United States            Office of Water          EPA-823-R-01-009
          Environmental Protection        4305              September 2001
&EPA  Mercury Maps

          A Quantitative Spatial Link Between
          Air Deposition and Fish Tissue
          Peer Reviewed Final Report

-------
       Mercury Maps
A Quantitative Spatial Link Between
   Air Deposition and Fish Tissue
        Peer Reviewed Final Report
                Paul Cocca
        Standards and Health Protection Division
          Office of Science and Technology
               Office of Water
                 9/10/01

-------
                                   Executive Summary

Mercury Maps is a tool that relates changes in mercury air deposition rates to changes in mercury
fish tissue concentrations, on a national scale. The tool utilizes a reduced form of accepted
mercury fate and transport models applied to watersheds in which air deposition is the sole
significant source.

The Mercury Maps model states that for long-term steady state conditions, reductions in fish
tissue concentrations are expected to track linearly with reductions in air deposition watershed
loads.  The model utilized in this project is a reduced form of the IEM-2M and MCM models
used in the Mercury Study Report to Congress (MSRC) (US EPA, 1997b), whereby the equations
of these models are reduced to steady state and consolidated into a single equation relating the
ratio of current/future air deposition rates to current/future fish tissue concentrations.

Mercury Maps is designed to work only with watersheds in which air deposition is the sole
significant source of mercury. A key step in the project then is to identify, and eliminate from the
analysis, watersheds in which mercury sources other than air deposition, such as gold mines and
chlor-alkali facilities,  are present and contribute loads  that are significant relative to the air
deposition load to that watershed.

-------
Table of Contents

Introduction	   1 of 31
Method Overview	  1 of 31
Model Development	   2 of 31
Mercury Maps Model	   4 of 31
Model Uncertainty	   7 of 31
Model Implementation	   10 of 31
Results and Interpretation	  14 of 31
Figure 1.  Fish Tissue Mercury Concentrations Averaged by Watershed	  15 of 31
Figure 2.  Percent Reduction in Air Deposition Load Necessary to Meet New Methylmercury
        Criterion: Watersheds with No Other Significant Mercury  Sources	  16 of 31
Figure 3.  Counts of Fish Tissue Mercury Data Records by Year	  17 of 31
References	     18 of 31

Appendix A:  Mercury Fish Tissue Data Layer	 22 of 31
Appendix B:  Hydrologic Cataloging Unit Boundaries Data Layer	 25 of 31
Appendix C:  Database of Significant Deposits of Gold, Silver, Copper, Lead, and Zinc in the
       U.S	  26 of 31
Appendix D:  Minerals Available System/ Mineral Industry Location (MAS/MILS) Data Layer
        	  27 of 31
Appendix E:  Permit Compliance System Data Layer	  28 of 31

Internal Peer Review: Response to Comments
la) Is the reduced-form model, presented in this report, an accurate characterization of the air deposition load / fish
       tissue concentration relationship, predicted by the IEM-2M and MCM models at steady state? ...    1 of 25
Ib) Is it accurate to say that in watersheds where the mercury load to water bodies is dominated by air deposition,
       mercury concentrations in fish tissue are expected to reduce in direct proportion to reductions in mercury
       deposition, at steady state?	    3 of 25
2a) Can the fish tissue data be used for the purposes outlined, given its origins, quality, and completeness?   7 of 25
2b) Is the scale of the analysis appropriate?  	  10 of 25
2c) Is the use of all other data layers in addressing the goals of the project, appropriate, taking into account their
       origins as well as their quality and completeness?	  10 of 25
3) Can the methods developed in this project be used to quantitatively assess impacts of air deposition reductions on
       fish tissue in air deposition dominated watersheds?	  13 of 25
4) Were the calculations performed correctly? Were the data processed without error?	  15 of 25
References 	   16 of 25
Figure 1 (revised) Fish Tissue Mercury Concentrations Averaged by Watershed	  18 of 25
Figure 2 (revised) Percent Reduction in Air Deposition Load Necessary to Meet New Methylmercury Criterion:
       Watersheds with No Other Significant Mercury Sources  	  19  of 25

Additional Data and Analyses for Appendix
Fish Tissue Database	 20 of 25
Database of Significant Deposits of Gold, Silver, Copper, Lead, and Zinc in the U.S	  22 of 25
Permit Compliance System Data Layer	 23 of 25
                                                11

-------
Introduction
The goal of the Mercury Maps project is to establish a tool to quantitatively evaluate the potential
impact of air mercury emission reduction rules on fish tissue mercury concentrations, nationwide.
Specifically, this project is proposed as a method to help evaluate the benefits of technology
based air emission reduction standards, and/or to be used to evaluate and/or design a risk-based
air emission reduction rule for mercury. In addition, another potential use of Mercury Maps is in
performing a regional or national TMDL analysis.  By relating  reductions in air deposition rates
to reductions in fish tissue concentrations, by watershed, this project is one component of an
overall emission reduction benefits analysis.  Estimates of percent air deposition reductions, by
watershed, would be needed and presumably generated from a regional air deposition model, to
relate air emission reductions to watershed air deposition reductions.
Method Overview
Previous modeling efforts in the MSRC (US EPA, 1997b) described the fate and transport of
mercury in the watershed and aquatic ecosystem in great detail.  The IEM-2M watershed model
and the MCM aquatic food chain model coalesced a considerable amount of scientific studies
into unified pictures of the fate and transport of mercury in the environment.  The MSRC
modeling studies represent the scientific understanding of the fate and transport of this complex
pollutant. This project builds on that considerable effort, and relies on the validation of those
models, in order to identify the minimum number of steps, or calculations, necessary to translate
air deposition rates to fish tissue concentrations as the environment approaches a steady state or
dynamic equilibrium in response to load reductions. The Mercury Maps model states that a ratio
reduction in air deposition watershed loads will produce an equivalent ratio reduction in average
fish tissue concentration in that watershed, at steady state.

To implement the model, a national data set of mercury fish tissue data were averaged across
USGS HUC-8 watersheds (Figure 1). To demonstrate the ratio reduction approach, the average
concentration was divided by the new methylmercury criterion, which is expressed as a
concentration in fish tissue (0.3 ppm).  The watersheds with potentially significant mine sources
were eliminated based on the presence of gold and mercury mines in the USGS Database for
Significant Deposits and the MAS/MILS databases, respectively, and the presence of mercury
cell chlor-alkali facilities. In addition, watersheds in which the total  estimated mercury discharge
rates from Publicly Owned Treatment Works (POTWs) and pulp and paper mills, in each
watershed exceeds 5% of a typical low level air deposition watershed load (10 ug/m2/yr) as
delivered to waterbodies, were eliminated from the analysis.

Results of the model implementation are presented in Figure 2. Together with air deposition
model output for both baseline and control scenario average watershed loads, this project
methodology could be used to relate  proposed emission control scenarios with expected long
term impacts on fish tissue.
                                      Page 1 of 31

-------
Model Development

Steady State Formulation of the Mercury Cycling Model (MCM)
Hudson, et al, 1994 details the Steady State equations of the MCM.  In this paper, they include
the equations for both mass balance and steady state concentrations for divalent mercury,
methylmercury, and elemental mercury. They also include steady state versions of the equations
used in the MCM to describe the following processes: total inputs of mercury to a lake; reduction
of Hg2+ to Hg°; methylation of Hg2+ in the water column and surficial sediments; scavenging and
sediment burial; out-seepage of dissolved  species; demethylation in water column; and the
volatilization loss of Hg from a lake.

The mass balance equations state that the sum of total inputs, reduction, methylation,
scavenging/sediment burial, demethylation, volatilization loss, and out seepage (as relevant for
each of the three species) must sum to zero.

The process equations, in words, state:
•      Total inputs of mercury to a lake are a function of air deposition and in-seepage,  and are
       independent of in lake concentrations;
•      Mass rate of reduction of Hg2+ to Hg° is proportional to dissolved Hg2+ concentration;
       Mass rate of methylation of Hg2+ in column and surficial sediments is proportional to
       dissolved Hg2+ concentration;
       Mass rate of scavenging and sediment burial is proportional to particulate concentrations
       (by species);
       Mass rate of out  seepage of each dissolved species is proportional to its dissolved
       concentration;
       Mass rate of demethylation  in the water column is proportional to the dissolved
       methylmercury concentration;
       Mass rate of volatilization loss of Hg from a lake is proportional to concentration of that
       species;

The equilibrium concentration equations can be reconstructed by inserting the mass rate process
equations into the mass balance equations, and solving for species concentration. Each of the
resulting dissolved mercury species steady state concentration equations shows a linear
relationship between concentration of the affected species and deposition rate.

The MCM describes fish methylmercury bioaccumulation as a linear function of dissolved
methylmercury concentration, i.e. the standard BAF equation.  Hudson, et al, 1994 also show
how they obtained the best correlation with observed data (r2 = 0.96) by taking into account
"speciation-independent influences of limnological parameters", in particular pH and  calcium
concentration, while still retaining the linear relationship between dissolved methylmercury
concentration and fish tissue levels.

While extreme events may unlock mercury from environmental pools, this is part of an expected


                                       Page 2 of 31

-------
uneven response over time to reduced mercury loadings to watersheds and waterbodies. Also
referred to as dynamic equilibrium, steady state in environmental systems means that
concentrations may vary season to season or even year to year, but that long term averages are
constant. While environmental media mercury concentrations are expected to trend downward,
random fluctuations in meteorological and other environmental patterns will cause uneven
responses.

Steady State Formulation of the IEM-2M Watershed Model
The IEM-2M watershed model, used in the MSRC (U.S. EPA, 1997b), describes the impact of a
steady state air deposition load on watersheds and waterbodies. Specifically, it tracks
concentrations of three mercury species in soil, soil pore water, and soil pore gas, as the
watershed is subject to air deposition, runoff, volatilization, and the interplay between the three
mercury species. IEM-2M tracks divalent, methyl-, and elemental mercury through methylation,
demethylation, reduction, oxidation, and mer demethylation (conversion of methylmercury to
elemental mercury).

Described in the MSRC (section 4.4, US EPA, 1997b), reaction rate constants for IEM-2M are
all independent of concentration.  The mass balance equations for each species, i.e. of the form:
                                       dt   '"

where:
       Vs = volume of watershed soil, and
       CS1 = concentration of species i in watershed soil

When set to zero, the equations of the above form can be solved simultaneously to demonstrate
that each species is a linear function of air deposition.

Additionally, the MSRC describes a parameter sensitivity analysis on the IEM-2M model (p. 6-
14, in Table 6-12, US EPA,  1997b).  Sensitivity is expressed as the relative change in total water
column mercury concentration divided by the relative change in the model parameter, in percent.
The above referenced table shows soil concentration sensitivity of+100% and -100% for positive
and negative changes, respectively, in the air deposition rate input parameter.  That is, given a
decrease in air deposition loading rate, the IEM-2M model shows the same decrease in total soil
mercury concentration.  The same result is shown for total water column mercury concentration
(Table 6-13, US EPA, 1997b), and predatory fish mercury concentration (Table 6-14, US EPA,
1997b).
                                      Page 3 of 31

-------
Mercury Maps Model

Building on the analysis of the steady state formulation of the MCM and IEM-2M model
equations, presented above, a simple reduced-form model is derived in order to relate percent
reductions in air deposition load to percent reductions in fish tissue concentration, at steady state.

The model is derived starting with the standard, steady state bioaccumulation equation:


                            Cfishil=BAF-Cwaterii     (1)

where:
       Cfishj! and CwaterJ1are methylmercury contaminant levels in fish and water at time tl, respectively;
       BAF is the site specific bioaccumulation factor, which is constant for a given age/length and species of fish
       in a specific waterbody.

For a future time, t2, when mercury contaminant levels have changed, but all other water quality
parameters remain the  same,  the equation is rewritten:
                             Cfishn=BAF-Cvatern     (2)

where:
       Cwater_t2 is the methylmercury contaminant level in water at time t2;
       C£sh a is methylmercury contaminant levels in fish of the same age/length and species as in Cfish tl
Combining equations 1 and 2:


                                    C       C
                                    ^       ^
                                    c       c
                                      fisht2      water^

That is, for the same age/length and species offish in the same waterbody, the ratio offish
concentrations and the ratio of water column concentrations at different times are equivalent.
Also of note is that the equation is independent of BAF, and thus can be applied from waterbody
to waterbody independent of the tremendous variability in mercury BAF, associated with inter-
waterbody variability in water quality parameters.  Given that methylmercury water column
concentrations are proportional to mercury air deposition load to a watershed, as  demonstrated
from the review of the IEM-2M model,  equation 3 can be rewritten:

                                      C       T
                                      ^
                                      C    ~ L
                                      ^f,Shtl   ^A,rtl

                                       Page 4 of 31

-------
where:
•      L^,. tl and L^ ,2 are the air deposition mercury loads to a waterbody at time tl and t2,
       respectively

That is, mercury fish concentrations will be reduced  from current levels in proportion to load
reductions for the watershed. For waterbodies in which air deposition is the sole significant
source, fish tissue mercury concentration reductions will be directly proportional to air deposition
load reductions.

In cases where air deposition is not the sole significant source, the equation describing fish tissue
fractional reduction would need to be modified as follows:

                             C    ~  (            \
                              f's^    (LA,rtl + L0ther)


where:
       1,0^ = non air deposition load of mercury to waterbody

and which reduces to the simple proportional reduction, when LO^ = 0.

For the above equation, it can be shown that:


               ^sh           Ai       LAir        L
                                   \'                    \
                -^i    \LAirtl+LOa*r)   A^    \LA,rtl+ Lather)
which is of the form Y = mX + b
where:
       Y = fractional reduction in fish tissue concentration;
       X = fractional reduction in air deposition load;
       m = slope; and
       b = y-intercept.

It can be shown that, for equation 6, m + b = 1 .  As well, the intercept (b) is just the other load as
a fraction of total load.  In the case of the Everglades mercury TMDL Pilot Project, m = 0.94,
while b = 0.06. That is, 6% of the total load is non-air-deposition load.

The method employed in this project, then is to eliminate from the analysis those watersheds
where sources other than air deposition are significant, thus keeping b close to zero, and m close
to one, reducing errors in using the direct proportion methodology (i.e. Y = X).
                                       Page 5 of 31

-------
Taking into account other loads, in other than a screening level manner, is infeasible due to large
uncertainties in estimating the absolute magnitude of these sources. A more detailed analysis, of
individual watersheds, in these cases, is more appropriate.  Screening out of watersheds with
significant or potentially significant sources allows one to proceed with much greater certainty
and credibility, though with a bias towards underestimating overall reductions.  Equation 6 could
be used to illustrate how excursions from the simple proportional reduction model, caused by the
presence of other sources, could influence the accuracy of Mercury Map  predictions. Given that
watersheds in which  all  potentially significant non-air-deposition sources (e.g. point sources and
historic mining activities) are eliminated from the analysis, these prediction errors are expected
to be quite small.

While the more complex model, presented above, would allow other significant sources, to be
included in the analysis, the quality of data characterizing loads from these sources is currently
insufficient for use in other than a screening level approach. In this project, loads from POTWs
and pulp and paper mills are simply estimates based on the product of facility flow rates and
average measured mercury effluent concentrations. There was insufficient data to use a similar
screening level approach for mercury cell chlor-alkali facilities and estimating loads from
abandoned mines would likely require numerous high quality samples at each site. This lack of
data quality was a key determinant in selecting an approach that included screening out
watersheds with significant or potentially significant non-air-deposition sources. While a national
statistical sampling of all permitted discharge facilities, by industry, would provide an adequate
base of information to improve on this approach, no such comprehensive data set as of yet exists.
In addition, it would  be extremely difficult and resource intensive to attain  a sufficiently precise
nationwide data set on mercury loads from other nonpoint sources such as abandoned mines and
bedrock erosion.
                                       Page 6 of 31

-------
Model Uncertainty

Comparison with Detailed Study
The Everglades Pilot Project Mercury TMDL (USEPA, 2000a), determined the relationship
between atmospheric Hg(n) deposition and long-term fish tissue mercury concentrations, in
Everglades Site WCA 3A-15. To determine the relationship, the researchers first calibrated the
E-MCM to the WCA 3A-15 site and ran the model for 100 years to achieve an approximate
steady state response to the current air deposition loading rate.  They compared these long-term
predicted results against current measured water, sediment, and fish concentrations and found an
acceptable match. They then reduced the mercury deposition load,  as well as watershed mercury
inflows, and ran the model for an additional 100 years; performing this process separately for
25%, 50%, 75%, and 85%  reductions in overall loads. In plotting the results of the above
analysis, the researchers found the relationship between fish mercury concentration and mercury
air deposition rate to be linear, and to fit the equation: Y = 0.9408X + 0.0611, where Y = the
fraction of current fish concentration; and X = fraction of current air deposition load (wet and
dry).  By comparison, the Mercury Maps model simply states Y = X (i.e. slope =1, and intercept
= 0).  The reasons for the discrepancy between the two models, as well as the effect on predictive
accuracy of the Mercury Maps model, in this case, are discussed below.

The researchers note the reason for the slope and intercept, of the fitted equation, not being equal
to one and zero, respectively, is that even after the initial 100 year simulation, deep sediment
concentrations had not yet reached a steady state response to current loading.  The sediment
concentrations are elevated with respect to true steady state, and thus represent an additional load
beyond the reduced air deposition load.  The researchers also state that were the simulations
carried out for much longer periods (as long as thousands of years), predicted concentrations
would approach a direct proportional response (i.e. Y = X).

The E-MCM model result, being not quite directly proportional, can be shown to have a
quantifiable effect on the predictive accuracy of the Mercury Maps  model, were it also applied to
the Everglades WCA 3A-15 site. For a 50% reduction in air deposition load, the Everglades Hg
TMDL application would predict a 47% reduction in fish concentrations, to Mercury Maps' 50%
reduction. Higher proportional reductions result in larger prediction errors, e.g. a 90% air
deposition load reduction causing a 84% fish tissue reduction in E-MCM, versus a 90%
reduction in Mercury Maps.

In conclusion,  sediment concentrations may, in some instances like the Everglades WCA 3A-15
site, be elevated with respect to current air deposition loads. These elevated concentrations,
unaccounted for in Mercury Maps, act as an additional source and, where significant, will cause
Mercury Maps to over-predict reductions in fish tissue concentrations. In the case of the
Everglades site WCA 3A-15, the quantified over-prediction is relatively small, on the order of up
to five percent at high air deposition rate reductions.  In addition, it can be reasonably expected
that similar situations exist in other watersheds throughout the U.S., where historic atmospheric
mercury deposition and continued mercury accumulation in sediments may cause current
                                      Page 7 of 31

-------
sediment concentrations to be elevated with respect to current deposition levels, but that these
elevated levels will cause fish concentrations to depart only slightly from the direct proportional
reduction used in Mercury Maps.  This small difference is likely reasonable for the purposes of
benefits assessments.
Linearity Assumption
The ES&T article (DiPasquale, et al, 2000) found that demethylation rates increased with
increased methylmercury concentration, in waterbodies with extreme mercury contamination,
caused by historic mining practices. While this study does appear to provide real evidence of a
nonlinear process with respect to mercury concentration, it is not clear whether net
methylmercury bioaccumulation in fish is expected to be a nonlinear function of mercury
concentration as a result. However, were nonlinear demethylation rates to affect net
bioaccumulation in this way (i.e. to cause fish tissue  concentration reductions to slow with
decreased load) the proportional reduction approach  in Mercury Maps would then overpredict
reductions in fish tissue.  That is, the potential implication of DiPasquale, et al, 2000 is that more
dramatic deposition reductions, than those derived in the current formulation of Mercury Maps,
would be required in order to achieve the methylmercury criterion in all watersheds. Conversely,
for a given technological standard based emission reduction, the estimated benefits would be
reduced.

Finally, it should be noted that any potential for non-linearity in the  selected air deposition model
should not be confused with the potential for non-linearity in watershed fate and transport. The
air deposition model need not be linear with respect to source loadings to the atmosphere in order
to be used in conjunction with Mercury Maps.
Use of Watershed Screening Method
It should be noted, however, that for the use of Mercury Maps in an emission reduction benefits
analysis, the effect of using the watershed elimination method is to under-predict the overall
benefits. The extent of this under-prediction can be bounded on the upper end as the percentage
of all watersheds that are eliminated (i.e. if 20% of all watersheds are eliminated, the under
prediction is, at a maximum, 20%).  It should also be noted that this expected under-prediction
should be within the realm of uncertainty brought in by other aspects of a benefits assessment.
This approach buys a higher degree of certainty at the expense of a potentially higher benefits
calculation.
Other Source Categories
Below, is a discussion on sources potentially not taken into account by this project, as well as
further explanation of the reasoning behind eliminating known or likely significant point sources
from the analysis.  Based on the studies summarized briefly below, it appears unlikely that
background sources contribute a significant fraction of the current load.  Also, the error likely to
                                       Page 8 of  31

-------
have been introduced, by not taking into account the effect of background load on predicted
response, is small relative to the prediction.

A number of studies have looked at the potential relationship between mercury concentrations in
soil and runoff, as well as soil/bedrock erosion as a potential mercury source. St. Louis, et al,
1996 estimated, based on comparisons with erosion rates of other minerals, that mercury
weathering rates could range from 0.4% to 6% of total mercury inputs to the watershed. They
deemed this amount insignificant for the purposes of the mass balance in their study. Finally,
Aastrup, et al,  1991 quantified mercury accumulation and transport in a number of soil layers in a
watershed, as fractions of total mercury deposited. By the use of a mass balance they
demonstrated that the content and fluxes of mercury, in interflow and groundwater flow, is
accounted for via the percolation of atmospherically deposited  mercury through the soil.  In
addition, while weathering and dissolution of mercury from bedrock may be significant in some
areas, these areas would likely be associated with mercury mining locations, and would be
screened from the analysis on that basis.
Emissions Inventory Period and Watershed Lag Time
Mercury air emission rates have decreased recently due to a contraction of the medical waste
incinerator industry in response to emission regulations, as well as recently due to new MACT
rules for Medical Waste Incinerators and Municipal Waste Combustors (SAI,  1998)
Since there is a time lag from emission to accumulation in fish, the time period over which fish
tissue concentrations are measured represents an earlier period of emissions.  A plot offish
sample records count by year is shown in figure 3, with  a distinct peak in 1993 and the bulk of
samples from 1990-1995. Given that emission rates have decreased significantly over the last 1-2
decades, and lag time may be as long as 1-2 decades, fish concentrations may be slightly elevated
with respect to  '95-'96 emission levels (recent air deposition modeling studies have used a '95-
'96 emissions inventory baseline year).

If '90-'95  fish tissue is elevated with respect to expected steady state fish tissue concentration
values for the '95-'96 emission levels, predicted future fish concentrations will be higher than
would ultimately occur. In terms of a benefits assessment, then, this model and data set is
expected to systematically under-predict reductions in ultimate fish tissue concentrations
resulting from an emissions rule, as compared against '95-'96 emission levels. Similarly, a risk-
based emission reduction rule, designed to achieve criterion levels, using this method, would be
expected to produce emission reductions beyond the minimum necessary to achieve the criterion.
The uncertainty caused by the effect of lag time and a gradual reduction in mercury emissions
over the last 1-2 decades, is a non-quantified margin of safety. This uncertainty could be reduced
and/or quantified by a more detailed analysis of trends in U.S. mercury emissions and studies of
lag times.  Other data quality issues and limitations, by data layer, are detailed in the Appendix.

In addition, while the accuracy of predictions offish tissue reductions in Mercury Maps will be a
direct reflection of the accuracy of the air deposition model in predicting relative rates of air
                                       Page 9 of 31

-------
deposition of mercury to watersheds, this project is not dependent on the use of any particular
deposition model. Rather, this project is intended to be coupled with the air deposition model
thought best able to predict percent reductions in mercury deposition on a national scale.
Model Implementation

The first step in implementation of the Mercury Maps model, was to average the fish tissue data
by watershed boundary.  As the exposure endpoint of concern is chronic consumption of
contaminated fish, an individual is expected to be exposed, over the long term, to average
concentrations.  The National Listing of Fish and Wildlife Advisories (NLFWA) fish tissue
database (described in more detail in the appendix) was processed in Arc View according to the
following steps:
       select only samples listed as fillets (inlcuding both skin on and skin off);
•      spatially merge fish tissue sampling locations with watershed boundary data;
       run summarize function to determine the average fish concentration by watershed;
•      create a new field and calculate the ratio of the criteria to the average concentration to
       show the percent reduction in total mercury load required to meet the criteria.

Since these fish tissue data are used by States in making decisions on whether to post fishing
advisories, the fish sampled may not be truly representative of the true population average across
the HUC. These samples are in fact generally based on areas that are most heavily fished
(anglingpressure) and/or those that are suspected of having higher than average potential to be
polluted (AFS, 2000). That is, the average concentration may be biased higher with respect to
the true average, but would be expected to be more reflective of the average concentrations in
consumed freshwater fish.  While a statistical sampling would produce  a less biased average
concentration, the goal of protecting human health is better served by sampling with a bias
towards areas, that with sound scientific reasons,  are suspected of having higher than average
concentrations, or that account for a disproportionate amount offish caught and consumed. The
manner in which fish tissue is sampled is a decision  made by the state agency. Additional
analyses on the fish tissue data are included in the Appendix. It should be noted that the median
number of samples per HUC is 9 (mean is 24), which is a reasonable statistical sample on which
to base a mean.

A statistical  analysis of the different sample types was performed. It was found that there were
large differences between fillet and whole fish  concentrations (0.40 ppm vs. 0.18 ppm,
respectively), as well as between fillets with skin off versus fillets with skin on (0.40 ppm vs.
0.31 ppm, respectively).  It was also shown that these differences in concentrations may be due in
large part to differences in species sampled. Whole  fish samples are more frequently trophic
level 3 fish, while having skin on or off appears to be a function of how a particular fish species
is typically consumed (see Appendix for details).  Based on this analysis, and because for the
purposes of benefits assessment, TMDLs, and risk-based emission control rules, the consumed
concentration is most relevant.  Whole fish samples, then, are eliminated from the analysis, and
                                     Page 10 of 31

-------
all fillets (skin on and skin off) are included. All unknown and unspecified sample types are
eliminated from the analysis.

A preliminary analysis of the effect of using a finer resolution watershed coverage was
performed. The HUC-11 watershed coverage, readily available for the Chesapeake Bay
watershed, was added to the project, and fish tissue data were aggregated to that level of
resolution. The resulting map of the average fish tissue concentration relative to the
methylmercury criterion showed a few local maxima that had been smoothed out at the higher
resolution HUC8 watershed scale, as is to be expected.  An analysis, presented in the Appendix
(fish tissue database section), however, shows that the median number of samples in the HUC11
watersheds is 3, while the median in HUC8 watersheds is 9.  That is, while the fish tissue
database is substantial, it is not large enough to justify averaging across watersheds smaller than
the HUC8 level. Analyses at smaller levels is justified on a case by case basis, and may be
appropriate in a sentinel-watershed  approach associated with a risk-based air emission reduction
rule analysis, or a TMDL.

In the next step in the model implementation, watersheds influenced by mercury loads from
historic gold and mercury mines, or by current chlor-alkali facilities, were eliminated from the
analysis.

Mercury Mines:
Plouffe, et al found elevated mercury concentrations in soils surrounding mercury mines, at
distances of up  to 20 km to 40 km.  Thus, it appears reasonable to suspect mercury mines as
likely sources of elevated environmental concentrations of mercury.

GoldMines:
The MAS/MILS database included  numerous locations, with gold as a commodity,  that were
likely insignificant sources and may have never used the mercury amalgamation process.  An
alternate database, the USGS Database of Significant Deposits of Gold, Silver, Copper, Lead,
and Zinc in the  United States was used instead (Long, et al, 1998).  This database was queried for
mines that have produced more than two tons of gold.  Any watershed containing one or more of
these mines was screened from the  analysis, on the basis of potential historical contamination of
the watershed due to use of the mercury amalgamation process.

Chlor-Alkali Facilities:
The Mercury Study Report to Congress (MSRC) (EPA, 1997) notes that only a fraction of chlor-
alkali facilities use the mercury cell process. MSRC lists the 14 facilities using the process at
that time and notes that no new mercury cell chlor-alkali facilities are planned to be built, while a
recent study (EPA, 2001) found a much lower loading rate than reported in PCS (see Appendix
for details). However, because, recent data were available for only one mercury cell chlor-alkali
plant, there were insufficient data to assign an average value across all plants, in a screening
approach, similar to that for pulp and paper mills.  Instead, the simple presence of a plant was
used to screen out watersheds.
                                      Page 11 of 31

-------
The spatial data processing steps were as follows:
       query mines databases (described in more detail in the appendix) for all mercury
       producing mines or mineral locations and all significant past producers of gold;
       export selected gold and mercury locations data and import back in as separate themes;
•      use "select by theme" command to identify and eliminate from the analysis, all
       watersheds that contained at least one gold or mercury location.

In the third and final step in model implementation, watersheds with significant point source
mercury discharge loads were eliminated. A number of reports have inventoried mercury
discharge to water suggesting alternative methods of taking inventory of mercury discharges to
surface waters. A recent study by EPA Office of Research and Development (US EPA, 2000b)
examined mercury use in five major categories of type of use, and estimated mercury release to
the environment by source category.  The report identified four sources of mercury discharge to
water with  estimated total annual loads: Chlor-Alkali Production (0.1 tons/yr), Utility Coal
Combustion (7 tons/yr), and Sewage Treatment and Sludge Incineration (15-30 tons/yr), and
Dental Offices (7.4 tons/yr).  The report used TRI data for chlor-alkali plants, an estimate of
measured concentrations from cooling tower blowdown for the utility coal combustion category,
and PCS data to estimate the  WWTP load. Dental offices, while estimated separately in the
report, would, for the most part, be incorporated in the WWTP value. The National Sediment
Inventory (NSI) (U.S. EPA, 1997) compiled an inventory of point source dischargers of mercury
to the waterways of the U.S.  In this study, they found that the Toxics Release Inventory (TRI)
showed 14  facilities discharging 271 Ibs/yr while the Permit Compliance System (PCS) showed
749 facilities discharging a total of 28,592 Ibs/yr, clearly indicating PCS is more comprehensive.
However, because of known uncertainties in PCS data for mercury, a new methodology for
screening out sources was developed.

Pulp and Paper Mills
A detailed review of a study conducted by the Maine DEP (ME DEP, 2001) found the expected
loading rate for both pulp mills and paper mills to be 3.1 Ib/yr. That is, the 14 pulp and paper
mills in the study had an average concentration of 13 ppt, and an average flow rate of 79 MOD
(see Appendix for details). By contrast, PCS data showed average loading data of 10 Ib/yr for
pulp mills and 40 Ib/yr for paper mills (see Appendix for details).  Additional data on POTW
facilities from PCS, in contrast with a clean sampling  and analysis study by AMSA,  indicates
further contamination problems with the PCS data.  As a result of these  efforts, discussed in
detail below, it was confirmed that the PCS data is likely based on non-ultra-clean techniques.
These data will not be used.

Publicly Owned Treatment Works (POTWs):
Using a study conducted by the Association of Metropolitan Sewerage Agencies (AMSA) on
mercury in POTW effluent, a sensitivity analysis was performed to examine the relative loading
from POTWs versus air deposition delivered to waterbodies. Nellor,  1999 cites an average
mercury concentration in POTW effluents of 7.25 ppt (ng/L).  This data is based on AMSA's
study of 24 POTW facilities in six states, using clean sampling and analytical techniques, for
                                     Page 12 of 31

-------
facilities with a range of flow rates from 0.65 MGD to 225 MGD.  The range in mercury effluent
concentrations was 0.7 ppt to 69.9 ppt, with a median value of 5.0 ppt.

For the purposes of this study, the mean value of 7 ng/L was applied to each POTW, at the PCS
reported flow rate, and their cumulative load was summed (across HUCs).  This loading rate is
compared to a typical air deposition load of 10 ug/m2/yr, with an assumed 20% delivery to
waterbodies (discussed below). If the sum of POTW mercury loads is greater than 5% of the air
deposition load, as  delivered to waterbodies, then the watershed is screened from the analysis.
79 watersheds were screened out solely on the basis of this procedure, all located west of the
Mississippi River.  Additional details on this analysis, as well  as on the mercury data available in
the PCS database, are available in the Appendix.

In order to compare the estimated point source mercury load data against average deposition rate,
the total estimated load for all key sources is compared against mercury load, delivered to
waterbodies.  That is, PCS data were used to eliminate watersheds as having significant point
source discharge only if the sum of PCS discharge  in the watershed was greater than 5% of the
typical (10ug/m2/yr) deposition rate as delivered to waterbodies. See Hydrologic Cataloging Unit
Boundary theme in appendix for calculations of average deposition load by watershed.

The following is a review of the literature on the percent of total atmospheric deposited mercury
which is transported from watersheds to receiving waterbodies.
      - Aastrup, et al,  1991 found yearly mercury transport to a lake from a forested
      subcatchment to be 17% of total mercury deposition.
      - Lindberg,  1996 found the  combination of runoff and leaching to total 3.5% of total
      atmospheric load of mercury to the Walker Branch watershed, TN.
      - St. Louis, et al, 1996 found that for five catchments in the Experimental Lakes Area
      (ELA) in northwestern Ontario, export of total mercury ranged from an average of 29.5%
      for a basin wetland to 61.1% for a riverine wetland.
      - Sherbatskoy, et al, 1998 found an export rate of 6% of total mercury from a small
      forested catchment in Vermont.
      - Swain, et al, 1992 found for catchments to seven headwater lakes in Minnesota and
      Wisconsin, the proportion of atmospheric mercury transported from catchment to lake to
      be 26% and 22% for modern and pre-industrial times, respectively.
      - Hurley, et al, 1995, found watershed total  mercury transfer efficiencies, for 39 river sites
      in Wisconsin to range from Fall means of 0.5% to 8% (depending on watershed type) to
      Spring means of 29% to 90%.
      - Tsiros and Ambrose, 1999 found delivery of mercury to canals in the Everglades
      Agricultural Area was 23% of atmospheric deposition.
      - Johansson et al. (1991) reported mercury transport fluxes from several small watersheds
      to be about 30% of atmospheric deposition.
      - Tsiros, 2001, in model simulations, found that mercury runoff flux was 2 to 3 times
      higher than  normal during wet years, and 5  to 7 times lower than normal during dry years,
      and that mercury runoff flux was 18% to 61% of atmospheric deposition for wet years,
                                     Page 13 of 31

-------
       and 1% to 4% of deposition for dry years. These values correspond to runoff flux for
       normal years to be between 6% and 30% of deposition.

Based on a review of the above studies, it appears that a delivery ratio of 20% is a reasonable
estimate of the central tendency of this value, an appropriate estimate for the purposes of this
study.  That is to say, on average, it's expected that only 20% of air deposited mercury reaches
waterbodies on a long-term average annual rate. So, watersheds in which the sum of loads from
pulp and paper mills and POTWs is greater than 5% of the nominal air deposition load to
waterbodies (20% of watershed deposited mercury) are screened from the analysis.
Results and Interpretation

The results of the model application are shown in Figure 2, which shows the ratio reduction in air
deposition, by watershed, required to meet the new methylmercury criterion.  Watersheds colored
red indicate where fish concentrations exceed the criterion, while those colored green indicate
watersheds in which no reductions are necessary and are unlikely to have a fish advisory.
Watershed outlines with no color indicate no available fish tissue data, while those highlighted
yellow are watersheds eliminated due to the presence of a significant estimated mercury load
from POTWs, and/or pulp and paper mills, and/or the presence of mercury mines, gold mines, or
chlor-alkali facilities. Of all the watersheds with reported fish tissue mercury concentration data,
only a few in the western third of the U.S. were not eliminated from the analysis, due to the
presence of mines and other sources

For an emission reduction rule benefits analysis, predicted percent reductions in air deposition
rates (average by watershed) would be applied to current average fish concentrations, to predict
future steady state fish concentrations at the proposed rule  emission level. Predicted average
concentration could then be compared against the points of departure for fish advisories to
determine if an advisory would be lifted. Tissue concentration reductions could also be used to
estimate incremental health benefit of reduced body burden as a result of the reductions.

In addition to the dataset-specific background information  and analysis, information on how each
data layer was processed is detailed in each section of the Appendix. It is recommended that
future users check the calculations and datasets used to determine whether all calculations and
data processing were performed correctly.
                                      Page 14 of 31

-------
                Fish  Tissue  Mercury  Concentrations
                        Averaged by Watershed
Average Fish Cone, (ppm)
\^\ 0.001 -0.149
|   10.150-0.299
QH °-300 ' °-449
    0.450 - 0.599
    0.6-3.3
    No Data
    States

    Note: New Criterion for mercury in fish is 0.3 ppm. Point of departure in fish advisories often in 0.15 ppm to 0.3 ppm range.
    Average value based on fillet samples only. See report text for details.
    Source: National Listing of Fish and Wildlife Advisories (NLFWA) Mercury Fish Tissue Database (June, 2001).


     Figure 1
                                    Page 15 of 31

-------
                    Percent Reduction in Air Deposition Load
               Necessary to Meet New Methylmercury Criterion
          Watersheds with No Other Significant Mercury Sources
% Reduction to Meet Criterion
   | Currently Meets Criterion
    10% Reduction Required
    15% Reduction Required
   j 20% Reduction Required
   | 25% Reduction Required
   | 50% Reduction Required
   | 75% Reduction Required
	| > 75% Reduction Required
    HNo Data
    States         ^ote: Watersheds highlighted yellow have significant non-deposition mercury sources defined as where the total
                 estimated load from Publicly Owned Treatment Works (POTWs) and pulp and paper mills is greater than 5% of
                 estimated waterbody delivered mercury at a typical deposition load (10 g/km2/yr), and/or where mercury cell
                 chlor-alkali facilities, mercury mines, or significant past producer gold mines are present. Additional reductions
                 would be required to meet most state fish advisory levels, which are often set below the methylmercury criterion.
                 Source: National Listing of Fish and Wildlife Advisories (NLFWA) Mercury Fish Tissue Database
                 (June, 2001). See text of report for data sources for point source dischargers and mines.
         Figure 2
                                            Page 16 of 31

-------
                   Mercury Fish Tissue Data by Year
•
o
   10000

    8000

    6000
       0

                                                               to
                                                               I   I   I   I
          1977  1980 1982  1984  1986  1988  1990  1992  1994  1996 1998
                                          Year
 Figure 3. Counts of Fish Tissue Mercury Data Records by Year.
 Source: National Listing of Fish and Wildlife Advisories (NLFWA) Fish Tissue Database
 (September, 2000).
                                 Page 17 of 31

-------
                                      References

Aastrup, M., J. Johnson, E. Bringmark, I. Bringmark, and A. Iverfeldt, Occurrence and Transport
of Mercury within a Small Catchment Area.  Water, Air, and Soil Pollution 56: 155-167, 1991

American Fisheries Society, 2000. Proceedings from the Forum on Contaminants in Fish,
October 18-20, 1999, Prepared by EVS Environment Consultants, Inc., Seattle, August 31, 2001.

AMSA, 2000.  Evaluation of Domestic Sources of Mercury. Association of Metropolitan
Sewerage Agencies (AMSA) August, 2000.  Available at:
http://www.amsa-cleanwater.org/pubs/mercury/mercury.pdf

Cadmus,  1997. The National Survey of Mercury Concentrations in Fish: Database Summary
1990-1995, September,29, 1997.  Prepared for US EPA, Standards and Applied Science
Division, by the Cadmus Group, Inc, Durham, NC 27713.

Dispasquale, M.M., J. Agee, C. McGowan, R.S. Oremland, M. Thomas, D. Krabbenhoft, and
C.C. Gilmour.  Methyl-Mercury Degradation Pathways: A Comparison Among Three Mercury-
Impacted Ecosystems.  Environ. Sci. Technol. 2000, 34, 4908-4916.

Hudson, Robert J.M.,  Steven A. Gherini, Carl J. Watras, and Donald B. Porcella,  1994.
Modeling the Biogeochemical Cycle of Mercury in Lakes: The Mercury Cycling Model (MCM)
and its Application to  the MTL Study Lakes. Mercury Pollution:  Integration and Synthesis. Ed.
Carl J. Watras and John W. Huckabee, CRC Press, Inc., Boca Raton, 1994.

Hurley, J.P., J.M. Benoit, C.L. Babiarz, M.M. Shafer, A.W. Andren, J.R. Sullivan, R. Hammond,
and D.A. Webb, Influence of Watershed Characteristics on Mercury Levels in Wisconsin Rivers.
Environ.  Sci. Technol., 1995, 29, 1867-1875.

Johansson, K., Aastrup, M., Anderson, A., Brinkman, L., Iverfeldt, A.,  1991. Mercury in
Swedish Forest Soils and Waters: Assessment of Critical Load. Water Air Soil Pollut. 56, 276-
281.

Johansson, K. and A. Iverfeldt. The Relation Between Mercury Content in Soil and the
Transport of Mercury  from Small Catchments in Sweden, in: Mercury Pollution: Integration and
Synthesis, Lewis Publishers, 1994.

Lindberg, 1996, Forests and the Global Biogeochemical Cycle of Mercury: The Importance of
Understanding Air/Vegetation Exchange Processes, in W. Baeyens et al. (eds), Global and
Regional Mercury Cycles: Sources, Fluxes and Mass Balances, 359-380.
                                    Page 18 of 31

-------
Long, K.R., J.H. Jr. DeYoung, and S.D. Ludington. Database of Significant Deposits of Gold,
Silver, Copper, Lead, and Zinc in the United States. Part A: Database Description and Analysis.
Open-File Report 98-206A.  USGS,  1998.

Maine Department of Environmental Protection, Status of Mercury Discharge from Wastewater
Treatment Facilities in Maine. Submitted to the Joint Standing Committee on Natural
Resources. January 15, 2001. DEPLW2001-5. Available at:
http://janus.state.me.us/dep/blwq/report/legisreport.htm.

Nellor, M., 1999.  Letter to Tudor Davies, Director EPA Office of Science and Technology, On
Mercury Effluent Sampling Results,  May 20, 1999.

Plouffe, A, G.E.M. Hall, and P. Pel chat (Geological Survey of Canada, Ottawa, Ontario, Canada
Kl A OE8; corresponding author: aplouffe@nrcan.gc.ca). Mercury Content of Soils in the
Vicinity of a Past-Producing Mercury Mine, Central British Columbia.
http ://www. sph.umich. edu/eih/heavymetals/Manuscripts/PlouffeA.htm

SAI, 1996. User's Guide to the Regulatory Modeling System for Aerosols and Deposition
(REMSAD). Systems Applications International, Inc. SYSAPP-96/42. September, 1996.

SAI, 1998. Development of Atmospheric Deposition Estimates of Mercury and Other
Chemicals. For E.H Pechan & Associates, Inc, for Submittal to US EPA. SYSAPP-98/02.
Systems Applications International, Inc. San Rafael, CA.

Sherbatskoy, T., J.B. Shanley, G.J. Keeler, Factors Controlling Mercury Transport in an Upland
Forested Catchment. Water,  Air, and Soil Pollution. 105: 427-438, 1998.

St. Louis, V., J.W.M Rudd, C.A. Kelly, K.G. Beaty, R.J. Flett, and N.T. Roulet, Production and
Loss of Methylmercury and Loss of Total Mercury from Boreal Forest Catchments Containing
Different Types of Wetlands. Environ. Sci. Technol.  1996, 30, 2719-2729.

Swain, E.B., D.R.  Engstrom, M.E. Brigham, T.A. Henning, and P.L. Brezonik, Increasing Rates
of Atmospheric Mercury Deposition in Midcontinental North America. Science, Vol. 257, 7
August, 1992.

Tsiros, I, 2001. A Screening Model-Based Study of Transport Fluxes and Fate of Airborne
Mercury Deposited onto Catchment Areas. Chemosphere 44, 99-107.

Tsiros and Ambrose, 1999, An Environmental Simulation Model for Transport and Fate of
Mercury in Small Rural Catchments, Chemosphere, 39(3):477-492.
                                     Page 19 of 31

-------
U.S. EPA., 1997'a.  The Incidence and Severity of Sediment Contamination In Surface Waters of
the United States. Volume 3: National Sediment Contaminant Point Source Inventory. EPA-
823-R-97-008.

US EPA, 1997b. Mercury Study Report to Congress. Volume HI: Fate and Transport of Mercury
in the Environment. EPA-452/R-97-005. December, 1997.

US EPA, 1998a. Better Assessment Science Integrating Point and Nonpoint Sources: BASINS
Version 2.0. User's Manual. EPA-823-B-98-006.

US EPA, 1998b. Better Assessment Science Integrating Point and Nonpoint Sources: BASINS
Version 2.0. Region 1 CD set. EPA-823-C-98-003.

US EPA, 1998c. Better Assessment Science Integrating Point and Nonpoint Sources: BASINS
Version 2.0. Region 2 CD set. EPA-823-C-98-004.

US EPA, 1998d. Better Assessment Science Integrating Point and Nonpoint Sources: BASINS
Version 2.0. Region 3 CD set. EPA-823-C-98-005.

US EPA, 1998e. Better Assessment Science Integrating Point and Nonpoint Sources: BASINS
Version 2.0. Region 4 CD set. EPA-823-C-98-006.

US EPA, 1998f. Better Assessment Science Integrating Point and Nonpoint Sources: BASINS
Version 2.0. Region 5 CD set. EPA-823-C-98-007.

US EPA, 1998g. Better Assessment Science Integrating Point and Nonpoint Sources: BASINS
Version 2.0. Region 6 CD set. EPA-823-C-98-008.

US EPA, 1998h. Better Assessment Science Integrating Point and Nonpoint Sources: BASINS
Version 2.0. Region 7 CD set. EPA-823-C-98-009.

US EPA, 1998L Better Assessment Science Integrating Point and Nonpoint Sources: BASINS
Version 2.0. Region 8 CD set. EPA-823-C-98-010.

US EPA, 1998J. Better Assessment Science Integrating Point and Nonpoint Sources: BASINS
Version 2.0. Region 9 CD set. EPA-823-C-98-011.

US EPA, 1998k. Better Assessment Science Integrating Point and Nonpoint Sources: BASINS
Version 2.0. Region 10 CD set. EPA-823-C-98-012.

USEPA, 2000a. Draft Florida Pilot Mercury Total Maximum Daily Load (TMDL) Study:
Application of the Everglades Mercury Cycling Model (E-MCM) to Site WCA 3A-15.  Prepared
for the United States Environmental Protection Agency and Florida Department of
                                   Page 20 of 31

-------
Environmental Protection.  Submitted by Reed Harris, Curtis D. Pollman, David Hutchinson and
Don Beals.  Tetra Tech Inc., Lafayette, CA, October, 2000.

US EPA, 2000b. Use and Release of Mercury in the United States. National Risk Management
Research Laboratory, Office of Research and Development, Cincinnati, OH.

US EPA., 2000c.  National Listing of Fish and Wildlife Advisories: Fish Tissue Database.
Available at: http://www.epa.gov/ost/fish/.

USEPA, 2000d. Guidance for Assessing Chemical Contaminant Data for Use in Fish Advisories:
Volume 1 - Fish Sampling and Analysis, Third Edition. November, 2000. EPA-823-B-00-007.

USEPA, 2001. Total Maximum Daily Load for Total Mercury in the Middle/Lower Savannah
River, GA.  February 28, 2001. USEPA Region 4.
                                    Page 21 of 31

-------
                                     Appendix A
                            Mercury Fish Tissue Data Layer
Source:
National Listing of Fish and Wildlife Advisories (NLFWA) Mercury Fish Tissue Database (as of
June, 2001) (US EPA, 2000c). The NLFWA database includes data from The National Survey of
Mercury Concentrations in Fish 1990-1995 (Cadmus, 1997), as well as other data provided by
states to the EPA Fish Advisory Program. In both cases, data were provided voluntarily by
states, in response to an EPA survey.  Data is currently uploaded by states directly to the fish
advisory program web site.

Data Processing:
The fish tissue database was cut to include only mercury concentration results and provided as a
single dbf data table.  Sample records with latitude/longitude values (georeferenced samples)
were added to the project using the Arc View Add Event Theme command.

Statistics:
The fish tissue database is apparently biased towards trophic level 4 fish (typically sport fish).
Looking at the top ten most frequent species of georeferenced samples, 12,422 (83%) are trophic
level 4 (largemouth bass, walleye, northern pike, channel catfish, yellow perch, and smallmouth
bass, in descending order of frequency), while 2,608 (17%) are trophic level 3 (common carp,
bluegill sunfish, white sucker, and black crappie, again in descending order of frequency).

Since methylmercury accumulates in fish muscle, rather than fat, skin, or organs, the manner in
which fish samples are analyzed affects the reported concentration. Using whole fish samples
will give a reduced concentration, relative to fillets, due to a dilution effect from lower
concentrations in non-fillet portions of the fish. While whole fish samples are relevant for
concerns over eco-system effects, whole fish sampling is not recommended for use in creating
fish consumption advisories (USEPA, 2000d).

Of the 21,571 geo-referenced fish tissue samples:
              17,826 (83%) samples were fillets,
              1,759 (8%) were whole, and
       •       1,986 (9%) were unknown/unspecified.
States in which whole fish sampling composes greater than 25% of all samples, are: GA, IN, KS,
MD, ME, NE,  and TX. States in which sample type is largely unknown or unspecified are MS
(unknown), and TX (year sampled, unknown). About a third of samples in GA are of unknown
sample type.

Of the fillet samples:
              3,969 specified with Skin Off,
              7,570 with Skin On, and
              5,984 did not specify.
                                     Page 22 of 31

-------
Of all 21,571 samples, only 8 indicated composite sampling, sampled at two separate locations,
all done with fillets.

Of the 21,571 geo-referenced samples:
       •      All fillet shows average concentration of 0.40 ppm;
       •      All whole shows average concentration of 0.18 ppm;
       •      Fillet Skin Off shows an average concentration of 0.40 ppm; while
       •      Fillet Skin On shows average concentration of 0.31 ppm.

In order to assess whether the fish tissue data were biased toward higher trophic level fish, fish
tissue samples were sorted by frequency according to species.  Of the top five most frequently
sampled species of georeferenced whole fish  samples, 53% were trophic level 4 fish versus 46%
as trophic level 3 fish. Of the five most frequently sampled species of georeferenced fillet
samples, 100% were trophic level 4 fish.

Thus, the large difference in concentrations from fillet to whole fish samples is certainly due in
part to the different species sampled.  However, dilution from using whole fish is still expected
to be significant and thus these samples will be removed from the analysis.

The top five most frequently sampled Fillet Skin On species were: walleye, northern pike, yellow
perch, largemouth bass, and common carp, in that order. While, the top five most frequently
sampled Fillet Skin Off species were: largemouth bass, channel catfish, white crappie, flathead
catfish, and blue catfish, in that order. Based  on this listing, it is  clear that skin removal is species
specific, and likely due to how the fish is commonly consumed.  While leaving skin on dilutes
the tissue concentration, it appears the choice to remove or retain the skin is representative of
how that fish is expected to be consumed, and thus most representative of the concentration
consumed.  That is, aft fillet samples, regardless of whether skin is specified as on, off,  or
unspecified, will be retained in this analysis.

Sensitivity to Scale (HUC8 versus HUC11) Analysis
The HUC11 watershed coverage was obtained for the Chesapeake Bay  drainage.  The HUC11
coverage contained 511 HUC11 watersheds compared to 65 in the HUC8 coverage, for the same
area, nearly a factor of 8 increase in resolution.

The georeferenced  fish tissue data were averaged across the HUC11 watersheds, and compared
against the HUC8 coverage. Of the 116 Chesapeake Bay HUC11 watersheds with fish sample
data, the median number of samples is 3 (mean of 5), with a range of 1  to 101 (26% have 1
sample). Of the 850 HUC8 watersheds (across the country) with fish tissue data, the median
number of samples is 9 (mean of 24), with a range of 1 to 959  (Everglades) samples (11% have 1
sample).  While 9 samples should provide a relatively solid statistical measure of the average
concentration in the watershed, 3 samples provides a much lower level  of statistical significance.
At the same time, the dramatic increase in 1 sample watersheds is even more problematic.  While
the HUC11 watersheds, then, show local maxima exceeding the  criterion within HUC8
                                     Page 23 of 31

-------
watersheds shown to be below that level, these high average concentrations are not likely
representative.
Notes on Data Quality and Interpretation:
This database is not a statistical sampling offish tissue across the country.  Instead, it is based on
sampling directed towards areas of suspected contamination or fishing pressure. In addition,
some states may not report data when resultant concentrations were found to be below state fish
advisory point-of-departure levels. For the above reasons, this data should not be considered a
complete characterization of the spatial distribution of mercury contaminant levels in fish.

Also, since advisories may be based on multiple pollutants, reducing fish mercury levels to below
advisory levels will not result in lifting the advisory in all cases. Sustained contaminant levels
for other pollutants could cause the advisory to remain in place.
                                       Page 24 of 31

-------
                                      Appendix B
                   Hydrologic Cataloging Unit Boundaries Data Layer

Source:
The USGS eight digit hydrologic cataloging unit (HUC-8) boundaries data layer is provided on
the BASINS 2.0 CDs (US EPA, 1998b through US EPA, 1998k) as EPA-regional coverages (in
Arc View shapefile format). Meta data is available for this data layer at the BASINS web page:
http://www.epa.gov/ost/basins/metadata/hydunits.htm

Data Processing:
Hydrologic cataloging unit boundary themes for each EPA Region were combined into a single
data layer using the Arc View merge command.  The theme was projected as Standard Albers
Equal Area projection. A field representing average deposition rate was added to the theme
attribute table. Values for this field were calculated as the area in square meters times the
"average" total mercury deposition rate of 10 ug/m2/yr (equivalent to 10 g/km2/yr). 10 g/km2/yr
is a typical low mercury deposition rate for the Eastern third of the U.S. (SAI,  1998). SAI, 1998
shows deposition rates for the Eastern third of the  U.S. at typically >5 g/km2/yr and with -95% of
the area at less than 50 g/km2/yr. The majority of the area receives deposition  in the range 5
g/km2/yr to 20 g/km2/yr.

Statistics:
Statistics on Area of HUC-8 Watersheds and a Calculated "Average Deposition Rate" Load
Statistic             Area (m2)            Ave.  Dep. Rate Load* (Ib/yr)
Count              3203
Mean               3.8E9m2                  84
Max                5.8E10m2                  1300
Min                9.6 E7m2                  2.1
St. Dev.             2.6 E9 m2
Sum                 1.2E13m2                  264,000 (132 tons)
* - Average deposition rate based on 10 g/km2/yr

Notes on Data Quality and Interpretation:
The use of an average deposition rate in screening out watersheds with significant point source
water dischargers is conservative in that urban areas typically show much higher deposition rates.
Higher total mercury discharges would also be expected in more urbanized watersheds, and thus
those watersheds would be more likely to be screened out unnecessarily by this approach. For a
benefits analysis, modeled baseline deposition rate data, averaged by watershed, could be used to
refine this screening process and thus to reduce this uncertainty. Additionally, the metadata,
referenced above, discusses the origins and original processing of the HUC-8 watershed theme
for delivery with BASINS 2.0.
                                     Page 25 of 31

-------
                                      Appendix C
    Database of Significant Deposits of Gold, Silver, Copper, Lead, and Zinc in the U.S.
Source:
This database was obtained directly from the database author (Long, et al, 1998), in excel
spreadsheet format.

Data Processing:
Fields in the spreadsheet, not needed in the project, were deleted, and the header rows were
reduced to a single row with a short, descriptive title.  The data was then exported from Excel as
a tab-delimited text file, imported to Arc View as a table, added to a View as an Event Theme,
converted to a shapefile, and converted to the project projection.  The field containing the amount
of gold in ounces, previously produced at the mine, was manipulated to remove all non-numeric
characters, and converted to number format.  This data was queried for mines having produced at
least 64,000 ounces (2 tons) of gold. The resultant selection was converted to a shapefile, and
imported to Mercury Maps as the gold_mines_sig_dep shapefile.

Notes on Data Quality and Interpretation:
The derivation and quality  of data in this database is discussed in considerable detail in Long, et
al, 1998.
                                     Page 26 of 31

-------
                                     Appendix D
     Minerals Available System/ Mineral Industry Location (MAS/MILS) Data Layer

Source:
Originally from U.S. Bureau of Mines, this database is provided as EPA regional data coverages
(ArcView shapefile format) on the BASINS 2.0 CDs (US EPA, 1998b through US EPA, 1998k).
Meta data is available for this coverage on the BASINS web page at:
http://www.epa.gov/ost/basins/metadata/mines.htm

Data Processing:
The mines themes for each EPA Region were combined into a single data layer using the
Arc View merge command and was projected to the Standard Albers Equal Area projection.
Mercury locations were identified, in separate queries, by queries of "gold" and "mercury" in any
of the five commodities fields ("COM1" through "COM5"). Mercury sites were exported as
Arc View shapefiles and re-imported to the project as separate national themes.

Statistics:
Number of Locations by Current Status
Current Status             Mercury Locations
Devel Deposit                    52
Exp Prospect                      500
Intermittent Producer
Mineral Location                  119
Other
Past Producer                     514
Producer                         27
Raw Prospect                     152
Temp Shutdown                   1
Unknown                        385

Notes on Data Quality and Interpretation:
No attempt was made to determine actual mercury loads from individual sites, or to rank
likelihood of impact based on size or type (e.g. "Past Producer" and "Raw  Prospect" were treated
as both having the potential to cause or contribute to elevated mercury contaminant levels in a
watershed). The above method is conservative, in that it may eliminate watersheds with minimal
mercury impact from mercury mines and mineral locations. Also, see metadata, referenced
above.
                                    Page 27 of 31

-------
                                      Appendix E
                         Permit Compliance System Data Layer
Source:
Permit Compliance System facility location coverage and pollutant loading tables are included in
BASINS 3.0 CDs (US EPA, 2001b through US EPA, 2001k). The PCS facility locations data is
provided as EPA regional data layers (ArcView shape file format). Loading data tables are
organized by HUC-8 watershed, with all data for all facilities and years 1990-1999. While this
loading data table format is different from the version 2.0, and the data has been updated through
1999, the data content is similar to version 2.0.  Additional detail on this theme and associated
tables can be found in the BASINS meta data available at the BASINS web page:
http://www.epa.gov/ost/basins/metadata/pcs.htm.

Data Processing:
PCS locations themes for each EPA Region were combined into a single data layer using the
Arc View merge command. The theme was projected to Standard Albers Equal Area projection.
Loading data tables (one per HUC-8 watershed) were placed in a single directory and processed,
using an AVENUE script that queries and extracts only data for "Mercury, Total (as Hg)"
(parameter 71900) or "Mercury, Total Recoverable" (parameter 71901), and places the data in a
single mercury loading data table. Loading data were linked to facility location by NPDES
identification number in order to identify, and separate out facilities that had loading data for
1995 or later, and were not publicly owned treatment works (POTWs) (SIC code = "4952").

Statistics:
Mercury  Loads for Non-POTW PCS facilities with 1995-1999 Data.
Statistic             Mercury Load (Ibs)
Count              228
Mean               14
Max                1324
Min                0.0
Var.                11669.
St. Dev.             108.0
Sum                3285

Notes on Data Quality and Interpretation:
While permits specify the use of test methods approved by the EPA, new mercury water column
test procedures (that have a substantially lower detection limit than old methods) were not
approved until July, 1999. Since permits would have to be reissued/modified to require use of
the new method, most of the data in the 1990-1999 PCS data base is likely based on the old
method.  Additionally, facilities may not use clean sampling techniques unless subject to
enforcement after high monitoring results. Instead of PCS mercury loading data, then, studies
with a statistical sample for POTWs and pulp and paper mills were located that could be used to
screen out watersheds with significant contributions from point source discharges.  Each of three
                                     Page 28 of 31

-------
PCS facility types, and associated non-PCS monitoring data sets and interpretation are discussed
below, by facility type.
Pulp and Paper Mills:
A separate theme for pulp and paper mills was created from the PCS theme by selecting facilities
with SIC = 2611 (Pulp mills), 2621 (Paper mills), or 2631 (Paperboard mills). The PCS database
includes 488 pulp and paper mill facilities, distributed as follows:
              160 Pulp mills
              244 Paper mills
       •       84 Paperboard mills

PCS Loading Data:
The PCS pulp and paper theme was joined with the mercury loading data table, and summarized
by year. For each year from 1990 to 1999, between 4 and 14 pulp and paper facilities reported
mercury loads.  The average load ranged from 5 Ibs/year to 228 Ibs/year, with the 228 Ibs/year
value appearing to be an outlier (next highest value is 22 Ibs/year and is associated with a facility
in Ohio reporting a load of 1324 Ibs, and another in NC reporting 261 Ibs in 1996).  For each of
the mill types, the data was summarized by facility (i.e.  average load over the ten year period).
Of the 27 facilities, average load was 0.0 Ibs/year for six of these facilities.
       •       Pulp mills (SIC = 2611) - average load (by year) ranged from 1.7 to 27 Ibs/year,
              with overall mean of 9.1 Ibs/yr.
       •       Paper mills (SIC = 2621) - average load (by year) ranged from 0.0 to 331 Ibs/year,
              where 7  out of 10 years, average load is less than 5 Ibs/year, with overall mean of
              35 Ibs/year.
              Paperboard mills (SIC = 2631) - average load (by year) ranged from 14 to 261
              Ibs/year, and overall mean of 44 Ibs/year.

Maine Study (1998) Data:
The PCS pulp and paper theme was queried to identify facilities in Maine (NPDES id starting
with "ME") and with SIC codes of 2611, 2621, or 2631, resulting in 17 facilities including:
              4 Pulp Mills (2611)
              13 Paper Mills (2621)
Ten of the 17 Maine PCS pulp and paper facilities have Average Limits for mercury effluent
concentrations (ME DEP, 2001).  The Average Limits are the 95th percentile probability limit on
the mean, based on a required number of samples and using EPA methods  1669 and 1631 (ultra-
clean techniques) for collection and analysis  of samples, respectively.  The average of the
Average Limit values is 13.0 ppt (ng/L). With a range of 4.5 ppt to  28.9  ppt, where 4.5 is the
minimum set by the regulation rather than that measurements.  The difference in Average Limits
by SIC is not large at 14.4 ppt for pulp mills and 12.4 ppt for paper mills. Of the 10, only three
had flow rate values in PCS: 34.0, 46.5, and  157 MOD, for an average of 79.2 MOD.
                                     Page 29 of 31

-------
In order to compare the PCS loads with loads based on average effluent concentration, a
conversion from effluent concentration in ppt (or ng/L) times flow rate in MGD to Ibs/yr is
required, as follows.
10-9 g/ng x lkg/10A3 g x 2.2046 Ib/kg x 3.785 L/gal x 10A6 gal/MG x 365 D/yr
= 0.003046 Ib/yr /(MGD - ng/L)
So the Average Limit for Pulp and Paper Mills is converted as follows
13.0 ng/L x 79.2 MGD x 0.003046 Ib/yr/ (MGD - ng/L) = 3.14 Ib/yr
So, an estimated mercury loading rate value of 3.1 Ib/yr will be applied to each pulp and paper
mill in PCS.
Publicly Owned Treatment Works (POTWs):
From the PCS data coverage, a query for SIC = 4952 resulted in a coverage of 23,629 POTW
facilities in the conterminous U.S.  For each watershed, then, the sum of the flows from POTWs
were summed, and multiplied by a mean effluent mercury concentration to get an estimated
direct waterbody discharge load  estimate (in Ibs/yr) (see Pulp and Paper Mills section for
conversion factor). A study by the Association of Metropolitan Sewerage Agencies (AMSA)
found an average concentration of 7 ppt in wastewater treatment plant effluent (Nellor, 1999).
This data is based on AMSA's study of 24 POTW facilities in six states, using clean sampling
and analytical techniques, for facilities with a range of flow rates from 0.65 MGD to 225 MGD,
serving populations ranging from 18,2000 to 1.74 million (median population - 384,000).

Applying this average concentration to the reported flow rate of each facility resulted in a mean
watershed load of 0.69 Ibs/yr, with a range from 0.0 to 128 Ibs/yr, and a total load of 472 Ibs/yr,
for the 681 watersheds with POTWs.  Based on this screening analysis, 79 watersheds had total
estimated POTW mercury discharge loads greater than 5% of an estimated typical air deposition
load delivered to waterbodies (i.e. 20% of a "typical" 10 ug/m2/yr deposition rate) or 1% of the
typical air deposition rate times the watershed area. All watersheds screened out by this
approach are located west of the Mississippi River. In these 79 watersheds, POTW effluent may
constitute a significant source of mercury to receiving waterbodies and thus these watersheds will
be eliminated from the analysis.

For comparison purposes, the data on POTW mercury loads in PCS are discussed below.  Of all
POTW facilities, 1946 have PCS mercury loading data with a range of average annual loads  (all
facilities per year) from 34 Ibs/yr to 1,200,000 Ibs/yr, with a maximum reported value of
1,000,000,000 Ibs/yr from a facility in GA. Of all reported mercury loading values (not
summarized by year or facility),  99% are 1000 Ibs/yr or less, while 37% are zero.  While there
has been a steady increase in the number of facilities reporting mercury loads, from 1990 to
1999, loading does not appear to have a specific pattern with respect to time. While one would
expect average loading rates to go down, as new mercury analytical and clean sampling
techniques came into use in the late 90's, data from 1997 and 1998 have the highest average
loading for the entire ten year period.  In sum, mercury loading data in the PCS database has
                                     Page 30 of 31

-------
some obviously suspect figures, as well as generally having reported loading rate values much
higher than values reported in the AMSA study.
Mercury Cell Chlor-Alkali Production Facilities:
The PCS theme was queried for the 14 mercury cell chlor-alkali facilities listed in the Mercury
Study Report to Congress (MSRC, 1997), and developed into a separate theme. The mercury cell
chlor-alkali facilities theme was linked to the PCS mercury loading data table.
Joining the mercury cell chlor-alkali theme with the PCS mercury loading data shows an average
annual load of 39.7 Ib/yr, with a minimum of 0.0 Ib/yr and a maximum of 1659 Ib/yr (one year
maximum, Occidental, Muscle Shoals, AL). The next highest value at that plant was 53.0 Ib/yr
and the next highest at any other plant was 152 Ib/yr.
While the PCS data is generally thought to be unreliable due to the use of non-ultra-clean
sampling and analytical techniques, no other data source, sufficient to derive a reliable screening
level estimate, was found. Evidence of faulty data in PCS can be found by comparing reported
loading rate data for the Olin Corp facility in Augusta, GA. Reported at 32.5 Ib/yr mercury
effluent load in PCS, this same facility was sampled recently, as part of the Savannah River
mercury TMDL and estimated to discharge 0.95 Ib/yr (EPA, 2001).
                                     Page 31 of 31

-------
                          Mercury Maps
A Quantitative Spatial Link Between Air Deposition and Fish Tissue
                       Internal Peer Review
                    Response to Comments
                             Paul Cocca
                  Standards and Health Protection Division
                     Office of Science and Technology
                           Office of Water
                              8/17/01

-------
la) Is the reduced-form model, presented in this report, an accurate characterization of the
air deposition load / fish tissue concentration relationship, predicted by the IEM-2M and
MCM models at steady state?

Comments:
Two reviewers declined to comment, stating it was out of their area of expertise.

la-1) One reviewer, familiar with the two models, stated that the models are in fact linear at
steady  state.

la-2) Another reviewer stated that the models are similar at steady state, though there would
need to be validation  and verification to confirm they are equivalent. The reviewer suggested, in
particular, looking at  the sensitivity of the two models (i.e. Mercury Maps model versus IEM-2M
and MCM), and how  they react to perturbations to same model components.
Response:
la-1) The acceptance of the reduced-form model, by the one reviewer, is of substantial weight, as
this reviewer has significant experience with the IEM-2M and MCM models from their work on
the Mercury Study Report to Congress (MSRC) (EPA, 1997).

la-2) In response to the other comment, the application of the E-MCM model in the Everglades
Pilot Mercury TMDL (USEPA, 2000), is compared below with what Mercury Maps would
predict for the same site.

The Everglades Pilot Project Mercury TMDL (USEPA, 2000), determined the relationship
between atmospheric Hg(n) deposition and long-term fish tissue mercury concentrations, in
Everglades Site WCA 3A-15. To determine the relationship, the researchers first calibrated the
E-MCM to the WCA 3A-15 site and ran the model for 100 years to achieve an approximate
steady state response to the current air deposition loading rate.  They compared these long-term
predicted results against current measured water, sediment, and fish concentrations and found an
acceptable match. They then reduced the mercury deposition load,  as well as watershed mercury
inflows, and ran the model for an additional 100 years; performing this process separately for
25%,  50%, 75%, and 85% reductions in overall loads.  In plotting the results of the above
analysis, the researchers found the relationship between fish mercury concentration and mercury
air deposition rate to be linear, and to fit the equation: Y = 0.9408X + 0.0611, where Y = the
fraction of current fish concentration; and X = fraction of current air deposition load (wet and
dry).  By comparison, the Mercury Maps model simply states Y = X (i.e. slope =1, and intercept
= 0).  The reasons for the discrepancy between the two models, as well as the effect on predictive
accuracy of the Mercury Maps model, are discussed below.
                                      Page 1 of 25

-------
The researchers note the reason for the slope and intercept, of the fitted equation, not being equal
to one and zero, respectively, is that even after the initial 100 year simulation, deep sediment
concentrations had not yet reached a steady state response to current loading. The sediment
concentrations are elevated with respect to true steady state, and thus represent an additional load
beyond the reduced air deposition load. The researchers also state that were the simulations
carried out for much longer periods (as long as thousands of years), predicted concentrations
would approach a direct proportional response (i.e. Y = X).

The E-MCM model result, being not quite directly proportional, can be shown to have a
quantifiable effect on the predictive accuracy of the Mercury Maps model, were it also applied to
the Everglades WCA 3 A-15 site. For a 50% reduction in air deposition load, the Everglades Hg
TMDL application would predict a 47% reduction in fish concentrations, to Mercury Maps' 50%
reduction.  Higher proportional reductions result in larger prediction errors, e.g. a 90% air
deposition load reduction causing a 84% fish tissue reduction in E-MCM, versus a 90%
reduction in Mercury Maps.

In conclusion, sediment concentrations may, in some instances like the Everglades WCA 3A-15
site, be elevated with respect to current air deposition loads. These elevated concentrations,
unaccounted for in Mercury Maps, act as an additional source and, where significant, will cause
Mercury Maps to over-predict reductions in fish tissue concentrations. In the case of the
Everglades site WCA 3A-15, the quantified over-prediction is relatively small, on the order of up
to five percent at high air deposition rate reductions. In addition, it can be reasonably expected
that similar situations  exist in other watersheds throughout the U.S.,  where historic atmospheric
mercury  deposition and continued mercury accumulation in sediments may cause current
sediment concentrations to be elevated with respect to current deposition levels, but that these
elevated  levels will cause fish concentrations to depart only slightly from the direct proportional
reduction used in Mercury Maps.  This small difference is likely reasonable for the purposes of
benefits assessments.

In addition to the Everglades pilot mercury TMDL study, a TMDL was performed and finalized
for the Savannah River using models that are similar to IEM-2M and MCM. In the TMDL
Determination section of this report (USEPA, 2001), the authors state that "Because the water
column mercury concentration response is linear with  respect to changes in load, a proportion
can be developed to calculate the total maximum mercury load from the watershed that would
achieve the derived water quality standard ..." The TMDL was determined by taking the ratio of
the existing stream concentrations to the water quality standard, and  equating that to ratio of the
current load to the TMDL load. And, in relating the TMDL stream load back to atmospheric
reductions (atmospheric sources accounted for 99% of the total load), no differentiation was
made between the percent reduction in atmospheric load to the watershed and the percent
reduction in atmospheric load delivered to the waterbody, implying a linear relationship.
                                      Page 2 of 25

-------
Ib) Is it accurate to say that in watersheds where the mercury load to water bodies is
dominated by air deposition, mercury concentrations in fish tissue are expected to reduce
in direct proportion to reductions in mercury deposition, at steady state?

Comments:
This question received the most discussion from the reviewers.
lb-1) A key issue raised by two of four reviewers is that other sources, if present in significant
quantities, would cause the relationship to be non-linear, and/or to not go through the origin (i.e.
fractional fish tissue concentration reductions would be of the form: a + b*[%_air_reduction]).
One reviewer pointed  out that not taking into account background sources constitutes a negative
margin of safety.  That is, the proportional reduction method would predict larger reductions than
would actually occur, were other sources to be of sufficient magnitude.  Potential other sources
listed by the reviewers included:
Ib-lA) Weathering and dissolution of mercury from bedrock;
Ib-lB) Vegetative capture of inert atmospheric elemental mercury; the reviewer states that
current air deposition models do not take this into account, and that elemental mercury is the
form of mercury that will contain the highest fraction of background mercury (i.e. from natural
and international sources) thus making this a potentially large source; and
Ib-lC) Non air deposition sources identified in the project as potentially significant (e.g.
mercury and gold mines, and mercury cell chlor-alkali facilities); the reviewer stated that
including a more sophisticated equation/model would allow these watersheds to be included in
the analysis.

lb-2) A third reviewer pointed to a recent study, published in ES&T that found an increased
demethylation rate in areas of high environmental mercury levels. That is, net methylation may
be lower in areas of high mercury concentrations, and thus fish tissue levels would be higher that
expected under a proportional decrease.

lb-3) Finally, a fourth reviewer pointed out that what might seem to be steady state conditions,
found while monitoring, could be upset by a number of extreme environmental/meteorological
events causing the release of mercury locked-up in the watershed, having been set there by
historical deposition. For example, extreme high river flows could cause resuspension of deep
riverine sediments containing elevated mercury concentrations.  Another example is re-
inundation of the littoral zone during the return to normal water levels following an extended
drought, releasing mercury bound up in vegetation and soils as it's re-exposed to a  methylating
environment.
Response:
It's important to distinguish between the two different types of non-linear responses mentioned
by reviewers. In one case, discussed in response lb-1, other sources cause the plot offish tissue
concentrations versus air deposition rates to not go through the origin, thus causing a simple
proportional reduction to over predict actual reductions in fish tissue concentration.  In another


                                      Page 3 of 25

-------
case, discussed in response lb-2, non-linear reaction rate kinetics cause biochemical processes to
alter methylmercury concentrations in the environment in a manner which is not linear with
respect to total load.  Finally, as discussed in lb-3, time to steady state may be exceedingly long,
or appear to have been achieved, only to be upset by the release of environmentally sequestered
mercury.
lb-1) In cases where air deposition is not the sole significant source, as in the Everglades
example cited previously, the equation describing fish tissue fractional reduction would need to
be modified as follows, as per the suggestion of the reviewer:
Eqn. 1.
                       \A4wj  + ^Other t
Where:
       Cf1Sh2 = concentration of mercury in fish at time 2
       Cfishi = concentration of mercury in fish at time 1
       L^ = air deposition mercury load to waterbody at time 2
       LAJJ.J = air deposition mercury load to waterbody at time 1
            = non air deposition load of mercury to waterbody
and which reduces to the simple proportional reduction, when L^,. = 0.

For the above equation, it can be shown that:


                                         L
    2-
                                      \ •                        \
                       (LAiri+L0ther)   ^Atr,    (LAiri+L0ther)
which is of the form Y = mX + b
where:
       Y = fractional reduction in fish tissue concentration;
       X = fractional reduction in air deposition load;
       m = slope; and
       b = y-intercept.

It can be shown that, for equation 2, m + b = 1 . As well, the intercept (b) is just the other load as
a fraction of total load. In the case of the Everglades mercury TMDL Pilot Project, m = 0.94,
while b = 0.06.  That is,  6% of the total load is non-air-deposition load.

The method employed in this project, then is to eliminate from the analysis those watersheds
where sources other than air deposition are significant, thus keeping b close to zero, and m close
to one, reducing errors in using the direct proportion methodology (i.e. Y = X).  Below, is the

                                       Page 4 of 25

-------
response to the possibility raised by comments lb-1 A and Ib-lB that there may be potential
sources not taken into account by this project, as well as, in response to comment Ib-lC, further
explanation of the reasoning behind eliminating known or likely significant point sources from
the analysis. Based on the studies summarized briefly in response lb-1 A, below, it appears
unlikely that background sources contribute a significant fraction of the current load. Also, as
discussed in response la-2, the error likely to have been introduced, by not taking into account
the effect of background load on predicted response,  is small relative to the prediction.  Finally,
this error may serve to offset the under-prediction of benefits discussed below in Ib-lC.

Ib-lA) A number of studies have looked at the potential relationship between mercury
concentrations in soil and runoff, as well as soil/bedrock erosion as a potential mercury source.
St. Louis, et al, 1996  estimated, based on comparisons with erosion rates of other minerals, that
mercury weathering rates could range from 0.4% to 6% of total mercury inputs to the watershed.
They deemed this amount insignificant for the purposes of the mass balance in their study.
Finally, Aastrup, et al, 1991 quantified mercury accumulation and transport in a number of soil
layers in a watershed, as fractions of total mercury deposited. By the use of a mass balance they
demonstrated that the content and fluxes of mercury, in interflow and groundwater flow, is
accounted for via the percolation of atmospherically deposited mercury through the soil. In
addition, while weathering and dissolution of mercury from bedrock may be significant in some
areas, these areas would likely be associated with mercury mining locations, and would be
screened from the analysis on that basis.

Ib-lB) The scenario described by the commenter, is one in which a deposition pathway (in this
case, vegetative capture of inert atmospheric elemental mercury, reflective of impacts from
background atmospheric mercury levels), is not taken into account. As deposition from local and
regional sources  diminishes, in response to source controls, the relative contribution from
background would become greater. As described above in equations 1 and 2, if this deposition
pathway were not taken into account, it would act as  an unaccounted for source, causing Mercury
Maps to over-predict an expected reduction in mercury concentrations  in fish tissue.  While this
scenario is plausible,  there are several reasons why it's not directly relevant to the Mercury Maps
project.

In the case of adsorption of elemental mercury by vegetation, the air deposition model REMS AD
does take into account vegetative properties in estimating dry deposition.  The deposition
velocity is the inverse sum (resistances in series) of aerodynamic, boundary layer, and surface
resistances. Stomatal resistance is included as a component in surface  resistances (SAI, 1996).
It should be noted that REMSAD does not take into account the reverse and balancing process of
volatilization of mercury from either the land surface or vegetation. Since this would tend to
increase net flux of elemental mercury to watershed surfaces, it would  appear that REMSAD may
in fact over-predict the impact of background elemental mercury levels thus diminishing the
impact of emission controls on deposition rate and thus fish concentration.
                                      Page 5 of 25

-------
In addition, while the accuracy of predictions offish tissue reductions in Mercury Maps will be a
direct reflection of the accuracy of the air deposition model in predicting relative rates of air
deposition of mercury to watersheds, this project is not dependent on the use of any particular
deposition model. Rather, this project is intended to be coupled with the air deposition model
thought best able to predict percent reductions in mercury deposition on a national scale.

Finally, it should be noted that any potential for non-linearity in the selected air deposition model
should not be confused with the potential for non-linearity in watershed fate and transport. The
air deposition model need not be linear with respect to source loadings to the atmosphere in order
to be used in conjunction with Mercury Maps.

Ib-lC) The approach suggested by the reviewer is outlined in response lb-1, discussed above.
Based on additional analyses of point source data, discussed in section 2c-NA and the Appendix,
a new approach of eliminating watersheds where the expected sum of loads from pulp and paper
mills and POTWs exceeds 5% of expected waterbody-delivered air deposition loads,  as well as
eliminating watersheds with mercury mines, significant gold mines, and/or chlor-alkali facilities,
is used. While the commenter's model, presented above, would allow other significant sources,
to be included in the analysis, the quality in the data characterizing loads from these sources is
insufficient for use in other than a screening level approach. Loads from POTWs and pulp and
paper mills are simply estimates based on the product of facility flow rates and average measured
mercury effluent concentrations.  There was insufficient data to use a similar screening level
approach for mercury cell chlor-alkali facilities and estimating loads from abandoned mines
would likely require numerous high quality samples.  This lack of data quality was a key
determinant in selecting an  approach that included screening out watersheds with significant or
potentially significant non-air-deposition sources. While a national statistical sampling of all
permitted discharge facilities, by industry, would provide an adequate base of information to
improve on this approach, no such comprehensive data set as of yet exists.  In addition,  it would
be extremely difficult and resource intensive to attain  a sufficiently precise nationwide data set
on mercury loads from other nonpoint sources such as abandoned mines and bedrock erosion.

It should be noted, however, that for the use of Mercury Maps in an emission reduction benefits
analysis, the effect of using the watershed elimination method is to under-predict the overall
benefits. The extent of this under-prediction can be bounded on the upper end as the percentage
of all  watersheds that are  eliminated (i.e. if 20% of all watersheds are eliminated, the under
prediction is, at a maximum, 20%).  It should also be noted that this expected under-prediction
should be within the realm of uncertainty brought in by other aspects of a benefits  assessment.
This approach buys  a higher degree of certainty at the expense of a potentially higher benefits
calculation.
lb-2) The ES&T article (DiPasquale, et al, 2000) found that demethylation rates increased with
increased methylmercury concentration, in waterbodies with extreme mercury contamination,
caused by historic mining practices. In addition, the researchers found that demethylation
                                       Page 6 of 25

-------
decreased over time as methylmercury concentrations in samples decreased, for all three test
areas, including the Everglades.  While this study does appear to provide real evidence of a
nonlinear process with respect to mercury concentration, it is not clear whether net
methylmercury bioaccumulation in fish is expected to be a nonlinear function of mercury
concentration as a result. Potentially, additional nonlinear  processes could serve to cancel out the
nonlinearity in demethylation, resulting in a linear net bioaccumulation function.  Finally, the
correlation of higher degradation rates with higher contaminant levels does not necessarily
connote a causality. There may be independent factors causing both.

However, were nonlinear demethylation rates to affect net bioaccumulation in this way (i.e. to
cause fish tissue concentration reductions to slow with decreased load) the proportional reduction
approach in Mercury Maps would then overpredict reductions in fish tissue. That is, the potential
implication of DiPasquale, et al, 2000 is that more dramatic  deposition reductions, than those
derived in the current formulation of Mercury Maps, would be required in order to achieve the
methylmercury criterion in all watersheds. Conversely, for a given technological standard based
emission reduction, the estimated benefits would be reduced.
lb-3) While extreme events may unlock mercury from environmental pools, this is part of an
expected uneven response over time to reduced mercury loadings to watersheds and waterbodies.
Also referred to as dynamic equilibrium, steady state in environmental systems means that
concentrations may vary season to season or even year to year, but that long term averages are
constant. While environmental media mercury concentrations are expected to trend downward,
random fluctuations in meteorological and other environmental patterns will cause uneven
responses.  The Everglades Mercury Pilot Project TMDL (discussed above in response to la-2) is
a good demonstration of how mercury from historic deposition, can remain in the sediments for
long periods of time before achieving true steady state.
2a) Can the fish tissue data be used for the purposes outlined, given its origins, quality, and
completeness?

Comments:
2a-l) Reviewers pointed out that the average fish concentration across a watershed is not
necessarily representative of the population offish eaten, nor of the true average for their
watershed. Reviewers mentioned a number of potential separate causes:
2a-lA) States might sample predominantly for sport fish (i.e. trophic level four fish), though
people likely consume a wider variety offish including those in trophic level three;
2a-lB) The fish sampled may not be truly representative of the true population average across the
HUC;
2a-lC) Some fish consumers may favor fish from a particular waterbody, i.e. one that is a local
maxima, and thus could be unduly exposed to a higher than acceptable body burden.
                                      Page 7 of 25

-------
2a-2) In addition, one reviewer pointed out that different sampling procedures (selection of site
and fish species) in different jurisdictions could distort the regional picture.

2a-3) Finally, another reviewer pointed out that not all samples will be of the same type (i.e.
fillets versus whole fish versus composites) and stated that they should be uniform across the
analysis.

Response:
2a-l) It is true that the fish sampled are not necessarily representative of the population offish
actually consumed, which could potentially introduce certain biases in the analysis.  The stated
purpose of this product is to establish a method of quantitatively predicting the spatial
distribution and extent of reductions in fish tissue mercury concentrations, in response to a given
spatially distributed reduction in air-deposited mercury watershed loads. Simply stated, then,
Mercury Maps will predict concentration reductions for those fish tissue data used.  The
commenter's suggestions are most relevant to inform the intended use of Mercury Maps,
particularly for the intended uses: assessment of benefits of reduced mercury emissions. The
responses below are intended to show how biases and potential biases in the fish tissue data used,
may influence the use of Mercury Maps in benefits assessment. It should be noted, there are two
entirely different approaches to performing a benefits assessment for reductions in fish tissue
mercury concentrations: 1. Estimate reductions in fish advisories and thus reduced economic
impact; or 2. Estimate reduced health impact from reduced fish tissue contaminant
concentrations. In addition, another potential use of Mercury Maps is in performing a regional or
national TMDL analysis, and/or in developing a risk-based mercury emission reduction rule. In
both cases, local maxima in fish tissue mercury concentrations are of primary concern.

2a-lA) The fish tissue database is apparently biased towards trophic level 4 fish (typically sport
fish). Looking at the top ten most frequent species of georeferenced samples, 12,422 (83%) are
trophic level 4 (largemouth bass, walleye, northern pike, channel catfish, yellow perch, and
smallmouth bass, in descending order of frequency), while 2608 (17%)  are trophic level 3
(common carp, bluegill sunfish, white sucker, and black crappie,  again in descending order of
frequency). One of the intended uses of the Mercury Maps project is as  a component in a
quantitative benefits analysis for national technology-based air emission reduction rules. For use
in such a benefits analysis, the Relative Source Contribution (RSC) approach suggested by the
reviewer would appear unnecessary, since the percent reduction in concentration is expected to
be the same, regardless of species, trophic level, or average concentration. However, were the
benefits analysis to take into account changes in diet  (e.g. as a result of lifting offish
consumption advisories), use of a RSC approach would be reasonable, given the clear bias in fish
sampling.

2a-lB) Since these data are used by States in making decisions on whether to post fishing
advisories, the fish sampled may not be truly representative of the true population average across
the HUC. These samples are in fact generally based on areas that are most heavily fished
(anglingpressure) and/or those that are suspected of having higher than average potential to be


                                      Page 8 of 25

-------
polluted (AFS, 2000). That is, the average concentration may be biased higher with respect to
the true average, but would be expected to be more reflective of the average concentrations in
consumed freshwater fish. While a statistical sampling would produce a less biased average
concentration, the goal of protecting human health is better served by sampling with a bias
towards areas, that with sound scientific reasons, are suspected of having higher than average
concentrations, or that account for a disproportionate amount offish caught and consumed. The
manner in which fish tissue is sampled is a decision made by the state agency.  Additional
analyses on the fish tissue data are included in the Appendix.  It should be noted that the median
number of samples per HUC is 9 (mean is 24), which is a reasonable statistical sample on which
to base a mean.

2a-lC) The reviewer notes that were a waterbody, with fish tissue concentrations elevated with
respect to the HUC-wide  average (i.e. a local maxima), to be the primary source offish for a
particular consumer group, these consumers could be unduly exposed to a higher than acceptable
body burden. While an average HUC-wide concentration is not representative of any particular
waterbody within the watershed, the few samples taken from any particular waterbody are less
likely to be representative of the average concentrations in that particular waterbody. This
aggregation of the data is appropriate given the density of the coverage.  That is, at a median
number of samples per HUC of 9 (see Appendix), the mean can be expected to reasonably
represent the watershed average much better than one to a few samples represent a waterbody
average.

2a-2) As discussed in 2a-lb, the average concentration may be biased higher with respect to the
true average, but would be expected to be biased towards areas that are most heavily fished
and/or those that are suspected of having higher than average mercury concentrations. In that
way the sampling site selection procedures are likely similar.  On the other hand, due to
differences in species abundance, fishing preferences, and consumption preferences, one would
expect States to adopt sampling practices that were different from their neighbors, but most
appropriate for their fishing citizenry. For this reason, data collected for a single purpose, would
be a much better measure of the actual distribution of mercury in fish throughout the country.
But for the purposes of evaluating the benefits of reduced mercury concentrations in fish,
particularly regarding the elimination of fish consumption advisories, the expected biases of the
data may actually improve its utility.

2a-3) In response to the commenters note that all samples were likely not of the same type and
that the data should be filtered to a single type for this analysis, a statistical analysis of the
different types was performed.  It was found that there were large differences between fillet and
whole fish concentrations (0.40 ppm vs. 0.18 ppm, respectively), as well as between fillets with
skin off versus fillets with skin on (0.40 ppm vs. 0.31 ppm, respectively).  It was also shown that
these  differences in concentrations may be due in large part to differences in species sampled.
Whole fish samples are more frequently trophic level 3 fish, while having skin on or off appears
to be a function of how a  particular fish species is typically consumed (see Appendix for details).
Based on this analysis, and because for the purposes of benefits assessment, TMDLs, and risk-


                                      Page 9 of 25

-------
based emission control rules, the consumed concentration is most relevant, whole fish samples
are eliminated from the analysis, and all fillets (skin on and skin off) are included. All unknown
and unspecified sample types are eliminated from the analysis.

2a-NA) Since the September, 2000 cut, the Fish Advisory fish tissue database was updated. An
additional 1,400 samples were added to the database since that time. These additional records
were added to Mercury Maps fish data, making it current as of June, 2001.

2b) Is the scale of the analysis appropriate?

Comments:
Two reviewers did not address this question.
The response of the remaining two reviewers, toward the HUC8 scale of analysis, however, was
positive with one reviewer stating that "the scale of the analysis gives a good picture of where
reductions in mercury deposition are needed" and another pointing out that the USGS is using a
similar scale in their analysis. However, one reviewer did point out that a watershed-wide
average may mask local minima and thus not be protective of consumers who get their fish from
a single water body or subwatershed area with higher than average concentrations.  Another
reviewer pointed out that the USGS is moving toward a finer resolution analysis, such that they
can evaluate issues of scale by aggregating data at different scales, thus addressing the issue of
scale up front by performing the analysis at the finest resolution and addressing the issues of
scale simply by viewing the results of different levels of aggregation.

Response:
A preliminary analysis of the effect of using a finer resolution watershed coverage was
performed. The HUC-11 watershed coverage, readily available for the Chesapeake Bay
watershed, was added to the project, and fish tissue data were aggregated to that level of
resolution.  The resulting map of the average fish tissue concentration relative to the
methylmercury criterion showed a few local maxima that had been smoothed out at the higher
resolution HUC8 watershed scale, as is to be expected. An analysis, presented in the Appendix
(fish tissue database section), however, shows that the median number of samples in the HUC11
watersheds is 3, while the median in HUC8 watersheds is 9. That is, while the fish tissue
database is substantial, it is not large enough to justify averaging  across watersheds smaller than
the HUC8 level. This effect was also discussed in response  2a-lC, with respect to waterbody
specific analyses.  Analyses at smaller levels is justified on a case by case  basis, and may be
appropriate in a sentinel-watershed approach associated with a risk-based  air emission reduction
rule analysis, or a TMDL.
2c) Is the use of all other data layers in addressing the goals of the project, appropriate,
taking into account their origins as well as their quality and completeness?

Comments:


                                     Page 10 of 25

-------
Two reviewers did not address this question other than to say that data quality and completeness
are important and should be taken into account.
2c-l) One reviewer pointed out that the screening of watersheds for NPDES permitted mercury
discharges, took 5% of the deposition load to the watershed, but that it should rather take 5% of
the air deposition load delivered to waterbodies. The reviewer stated that the percent delivered is
on the order of 1-10% of the total deposition.
2c-2) Another reviewer suggested taking into account pulp and paper mills, that though they are
not likely significant mercury dischargers now, they may have been earlier, and thus affected fish
tissue samples from 1990 to 1995.

Response:
2c-l) The reviewer correctly pointed  out that pollutant discharge loads to waterbodies were being
compared  directly with air deposition loads to watersheds. The following is a review of the
literature on the percent of total atmospheric deposited mercury which is transported from
watersheds to receiving waterbodies.
       - Aastrup, et al, 1991 found yearly mercury transport to a lake from a forested
       subcatchment to be 17% of total mercury deposition.
       - Lindberg, 1996 found the combination of runoff and leaching to total 3.5% of total
       atmospheric load of mercury to the Walker Branch watershed, TN.
       - St. Louis, et al, 1996 found that for five catchments in the Experimental Lakes Area
       (ELA) in northwestern Ontario, export of total mercury ranged from an average of 29.5%
       for a basin wetland to 61.1% for a riverine wetland.
       - Sherbatskoy, et al, 1998 found an export rate of 6% of total mercury from a small
       forested catchment in Vermont.
       - Swain, et al, 1992 found for catchments to seven headwater lakes in Minnesota and
       Wisconsin, the proportion of atmospheric mercury transported from catchment to lake to
       be  26% and 22% for modern and pre-industrial times, respectively.
       - Hurley, et al, 1995, found watershed total mercury transfer efficiencies, for 39 river sites
       in Wisconsin to range from Fall means of 0.5% to 8% (depending on watershed type) to
       Spring means of 29% to 90%.
       - Tsiros and Ambrose, 1999 found delivery of mercury to canals in the Everglades
       Agricultural Area was 23% of atmospheric deposition.
       - Johansson et al. (1991) reported mercury transport fluxes from several small watersheds
       to be about 30% of atmospheric deposition.
       - Tsiros, 2001, in model simulations, found that mercury runoff flux was 2 to 3 times
       higher than normal during wet years, and 5 to 7 times lower than normal during dry years,
       and that mercury runoff flux was 18% to 61% of atmospheric deposition for wet years,
       and 1% to 4% of deposition for dry years. These values correspond to runoff flux for
       normal years to be between 6% and 30% of deposition.

Based on a review of the above studies, it appears that a delivery ratio of 20% is a reasonable
estimate of the central tendency of this value, an appropriate estimate for the purposes of this
study.  That is to say, on average, it's expected that only 20% of air deposited mercury reaches


                                     Page 11 of 25

-------
waterbodies on a long-term average annual rate. Rather than comparing water discharge loads to
5% of air deposition, then, they will be compared to 20% of 5% or 1% of the typical air
deposition rate.

2c-2) In response to this comment, pulp and paper mills were also considered as potentially
significant sources.  In addition, due to known uncertainties in PCS data for mercury, a new
methodology for screening out sources was developed. A detailed review of a study conducted
by the Maine DEP (ME DEP, 2001) found 3.1 Ib/yr to be the expected loading rate for both pulp
mills and paper mills. That is, the 14 pulp and paper mills in the study had an average
concentration of 13 ppt, and an average flow rate of 79 MGD (see Appendix for details). By
contrast, PCS data showed average loading data of 10 Ib/yr for pulp mills and 40 Ib/yr for paper
mills (see Appendix). As the PCS data is likely based on a non-ultra-clean techniques it will not
be used. While of the 14 current chlor-alkali facilities, only one is classified as a pulp or paper
mill, pulp and paper mills more commonly used in-house mercury cell chlorine production
facilities in the past.  Past mercury sampling and analytical techniques, in which the use of non-
ultra-clean techniques resulted in sample contamination, and falsely high results may also have
influenced the perception of pulp and paper mills as significant mercury dischargers. The mid-
90's date is the time in which ultra-clean sampling techniques were becoming recognized as
producing more accurate mercury water column concentration data.

2c-NA) In addition, though not discussed by the reviewers, additional data analyses were
performed in order to address data quality issues identified in the PCS data.  The watershed
screening methodology was improved by using additional data sources for POTWs, chlor-alkali
facilities, and gold mines.

Publicly Owned Treatment Works (POTWs):
Using a study conducted by the Association of Metropolitan Sewerage Agencies (AMSA) on
mercury in POTW effluent, a sensitivity analysis was performed to examine the relative loading
from POTWs versus air deposition delivered to waterbodies.  Nellor, 1999 cites an average
mercury concentration in POTW effluents of 7.25 ppt (ng/L).  This data is based on AMSA's
study of 24 POTW facilities in six states, using clean sampling and analytical techniques, for
facilities with a range of flow rates from 0.65  MGD to 225 MGD.  The range in mercury effluent
concentrations was 0.7  ppt to 69.9 ppt, with a median value of 5.0 ppt.

For the purposes of this study, the mean value of 7 ng/L was applied to each POTW, at the PCS
reported flow rate, and  summed their cumulative load (across HUCs). This loading rate is
compared to a typical air deposition load of 10 ug/m2/yr, with an assumed 20% delivery to
waterbodies. If the sum of POTW mercury loads is greater than 5% of the air deposition load, as
delivered to waterbodies, then the watershed is screened from the analysis.  79 watersheds were
screened out solely on the basis of this procedure,  all located west of the Mississippi River.
Additional details on this analysis, as well as on the mercury data available in the PCS database,
are available in the Appendix.
                                      Page 12 of 25

-------
Chlor-Alkali Facilities:
The screening Procedure for chlor-alkali facilities was also examined and improved as follows.
The Mercury Study Report to Congress (MSRC) (EPA, 1997) notes that only a fraction of chlor-
alkali facilities use the mercury cell process. MSRC lists the 14 facilities using the process at
that time and notes that no new mercury cell chlor-alkali facilities are planned to be built, while a
recent study (EPA, 2001) found a much lower loading rate than reported in PCS (see Appendix
for details).  However, because, recent data were available for only one mercury cell chlor-alkali
plant, there were insufficient data to assign an average value across all plants, in a screening
approach, similar to that for pulp and paper mills. Instead, the simple presence of a plant will be
used to screen out watersheds.

GoldMines:
Locations with gold as a commodity in the MAS/MILS database included numerous locations
that were likely insignificant sources and may have never used the mercury amalgamation
process.  An alternate database, the USGS Database of Significant Deposits of Gold, Silver,
Copper, Lead, and Zinc in the United  States was used instead (Long, et al, 1998).  This database
was queried for mines that have produced more than two tons of gold. Any watershed containing
one or more of these mines was screened from the analysis, on the basis of potential historical
contamination of the watershed due to use of the mercury amalgamation process.

Mercury Mines:
Plouffe, et al found elevated mercury concentrations in soils surrounding mercury mines, at
distances of up to 20 km to 40 km.  Thus, it appears reasonable to suspect mercury mines as
likely sources of elevated environmental concentrations of mercury.
3) Can the methods developed in this project be used to quantitatively assess impacts of air
deposition reductions on fish tissue in air deposition dominated watersheds?

Comments:
One reviewer simply said "yes, I think so." Another reviewer did not address this question.

3-1) A third reviewer said this approach cannot be used with certainty due to current air
deposition models not taking into account orographic effects, and that the only way to know true
deposition rates of mercury is to sample them.

3-2) A fourth reviewer suggested a modification to the proportional reduction equations as shown
in section lb-1. The reviewer stated that this equation should be used instead, and that
watersheds with other sources should then not be excluded from the analysis.

Response:
3-1) While it's true that measured deposition rates would be considered a more accurate estimate
of true deposition than a modeled result, air deposition models can be compared to and calibrated


                                      Page 13 of 25

-------
against monitoring data, and have the added advantage of providing a more complete data
coverage. As discussed previously, the Mercury Maps technique relies on the accuracy of air
deposition model results, and incorrect air deposition predictions will be reflected in fish tissue
concentration reduction estimates. However, if the air deposition model inaccuracies are
proportional to the total load, taking the ratio of future over current deposition load will cancel
out these inaccuracies.

Furthermore, REMSAD, the model most likely to be used for the air deposition input to Mercury
Maps, has features that represent orographic effects reasonably well. Orographic effects occur
when a parcel of warm air meets a mountainside, and is lifted vertically (orographic lifting), thus
cooling and condensing, and losing moisture due to precipitation.

In addition to features of its parent model, UAM-V, the REMSAD aerosol and toxics deposition
model (ATDM) includes vertical mass redistribution within convective and stratiform clouds, as
well as incorporation of a cloud submodel. REMSAD uses 8 vertical layers with sigma (terrain
following) vertical coordinates. The first four layers of the model cover the maximum expected
daytime planetary boundary layer (typically up to 3.5 km) (SAI, 1996). Since individual cumulus
cells are small (l-10km) relative to REMSAD grid cells (50-100km), cloud effects are
parameterized in ATDM. The cloud submodel includes a diagnosis of cloud type, and
determination of whether deep convection is possible, and if so the mixing coefficient is
calculated, fractional cloud cover in  the grid column, and net vertical distribution are determined.

3-2) Taking into account Other loads, in other than a screening level manner, is infeasible due to
large uncertainties in estimating the absolute magnitude of these sources.  A more detailed
analysis, of individual watersheds, in these cases, is more appropriate. Screening out of
watersheds with significant or potentially significant sources allows one to proceed with much
greater certainty and credibility, though with a bias towards underestimating overall reductions.
The equation suggested by the  reviewer, was used in section lb-1 to illustrate how excursions
from the simple proportional reduction model, caused by the presence of other sources, could
influence the accuracy of Mercury Map predictions.  Given that watersheds in which all
potentially significant non-air-deposition sources (e.g. point sources and historic mining
activities) are eliminated from the analysis, these prediction errors are expected to be quite small.
                                      Page 14 of 25

-------
4) Were the calculations performed correctly?  Were the data processed without error?

Comments: None of the four reviewers addressed this question.

Response: In the revised analysis, I have made the final presentation as clear as possible to a
future user. It is recommended that future users check the calculations and datasets used to
determine whether all calculations and data processing were performed correctly.  In addition to
the dataset-specific background information and analysis, information on how each data layer
was processed is detailed in each section of the Appendix.
                                      Page 15 of 25

-------
                                      References

Aastrup, M., J. Johnson, E. Bringmark, I. Bringmark, and A. Iverfeldt, Occurrence and Transport
of Mercury within a Small Catchment Area. Water, Air, and Soil Pollution 56: 155-167, 1991

American Fisheries Society, 2000. Proceedings from the Forum on Contaminants in Fish,
October 18-20, 1999, Prepared by EVS Environment Consultants, Inc., Seattle, August 31, 2001.

Dispasquale, M.M., J. Agee, C. McGowan, R.S. Oremland, M. Thomas, D. Krabbenhoft, and
C.C. Gilmour.  Methyl-Mercury Degradation Pathways: A Comparison Among Three Mercury-
Impacted Ecosystems.  Environ. Sci. Technol. 2000, 34, 4908-4916.

Hurley, J.P., J.M. Benoit, C.L. Babiarz, M.M. Shafer, A.W. Andren, J.R. Sullivan, R. Hammond,
and D.A. Webb, Influence of Watershed Characteristics on Mercury Levels in Wisconsin Rivers.
Environ.  Sci. Technol., 1995, 29, 1867-1875.

Johansson, K., Aastrup, M., Anderson, A., Brinkman, L., Iverfeldt, A., 1991. Mercury in
Swedish Forest Soils and Waters: Assessment of Critical Load. Water Air Soil Pollut. 56, 276-
281.

Johansson, K. and A. Iverfeldt.  The Relation Between Mercury Content in Soil and the
Transport of Mercury from Small Catchments in Sweden, in: Mercury Pollution: Integration and
Synthesis, Lewis Publishers, 1994.

Lindberg, 1996, Forests and the Global Biogeochemical Cycle of Mercury: The Importance of
Understanding Air/Vegetation Exchange Processes, in W. Baeyens et al. (eds), Global and
Regional Mercury Cycles: Sources, Fluxes and Mass Balances, 359-380.

Long, K.R., J.H. Jr. DeYoung, and S.D. Ludington.  Database of Significant Deposits of Gold,
Silver, Copper, Lead, and Zinc in the United States. Part A: Database Description and Analysis.
Open-File Report 98-206A. USGS, 1998.

Maine Department of Environmental Protection, Status of Mercury Discharge from Wastewater
Treatment Facilities in Maine. Submitted to the Joint Standing Committee on Natural
Resources. January 15, 2001.  DEPLW2001-5. Available at:
http://janus.state.me.us/dep/blwq/report/legisreport.htm.

Nellor, M., 1999.  Letter to Tudor Davies, Director EPA Office of Science and Technology, On
Mercury Effleuent Sampling Results, May 20,  1999.

Plouffe, A, G.E.M. Hall, and P. Pel chat (Geological Survey of Canada, Ottawa, Ontario, Canada
Kl A OE8; corresponding author: aplouffe@nrcan.gc.ca). Mercury Content of Soils in the
                                     Page 16 of 25

-------
Vicinity of a Past-Producing Mercury Mine, Central British Columbia.
http ://www. sph.umich. edu/eih/heavymetals/Manuscripts/PlouffeA.htm

SAI, 1996. User's Guide to the Regulatory Modeling System for Aerosols and Deposition
(REMSAD). Systems Applications International, Inc. SYSAPP-96/42. September, 1996.

Sherbatskoy, T., J.B. Shanley, GJ. Keeler, Factors Controlling Mercury Transport in an Upland
Forested Catchment. Water, Air, and Soil Pollution. 105: 427-438, 1998.

St. Louis, V., J.W.M Rudd, C.A. Kelly, K.G. Beaty, RJ. Flett, and N.T. Roulet, Production and
Loss of Methylmercury and Loss of Total Mercury from Boreal Forest Catchments Containing
Different Types of Wetlands. Environ. Sci. Technol. 1996, 30,  2719-2729.

Swain, E.B., D.R. Engstrom, M.E. Brigham, T.A. Henning, and P.L. Brezonik, Increasing Rates
of Atmospheric Mercury Deposition in Midcontinental North America. Science, Vol. 257, 7
August, 1992.

I. Tsiros, 2001. A Screening Model-Based Study of Transport Fluxes and Fate of Airborne
Mercury Deposited onto Catchment Areas. Chemosphere 44, 99-107.

Tsiros and Ambrose, 1999, An Environmental Simulation Model for Transport and Fate of
Mercury in Small Rural Catchments, Chemosphere, 39(3):477-492.

USEPA, 2000. Draft Florida Pilot Mercury Total Maximum Daily Load (TMDL) Study:
Application of the Everglades Mercury Cycling Model (E-MCM) to Site WCA 3A-15. Prepared
for the United States Environmental Protection Agency and Florida Department of
Environmental Protection. Submitted by Reed Harris, Curtis D. Pollman, David Hutchinson and
Don Beals. Tetra Tech Inc., Lafayette, CA, October, 2000.

USEPA, 2000b. Guidance for Assessing Chemical Contaminant Data for Use in Fish Advisories:
Volume 1 - Fish Sampling and Analysis, Third Edition. November, 2000.  EPA-823-B-00-007.

USEPA, 2001. Total Maximum Daily Load for Total Mercury in the Middle/Lower Savannah
River, GA. February 28, 2001. USEPA Region 4.
                                    Page 17 of 25

-------
               Fish Tissue Mercury Concentrations
                        Averaged  by Watershed
Average Fish Cone, (ppm)
     0.00-0.14
     0.15-0.29
     0.30-0.44
     0.45-0.59
     0.60-3.30
     States

     Note: New Criterion for mercury in fish is 0.3 ppm. Point of departure in fish advisories often in 0.15 ppm to 0.3 ppm range.
     Average value based on fillet samples only. See report text for details.
     Source: National Listing of Fish and Wildlife Advisories (NLFWA) Mercury Fish Tissue Database (June, 2001).
     Figure 1 (revised)
                                   Page 18 of 25

-------
                    Percent Reduction in Air Deposition  Load
               Necessary to Meet New Methylmercury  Criterion
          Watersheds with No Other Significant Mercury Sources
% Reduction to Meet Criterion
    Currently Meets Criterion
    10% Reduction Required
    15% Reduction Required
    20% Reduction Required
    25% Reduction Required
    50% Reduction Required
    75% Reduction Required

                 Note:  Watersheds defined as having no significant mercury sources if total estimated load from Publicly Owned
                 Treatment Works (POTWs) and pulp and paper mills is less than 5% of estimated waterbody delivered mercury
                 at a typical deposition load (10 g/km2/yr), and no mercury cell chlor-alkali facilities, mercury mines, or significant
                 past producer gold mines are present. Additional reductions would be required to meet most state fish advisory
                 levels, which are often set below the methylmercury criterion.
                 Source: National Listing of Fish and Wildlife Advisories (NLFWA) Mercury Fish Tissue Database
                 (June, 2001). See text of report for data sources for point source dischargers and mines.
States
        Figure 2 (revised)
                                           Page 19 of 25

-------
                        Additional Data and Analyses for Appendix
                           (to be merged with report appendix)

                                 Fish Tissue Database

Since methylmercury accumulates in fish muscle, rather than fat, skin, or organs, the manner in
which fish samples are analyzed affects the reported concentration. Using whole fish samples
will give a reduced concentration, relative to fillets, due to a dilution effect from lower
concentrations in non-fillet portions of the fish. While whole fish samples are relevant for
concerns over eco-system effects, whole fish sampling is not recommended for use in creating
fish consumption advisories (USEPA, 2000b).

Of the 21571 geo-referenced fish tissue samples:
              17,826 (83%) samples were fillets,
              1,759 (8%) were whole, and
       •       1,986 (9%) were unknown/unspecified.
States in which whole fish sampling composes greater than 25% of all samples, are: GA, IN, KS,
MD, ME, NE,  and TX.  States in which sample type is largely unknown or unspecified are MS
(unknown), and TX (year sampled, unknown). About a third of samples in GA are of unknown
sample type.

Of the fillet samples:
              3,969 specified with Skin Off,
              7,570 with Skin On, and
              5,984 did not specify.

Of all 21,571 samples, only 8 indicated composite sampling, sampled at two separate locations,
all done with fillets.

Of the 21,571 geo-referenced samples:
       •       All fillet shows average concentration of 0.40 ppm;
       •       All whole shows average concentration of 0.18 ppm;
       •       Fillet Skin Off shows an average concentration of 0.40 ppm; while
       •       Fillet Skin On shows average concentration of 0.31 ppm.

In order to assess whether the fish tissue data were biased toward higher trophic level fish, fish
tissue samples were sorted by frequency according to species. Of the top five most frequently
sampled species of georeferenced whole fish samples, 53% were trophic level 4 fish versus 46%
as trophic level 3 fish. Of the five most frequently sampled species of georeferenced fillet
samples, 100% were trophic level 4 fish.
                                     Page 20 of 25

-------
Thus, the large difference in concentrations from fillet to whole fish samples is certainly due in
part to the different species sampled. However, dilution from using whole fish is still expected
to be significant and thus these samples will be removed from the analysis.

The top five most frequently sampled Fillet Skin On species were: walleye, northern pike, yellow
perch, largemouth bass, and common carp, in that order. While, the top five most frequently
sampled Fillet Skin Off species were: largemouth bass, channel catfish, white crappie, flathead
catfish, and blue catfish, in that order. Based on this listing, it is clear that skin removal is species
specific, and likely due to how the fish is commonly consumed. While leaving skin on dilutes
the tissue concentration,  it appears the choice to remove or retain the skin is representative of
how that fish is expected to be consumed, and thus most representative of the concentration
consumed. That is, all fillet samples, regardless of whether skin is specified as on, off,  or
unspecified, will be retained in this analysis.
Sensitivity to Scale (HUC8 versus HUC11) Analysis
The HUC11 watershed coverage was obtained for the Chesapeake Bay drainage. The HUC11
coverage contained 511 HUC11 watersheds compared to 65 in the HUC8 coverage, for the same
area, nearly a factor of 8 increase in resolution.
The georeferenced fish tissue data were averaged across the HUC11 watersheds, and compared
against the HUC8 coverage. Of the 116 Chesapeake Bay HUC11 watersheds with fish sample
data, the median number of samples is 3 (mean of 5), with a range of 1 to 101 (26% have 1
sample). Of the 850 HUC8 watersheds (across the country) with fish tissue data, the median
number of samples is 9 (mean of 24), with a range of 1 to 959 (Everglades) samples (11% have 1
sample). While 9 samples should provide a relatively solid statistical measure of the average
concentration in the watershed, 3 samples provides  a much lower level of statistical significance.
At the same time, the dramatic increase in 1 sample watersheds is even more problematic. While
the HUC11 watersheds, then, show local maxima exceeding the criterion within HUC8
watersheds shown to be below that level, these high average concentrations are not likely
representative.
                                     Page 21 of 25

-------
    Database of Significant Deposits of Gold, Silver, Copper, Lead, and Zinc in the U.S.
Source:
This database was obtained directly from the database author, in excel spreadsheet format.

Data Processing:
Fields in the spreadsheet, not needed in the project, were deleted, and the header rows were
reduced to a single row with a short, descriptive title.  The data was then exported from Excel as
a tab-delimited text file, imported to Arc View as a table, added to a View as an Event Theme,
converted to a shapefile, and converted to the project projection.  The field containing the amount
of gold in ounces, previously produced at the mine, was manipulated to remove all non-numeric
characters, and converted to number format.  This data was queried for mines having produced at
least 64,000 ounces (2 tons) of gold. The resultant selection was converted to a shapefile, and
imported to Mercury Maps as the gold_mines_sig_dep shapefile.

Notes on Data Quality and Interpretation:
The derivation and quality of data in this database is discussed in considerable detail in Long, et
al, 1998.
                                     Page 22 of 25

-------
                               Permit Compliance System

                                  Pulp and Paper Mills

A separate theme for pulp and paper mills was created from the PCS theme by selecting facilities
with SIC = 2611 (Pulp mills), 2621  (Paper mills), or 2631 (Paperboard mills). The PCS database
includes 488 pulp and paper mill facilities, distributed as follows:
             160 Pulp mills
             244 Paper mills
             84 Paperboard mills

PCS Loading Data:
The PCS pulp and paper theme was joined with the mercury loading data table, and summarized
by year. For each year from 1990 to 1999, between 4 and 14 pulp and paper facilities reported
mercury loads.  The average load ranged from 5 Ibs/year to 228 Ibs/year, with the 228 Ibs/year
value appearing to be an outlier (next highest value is 22 Ibs/year and is associated with a facility
in Ohio reporting a load of 1324 Ibs, and another in NC reporting 261 Ibs in 1996).  For each of
the mill types, the data was summarized by facility (i.e. average load over the ten year period).
Of the 27 facilities, average load was 0.0 Ibs/year for six of these facilities.
             Pulp mills (SIC = 2611) - average load (by year) ranged from 1.7 to 27 Ibs/year,
             with overall mean of 9.1 Ibs/yr.
             Paper mills (SIC = 2621) - average load (by year) ranged from 0.0 to 331 Ibs/year,
             where 7 out of 10 years, average load is less than 5 Ibs/year, with overall mean of
             35 Ibs/year.
       •      Paperboard mills (SIC = 2631) - average load (by year) ranged from  14 to 261
             Ibs/year, and overall  mean of 44 Ibs/year.

Maine Study (1998) Data:
The PCS pulp and paper theme was queried to identify facilities in Maine (NPDES id starting
with "ME") and with SIC codes of 2611, 2621, or 2631, resulting in 17 facilities including:
             4 Pulp Mills (2611)
             13 Paper Mills (2621)
Ten of the 17 Maine PCS pulp and paper facilities have Average Limits for mercury effluent
concentrations (ME DEP, 2001).  The Average Limits are the 95th percentile probability limit on
the mean, based on a required number of samples and using EPA methods 1669 and 1631 (ultra-
clean techniques) for collection and analysis of samples, respectively. The average of the
Average Limit values is 13.0 ppt (ng/L). With a range of 4.5 ppt to 28.9  ppt, where 4.5 is the
minimum set by the regulation rather than that measurements.  The difference in Average Limits
by SIC is not large at 14.4 ppt for pulp mills and 12.4 ppt for paper mills. Of the 10, only three
had flow rate values in PCS: 34.0, 46.5, and 157 MOD, for an average of 79.2 MOD.
                                     Page 23 of 25

-------
In order to compare the PCS loads with loads based on average effluent concentration, a
conversion from effluent concentration in ppt (or ng/L) times flow rate in MGD to Ibs/yr is
required, as follows.
10-9 g/ng x lkg/10A3 g x 2.2046 Ib/kg x 3.785 L/gal x 10A6 gal/MG x 365 D/yr
= 0.003046 Ib/yr /(MGD - ng/L)
So the Average Limit for Pulp and Paper Mills is converted as follows
13.0 ng/L x 79.2 MGD x 0.003046 Ib/yr/ (MGD - ng/L) = 3.14 Ib/yr
So, an estimated mercury loading rate value of 3.1 Ib/yr will be applied to each pulp and paper
mill in PCS.
                        Publicly Owned Treatment Works (POTWs)

From the PCS data coverage, a query for SIC = 4952 resulted in a coverage of 23,629 POTW
facilities in the conterminous U.S.  For each watershed, then, the sum of the flows from POTWs
were summed, and multiplied by a mean effluent mercury concentration to get an estimated
direct waterbody discharge load estimate (in Ibs/yr) (see Pulp and Paper Mills section for
conversion factor). A study by the Association of Metropolitan Sewerage Agencies (AMSA)
found an average concentration of 7 ppt in wastewater treatment plant effluent (Nellor, 1999).
This data is based on AMSA's study of 24 POTW facilities in six states, using clean sampling
and analytical techniques, for facilities with a range of flow rates from 0.65 MGD to 225 MGD,
serving populations ranging from 18,2000 to 1.74 million (median population - 384,000).

Applying this average concentration to the reported flow rate of each facility resulted in a mean
watershed load of 0.69 Ibs/yr, with a range from 0.0 to  128 Ibs/yr, and a total load of 472 Ibs/yr,
for the 681 watersheds with POTWs.  Based on this screening analysis,  79 watersheds had total
estimated POTW mercury discharge loads greater than 5% of an estimated typical air deposition
load delivered to waterbodies (i.e. 20% of a "typical" 10 ug/m2/yr deposition rate) or 1% of the
typical air deposition rate times the watershed area. All watersheds screened out by this
approach are located west of the Mississippi River. In these 79 watersheds,  POTW  effluent may
constitute a significant source of mercury to receiving waterbodies and thus  these watersheds will
be eliminated from the analysis.

For comparison purposes, the data on POTW mercury loads in PCS are discussed below.  Of all
POTW facilities, 1946 have PCS mercury loading  data with a range of average  annual loads  (all
facilities per year) from 34 Ibs/yr to 1,200,000 Ibs/yr, with a maximum reported value of
1,000,000,000 Ibs/yr from a facility in GA. Of all reported mercury loading values (not
summarized by year or facility), 99% are 1000 Ibs/yr or less, while 37% are  zero.  While there
has been a steady increase in the number of facilities reporting mercury loads, from  1990 to
1999, loading does not appear to have a specific pattern with respect to  time. While one would
expect average loading rates to go down, as new mercury analytical and clean sampling
techniques came into use in the late 90's, data from 1997 and 1998 have the  highest  average
loading for the entire ten year period.  In sum, mercury loading data in the PCS  database has
                                     Page 24 of 25

-------
some obviously suspect figures, as well as generally having reported loading rate values much
higher than values reported in the AMSA study.
                      Mercury Cell Chlor-Alkali Production Facilities

The PCS theme was queried for the 14 mercury cell chlor-alkali facilities listed in the Mercury
Study Report to Congress (MSRC, 1997), and developed into a separate theme. The mercury cell
chlor-alkali facilities theme was linked to the PCS mercury loading data table.
Joining the mercury cell chlor-alkali theme with the PCS mercury loading data shows an average
annual load of 39.7 Ib/yr, with a minimum of 0.0 Ib/yr and a maximum of 1659 Ib/yr (one year
maximum, Occidental, Muscle Shoals, AL). The next highest value at that plant was 53.0 Ib/yr
and the next highest at any other plant was 152 Ib/yr.
While the PCS data is generally thought to be unreliable due to the use of non-ultra-clean
sampling and analytical techniques, no other data source, sufficient to derive a reliable screening
level estimate, was found. Evidence of faulty data in PCS can be found by comparing reported
loading rate data for the Olin Corp facility in Augusta, GA. Reported at 32.5 Ib/yr mercury
effluent load in PCS, this same facility was sampled recently,  as part of the Savannah River
mercury TMDL and estimated to discharge 0.95 Ib/yr (EPA, 2001).
                                     Page 25 of 25

-------