United States
        Environmental Protection
        Agency	
Office of Water
(4305)
EPA-823-R-00-008
September 2000
&EPA  AQUATOX FOR WINDOWS

          A MODULAR FATE AND EFFECTS
        MODEL FOR AQUATIC ECOSYSTEMS

                   RELEASE 1
        VOLUME 3: MODEL VALIDATION REPORTS

-------
 AQUATOX FOR WINDOWS
 A MODULAR FATE AND EFFECTS MODEL
      FOR AQUATIC ECOSYSTEMS

            RELEASE 1
VOLUME 3: MODEL VALIDATION REPORTS
           SEPTEMBER 2000

   U.S. ENVIRONMENTAL PROTECTION AGENCY
           OFFICE OF WATER
     OFFICE OF SCIENCE AND TECHNOLOGY
         WASHINGTON DC 20460

-------
                                   DISCLAIMER

This document has been approved for publication by the Office of Science and Technology, Office
of Water, U.S. Environmental Protection Agency. Mention of trade names, commercial products
or organizations does not imply endorsement or recommendation for use.

This document describes a new aquatic ecosystem simulation model.  It is not intended to serve as
guidance or regulation, nor is the use of this model in any way required. This document cannot
impose legally binding requirements on EPA, States, Tribes, or the regulated community.
                              ACKNOWLEDGMENTS

       This model has been developed and documented by Dr. Richard A. Park of Eco Modeling;
most of the programming has been by Jonathan S. Clough under subcontract to Eco Modeling. It
was funded with Federal funds from the U.S. Environmental Protection Agency, Office of Science
and Technology under contract number 68-C4-0051 to The Cadmus Group, Inc. Work assignment
managers for The Cadmus Group have been Paul Jacobson, Jonathan Butcher, and William Warren-
Hicks; their help in expediting the contractual arrangements  and in reviewing the scientific
approaches is appreciated. Revision of the documentation has been performed under subcontract
to AQUA TERRA Consultants, Anthony  Donigian, Work Assignment Manager, under EPA
Contract 68-C-98-010.

       Additional Federal funding for program development has come from the U.S. Environmental
Protection Agency, Office of Pollution Prevention and Toxics, through Purchase Orders 7W-0227-
NASA and 7W-4330-NALX to Eco Modeling.

       The assistance, advice,  and comments of the EPA work assignment manager, Marjorie
Coombs Wellman of the Exposure Assessment Branch, Office of Science and Technology has been
of great value in developing this model and preparing this report. Further technical and financial
support from David A. Mauriello and Rufus Morison of the Office of Pollution Prevention and
Toxics is gratefully acknowledged.

       In an earlier version of the model developed at Abt Associates, Brad Firlie facilitated the
programming; Rodolfo Camacho developed and programmed the inorganic sediment constructs; and
review was provided by Lisa Akeson, Elizabeth Fechner-Levy, and Keith Sappington.
                                          11

-------
                             TABLE OF CONTENTS

VALIDATION OF THE AQUATOX MODEL, VERSION 1.66,
WITH DATA FROM LAKE ONONDAGA, NEW YORK	1-1
      Introduction 	1-1
      Input Data	1-1
      First-Level  Results 	1-5
      Second-Level Results	1-8
      Third-Level Results	1-11
      Conclusions	1-19
      References  	1-19
VALIDATION OF THE AQUATOX MODEL, VERSION 1.66,
WITH DATA FROM CORALVILLE RESERVOIR, IOWA	2-1
      Introduction 	2-1
      Input Data	2-1
      Preliminary Results and Modification	2-4
      Results	2-6
      Conclusions	2-10
      References  	2-10
VALIDATION OF AQUATOX 1.68 FOR PREDICTING BIOACCUMULATION
 OF PCBS IN THE LAKE ONTARIO FOOD WEB	3-1
      Introduction 	3-1
      Model Structure	3-1
            Partition Coefficients	3-2
            Nonequilibrium Kinetics	3-8
                   Sorption and Desorption to Sedimented Detritus  	3-8
                   Bioconcentration in Macrophytes and Algae  	3-11
                         Macrophytes	3-11
                         Algae 	3-12
                   Bioaccumulation in Animals  	3-13
                         Gill Sorption  	3-13
                         Dietary Uptake	3-15
                         Elimination	3-16
      Code and Parameter Changes to Facilitate Analyses	3-18
      Results and Discussion  	3-19
            Comparison of Predicted and Measured BAFs 	3-20
            Sensitivity Analysis	3-25
            Uncertainty Analysis 	3-29
      Conclusions	3-33
      References  	3-34
                                        in

-------
                                       PREFACE

       The Clean Water Act— formally the Federal Water Pollution Control Act Amendments of
1972 (Public Law 92-50), and subsequent amendments in 1977,1979,1980,1981,1983, and 1987—
calls for the identification, control, and prevention of pollution of the nation's waters. In the
National Water Quality Inventory: 1996 Report to Congress, 36 percent of assessed river lengths
and 39 percent of assessed lake areas were impaired for one or more of their designated uses (US
EPA 1998). The most commonly reported causes of impairment in rivers and streams were siltation,
nutrients, bacteria, oxygen-depleting substances, and pesticides;  in lakes and reservoirs the causes
also  included  metals  and noxious  aquatic  plants.  The most commonly reported sources of
impairment were agriculture, nonpoint sources, municipal point sources, atmospheric deposition,
hydrologic modification, habitat  alteration and resource extraction.   There  were  2196 fish
consumption advisories, which may include outright bans, in 47 States, the District of Columbia and
American Samoa. Seventy-six percent of the advisories were due to mercury, with the rest due to
PCBs,  chlordane, dioxin, and DDT (US EPA 1998). States are not required to report fish kills for
the National Inventory;  however, available  information for 1992 indicated 1620 incidents in 43
States, of which 930   were attributed to pollution,  particularly oxygen-depleting substances,
pesticides, manure, oil and gas, chlorine, and ammonia.

       New approaches and tools, including appropriate technical guidance documents, are needed
to facilitate ecosystem analyses of watersheds as required by the Clean Water Act. In particular,
there is a pressing need for refinement and release of an ecological risk methodology that addresses
the direct, indirect, and synergistic effects of nutrients, metals, toxic organic chemicals, and non-
chemical stressors on aquatic ecosystems, including streams, rivers, lakes, and estuaries.

       The ecosystem model AQUATOX is one of the few general ecological risk models that
represents the  combined environmental  fate and effects of toxic chemicals.  The model also
represents conventional pollutants, such as nutrients and sediments, and considers several trophic
levels,  including attached and planktonic algae, submerged aquatic vegetation, several  types of
invertebrates, and several types offish. It has been implemented for streams, small rivers, ponds,
lakes, and reservoirs.

       The AQUATOX model is described in  these  documents.  Volume 1: User's Manual
describes the usage of the model.  Because the model is menu-driven and runs under Microsoft
Windows on microcomputers, it is user-friendly and little guidance is required.   Volume 2:
Technical Documentation provides detailed  documentation of the concepts and constructs of the
model so that its suitability for given applications can be determined. Volume 3: Model Validation
Reports presents three model validation studies performed for different environmental stressors and
in different waterbody types. The validations were performed using test versions of the model which
had only very minor differences from Release Version 1; the specific test version is noted in the title
of each report.
                                           IV

-------
MODEL VALIDATION REPORTS	CHAPTER 1


    VALIDATION OF AQUATOX VERSION 1.66, WITH DATA FROM
                       LAKE ONONDAGA, NEW YORK

                                     Introduction

"Lake Onondaga is arguably the most polluted lake in the United States" according to Effler (1996)
in the preface to his comprehensive book, which serves as the primary reference for the following
information and data on the lake. The shore of this lake in central New York State was industrialized
before 1800, and over the last hundred years at least thirty different chemicals were produced from
nearby salt and limestone deposits. Unfortunately, the lake was a convenient dumping ground for
waste products.  Production of soda ash resulted in waste beds as much as 21 m deep and 8.1 km2
in area along 30% of the lake shore; the wastes include NaCl and CaCl2 that easily leach into the
lake. The salinity of the lake was around 3%o (parts per thousand) prior to closure of the soda ash
plant in 1986; by 1990 the salinity had decreased to 1.3%o. Nevertheless, this salinity creates unusual
density gradients and intense stratification of the lake.  A chlor-alkali plant produced NaOH and Cl
by electrolysis, using Hg as the cathode. From 1946 to 1970 as much as 75,000 kg of Hg were
discharged into the lake. Aside from an advisory against eating fish from the lake, the high mercury
levels may have adversely affected the functioning of the lake ecosystem.

The lake has been a receptacle for most of the domestic waste and urban runoff from Syracuse and
the surrounding area. Prior to 1960 untreated and poorly treated sewage was discharged directly to
the lake. In 1960 the Metropolitan Sewer District (METRO) primary treatment plant was completed;
in 1979 it was upgraded to  secondary treatment; and in 1981 tertiary treatment (removal of
phosphorus) was instituted. By design, there is little reduction in ammonia in the sewage effluent.
At present nearly 20% of the annual inflow to the lake is from METRO. Most troubling are the
combined sewer overflows (CSOs) that carry storm water and raw sewage into tributary creeks about
50 times a year.  In 1991 there were 45 CSOs discharging into Onondaga Creek, 19 into Harbor
Brook, and 2 into Ley Creek. The purpose of this study is to evaluate the validity of the AQUATOX
model in representing eutrophication of a stratified urban lake receiving large amounts of point- and
nonpoint-source nutrients and organic matter.

Onondaga is a glacial lake that is 7.6 km long and has a maximum width of 2 km; the watershed is
652 km2. The surface area of the lake is 12E6 m2, and the mean depth is 10.9 m, with a volume of
131E6 m3; the maximum depth is 19.5 m. For a lake of this size it is appropriate to apply a one-
dimensional model such as AQUATOX, which represents stratification into epilimnion and
hypolimnion but assumes that the lake is well mixed horizontally.

                                      Input Data

As a test of the application of AQUATOX, three levels of analysis were implemented using data that
are generally available.  With Effler's (1996) 832-page book  as a resource,  even more detailed
analyses could have been performed, but they would have been beyond the scope of the present
project.  Therefore, some simplifications were taken in computing loadings and site characteristics.
As  shown in Table 1, mean annual values for nutrient and organic matter mass loadings and

                                          1-1

-------
MODEL VALIDATION REPORTS
CHAPTER 1
concentrations were used: first with mean monthly inflow data (first-level analysis) and then with
daily inflow data (second- and third-level analyses).
                                   Table 1. Input Data
Variable
Inflow
Phosphorus, NFS
METRO
NOX&NH3, NFS
METRO
Org. matter, NFS
METRO
Epilimnion
temperature
Hypolimnion
temperature
Wind
Solar radiation
Initial conditions
Source
Efflerl996,p. 103
www.waterdata.usgs .gov
Effler 1996, calc. from p. 162
Efffler 1996, calc. from p. 159
Effler 1996, p. 162
Effler 1996, calc. from p. 138
Effler 1996, calc. from p. 128
Effler 1996, calc. from p. 138
Effler 1996, calc. from p. 138
Effler 1996, calc. from p. 128
Effler 1996, calc. from p. 138
Effler 1996, p. 207
Effler 1996, p. 247
Effler 1996, p. 248
unpub. data, Lake George,
N.Y.
Effler 1996
Format
1 monthly values based on 28-yr means
2 daily values for 4 gauged streams;
extrapolated to ungauged streams
1 mean loads, April-September, 1990
2 mean annual cone., 7 tributaries,
1989-1990; mult, by respective inflow
1>2 mean loads, April-September, 1990
1 mean annual loads for 1989
2 mean annual concentrations for 1989
for 4 tributaries
1>2 mean annual loads for 1989
1 back-calculated from organic-N
2 back-calculated from organic-N
1>2 mean annual loads for 1989
annual mean and range interpolated
from figure for 1989
annual mean and range interpolated
from figure for 1981
mean value est. from figure for 30
years
observed annual mean and range
obs. data and professional judgment
 Ist-level analyses with monthly inflow data;  2nd- and 3rd-level analyses with daily inflow data

AQUATOX can accept loadings data in a variety of formats. For the first-level analyses, constant
mass loadings expressed as g/d were used for nutrients and organic matter. Mean monthly inflow
values (Figure 1), based on a 28-yr dataset, primarily affected the retention time or flow-through
rates for the lake.  The intent was to use the model as if only general information was available.
                                          1-2

-------
MODEL VALIDATION REPORTS
                                              CHAPTER 1
                      Figure 1. Lake Onondaga mean daily inflow for
                      each month, based on 28-yr record.
J™
E 2E+06
j|2E+06
§ 2E+06
O
u! 1E+06
-?
- 1E+06
w 1E+06
8E+05

n
/ \
/• A
/
/ • • V . /
\ ^
\ /
• ; • ^-^ •








Jan Mar May Jul Sep Nov
Feb Apr Jun Aug Oct Dec
For the second- and third-level analyses,  discharge data from the four gauged streams in the
watershed (Onondaga Creek, Ninemile Creek, Ley Creek, and Harbor Brook, listed in order of
importance) were downloaded from the U.S. Geological Survey Web site (see Table 1). Discharge
from four ungauged streams was estimated, assuming that they had an aggregate flow rate that was
94% of the discharge of Ley Creek and Harbor Brook based on data in Effler (1996, p. 102). The
loadings were then computed using average concentrations for the respective streams, assuming a
constant relationship between concentration and discharge. Different average phosphate values were
used for 1989 and 1990 for Onondaga and Ninemile Creeks, which varied considerably between the
two years due to combined sewer overflows. Also, the concentration of ammonia in Ninemile Creek,
which flows through the soda ash waste beds, exhibits an inverse relationship to flow rate according
to Effler  (1996, p. 131); therefore, his  Equation 3.12 was used to compute the  ammonia
concentrations:
                                [T JVHl]   0.20
                       0.73
                       Flaw
where:
       Flow
concentration of total ammonia (mgN/L),
flow rate (mVs).
Given the readily available hydrologic data, both 1989 and 1990 were simulated with daily loadings.
Examination of the loading plots confirms that the streams draining into Lake Onondaga are indeed
"flashy" or subject to fast runoff with distinct peaks (Figure 2), and the nutrient and organic matter
loadings vary accordingly (Figures 3-6).
                                          1-3

-------
MODEL VALIDATION REPORTS
                                                                       CHAPTER 1
      Figure 2. Lake Onondaga inflow.
                                         Figure 3. Lake Onondaga phosphate loadings.
     7,000,000

     6,000,000

     5,000,000

     4,000,000

     3,000,000

     2,000,000

     1,000,000
          01/01/89    09/24/89     06/17/90
               05/14/89     02/04/90     10/28/90
                                                  01/01/89     09/24/89      06/17/90
                                                        05/14/89     02/04/90     10/28/90
 Figure 4.  Lake Onondaga ammonia loadings.      Figure 5. Lake Onondaga DOM loadings.
     1
     1
     1
   I1
   < 1
   o
,800,000
,600,000
,400,000
,200,000
,000,000
800,000
600,000
400,000
200,000
     0
          01/01/89    09/24/89     06/17/90
               05/14/89     02/04/90     10/28/90
- 50,000,000
                                                    01/01/89    09/24/89    06/17/90
                                                         05/14/89    02/04/90    10/28/90
In the second-level analysis no attempt was made to fine-tune the model.  In the third-level analysis
the model input was specifically modified to represent better the known site-specific characteristics
of this unusual lake. No changes were made in the computer program, but several work-arounds
were used; these are available to any model user. Also, another algal group was parameterized—
something any knowledgeable user could implement.  Altogether, one  additional day was spent
incorporating the greater detail into the simulations; more time could have been spent to obtain an
even better calibration.

It appears that the salinity gradient in the lake is restricting the mixing depth. The model computes
the depth of the well mixed layer (epilimnion) using a robust regression equation with the fetch
(distance across which the wind can blow) as the independent variable. This equation is based on
a dataset for 167 lakes.  By back-calculating from the regression equation, a fetch (Length) of 0.779
km was found to give the observed well mixed depth (MaxZMix) of 7.75 m:
                                            1-4

-------
MODEL VALIDATION REPORTS                                         CHAPTER 1
MaxZMix
\og(Length)
Length
Length0336
log(7.75)
0.336
779 m
• 0.569
0.245
The spring bloom was reported to be due to cryptomonads, a flagellated algal group. Using values
from Collins and Wlosinski (1983), a cryptomonad compartment was parameterized. Version 1.66
of AQUATOX can simulate three algal groups, generally diatoms, green algae and blue-green algae.
Diatoms and green algae are more important than blue-greens in Lake Onondaga, so cryptomonads
were substituted for blue-greens. This is appropriate because the model assumes that blue-greens
occupy the top meter of water unless the wind exceeds 3 m/s, when Langmuir stripes form, and
cryptomonads also tend to move toward the surface.

The first- and second-level implementations had cladocerans (Daphnia) and predatory zooplankton
as the two zooplankton compartments. However, rotifers are important grazers on cryptomonads,
and predatory zooplankton probably are unimportant in the lake, so rotifers were substituted for
predatory zooplankton in the third-level analysis. Furthermore, the food preferences for rotifers were
changed to force them to "eat" cryptomonads in the model.  In addition, the zoobenthos, (Tubifex
tubifex) feeding rates and catfish initial conditions were calibrated in order to model sediment
oxygen demand.

                                  First-Level Results

Initially, the lake was simulated using monthly averages for inflow and annual averages for nutrient
and organic-matter loadings, similar to a screening-level application. The maximum chlorophyll a
predictions were of the same magnitude as those observed in Lake Onondaga, and the observed
summer biomass values were bounded by the predictions.  However, the algal bloom in early May
was missed entirely and several summer fluctuations were missed (Figure 6).  The spring bloom was
due to cryptomonads, which were not modeled. Evidently the fluctuations in summer biomass were
due to a succession of algal blooms and crashes involving greens, flagellated greens, and diatoms;
whereas the model predicted diatoms to be the dominant phytoplankton. Site-specific calibration
(third-level analysis of this study) would be necessary to represent this detailed succession. Plots
of cumulative distributions emphasize the differences between the predicted (Figure 7) and observed
(Figure 8) chlorophyll values.  The lack of  low predicted values compared  to  the uniform
distribution of observed values between 5 and 80 |J,g/L is obvious.
                                          1-5

-------
MODEL VALIDATION REPORTS
                             CHAPTER 1
                     Figure 6. Predicted and observed chlorophyll in
                     Lake Onondaga, New York, in 1990.
70
G~60
^5>
-3=;n
CO
_i An
Q- ^n
o30
O9n
_i
Om
Q


A.






•

*
\^ xTX-
- " \
\
\
• . "\
,\
,\
\
_\
I . -. -^
-J- \
.-7- v_
01/01/90 05/14/90 09/25/90
03/08/90 07/20/90 12/01/90
Predicted — Observed

Figure 7. Cumulative distribution of predicted   Figure 8. Cumulative distribution of observed
chlorophyll in Lake Onondaga in 1990.
LU
Omn
££
g 80
Q
LU 60
Q_
^ 40
^
<
D
1 0
o






*s
(









l^--—























. 	 —












— -











. *












• •-















































^









/
{_















































0 10 20 30 40 50 60 70
CHLOROPHYLL a (ug/L)
— Predicted
chlorophyll from Lake Onondaga in 1990.
LU
0 100
£
~Z. on
O
K fin
Q.
[IJ 40
"^i 9O
1 0










/







/"
^







y
^^







J


















f








s*








^


















_/









--^








x









— — —









-










° 0 10 20 30 40 50 60 70
CHLOROPHYLL a (ug/L)
— Observed
Examination of the epilimnetic diatom rates (Figure 9, an area graph that displays individual rates
for each day) reveals that growth — represented by photosynthesis (PhotosynS) balanced against
respiration (RespirS) and, to a minor degree, excretion (ExcretS) — occurs over the entire summer
and early fall. Nonpredatory mortality (MortS) is  important, and predation (PredationS) is less
important. Washout  (WashoutS),  loss  to  bottom sediments  (SedimentS), mixing with the
hypolimnion (TurbDiffS), and sinking to the hypolimnion (SinkToHypoS) are unimportant.
                                          1-6

-------
MODEL VALIDATION REPORTS
                                   CHAPTER 1
                       Figure 9. Predicted rates for diatoms in Lake
                       Onondagain 1990.
                         01/02/90       05/30/90        10/25/90
                                03/17/90        08/12/90
                         | PhotosynS
                         |Mort8
                         I Sediments
I RespirS
] PredationS
I TurbDiffS
] ExcretS
] Washouts
| SinkToHypoS
Dissolved oxygen is another environmental variable that is important from a regulatory perspective,
with 5 mg/L being the standard applicable to Lake Onondaga. The predicted epilimnion values
(Figure 10) indicate a late winter dissolved oxygen sag under the ice that cannot be confirmed due
to a lack of observed data. However, observed low values during the summer were not represented
by the first-level model application. Interestingly, these low values are not represented by the site-
specific Lake Onondaga model either (Effler 1996, p. 712). The lack of agreement maybe due to
several factors. The crashes of the successive algal blooms observed in the lake undoubtedly created
labile detritus that exerted oxygen demand; furthermore, reduced chemical species, especially
methane and Fe++, transported from the anoxic hypolimnion are not modeled in AQUATOX. The
pronounced sag in October is due to the admixture of a large volume of anoxic hypolimnetic waters
during overturn. The hypolimnion predictions indicate anoxic conditions in the middle of summer,
and the episode is remarkably close to the observed conditions (Figure 11).
   Figure 10. Dissolved oxygen in the Lake
   Onondaga epilimnion in 1990.
        Figure 11.  Dissolved oxygen in the Lake
        Onondaga hypolimnion in 1990.
-16
en . .
El 4
Z -19
LJJ '*
o
£10
o
Q 8
LJJ
> R
3 b
r~ ^
• .•*? — «•* .

•
• •
^

1/90 05/14/90 09/25/90
03/08/90 07/20/90 12/01/90
— Predicted -•- Observed


14
i,.
OXYGEN (me
n oo o N
a u
LJJ
> A
_l t
O
W 9
w ^
Q
0
01


.• ^

\
\
\
\
\
\
\ / \
' — *v /
•\
\
\
\
\
.•\
A
• \
• \
f








...: L 	 /•
01 03/08 05/14 07/20 09/25 12/01
— Predicted -— Observed

                                          1-7

-------
 MODEL VALIDATION REPORTS
                            CHAPTER I
                                  Second-Level Results

The prediction of chlorophyll concentrations is no better with the daily data (Figure 12) than with
monthly loadings. The maximum values of the 1990 summer bloom are predicted reasonably well,
but neither of the spring blooms is predicted.  The algal rates (not shown) do not differ significantly
from those predicted with the monthly loadings. The summer epilimnetic oxygen sags again are
poorly represented (Figure 13).  The 1989  hypolimnetic anoxia occurs too late in the summer,
although its magnitude is accurately predicted. The 1990 period of anoxia is well represented (Figure
14).

                        Figure 12. Chlorophyll in Lake Onondaga
                        based on daily loadings.
mn
_i
15) Rn
(8
— I en
LOROPHYL
J .b. C
DOC
I ^
O
Q

•

: A .

•

•_
-

•
w \
\
•^
-i- \
: \
V 3

m\
rr
-m-\
• \
- \
V

01/01/89 08/14/89 03/28/90 11/09/90
04/23/89 12/05/89 07/19/90
— Predicted - Observed
Figure 13. Dissolved oxygen in Lake
Onondaga epilimnion, based on daily loadings.
1fi
^J
~rji1A
~7 19
LU
£lO
|
Q °
LU
> R
O
W A
co t
Q
2

• •

.-•.. *

\
\
\
\
\
\
\ . / \
	 ^ / \
.X, s „
. x^ _Z
•^ 	 ^ •

•
• •



01/01/90 05/15/90 09/26/90
03/09/90 07/21/90 12/02/90
— Predicted -— Observed
Figure 14.  Dissolved oxygen in Lake Onon-
daga hypolimnion, based on daily loadings.
OXYGEN (mg/L)
n oo o fo ji.
Q u
>
> A
0
w 0
w ^
Q
0
01


A- Ji— f,


\
\
\
\
\
\
\
\

V- 7\
r\ / j
1
\
\
5ft
\
• \
. \ '
* V '
t 3 .
• \^
( ' \
\
t \
\
\
V /
S / }
•\
.]

•\
\
' -\
A .
*\
A, i
5 H «










01 05/13 09/23 02/03 06/16 10/27
— Predicted -~- Observed


-------
 MODEL VALIDATION REPORTS
                                                                           CHAPTER 1
Examination of multiple model runs suggests that the model is sensitive to the feeding rate of the
zoobenthos (modeled as Tubifex tubifex). As formulated, feeding gradually converts refractory detrital
sediment to labile detritus, which creates additional oxygen demand. Slower feeding delays the onset
of hypolimnetic anoxia. If the 1989 period is considered as "spin-up" so the simulation can recover
from the effects of poor initial conditions and only 1990 results are considered, then the model can
be judged as an adequate representation.

The herbivorous zooplankton exhibit realistic seasonal fluctuations  in biomass.  Daphnia is the
surrogate herbivorous zooplankter represented in the model run, and the observed mean biomass of
0.625 mg/L for cladocerans during the growing season is about four times higher than predicted. The
fish are dominated by catfish in the simulation, compared to dominance by planktivorous fish in the
lake,  although the initial condition for  catfish of 5 mg/L is  forcing that result in the two-year
simulation. The gradual decrease in the fish stocks over the two-year period also could be a function
of inappropriate initial conditions, or it may indicate that zoobenthos production is not simulated
properly—another consequence  of the  sensitivity to zoobenthic  feeding rates.  Unfortunately,
zoobenthic biomass and production data are unavailable.
Figure 15. Predicted herbivorous zooplankton
(Daphnia) biomass in Lake Onondaga.
                                                Figure 16. Predicted fish biomass in Lake
                                                Onondaga.
O
    0.3

  " 0.25
  )
    0.2

   0.15
 s
 o
   0.05
     1/01  05/13   09/23   02/03   06/16   10/27

                  — Predicted
                                                    10
                                                ^5)

                                                w
                                                 O
                                                 00
                                                    0.1
                                                   0.01
                                                                        Catfish
                                                                           White Perch
                                                                               -~—
                                                                                YOY Bass
                                                    01/01/89      09/23/89      06/16/90
                                                          05/13/89      02/03/90     10/27/90

                                                            -Catfish    —White Perch
                                                           - - YOY Bass  — Adult Bass
                                   Third-Level Results

Detailed site information was used to obtain a satisfactory calibration to 1989 data for Lake
Onondaga.  This is demonstrated by both the dissolved oxygen and chlorophyll a trends.  The
epilimnetic dissolved oxygen (Figure 17) exhibits a sag during the fall overturn, and the summer
values are realistically lower than in the first- (Figure 10) and second-level (Figure 13) analyses.
These reflect the simulation of a larger volume of anoxic hypolimnion admixed during overturn and
the oxygen demand created by successive crashes of algal blooms. The high observed oxygen values
are not predicted well; however, it is generally more important to predict the low values than the high
values. The hypolimnetic anoxia is represented even better than in the first- (Figure 11) and second-
level  (Figure 14) analyses.  The calibration of Tubifex tubifex and adjustment of catfish initial
conditions were necessary to model adequately the dynamics of sediment oxygen demand.
                                            1-9

-------
 MODEL VALIDATION REPORTS
                           CHAPTER 1
   Figure 17.  Dissolved oxygen in Lake
   Onondaga epilimnion based on calibrated
   model.
Figure 18.  Dissolved oxygen in Lake
Onondaga hypolimnion based on calibrated
model.
1jd
G*
^1?
*§.
in in
O
5 8
O °
Q
yj K
5 6
O
CO A
w 4
Q
2
01/0
•

.. - . /\
I ^ / \
\
\
\
\
\
\
\
\
I ^~^^\ •• /__*__
\ . ./•'I/
•^VLvr^-^.. .
• • • • V

• •


1/89 05/14/89 09/25/89
03/08/89 07/20/89 12/01/89
— Predicted — Observed

14
0)12
Z1 n
LU
O _
£ 8
S 6
Q D
LU
^ 4
O
(/> _
W 2
Q
0
01


.-"...
S\
\ /^ • s \
\ / ^-\ / .. V
\
\
\
\
\
\
V

A
A
\
A
5
•\
•\
-.\ 	 /









01 03/08 05/14 07/20 09/25 12/01
— Predicted — Observed

The predicted chlorophyll a trends of the improved application are reasonably representative of the
trends observed during 1989; they are much better than those simulated in the first- (Figure 6) and
second-level (Figure 12) analyses. The factors contributing to the improvement are the forcing of
a shallow epilimnion, inclusion of cryptomonads and rotifers that account for the spring algal bloom
and subsequent crash, and more dynamic nutrient cycling due to improved loadings data and dynamic
sediment release. All these changes could have been made by a knowledgeable user; they did not
involve changing the AQUATOX code.

                        Figure 19. Chlorophyll in Lake Onondaga
                        based on calibrated model.
inn
_i
13) Rn
m
— ' Rn
>-
5; 40
O ^u
IT
3 20
I zu
O
01/0
•

•
J
•

-.
-
-



m m1^
/-v^V^x^i-\"
1" m" ^*\\
1 ^

1/89 04/24/89 08/16/89 12/08/89
02/26/89 06/20/89 10/12/89
— Predicted • Observed
The calibrations were preformed using 1989 data and keeping 1990 data separate from the process.
As a further check on the model validity, both 1989 and 1990 were simulated. As might be expected,
the test revealed some transient behavior, especially with respect to zoobenthos and fish populations.
However, the validity of the epilimnetic and hypolimnetic dissolved oxygen simulations held (Figure
20 and Figure 21).  Also, the simulations of chlorophyll represented significantly different algal
                                          1-10

-------
 MODEL VALIDATION REPORTS
                           CHAPTER 1
responses in the two years (Figure 22), and the simulated 1990 algal blooms—although larger than
observed—were within historic ranges observed in the lake.
   Figure 20. Dissolved oxygen in Lake
   Onondaga epilimnion in 1989 and 1990;
   simulation based on calibration for 1989.
16
~™ -I/I
OXYGEN (m<
» o ro 4
Q "
LU
5 6
O
if) A
(f) 1
Q
01/0



• •
•
•


\
\
\
\
\
\
_\
f* / \
^. .- f-1
1' ; A \
J&/J . \
• - 3 \
- V
•• 3
L
V j
XT*- . '
> #.
\/v-
•
A
( \
/
i



• •



1/89 09/24/89 06/17/90
05/14/89 02/04/90 10/28/90
— Predicted — Observed
Figure 21. Dissolved oxygen in Lake
Onondaga hypolimnion in 1989 and 1990;
simulation based on calibration for 1989.

l5> 19

Zm
LU
O
S_ Q
N
c
Q °
LU
>, 4
O
W o
W ^
0
01/

-*-


\
\
\
\
\
\
\

f* . L\
5
3
5
A
\
A !
.\
»\ *
9\
•\ &2
r \
\
. \
\
^ \
^

r A
\_- { \
\ /»
3
1
5

\
t
i * *
i *
•








01 05/13 09/23 02/03 06/16 10/27
— Predicted — Observed

                        Figure 22. Chlorophyll in Lake Onondaga in
                        1989 and 1990; based on 1989 calibration.
                             01/01/89
                                      08/14/89
                                                03/28/90
                                                          11/09/90
                                  04/23/89     12/05/89     07/19/90

                                     — Predicted  •  Observed
                                          1-11

-------
 MODEL VALIDATION REPORTS
                           CHAPTER 1
In order to test the null hypothesis that the predicted and observed distributions are the same,
cumulative distributions  were  obtained  (Figures  23-25).   These  were compared using the
Kolmogorov-Smirnov test. The distributions for epilimnetic oxygen were significantly different, but
those for hypolimnetic oxygen and chlorophyll were not shown to be significantly different at the 0.05
level (Tables 2-5).
  Figure 23. Cumulative distributions of
  predicted and dissolved oxygen in Lake
  Onondaga epilimnion in 1989 and 1990.
Figure 24. Cumulative distributions of
predicted and dissolved oxygen in Lake
Onondaga hypolimnion in 1989 and 1990.




3?
>
'is
3
|
u






100
90
80^
70^
60
50 H
40^
30 :
20 i
0





^^~~
/ ,-^~^
/ //
/r
J/
I /
/f
/ 1
j-/ }
— -^ ,--'''
2 4 6 8 10 12 14 16
Dissolved Oxygen (mg/L)
, J TI- J'_t J
Observed ~ rredicted


















e?
u
1
|
§
U






1UO
90
80
70
60
50
40
30
20
10
0





/ /
__^

r~~~~^ ;--'"'
r^-"'"
s~~~'J
/~~^

0 2 4 6 8 10 12 14
Dissolved Oxygen (mg/L)














                        Figure 25.  Cumulative distributions of
                        predicted and observed chlorophyll in Lake
                        Onondaga epilimnion in 1989 and 1990.
                            100
                            90
                          S? 80
                          S 70
                          I 60
                          | 50
                          I 40
                          J 30
                          U 20
                            10
                             0
                               0   20   40   60   80   100  120
                                      Chlorophyll a (ug/L)
                                  Observed
 Predicted
                                          1-12

-------
 MODEL VALIDATION REPORTS
CHAPTER 1
       Table 2. Summary Statistics for Dissolved Oxygen in Lake Onondaga Epilimnion
Time Period
1989-1990
Group
Observed
Predicted
Oxygen (mg/L)
No. of Obs.
82
82
Mean
8.46
8.53
Median
8.00
8.45
Std. Dev.
2.82
1.41
Std. Err.
0.31
0.16
p-value from
KS Test
0.025*
  : Significantly different at 
-------
 MODEL VALIDATION REPORTS
                            CHAPTER 1
  Figure 26. Predicted epilimnetic dissolved
  oxygen in Lake Onondaga with and without
  METRO sewage effluent.
DISSOLVED OXYGEN (mg/L)
§K> 4^ 05 03 0 fo it
a



yA
\
\
\
\
\\
\ '
\
\
"^x^. / \
I "^3^. / \
V j\ I

~^r^f^r^/


I






1/89 05/14/89 09/25/89
03/08/89 07/20/89 12/01/89
- Predicted — Without METRO
Figure 27. Predicted hypolimnetic dissolved
oxygen in Lake Onondaga with and without
METRO sewage effluent.

l5i 19
Zm
LLI
O .
S_ Q
N
c
Q °
LU
>, 4
O
w ~
w ^
Q
0
01;


/v j^\
\
\
\
\
\v
\ >
\
\
\

r^~-\. s \
i \
\
\\^
\ \^
\ >y
\ X
\ X
_s^
\
\

r '-









01 03/08 05/14 07/20 09/25 12/01
- Predicted — Without METRO

                        Figure 28. Predicted chlorophyll in Lake
                        Onondaga with and without METRO sewage
                        effluent.
                           01/01/89      04/24/89      08/16/89     12/08/89
                                02/26/89      06/20/89      10/12/89

                                - Predicted     — Without METRO
The model seemed to be sensitive to zoobenthic feeding, so sensitivity analysis was performed. The
maximum consumption  rate for Tubifex tubifex was calibrated at 0.25 g/g'd for the third-level
analysis, reflecting the interference of low oxygen levels with feeding behavior. For the sensitivity
analysis, that value was taken as the mean for a normal distribution with a standard deviation of 0.1
and was used in a Latin hypercube sampling with ten iterations. The parameter values spanned a
range from no feeding, and therefore limited sediment-water interaction, to doubled feeding rates with
accelerated sediment-water interaction. The results demonstrate the influence that simulated feeding
has on  utilization and mobilization of detrital sediments  and subsequently on the predicted
hypolimnetic anoxia (Figure 29). Given the pool of nutrients tied up in the bottom sediments, the
                                          1-14

-------
 MODEL VALIDATION REPORTS
                                                     CHAPTER 1
potential for release, and the annual overturn that mixes hypolimnetic and epilimnetic water, it is not
surprising that increased recycling affects all the ecosystem; this includes epilimnetic nutrients
(Figure 30), cryptomonad blooms (Figure 31), and epilimnetic dissolved oxygen (Figure 32).
  Figure 29.  Sensitivity of hypolimnetic
  dissolved oxygen to zoobenthic feeding in
  AQUATOX.
     01/01/89       09/24/89      06/17/90
           05/14/89      02/04/90       10/28/90
              • Minimum
              • Maximum
   Mean
-- Deterministic
                            Figure 30. Sensitivity of epilimnetic
                            phosphate to zoobenthic feeding in
                            AQUATOX.
01/01/89      09/24/89      06/17/90
      05/14/89      02/04/90       10/28/90

       — Minimum      Mean
       — Maximum   -- Deterministic
Figure 31. Sensitivity of cryptomonad blooms
to zoobenthic feeding in AQUATOX.
7
?
T3)c
E6
,n ^ -
W5
< A
o •>
53
07
i- *•
CL
f21
°0
01/0



1
1 1
11 n
I /
I! /J
11 III
1 1 /I
II I/I /
11 III
I !fi*4f<*A An ttoc»»»3

] \^ Jj \^

A
A
oS
<\
\

1/89 09/24/89 06/17/90
05/14/89 02/04/90 10/28/90
— Minimum Mean
— Maximum - Deterministic
                       Figure 32. Sensitivity of epilimnetic oxygen to
                       zoobenthic feeding in AQUATOX.
                                                    01/01/89       09/24/89      06/17/90
                                                          05/14/89       02/04/90       10/28/90
                                                              • Minimum
                                                              • Maximum
                                                   Mean
                                                - - Deterministic
                                        Conclusions

AQUATOX is seen to be a powerful tool for assessing eutrophication problems. It can be applied
as a screening-level model with readily available data; it can import time series data for more detailed
analyses;  and it can  be calibrated to represent site-specific  conditions.  These  approaches are
illustrated by increasingly detailed applications to Lake Onondaga, New York, a heavily polluted lake
                                            1-15

-------
 MODEL VALIDATION REPORTS	CHAPTER 1

affected by both municipal sewage effluent  and urban runoff.  Paired control and perturbed
simulations provide insights and estimates of impacts suitable for quantitative risk assessment and
environmental decision making. Uncertainty analysis with efficient Latin hypercube sampling can
be used to assess sources of uncertainty, including sensitivity to key parameters.
                                      References

Collins, Carol D., and Wlosinski, Joseph H.  1983. Coefficients for Use in the U.S. Army Corps of
       Engineers Reservoir Model, CE-QUAL-RL Vicksburg, Miss.: Environmental Laboratory,
       U.S. Army Engineer Waterways Experiment Station, 120 pp.

Effler, Steven W. 1996.  Limnological and Engineering Analysis of a Polluted Urban Lake.  New
       York: Springer-Verlag, 832 pp.
                                          1-16

-------
MODEL VALIDATION REPORTS	CHAPTER 2


                 VALIDATION OF AQUATOX VERSION 1.66
          WITH DATA FROM CORALVILLE RESERVOIR, IOWA


                                    Introduction

Coralville Reservoir is a shallow, eutrophic, run-of-the-river reservoir formed when the U.S. Army
Corps of Engineers impounded the Iowa River in 1958 for flood control. The dam is located about
three miles from Iowa City, and the reservoir was closely monitored by the University of Iowa under
contract to the Corps of Engineers. Four years of data (10/1/74-9/30/78) from the Coralville Water
Quality Study (McDonald and McDonald, 1976; McDonald, 1977, 1978, 1979) were used for
validation.

The extent and depth of the lake are quite variable, depending on the storage capacity. At spillway
level (712 ft elevation), the lake extends 35.1  mi (56 km) upstream,  and it covers 24,800 acres
(10,000 ha); at conservation pool level (680 ft) it extends 21.7 mi (34.7 km), and it covers 4,900
acres (1,990 ha). At its lowest level (670 ft), in late winter because of flood control requirements,
the reservoir covers 1,820 acres (735 ha). Going from spillway to conservation to minimum pool
levels, the volume varies from 469,400 acre-ft (57,900 ha-m) to 40,300 acre-ft (4,970 ha-m) to
10,600 acre-ft (1,310 ha-m), and the mean depth varies from 19 ft (5.8 m) to 8 ft (2.5 m) to 5.8 ft (1.8
m).
In 1973, just prior to the period simulated (10/1/74 to 10/1/78), 90% of the drainage basin of 4,770
mi2 (12,350 km2) was in agriculture; most of the remaining  10% was urban, suburban, and road
right-of-way. Of the total land, 77% was in crops, including 30% in corn, and 13% was pastureland
(McDonald and McDonald 1976).  Agricultural runoff in  the basin carries  large amounts of
fertilizers, pesticides, animal wastes, and silt—all of which have adverse impacts on the Coralville
Reservoir ecosystem. Such impacts can be examined with the AQUATOX model. In this validation,
the emphasis is on eutrophication as a consequence of nutrient and organic matter loadings.

                                     Input Data

Input data included the above site characteristics and time series of loadings taken from  the Water
Quality Study reports, time series of pool elevations downloaded from the Army Corps of Engineers
World Wide Web  (Web or www) Internet site, Iowa River flow data downloaded from the U.S.
Geological Survey Web site, and meteorological data downloaded from the Web (Table 1).
                                         2-1

-------
MODEL VALIDATION REPORTS
CHAPTER 2
                                Table 1.  Data Sources
Variable
Pool elevations
Volumes
Inflow and discharge
Phosphorus
Nitrate
Ammonia
Biochemical oxygen demand
Temperature
Wind
Solar radiation
Initial conditions
Source
www.water.mvr.usace.army.
mil/doc/desc/crvi4 .html
computed from pool
elevations using a regression
with 3 observed volumes
www.waterdata.usgs .gov/nw
is-w/IA/
Coralville Water Quality
Studies
Coralville Water Quality
Studies
Coralville Water Quality
Studies
Coralville Water Quality
Studies; converted to detrital
values in the model
Coralville Water Quality
Studies
www.solstice.crest.org/cgi-
bin/solrad/radiate-form.cgi
www.solstice.crest.org/cgi-
bin/solrad/radiate-form.cgi
Coralville Water Quality
Studies
Format
daily
daily
daily values upstream at
Marengo and at dam
approximately twice monthly
approximately twice monthly
approximately twice monthly
approximately twice monthly
weekly
average value
average value and range
observed values at beginning
of simulation
The AQUATOX  model was used to simulate the impacts of agricultural runoff on Coralville
Reservoir, so loadings of nutrients and organic matter (represented by biological oxygen demand,
BOD) were calculated from time series of water discharge (Figure 1) and nutrient and BOD
concentrations (Figures 2-5) for the Iowa River at Marengo, just upstream from the reservoir. As
seen from the discharge (Figure 1), the second and third years of the simulation period were drought
years; the fourth year was characterized by highly variable inflow (Figure 1; McDonald, 1979) and
prolonged high temperatures (Figure 6).
                                         2-2

-------
MODEL VALIDATION REPORTS
                               CHAPTER 2
  Figure 1. Iowa River discharge at Marengo,
  Iowa.
Figure 2. Concentrations of orthophosphate in
the Iowa River at Marengo, Iowa.
    50000000
   •40000000
   ',30000000
  HI
  o
  < 20000000
  o

  Q 10000000
        10/01/1974      05/05/1976      12/08/1977
               07/19/1975      02/20/1977      09/25/197!
   10/15/74       08/09/76       02/28/78
         09/15/75       07/12/77      08/01/78
  Figure 3. Concentrations of nitrate in Iowa
  River at Marengo, Iowa.
Figure 4. Concentrations of ammonia in Iowa
River at Marengo, Iowa.
                   10/26/76        05/02/78
            10/13/75        09/27/77
     10/15/74        10/26/76        05/02/78
            10/13/75        09/27/77
                                             2-3

-------
MODEL VALIDATION REPORTS
                         CHAPTER 2
 Figure 5. Biochemical oxygen demand, Iowa   Figure 6. Temperature in Coralville Reservoir,
 River at Marengo, Iowa.                      Iowa.
     10/15/74       10/26/76       05/02/78
           10/13/75        09/27/77
10/15/74
10/26/76
05/02/78
       10/13/75
       09/27/77
                          Preliminary Results and Modification

Two levels of simulations were used. First, simulations were run for the period of October 1,1974
to October 1,1978, using a constant volume for the reservoir, and adjusting the discharge according
to the varying inflow. The model predicted time-varying concentrations of nutrients, detritus, algae,
invertebrates, and fish as mg/L (dry weight for the organisms). The observed algae were reported as
cells/ml in the Water Quality reports. These values were converted to mg dry wt/L by assuming that
an average algal cell = 2E-10 g dry wt; therefore, cells/ml  * 2E-4 = mg dry wt/L.  The model
predicted the average biomass of algae for most of the period, but missed the timing of the algal
blooms; furthermore, this implementation almost completely missed a large bloom that occurred
during low-flow conditions in the summer of 1977.

                      Figure 7. Algal biomass in Coralville Reservoir
                      with constant volume.
50
?40
D)
E
830

-------
MODEL VALIDATION REPORTS
                        CHAPTER 2
In a second level of analysis, the large seasonal and annual changes in volume were incorporated into
the simulation.  Direct, time-varying observations of volume were not available, but volume
estimates were available for three pool elevations (see Introduction). The changing pool elevations
(Figure 8) and a linear regression of volume on pool elevation (Figure 9) were used to estimate
changing volume (Figure 10).

Figure 8. Observed Coralville pool elevations;  Figure 9. Regression of volume on Coralville
values are elevation above mean sea level.       pool elevation.
    705
    665
     10/01/74
             10/01/75
                      09/30/76
                              09/30/77
                                       09/30/78
                                                = 8.5
                                                d7'5
670    680     690    700    710
            POOL ELEVATION
                                                            -•- Observed -^- Estimated
                                                                                     720
                        Figure 10. Coralville volumes estimated from
                        pool elevations.
                           2.5E+08
                           O.OE+00
                              10/01/74   10/01/75   09/30/76   09/30/77   09/30/78
                                         Results

The time-varying volume and associated depth estimates provided more accurate representations of
nutrient levels and light penetration, resulting in more realistic simulations of algal biomass, as seen
from the arithmetic (Figure 11) and log plots (Figure 12).  The model was off by a factor of only
two in predicting the large algal bloom in 1977. A more rigorous test is the Kolmogorov-Smirnov
statistic, which compares the cumulative distributions of the predicted and observed data (Figure
                                           2-5

-------
MODEL VALIDATION REPORTS
                           CHAPTER 2
13). As seen in Table 2, the distributions are not significantly different. The behavior of the model
is deemed acceptable, considering that the model was not calibrated or "fit" to the observed data, and
the simulation period spanned high-flow, low-flow, and normal hydrologic periods. The large bloom
may have been predicted even better had the simulation been set up to represent the temporary
stratification that  occurred at that time. (AQUATOX can represent stratified conditions; but
Coralville Reservoir was modeled as a completely mixed system because it stratifies so seldom.)
  Figure 11. Algal biomass (arithmetic) in
  Coralville Reservoir with changing volume.
50
^ An -
0>
^30
ro
E
o on
m20
ji 10

•


•

I*
ftj J
I \l\
n .*/ trr A
^ A^ft**^ ^f^^-^f 1 A» .J^Ji^!
1 0/0 1 /74 05/1 5/76 1 2/28/77
07/24/75 03/07/77
— Predicted Algae • Observed Algae

Figure 12. Algal biomass (logarithmic) in
Coralville Reservoir with changing volume.
100
Total Biomass (mg/L)
10
1

0.1
n n-i



\ J v»\ » / *^T •
5 [• • \ i \
\. / . — 3 — • — \-
\^ ^^ =5


/ ^ r •
\1 *S
*J •*
V*





10/01/74 05/15/76 12/28/77
07/24/75 03/07/77
— Predicted Algae • Observed Algae
                          Figure 13. Cumulative distribution of
                          Algal biomass in Coralville Reservoir.



^0
U

1

§
O






100
90
80 H
70^
60
50^
40^
30^
20^
0J





,.--• ^ 	 — 	 	
^
/f
(
{
1
j


I
0 10 20 30 40 50
Algae (mg/L)
/~\lT.cijai-Trjij-l "Dr-ja.rli.n+.n A
Observed .Predicted















                                          2-6

-------
MODEL VALIDATION REPORTS
                            CHAPTER 2
                Table 2. Summary Statistics for Algae in Coralville Reservoir
Time Period
1974-1978
Group
Observed
Predicted
Algae (mg/L)
No. of Obs.
61
61
Mean
4.82
3.81
Median
1.51
2.27
Std. Dev.
9.31
4.32
Std. Err.
1.19
0.55
p-value from
KS Test
0.08 la
 "• Not significantly different at 
-------
MODEL VALIDATION REPORTS
                             CHAPTER 2
  Figure 16. Predicted biomass of Coralville
  buffalofish, compared with mean observed
  value.
i9n
G* 100
^)
<§ 80
in
m
g 60
0
m 40
° 20
10/0




4_
n
-J i
•




	
fl(V
,.
fl

—

I


J^_ 	 J\l
"^ I
rV
\

\-~
fli

i
ii
I,*
v i
I









1/74 12/27/75 03/23/77 06/18/78
05/15/75 08/09/76 11/04/77
— Predicted buffalofish - - Mean predicted
Mean observed
Figure 17. Predicted rates for buffalofish (the
4th animal in the simulation) in Coralville
Reservoir for 1975.
                                                               gamete loss 7.3
                                                  0
                                                 01/01/75      05/15/75       09/26/75
                                                       03/09/75       07/21/75       12/02/75
                                                   | Consumption^ Q Defecation4   m Respiration4
                                                   ] Excretion4   Q Mortality4    B GameteLoss4
The predicted buffalofish biomass is a function of consumption (the only plus term), defecation,
respiration, minimal excretion, and, at times, spawning, and mortality.  The predicted in situ rates
for 1975 (Figure 17) indicate that the rate of consumption (Consumption4) was quite variable and
that loss due to defecation (Defecation4) was equally variable. At times, such as in early May, the
predicted diet was largely zoobenthos and defecation loss was very low; at other times detritus was
an important component of the diet,  and defecation of this low-quality  food source was high.
Predicted respiration rate (Respiration4) was low during the winter months, and mortality rate
(Mortality4) was high in late winter and early spring. Due to the way spawning is represented in the
model (as a function of temperature and growth), the predicted rate of gamete loss (GameteLoss4)
had a sharp peak in May and a smaller peak in September. Detailed observations offish dynamics
are sparse, but Mitzner (1972) reports that the success of buffalofish age classes is highly variable
in the reservoir and that mortality due to anoxia occurs in some years when spring runoff causes high
BOD.

Biomass data for other organisms in  Coralville Reservoir are scarce, so that  only qualitative
comparisons can be made with model predictions (Figure 18 and Figure 19). Zooplankton biomass
is small in both the simulations and in the actual Coralville ecosystem; a maximum of 2.3 mg/L was
observed in the lake (McDonald 1979) and that same value was predicted by the model. Zoobenthos
biomass fluctuates in the simulation depending on sediment conditions and predation; a maximum
of 2.6 mg/L is predicted. No quantitative estimates of either biomass or numbers were available
from the field, but McDonald (1979) states  that zoobenthos are sparse.

Bluegills are unimportant in both the lake and in the simulation (predicted mean of 0.5 mg/L for
1977-78). However, adult bass (but not young-of-year) are predicted to be important (predicted mean
of 9.9 mg/L for  1977-78). Interestingly, bass are absent in Coralville sport fish harvest data (Leidy
and Jenkins,  1977) and in gill net catches  (Mitzner 1972).  Perhaps their absence is due to the
periodic low dissolved oxygen levels—the known sensitivity of bass to low oxygen levels is not well
simulated by the model. Alternatively, agricultural pesticides, such as dieldrin, are known to be
                                           2-8

-------
MODEL VALIDATION REPORTS
                          CHAPTER 2
present, and may be causing toxicity. In order to test this possibility, dieldrin would have to be
included as a state variable in the simulation.

 Figure 18. Predicted Coralville invertebrates.     Figure 19. Other predicted Coralville fish.
        01/74      12/27/75      03/23/77      06/18/78
            05/15/75     08/09/76      11/04/77
     — Zoobenthos        — Herb. Zooplankton
      - Predatory zooplankton
0/01/74      12/27/75      03/23/77      06/18/78
     05/15/75      08/09/76      11/04/77

   — Bluegill   --YOYbass —Adult bass
The simulated and reported ranges in nutrient levels are comparable. The simulated nutrient levels
vary widely during the period as a function of the varying hydrologic regime and assimilation by
algae (Figure 20). Time varying nutrient data were not available, however, the observed maxima
correlated well with the predicted maxima.  In 1977-78 the maximum observed nitrate concentration
was 13 mg/L, compared to 11.2 in the simulation; the maximum observed ammonia value was 1.4,
compared to 1.04 in the simulation; and the maximum observed phosphate value was 0.85, compared
to 0.6 in the simulation.
                        Figure 20. Predicted nutrient concentrations.
100
:? 10
O)
E 1
c
'•s °-1
il
-i-»
§ 0.01
c
o 0.001
0 0001


*\ /***-. J^\ r^T1 \|
V \ i \ i \|
r~*~f<\ n 1 1 — \ — /vl — Til
14 r l\ A Gy
f ^7^1 '''^ \ /•' J i ~ /iii i' • nfl

A I1 '^Vrf >\|§ 4f= —
g V^ T 1 '

1
10/01/74 05/03/76 12/04/77
07/18/75 02/17/77 09/20
- NH4 — N03 — P04
                                            2-9

-------
MODEL VALIDATION REPORTS	CHAPTER 2

                                     Conclusions

When applied to Coralville Reservoir, a eutrophic run-of-the-river reservoir, AQUATOX yielded
results for algal biomass that were similar to those observed—predicting blooms under varying
hydrologic conditions. The model  also predicted the biomass of the dominant fish, buffalofish,
within 13% of the mean observed value.  Furthermore, with the exception of bass, AQUATOX
satisfactorily represented the qualitative importance of diverse groups of algae, invertebrates, and
fish. The model is shown to be suitable for simulating reservoir eutrophication, given standard water
quality data.

                                      References

Leidy, G.R., and R.M. Jenkins. 1977. The Development of Fishery Compartments and Population
       Rate Coefficients for Use in Reservoir Ecosystem Modeling. Contract Rept. CR-Y-77-1,
       U.S. Army Engineer Waterways Experiment Station, Vicksburg Mississippi, 134 pp.

McDonald, D.B. 1977. Coralville Water Quality Study Annual Report Water Year October 1,1975
       to September 30,  1976.  Report No. 200, Iowa Institute of Hydraulic Research,  The
       University of Iowa, Iowa City, Iowa, 48 pp.

McDonald, D.B. 1978. Coralville Water Quality Study Annual ReportWaterYear October 1,1976
       to September 30,  1977.  Report No. 213, Iowa Institute of Hydraulic Research,  The
       University of Iowa, Iowa City, Iowa, 17 pp. + 35 tables.

McDonald, D.B. 1979. Coralville Water Quality Study Annual Report Water Year October 1,1977
       to September 30,  1978.  Report No. 222, Iowa Institute of Hydraulic Research,  The
       University of Iowa, Iowa City, Iowa, 22 pp. + 25 tables.

McDonald, D.B. and M.P. McDonald. 1976. Coralville Water Quality Study Annual Report Water
       Year October 1, 1974 to September 30, 1975. Report No. 187, Iowa Institute of Hydraulic
       Research, The University of Iowa, Iowa City, Iowa, 78 pp.

Mitzner, L.  1972.  Population Studies ofBigmouth Buffalo in Coralville Reservoir with Special
       Reference to Commercial Harvest. Iowa Fisheries Research Technical Series No. 72-3, Iowa
       Conservation Commission, 36 pp.
                                         2-10

-------
MODEL VALIDATION REPORTS                                         CHAPTER
    VALIDATION OF AQUATOX 1.68 FOR PREDICTING BIOACCUMULATION
                   OF PCBS IN THE LAKE ONTARIO FOOD WEB

                                     Introduction

Certain chemicals are persistent in aquatic systems and tend to accumulate in the tissue offish and
other  aquatic organisms,  sometimes to levels that  make them unsafe  for human or wildlife
consumption. The "bioaccumulative" compounds, such as polychlorinated biphenyls (PCBs), are
the cause of a significant number of advisories against  fish consumption throughout the U.S.
Exposure to bioaccumulative compounds may be through  water, contaminated sediment, or diet.

The extent to which a chemical bioaccumulates depends upon many factors, such as trophic structure
(food-chain and food-web relationships), ambient water and sediment characteristics, the chemical
characteristics of the pollutant, and the lipid content of the exposed organisms. The bioaccumulation
factor (BAF) relates tissue concentration to ambient water concentration of the chemical. BAFs may
be useful in setting ambient water quality standards for bioaccumulative chemicals.  The various
chemical and biological processes that drive bioaccumulation may vary over time. For example,
food consumption and therefore dietary exposure will vary with seasonal cycles of prey organisms.
However, for purposes of setting water quality standards, regulatory agencies are usually concerned
with protection based on lifetime average consumption, and thus with long-term average conditions
of the water body and of the fish that are consumed.

Several mechanistic models have been developed to predict BAFs and fish tissue concentrations. For
lipophilic compounds, the octanol-water partition coefficient (Kow) is often used as an indication of
the compound's bioaccumulative potential. Thomann et al. (1992) and Gobas (1993) developed
steady-state models that rely on Kow, organic content of the sediment, lipid content of the organisms,
and position in the food web to predict tissue concentrations in the organisms.  A recent paper
(Burkhard 1998) compared the Thomann and Gobas models against an observed dataset from Lake
Ontario  (Oliver and Niimi  1988).  The two models performed similarly for compounds with log
Kows in  the range of three to eight, and they were in good agreement with the observed data.  The
models were shown to be highly sensitive to several input parameters. Sources of uncertainty also
were identified. The purpose of this study is to evaluate the performance of the AQUATOX model
using the same assumptions and dataset in so  far as possible.

                                   Model Structure

The AQUATOX model is a general ecological risk assessment model that represents the combined
environmental fate and effects of conventional pollutants, such as nutrients and sediments, and toxic
chemicals in aquatic ecosystems.  It considers  several trophic levels, including attached and
planktonic algae and submerged aquatic vegetation, invertebrates, and forage, bottom-feeding, and
game fish; it also represents associated organic toxicants. It can be implemented as a simple model
(indeed, it has been used to simulate an abiotic flask) or as a truly complex food-web model.  Food-
web modeling is now considered necessary for bioaccumulation studies other than screening level
(Abbott et al. 1995). "Food web models provide a means for validation because they mechanistically
                                          3-1

-------
MODEL VALIDATION REPORTS	CHAPTER 3

describe the bioaccumulation process and ascribe causality to observed relationships between biota
and sediment or water" (Connolly and Glaser 1998).

The  fate portion  of the model,  which is  applicable especially to organic  toxicants, includes:
partitioning among organisms, suspended and sedimented detritus, suspended and sedimented
inorganic sediments, and water;  volatilization; hydrolysis;  photolysis; ionization; and microbial
degradation.  Earlier versions of AQUATOX had two modes for representing partitioning of a
pollutant, equilibrium fugacity and kinetic partitioning. However, the equilibrium fugacity mode
was found to have limited applicability and was discontinued.  The effects portion of the model
includes: chronic and acute toxicity to the various organisms modeled; and indirect effects such as
release of grazing and predation pressure, increase in detritus and recycling of nutrients from killed
organisms, dissolved oxygen sag due to increased decomposition, and loss of food base for animals.

Partition Coefficients

Although AQUATOX is a kinetic model, steady-state partition coefficients for organic pollutants
are computed in order to place constraints on competitive uptake and loss processes, speeding up
computations. They are estimated from empirical regression equations and the pollutant's octanol-
water partition coefficient.

Natural organic matter is the primary sorbent for neutral organic pollutants. Hydrophobic chemicals
partition primarily in nonpolar organic matter (Abbott et al. 1995).  Refractory detritus is relatively
nonpolar; its partition coefficient is a function of the octanol-water partition coefficient (N = 34, r2
= 0.93;  Schwarzenbach et al. 1993):

                                          = 1.38 • KOW^                             (1)
where:
       KOMRefrDetr   =     suspended refractory detritus-water partition coefficient (L/kg); and
       KOW        =     octanol-water partition coefficient (unitless).

This and the following equations are extended to polar compounds, following the work of Smejtek
and Wang (1993):


               KOMp,nfr =  1.38 •  KOW0*2 •  Nondissoc
                    RefrDetr                                                               /*\
                              +  (1 -Nondissoc) •  lonCorr •  1.38  • KOW082

where:
       Nondissoc    =     un-ionized fraction (unitless); and
       lonCorr      =     correction factor for decreased sorption, generally 0.1 (unitless).

Partitioning of bioaccumulative chemicals on organic carbon in sediments in Lake Ontario,  as
represented by the Oliver and Niimi (1988) data, exhibits a weak relationship with KOW (US EPA
1995, Burkhard 1998):

                                           3-2

-------
MODEL VALIDATION REPORTS _ CHAPTER 3


                                   KOC = 25  • KOW                                  (3)


where:
       KOC   =     the partition coefficient for particulate organic carbon-water (L/kg).

Converting to organic matter (assuming a conversion factor of 0.526) and generalizing to include
polar compounds, this relationship is used in AQUATOX for this project, and only this project, to
represent the partitioning of chemicals between water and refractory detritus in sediments:

           KOMRDetrSed =  13 ' KOW + t1 ~ Nondissoc) • lonCorr  - 13  • KOW          (4)


where:
       KOMRDetrSed   =     sedimented refractory detritus-water partition coefficient (L/kg).

There appears to be a dichotomy in partitioning; data in the literature suggest that labile detritus does
not take up hydrophobic compounds as rapidly as refractory detritus. Algal cell membranes contain
polar lipids, and it is likely that this polarity is retained in the early stages of decomposition.  KOC
does not remain the  same upon aging, death, and decomposition, probably  because of polarity
changes. In an experiment using fresh and aged algal detritus, there was a 100% increase in KOC
with aging (Koelmans et al. 1995).  KOC increased as the C/N ratio increased, indicating that the
material was becoming more refractory. In another study, KOC doubled between day 2 and day 34,
probably due to deeper penetration into the organic matrix and lower polarity  of the partially
decomposed material (Cornelissen et al. 1997).

Polar substrates increase the pKa of the compound  (Smejtek and Wang 1993). This is represented
in the model by lowering the pH of polar particulate material by one pH unit, which changes the
dissociation accordingly.  The partition equation for labile detritus (N = 3, r2 = 1.0;) is based  on a
study by Koelmans et al. (1995) using fresh algal detritus:
                              KOCLabPart = 23'44 ' KOW™1                            (5)
where:
       KOCLabPart    =     partition coefficient for suspended labile organic carbon (L/kg).

The equation is generalized to polar compounds and transformed to an organic matter partition
coefficient:

          KOMLahDetr  = (23.44 •  KOW061  • Nondissoc
                        +  (1 -Nondissoc)  • lonCorr - 23.44 • KOW0-61) •  0.526
where:
       KOCLabPart    =     partition coefficient for suspended labile organic carbon (L/kg);
       KOMLabDetr    =     partition coefficient for suspended labile detritus (L/kg); and

                                           3-3

-------
MODEL VALIDATION REPORTS	CHAPTER 3

       0.526        =     conversion factor for organic carbon to organic matter.

O'Connor and Connolly (1980; see also Ambrose et al.,  1991) found that the sediment partition
coefficient is the inverse of the mass of suspended sediment, and DiToro (1985) developed a
construct to representthe relationship. However, AQUATOX models partitioning directly to organic
detritus and ignores inorganic sediments, which are seldom involved directly in sorption of neutral
organic pollutants. Therefore, the partition coefficient is not corrected for mass of sediment.

Association of hydrophobic compounds with colloidal and dissolved organic matter (DOM) reduces
bioavailability; such contaminants are unavailable for uptake by organisms (Stange and Swackhamer
1994, Gilek et al. 1996). Therefore, it is imperative that complexation of organic chemicals with
DOM be modeled correctly.  In particular, contradictory research results can be reconciled by
considering that DOM is not homogeneous: refractory humic acids, derived from decomposition of
terrestrial and wetland organic material, are quite different from labile exudates from algae and other
indigenous organisms.

Humic acids exhibit high polarity and do not readily complex neutral compounds. Natural humic
acids from a Finnish lake with extensive marshes were spiked with a PCB, but  a PCB-humic acid
complex could not be demonstrated (Maaret et al. 1992). In another study, Freidig et al. (1998) used
artificially prepared Aldrich humic acid to determine a humic acid-DOC partition coefficient for
several dissimilar chemicals (n = 5, r2, = 0.80), although they cautioned about extrapolation to the
field:

                                         =  28.84  - KOW™                            (7)
where:
       KOCRefrDOM   =     refractory dissolved organic carbon distribution coefficient (L/kg).

Until a better relationship is found, we are using a generalization of their equation to include polar
compounds, transformed from organic carbon to organic matter, in AQUATOX:

         KOMK ,nnM  =  (28.84 • KOW0-67 • Nondissoc
              KejrDUM    ^-                                                              fo\
                        +  (1  - Nondissoc) • lonCorr •  28.84 • KOW061) • 0.526
where:
       KOMRefrDOM   =     refractory dissolved organic matter distribution coefficient (L/kg).

Nonpolar lipids in algae occur in the cell contents, and it is likely that they constitute part of the
labile dissolved exudate, which may be both excreted and lysed material. Therefore, the stronger
relationship reported by Koelmans and Heugens (1998) for partitioning to algal exudate (n = 6, r2
= 0.926) is:

                               KOCLabDOC - °'88 ' KOW                              (9)
                                           3-4

-------
MODEL VALIDATION REPORTS	CHAPTER 3


 which was also generalized for polar compounds and transformed for organic matter:

           KOMT ,nn,, = (0.88 • KOW • Nondissoc
                LabDOM   ^                                                           (\(\\
                         +  (1  - Nondissoc) • lonCorr • 0.88 •  KOW)  • 0.526          l  '

where:
       KOCLabDOC    =     partition coefficient for labile dissolved organic carbon (L/kg); and
       KOMLabDOM   =     partition coefficient for labile dissolved organic matter (L/kg).

Unfortunately, older data and modeling efforts failed to distinguish between hydrophobic compounds
that were truly dissolved and those that were complexed with DOM. For example, the PCB water
concentrations for Lake Ontario, reported by Oliver and Niimi (1988) and used by many subsequent
researchers, included both dissolved and DOC-complexed PCBs (a fact which they recognized).
AQUATOX distinguishes  between truly dissolved and complexed compounds; therefore, the
partition coefficients may be larger than those used in older studies.

Bioaccumulation  of PCBs in algae depends on solubility,  hydrophobicity  and molecular
configuration of the compound, and growth rate, surface area and type, and content and type of lipid
in the alga  (Stange and Swackhamer 1994). Phytoplankton may double or triple in one day and
periphyton  turnover may be so rapid that some PCBs will not reach equilibrium (cf. Hill and
Napolitano 1997). Therefore, one should use the term "bioaccumulation factor" (BAF) rather than
"bioconcentration factor," which implies equilibrium (Stange and Swackhamer 1994).

Hydrophobic compounds partition to lipids in algae, but the relationship is not a simple one.
Phytoplankton lipids can range from 3 to 30% by weight (Swackhamer and Skoglund 1991), and
not all lipids are the same. Polar phospholipids occur on the surface. Hydrophobic compounds
preferentially partition to internal neutral lipids, but those are usually a minor fraction of the total
lipids, and they vary depending on growth conditions and species (Stange and Swackhamer 1994).
Algal lipids have a much stronger affinity for hydrophobic compounds than does octanol, so that the
algal BAFlipid > Kow (Stange and Swackhamer 1994, Koelmans et al. 1995,  Sijm et al. 1998).

For algae, the approximation to estimate the dry-weight bioaccumulation factor (r2 = 0.87), computed
from  Swackhamer &  Skoglund's (1993) study  of numerous PCB congeners  with  a natural
phytoplankton assemblage, is:

                         log(KBAlga) - 0.41 + 0.91 • LogKOW                          (11)


where:
       KBAlga        =     partition coefficient between phytoplankton and water (L/kg).

Rearranging and extending to hydrophilic and ionized compounds:
                                          3-5

-------
MODEL VALIDATION REPORTS                                         CHAPTER
                 KB.,   =  2.57 • KOW091 •  Nondissoc
                    Alga
                           + (1 -Nondissoc) •  lonCorr •  2.57 • KOW091
Comparing the results of using these coefficients, we see that they are consistent with the relative
importance of the various substrates in binding organic chemicals (Figure 1). Binding capacity of
detritus is greater than dissolved organic matter in Great Lakes waters (Stange and Swackhamer
1994, Gileketal. 1996).  In a study using Baltic Sea water, less than 7% PCBs were associated with
dissolved organic matter and most were associated with algae (Bjork and Gilek 1999). In contrast,
in a study using algal exudate and a PCB, 98% of the dissolved concentration was as a dissolved
organic matter complex and only 2% was bioavailable  (Koelmans and Heugens 1998).

                      Figure 1. Partitioning to various types of organic
                      matter as a function of KOW.

                          1E+10 ,
                          1 E+09 ,                           • "
                        O 1 E+05 |   m  ••      _^.^-~f  *
                          1 E+04 f    _-— -"T~"~ *
                          1E+03 r""l
                          1E+02 *
                               345678
                                            Log KOW

                        — humic acids    algae         exudate
                         ' algal detritus  * refr. detritus  - sediments
For macrophytes, an empirical relationship reported by Gobas et al. (1991) for 9 chemicals with
LogKOWs of 4 to 8.3 (r2 = 0.97) is used:
                         log(KBMacro)  - 0.98 • LogKOW -  2.24                          (13)


Again, rearranging and extending to hydrophilic and ionized compounds:

                    KEMacro = 0.00575  • KOW°98 •  (Nondissoc + 0.2)                  (14)
       For the invertebrate bioconcentration factor, the following empirical equation is used, based
on 7 chemicals with LogKOWs ranging from 3.3 to 6.2 and bioconcentration factors for Daphnia
pulex (r2 = 0.85; Southworth et al., 1978; see also Lyman et al., 1982), converted to dry weight:
                                          3-6

-------
MODEL VALIDATION REPORTS _ CHAPTER 3


              \og(KBInvertebrate) =  (0.7520 • LogKOW -  0.4362)  • WetToDry              (15)


where:
          invertebrate     =     partition coefficient between invertebrates and water (L/kg); and
       WetToDry     =     wet to dry conversion factor (unitless, default = 5).

Extending and generalizing to ionized compounds:

           KB invertebrate =  WetToDry •  0.3663 •  KOW0-7520 • (Nondissoc +  0.01)           (16)


Fish take longer to reach equilibrium with the surrounding water; therefore, a nonequilibrium
bioconcentration factor is used. For each pollutant, a whole-fish bioconcentration factor is based on
the lipid content of the fish extended to hydrophilic chemicals (McCarty et al., 1992), with provision
for ionization:

                 KBFish = Lipid • WetToDry •  KOW • (Nondissoc + 0.01)                 (17)


where:
       KBFish         =     partition coefficient between whole fish and water (L/kg);
       Lipid         =     fraction offish that is lipid (g lipid/g fish); and
       WetToDry     =     wet to dry conversion factor (unitless, default = 5).

Lipid content offish is varied depending on the potential for growth as predicted by the bioenergetics
equations; the initial lipid values for the species are given. The bioconcentration factor is adjusted
for the time to reach equilibrium as a function of the clearance or elimination rate and the time of
exposure (Hawker and Connell, 1985; Connell and Hawker, 1988):
                       ECF    -  fCR    • (l  - P (-Eltmination • TElapsed)]                       ,., „,
                       n^-r Fish    fU3Fish   V1    K                 '                       (L°)
where:
       BCFFish       =     quasi-equilibrium bioconcentration factor for fish (L/kg);
       TElapsed     =     time elapsed since fish was first exposed (d); and
       Elimination   =     combined clearance and biotransformation, see (40) (1/d).
                                           3-7

-------
MODEL VALIDATION REPORTS
                                               CHAPTER
                                       Figure 2
                            Bioconcentration factor for fish
                           as a function of time and log KOW
                   O
                   tr
                   LU
                   o
                   z
                   o
                   o
                   o
                   CQ
                           1E7
                           1E5
1E4
                           1E3
                                     tog
          teg
             KOW =
                              0    200   400  600  800  1000  1200
                                              DAY
Nonequilibrium Kinetics

Often there is an absence of equilibrium due to growth or insufficient exposure time, metabolic
biotransformation,  dietary  exposure, and nonlinear  relationships  for  very  large  and/or
superhydrophobic compounds (Bertelsen et al. 1998). Although it is important to have a knowledge
of equilibrium partitioning because it is an indication of the condition toward which systems tend
(Bertelsen et al. 1998), it is often impossible to determine steady-state potential due to changes in
bioavailability and  physiology (Landrum  1998).  PCBs may not be at steady  state even in large
systems such as Lake Ontario that have been polluted over a long period of time—the challenge is
to obtain sufficient data for a kinetic model (Gobas et al.  1995). In fact, PCBs  in Lake Ontario
exhibit a 25-fold disequilibrium (Cook and Burkhard 1998).

Sorption and Desorption to Sedimented Detritus

Partitioning to sediments appears to involve rapid sorption to particle surfaces, followed by slow
movement into, and out of, organic matter and porous aggregates (Karickhoff and Morris, 1985);
therefore, attainment of equilibrium may be slow.  This applies to suspended detritus compartments
as well. Because of the need to represent sorption and desorption separately  in detritus, kinetic
formulations are used (Thomann and Mueller, 1987), with provision for ionization:
            Sorption = klDetr •  Toxicantwater •  DifflCarrier
                       •  Org2C •  Detr •  le-6
                              (Nondissoc + 0.01)
                                                            (19)
                      Desorption = k2Detr •  Diff2Carrier •  ToxicantDetr
                                                            (20)
                                          3-8

-------
MODEL VALIDATION REPORTS
                                                 CHAPTER
where:
       Sorption
       kl
       Nondissoc
       ToxicantWater
       Diff2Carner     =
       Org2C

       Detr
       le-6
       Desorption    =
       k2
       Toxicant
               'Detr
rate of sorption to given detritus compartment (• g/L*d);
sorption rate constant for given compartment (L/kg«d);
fraction not ionized (unitless);
concentration of toxicant in water (• g/L);
factor  to  normalize rate  constant  for  given  carrier  (detritus
compartment in this case)  based on all  competing  uptake rates
(unitless);
factor to normalize loss rates (unitless);
conversion factor for organic matter to carbon (= 0.526 g C/g organic
matter);
mass of each of the detritus compartments per unit volume (mg/L);
units conversion (kg/mg);
rate  of desorption  from given  sediment  detritus compartment
C g/LM);
desorption rate constant for given compartment (1/d); and
mass of toxicant in each of the detritus compartments (• g/L).
Because there are several processes competing for the dissolved toxicant, the rate constants for these
processes are normalized in order to preserve mass balance.  The Diffl factor is computed for each
direct uptake process by  the various carriers, including sorption to detritus and algae, uptake by
macrophytes, and uptake across animals' gills:
                             Diffl
                                             RateDiffl
                                                      Carrier
                                  Carrier
                                               RateDiffl
                                                             (21)
                                                        Carrier
                        RateDiffl r   .  =  Gradientl
                             • kl
                        Carrier     Carrier
(22)
                   Gradientl
                                     Toxicant•„,_.__ •  kpr__,__ - PPB
                                             Water
                                                      Carrier
                                                                  Carrier
                            Carrier
                                           Toxicant
                                                    Water
                                                            Carrier
                                                             (23)
where:
       RateDiffl Camer =


       Gradient lcamer =


       ^Carrier        ~
       PPfi          =
       1 -* ^Carrier
maximum rate constant for uptake given the concentration gradient
(L/kg-d);
gradient between potential and actual concentrations of toxicant in
each carrier (unitless);
partition or bioconcentration factor for each carrier (L/kg);
concentration of toxicant in each carrier (• g/kg).
                                            3-9

-------
MODEL VALIDATION REPORTS
                                                                           CHAPTER
Likewise, the loss rate constants are normalized; the equations parallel those for uptake, with the
gradient being reversed:
                                            RateDiffl >Carrier
                                  Carrier
                                                                                        (24)
RateDiff2Carrier = Gradient2
                                                  Carrigr •  k2Carrier
                                                                                        (25)
                                                PFB

                                                     Carrier
where:
       RateDiff2Carrier =      maximum rate constant for loss given the concentration gradient
                           (L/kg«d); and
       Gradient2Carrier =     gradient between actual and potential concentrations of toxicant in
                           each carrier (unitless).

       Desorption of the slow compartment is the reciprocal of the reaction time, which Karickhoff
and Morris (1985) found to be a linear function of the partition coefficient, expressed in hours, over
three orders of magnitude (r2 = 0.87):

                                — « 0.03 • 24  • KPSed
                                k2
                                                                                        (27)
                                                                                           '
       So k2 is taken to be:
                                             1.39
where:
       KPSed
       24
                           detritus-water partition coefficient (L/kg, see Eq. (2)); and
                           conversion from hours, as used by Karickhoff and Morris (1985), to
                           days.
The slow compartment may be involved in 40 to 90% of the sorption so, as a simplification, fast
desorption  of the labile compartment is  ignored.  This compensates in  part for the fact that
AQUATOX models the top layer of bottom  sediments as if it were in close contact with the
overlying water column (interstitial water is not modeled at this time).
                                          3-10

-------
MODEL VALIDATION REPORTS
                                                                    CHAPTER
The sorption rate constant is set to 1200 L/kg«d in the code, representing the very fast sorption of
most chemicals.
Bioconcentration in Macrophytes and Algae

Macrophytes—As Gobas et al. (1991) have shown, submerged aquatic macrophytes take up and
release organic chemicals over a measurable period of time at rates related to the octanol-water
partition coefficient. Uptake and elimination are modeled assuming that the chemical is transported
through both aqueous and lipid phases in the plant, with rate constants using empirical equations fit
to observed data (Gobas et al., 1991), modified to account for ionization effects (Figure 3, Figure
4):
               Uptakeplant  = kl •  Diffl
                                     Plant
                                     Toxicant
                                                    Water
                              StVar
                                                        Plant
le-6
(29)
                         ClearPlant  - k2 •  Toxicantplant •  Diff2plant
                                                                                  (30)
                            kl  =
                                  0.0020 +
                                                   500
                                            KOW • Nondissoc
                                                                                  (31)
                       k2  =
                                               1
                              1.58  + 0.000015  • KOW •  Nondissoc
                                                                                  (32)
where:
Uptakeplant
Clearplant
StVarPlant
1 e-6
Toxicant
kl
k2
               'Plant
           Plant

       KOW
       Nondissoc
uptake of toxicant by plant (• g/L«d);
clearance of toxicant from plant (• g/L»d);
biomass of given plant (mg/L);
units conversion (kg/mg);
mass of toxicant in plant (• g/L);
sorption rate constant (L/kg»d);
elimination rate constant (1/d);
factor  to normalize rate constant for given plant based  on all
competing uptake rates (unitless);
octanol-water partition coefficient (unitless); and
fraction of un-ionized toxicant (unitless).
                                          3-11

-------
MODEL VALIDATION REPORTS
                                     CHAPTER
   Figure 3. Uptake rate constant for
   macrophytes (after Gobas et al.,  1991)
    500

    400

    300

    200

    100

      0
                   468
                    Log KOW


                - Predicted •  Observed
10
             Figure 4. Elimination rate constant for
             macrophytes (after Gobas et al., 1991)
              0.7
              0.6
              0.5
              0.4
              0.3
              0.2
              0.1
               0
2468
       Log KOW


 — Predicted -  Observed
10
Algae—There is probably a two-step algal bioaccumulation mechanism for hydrophobic compounds,
with rapid surface sorption of 40-90% within 24 hours and then a small, steady increase with transfer
to interior lipids for the duration of the exposure (Swackhamer and Skoglund 1991).  Uptake
increases with increase in the surface area of algae (Wang et al. 1997). Therefore, the smaller the
organism the larger the uptake rate constant (Sijm et al. 1998). However, in small phytoplankton,
such as the nannoplankton that dominate the Great lakes, a high surface to volume ratio can increase
sorption, but high growth rates can limit internal contaminant concentrations (Swackhamer and
Skoglund 1991). AQUATOX uses a generalized uptake construct, but explicitly models growth rate
and the effect on the BAFs.

The kinetics of partitioning of toxicants to algae is based on studies on PCB congeners  showing
uptake to be very rapid. Sijm et al. (1998) presented data on several congeners that were used in this
study to develop the following relationship for phytoplankton (Figure 5):
                         kl  =
                                               1
                                1.8E-6 + \I(KOW - Nondissoc}
                                                 (33)
Based in part on Skoglund et al. (1996), but ignoring surface sorption and recognizing that growth
dilution is explicit in AQUATOX, the elimination rate constant (Figure 6) is computed as:
                                      k2  =
        kl
       KOW
                           (34)
                                          3-12

-------
MODEL VALIDATION REPORTS
                                               CHAPTER
   Figure 5. Algal sorption rate constant.
        FIT TO DATA OF SUM ET AL. 1*98
    600000
  -oSQOQOO
^300000 :
a 2QQOGG •
5 100000 -
0
::::•>:
246
LOG KOW
                       Figure 6. Rate of elimination by algae.

                        1.2       	      •      |

                         1

                      |"0.8

                      20,6
                                     10
                                                                  4     6
                                                                   Log KOW
                                                          10
                 •  ObsKI ^

Bioaccumulation in Animals
Animals can absorb toxic organic chemicals directly from the water through their gills and from
contaminated food through their guts.  Direct sorption onto the body is assumed to be negligible in
this version of AQUATOX. Reduction of body burdens of organic chemicals is accomplished
through excretion and biotransformation, which are  often  considered together as  empirically
determined elimination rates. "Growth dilution" occurs when growth of the organism is faster than
accumulation of the toxicant.  Fecal  loss is important as an input to the detrital toxicant pool.
Inclusion of mortality and promotion terms is necessary for mass balance, but emphasizes the fact
that average concentrations are being modeled for any particular compartment.

Gill Sorption—An important route of exposure is by active transport through the gills (Macek et
al., 1977). This is the route that has been measured so often in bioconcentration experiments with
fish.  As the organism respires, water is passed over the  outer surface of the gill and blood is moved
past the inner surface.  The exchange of toxicant through the gill membrane is assumed to be
facilitated by the same mechanism as the uptake of oxygen, following the approach of Fagerstrom
and Asell  (1973, 1975), Weininger (1978), and Thomann and Mueller (1987; see also Thomann,
1989).  Therefore, the uptake rate for each animal can be calculated as a function of respiration
(Leung, 1978; Park et al., 1982):
                    GillUptake = KUptake  • Toxicant
                                                   'Water
                                                               Carrier
                                                             (35)
                              -  WEffTox • Respiration • O2Biomass
                                          Oxygen •  WEffO2
                                                             (36)
where:
       GillUptake   =
       KUptake
       ToxicantWater  =
       Difflcarner    =
       WEfJTox
uptake of toxicant by gills (• g/L - d);
uptake rate (1/d);
concentration of toxicant in water (• g/L);
factor  to  normalize  rate  constant  for  given  carrier  (animal
compartment in this case) based on all  competing uptake rates
(unitless);
withdrawal efficiency for toxicant by gills (unitless);
               3-13

-------
MODEL VALIDATION REPORTS
                                              CHAPTER
      Respiration
      O2Biomass

      Oxygen
      WEffO2
respiration rate (mg biomass/L»d);
ratio of oxygen to organic matter (mg oxygen/mg biomass; generally
0.575);
concentration of dissolved oxygen (mg oxygen/L); and
withdrawal efficiency for oxygen (unitless).
The oxygen uptake efficiency WEffO2 is assigned a constant value of 0.62 based on observations of
McKim et  al.  (1985).  The toxicant  uptake efficiency can be expected to have a sigmoidal
relationship to the log octanol-water partition coefficient based on aqueous and lipid transport
(Spacie and Hamelink, 1982); this is represented by an inelegant but reasonable,  piece-wise fit
(Figure 7) to the data of McKim et al.  (1985) using 750-g trout, corrected for ionization:
                                If  LogKOW < 1.5  then
                                    WEffTox  = 0.1

                             If 1.5 < LogKOW > 3.0 then
                 WEffTox =  0.1  + Nondissoc •  (0.3 •  LogKOW -  0.45)

                             If 3.0 < LogKOW < 6.0 then
                          WEffTox = 0.1 + Nondissoc  • 0.45

                             If 6.0 < LogKOW < 8.0 then
             WEffTox =  0.1  + Nondissoc •  (0.45  -  0.23  • (LogKOW -  6.0))

                                If LogKOW > 8.0  then
                                    WEffTox  = 0.1
                                                           (37)
where:
      LogKOW     =     log octanol-water partition coefficient (unitless); and
      Nondissoc    =     fraction of toxicant that is un-ionized (unitless).

                      Figure 7. Piece-wise fit to observed toxicant
                      uptake data; modified from McKim et al., 1985.
                                       345
                                          LOG KOW
                                         3-14

-------
MODEL VALIDATION REPORTS
                                                                CHAPTER
lonization decreases the uptake efficiency. This same algorithm is used for invertebrates. Thomann
(1989) has proposed a similar construct for these same data and a slightly different construct for
small organisms, but the scatter in the data do not seem to justify using two different constructs.

Dietary Uptake—Hydrophobic chemicals usually bioaccumulate primarily through absorption from
contaminated food.  Persistent, highly hydrophobic chemicals demonstrate biomagnification or
increasing concentrations as they are passed up the food chain from one trophic level to another;
therefore, dietary exposure can be quite important (Gobas et al., 1993). Uptake from contaminated
prey can be computed as (Thomann and Mueller, 1987; Gobas, 1993):
                        DietUptakeprey = KDprey • PPBprey • le-6                        (38)
                           KDP   = GutEffTox • Ingestion
                               Prey
                                               Prey
                                                             (39)
where:
       DietUptakePrey =
       KDj
       PPBt
       1 e-6
       GutEfjTox
       Ingestiorip
Prey

 Prey
uptake of toxicant from given prey (• g toxicant/L«d);
dietary uptake rate for given prey (mg prey/L»d);
concentration of toxicant in given prey  (• g toxicant/kg prey);
units conversion (kg/mg);
efficiency of sorption of toxicant from gut (unitless); and
ingestion of given prey (mg prey/L»d).
Gobas (1993) presents an empirical equation for estimating GutEffTox as a function of the octanol-
water partition coefficient.  However, data published by Gobas et al. (1993) suggest that there is no
trend in efficiency between LogKOW4.5 and 7.5 (Figure 8); this is to be expected because the
digestive system has evolved to assimilate a wide variety of organic molecules. Therefore, the mean
value of 0.63 is used as a constant for small  fish. Nichols et al. (1998) demonstrated that uptake is
more efficient in larger fish; therefore, a value of 0.90 is used for large game fish. Invertebrates

                     Figure 8. GutEffTox constant based on mean value
                     for data from Gobas et al., 1993.
generally  exhibit
Dietary Absorption Efficiency
1
0.75
0.5
0.25
0
4.

•
• * «
»
5 5 5.5 6 6.5 7 7.5
Log KOW
Guppies Goldfish — Mean = 0.63
                                                            lower efficiencies;
                                          3-15

-------
MODEL VALIDATION REPORTS	CHAPTER 3

Landrum and Robbins (1990) showed that values ranged from 0.42 to 0.24 for chemicals with log
KOWs from 4.4 to 6.7; the mean value of 0.35 is used for invertebrates in AQUATOX.

Elimination—Elimination includes  both excretion  and biotransformation  of a toxicant  by
organisms.  Biotransformation is difficult to model separately and may cause underestimation of
elimination (McCarty et al., 1992).  Therefore, an overall elimination rate constant is estimated and
the derived value is reported in the toxicity record. The user may then modify the value based on
observed data.

For purposes of  estimating elimination, a modification of Eq. (35) is used to compute uptake,
assuming a generalized allometric relationship between respiration and the mean  weight of the
animal (Thomann, 1989):

                           kl  = 1000 • WetWt'02 •  WEffTox                            (40)

where:
       kl            =      uptake rate (L/kg»d);
       WetWt        =      mean wet weight of organism (g);
       1000         =      units conversion (g/kg);
       WEfJTox      =      withdrawal efficiency for toxicant by gills, see Eq. (36) (unitless).

If, as Thomann (1989) assumes, lipid-normalized bioconcentration is equal to the  octanol-water
partition coefficient at equilibrium  and zero growth, then:

                 ,,                           kl
                k2  -  	                 (4i\
                      KOW • LipidFrac •  WetToDry  • (Nondissoc +  0.1)                 ^  '

where:
       k2            =      elimination rate constant (1/d); and
       KOW        =      octanol-water partition coefficient (unitless);
       LipidFrac    =      fraction of lipid in organism (g lipid/g organism);
       Wet2Dry      =      wet to dry weight ratio (5); and
       Nondissoc    =      fraction of compound un-ionized (unitless, 1.0 for PCBs).

Note that this is the only place where the lipid  fraction is used in modeling bioaccumulation in
AQUATOX. This simple relationship, although weak, has been used in AQUATOX for both
invertebrates and fish (Figure 9).  However, the fish curve seems to drastically underestimate
clearance at higher KOWs.  Therefore, as an alternative until the formulation is changed,  k2
estimates may be entered manually using as guides regression equations for Daphnia:
                        Log k2  =  -0.5688 • Log  KOW +  3.6445                       (42)


and small fish:
                         Log k2 =  -0.503 • Log KOW + 1.45                         (43)
                                         3-16

-------
MODEL VALIDATION REPORTS
                                              CHAPTER
                  Figure 9. Elimination rate constants for Daphnia and for
                  10-g fish; see Thomann, 1989.
4
~ 2
?= 0
2-2
0
-•-6
-8
0

**-»^
	












^^^
~~~~~~~~'~>^S-^

















^--^

f~- — .
• m








^.
•w
5^










•^^
1





246




-

R^**
^-








• 	
—

< 	
^~x^








~~ - -



\^
^














8 10
Log KOW
• Observed Fish — Estimated Fish
A Observed Invert. — Estimated Invert.
Daphnia regr Sm fish regr
For any given time the clearance rate is:
                            ClearAnimal = k2 '  ToxtcantAnimal
                                                            (44)
where:
       ClearAmmal
       ToxicantAmmal  =
clearance rate (• g/L«d); and
mass of toxicant in given animal (• g/L).

 Data Used for Model Evaluation
Data presented by Oliver and Niimi (1988) for various PCB congeners in Lake Ontario sediments,
water, and organisms were used to characterize pollutant distributions in that Great Lake food web.
Although the data are high quality, they are not synoptic; rather, water samples were taken in April,
1984; sediment samples were taken in May, 1981; suspended sediment samples from November to
April for 1982 to 1986; plankton (a mixture of phytoplankton and zooplankton) in July, 1982;
mysids in July, 1981 and October, 1984; benthos in June, 1985; sculpin in spring, 1986;  alewives
and smelt in May, 1982; smaller smelt (not used in the present study) in April, 1986; and salmonids
in fall, 1981, and April, 1982. Furthermore, water, sediment, and plankton samples were taken from
all three major basins in Lake Ontario; but fish and benthos samples were taken from the western
Niagara Basin.

Seventy-two PCB congeners were studied by Oliver and Niimi (1988). Because of computational
load, 16 congeners were selected for use in the evaluation of AQUATOX.  They were selected to
span the range  of Kows  and included  congeners with higher,  and therefore more  reliable,
concentrations.   The freely dissolved concentrations were computed from the  reported water
concentrations, which included both dissolved PCBs and those associated with dissolved organic
matter, using the approach of US EPA (1995).
                                         3-17

-------
MODEL VALIDATION REPORTS	CHAPTER 3

The feeding relationships of phytoplankton, mysids, the amphipodDiporeia, sculpin, alewife, smelt,
and salmonids were taken from Burkhard (1998, based on Flint 1986 and Gobas 1993). However,
feeding ratios are not constants; what is actually consumed in Nature depends on time-varying prey
biomass. Consumption as modeled in AQUATOX treats the preferences as weights for available
prey, mimicking the way predation changes in the real world. The assumption made by both Gobas
(1993) and Burkhard  (1998)  is that the plankton  analyzed by Oliver and Niimi  (1988) were
phytoplankton, although Oliver and Niimi state (p. 388) that a plankton sample from a depth of 10
m "would contain a mixture of phytoplankton and zooplankton, phytoplankton should predominate."
Furthermore, Mysis relicta is treated in the models as if it is the herbivorous or next trophic level,
although mysids are opportunistic feeders and prey heavily on zooplankton (www.fw.umn.edu/
nresexotics3001/mysisrelicta.html).  Therefore, cladocerans were included as the intermediate,
herbivorous trophic level, as they are in most implementations of AQUATOX. Feeding preferences
for both cladocerans and mysids were based on the literature. The assignment of animals to guilds
in AQUATOX  is as follows:  amphipods, detritivorous invertebrates; cladocerans, herbivorous
invertebrates; mysids,  predatory invertebrates; sculpin, benthic fishe' alewife,  forage fish; smelt,
nominally small game  fish; trout, large game fish.

Lipid content for the various organisms was based  on Oliver and Niimi (1988).  Using these as
constant lipid fractions is even more misleading; there is ample evidence, presented in Arts and
Wainman (1999) and elsewhere, that lipid content is a time-varying function of nutritional state at
all trophic levels. Furthermore,  many literature values for lipid content (cf Arts and Wainman 1999),
even in Lake Ontario, are two  or more times those reported by  Oliver and Niimi (1988), allowing
for wet and dry weights. Therefore, in this study the lipid values were held constant at the reported
values with some misgivings.  For comparison purposes, BAFs on a lipid-normalized and freely-
dissolved basis were  taken from  the tables in US EPA  (1995) and were checked using the
computational procedure presented in that document.

                   Code and Parameter Changes to Facilitate Analyses

AQUATOX is more mechanistic than the implementations of the Gobas and Thomann models used
by Burkhard (1998).  Several changes were made to the AQUATOX code, and even more changes
were made to the user-supplied  parameter values to facilitate the comparison with Burkhard's (1998)
results. Because of the requirement that the  simulations  be run to steady-state with a constant
dissolved concentration, the code was changed to disable the differential equation for the dissolved
phase.  Burkhard (1998) set the dissolved concentration in water to 1  ng/L for all chemicals,
primarily because there is no concentration-dependent feedback in the Gobas and Thomann models.
However,  several of the congeners exhibit dioxin-like toxicity, and the model predicted high
bioaccumulation levels that caused chronic toxicity to be manifested over the course of a seven-year
simulation period.  Therefore, the  dissolved concentrations were set to congener-specific levels
calculated from the observed concentrations.

The constant dissolved concentration constraint also removed  volatilization as a loss term.  The
model was parameterized so that microbial degradation was not a factor. Version 1.68, which was
the version used, does  not model biotransformation separately from depuration.

Burkhard (1998) assumed well mixed conditions, but AQUATOX simulates the dynamics of the
ecosystem as well as the fate and effects of the pollutant.  Stratified conditions characterize Lake
                                         3-18

-------
MODEL VALIDATION REPORTS	CHAPTER 3

Ontario during most of the growing season. A simulation without stratification showed that the
pelagic ecosystem in this deep lake requires a well defined epilimnion in order for phytoplankton
dynamics to be represented realistically.  Therefore, the simulations were driven by time-varying
epilimnetic and hypolimnetic temperatures with mean values (7.4° and 3.6°) and ranges taken from
the literature (Canada Centre for Inland Waters, 1979, p 97), and stratification was correctly
modeled.

AQUATOX models time-varying lipid fractions. That function was disabled in the code so that lipid
fractions were held constant at the initial values. However, in AQUATOX, lipid fractions in animals
only affect the estimations of the k2 (elimination) parameters. Based on initial runs, the estimates
of k2 values were found to be too low at higher Kows; therefore, they were replaced by estimated
values using (41) and (42). The result is that the model is insensitive to lipid values in all but algae.

The model was modified so that the partition coefficient for refractory detrital carbon in sediment
would be 25 times Kowusing (4). This was done only in this application to facilitate the comparison.
In the first simulations the only PCBs were those in water; however, steady-state was not reached
even after 16 years.  Therefore, the initial PCB concentrations in refractory detrital sediments were
set to observed values, correcting for organic carbon content in the sediment (2.7%), and greatly
shortening the time to steady-state.

In order to accommodate the numerous simulations required for the sensitivity analyses, the code
was modified to run in batch mode, and then copy the predicted BAFs at the end of the simulation
to a text file suitable for importing directly into an Excel spreadsheet. The initial simulations were
run for 16 years to determine time to steady-state. Inspection of the output  suggested that seven
years was more than sufficient to achieve steady state, even with large perturbations.  Therefore,
seven years was used for the  standard and sensitivity simulations.  Because of the time required to
run 100 simulations for the uncertainty analysis,  a four-year simulation period was used for those
simulations. That was determined to  be the minimum time  required to reach steady state for a
chemical with a log Kow of 6.5.

                                 Results and Discussion

AQUATOX simulations were calibrated for the Lake Ontario ecosystem with some difficulty
because the model had never before been applied to a Great Lake.  In particular, bioenergetic
parameters, such as maximum consumption rate (CMax), minimum biomass for feeding (BMin), and
respiration rates were only approximated for sculpin and smelt using other Great Lakes species. The
model was run for seven years with 1972-1973 loadings repeated; and, after a transient period, the
annual patterns were stable (Figure 10).  The phytoplankton  and zooplankton biomass levels and
seasonal trends were similar to those in the literature (Canada Centre for Inland Waters, 1979, Scavia
1980). Unfortunately, no biomass data were found for fish  to verify the calibrations,  and catch
statistics only covered lake trout, cisco and whitefish (Robertson and Scavia  1979).
                                          3-19

-------
MODEL VALIDATION REPORTS
                                                                    CHAPTER
Figure 10. Seven-year simulation of Lake Ontario with predicted epilimnetic biomass patterns.
             Oaamittia JD ,T.jyX tatiiai* in u&l unteta aMawag Xxilcsiea.
irt ]
                                     ynnt Sftup |  Prtnt CMrt (
       -  
-------
MODEL VALIDATION REPORTS
CHAPTER 3
Figure 11. Seven-year simulation of Lake Ontario with predicted bioaccumulation factors.
                      LAKE ONTVVPO pee IEO (PERTUREEDJ
The contributions of the various uptake and loss processes differ from one group to another (Figure
12,Figure 13).  Given the dynamics of uptake and loss, especially in the short-lived zooplankton,
and the averaging effects of disparate epilimnetic and hypolimnetic sediment concentrations, it is not
surprising that there are fluctuations in the BAFs for many compartments. Therefore, the tabulated
BAFs were taken at the end of the simulation period, March 31, which corresponded to the time
when many of the observed data were collected.

As can be seen from Figure 14 and Table 1, some predictions are remarkably close and others are
off by factors of as much as 8.6, as in the case of sculpin.  The only predictions that exhibited a
noticeably different trend were those for the phytoplankton BAFs, which diverged at lower Kows
(Figure 14A). The irregularities in the predictions mirror the fluctuations in the observed data and
are related to the varying concentrations of PCBs in the sediments. The best predictions are for lake
trout (Figure 14G), which are the most important fish in terms of human health hazard.  Mysid
BAFs are predicted very well (Figure 14B), as are smelt BAFs (Figure 14F). Phytoplankton BAFs
are underestimated (Figure 14A); this may reflect the fact that the observations are actually for
combined phytoplankton and zooplankton.  Amphipods are over-estimated (Figure 14C);  as will
be shown later, the model is sensitive to changes in their bioenergetic parameters, such as maximum
consumption  rate.  The overestimates of sculpin (Figure 14D) and alewife (Figure 14E) BAFs
emphasize the sensitivity  of the benthic food web as modeled.  The predicted partition coefficients
for refractory detrital  sediments (Figure 14H) are very close to the observed values.
                                          3-21

-------
MODEL VALIDATION REPORTS
                               CHAPTER
Figure 12. Predicted transfer rates for PCB
180 in Lake Ontario diatoms.
                   DIATOMS
                 11121/72     OTMIi'73     B2V3/74
           07/28^72      03H7/73      11, '0*73
          B Uptak«i2
           _J Depurrfwt2 | Excretion! B S«)"nerrt2
Figure 13. Predicted transfer rates for PCB
180 in Lake Ontario smelt.
                    SMELT
                                                           07/28^72
                           OT'Hi'73
                      Oi'17,73      11, '0*73
             PredationS H MortalityS
Figure 14. Observed and predicted lipid-normalized and freely dissolved BAFs for PCBs in Lake
Ontario ecosystem components.	
    11
    10 -

    9 -
  g
  03
A.
                 Phytoplankton
                   * *
                 6     7
                  Log KOW
                                                    11
    10 -

     9 -
B.
                    Mysids
                 6     7
                  Log KOW
                                            3-22

-------
MODEL VALIDATION REPORTS
CHAPTER
Figure 14. continued
Amphipods
11 -
10 -

9 -
m
2 7-
6 -
5 -
4 -
4

r
jsm
/s.
*


56789
Log KOW
c.
11 -
10 -
9 -
LL p
2 7 •
B -
5 -
4 -
L
Ale wife
m &, \
JS I
s s s * >
/\x** * ^
« » * I
f i

56789
Log KOW
E.
Lake Trout
11 -
10 -
9 -
< 8~
O5
2 7~
6 -
5 -
4 -

f i
/*ii i
*•* * i
» .,«» .-* i
ff '•
f *'* i
f*'-*l :
*/ k !
?" !
* i

» 5 6 7 8 9
Log KOW
G.








* Observed
s Predicted







* Observed
-s- Predicted










* Observed
^^ Predicted

























T

Sculpin
11 •
10 -

9 -
% B-
1 7-
6 -
5 -
4 -
i

H
/ a
H *

• ' * ** * Observed
/"'***» « Predicted
«

5 6 7 B 3
Log KOW
D.
11 -
10 -
9 -
< 8-
1 7«
B -
S -
4 -
Smelt

/«"
* «.«* t"1*
f *I»'lr * Observed
•j^**" -•— Predicted
s
s
«'
56789
Log KOW
F.
Refractory Sediment Detritus
11-
10 -
9 -
LL O
03
US)
5 7-
6 -
5 -
4 -


f
f \
»* -**
a-' * 'J * Observed
' s Predicted


456789
Log KOW
H.
3-23

-------
MODEL VALIDATION REPORTS
CHAPTER
he predicted and observed BAFs can be compared statistically by taking the ratio of the predicted
to the observed, as summarized in Table 1.  A ratio of 1.0 represents perfect correspondence
between the predicted and observed BAFs.  This is a quantitative  summary of the results shown
diagrammatically  in  Figure 14, with sculpin  and alewife  biasing the overall results.  The
bioenergetics equations for these fish most likely are not parameterized properly; additional literature
survey and calibration of the bioenergetic parameters would help correct the discrepancies.

Table 1.  Ratio of observed to predicted bioaccumulation factors.
Congener

18
28+31
84
70+76
66
101
110
105
149
118
138
153
187+182
180
203+196
194
Grand
Mean
Mean
Std Dev
Count
Minimum
Median
Maximum
Skewness
Kurtosis
Log KOW

5.24
5.67
6.04
6.17
6.2
6.38
6.48
6.65
6.67
6.74
6.83
6.92
7.19
7.36
7.65
7.8









Phyto-
plankton

0.07
0.13
0.05
0.19
0.19
0.49
0.37
0.40
0.04
0.51
0.68
0.71
1.15
1.41
0.39
1.78

0.53
0.51
16
0.04
0.39
1.78
1.34
1.20
Mysids

0.24
0.55
0.44
1.35
1.82
0.98
1.15
0.89
0.87
0.63
0.79
0.72
1.00
1.38
4.68
3.89

1.34
1.22
16
0.24
0.93
4.68
2.11
3.89
Amphipods

1.02
1.55
2.69
4.57
4.79
2.34
4.37
2.75
2.40
2.34
1.91
1.82
2.04
0.95
5.75
3.72

2.81
1.42
16
0.95
2.37
5.75
0.70
-0.43
Sculpin

3.89
18.62
2.51
21.88
13.49
3.47
9.55
5.62
16.60
3.72
3.39
3.89
6.31
4.07
10.96
10.00

8.62
6.14
16
2.51
5.97
21.88
0.98
-0.18
Alewife

0.71
4.57
1.86
6.61
5.50
2.19
4.57
4.07
3.24
3.09
2.95
3.89
4.68
5.25
15.14
13.80

5.13
3.94
16
0.71
4.32
15.14
1.85
3.05
Smelt


0.17
0.07
0.28
0.30
0.23
0.33
0.29
0.34
0.22
0.30
0.35
0.63
0.93
3.63
3.98

0.80
1.24
15
0.07
0.30
3.98
2.30
4.06
Trout

0.38
0.49
0.18
0.95
0.87
0.41
0.74
0.55
0.69
0.43
0.46
0.52
0.79
1.00
3.24
3.80

0.97
1.03
16
0.18
0.62
3.80
2.32
4.55
Mean for
congeners

1.05
3.73
1.11
5.12
3.85
1.44
3.01
2.08
3.45
1.56
1.50
1.70
2.37
2.14
6.26
5.85
2.89








A similar tabular summary was presented by Burkhard (1998) for the Gobas and Thomann models.
It is instructive to compare the results of the three models (Figure 15).  The heavy line indicates
unity for the mean and predicted ratios. This is not a rigorous comparison because AQUATOX was
only used to simulate 16 PCB congeners, in contrast to the  72 congeners simulated by Burkhard
(1998), although there is not any apparent bias in the smaller sample. Burkhard concluded that the
Gobas model seems to represent Lake Ontario bioaccumulation better than the Thomann model.
However, AQUATOX did equally as well for smelt and lake trout, and the best of the three models
for phytoplankton and mysids.
                                         3-24

-------
MODEL VALIDATION REPORTS
CHAPTER
Figure 15. Comparison of predicted/observed BAF ratios for the AQUATOX, Gobas, and
Thomann models.
      8.00
                                                                    DAQUATOX
                                                                    • GOBAS
                                                                    D THOMANN
      0.00
Sensitivity Analysis

Following the approach of Burkhard (1998), sensitivity analyses were run on each key parameter,
using the equation:
                               sensitivity
                                           BAF. • AP
            (45)
where:
      Pt     =     nominal value of input parameter /',
      • Pt   =     deviation in the input parameter /',
      BAFt  =     the BAF predicted using the nominal input values, and
      • BAFt =     the deviation in the predicted BAFs using the nominal and modified input
                   parameter /'.

A sensitivity of 1.0 means that a change in a parameter results in an equal change in the BAFs
predicted by the model.  A negative sign indicates that  the change is opposite in direction to the
parameter change.  Because the simulations were run for seven years to ensure steady-state,
sensitivity analysis for each parameter for the 16 PCB congeners took eight hours on a Pentium HI
                                        3-25

-------
MODEL VALIDATION REPORTS
CHAPTER
500 mHz machine. Therefore, only the following representative parameters were analyzed: Kow, k2,
preference of amphipods for refractory detritus, Cwaxamphipod, and BMinsculpin.

Burkhard (1998) found that Thomann's and Gobas' models were sensitive to variable Kow values
within a factor of+/- 2.0.  As shown in Figure 16, in AQUATOX phytoplankton are highly sensitive
to Kow\ furthermore, as shown in Figure 14A, the predicted algal BAF trend does not parallel the
observed trend  Both the phytoplankton kl uptake and k2 values are a direct function of the Kow
values. In contrast, none of the animals show any  sensitivity, reflecting the fact that by manually
entering k2 elimination values the dependence on Kow was overridden.
          Figure 16. Sensitivity of AQUATOX BAFs to a +10% change in Kt
                                                                      ow
n ?n -i
n nn -
n ">r\ -
>i n An -
.£ n en -
S2 n Rn -
* i nn -
1 9fl -
1 in -

1 en .
£
KOW + 10%


™4- ™~
8"

^


£

5 5.5 G 6.5 7 7.5 £
Log KOW











3

	 » 	 Phyto

— * — Amphipods
Mysids
Sculpin
^^ Smelt
..JP — Trout

— i — Ale wife


One of the key parameters is the elimination coefficient for each of the organisms. As explained
above, the k2 values can be estimated by the model, see (40). However, those estimates were found
to underestimate elimination, so simpler regression equations (41) and (42) were used to estimate
the k2 values, and those were entered manually in the toxicity parameter screens. In order to test the
sensitivity, k2 +10% values were used in the analyses (Figure 17). The sensitivity was small and
affected only the lower Kow congeners.
                                         3-26

-------
MODEL VALIDATION REPORTS
CHAPTER
          Figure 17.  Sensitivity of AQUATOX BAFs to a +10% change in K2
          values for all organisms.
K2 + 10%
0 1 g -, 	 	 .
0.05 -
£, 0.00 -
"> n ne _
"1 n m
CD
<* n 1^

-0.20 -

-0.25 -I
,-*,

.*f"J

_'---,__ r/l
\""~^y i
\ /
\ /
\ /



i i i i i








	 » 	 Phyto
—»— Amphipods
Mysids
Sculpin
^»^ Smelt
-^ Trout

— i — Alewife

5 5.5 6 6.5 7 7.5 8
Log KOW
Amphipods provide the primary link between the sediments and the higher trophic levels. They feed
on detritus,  converting some refractory detritus to labile detritus that can be assimilated or
decomposed more rapidly.  Therefore, the preference of amphipods for refractory detritus was
increased by 10%, from 0.05 to 0.055 and the corresponding preference for labile detritus (primarily
freshly sedimented algae) was decreased by 10% in a sensitivity analysis. There were seemingly
random fluctuations in the animal BAFs,  probably representing differences in the initial PCB
concentrations in the sediments (Figure 18).

          Figure 18.  Sensitivity of AQUATOX BAFs to a +10% change in
          amphipod preference for refractory detrital sediments .
AMPHIPOD PREFERENCE
n pn 	
Ocn
>i n An -
~ n on -
« u-^
t/5 0.00 -
n nn _
OAn


A »-A
^S^M^^-

5 5.5 6 6.5 7 7.5 8
Log KOW

= Phyto
* Amphipods
Mysids
Sculpin
	 * 	 Smelt
-•-Trout
• Alewife

In reviewing the literature on Lake Ontario, the observations of Landrum and Robbins (1990) on the
feeding rate ofDiporeia (the common amphipod in the Great Lakes) were found to be at variance
with the maximum consumption rate (CMax) used in AQUATOX.  Therefore, an analysis was
                                         3-27

-------
MODEL VALIDATION REPORTS
CHAPTER
performed in which CMax was changed from 1.3 g/g d to 0.288 g/g d. A systematic response was
found from one trophic level to the next (Figure 19), suggesting that indeed amphipods are
important in the transfer of contaminants, that the effects are magnified at the higher trophic levels
and that the CMax value should probably be changed in future implementations. Smelt are affected
the most, probably due to amphipod biomass dropping below the minimum biomass level (BMiri)
for consumption and uptake of PCBs by smelt.  In this and other analyses  of sensitivity to
bioenergetic parameters, interpretations of the effects on B AFs are difficult because those effects are
indirect.
          Figure 19.  Sensitivity of AQUATOX BAFs to a -77% change in
          amphipod maximum consumption rate.
1 An -,

1 "J!~l -
1 nn -
>i n nn -
.£ n fin -
w n An
s U.4U -
* mn
(/) 0.20 -
Onn -
n 9n -

OAn
£
Amphipod CMax


-»j|p::::::::::::_^jg^^^^^gte^3g|,||..|| 	 ^...m^mmm P ^


^-~
~~~~~*^^-^^-**^-*^-*~~ •+— *— —+~*

— e.
™,

5 5.5 6 6.5 7 7.5 8
Log KOW


—m 	 Phyto

— * — Amphipods
Mysids
Sculpin
—*— Smelt
	 » 	 Trout

— i — Ale wife


Because the trophic feeding relationships seem to be important in the transfer of PCBs in the Lake
Ontario food web, the effect of changing the sculpin BMin was investigated by changing the value
from 0.01 to 0.1 mg/L.  This has the effect of providing a refuge from predation for amphipods and,
to a lesser extent, zooplankton; mysids were found to respond the opposite to the fish. Because
sculpin compete to a certain extent with alewives, and both compete with and are fed on by smelt,
changing the feeding dynamics for sculpin affects the other fish species, although the sensitivities
are low (Figure 20).

In summary, changes in Kow and k2 values had little effect on the BAFs; and changes in feeding
preferences in amphipods had no systematic effect on BAFs, but changes in CMax and BMin values
were important in regulating the transfer of PCBs in the Lake Ontario ecosystem.
                                         3-28

-------
MODEL VALIDATION REPORTS
CHAPTER
          Figure 20.  Sensitivity of AQUATOX BAFs to a change in sculpin
          minimum biomass for feeding from 0.01 to 0.1 mg/L.
SCULPIN BMIN = 0.1
Q "3Q 	 .
n ic .
0~>ri .
£ n 15 -
2 n in -
M U-IIJ
oa n n£ -
(/3
Onn
-0.05 -
n m -



•^•i* 	 »-— — «Sipt-J«
'"X ^^ *-•»*»_
--- "iH
""*'*"*"»>»,«, ™^~— ~_l.— , jH
•^"™lK"B*::qfe,. _ ^"-»
*•$>»».... •.
**»-tei-.^ 	 „_
^"^
t * 	 t-»— -*•-*— ***-» — *-<+— »-*•

	 « 	 Phyto
^»— Amphipods
Mysids
Sculpin
^^ Smelt
^^ Trout
— ' — Alswifs

5 5.5 6 6.5 7 7.5 8
Log KOW
Uncertainty Analysis

Uncertainty analysis was conducted using the same parameter distributions as were used by Burkhard
(1998), in so far as appropriate. However, the contributions of individual parameters were not
investigated because of the computational load.  Rather, the aggregate uncertainties  due to all
selected input variables were analyzed simultaneously.  Burkhard (1998) used a Monte Carlo
procedure with random values taken from normal and lognormal distributions; the simulations were
run for 100,000 iterations.  Such an approach is neither desirable nor warranted for a complex model
such as AQUATOX.  Instead, Latin hypercube sampling was performed for  100 iterations; that
algorithm took a random sample from each of 100 segments of the normal  distributions for each
selected input parameter, ensuring that the distributions were well represented.  Each of the 100
simulations was run for a four-year period to obtain steady-state. Because of the model complexity
and the length of the simulation period, the uncertainty analysis took 40.5 hours to complete on a
Pentium  III 500 mHz machine.  The input  parameters are shown in Table 3, taken from the
AQUATOX Setup. Two parameters, LogKow and the sediment detritus-water partition coefficient,
were converted from lognormal to normal distributions to facilitate the analysis.
                                         3-29

-------
MODEL VALIDATION REPORTS
CHAPTER
Table 3. Parameters used in uncertainty analysis of AQUATOX; Parameter 1 is the mean and
Parameter 2 is the standard deviation.
Distribution Name
Octanol Water Partition Coeff (log)
Type
Normal
Sed.'Detritus-Water Partition Coeff. (rng.'l Normal
D invert: Max Consumption: (g ,' g d)
P invert: Max Consumption: (g ,' g d)
F fish: Max Consumption: (g / g d)
B fish: Max Consumption: (g / g dj
Sm g fish: Max Consumption: (g / g d)
Lg g fish: Max Consumption: (g / g d)
D invert: Respiration Rate: (L / d)
P invert: Respiration Rate: (L / d)
F fish: Respiration Rate: (L / d)
B fish: Respiration Rate: (L / d)
Sm g fish: Respiration Rate: (L / d)
Lg g fish: Respiration Rate: (L I d)
Normal
Normal
Normal
Normal
Normal
Normal
Normal
Normal
Normal
Normal
Normal
Normal
R detr sed(g/m2): Initial Condition (g/sg.i Normal
L detr sed(g>'rn2): Initial Condition (g/sq.i Normal
Temp: Multiply Loading by
B invert: Lipid Frac
P invert: Lipid Frac
F fish: Lipid Frac
B fish: Lipid Frac
Sm g fish: Lipid Frac
Lg g fish: Lipid Frac
Normal
Normal
Normal
Normal
Normal
Normal
Normal
Param. 1
6.S
79056941
1.3
0.085
0.299
0.65
0.3157
0.0188
0.02
0.0023
0.0031
0.0019
0.0033
0.001
600
160
1
0.03
0.05
0.07
0.08
0.04
0.11
Param. 2 Param. 3
0.2
11900000
0.182
0.0119
0.0343
0.1242
0.0311
0.0008
0.0012
0.0003
0.000285
3.78E-5
0.000186
0.00011
378
100.8
0.1
0.0015
0.0025
0.0035
0.004
0.002
O.OOSS
Param. 4 Used?
VES
VES
VES
VES
VES
VES
VES
VES
VES
VES
VES
VES
VES
VES
VES
VES
VES
VES
VES
VES
VES
VES
VES
The bioenergetic parameters have a direct effect on the ecosystem components, such as the biomass
of lake trout shown in Figure 22.  Given the number of parameters, the uncertainty exhibited is
surprisingly small.
                                        3-30

-------
MODEL VALIDATION REPORTS
                                                                            CHAPTER
Figure 22. Effect of uncertainty on biomass of lake trout over four years; the mean and
deterministic results are very close and are bounded by +/- 1 standard deviation; the top and
bottom curves represent the high and low values out of 100 simulations.
               |
                            . hKWirw ayi. wifess tWentus invKatsX
                                       ! Vww • ffimvr*
                                       LsVe Trout
                                                                                *  Metn-StOev
                                                                                0  DttBITfcbiL
The low uncertainty in the trout bioenergetics contributes to the low uncertainty in the trout BAFs
(Figure 23). Because the trout had no contaminant burden at the beginning of the simulation, it took
three or more years for steady state to be reached. The seasonal fluctuations that are evident in the
short-lived animals are barely discernible in the large, long-lived trout due to the slow dietary uptake
and even slower depuration.
                                           3-31

-------
MODEL VALIDATION REPORTS
CHAPTER
  Figure 23. Effect of uncertainty on the log BAFs for lake trout over a four-year period.
The 90th and 10th percentiles, representing the tails of the distributions, were computed from the
mean and standard deviations of the BAFs for the various simulated organisms. The ratios of these
percentiles were then calculated (Table 4).

Table 4.  Lipid-normalized, freely dissolved log bioaccumulation factors using uncertainties.
State variable
Diatoms
Mysids
Sculpin
Alewife
Smelt
Lake trout
Refr. sed. detritus
Labile susp. detritus
Mean
6.77
7.84
8.88
8.63
7.78
8.12
8.63
7.66
Standard Deviation
0.00004
0.36
0.29
0.31
0.36
0.31
0.23
0.28
Ratio 90th: 10th
1.00
1.13
1.09
1.10
1.12
1.10
1.07
1.10
                                         3-32

-------
MODEL VALIDATION REPORTS	CHAPTER 3

These low ratios are misleading because manually entered k2 values desensitize the model to
changes in weights and lipid fractions in organisms. Therefore, although the ratio of 1.10 for lake
trout is for all selected input parameters, it is not directly comparable with the values of 3.63 and
3.98 for all parameters obtained with the Gobas and Thomann models (Burkhard 1998).

                                      Conclusions

The subset of 16 PCB congeners from the 72 used by Burkhard (1998) provided an adequate basis
for evaluating the validity of AQUATOX in predicting bioaccumulation of PCBs in Lake Ontario.
Unlike the Gobas and Thoman models, AQUATOX had never been applied to such a large system,
but in general, AQUATOX gave acceptable results. It provided better fits to observed data for
phytoplankton and mysids than those provided by the Gobas (1993) and Thomann (1989) models
as implemented by Burkhard (1998), and equally acceptable results for smelt and lake trout when
compared to the Gobas model.

Several  modifications to the code were necessary to facilitate direct comparison to the Burkhard
(1998) study. Specifically, the freely dissolved contaminant concentration and lipid fractions in
organisms were held constant. A procedure for computing the sediment-water partition coefficient
was added, and capability for running in batch mode also was provided.  All these features were
made options in Version 1.68 and subsequent versions.

AQUATOX seems to have been successfully calibrated to represent the Lake Ontario ecosystem.
Most predicted bioaccumulation factors were reasonably close to observed values. Phytoplankton
BAFs were underestimated according to the observed data,  which may be biased by inclusion of
zooplankton; predictions were  noticeably better than  the Gobas  and Thomann models  as
implemented by Burkhard (1998).  Mysid BAFs were predicted very well, in contrast to the
underestimations in the Gobas and Thomann models, which may reflect the lack of herbivorous
zooplankton in those implementations. Amphipods, sculpin, and alewife BAFs were overestimated,
suggesting that the bioenergetics parameters may not be well calibrated; indeed, changing the CMax
for amphipods to  a value measured with the Lake Ontario species greatly improved the BAF
prediction.  The smelt and lake trout BAF predictions were very close to the observed and were
better than the predictions of the Thomann model, and equivalent to the Gobas model results.

AQUATOX has many nonlinear relationships in the ecosystem, fate, and effects portions of the
model. Therefore, it does not exhibit one-to-one sensitivities to input parameters as do the simpler
Gobas  and Thomann models.   Furthermore,  by manually entering estimated k2  values,  the
simulations were desensitized to lipid and weight factors. The algal BAFs are relatively sensitive
to changing  Kow values, although  only  one congener exceeded  a sensitivity of 1.0  (perfect
correspondence). BAFs exhibited low sensitivities to benthic feeding relationships.  The almost
negligible effect of changing preference of amphipods for labile organic  matter, which is mostly
sedimented phytoplankton, was in sharp contrast to the response of Thomann's model to a similar
preference for  phytoplankton (Burkhard 1998).   This emphasizes the unique  capability  of
AQUATOX in realistically representing ecosystemic relationships affecting contaminant fate. The
model properly represented the cascading effect that changing amphipod consumption has on BAFs
in the benthic food web.  A future task should be to fine-tune the amphipod and fish bioenergetic
parameters for Great Lakes species.

                                         3-33

-------
MODEL VALIDATION REPORTS	CHAPTER 3

AQUATOX exhibited a low degree of uncertainty, especially compared to the Gobas and Thomann
models (Burkhard 1998). However, this is probably in part an artifact of the constraints imposed by
specifying the k2 elimination values.

In conclusion, this study has demonstrated the validity and robustness of AQUATOX in estimating
bioaccumulation factors for PCBs in Lake Ontario.
                                     References

Abbott, J.D., S.W. Hinton, and D.L.  Borton. 1995. Pilot Scale Validation of the RIVER/FISH
      Bioaccumulation Modeling Program forNonpolarHydrophobic Organic Compounds Using
      the Model Compounds 2,3,7,8-TCDD and 2,3,7,8-TCDF. Environmental Toxicology and
      Chemistry, 14(11): 1999-2012.

Ambrose, R.B., Jr., T.A. Wool, J.L. Martin, J.P. Connolly, and R.W. Schanz. 1991. WASPS.x, A
      Hydrodynamic and Water Quality Model—Model Theory, User's Manual, and Programmer's
      Guide. U. S. Environmental Protection Agency, Environmental Research Laboratory, Athens,
      Georgia.

Arts, M.T., and B.C. Wainman, eds.  1999. Lipids in Freshwater Ecosystems. New York: Springer.

Bertelsen, S.L., A.D. Hoffman, C.A. Gallinat, C.M. Elonen, and J.W. Nichols. 1998. Evaluation of
      Log Kow and Tissue Lipid Content as Predictors of Chemical Partitioning to Fish Tissues.
      Environmental Toxicology and Chemistry., 17(8): 1447-1455.

Bjork, M., and M Gilek. 1999. Efficiencies of Poly chlorinated Biphenyl Assimilation from Water
      and Algal Food by the  Blue  Mussel (Mytilus edulis). Environmental Toxicology and
      Chemistry, 18(4):765-771.

Burkhard, L.P. 1998. Comparison of Two Models for Predicting Bioaccumulation of Hydrophobic
      Organic Chemicals in a Great Lakes Food Web. Environmental Toxicology and Chemistry,
      17(3):383-393.

Canada Centre for Inland Waters. 1979. Assessment of Water Quality Simulation Capability in Lake
      Ontario.  CCIW Scientific Series No. Ill, Burlington, Ontario.

Connolly,  J.P., and D. Glaser.  1998.  Use of Food Web Models to Evaluate  Bioaccumulation.
      National  Sediment Bioaccumulation  Conference Proceedings. U.S.  Environmental
      Protection Agency Office of Water EPA 823-R-98-002, p.4-5-4-17.

Connell, D.W.,  and D.W.  Hawker.  1988.   Use of Polynomial Expressions to Describe the
      Bioconcentraton of Hydrophobic Chemicals by Fish.  Ecotoxicology and Environmental
      Safety 16:242-257.

Cook, P.M., and L.P. Burkhard.  1998. Development of Bioaccumulation Factors for Protection of
      Fish and Wildlife in the Great Lakes. National Sediment Bioaccumulation Conference
                                         3-34

-------
MODEL VALIDATION REPORTS	CHAPTER 3

      Proceedings. U.S. Environmental Protection Agency Office of Water EPA 823-R-98-002,
      p. 3-19-3-27.

Cornelissen, G., P.C.M.  van Noort, and  H.A.J.  Govers.   1997.  Desorption Kinetics of
      Chlorobenzenes, Polycyclic Aromatic Hydrocarbons, and  Polychlorinated Biphenyls:
      Sediment Extraction with Tenax and Effects of Contact Time and Solute Hydrophobicity.
      Environmental Toxicology and Chemistry, 16(7):1351-1357.

Di Toro, D.M. 1985.   A Particle Interaction Model of Reversible Organic Chemical Sorption.
      Chemosphere, 14(10): 1503-1538.

Fagerstrom, T., and B. Asell.  1973. Methyl Mercury Accumulation in an Aquatic Food Chain, A
      Model and Some Implications for Research Planning. Ambio, 2(5): 164-171.

Fagerstrom,  T., R. Kurten ,and B. Asell.  1975. Statistical  Parameters as  Criteria in Model
      Evaluation: Kinetics of Mercury Accumulation in Pike Esox lucius. Oikos 26:109-116.

Flint, R.W. 1986.  Hypothesized Carbon Flow Through the Deep Water Lake Ontario Food Web.
      Jour. Great Lakes Res. 12:344-354.

Freidig,  A.P., E.A. Garicano, and F.J.M. Busser. 1998. Estimating Impact of Humic Acid on
      Bioavailability and Bioaccumulation of Hydrophobic Chemicals in Guppies Using Kinetic
      Solid-Phase Extraction. Environmental Toxicology and Chemistry, 17(6):998-1004.

Gilek, M., M. Bjork, D. Broman, N. Kautsky, and C. Naf. 1996. Enhanced Accumulation of PCB
      Congeners by Baltic Sea Blue Mussels, Mytilus edulis, with Increased Algae Enrichment.
      Environmental Toxicology and Chemistry, 15(9): 1597-1605.

Gobas, F.A.P.C. 1993. A Model  for Predicting the Bioaccumulation Hydrophobic  Organic
      Chemicals in Aquatic Food-webs: Application to Lake Ontario. Ecological Modelling, 69:1-
      17.

Gobas, F.A.P.C., E.J. McNeil, L. Lovett-Doust, and G.D. Haffner. 1991. Bioconcentration of
      Chlorinated Aromatic Hydrocarbons in Aquatic Macrophytes (Myriophyllum spicatum).
      Environmental Science & Technology, 25:924-929.

Gobas, F.A.P.C.,  Xin Zhang,  and Ralph Wells.  1993. Gastrointestinal Magnification:  The
      Mechanism of Biomagnification and Food Chain Accumulation of Organic Chemicals.
      Environmental Science & Technology, 27:2855-2863.

Gobas, F.A.P.C., M.N. Z-Graggen, X. Zhang. 1995. Time response  of the Lake Ontario Ecosystem
      to Virtual Elimination of PCBs. Environmental Science & Technology, 29(8):203 8-2046.

Hawker,  D.W.  and D.W. Connell.   1985.  Prediction of Bioconcentration Factors Under
      Non-Equilibrium Conditions.  Chemosphere 14(11/12):1835-1843.
                                         3-35

-------
MODEL VALIDATION REPORTS	CHAPTER 3

Hill, W.R., and Napolitano, G.E. 1997. PCB Congener Accumulation by Periphyton, Herbivores,
       and Omnivores. Archives Environmental Contamination Toxicology, 32:449-455.

Koelmans, A.A., and  E.H.W. Heugens. 1998. Binding  Constants  of Chlorobenzenes  and
       Polychlorobiphenyls for Algal Exudates. Water Science Technology., 37(3):67-73.

Koelmans, A.A., S.F.M. Anzion,  and L. Lijklema.  1995. Dynamics of Organic Micropollutant
       Biosorpti on to Cyanobacteria and Detritus. Environmental Science & Technology, 29(4):933-
       940.
Karickhoff, S.W., and K.R. Morris.  1985.  Sorption Dynamics of Hydrophobic Pollutants in
       Sediment Suspensions. Environmental Toxicology and Chemistry, 4:469-479.

Landrum,  P.P.  1998.  Kinetic Models for Assessing Bioaccumulation. National Sediment
       Bioaccumulation Conference Proceedings. U.S. Environmental Protection Agency Office
       of Water EPA 823-R-98-002, p.  1-47-1-50.

Landrum, P.P., and J.A. Robbins.  1990. Bioavailability of Sediment-Associated Conyaminants to
       Benthic Invertebrates.  In R. Baudo, J.P. Giesy, and H. Muhtau, eds., Sediments: Chemistry
       and Toxicology ofln-Place Pollutants, Chelsea MI: Lewins Pub., pp. 237-263.

Leung, D.K. 1978. Modeling the Bioaccumulation of Pesticides in Fish. Report N.  5, Center for
       Ecological Modeling, Rensselaer Polytechnic Institute, Troy, N. Y.

Lyman, W. J., W.F. Reehl, and D.H. Rosenblatt. 1982. Handbook of Chemical Property Estimation
       Methods.  McGraw-Hill, New York.

Maaret, K., K, Leif, and H. Bjarne.  1992. Studies  on the Partition Behavior of Three Organic
       Hydrophobic Pollutants in  Natural Humic Water. Chemosphere, 24(7):919-925.

Macek, K.J., M.E. Barrows, R.F. Frasny,  and B.H. Sleight III. 1977. Bioconcentration of  14C-
       Pesticides by Bluegill Sunfish During Continuous Aqueous Exposure. In Structure-Activity
       Correlations in Studies of Toxicity and Bioconcentration with Aquatic Organisms, G.D.
       Veith and D. Konasewick,  eds.

McCarty, L.S., G.W. Ozburn, A.D. Smith, and D.G. Dixon.  1992. Toxicokinetic  Modeling of
       Mixtures of Organic Chemicals. Environmental Toxicology and Chemistry, 11:1037-1047.

McKim, J.M., P. Schmeider,  and  G.  Veith.  1985.  Absorption Dynamics of Organic Chemical
       Transport Across Trout Gills as Related to Octanol-Water Partition Coefficient. Toxicology
       and Applied Pharmacology, 77:1-10.

Nichols, J.W., K.M. Jensen, I.E. Tietge, and R.D. Johnson.   1998.   Physiologically Based
       Toxicokinetic Model for Maternal Transfer of 2,3,7,8-Tetrachlorodibenzo-p-Dioxin in Brook
       Trout (Salvelinusfontinalis). Environmental Toxicology and Chemistry, 17:2422-2434.

O'Connor, D.J., and J.P. Connolly. 1980. The Effect of Concentration of Adsorbing Solids on the
       Partition Coefficient. Water Research, 14:1517-1523.
                                         3-36

-------
MODEL VALIDATION REPORTS                                         CHAPTER
Oliver, E.G., and AJ.  Niimi.   1988. Trophodynamic Analysis of Polychlorinated  Biphenyl
       Congeners and Other Chlorinated Hydrocarbons in the Lake Ontario Ecosystem. Environ.
       Sci. Technol.,22(4):388-397.

Park, R.A., C.I. Connolly, J.R. Albanese, L.S. Clesceri, G.W. Heitzman, H.H. Herbrandson, B.H.
       Indyke, J.R. Loehe, S. Ross, D.D. Sharma, and W.W. Shuster. 1982. Modeling the Fate of
       Toxic Organic Materials in Aquatic Environments. U.S. Environmental Protection Agency
       Rept. EPA-6OO/S3-82-028, Athens, Georgia.

Robertson, A., and D. Scavia.  1979. The Examination of Ecosystem Propertiesof Lake Onatrio
       Through the Use of an Ecological Model. In D. Scavia and A. Robertson, eds., Perspectives
       on Lake Ecosystem Modeling, Ann Arbor MI: AnnArbor Science Pub., pp. 281-292.

Scavia, D. 1980.  An Ecological Model of Lake Ontario. Ecological Modelling 8:49-78.

Schwarzenbach, R.P., P.M. Gschwend, and D.M. Imboden.   1993.  Environmental Organic
       Chemistry. Wiley and Sons, Inc., New York.

Sijm, D.T.H.M., K.W. Broersen, D.F de Roode, and P. Mayer. 1998.  Bioconcentration Kinetics of
       Hydrophobic Chemicals in Different Densities of Chlorella Opyrenoidosa. Environmental
       Toxicology and Chemistry 17:9:1695-1704.

Skoglund, R.S., K. Stange, and D.L. Swackhamer.   1996.  A Kinetics Model for Predicting the
       Accumulation of PCBs  in Phytoplankton.   Environmental Science  and  Technology
       30:7:2113-2120.

Smejtek,  P., and S. Wang.  1993. Distribution of Hydrophobic lonizable Xenobiotics Between
       Water and Lipid Membranes: Pentachlorophenol and Pentachlorophenate. A Comparison
       with Octanol-Water Partition. Archives of Environmental Contamination and Toxicology,
       25(3):394.

Southworth,  G.R., JJ. Beauchamp, and P.K. Schmieder.   1978. Bioaccumulation Potential of
       Polycyclic Aromatic Hydrocarbons in Daphniapulex. Water Res., 12:973-977.

Spacie, A., and J.L. Hamelink.  1982. Alternative Models for Describing the Bioconcentration of
       Organics in Fish. Environmental Toxicology and Chemistry, 1:309-320.

Stange, K., and D.L.  Swackhamer.  1994.  Factors Affecting Phytoplankton Species-Specific
       Differences in Accumulation of 40  Polychlorinated Biphenyls (PCBs).  Environmental
       Toxicology and Chemistry, 13(11): 1849-1860.

Swackhamer, D.L., and R.S. Skoglund. 1993. Bioaccumulation of PCBs by Algae: Kinetics versus
       Equilibrium. Environmental Toxicology & Chemistry, 12:831-838.

Thomann, R.V. 1989. Bioaccumulation Model of Organic Chemical Distribution in Aquatic Food
       Chains. Environmental Science & Technology, 23:699-707.
                                         3-37

-------
MODEL VALIDATION REPORTS	CHAPTER 3

Thomann, R.V., J.P. Connolly, and T. Parkerton. 1992. Chapter 7, Modeling Accumulation of
      Organic Chemicals in Aquatic Food Chains. In Gobas, F.A.P.C., and J.A. McCorquodale,
      eds., Chemical Dynamics in Fresh Water Ecosystems. Pp. 153-186.

Thomann, R.V., and J.A. Mueller. 1987.Principles of'Surface Water Quality Modeling and Control,
      Harper Collins: new York N.Y., 644 pp.

US EPA.  1995.  Great Lakes Water Quality Initiative Technical Support Document for  the
      Procedure to Determine Bioaccumulation Factors. EPA-820-B-95-005, U.S. Environmental
      Protection Agency, Washington, D.C.

Wang, X., Y. Ma, W. Yu, and HJ. Geyer.  1997. Two-compartment Thermodynamic Model for
      Bioconcentration of Hydrophobic  Organic Chemicals by Algae. Chemosphee, 35(8): 1781-
      1797.

Weininger, D. 1978. Accumulation of PCBs by Lake Trout in Lake Michigan. Ph.D. Dissertation,
      University of Wisconsin, Madison, 232 pp.
                                        3-38

-------