be conducted at a single location in
each tributary. In order to establish the representativeness of the locations,
cross-sectional samples will be taken for suspended solids during regular sampling
visits, and compared with values obtained from ISCO samples.

Point Source Loadings

Tributary monitoring sites have been selected, purposely, to be downstream of most
major point source discharges. The Great Lakes States are currently evaluating the
potential contribution of point sources on Great Lakes tributaries as well as point source
contributions direct to Lake Michigan. The approach taken is to utilize any available
concentration measurements for the contaminant of interest sampled from the point
sources. Where contaminant concentrations are below the limit of detection (LOD) a
range of estimates using the LOD, one-half the LOD, and zero will be evaluated against
estimated total tributary loads. Point source loadings are an important component of
watershed models, which may follow the mass balance modeling effort.

Research Issues

An area of research being planned is the use of automated Infiltrex samplers for
continuous monitoring of toxic organic compounds. The samplers are self-contained,
employing a pump and in-line filter cartridge and XAD2 resin column (small), controlled
by a programmable microprocessor. As part of the regular sampling program, Infiltrex
samplers will be evaluated against quarter point sampling to determine the
representativeness of samples taken at a single point. There is some potentially cost
saving in using these samplers as an alternative to sampling crews on standby for
sampling during rain events. The automated samplers will be installed in selected tribs
in spring 1995.

ATMOSPHERIC MONITORING OVERVIEW

Introduction

Atmospheric deposition has been shown to be a significant source of target organic
compounds to the Great Lakes, particularly the upper lakes, Superior, Michigan and
Huron (Strachan and Eisenreich, 1988; Eisenreich and Strachan, 1992). Atmospheric
monitoring for the Lake Michigan Mass Balance will be conducted to assess the
contribution of atmospheric deposition and exchange to the concentration of toxic
contaminants in the lake. The atmospheric data set will be used to calculate the
atmospheric load to the system and to calibrate air models linked with the mass balance
water models.

Contaminants are removed from the atmosphere as wet and dry deposition and
exchanged across the air-water interface through vapor absorption and volatilization. To
address these processes, the atmospheric monitoring to be conducted on Lake
Michigan consists of several components: approximately one and one-half year of
routine land-based monitoring, and special research/monitoring studies necessary for
the mass balance. The special studies include the following: intensive seasonal
monitoring and research studies off of Chicago; over-water atmospheric monitoring
from the R/V Lake Guardian during intensives and open-water surveys; and mercury
monitoring at four land-based sites for the 1.5 year period of the loading study and
extensive land-based and over-water monitoring during the intensive studies.

These studies have been designed to:

• Assess the impact of the Chicago urban area on atmospheric
  deposition and exchange with Lake Michigan including
  categorization of major urban source categories.
• Compare over-water and land-based sites to assess whether
  land-based sites are representative of the bulk deposition to
  the lake surface.
• Estimate the air-water exchange of contaminants including
  seasonal direction and magnitude.
• Improve estimates of dry deposition including the large
  particle contribution from urban areas.

Relatively little is known of the spatial and temporal variability of atmospheric
concentrations of the target compounds over Lake Michigan. It is therefore not
possible to design the atmospheric network to assess loads with predetermined
accuracy and precision. Measurement uncertainty (sampling and analytical) is
minimized through the use of the best sampling and analytic techniques available. For
modeling purposes, the goal of the data collection is a combined sampling and
analytical uncertainty of + 20-30% at 90% confidence.

Parameters

The parameters for the atmospheric component of the mass balance are those
previously identified: congener-level PCBs, trans-nonachlor, atrazine, and mercury on a
research basis. The longer list of parameters in Appendix 2 will also be monitored for
the atmospheric loading study for Lake Michigan and, in some cases, to provide
supplemental information for the mass balance. Data collection for the loading study list
of parameters will allow estimates of the ratio of the atmospheric load (wet and dry
deposition) to the total load (tributaries and atmosphere). Meteorological data
collected includes air temperature, wind speed, wind direction, relative humidity, and
solar radiation. Additional research and ancillary data will also be collected for the
special studies.

Routine Monitoring at Land-based Sites

Data collected during the routine monitoring portion of the mass balance will be used to
estimate annual and seasonal loadings and calibrate deposition models. Routine
monitoring will be conducted at nine land-based sites for the period beginning February
1994 and ending October 1995 (Figure 6). The number and location of the sites
were chosen at workshops and through discussions with experts working on
atmospheric deposition in the Great Lakes (USEPA, 1992a and b). Unpublished data
(Steve Eisenreich personal communication, 1992) indicate that the atmospheric
concentrations of PCBs exhibit a strong gradient over Lake Michigan, with
concentrations in the southern portion of the lake approximately 3 to 5 times that in the
north. A higher density of sites is located in the southern portion of the lake due to the
higher contaminant concentration and load variability attributable to urban areas.

The site classifications (urban/urban-influence, rural/background and remote)
correspond to proposed sampling frequency and are identified in Table 4. The urban
site is located at Illinois Institute of Technology in Chicago, Illinois, with a background
site located upwind at Bondville, Illinois to monitor the impact of contaminants carried to
the Chicago urban area from outside the region (i.e., St. Louis area). Duplicate
samplers will be located at the IADN and/or IIT sites. The possible inclusion of a
routine site located on a NOAA buoy is being explored.

Table 4
Atmospheric Monitoring Sites and Sampling Frequency

Figure 6
Land-based and Intensive Atmospheric Sampling Sites

Particle-phase, gas-phase and precipitation samples will be collected at each routine site
according to the frequencies in Table 5. A modified Andersen high-volume sampler (flow
rate of 20 cfm) with GF/F glass fiber filter and XAD-2 resin cartridge will be used to collect
the particulate and gas-phase SVOCs, respectively. The average sampling frequency for
particulate and gas-phase concentrations is once every six days based on a range of every
twelfth day for rural sites to every three days for urban sites. Modified hi-vol samples will be
composited on a monthly basis at all sites during the routine monitoring. This sampling
scheme is intended to address the variability expected in urban areas without increasing the
required laboratory capacity. Precipitation is collected as an integrated monthly sample
using a modified MIC with XAD-2 resin column which is analyzed for the SVOCs.

During the period of routine sampling, dry deposition plates will be located at those sites
also identified for the mercury studies. Each dry deposition sampler consists of two plates
pointed into the prevailing wind by a wind vane. Four greased Mylar strips are located on
each of the plates for a total of eight per sampler. The samplers will collect integrated
monthly samples. One strip from each site will be analyzed for trace metals. Strips are
weighed before and after sampling to determine the total mass collected. The compositing
and analysis schemes for the other strips has not yet been determined.

Additional parameters are collected at the routine sites either as support for the mass
balance or the loading study. A dichotomous sampler is used to collect coarse and fine
particulate for trace metal analysis by XRF. Precipitation is collected weekly in a modified
Aerochem Metric sampler with a Teflon coated sampling train and analyzed for trace metals
using ICP/MS. Nutrients and major ions are collected weekly at the GLAD sites only.
Meteorological data is recorded continuously and averaged for hourly values at all
land-based sites.

Specific details of the sample collection and analysis will be covered in the Quality
Assurance Project Plans (QAPjPs) and Standard Operating Procedures (SOPs) developed
for the mass balance and loading studies. The QAPjP and SOPs will be based on the
procedures and quality assurance/quality control measures that were used in the Green
Bay Mass Balance Study (Swackhamer, 1988) and those which are currently used for the
Integrated Atmospheric Deposition Network (IADN) (Sweet, 1992). XAD-2 resin is used to
concentrate the vapor phase and dissolved (precipitation) phase SVOCs. Whatman GF/F
filters are used to collect particle phase SVOCs. Each sample is analyzed for all the
loading study parameters including PCBs, trans-nonachlor and atrazine. Following
extraction, a portion of the sample is analyzed for atrazine and its two major degradation
products, de-ethylatrazine and de-isopropylatrazine, using gas chromatography/mass
spectrometry (GC/MS) with selected ion monitoring (SIM). The remaining sample is separated into two fractions with silica column chromatography. The first fraction is analyzed by gas
chromatography/electron capture detection (GC/ECD) for PCBs, DDE, HCB and aldrin,
while the second fraction is analyzed by GC/MS for the PAHs, trans-nonachlor, chlordane,
dieldrin, DDT and DDD.

Table 5
Atmospheric Monitoring Sampling Frequency

Atmospheric Component of the Open-water Surveys

Air and water samples above and below the air-water interface will be collected
simultaneously to assess the volatilization of HOCs. This sampling component is similar to
that used for the volatilization study conducted as part of the Green Bay Mass Balance
(Achman, 1993; Hornbuckle 1993). A modified high-volume sampler will be mounted on
the bow of the Lake Guardian for collection of 12-hour air samples (approximately 400
m3). Concurrent water samples are collected as discussed in the plan's section on
Open-Water sampling. Air and water temperature, wind speed and direction, barometric
pressure and wave height are also recorded while on station. Depending on the sea
conditions, the ship will be at anchor while on station to ensure that prevailing winds are
from the bow of the ship and to minimize contamination from the ship's exhaust.
Alternately, the sampler will be operated only when wind is <60° off of the bow. All station
locations have not yet been selected but include the master stations for the open-water
survey as a minimum. The magnitude and direction of the flux is estimated by comparison
of the vapor-phase air and dissolved water concentrations with the expected equilibrium
values as discussed in the section on atmospheric loading calculations.

We will evaluate the representativeness of land-based sites as surrogates for over-water
measurements. Land-based samples (hi-vol and dichot) will be collected concurrently with
over-water samples (hi-vol and dichot) when the ship is on a station near to that land site.
This will allow for the comparison of over-water and over-land samples and an assessment
of the representativeness of land-based sites.

During the survey, event-based precipitation samples for SVOCs will be collected using a 1
m2 steel funnel draining to an XAD-2 resin column. This is a modification of the Baker
sampler, which allows for manual operation.

Intensive Study

The Lake Michigan Urban Air Toxics Study (LMUATs) indicated that the concentrations of
several contaminants were significantly higher in the Chicago urban area than at sites
upwind (Kankakee, IL) and downwind (South Haven, MI) (Keeler, 1993) In addition, a study
of dry depositional flux of PCBs indicated that the flux from the Chicago urban area may be
up to three orders of magnitude higher than that of nonurban areas. (Holson, 1991) The
intensive studies are designed to further assess the impact of the Chicago Urban area on
the atmospheric deposition to the lake, to address process oriented research issues, and
to provide data in support of source apportionment and trajectory modeling.

Three intensive studies were conducted: spring 1994, summer 1994 and winter 1995.
Monitoring locations include three land-based sites: IIT (urban), South Haven (downwind),
and Champaign/Bondville (upwind) and one over-water site, the R/V Lake Guardian
approximately 5 miles off of Chicago. During each intensive, the R/V Lake Guardian wase
used for a period of one to two weeks for frequent sampling. The land-based sites
operated for several additional days on either side of the Lake Guardian operations,
resulting in approximately three to four weeks of intensive sampling at land-based sites.
These sampling periods are to provide information to track plumes/events over and across
the lake.

The monitoring equipment included versions of that used at the routine monitoring sites and
additional equipment for research studies and source apportionment analysis. Precipitation
was sampled for SVOC and trace metals on a daily (24-integrated) basis, as warranted.
Vapor and particulate phase SVOCs were collected, at a minimum, as two 12-hour
samples each day of the intensive. Two 12-hour integrated aerosol samples (coarse and
fine) were collected each day for trace metal analysis. Meteorological data and
dichotomous sampler data were used to select those samples to be analyzed and to define
any compositing scheme which may be employed for the intensive studies. Dry deposition
plates and a micro-orifice impactor/Noll rotary impactor combination collected 24-hour
integrated samples daily. The latter equipment is included to address the impact of large
particle deposition collecting size segregated aerosol up to 150 mm. Mercury speciation
was determined in precipitation, vapor, aerosol less than 2.5 microns, and total aerosol.
Fine particulate samplers for carbon (elemental and volatile), VOC canisters, and annular
denuders for acid gases were used during the intensives for source apportionment
analysis. Open water samples was collected to address exchange at the air-water
interface. Additional research was conducted to address the research issues discussed
below. The intensive studies were coordinated with the open-water surveys so that during
the two weeks following or preceding the intensives, atmospheric monitoring were
conducted aboard the R/V Lake Guardian during the open water surveys.

Water column data collected in 1976-77 indicate strong gradients off Chicago for several
conventional water quality parameters (Rockwell, et. al., 1980). As there is no consistent
hydrologic connection between Chicago and Lake Michigan, the origin is likely
atmospheric. Sediment traps radiating from Chicago are also proposed to monitor the
impact of atmospheric deposition from the Chicago Urban area. The details of this
sediment monitoring are discussed in the sediment section of the study plan.

Loading Calculations

Atmospheric deposition and exchange with a lake, which includes wet and dry deposition,
net gas transfer, resuspension from the lake and the atmospheric component of the
tributary contribution, may be expressed as the equation in Appendix 3.

The atmospheric component of the Green Bay Mass Balance Study divided the bay into
four surface segments corresponding to the nine used for the surface water. Wet
deposition was calculated as an external input to each of the four segments using
monitoring data to generate the input series. The volatilization was incorporated in the
water balance model. Two air-water mass transfer sub-models (O'Connor (1983) and
Mackay and Yeun (1983)) were evaluated to compute the overall mass transfer coefficient.
Predictions from both models were compared with measurements of instantaneous
air-water fluxes. Results from the O'Connor model were found to be in better agreement
with observations, particularly at high wind speeds. This sub-model coupled the
atmosphere and water.

Based on the ambient data, wet and dry deposition loads to the lake and the atmospheric
boundary conditions will be assessed. In the simplest terms, wet deposition is assessed
from the concentration in the precipitation, amount of precipitation, and area of lake covered
by precipitation. The dry deposition flux is calculated by dividing the particle distribution into
a number of intervals and assigning the appropriate deposition velocity.(Holsen 1993) The
flux for each interval is summed for the total deposition. Several models exist for the
determination of the deposition velocity and the intensive studies are expected to advance
the state of these models. The volatilization component will be addressed as a sub-model
as in the Green Bay Mass Balance. However, it will be improved upon by the specific
research studies of the Lake Michigan Mass Balance and calibrated with ambient data.

Linking an atmospheric mass balance/transport model with the water mass balance model
requires emission inventories and process information which are not presently available for
comprehensive atmospheric models. Simple atmospheric deposition models are currently
being developed, such as RELMAP which is being developed to use a mercury emission
inventory. However, while these models are being developed, the Lake Michigan Mass
Balance model will use loads based on ambient data.

Research Issues/Areas

Process related atmospheric research to improve mass balance estimates for SVOCs
include (approximately in order of importance by category):

Wet deposition:

Gas/aerosol distribution, and aerosol scavenging coefficients
Total atmospheric concentration
Total precipitation concentration
Gas scavenging coefficient

Dry deposition:

Aerosol deposition velocity
SOC aerosol size distribution

Gas Exchange:

SOC speciation in water
Mass transfer coefficients, including the applicability of existing mass transfer coefficients
to the Great Lakes
Total SOC concentration in water
Henry's Law constant/temperature dependence
Comparison of different models (two film vs. surface renewal)
Investigation of surface microlayer in gas exchange

OUTPUTS

There are three potential removal paths for the targeted chemicals in Lake Michigan. These
are burial in the bottom sediments, volatilization to the atmosphere (see Atmospheric
Loadings), and discharge through the Straits of Mackinaw. Volatilization is covered under
Atmospheric Loadings and discharge through the Straits will be calculated from water
column measurements. This section will describe monitoring to quantify sedimentation.

SEDIMENT AND PARTICLE FLUX

Data Quality Objective

The goal is to measure the sediment-water exchange of the target compounds to
within an error of 30% with a confidence level of 90%. All data collected will be
acquired with generally acceptable or peer reviewed sample/data collection,
handling and analytical techniques. All of the data to be collected for this program
will be subject to EPA QA/QC oversight.

Sediment Project Components

The annual cycle of particle production and transport plays a major role in the
seasonal and long term behavior of contaminants in lakes. Compounds entering
the lakes are removed to the sediments at a rate proportional to their affinity for
settling particles. Since particle residence times in the water column are relatively
short (even in deep systems, such as Lake Michigan, particle settling times are less
than one year), particle-associated contaminants are efficiently scavenged and
removed to the sediments. After reaching the bottom, the settled materials are
mixed by the feeding activities of bottom dwelling organisms into an homogenized
pool representing years to decades of recent sedimentation (Robbins, 1982). It is
apparent from the relatively slow decline in the concentrations of particle-associated
constituents in water and biota in recent years, that sediments are a leaky sink; small
concentrations persist in the water for decades because of processes that
remobilize materials from the bottom.

In regions where sediments are accumulating, the extent of this pool is the sediment
mixed layer (except for constituents with a rate of decomposition greater than the
layer mixing time (approx. seasonal). In regions where there is no apparent long
term accumulation of sediments, the exchangeable pool is the material temporarily
deposited and in transit to the depositional regions. The critical parameters required
to estimate the sediment-water exchange of contaminants are:

1. the concentration of total contaminant in the sediment mixed layer
   (the material available for exchange);
2. the time constant (sediment accumulation rate/thickness of the mixed layer)
   for changing the concentration within this layer;
3. the amount of resuspension of the local sediments;
4. the distribution coefficients for the contaminant in local sediments;
5. the gross downward sediment and associated contaminant flux;
6. the dissolved and DOC bound contaminant sediment-water exchange.

Sediment Core and Surface Sediment Analyses

Many of the target compounds in this LMMB Program accumulate in sediments of
lakes and, as a result of resuspension/benthic food web processes, this
exchangeable inventory effectively buffers the temporal behavior of these
contaminants to changes in loadings. In order to model the behavior of the
programs' target compounds, a careful measurement of the concentrations in the
sediment mixed-layer and long term burial must be made. Radionuclides, principally
210Pb and 137Cs, have been used to: 1)determine the geochronology over the
last 100-120 years of such sediment records; 2) estimate the extent of surficial
mixing due to physical or biological process; to estimate the rate of movement of
contaminated sediment from non-depositional to depositional areas (focusing); and
3) calculate fluxes to the sediments and relate them to input functions. Radionuclide
measurements will be used to address:

1. the concentration and inventory of target contaminants in the
   sediment mixed layer (the material available for exchange)
   through determination of the thickness of the mixed layer, and
2. the time constant (sediment accumulation rate/thickness of the
   mixed layer) for changing the concentration within this layer

This project will consist of four components:

(a)the collection of vertically undisturbed sediment cores representative of all of the depositional zones in Lake Michigan;

(b) the sampling of these cores in the best manner to provide samples that:
(i) may be used to measure the sedimentation rate and mixed-layer depth;
(ii) will be analyzed for mercury and specific organic compounds in the mixed layer;

(c) the analysis of samples from fully-sectioned cores for water content, 137Cs, and 210Pb;

(d) the evaluation of the data on a core-by-core basis obtained in component (3) and the calculation of sedimentation rates and mixed depths using established best practice.

Sampling

A sampling grid covering all depositional areas of the lake has been established
based on the locations of the 40 stations already created for the EMAP program. A
comparison of these stations with the grid established by Argonne National
Laboratory in 1972, and expanded upon in 1982 and 1992 by the Center for Great
Lakes Studies, University of Wisconsin-Milwaukee and NOAA-GLERL is shown in
Figure 7. Representative samples will be collected at each station on the complete
EMAP grid provided there is a sufficient depth of sediment to sample using the
CGLS-UWM Box Corer. In addition sediment samples will be collected at
approximately 30 stations sampled last in 1992 where the 137Cs distribution, and
therefore the effectiveness of this particular location to reflect focusing of
contaminated sediment and the depth of the mixed layer, is already known. This
arrangement will provide materials that are already known to reflect significant
focusing and permit the timely start for the analysis of organics (and mercury) in the
mixed layer during the first year of the study without having to wait for the evaluation
of the samples from the new stations based on the EMAP grid. (The selection of
stations will be arranged among the P.I.'s involved based on an analysis of both
published and unpublished information).

Each retrieved box core, one per station, will be sub-sampled to provide 4 - 10 cm
diameter cores using best practices to prevent core shortening. The disposition of
these cores will be as follows:

Sub-core 1 This core will be sectioned at 1 cm intervals to the bottom and the samples analyzed for water content, 137Cs, and 210Pb, diatoms (EMAP) and mercury

Sub-core 2 The core will be sectioned in a similar manner to the one above, frozen, and archived.

Sub-core 3 and 4 These cores will be sectioned in 1 cm intervals and combined in
order to provide sufficient sample for the analysis of the organic compounds
included in the Lake Michigan Mass Balance Program.

Sediment Resuspension

3. quantifying the resuspension of sediments,

The sediment-water exchange component is critical and all current approaches to
quantifying sediment resuspension are imperfect. To add confidence to these estimates,
three approaches will be taken:.

a. Laboratory flume measurements of sediment resuspension potential or initiation velocity have been made and will be continued. Undisturbed cores will be collected and limited measurements will be made. These would be useful in constraining any modeling.

b. In-situ flume measurements of sediment resuspension will be made at several locations in sediment depositional areas.

c. In-situ time series of light transmission (calibrated to TSS) and current velocity provide the only direct evidence of resuspension events. Vertical arrays of transmissometers and sediment traps (which passively sample the settling particle pool) can provide information on the vertical extent of the bottom nepheloid layer during the stratified period and directly measure the resuspended particle flux during the period when the lake is well-mixed. Sites will be located near water column master stations. The quantity of resuspended sediment in traps can be estimated by measuring their 137Cs activity. This tracer is all sediment associated and virtually all 137Cs in traps has come from sediment resuspension (Eadie et al., 1984).

Fine-grained sediments are transported primarily as suspended load, so once the
material is in the water column a circulation model can be used to track the
movement of the sediment, but determining under what conditions the sediment is
deposited or eroded is considerably more difficult. This effort will consist of field
measurements designed to establish the conditions necessary for the resuspension
of fine?grained bottom sediments in Lake Michigan and to assess the relative
importance of local resuspension versus advective processes in the deeper parts of
the lake.

(a) Instrument platforms have been deployed at various locations in the lake.
The platforms support sensors that measure water temperature and water
transparency at several heights above the bottom, as well as current velocity and
water depth. The attached Table 6 shows the positions of the sensors at the three
stations deployed for the winter on October 31, 1994. For logistical reasons the
three tripods were deployed near Muskegon, MI in water depths of 30, 58, and
100m along a transect running from Muskegon harbor to Brian Eadie's sediment trap
station. The 100m station is near Brain Eadie's set of sequencing traps, while the
shallower stations will allow the observation of both the changes in conditions with
depth and the amount of cross?shelf transport. Weekly vertical temperature and
transmissometer profiles taken during the summer, 1995, will be used to correct the
time series measurements for any fouling that may occur, and to assess the
representativeness of the observations. Beginning May 1995, the tripods will be
serviced at approximately 4 week intervals until October 1995, so that a full year of
data is collected. Supporting weather data will be obtained from NOAA's NOMAD
buoys and CMAN stations, and from the weather station established at NOAA's
Muskegon facility.

Moorings have been deployed at three sites; the instruments at each site are at the
following elevations above the bottom (in meters, mab). All instruments will sample
for one minute every hour at one Hertz. The average of these measurements will be
recorded. The current meters are electromagnetic (either Marsh?McBirney 585s, or
Interocean S4s). Temperature measurements are made using YSI thermocouples.
Water transparency measurements are made using either Sea Tech
transmissometers (25 cm pathlength) or Sea Tech light-scattering sensors.
Paroscientific pressure sensors are being used to record water depth.

(b) Data from the tripods will be augmented by measurements from a
bottom?resting flume. This device allows in?situ measurements of the critical
velocity required for erosion by creating a controlled flow across the bottom and
monitoring when sediment resuspension occurs (Hawley, 1991). Using this device
will allow the critical erosion velocity to be measured at a large number of sites in a
relatively short time. Deployments will first be made at the tripod sites so that the
flume results can be compared to naturally? occurring erosion events. Once this is
done the flume can be used at other sites in the lake, so that the erosion velocity of
different sediment types can be determined. Box cores will also be taken at each
site where the flume is deployed in order to determine the sediment properties.
These will include porosity and grain size.

Contaminant Distribution Coefficients

4. the distribution coefficients and bioavailability of the target
contaminants in mixed-layer sediments,

The coupling of a physical sediment model with concentrations of contaminants
associated with sediment particles will provide an estimate of the sediment-water
exchange of persistent hydrophobic contaminants. Equilibrium phase distribution
coefficients are available for the PCBs from the Green Bay monitoring program
(DiPinto et al., 1991). other Great Lakes field measurements (Baker et al., 19xx) and
from laboratory experiments (Eadie et al., 1990). Attempts to measure distribution
coefficients for phytoplankton were pioneered as part of the Green Bay Mass
Balance Study and are continuing for Lake Michigan.

Downward Flux of Sediments and Contaminants

5. the gross downward sediment and associated contaminant flux

The objectives of this effort are:

a) to measure the gross downward fluxes of particulate material and
organic carbon and

b) to collect samples of the resuspendable pool of materials in regions
of the lake where modern sediments do not accumulate and

c) to provide samples of these materials for target compound analysis.

In the Great Lakes, as in most aquatic systems, the rapid and efficient processes of
sorption and settling scavenge contaminants from the water column with the result that the
largest fraction of persistent trace contaminant inventories reside in sediments. However,
studies of the long-term behavior of certain fallout radionuclides and stable contaminants in the Great Lakes have shown that higher levels persist in the lakes than expected if settling and burial were the sole transport process. Materials return from sediments due primarily
to resuspension. Constituents initially transferred to sediments are homogenized via
bioturbation creating a mixed layer corresponding to a decade or more of accumulation.
These are resuspended back into the water column during the isothermal period and are
available for uptake by pelagic biota. It is now accepted that the internal recycling caused
by the coupled processes of bioturbation and resuspension is responsible for the
continuing elevated concentrations of trace contaminants (e.g. PCB, DDT) in fish and the
lag in lake response to nutrient abatement.

Since 1977, GLERL has been examining the processes of particle flux and resuspension
through the use of sediment traps, passive cylinders deployed to intercept materials
settling to the bottom. Traps provide an efficient tool for the collection of integrated
samples of settling materials for detailed analysis. Measuring the mass collected allows us
to calculate the gross downward flux of particulate matter and associated constituents and
to calculate settling velocities.

Twelve traps having sequencing capability for multiple samples per deployment,
(autosequencing sediment traps) will be deployed with eight in four 2 trap arrays (5m above
bottom, and 30m below surface). The remaining 4 traps will be deployed at 5m above
bottom in regions of the lake that do not accumulate recent sediments and are not suitable
for coring. These will provide samples of the mobile pool of particulate matter in the
benthic nepheloid layer, materials resuspended during the unstratified period and materials
settling out of the epilimnion during stratification. Figure 7 shows trap locations and the
attributes of the selected stations are listed in Appendix 7.

For this project, the samplers will be programmed as described in Appendix 7. The
simpler design is also