United States Office of Water EPA-823-R-09-003
Environmental Protection 4305 February 2009
Agency
&EPA AQUATOX Technical Note 1
A Calibrated Parameter Set for
Simulation of Algae in Shallow Rivers
-------
AQUATOX Technical Note 1
A Calibrated Parameter Set for Simulation of Algae in Shallow Rivers
Richard A. Park1, James N. Carleton2, Jonathan S. Clough3, and Marjorie C. Wellman2
Modeling, Diamondhead MS 39525
2U.S. Environmental Protection Agency, Washington, DC 20460
3Warren Pinnacle Consulting, Inc., Warren VT 05674
January, 2009
INTRODUCTION
The AQUATOX aquatic ecosystem model contains constructs that represent responses of
biological entities to numerous environmental factors, and simulations can include many
taxa. Because large numbers of parameters must be chosen before simulations can take
place, calibration of the model can be a daunting task, especially for novice users. The
existence of a parameter set that can be used "off the shelf potentially represents a great
time and effort saver. This note describes development of one such parameter set, for
photosynthetic algae in small to medium-sized shallow rivers in temperate North
America.
SUMMARY OF APPROACH
AQUATOX was calibrated against data from three shallow rivers in Minnesota: the Crow
Wing, Rum, and Blue Earth Rivers (Figure 1). These rivers are, respectively, nutrient-
poor clear-water, moderately nutrient-enriched clear-water, and nutrient-enriched turbid.
Simulations were run with a shared parameter set using AQUATOX Release 3, to obtain
acceptable fits to observed data across all three sites. Goodness-of-fit to observed data
was evaluated visually, and with relative bias and F tests. The resulting parameter set
was verified by simulating a site on the Cahaba River, Alabama. Further verification was
obtained by applying the original parameter set, without change, to nutrient-poor and
nutrient-enriched, clear-water and turbid sites on the Lower Boise River, Idaho.
BACKGROUND
AQUATOX has been designed as a general ecological risk assessment model capable of
representing the combined environmental fate and effects of conventional pollutants (i.e.
nutrients, sediments) and toxic chemicals in aquatic ecosystems (Park etal. 2008). The
model explicitly simulates competition, predation, and other kinds of interactions among
user-definable groups in several trophic levels, including attached and planktonic algae,
submerged aquatic vegetation, invertebrates of various guilds, and multiple size or age
classes of several species offish. Equations that represent these processes involve the
biological components themselves (state variables), environmental drivers (e.g. flow,
temperature), and chemical and biological factors that affect the state variables.
-------
A challenge for users in setting up new simulations is assigning suitable values for
parameters that govern state variables. Representative values may be difficult to find in
the literature, and users may have difficulty knowing which parameters to modify during
calibration, as well as by how much to vary them and in what order. To help address this
concern AQUATOX comes bundled with libraries of parameter sets for many species,
and these libraries are constantly being expanded and refined. In order to simulate a
particular water body, however, modification of various parameters may still be
necessary. Parameters may represent properties that are affected by ambient conditions
or vary with local ecotypes. As an example, maximum photosynthetic rate (Pmax) would
seem to be a basic property of an algal taxon, but at any given time the photosynthetic
rate is affected by temperature, nutrients, light, etc. A well-chosen value for Pmax would
be expected to be applicable to most sites where a specific algal group is present. This is
the goal of a "global" parameter, and its existence would imply that new simulations
could be set up without the need for significant re-calibration.
METHODS
To develop the initial calibration we worked with data provided by the Minnesota
Pollution Control Agency (MPCA), which had found "significant and predictable" linear
relationships between nutrients, algae, and biochemical oxygen demand in five medium
to large rivers, in samples collected during 1999 and 2000 (Heiskary and Markus 2001,
Heiskary and Markus 2003). Trophic conditions in these rivers span a nutrient gradient,
from a relatively unenriched, predominantly forested watershed in the "Northern Lakes
and Forests" ecoregion (Omernik 1987), to a relatively high-nutrient, predominantly row
crop agricultural watershed in the "Western Corn Belt Plains" ecoregion. We first
calibrated AQUATOX against data from each of these rivers simultaneously, i.e. using a
single parameter set. We then tested the robustness of this parameter set by applying it to
the Cahaba River in Alabama and the Lower Boise River in Idaho. Table 1 presents
observed mean annual nutrient concentrations, temperatures and discharge in each of
these rivers at the locations of interest.
Table 1. Mean annual conditions at the calibration and verification sites.
Site
MN
Crow Wing River
Rum River
Blue Earth River
Lower Boise River, ID
Eckert
Middleton
Parma
Cahaba River, AL
Total N Total P
(mg/L) (mg/L)
0.76 0.033
1.18 0.115
6.80 0.204
TSS Temp
(mg/L) (deg. C)
Discharge
(m3/d)
2.37
13.99
82.48
9.3
12.0
10.6
1.52E+06
1.02E+06
1.49E+06
0.128
1.763
2.643
1.131
0.060
0.325
0.395
0.198
5.15
13.89
47.81
15.77
10.5
11.7
11.9
18.0
3.90E+06
1.07E+06
3.80E+06
6.63E+05
-------
INITIAL CALIBRATION
We chose to focus our modeling efforts on specific sampled reaches within three of the
five rivers monitored by MFC A. All three rivers are shallow (mean depth 1 m or less)
and support diverse periphyton communities, which appear to vary in composition among
rivers according to their position along enrichment and turbidity gradients (Heiskary and
Markus 2001, Heiskary and Markus 2003). Based on MFC A data the Crow Wing River,
which drains a heavily forested watershed, has relatively low concentrations of nutrients,
and low water-column turbidity. The Rum River, draining a watershed with numerous
dairy farms, has moderate nutrient concentrations and low turbidity. The Blue Earth
River, whose watershed is composed largely of corn and soybean acreage with extensive
tile drainage, has high nutrient concentrations and periodically high turbidity. The
phytoplankton composition varies from river to river in a predictable fashion, apparently
correlated with nutrient conditions. For example, in sampling conducted in 2000,
cyanobacteria constituted only a tiny fraction of the sestonic algal biomass in the Crow
Wing River but an important fraction in the Rum River, and were the dominant type in
the Blue Earth River.
MPCA sampling data for nutrients, BODs, and total suspended solids (TSS) were
available from two to five separate locations in each river on six to eight separate
occasions between June and September of each of the years 1999 and 2000. These data
were used to provide influent concentrations that drove the reach simulations. Biotic
state variables were chosen to represent nutrient-poor, clear-water conditions
characteristic of forested regions as well as nutrient-enriched, sporadically turbid
conditions characteristic of agricultural regions. Because the objective was to obtain a set
of state variables that would span broad conditions, the number of state variables was
larger than might be necessary if a single river with static conditions were being
simulated. By seeding the simulations with a range of taxa likely to thrive under a
variety of conditions, the algal community responses to widely differing and changing
conditions could be simulated using a single parameter set.
Simulated periphyton and phytoplankton communities consisted of broad taxonomic
groups of green algae and cyanobacteria ("blue-greens"), as well as "low-nutrient" and
"high-nutrient" adapted diatoms (Table 2). The simulated periphyton assemblage also
specifically included the diatom Nitzschia and the filamentous green Cladophora, and the
phytoplankton assemblage included the diatom Navicula, and Cryptomonas, for a mixture
of organisms intended to represent potential algal responses throughout the range of
simulated environmental conditions. Periphyton and phytoplankton groups were linked
to each other in the model through sloughing and sinking, such that detached green algae
become part of the sestonic green algal biomass, for example. Invertebrates were
represented by broad guilds, with representative genera for shredders, suspension feeders,
and grazers among the zoobenthos and zooplankton, as well as clams and snails. Fish
species represented small and large forage, bottom, and game fish that are characteristic
of different environmental conditions. All of the fauna were parameterized using the
default parameter sets provided with AQUATOX version 3.
-------
Minneapolis
St. Paul*!/-
r~^
I ! ' ! I ! ! ! I
0 25 50 100 Miles
Figure 1. Locations of simulated Minnesota river reaches, with watersheds (HUC8), and
aggregate level 3 nutrient ecoregions indicated.
-------
Table 2. Select algal parameters employed in AQUATOX simulations.
Topt Tmax Tresp LightSat Pmax Lightex P half-sat N half-sat C half-sat ExpMoCo
20
20
25
30
30
25
15
20
26
27
8
39
35
42
42
50
39
39
35
42
50
30
2
1.8
2
2
2
2
2
2
2
2
2
64
22.5
70(110)*
135
45
82.5
64
18
50(110)
60
80(110)
0.65
2.3
1.7 (2)
0.7(1.4)
1.4
0.5(1.75)
0.7
1.87
1.5(1.65)
2.2
3
0.03
0.03
0.03
0.05
0.03
0.03
0.14
0.14
0.24
0.09
0.144
0.006
0.055
0.1
0.04
0.1
0.095
0.006
0.055
0.1
0.03
0.076
0.07
0.2
0.8
0.1
0.8
0.4
0.0154
0.117
0.8
0.4
0.03
0.054
0.054
0.054
0.054
0.024
0.054
0.054
0.054
0.054
0.024
0.054
0.01
0.01
0.01
0.05
0.01
0.05
0.05
0.05
0.04
0.12
0.04
0.001 (0.003)
0.004
0.004 (0.003)
0.004
0.004
0.001 (0.003)
NA
NA
NA
NA
NA
75 (90)
60
20 (60)
90
90
50 (90)
NA
NA
NA
NA
NA
Periphyton
Low-nutrient diatoms
High-nutrient diatoms
Greens
Cladophora
Blue-Greens
Nitzschia
Phytoplankton
Low-nutrient diatoms
High-nutrient diatoms
Greens
Blue-Greens
Cryptomonas
Key:
Topt = optimal temperature (deg C)
Tmax = maximum temperature (deg C)
Tresp = temperature response slope
LightSat = saturating light (Ly/day)
Pmax = maximum photosynthetic rate (I/day)
Lightex = light extinction coefficient l/m-g/m3
P half-sat = phosphorus half-saturation constant (mg/L), Michaelis-Menten kinetics
N half-sat = nitrogen half-saturation constant (mg/L), Michaelis-Menten kinetics
C half-sat = inorganic carbon half-saturation constant (mg/L), Michaelis-Menten kinetics
ExpMoCo = exponential mortality coefficient (g/g-day)
Fcrit = critical force for periphyton scour (Newtons)
%sl = percent periphyton lost in slough event (%)
Fcrit
* Parameters in boldface were modified during calibration. Parameters enclosed in parentheses are the corresponding defaults. All other numbers are also model defaults.
-------
Flow Data and Simulations
Biological responses in stream ecosystems can be highly sensitive to flow conditions. In
the ideal case a modeler has access to reach-specific flow data, but if not flows may be
estimated using other methods. For this exercise, USGS flow-gauging stations were
conveniently located at the Rum-18 (Rum River, mile 18) and CWR-72 (Crow Wing
River, mile 72) sampling sites (gauges 05286000 and 05244000 respectively). Recorded
mean daily discharges at these locations supplied flow values for simulating these two
sites. Unfortunately a gauging station was not present at BE-54 (Blue Earth River, mile
54); the nearest station on the Blue Earth River was 42 miles downstream, at gauge
05320000 near Rapidan Dam. Therefore, to provide flow estimates for the BE-54
AQUATOX simulation, a previously-derived Hydrologic Simulation Program-Fortran
(HSPF) simulation of the river and its watershed at this location was employed (Donigian
et al. 2005).
Nutrient and Suspended Solids Estimation
Nutrients support the bases of aquatic food chains, and nutrient state variables are
required for AQUATOX to run. The available TP, total Kjeldahl nitrogen (TKN), nitrate
(NCV), and BOD5 data collected by MFC A supplied the concentrations of these variables
that were used to drive all three river simulations. Because AQUATOX is designed to
process daily values of all inputs, it automatically applied linear interpolation between
sampled dates to estimate concentrations of these constituents for the intervening days.
AQUATOX applies this method for all input data where daily values are not provided by
the user.
Total Suspended Solids (TSS) is a measure of suspended matter in the water, which can
have significant effects on light transmission and consequently photosynthesis.
Possession of daily site specific monitoring data would have been ideal, but was not the
case. However, the available TSS observations were found to be correlated with flow in
both the Blue Earth and Rum Rivers, such that daily estimates of TSS could be generated
for use in the model. Because there was no gauge at BE-54, the relationship for BE-54
was based on a linear regression of TSS against In-transformed flow at downstream
gauge 05320000. The resulting time series provided a close match to most of the
observed data (Figure 2). For the Rum River, a linear regression of TSS against flow was
used, which provided reasonably good approximations of the observed data at this site as
well (Figure 3). In contrast to the Rum and Blue Earth Rivers, the Crow Wing River
exhibited no correlation between TSS and flow (this may have been because the river
drains glacial outwash sands, only about 15% of the watershed for the reach is composed
of agricultural land, and an estimated 57% is forest). In order to drive all three
simulations with analogous forcing functions, an HSPF model of daily TSS in the Crow
Wing reach (Donigian et al. 2005) was employed. It was felt that a well-calibrated HSPF
simulation would provide a better representation of temporal trends in TSS than linear
interpolation between the sampled points would have, despite discrepancies between
some of the simulated and measured concentrations (Figure 4).
-------
a)
b)
350
300 -
250 -
150 -
100 -
50 -
0
R2=0.45
0 2000000 4000000 6000000 8000000 10000000 12000000
Daily Mean Flow (m /d)
05/99
08/99
12/99
04/00
08/00
12/00
Figure 2. TSS at Blue Earth River 54: a) regression against In-transformed daily flow at
gauge 05320000; b) resulting simulated daily time series (line), and observed values
(symbols).
-------
a)
b)
25
20 -
15 -
10 -
140
H
120
100
80
60
40
20 -
R2=0
400000 800000 1200000 1600000 2000000
Daily Mean Flow (m /d)
0
01/99 05/99 08/99 12/99 04/00 08/00 12/00
Figure 3. TSS at Rum River 18: a) linear regression against daily flow at gauge
05286000; b) resulting simulated daily time series (line), and observed values (symbols).
-------
a) 9T
8 -
7 -
,-x 6 -
"M 5 -
GO 4 -
GO
H 3-
2 -
1 -
0 -
500(
•
•
R2=0.08
• ^ —
— " » »
** » • ^
)00 900000 1300000 1700000
Daily Mean Flow (m /d)
b>
20 -
2s 15 -
00
GO
H 10 -
5 -
0 -
01
j
1 1 » " ^^^^^^
799 05/99 08/99 12/99 04/00
•
2100000
•
mm
I
\i
08/00
2500000
„
12/00
Figure 4. TSS at Crow Wing River 72: a) plot of TSS against daily flow at gauge
05244000; b) HSPF-simulated daily time series (line) and observed values (symbols).
-------
Calibration Process
Calibration of AQUATOX for the Minnesota Rivers used observed sestonic and
periphytic chlorophyll a as the primary target against which model output was optimized
(via parameter modifications indicated in Table 2). Because there were only five to eight
sestonic and one benthic chlorophyll a observations at each location in each of the two
target years, calibration adequacy was evaluated subjectively, based on generally
expected behavior (e.g. blooms occurring during summer) and approximate concordance
with observed values (in terms of both magnitude and timing), as determined through
graphical comparisons of model output and data (Figure 5).
Evaluation of Calibration
We also employed quantitative measures to evaluate the adequacy of the calibration and
model performance. Relative bias is a robust measure of how well central tendencies of
predicted and observed results correspond; a value of zero indicates that the means are
the same (Bartell etal. 1992):
rB = (Pred Bar - Obs Bar)/Sobs
or:
rB = relative bias (standard deviation units);
Pred Bar = mean predicted value;
Obs Bar = mean observed value; and
Sobs = standard deviation of observations.
The F test is the ratio between the variance of the model output and the variance of the
data. A value of unity indicates that the variances are the same:
F = Var Pred/Var Obs
where:
Var Pred = variance of predictions;
Var Obs = variance of observations.
Very large F values indicate that the predictions are imprecise (Bartell et al. 1992).
Large F values also may indicate that the model is predicting greater fluctuations than can
be supported by sparse data. Small F values may indicate highly variable or uncertain
observed data. Assuming normal distributions, the probability that the observed and
predicted distributions are the same can be evaluated (Figure 6). Putting the two tests
together, if a comparison has rB = 0 and F = 1, then the predicted and observed results
are identical.
10
-------
400
350 -
300 -
250
00
50-
=50
01/99 05/99 08/99 12/99 04/00 08/00 12/00
01/99 05/99 08/99 12/99 04/00 08/00 12/00
c)
30
25 -
20 -
i,
5 -
no -
5 -
0
01/99 05/99 08/99 12/99 04/00 08/00 12/00
Figure 5. Observed (symbols) and calibrated AQUATOX simulations (lines) of sestonic
chlorophyll a in three Minnesota rivers: a) Blue Earth at mile 54, b) Rum at mile 18, c)
Crow Wing at mile 72. Note the order-of-magnitude range in scale among the figures.
11
-------
O.2
-202
Relative Bias
Figure 6. Overlap between model and data distributions based on relative bias and ratio
of variances, F; 1 = Blue Earth River, 2 = Crow Wing River, 3 = Rum River. Isopleths
indicate the probability that the predicted and observed distributions are the same,
assuming normality.
Application of these statistics to the three Minnesota simulations for the periods covered
by the observed data indicates that in two simulations there is good overlap between the
predicted and observed distributions, and in one simulation (BE-54) the model predicts
greater variance (Figure 6). The central tendencies are similar between predicted and
observed distributions for all three sites, as shown by the relative bias. Despite the
fluctuations in predicted chlorophyll a, the predicted and observed variances are similar
for the CWR-72 and Rum-18 simulations. Predicted periphyton sloughing events played
a major role in determining the timing of chlorophyll a peaks in both simulations. The
variance in predicted values is high in the BE-54 simulation, in which summer peak
concentrations in 1999 appear to be overestimated by a factor of about two. The reason
for this overestimation is not known, but may reflect uncertainties inherent in the HSPF-
simulated flow and TSS values, the sparseness of water chemistry sampling data, and/or
other limitations. Using the procedure described above, the probability that the BE-54
predictions and observations have the same distribution was found to be greater than 0.8.
For the purpose of this analysis we therefore judged the calibration to be adequate.
12
-------
Sensitivity Analysis
Kpost-hoc nominal-range sensitivity analysis of the BE-54 simulation identified the most
sensitive cyanobacterial parameters (Figure 7). Two of these parameters, 'Saturating
Light' and 'Maximum Photosynthetic Rate', were modified as part of the calibration
process. The phytoplankton and periphyton parameters were linked in the analysis—that
is, a 10% change in a parameter for the phytoplankton was also a 10% change in the same
parameter for periphyton. When interpreting Figure 7, the vertical line at the middle of
the 'tornado' diagram represents the deterministic model result. Red lines represent
model results when the given parameter is reduced by 10%, while blue lines represent a
positive 10% change in the parameter. The sensitivity statistic represents the average
absolute percent change in model results divided by the percent change used to test model
parameters. For example, if a 10% change in the parameter resulted in a 10% change in
model results (in either the positive or negative direction), the sensitivity would be
calculated as 100%. Similar to the findings of other investigators (Sourisseau et a/.,
2008; Rashleigh et a/., 2009), the results of our analysis indicate that water column
chlorophyll a is most sensitive to the cyanobacterial parameters 'Optimal Temperature'
and 'Maximum Photosynthetic Rate'.
Sensitivity of Sestonic Chlorophyll (ug/L) to 10% change in tested parameters
1047.3% - Phyt, Blue-Gre: Optimal Temperature (deg. C) * Linked *
859% - Phyt, Blue-Gre: Max Photosynthetic Rate (1/d) * Linked '
374% - Phyt, Blue-Gre: Temp Response Slope * Linked *
314% - Phyt, Blue-Gre: Saturating Light (Ly/d) * Linked '
126% - Phyt, Blue-Gre: Exponential Mort. Coefficient: (max / d) * Linked *
86.9% - Phyt, Blue-Gre: Inorg. C Half-saturation (mg/L) * Linked *
75.1% - Phyt, Blue-Gre: Maximum Temperature (deg. C) * Linked *
30.6% - Phyt, Blue-Gre: P Half-saturation (ng/L) * Linked "
29.2% - Phyt, Blue-Gre: Sedimentation Rate (1/d) * Linked "
26.8% - Phyt, Blue-Gre: Respiration Coefficient (1/d) * Linked '
20 40 60
Sestonic Chlorophyll (ug/L)
80
Figure 7. Tornado diagram showing relative sensitivity of sestonic chlorophyll a
predictions to cyanobacteria parameters in simulations of the Blue Earth River. Red lines
represent model results with a negative change in a parameter, and dark blue lines
indicate model results with an increase in the parameter. The sensitivity statistic
represents the average absolute percent change in model results divided by the percent
change used to test model parameters.
13
-------
PARAMETER SET VERIFICATION
The Cahaba River, AL
As a limited verification, the calibrated model was applied to a site on the Cahaba River
south of Birmingham, Alabama, with modifications made to two parameters. The Crow
Wing and Rum Rivers have cobbles and boulders and are more sensitive to high current
velocities than the bedrock outcrops in the Cahaba River: not only is the bedrock
hydrodynamically stable, it also provides abundant crevices and lee sides that serve as
protected refuges for periphyton. For these reasons greater water velocity should be
required in order to initiate periphyton scour in the Cahaba River than in the Crow Wing
and Rum Rivers. With this rationale in mind, the critical force parameter for scour of
periphyton (Fcrit) was increased by about two-fold in the Cahaba River simulation to
match model predictions against data. In locations with climates as different as
Minnesota and Alabama one would expect different local ecotypes in resident algal
species, with differing adaptations to temperature. With this rationale in mind, the
optimum temperature values (Topt) for green algae and cyanobacteria were also
increased, by SEC to 31EC and 32EC respectively. The resulting fit to observed data
(Figure 8) was better than that obtained in the prior site-specific calibration (Park et al.
2002).
20
0
1/1/01
2/5/02
8/24/02
Figure 8. Observed (symbols with 1 standard deviation) and predicted (line) benthic
chlorophyll a in the Cahaba River, Alabama.
The Lower Boise River, ID
As a further parameter set verification, the calibrated algal model was also applied to
three dissimilar sites (Table 1) on the Lower Boise River, Idaho, without modification
from the Minnesota calibration. The three sites cover a broad range of nutrient and
turbidity conditions over a distance of 90 km (Figure 9). Eckert is a low-nutrient, clear-
water site upstream of Boise; Middleton receives wastewater treatment effluent and is a
14
-------
nutrient-enriched, clear-water site; and Parma is a nutrient-enriched, turbid site impacted
by irrigation return flow from agricultural areas. Although the model overestimates
periphyton at the Eckert site, the fit of this initial application (Figure 10) provides a
promising starting point for further river-specific calibration.
Figure 9. USGS gauging stations (triangles) and biological sampling stations (+s) on the
Lower Boise River, Idaho, with the three simulated sites indicated.
15
-------
b)
c)
200
1/1/98
250
200
1/1/98
35796
5/16/99
9/27/00
5/16/99
9/27/00
Figure 10. Predicted (line) and observed (symbols) benthic chlorophyll a (a) at Eckert
Road, (b) near Middleton, (c) near Parma, Lower Boise River, Idaho.
16
-------
CONCLUSIONS
Based upon evaluation of the simulations described in this note, the algal parameter set
originally developed for Minnesota rivers (Table 2) appears to be fairly robust across
wide nutrient and turbidity gradients in similarly-sized rivers, even in different regions of
the country. These values may represent a global parameter set for algae in medium-
sized rivers, and are suggested as a starting point for calibration of AQUATOX when
simulating such systems. Future model comparisons against data from additional rivers
may provide opportunities for further verification or refinement of these values.
Sensitivity analysis produced two parameters relating to photosynthesis and temperature
optima, to which special attention should be paid when further calibration is necessary.
ACKNOWLEDGMENTS
We are grateful to Tony Donigian and Jason Love for providing simulated influent
loadings for the Minnesota rivers, and we are indebted to Steve Heiskary for providing us
with MPCA's monitoring data, and for the logistical assistance he graciously provided
during our site visits. We are grateful to Don Blancher, Susan Sklenar, and Lynn Wood
for data and assistance on the Cahaba River; and to Tom Dupuis, Ben Nydegger, Kate
Harris and Robbin Finch for data and assistance on the Boise River. We also thank
Brenda Rashleigh for her review and useful suggestions. This work was conducted in
part with Federal funds from the U.S. Environmental Protection Agency, Office of
Science and Technology under contracts to AQUA TERRA Consultants and CH2M Hill.
Disclaimer
This document describes the development of a set of calibration parameters for the
AQUATOX model, for the purpose of simulating algal growth in rivers. Anticipated
users of this document include persons who are interested in using the model for this
purpose, including but not limited to researchers and regulators. The model described in
this document is not required, and the document does not change any legal requirements
or impose legally binding requirements on EPA, states, tribes or the regulated
community. 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.
17
-------
REFERENCES
Bartell, S. M., R. H. Gardner, and R. V. O'Neill, 1992. Ecological Risk Estimation. Lewis
Publishers, Boca Raton, Florida.
Donigian, A. S., J. T. Love, J. S. Clough, R. A. Park, J. N. Carleton, P. A. Cocca, and J.
C. Imhoff, 2005. Nutrient Criteria Development with a Linked Modeling System:
Watershed and Ecological Model Application and Linkage, in 2005 TMDL
Conference. Water Environment Federation, Philadelphia PA.
Heiskary, S., and H. Markus, 2001. Establishing relationships among nutrient
concentrations, phytoplankton abundance, and biochemical oxygen demand in
Minnesota, USA, rivers. Journal of Lake and Reservoir Management 17:251-262.
Heiskary, S., and H. Markus, 2003. Establishing Relationships Among In-Stream
Nutrient Concentrations, Phytoplankton Abundance and Composition, Fish IBI
and Biochemical Oxygen Demand in Minnesota USA Rivers. Minnesota
Pollution Control Agency, St. Paul, MN.
Omernik, J., 1987. Ecoregions of the conterminous United States. Annals of the
Association of American Geographers 77:118-125.
Park, R. A., E. C. Blancher, S. A. Sklenar, and J. L. Wood, 2002. Modeling the Effects of
Multiple Stressors on a Use-Impaired River, in Society of Environmental
Toxicology and Chemistry, Salt Lake City.
Park, R. A., J. S. Clough, and M. C. Wellman, 2008. AQUATOX: Modeling
environmental fate and ecological effects in aquatic ecosystems. Ecological
Modelling 213:1-15.
Rashleigh, B., M.C. Barber, and D.M. Walters, 2009. Foodweb modeling for PCBs in the
Twelvemile Creek Arm of Lake Hartwell, South Carolina, USA. Ecological
Modelling 220: 254-264.
Sourisseau, S., A. Basseres, F. Perie, and T. Caquet, 2008. Calibration, validation and
sensitivity analysis of an ecosystem model applied to artificial streams. Water
Research 42:1167-1181.
18
------- |