JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION
AMERICAN WATER RESOURCES ASSOCIATION
VOL. 33, NO.2
APRIL 1997
SENSITWITY ANALYSIS OF SIMULATED
CONTAMINATED SEDIMENT TRANSPORT'
Thomas A. Fontaine and Vanusa M. F Jacomino2
ABSTRACT: A simulation analysis of contaminated sediment
transport involves model selection, data collection, model calibration and verification, and evaluation of uncertainty in the results.
Sensitivity analyses provide information to address these issues at
several stages of the investigation. A sensitivity analysis of simulated contaminated sediment transport is used to identify the most
sensitive output variables and the parameters most responsible for
the output variable sensitivity. The output variables included are
flow almost no sediment flux occurs. Sediment eroded
from hillslopes and stream channels during small to
moderate floods (e.g., < 10-year events) can deposit in
the channel network behind debris jams, weirs and
dams, and in the streambed. Major floods (e.g., > 40year events) can scour these sediment deposits as well
as additional hillslope and channel sediment, and
result in the movement of thousands of tons of sedi-
streamfiow and the flux of sediment and Cs137. The sensitivities of
these variables are measured at the field and intermediate scales,
ment per day. The movement of large volumes of sedi-
for flood and normal flow conditions, using the HSPF computer
model. A sensitivity index was used to summarize and compare the
results of a large number of output variables and parameters. An
extensive database was developed to calibrate the model and conduct the sensitivity analysis on a 6.2 mi2 catchment in eastern Tennessee. The fluxes of sediment and Cs137 were more sensitive than
streamfiow to changes in parameters for both flood and normal flow
conditions. The relative significance of specific parameters on output variable sensitivity varied according to the type of flow condi-
ment results in extensive contaminant transport,
tion and the location in the catchment. An implications section
illustrates how sensitivity analysis results can help with model
selection, planning data collection, calibration, and uncertainty
sediment transport in river basins has been exten-
even if the concentrations of particle reactive contaminants are relatively low.
The simulation of contaminated sediment trans-
port at the catchment, or intermediate, scale has
received relatively little attention compared to larger
and smaller scales. At larger scales (e.g., > 500 n-li2)
sively studied and many models have been developed
and tested for this case (Vanoni, 1984; Dawdy and
analysis.
(KEY TERMS: sensitivity analysis; sediment transport; contami-
Vanoni, 1986). Several water quality models for river
applications have been available and evaluated. At
smaller scales (e.g., < 4 mi) field and hillslope applications have been extensively studied by agricultural
nated sediment; HSPF; Cs137.)
engineers and soil scientists. For example, the
INTRODUCTION
CREAMS model (Knisel, 1980; USDA-SCS, 1984) is
one of several contaminant transport models available for small scale hillslopes (Yoon et al., 1992).
Interest in catchment scale investigations and models, which require simulation of both hillslope and
stream channel processes, has increased because of
environmental concerns about nonpoint source pollution associated with sediment and contaminated sedi-
Modeling contaminated sediment transport at the
catchment scale involves the simulation of streamflow, channel hydraulics, erosion, sediment transport,
and interaction between the sediment and contaminants. At this intermediate scale, in-between the field
and river basin scales, streamfiow can vary by orders
of magnitude, making the transport of contaminated
sediment an erratic process. During periods of low
ment (Allen, 1981).
'Paper No. 95126 of the Journal of the American Water Resources Association (formerly Water Resources Bulletin). Discussions are open
until October 1, 1997.
2Respectively, Assistant Professor, Civil and Environmental Engineering Department, South Dakota School of Mines and Technology, 501
East St. Joseph St., Rapid City, South Dakota 57701-3995; and Scientist, Instituto de Pesquisas Energeticas e Nucleares, Travessa R No. 400,
Cidade Universitaria, Sao Paulo, SP, CEP 05508, Brazil.
JOURNAL OF THE AMERICAN WATER RESOURCES AssOcIAnoN
313
JAWRA
Fontaine and Jacomino
Accurately simulating contaminated sediment
transport at the catchment scale requires a carefully
selected computer model and a comprehensive
database. Well designed model algorithms must handle a wide range of conditions for hilislope hydrology,
channel hydraulics, sediment characteristics and contaminant chemistry A substantial database is needed
to setparameter values and to verify that the model
is working correctly for the specific catchment, contaminants and hydrologic events of interest.
Conceptual models for simulating contaminated
sediment transport at the catchment scale can have
over 100 parameters that must be set for a specific
application. The values for these parameters are
either set directly using field measurements or estimated by calibrating and verifying the model with
observed input and output data. Several types of data
critical for simulating contaminated sediment transport are rarely collected on a routine basis; therefore,
sitespecific programs to obtain data to evaluate
parameters can be very expensive and require several
years or more of fieldwork. Parameters that are critical for correct simulation of extreme conditions (e.g.,
floods or droughts) often require databases with periods of 10 years or more to obtain an event of sufficient
magnitude in the observation record.
In order to ensure that an adequate database is
developed to set parameter values, and because the
sensitivity analysis of parameters related to hilislope
erosion in an evaluation of simulated sedimenttransport on a 2.7 mi2 agricultural watershed. At the river
basin scale, a sensitivity analysis of simulated sediment transport has been done by Hrissanthou(1988).
More information is needed on catchment scale appli-
cations, where both hilislope and stream channel
processes are important for the simulation of contami-
nated sediment transport. Studies are also needed
that include short time frames (e.g., single flood
events) as well as longer time frames such as monthly
or annual averages. The experiments reported on in
this paper were designed to provide this type of information.
It is difficult to design a sensitivity analysis producing results that are general enough to apply to a
number of models and specific site conditions. This
research involves a case study using a specific computer model and catchment. The sensitivity of primary output variables is evaluated for two streamfiow
conditions: a month of typical rainfall without severe
storms and the eight-day streamfiow related to a 100year storm with relatively normal antecedent condi-
tions. The sensitivity analysis includes output
variables and parameters related to streamfiow generation, sediment transport, and particle reactive con-
taminants. The role of input data and processes
involving soluble contaminants are issues outside the
field and lab work required to obtain some of this data
is very expensive, it is essential to plan the data col-
scope of this study.
The objectives of this paper are to evaluate the sen-
lection program carefully. Examples of important
sitivity of primary output variables from computer
simulations of contaminated sediment transport at
the catchment scale, and to identify parameters caus-
issues to evaluate during the planning process include
the required length of record and the spatial and tem-
poral resolution, as well as the allowable measurement error for various types of data.
A sensitivity analysis provides information that
can help address these issues. Sensitivity analyses
ing output variable sensitivity. After the results of the
sensitivity analysis are discussed, the application of
the results for model selection, planning data collec-
measure the variability of output variables caused by
tion programs, model calibration, and evaluation of
uncertainty, are illustrated in the implications sec-
perturbations in parameter values and input data.
tion.
Knowing how sensitive each of the important output
variables are, and which of the parameters and input
data have the most impact on output variable sensitivity, improves model selection, planning for the data
collection program, model calibration, and evaluation
ofthe potential uncertainty in the final simulation
results. Rogers et al. (1985) provide an example of
how a sensitivity analysis can be used for these purposes in the application of a physically based runoff
model.
Model Description
For this case study the Hydrological Simulation
Program — Fortran (HSPF)model (Johanson et at.,
1984) was used. The HSPF model was selected
A review of the literature on sensitivity analysis of
because it provides a comprehensive approach to simulate transport in the hydrologic system on a continu-
contaminated sediment transport at the catchment
scale indicates that little has been published on this
subject. At the hilislope scale, investigations using
sensitivity analysis have been done on sediment
simulation (Young et at., 1987) and streamfiow
ous basis. The model has linked components for
rainfall-runoff processes (including overland, shallow
subsurface, and ground water pathways for streamflow generation and contaminant transport); hiflslope
simulation (Calver, 1988). Arihood (1989) included a
JAWRA
METHODS
314
JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION
Sensitivity Analysis of Simulated Contaminated Sediment Transport
erosion; scour, transport and deposition of sediment in
the channel system; and water quality algorithms for
both particle reactive and soluble contaminants. The
model also has options for detailed reporting of state
variables and intermediate fluxes of water, sediment
network. There are specific routines for commonly
analyzed constituents (e.g., water and soil, temperature, pesticides, nitrogen, phosphorus, dissolved oxy-
and contaminants at each time step, and time series
of a large number of output variables. This information is important for diagnosing simulation problems
and completing the calibration and verification of the
can be modeled using fixed concentrations (i.e., constant mg contaminantlmg sediment) or equilibria con-
model.
ment-associated phases.
Over 50 parameters are involved in the HSPF algorithms for simulating the transport of sediment con-
gen, BOD, pH, C02, and alkalinity) as well as options
for simulating general constituents. Contaminants
ditions can be modeled using adsorption and rate
coefficients for transfer between soluble and sedi-
HSPF uses a streamflow generation component
based on the Stanford Watershed Model (Linsley et
al., 1982), which allows detailed modeling of the over-
taminated with a single constituent. An extensive
land, shallow subsurface, and ground water flow
lope erosion as either washoff of soil detached by raindrop impact or as gully erosion. We selected the latter
option because the catchment is almost entirely covered by vegetation, and the visible landforms indicate
database is required to set the values of these parameters. Many parameters are evaluated using information from maps, existing field surveys, and published
reports of comparable studies. The remaining parameter values are estimated during calibration and verification using observed input and output data.
that gully erosion is a primary geomorphologic process at this site. The equation used in HSPF for this
approach is
Catchment Description
pathways. The sediment component simulates hills-
SCRSD = DELT6O*KGER*(SURO/DELT6O)**JGER
The site used for this case study is a 6.2 mi2 catchment located at the Oak Ridge National Laboratory
(ORNL) 20 miles west of Knoxville, Tennessee. A map
of the catchment is shown in Figure 1. The catchment
is 80 percent forest, 10 percent grass field, and 10 percent developed (i.e., similar to urban development).
The terrain is 50 percent steep slopes and ridges and
50 percent valley bottoms and mild slopes. Soils have
a silty or very fine loam texture and are underlain by
dolomite, limestone, sandstone, and shale. Infiltration
capacity is relatively high at the surface, and generally decreases enough in the top 3 feet of soil to create
shallow subsurface stormflow during moderate rainfalls. Overland flow is normally seen only during very
(1)
where SCRSD is the scour of the soil matrix (tons/actime step), DELT6O is the hours/time step, KGER is a
coefficient, SURO is overland runoff (in/time step),
and JGER is an exponent. The development of the
hillslope sediment erosion and transport algorithms
in HSPF were based on the Agriculture Runoff Management Model (ARM) (DOnigian and Davis, 1978)
and Nonpoint Source Pollutant Loading Model (NPS)
(Donigian and Crawford, 1976). The ARM and NPS
models were developed for simulation of runoff, sediment, and pollutants from small watersheds.
The scour, transport, and deposition of sediment in
the channel system is simulated for cohesive and noncohesive material in three size classes. Deposition of
cohesive sediment is based on an approach developed
by Krone (1962). Scour of cohesive sediment is based
wet conditions and is normally located where hillslopes converge.
The sections of the channel network included in the
modeling analysis have streambeds composed of sand
and silt. A number of weirs and flumes used for
streamfiow measurement are located on the tribu-
on an approach developed by Partheniades (1962).
Three options are available for the transport of sand
sized sediment in the channel: the Toffaleti method,
the Colby method, and a simple power function of
taries and main stream channel. Sediment deposits
accumulate behind these weirs as well as upstream of
debris dams formed by trees that have fallen into the
channels. A small, shallow lake is located near the
velocity. We used the Toffaleti method (Toffaleti, 1969)
because it was more appropriate for the conditions at
outlet of the catchment. The seasonal climate generally involves cool winters with occasional heavy rain-
this site than the other options. The channel
hydraulics component routes floods using a kinematic
method and calculates variables required for the simulation of sediment transport in the channel.
The water quality component includes a wide variety of options for simulating the transport of soluble
falls from frontal systems, and hot summers with
occasional intense rainfalls from thunderstorms.
Average annual precipitation is 54 in.
Cs137 was selected as an indicator contaminant for
•
the sensitivity analysis because it is highly particle
reactive, relatively easy to measure, and because the
spatial distribution and transport of Cs137 adsorbed
and particle reactive contaminants in pervious and
impervious hillslope segments and in the channel
JOURNAL OF THE AMERICAN WATER RESOURCES ASsOcIATION
315
JAWRA
Fontaine and Jacomino
to sediment in this catchment has been well docu-
catchment (Borders and Frederick, 1993). A three-
mented by Lomenick and Gardiner (1965), Lomenick
and Tamura (1965), Cerling and Spalding (1982), Cer-
year field data collection program specifically
ling and Thrner (1982), Cerling et al. (1990), and
Sobocinski et al. (1990).
in the catchment at 30-minute intervals using stan-
— CHANNEL
dard methods from Edwards and Glysson (1988). The
type of events sampled included low flow conditions
SUB-BASIN
BOUNDARY
o
as well as a number of floods at and slightly above
MONITORING
bankfull stage, and one five-year flood. Samples were
analyzed for concentrations of sediment and of Cs137
on sediment, and for particle size distributions. Relationships were developed between concentrations of
sediment and Cs137, particle size distributions, and
STA77OIV
0
0.5
1
mile
CATCHMENT
designed for the simulation of contaminated sediment
was used to measure sediment transport and the concentration of Cs137 on sediment during normal and
flood streamfiow conditions (Fontaine, 1991). Manual
and automatic samples were collected at seven sites
BOUNDARY
streamfiow. The collection of these data at such a
detailed spatial and temporal scale, combined with
the excellent historic data for the distribution of
HILLSLOPE SITE
Cs137 in the catchment, provided an unusually comprehensive database for the investigation of contaminated sediment transport.
The entire dataset was used to calibrate the model
by adjusting parameter values until the simulated
streamfiow and concentrations of sediment and
TENNESSEE
Cs137 matched the observed data as closely as possible. Following the procedures from the user's manuals, the model was first calibrated for hydrology, then
for sediment transport, and finally for the Cs137 and
C'
CATCHMENT SITE
sediment relationship. Information from similar
applications of HSPF reported in Donigian et al.
(1984) and Chew et al. (1991) were used to design and
evaluate the calibration of HSPF for this site. The cal-
Figure 1. Map of Whiteoak Creek Catchment.
ibrated model was then verified with additional data
to confirm that the components were set up correctly
and that the parameter values were appropriate. In
the sensitivity analysis that follows the model is used
to simulate hypothetical floods up to the 100-year
event. The maximum streamfiow that occurred during the data collection program used to calibrate the
model was a five-year event. In our opinion this is not
a serious limitation for the objectives of this paper.
However, the results of the sensitivity analysis at the
100-year event should be evaluated with this constraint in mind. Final values of the parameters are
Model Calibration
In order to prepare HSPF for the sensitivity analy-
sis, the catchment was divided into four subcatchments and seven channel sections (Figure 1). This
configuration takes advantage of locations where the
best field data has been collected and distinguishes
between tributaries with and without Cs137 contamination. Preliminary values of the parameters involved
in contaminated sediment transport were set using
the relatively extensive database available from previous and ongoing studies at ORNL. These studies
given in Table 1.
provided information on streamfiow generation mechanisms, the groundwater system, current and historic
rates of erosion, location and magnitude of the source
Sensitivity Analysis
Three different hydrologic scenarios were developed to compare the sensitivity of output variables at
normal flows and during extreme floods. A month of
wet-season baseflow with brief increases in streamflow due to a few minor rainstorms was used to represent contaminated sediment transport at normal flow.
terms for Cs137, and previous computer modeling
investigations (Clapp et al., 1994).
Hydrologic data used for this analysis included
hourly precipitation, hourly streamflow, and daily
evaporation collected at several locations within the
JAWRA
316
JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION
Sensitivity Analysis of Simulated Contaminated Sediment Transport
The 24-hour, 100-year rainfall for the site (6.6 in) was
used to represent the transport occurring during
extreme floods. The first flood scenario used wet
antecedent conditions caused by a two-inch rainfall in
one hour, followed by the 100-year rainfall four days
later. The second flood scenario involved the 100-year
rainfall with relatively normal antecedent conditions.
The primary output variables used for measuring
the sensitivity of HSPF were (1) daily average
streamfiow in cfs; (2) the flux of suspended sediment
in tons/day; and (3) the flux of Cs137 on sediment in
mCi/day. The sensitivities of these three variables
were measured at the outlet of hillslope No. 1 (or the
"hillslope" site) and the outlet of the entire catchment
The historic rainfall and streamflow record for
March 1 to June 8, 1990, with some modifications,
was used as the basis for these three scenarios. The
at the end of reach No. 7 (or the "catchment" site).
Each variable was evaluated for the normal flow and
model was run for the month of March before any of
ty.
the scenarios in the sensitivity analysis occurred.
This is a standard practice used to improve the
results of continuous runoff models that require
parameter values to be specified for initial conditions
such as antecedent soil moisture. April was used for
the month of typical wet-season baseflow, or the nor-
mal flow condition. The 24-hour, 100-year design
storm was inserted into the rainfall record on May 5
with wet antecedent conditions and on June 1 with
normal antecedent conditions. The sensitivity results
of the two extreme flood analyses were so similar that
only the normal antecedent case will be discussed
here. Therefore, the two scenarios used for the results
and discussion sections which follow below will be
referred to as the normal flow and flood cases. Plots of
the hydrologic conditions are given in Figure 2.
flood conditions, resulting in 12 measures of sensitivi-
The sensitivity analysis involved multiple simula-
tion runs, increasing each parameter one at a time
from the calibrated value by 10 percent, except for the
ground water recession coefficient (AGWRC) which
had to be reduced by 10 percent in order to avoid set-
ting the value greater than 1.0. Definitions of the
parameters used in this analysis are given in Table 1.
Using a 10 percent variation in each parameter is
somewhat arbitrary. In a typical modeling analysis,
some parameter values could have a reasonable range
of more than 10 percent, while other parameters may
only vary by 1 or 2 percent. For a standard sensitivity
analysis the use of 10 percent is adequate. The rea-
sonable range that a parameter value would be
expected to have in a model application is an issue
related 'to uncertainty analysis and is briefly discussed later in the implications section.
The results are given for hillslope parameters in
Table 2(A) and for channel reach parameters in Table
2(B). For the normal flow case the number in Table 2
represents the percent change in that output variable
over a one day period. For the flood case, the number
in Table 2 represents the percent change in that out-
2
0
a
0
a,
a.
4
6
put variable over the eight-day period that stream-
8
flow exceeded normal rates. None of the reach
parameters influence variables at the hilislope site,
and reach parameters only influence the flux of sediment and Cs'37; therefore, Columns 2-8 and 11 are
500
400
blank in Table 2(B).
Parameters not included in the results in Table 2
because the impact on every output variable was less
than 1.0 percent are forest cover (FOREST); interfiow
recession coefficient (IRC); hillslope length (LSUR);
300
0'
0
200
slope and Manning's n for hillslopes (SLSUR and
100
0
10
20
30
40
50
60
70
80
90
NSUR); several initial conditions (interfiow (IFWS),
upper soil moisture (UZS), suspended sediment concentration (INIT-[SSID, and concentration of Cs137 in
100
solution (DQUAL)1; diameter of clay and silt (D-CLAY,
Day of Simulation
D-SILT); density of sand, silt and clay; fall velocity of
clay (W-CLAY); some of the channel bed information
(POROSITY, BED DEPTH, and BED WIDTH); distribution coefficient for Cs137 (Kd); and adsorption rate
Figure 2. Precipitation and Discharge
for Sensitivity Analysis.
for solubility of Cs137 (ADSORP-RATE).
A sensitivity index (SI) was used to summarize the
information in Table 2, providing a way to compare
JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION
317
JAWRA
Fontaine and Jacomino
TABLE 1. List of Parameters with Calibration Values.
Calibrated
Value5
Parameter Description
Hifislope Parameters
AGWRC
AGWS
0.990
FOREST
INFILT
INTFW
IFWS
0.10
0.08
4.0
0.05
0.30
0.64
0.132
IRC
JGER
KGER
LSUR
LZS
LZSN
MON-LZET
NSUR
POTFS
SLSUR
UZS
UZSN
Ground water recession coefficient (1/day)
Initial condition for ground water storage (in)
Percent forest cover (fraction)
Infiltration (in/hr)
Interfiow (none)
Initial condition for interfiow storage (in)
Interfiow recession coefficient (1/day)
Exponent for gully erosion (none)
Coefficient for gully erosion (none)
Hillslope length for overland flow (It)
Initial condition for lower zone soil moisture storage (in)
Nominal lower zone soil moisture storage (in)
Coefficient for monthly lower zone evapotranspiration (none)
Manning's n value for overland flow
[Cs] on hillslope sediment (mCi/mg)
Slope of hillslope for overland flow
Initial condition for upper zone soil moisture storage (in)
Nominal upper zone soil moisture storage (in)
7.0
1440
10.8
7.0
0.37
0.30
2.1*10-9
0.14
0.80
0.80
Reach Parameters
ADSORP-RATE
BED DEPTH
BED WIDTH
D-CLAY
D-SAND
D-SILT
density-clay
density-sand
density-silt
DQUAL
INIT-[SS]
IBC
Kd
M-clay
M-silt
POROSITY
SEDCONC
Tau-dep:clay
Tau-dep:silt
Tau-sc:clay
Tau-sc:silt
W.clay
W-sand
W-silt
1.0
Adsorption rate for solubility (1/day)
Depth of bed sediment (ft)
Width of bed (It)
Median diameter of clay (in)
Median diameter of sand (in)
Median diameter of silt (in)
Density of clay (gm/cm3)
Density of sand (gm/cm3)
Density of silt (gm/cm3)
Initial condition for [Cs] in solution (mCi/l)
Initial condition for [ss] in suspension (mg/l)
9.7
35
7.910
0.01
0.00063
2.0
2.65
2.2
1* io15
50/30/20
0.0003
0.60
0.60
0.50
3*10.9
Initial bed composition (percent sand, silt, and clay)
Distribution coefficient for CS-137 (silt and clay; I/mg)
Erodibility of clay (lb/ft2-day)
Erodibility of silt (lb/ft2-day)
Porosity of stream bed (fraction)
Initial condition of[Cs] on bed material (mCilmg)
Critical shear stress for deposition of clay (lb/It2)
Critical shear stress for deposition of silt (lb/It2)
Critical shear stress for scour of clay (lb/It2)
Critical shear stress for scour of silt (lb/It2)
Terminal fall velocity of clay (in/sec)
Terminal fall velocity of sand (in/sec)
Terminal fall velocity of silt (in/see)
0.046
0.051
0.081
0.061
0.00011
1.7
0.007 1
Vaiue may represent an average.
JAWRA
318
JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION
Sensitivity Analysis of Simulated Contaminated Sediment Transport
TABLE 2. Sensitivity of Output Variables (percent change) to a 10 Percent Increase in Hilislope and Reach
Parameters (10 percent decrease for AGWRC). Output variables are streamfiow (Q), sediment (SS),
and Cs-137 (Cs). Parameter abbreviations are fully defmed in Table 1.
Output Variables at Hillslope Location
Flood Flow
Normal Flow
Parameter
(1)
(3)
55
(2)
.
Q
(4)
(5)
Cs
Q
(8)
Q
(9)
SS
(10)
0.0 -48.9
-19.5
-15.0
-45.2
-26.0
(7)
(6)
SS
Output Variable at Catchment Outlet
Flood Flow
Normal Flow
Cs
Cs
(12)
Q 55
(13)
2.5
2.5
(11)
Cs
(A) Hhislope Parameters
AGWRC (ground water recession cx)eff.)
JGER (gulley erosion exponent)
LZSN (lower soil moisture capacity)
UZSN (upper soil moisture capacity)
LZS (initial moisture in lower soil zone)
INFILT (infiltration)
KGER (gulley erosion coefficient)
AGWS (initial condition for baseflow)
INTFW (interfiow)
POTFS ([Cs137] on hilislope sediment)
LZET(lowerzoneETcoefficient)
-49.4
0.0
0.9
0.4
-1.3
1.1
Sensitivity Index: SI (percent)
0.0
7.4
0.0
0.0
0.0
7.6
0.0
-45.2
-25.3
16.0
11.4
-12.0
10.0
0.0
-45.2
-25.3
16.0
11.4
-11.9
10.0
0.0
0.0
-6.2
0.0
-0.2
-6.2
15.8
0.0
3.1
0.0
-45.2
-26.2
17.6
12.7
-12.0
10.0
-1.1
0.0
7.3
0.0
0.0
0.0
-0.2
16.9
12.2
-12.1
10.0
2.2
-6.3
5.7
-0.3
5.1
5.2
7.5
15.4
13.8
1.7
*
*
*
-5.3
-2.8
*
0.0
0.0
0.0
0.0
0.0
0.0
0.0
-1.7
-1.5
0.0
0.0
0.6
-0.9
-1.7
0.0
0.0
0.0 -0.1
0.8
0.7
2.0 3.3
-4.5
-5.4
-2.2
-4.5
-5.4
-2.2
3.2
6.4
6.4
-1.6
-6.0
10.0
-0.2
0.0
0.0
-0.6
-6.0
10.0
0.0
-5.2
0.0
-1.1
14.7
1.7
6.1
0.0
-0.7 -0.4
-2.9 -2.8
-1.8 -1.8
0.0
0.9
0.3
-1.3
1.2
-2.9
-2.0
0.0
0.6
2.8
0.0
-5.2
6.1
3.9
-6.2
0.0
-2.8
-1.6
3.2
-1.6
0.0
0.5
0.0
0.0
-0.6
3.3
3.2
-1.4 -1.1
1.7
1.0
0.2
0.7
0.2
1.0
0.0 0.6
-0.6 -0.6
1.6
1.4
0.0
0.0
(B) Reach Parameters
Tau-dep:silt (silt deposition)
Tau-sc:silt (silt scour)
Tau-sc:clay (clay scour)
M-silt (silt erodibility)
M-clay (clay erodibility)
SEDCONC (initial [CS] on bed material)
IBC (initial %sand, silt & clay in streambed)
W-silt (silt fall velocity)
Tau-dep:clay (clay deposition)
W-sand (sand fall velocity)
D-sand (sand median diameter)
*
*
S
Sensitivity Index: SI (percent)
0.0
0.0
0.0
1.6
-6.7 -7.1
-4.9 -5.2
5.0
5.4
3.3
3.5
9.0
5.6
0.0
0.0
0.0
0.0
0.0
0.0
0.0
0.0
5Reach parameters only affect SS and Cs137, and only at the catchment site.
and POTFS) would never have an impact on streamflow. The SI for each output variable indicates the relative sensitivity of that variable compared to other
variables in the same table. For example, in Table 2A
for the normal flow case at the hillslope location, the
SI for streamfiow is 7.6 percent and the SI of sediment flux is 15.8 percent. Therefore, the overall sensitivity of sediment flux to the 10 percent change in
parameters was significantly greater than the sensitivity of streamfiow under these conditions. The SI
values are only comparable for the same set of parameters (i.e., a SI in Table 2A is not comparable to a SI
the relative sensitivities of output variables. The sensitivity index (SI) for each output variable OV(i) in
Table 2 is calculated using
N
SI=(1/N)*
I change in OV(i) I
i= 1
(2)
where N is the total number of theoretically possible
occurrences for OV(i). For example, N = 8 in the
calculation of SI = (60.5/8) = 7.6 for streamfiow at
normal flow at the hilislope location in Table 2A,
because three of the 11 parameters (JGER, KGER
JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION
in Table 2B).
319
JAWRA
Fontaine and Jacomino
DISCUSSION
the initial condition for lower zone soil moisture
(LZS), also have a larger impact on sediment for the
The results given in Table 2 can be used to evaluate two issues: (1) identify the most sensitive output
variables for this case study, and (2) identify parameters that have the most significant impact on output
variable sensitivity for this case study.
Significant Output Variables
The output variables that are most sensitive to
model parameters can be identified by comparing sen-
sitivity indices (Table 3). The relative sensitivity of
the variables in Table 3(A) related to changes in hills-
lope parameters indicates that the most sensitive
variables are sediment at the hillslope and catchment
sites in the normal flow case, and Cs137 at the hills-
lope and catchment sites in the normal flow case.
Moderately sensitive variables are streamfiow at the
hillslope and catchment in the normal flow case,
Cs137 at the hillslope for the flood case, and sediment
at the hillslope for the flood case.
Comparing the relative sensitivity values of the
variables in Table 3(B) related to changes in channel
reach parameters indicates that the most sensitive
variable is Cs137 at the catchment site for the flood
case. Sediment at the catchment site for the flood case
is a moderately sensitive variable.
The variables in the most sensitive 50th percentile
in Table 3(B) all occur during the flood scenario
because the table only reflects the influence of the
reach parameters, which in this case are primarily
the scour and deposition of sediment in the channel
system. In the scenarios used for this analysis, the
normal flow case involved a little deposition and no
scour, and the flood case involved no deposition and
significant scour. Although there is some impact on
sediment and Cs137 in the normal flow case due to the
(SI, percent) for Output Variables.
increase in the parameter for deposition of silt
Normal
Flow
(Tau:dep-silt), the reach parameters having the primary impact on sediment and Cs137 are related to the
scour of contaminated silt and clay, and scour only
Flood Normal
Flow
Flow
Flood
Flow
occurs during the flood case.
An important conclusion from Table 3 is that the
variability of contaminated sediment at the catch-
(A) Sensitivity to Hillslope Parameters
Sediment
15.8
5.1
Cs'37
14.7
5.2
15.4
13.8
7.6
1.7
7.5
1.6
ment outlet depends on two different sets of processes
1.4
1.7
and parameters. Hilislope parameters dominate the
sensitivity at normal flows, and stream channel
parameters dominate at flood flows. This result
(B) Sensitivity to Reach Parameters
Sediment
*
Cs137
*
*
*
0.8
2
implies that different algorithms or components of a
single model (or even different models in some cases)
0.7
3.3
and different parameters and types of data are
required for accurate contaminated sediment trans-
*ach parameters do not affect variables at the hilislope site.
port simulation over a range of streamfiows.
The difference in processes and parameters having
the most impact on output variable sensitivity for nor-
The variables in the most sensitive 50th percentile
in Table 3(A) all occur during the normal flow sce-
nario. For sediment and Cs137 this result occurs
because overland runoff (SURO) is near zero in the
normal flow case. Based on the equation for soil
mal flow and flood conditions also has important
implications for the adequacy of the dataset used to
calibrate a transport model. Normally the calibration
dataset will represent normal flow conditions better
than floods because of the relative infrequency of
floods. Reliable data during floods is also relatively
matrix scour (Equation 1), a change in the exponent
for gully erosion (JGER) has a much larger percent
impact on sediment flux at low values of overland
runoff than at high values. The parameters for the
volume of lower and upper soil moisture storages
(LZSN and UZSN), the infiltration rate (INFILT), and
JAWRA
impact on overland runoff, and hence sediment, during normal flow compared to during floods. Streamflow is more sensitive for the normal flow case than
the flood case because a change in the parameters for
groundwater simulation (AGWRC and AGWS) has a
strong impact on streamfiow at baseflow conditions in
HSPF. This issue is discussed later.
TABLE 3. Comparison of Sensitivity Indices
Hifislope Location Catchment Outlet
Discharge
normal flow case versus the flood case because a
change in these parameters has a larger percent
scarce because of the additional problems associated
with collecting high quality hydrologic and sediment
transport data during extreme events, especially
when uncontrolled contamination presents additional
hazards. Uncertainty in final simulation results will
320
JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION
Sensitivity Analysis of Simulated Contaminated Sediment Transport
between the two sites with the exception of parameters for groundwater recession (AGWRC and AGWS)
and silt deposition (Tau-dep:silt). The most critical
parameter is the exponent in the equation for gully
erosion (JGER) because the overland flow values are
close to zero, as discussed earlier. Other important
parameters in the normal flow case are related to the
generation of overland flow on the hillslope (LZSN,
UZSN, LZS, and INFILT), and the coefficient in the
increase as a model calibrated with small events is
used to predict transport during larger events. Sensitivity analyses involving a wide range of flow conditions will reveal the output variables most susceptible
to this source of uncertainty, and should be used during the planning stage of a simulation project to select
appropriate models and design adequate data collection programs.
equation for gully erosion (KGER).
The significance of the ground water and silt depo-
Significant Parameters
sition parameters at the catchment site but not at the
hillslope site is consistent with the configuration of
the model and the catchment used for this analysis.
Ground water and channel deposition have no impact
The results in Table 2 can be used to identifS' the
most significant parameters for the sensitivity of each
output variable. In the following discussion streamflow is considered first, followed by sediment and then
on hillslope overland flow, and therefore would not be
represented in the sediment results at the hillslope
Cs137.
outlet. The ground water recession parameters
Columns 2, 5, 8, and 11 of Table 2 indicate the
(AGWRC and AGWS) are significant for the sediment
impact of each parameter on streamfiow. The results
for the normal flow case are essentially the same at
the hillslope and catchmeñt sites. The results for the
flood case are also very similar at both sites. Similar
results for streamflow sensitivity at these sites for a
given flow condition are expected because the parameters and routing process in the reach component do
not significantly impact daily average streamfiow on
this catchment.
flux at the catchment site because in this case study
the streamfiow generated in the normal flow condition is similar to the threshold for sediment deposition in the stream channel. This was discovered by
comparing the time series of shear stress (tau) at the
channel bed to the threshold for sediment deposition
for silt (Tau-dep:silt). Changing the ground water
recession parameters caused a slight decrease in the
streamfiow, and hence the shear stress, during the
month of the normal flow case, increasing the amount
The critical parameters for the sensitivity of
streamfiow in the normal flow condition involve the
baseflow process (AGWRC and AGWS). Less significant are the initial conditions for lower zone soil moisture (LZS) and the infiltration rate (INFILT). These
results occur because the streamfiow during the normal flow condition is primarily wet season baseflow
with a small amount of overland and shallow subsurface stormflow from small rainfalls.
of silt deposition. This relationship between the
threshold shear stress for deposition and sediment
flux is also illustrated by the decrease in sediment
flux of 5.3 percent at the catchment site caused by the
10 percent increase in the silt deposition parameter
(Tau-dep:silt).
The sediment results for the flood condition indicate that the critical parameters for sensitivity of sed-
The critical parameters for the sensitivity of
iment at the hillslope site are different than the
streamflow for the flood case involve the lower zone
soil moisture (LZS and LZSN), the baseflow recession
coefficient (AGWRC), the volume of the upper soil
moisture zone (UZSN) and infiltration rate (INFILT).
The relatively high influence of AGWRC on the sensitivity of runoff during a 100-year flood is due to the
fact that a small change in AGWRC can have a significant impact on streamfiow generation in the HSPF
algorithm, and that the output variables for the flood
case are evaluated over an eight-day period, so that
baseflow recession is a factor in this simulation. The
critical parameters at the catchment site. The critical
parameters for the sensitivity of sediment flux at the
hilislope involve hilislope erosion (KGER and JGER),
and the amount of overland runoff generated (LZS,
INFILT, LZSN, and INTFW). The critical parameters
at the catchment site are related to the scour of cohesive bed sediment (Tau-sc:silt, Tau-sc:clay, M-silt, and
M-clay) and the magnitude of streamfiow generated
(LZS and LZSN).
Columns 4, 7, 10, and 13 of Table 2 indicate the
impact of each parameter on Cs137. Comparing the
sediment and Cs137 results in Table 2 indicates that
changing parameter values by 10 percent has a similar impact on sediment and Cs137 for most of the
parameters. With the exception of the parameter for
concentration of Cs137 on sediment eroded from hillslopes (POTFS) the impacts of the various parameters
on sediment and Cs137 are the same for the normal
large impact on simulated streamfiow of small
changes in AGWRC does not normally cause problems
with HSPF applications because this parameter is rel-
atively easy to calibrate correctly using observed
streamfiow data.
Columns 3, 6, 9, and 12 of Table 2 indicate the
impact of each parameter on sediment. The results for
the normal flow condition are fairly consistent
JOURNAL OF THE AMERICAN WATER RESOURCES AssociATioN
321
JAWRA
Fontaine and Jacomino
flow and flood conditions at the hilislope site. At the
catchment site the impacts of the various parameters
on sediment and Cs137 are similar for the normal flow
and flood condition with the exception of parameters
for the initial size distribution of bed sediment (IBC)
and the initial concentration of Cs137 in the bed sediment (SEDCONC).
The IBC parameter is critical because different
particle size classes of bed sediment are scoured or
deposited at different streamfiow conditions. The rela-
tive significance of these two parameters (IBC and
SEDCONC) compared to the other parameters in the
Cs137 sensitivity at the catchment site for the flood
case was initially unexpected; however, it appears
consistent in hindsight given that the channel scour
processes are more important than hilislope processes
for sediment and Cs137 transport out of the catchment
during floods.
The similarity of results for sediment and Cs137
illustrates the importance of accurate sediment simulation for achieving accurate simulation of contaminated sediment. The high potential for uncertainty of
sediment is illustrated by the large number of parameters that cause a change in sediment of more than 10
percent. These same parameters have similar impacts
on the Cs137 variability, suggesting that the variability of Cs137 flux is closely related to the variability of
sediment flux. Errors in simulating sediment could be
compounded in some applications by the uncertainty
in the relationship between the contaminants and
sediment.
Table 2 shows that the most important parameters
for Cs137 flux at both sites in the normal flow case are
related to scour of sediment off the hillslope (JGER
and KGER) and overland runoff (LZSN, UZSN, LZS,
and INFILT). Baseflow is significant only at the
catchment site, for the same reasons given earlier.
The most important parameters for Cs137 flux at the
hillslope site for the flood case are the coefficient for
gully erosion (KGER), overland runoff parameters
(LZS and INFILT), and the concentration of Cs137 on
hilislope sediment (POTFS). The critical parameters
for the sensitivity of Cs137 flux at the catchment site
for the flood case are related to the initial concentration of Cs137 on the channel sediment (SEDCONC),
initial size distribution of bed material (IBC), and
channel scour (critical shear stress for scour of silt
The parameters determining the overland flow and
hillslope scour have a strong non-linear effect on sedi-
ment and Cs137 flux at normal flows, as discussed
earlier in the results of Table 3, and therefore the
POTFS parameter is relatively insignificant.
In the flood case at the hillslope scale, the overland
flow and hillslope scour variables and parameters do
not have the same strong non-linear relationship as
in the normal flow case, and the POTFS parameter
has a higher impact on Cs137 flux relative to most of
the other parameters in the flood case at the hillslope
site. At the catchment scale the primary source of
contaminated sediment during floods is channel sediment. As Table 2 indicates, the impact on Cs137 flux of
the concentration of Cs137 on channel sediment (SED-
CONC) is relatively high compared to the other
parameters for floods at the catchment site.
IMPLICATIONS
These results for significant variables and parameters can be used for planning and evaluating model
applications for contaminated sediment transport. At
the outset of an analysis, one or more models must be
identified as having the potential to adequately simulate the important processes involved. The model(s)
used will partially determine how much time and
money are required for the data collection and model
calibration efforts. Decisions about the model selection and the extent of data collection and calibration
will influence the potential uncertainty in the final
simulation results, and therefore the usefulness of
those results. The results of a sensitivity analysis
have the following implications for each of these
issues.
The goal of model selection should be to include all
of the critical processes but avoid unneeded complexi-
ty, as demonstrated recently in Beven (1989) and
Grayson et al. (1992). A simpler model that is well
understood and calibrated is superior to a more complex model that is less understood or can not be ade-
quately calibrated. A sensitivity analysis ofthe
models under consideration will indicate whether the
important output variables adequately respond to the
and clay {Tau:sc-silt and Tau:sc-clay], and the erodi-
expected range of input conditions. Models having
bility of silt [M-silt]).
The parameters for concentration of Cs137 on sedi-
overly sensitive output variables or unnecessary com-
plexity will require more effort to achieve useful
results. In this case study, the threshold for critical
ment at both hilislope and catchment scales (POTFS
and SEDCONC) are relatively insignificant for Cs137
flux in the normal flow case but become very significant in the flood case. In the normal flow scenario for
shear stress of cohesive sediment, and the nonlinear
impact of hilislope parameters for normal and flood
flows, are examples of components in HSPF that
require special attention to be sure the calibration
both the hillslope and catchment cases, this result
occurs because the primary source of sediment is
JAWRA
hillslope scour during brief periods of overland flow.
and final simulation results are valid.
322
JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION
Sensitivity Analysis of Simulated Contaminated Sediment Transport
Early in the modeling analysis, before a comprehensive database has been developed and a valid calibration effort completed, the sensitivity analysis may
be approximate or preliminary. In this case, an iterative approach may be used to (1) select one or more
models for evaluation, (2) identif' and acquire critical
data, (3) calibrate the model, (4) apply the sensitivity
analysis, (5) evaluate whether the models are appropriate, and then repeat the sequence, sorting models
and collecting more data as needed.
A sensitivity analysis done on the specific model(s)
involved will help determine the type and amount of
data required to set values for the parameters. Data
that are important and are expensive to collect, or
that require a long data collection program (e.g., three
years or more), must be identified early in the planfling process to develop accurate budgets and time
schedules. Parameters that can be set with existing
data, or that have a relatively insignificant impact on
output variables, should be identified to minimize
calibration phase. The relative possibility of acquiring good calibration data and a reasonably accurate
parameter value, and the expected effort required for
the calibration, are given for the 13 most significant
parameters in this case study. For this illustration
the possibility of acquiring good data was assumed to
depend on relative cost (Column 8) and how accurately the critical data could be measured. A parameter
value was assumed to be relatively accurate if good
calibration data were likely and if a calibration technique exists that could normally result in a unique
and accurate parameter value (e.g., no problems with
parameter correlation or thresholds).
Model simulations are more useful if the uncertainty of the results is known. Uncertainty can be caused
by incorrect model structure, natural variability of
processes and parameters in space and time, and
errors in input data and parameters. Uncertainty
related to parameters has two components: parameter
impact and parameter accuracy. Sensitivity analyses
measure parameter impact on output variables, while
unnecessary data collection.
Columns 1 to 8 of Table 4 illustrate the use of sensitivity analysis results to plan a data collection program (Columns 9 to 12 are described shortly). The 13
evaluating parameter accuracy requires measuring
the difference between true parameter values and values estimated by the calibration process. A full uncer-
parameters having the most impact on output variability in this case study are listed along with information about the data type, sampling details (eg.,
location, interval, period of record, and method), and
tainty analysis involving both aspects is best done
relative cost, that would be recommended to generate
ple).
the database required for calibrating these parameters. Period of record depends on streamfiow condi-
analysis using a sensitivity analysis and estimates of
using a probabilistic analysis involving frequency dis-
tributions for parameter values plus a Monte Carlo
sampling approach (see Binley et al., 1991 for exam-
In some applications a preliminary uncertainty
parameter accuracy can be useful, avoiding the
tions involved; the values in Column 6 are minimums
for simulations of streamfiow up to small floods (e.g.,
< 10 years). For the purpose of this illustration, rela-
expense and effort of a full uncertainty analysis. This
simple approach is illustrated in Column 12 of Table
4, based on the results of this case study. The relative
potential uncertainty rating is a combination of three
tive cost was assumed to be low if the type of data
was often available from a public database (e.g., routine data provided by the U.S. Geological Survey or
the National Oceanic and Atmospheric Administration). Relative cost would be high if an expensive inhouse program of fieldwork and lab analyses were
required (e.g., information about critical shear stress
of bed material, concentration of contaminants on
factors: (1) the impact of the parameter on output
variable sensitivity, (2) the range that the parameter
value could reasonably have in this case study, and
(3) the possibility of setting the parameter value correctly. A detailed explanation of this approach is outside the scope of this paper; however, the results in
Column 12 show that if model output uncertainty in
sediment, or particle size distributions).
this application had to be reduced, the first task
During the planning stage, a sensitivity analysis
can be used to estimate the amount of effort required
to calibrate each parameter, and to detect threshold
values where certain parameters have a discontinuous impact on output (e.g., the threshold for scour of
cohesive sediment from channel beds in HSPF). During the actual calibration effort, having information
about the relative impact of each parameter on output
variables will help determine when the calibration
and verification are adequate for the objectives of the
modeling analysis.
should concentrate on evaluating the collection of
data and the calibration of parameters for hillslope
erosion (JGER and KGER), and the scour of silt and
clay in the channel (Tau-sc:silt, Tau:sc-clay, and Msilt).
Table 4 is only intended to illustrate in general how
a sensitivity analysis can be used to improve a model
application. The specific modeling objectives for a
given application would determine which parameters
are most significant, the type of data and calibration
effort required, and how the potential uncertainty
Columns 9 to 11 of Table 4 illustrate the use of
sensitivity analysis results to plan and conduct the
JOURNAL CF THE AMERICAN WATER RESOURCES AssociATioN
would be estimated.
323
JAWRA
TABLE 4. Example of Application of Sensitivity Results to Data Collection, Calibration, and
Uncertainty for Simulation of Contaminated Sediment Transport.
Data Needed
Channel
Location
Sample
Collection
Method
Relative
Cost
.
Relative
Potential
Uncertainty
Parameter
Type
Interval
(1)
(2)
(3)
Record
(yrs)
(4)
(5)
(6)
(7)
(8)
(9)
(10)
(11)
(12)
AGWRC
P, Q, ET
yes
no
hi', day
3
auto.
low
high
high
low
low-moderate
LZSN & UZSN
P, Q, ET
yes
no
hr
3-5
auto,
low
high
moderate-high
low-moderate
moderate
LZS
P, Q, ET
yes
no
hr
1-2
auto,
low
high
high
low
low-moderate
[SS], Q
yes
no
hr
3-5
auto., man.
moderate
moderate
low-moderate
moderate-high
high
P, Q, ET
yes
no
hr
3-5
auto.
low
high
moderate-high
low-moderate
moderate
Tau-sc:silt & clay
[SS], Q, core
no
yes
0.25 hr
3-5
auto., man,
high
low
low
high
high
M-silt
[SS], Q, core
no
yes
0.25 hr
3-5
auto., man,
high
low
low
high
high
POTFS
core, [Cs]
yes
no
0.5 hr
3-5
auto., man,
moderate-high
moderate
moderate
moderate
moderate-high
SEDCONC
core, [Cs]
no
yes
0.5 hr
3-5
auto., man,
moderate-high
moderate
moderate
moderate
moderate-high
IBC
core, psd
no
yes
N/A
N/A
man.
low
high
high
low
moderate
INFILT
0-n
P
Q
ET
= precipitation
= streamfiow
= evapotranspiration
[SS] =
[Cs] =
core =
psd =
N/A =
m
Relative CalibrationEffort
Availability Accuracy
of
of
Good
Parameter
Effort
Data
Value
Required
Hilislope
Location
JGER & KGER
C
Period
of
auto. =
man. =
suspended sediment
concentration
concentration of Cs on mobile sediment
core of bed material or soil
particle size distribution
not applicable
automatic
manual
'Ti
0
a
0
a
0C-
a
8
S
00
-
Sensitivity Analysis of Simulated Contaminated Sediment Transport
The flux of sediment and particle reactive contami-
CONCLUSION
nation out of a catchment can be a highly variable
process. The processes and model parameters critical
for simulating these fluxes at low streamflows can be
quite different than the critical processes and parameters at higher flows. A single model may not be able
to adequately simulate these fluxes for both streamflow conditions, and in most cases much more data
This investigation provides a case study of a sensi-
tivity analysis of simulated contaminated sediment
transport. Some of the concepts and sensitivity
results can be applied to other simulation studies
involving similar models, hydrology, and catchment
conditions. The database used is relatively comprehensive because of the extensive field data collection
program specifically designed for this purpose, as well
as the amount of information available on the distribution of the Cs137 contamination from other studies
at Oak Ridge National Laboratory.
The sensitivities of streamfiow, and the flux of sediment and Cs137 were evaluated at the outlets of the
hillslope and catehment for flood and normal flow scenarios. A sensitivity index was developed to summa-
will be available for calibrating and verifying the
model at the lower flows compared to the higher
flows. These issues have implications for model selection, data collection, calibration and verification of the
model, and uncertainty in the results.
The results illustrate that the accuracy of simulated sediment flux depends on accurate streamfiow sim-
ulation, and that both streamfiow and sediment flux
must be accurately simulated to achieve good simulation of particle reactive contamination transport. At
rize the results and provide a comparison of
low flows in particular, at both the hillslope and
catchment scales, the parameters for contaminant
sensitivities between a large number of parameters
and the output variables. The most sensitive output
variables for the normal flow scenario are sediment
and Cs137 at both the hillslope and catchment outlet
locations. The most sensitive output variables for the
flood scenario at the hillslope location are sediment
concentration were relatively insignificant compared
to the parameters for streamfiow and sediment flux.
One suggestion for additional study is to compare
the quality of modeling results using the relatively
extensive database from this study with a more limit-
and Cs137. At the catchment outlet for the flood case,
all three output variables are equally sensitive.
ed database that would represent the information
available in a typical modeling investigation. The
results from this type of study would provide guid-
The impact of model parameters on the output
variables was evaluated at the hillslope and catch-
ance to agencies involved with the routine collection
ment sites for the same flood and normal flow condi-
of sediment and water quality data. In the current
tions. For the normal flow scenario at the hilislope
location the most important parameters are (1) the
ground water recession coefficient (AGWRC) for
streamfiow, and (2) the exponent in the gully erosion
equation (JGER), and lower and upper soil moisture
period of tight budgets, it is important to ensure that
data critical to contaminated sediment transport is
acquired over a long period of time involving a range
of conditions, including extremes of streamflow and
seasonal variations in land surface conditions and
storm types. Collecting these data on a short term
basis to satisfy the requirements of a specific project,
for just a few months at a specific site for example,
will not provide adequate results because of the vari-
storage volumes (LZSN and UZSN) for sediment and
Cs'37. For the normal flow scenario at the catchment
outlet, the most important parameters are (1)
AGWRC for streamfiow; (2) JGER, LZSN, AGWRC,
and UZSN for sediment; and (3) JGER, LZSN, UZSN,
ability of contaminated sediment transport, as
and AGWRC, for Cs137.
demonstrated in this investigation.
For the flood scenario at the hillslope location the
most important parameters are (1) initial condition
for lower soil moisture storage (LZS), AGWRC, LZSN,
and UZSN for streamfiow; (2) coefficient in the gully
ACKNOWLEDGMENTS
erosion equation (KGER), LZS, and infiltration rate
(INFILT) for sediment; and (3) KGER, LZS, concentration of Cs137 on hillslope sediment (POTFS), and
INFILT for Cs137. For the flood scenario at the catch-
Telena Moore set up the channel routing component and helped
with the collection and analysis of field data. Tim Diehl, Rollin
Hotchkiss, and Telena Moore provided helpful comments on earlier
versions of the manuscript. Primary sponsors of this research were
the Office of Environmental Restoration and Waste Management,
ment outlet the most important parameters are
U.S. Department of Energy (under contract with Martin Marietta
Energy Systems, Inc.), and the Instituto de Pesquisas Energeticas e
Nucleares in Brazil. Additional support was provided by the South
(1) LZS, AGWRC, and LZSN for streamflow; (2) scour
of cohesive bed sediment (Tau-sc:silt, M-silt, and Tau-
sc:clay) for sediment; and (3) initial concentration of
Dakota School of Mines and Technology.
Cs137 on bed sediment (SEDCONC), initial size distri-
bution of bed material (IBC), and scour of cohesive
bed sediment (Tau-sc:silt, M-silt, and Tau-sc:clay) for
Cs137.
JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION
325
JAWRA
Fontaine and Jacomino
Johanson, R. C., J. C. Imhoff, H. H. Davis, Jr., J. L. Kittle, Jr., and
A. S. Donigian, 1984. User's Manual for the Hydrological Simulation Program-Fortran (HSPF). U.S. Environmental Protection
LITERATURE CITED
Allen, P. B., 1981. Measurement and Prediction of Erosion and Sed-
iment Yield. Agricultural Reviews and Manuals, ARM-S-15,
Agency, Athens, Georgia.
U.S. Department of Agriculture, 23 pp.
Arihood, L. D., 1989. Evaluation of a Watershed Model to Simulate
Sediment Transport in a Small Agricultural Watershed in Indiana. U. S. Geological Survey, Water Resources Investigations
Report 88-4222, 40 pp.
Beven, K., 1989. Changing Ideas in Hydrology — The Case of Physically-Based Models. Journal of Hydrology 105:157-172.
Knisel, W. G. (Editor), 1980. CREAMS: A Field Scale Model for
Chemicals, Runoff, and Erosion from Agricultural Management
Systems. Conservation Research Report No. 26, U.S. Depart-
Binley, A. M., K. J. Beven, A. Calver, and L. G. Watts, 1991.
California, Berkeley, California.
Linsley, R. K., M. A. Kohler, and J. L. H. Paulhus, 1982. Hydrology
for Engineers. McGraw-Hill, New York, New York.
Changing Responses in Hydrology: Assessing the Uncertainty in
Physically Based Model Predictions, Water Resources Research
27(6):1253-1261.
Borders, D. M. and B J. Frederick, 1990-1993. Hydrologic Data
Summary for the White Oak Creek Watershed at Oak Ridge
National Laboratory, Oak Ridge, Tennessee.
Calver, A., 1988. Calibration, Sensitivity and Validation of a Physically-Based Rainfall-Runoff Model. Journal of Hydrology 103:
103-115.
Cerling, T. E., S. J. Morrison, R. W. Sobocinski, and I. L. Larsen,
1990. Sediment-Water Interaction in a Small Stream: Adsorption of Cs137 by Bed Load Sediments. Water Resources Research
26(6):1165-1176.
Cerling, T. E. and B. P Spalding, 1982. Distribution and Relationship of Radionuclides to Streambed Gravels in a Small Water-
ment of Agriculture, Washington, D.C.
Krone, R. B., 1962. Flume Studies of the Transport of Sediment in
Estuarial Shoaling Processes. Hydraulic Engineering Laboratory and Sanitary Engineering Research Laboratory, University of
Lomenick, T. F. and D. A. Gardiner, 1965. The Occurrence and
Retention of Radionuclides in the Sediments of White Oak Lake,
Health Physics 11:567-577.
Lomenick, T. F. and T. Tamura, 1965. Naturally Occurring Fixation
of Cesium-137 on Sediments of Lacustrine Origin. Soil Science
Society Proceedings, pp. 383-387.
Partheniades, E., 1962. A Study of Erosion and Deposition of Cohesive Soils in Salt Water. Ph.D. Thesis, University of California,
Berkeley, California.
Rogers, C. C. M., K. J. Beven, E. M. Morris, and M. G. Anderson,
1985. Sensitivity Analysis, Calibration and Predictive Uncertainty of the Institute of Hydrology Distributed Model. Journal
of Hydrology 81:179-191.
Sobocinski, R. W., T. E. Cerling, S. J. Morrison, and I. L. Larsen,
1990. Sediment Transport in a Small Stream Based on Cs-137
shed. Environ. Geol. 4:99-116.
Cerling, T. E. and R. R. Turner, 1982. Formation of Freshwater Fe-
Mn Coatings on Gravel and the Behavior of Co-60, Sr-90, and
Inventories of the Bed Load Fraction. Water Resources Research
Cs137 in a Small Watershed. Geochimica et Cosmochimica Acta
46:1333-1343.
Chew, C. Y., L. W. Moore, and R. H. Smith, 1991. Hydrological Sim-
Toffaleti, F. B., 1969. Definitive Computation of Sand Discharge in
Rivers, ASCE Journal of Hydraulic Engineering 95(HY1):225-
26(6):1177-1187.
ulation of Tennessee's North Reelfoot Creek Watershed.
246.
Research Journal of the Water Pollution Control Federation,
U.S. Department of Agriculture Soil Conservation Service, (USDA-
SCS), 1984. User's Guide for the CREAMS Computer Model.
63(1):10-16.
Clapp, R. B., J. A. Watts, and M. A. S. Guth (Editors), 1994. Third
Annual Environmental Restoration Monitoring and Assessment
Report for FY 1994 of the Oak Ridge National Laborathr Oak
Ridge, Tennessee. DOE/ORIO1-1290&D1, ORNIJER-250.
Dawdy, D. R. and V. A. Vanoni, 1986. Modeling Alluvial Channels,
Water Resources Research 22(9):71S-81S.
USDA-SCS Engineering Division, Technical Release 72, Washington, D.C.
Vanoni, V. A., 1984. Fifty Years of Sedimentation. ASCE Jr. of
Hydraulic Engineering 110(8): 1022-1057.
Yoon, K. S., K. H. Yoo, J. M. Soileau, and J. T. Touchton, 1992. Sim-
Donigian, A. S., J. C. Imhoff, B. R. Bicknell, and J L. Kittle, Jr.,
1984. Application Guide for Hydrological Simulation Program-
ulation of Sediment and Plant Nutrient Losses by the CREAMS
Water Quality Model. Water Resources Bulletin 28(6):10131021.
Fortran (HSPF). U.S. Environmental Protection Agency, Athens,
Georgia.
Donigian,Jr., A. S. and N. H. Crawford, 1976. Modeling Nonpoint
Young, R. A., C. A. Onstad, D. D. Bosch, and W. P. Anderson, 1987.
Pollution from the Land Surface. Environmental Protection
Dept. of Agriculture, Agriculture Research Service, Morris, Minnesota.
Agency, Athens, Georgia.
Donigian, A. S. and H. H. Davis, Jr., 1978. User's Manual for Agricultural Runoff Management (ARM) Model. U.S. Environmental
Protection Agency, Athens. Georgia.
AGNPS, Agricultural Nonpoint Source Pollution Model: A
Watershed Analysis Tool. Conservation Res. Report 35, U.S.
Edwards, T. K. and G. D. Glysson, 1988. Field Methods for Measurement of Fluvial Sediment. U.S. Geological Survey Open File
Report 86-531, U.S. Geological Survey, Reston, Virginia.
Fontaine, T. A., 1991. Monitoring and Modeling Contaminated Sediment Transport in the White Oak Creek Watershed. Oak Ridge
National Laboratory, ORNL'ER-65, Oak Ridge, Tennessee, 19
pp.
Grayson, R. B., I. D. Moore, and T. A. McMahon, 1992. Physically
Based Hydrologic Modeling 2. Is the Concept Realistic? Water
Resources Research 26(10):2659-2666.
Hrissanthou, V., 1988. Simulation Model for the Computation of
Sediment Yield Due to Upland and Channel Erosion from a
Large Basin. In: Sediment Budgets. lABS Publication No. 174,
International Association of Hydrological Sciences, Washington
D.C., pp. 453-462.
JAWRA
326
JOURNAL OF THE AMERICAN WATER RESOURCES ASSOCIATION
© Copyright 2026 Paperzz