Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
John E. McCarthy
Master of Science
August 2012
John E. McCarthy
Department of Civil and Environmental Engineering, BYU
Master of Science
1 Introduction ........................................................................................................................... 1
v
2.7.3 Calibrating with the SED_K Parameter .................................................................... 23
3.10.1 Rerunning with the Slope and Unit Discharge and Generic Sediment Fractions ..... 43
4 Conclusion ........................................................................................................................... 57
REFERENCES ............................................................................................................................ 61
APPENDIX .................................................................................................................................. 63
vi
LIST OF TABLES
Table 2-1: Total wash-load outlet discharge results with various transport capacity equations for
previously calibrated model ...................................................................................................11
Table 2-2: Soil erodibility factor for the detachment by rainfall impact ...........................................15
Table 2-3: Sensitivity analysis washload discharge results for soil erosion parameters ...................17
Table 2-4: Sensitivity analysis sand discharge results for soil erosion parameters ...........................17
Table 2-5: Soil erosion with varying max erosion depths .................................................................18
Table 3-3: Sediment discharge results for land use modification scenarios .....................................49
Table 3-4: Change in sediment discharge for connected and disconnected scenarios ......................51
Table 3-5: Normalized sediment discharges for hurricane and 1983 models ....................................55
vii
viii
LIST OF FIGURES
Figure 1-1: GSSHA 2D overland flow grid and stream interaction. ...................................................3
Figure 2-1: Net sediment accumulation [tons/ha] with different cell sizes . .......................................9
Figure 2-2 Spatial distribution for the net sed. transfer for each transport capacity equation. ..........12
Figure 2-3: Example area of high erosion (blue is deposition and red is erosion). ...........................20
Figure 3-2: GSSHA finite difference grid for the Aguacate watershed.............................................31
Figure 3-7: Max erosion parameter in the arc properties window .....................................................36
Figure 3-8: Areas of high erosion (red) and deposition (blue) ..........................................................37
Figure 3-9: Sedograph for uncalibrated Kilinc-Richardson GSSHA sediment model ......................39
Figure 3-10: Log scale sedograph comparison of GSSHA model results .........................................42
Figure 3-11: Spatial distribution of erosion and deposition with both GSSHA models ....................43
Figure 3-12: Land use index maps before and after alterations .........................................................46
Figure 3-13: Soil erosion parameters for new deforested area ..........................................................48
Figure 3-14: Spatial variations of net sediment transfer for different LU scenarios .........................49
Figure 3-15: Connected to stream arc and disconnected to stream LU index maps ..........................50
Figure 3-17: Stream buffer zone land use index map ........................................................................53
Figure 3-18: Comparison of the net sed. transfer for hurricane and 1983 models ............................55
ix
x
1 INTRODUCTION
The purpose of this report is to outline the methods and issues involved with using the
Analysis (GSSHA) sediment transport model. Furthermore, a case study is presented in which
the GSSHA sediment transport capabilities are demonstrated for the Aguacate watershed in the
Dominican Republic.
The GSSHA model is a reformulation and enhancement of the distributed runoff model
Cascade of Planes in Two Dimensions (CASC2D) (Ogden 2000). GSSHA is a physically based,
two-dimensional model in which the watershed is discretized into a finite difference grid of any
size.
Processes that occur before, during, and after a rainfall event are calculated for each grid
cell and then the response from individual grid cells are integrated to produce the watershed
nutrient transport, groundwater surface-water interaction, wetland simulation, and lumped and
distributed rainfall by solving the finite difference equations. The model solves transport
equations using finite difference and finite volume techniques and it uses 2D diffusive-wave
overland flow routing and 1D diffusive-wave channel routing (Downer et al., 2002).
1
GSSHA is a process-based model with an option to select the specific processes to be
modeled for a particular application. Among the processes that can be simulated are precipitation
water retention, infiltration, overland flow runoff, overland erosion and deposition, channel
unsaturated groundwater flow (vadose zone modeling), saturated groundwater flow, stream
During an event, rainfall is spatially and temporally distributed over the watershed.
Rainfall may be intercepted by vegetation before reaching the land surface. Once an initial
interception demand is reached, a fraction of the precipitation will reach the land surface. Upon
reaching the land surface, precipitation may infiltrate due to gravity and capillary forces. Water
remaining on the land surface may runoff as 2D overland flow, after a specified retention depth
representing micro-topography has been reached (see Figure 1-1). This water may eventually
enter a stream and be routed to the watershed outlet as 1D channelized flow. Between
flow can be simulated. When precipitation falls in the form of snowfall, the water equivalent
volume remains on the land surface and is released as melted water according to an energy
budget calculation (Downer et al., 2002; Downer et al., 2008; Wiki, 2009).
2
Figure 1-1: GSSHA 2D overland flow grid and stream interaction.
Many methods exist for predicting sediment transportation and watershed erosion. The
most common methods often involve empirical equations such as the Universal Soil Loss
Equation (USLE) or one of its modified variants. These empirical equations develop a
watershed’s sediment yield from a variety of factors including rainfall, land use, soil type,
ground cover, topography, runoff, drainage area, and sediment characteristics. However, because
the equations are lumped empirical models of erosion, they do not account for spatial variation in
rainfall, land use, topography, and soil type. Furthermore, the empirical equations are often
based on certain watersheds and the applicability of such equations is often limited to watersheds
with similar characteristics. Many distributed processed-based models that contain both
empirical and physically based components, including GSSHA, have been developed to account
for spatial variability and to understand watershed evolution (Yang et al, 2006).
3
1.3 GSSHA Sediment Transportation Background
The sediment transport process in GSSHA originated from the CASC2D model.
Investigations by Ogden and Heilig (2001) and later by Kalin and Hantush (2006) found
problems with the sediment transport methods used by GSSHA. Many of the concerns have
since been addressed by the developers of GSSHA (Downer 2010). Currently the sediment
transport capabilities in GSSHA are still under development and refinement. The build of
GSSHA that was used throughout this report was GSSHA 5.7s.
The sediment transport methodologies used in GSSHA allow for the modeling of any
particles smaller than gravel with a specific gravity greater than zero. The basic process used in
GSSHA to model sediment transport is first, the detachment of the particles either though
raindrops or surface runoff, and second, the calculation of the transport capacity. The amount of
sediment that is transported is calculated by comparing the detached sediment with the transport
capacity. Furthermore, the transported particles are treated as either wash load or bed load,
(Downer 2010).
First, the user must define the soil types, the size and density of the particles in the flow
plane, and the soils’ fractional composition for each particle type. As GSSHA runs, it will store
the sediment data in three bins: parent soil, suspended sediment, and deposited material.
The sediment is detached in a rainfall event either through the splash of raindrops or
through overland flow. The detachment due to raindrops is a function of rainfall intensity and it
is calculated following a similar approach as presented by Wicks and Bathurst (1996). Some of
the parameters needed to calculate the detachment due to rainfall splash are calculated through
4
WMS. The parameters that are not calculated with WMS including the canopy cover, ground
cover, and soil dependent raindrop erodibility factor are multiplied together to create one
aggregate input parameter for each combination of land use and soil type in GSSHA in order to
The surface runoff detachment follows the method used in the Water Erosion Prediction
Project (WEPP) model which focuses on the shear stress exerted by the runoff on the bonds
between particles and the available sediment transport capacity of the runoff. Through GSSHA
the flow shear stress is calculated. The user must input three parameters, the critical shear stress
and two empirical coefficients. These values can be determined from published WEPP literature.
GSSHA first applies the detachment forces from raindrops and overland flow to the deposited
material. Once all of the remaining deposited sediments are eroded, the parent soil can then be
eroded.
The sediment transport capacity is then calculated. GSSHA allows users to select which
sediment transport capacity equation to use. Currently three equations are ready for use. The
most commonly used and tested equation is the Kilinc-Richardson (1973) sediment transport
equation, as modified by Julien (1995) and Ogden & Heilig (2001). However, this equation
calibration data is available. The Slope and Unit Discharge equation is recommended when
calibration data is not available. To calculate the sediment transport capacity, the user must
define the soil erodibility coefficient. Like the previously mentioned rainfall impact detachment
input parameter, the soil erodibility coefficient is the product of three parameters that can be
determined by referencing land use and soil type. In newer versions of GSSHA this aggregated
parameter is used solely to calibrate models that use the Kilinc-Richardson and Engelund-
5
Hansen transport capacity equations. Models that use the Slope and Unit Discharge transport
equation do not need an input for this parameter. Once the transport capacity is determined, the
available sediment, up to the amount of sediment needed to reach the transport capacity, is then
If the suspended sediments exceed the sediment transport capacity, possibly due to a
decrease in velocity, then the sum of the excess sediments will be deposited according to the trap
efficiency. The trap efficiency for the excess suspended sediment is calculated for each particle
size fraction using an equation developed by Johnson et al (2000). The excess sediment that was
neither transported nor deposited will remain in suspension. In a later time step it can be
transported or deposited. Once the overland flow reaches a stream node, the suspended and
deposited materials in the overland flow cell are added to the stream flow.
In the channel the particles are determined if they are washload or bedload, depending on
how the particles sizes compare to the user defined size of sand. If the particles are sand or
larger, they are considered bedload. If the particles are smaller than sand then they are
considered washload. The bedload transport is calculated using Yang’s unit stream power
method (Yang 1973). The most important parameter for the channel erosion is the max erosion
depth of the stream bed. The advection-dispersion equation is used to transport the washload
sediment. The washload sediments are assumed to not settle in the stream. Because it is assumed
that only the stream bed is mobile and the banks are fixed, the only sediments that are detached
in the stream are those with a size of sand or larger. Ultimately this means that the channel is
neither a source nor a sink for fines (GSSHAwiki, 2009c). This is a limitation for watersheds in
6
2 GSSHA SEDIMENT TRANSPORTATION METHODS AND ISSUES
The following are common methods and issues associated with sediment modeling with
GSSHA is not within the scope of this paper, there are a few important aspects to keep in mind
when constructing a model that will ultimately be used for sediment transport.
Spatially distributed 2D erosion models tend to be less accurate at predicting erosion for
single rainfall events. This is likely in part due to the limitations that 2D models have in
capturing watershed complexity and natural variability too small and intricate to be represented
in a model. However, as many events are lumped together into a long term simulation, the
inaccuracies tend to average themselves out (Yang et al, 2006). Furthermore, the generation and
movement of sediment from overland erosion frequently takes more time than would be allotted
in a single event simulation. Often it takes a few months of simulation for the model to build up
to a stable condition. Because of these issues it is recommended that calibrated long term
hydrologic models are constructed before modeling sediment transport. Unlike short term
7
simulations, long term simulations require continuous HMET data for the duration of the
simulation. Another consideration in a long term simulation is the amount of CPU time required
to run the model. The most important factor of the simulation run time is the grid cell size.
According to Rojas et al. (2008), when doing sediment transport the overland erosion is
especially sensitive to the grid cell size. This is due to the reduction in land surface slopes and
changes in the channel network topology. Even if the hydrologic parameters were adjusted for
the models with larger grid cells in order to calibrate to a measured runoff, significant differences
in erosion results remained. Figure 2-1demonstrates how sensitive the erosion can be to different
For the model of interest in the study performed by Rojas et al. (2008) it was determined
that grid cell sizes of 150 meter or more produced inaccurate erosion models. However it should
be noted that the size of the watershed in the study was about 20 square kilometers. While no
suggestions were given for larger watersheds, it may be reasonable to assume that as the
watershed size increases, the allowable cell size will also increase. Nevertheless, it should be
kept in mind that the grid cell size should be small enough to capture the essential features of a
watershed (GSSHAwiki, 2008). For a sediment model this would include the details of land use
and soil type, as well as the channel network topology and land surface slopes.
8
Figure 2-1: Net sediment accumulation [tons/ha] with different cell sizes (Rojas et al.,
2008).
As mentioned above, the grid cell size has a large influence on the simulation run time,
especially when doing long term simulations. In order to create a long term simulation that can
run in a reasonable amount of time, grid cell sizes can be increased. This is in tension with the
importance of having cells small enough to capture important hydrologic details. Each model
will have unique conditions and a good balance between the needed detail and the available time
It is important that a stream network is dense enough to represent the actual channels. Not
only will this help resolve problems associated with digital dams but it will also help stabilize
areas of the model that produce extremely high and unrealistic erosion. The stream network can
9
have a large impact on the sediment discharge results. Generally, as the density of the stream
network increases, there will be less washload discharge and more sand outlet discharge. For
GSSHA currently offers five different equations as options for sediment transport capacity.
Of the five equations, the three that are currently ready for modeling are Kilinc-Richardson,
The Kilinc-Richardson equation is the most tested of the transport capacity equations.
Through extensive testing it has been determined that the Kilinc-Richardson equation should
only be used when calibration data are available. The Engelund-Hansen equation is also
recommended to be used with calibrated data. If calibrated data are unavailable the Engelund-
Hansen equation can still be used to get a rough estimate of the erosion. However, in most cases
the Slope and Unit Discharge method is recommend to be used if calibration data are not
available. According to the GSSHA developers, the Slope and Unit Discharge equation was able
to predict sediment runoff volume with +/- 200% of measured sediment volume for a research
watershed. However, the Slope and Unit Discharge method can be sensitive to the sediment
fractions in the soil erosion mapping table. If a GSSHA model with the Slope and Unit
Discharge equation results in unrealistic sediment discharge volumes then it may be a good idea
A sensitivity analysis was performed on the various sediment transport capacity equations
available using the Goodwin Creek long term calibrated sediment GSSHA model provided by
the developers of GSSHA. In the sensitivity analysis a calibrated model with the Kilinc-
10
Richardson equation is compared to uncalibrated Engelund-Hansen and Slope and Unit
Table 2-1: Total wash-load outlet discharge results with various transport capacity
equations for previously calibrated model.
Since the Kilinc-Richardson equation was used in calibration, the resulting wash-load
outlet discharge was used as a gage to determine the accuracy of the other transport capacity
equations. The washload outlet discharge for the Engelund-Hansen equation was significantly
different than that of the calibrated Kilinc-Richardson equation. This reinforces the suggestion
from the GSSHA developers that the Engelund-Hansen equation is usually only appropriate
when calibration data are available. The Slope and Unit Discharge methods provided much
closer washload discharge results. In fact, when taking into account the uncertainty involved in
sediment models, the magnitude of difference between the calibrated Kilinc-Richardson equation
and the uncalibrated Slope and Unit Discharge equation is surprisingly small. In regards to the
sand outlet discharge, the type of transport capacity equation used does not have much influence.
While some sand is transported through overland erosion, most sand is transported through
Further investigations were done to determine the influence that transport capacity
equations had on the spatial distributions of sediment erosion and deposition. Figure 2-2
11
compares the differences in spatial sediment erosion and deposition for a research sediment
transportation model. Note that red represents erosion and blue represents deposition.
Figure 2-2 Spatial distribution for the net sediment transfer for each transport capacity
equation.
While the magnitude of difference in net sediment transfer between the Engelund-Hansen
and Slope and Unit Discharge equations is significant (see Table 2-1) the spatial distributions as
seen in Figure 2-2 are similar. Both have well distributed erosion at steeper cells and deposition
at cells with mild slopes. However, the net sediment transfer for the calibrated Kilinc-Richardson
model is a little bit less distributed throughout the mode. The sediment erosion and deposition
that does occur seems to be focused in certain areas. Understanding the geomorphic dynamics of
the watershed of interest may help one determine which transport capacity equation is most
appropriate. It may even be appropriate to use each equation to see which sedimentation
12
2.3 Adjust Elevation Card
When choosing which transport capacity equation to use, GSSHA also allows one to use
the adjust elevation card (ADJUST_ELEV). This card tells GSSHA to output the new elevations
for each time step based on the sediment erosion and deposition (GSSHAwiki, 2009b). Including
this card increases the simulation run time. Often the results are not significantly different,
however if one is interested in examining the potential elevation changes, it may be worth the
The soil erosion mapping table in GSSHA is based off of the combination of land uses
and soil types. This means that a combination land use and soil type index map should be
Currently, GSSHA uses five parameters in its mapping tables for soil erosion plus the soil
composition proportions for each defined soil particle. The soil erosion parameters can be
grouped into three categories; detachment by surface runoff, detachment by raindrops, and
calibration. When the runoff and raindrop detachments summed together they account for all of
DETACH_CRIT which is label as “Critical rill detachment (Pa)” in the soil erosion map
table is the critical shear stress needed to detach the sediment (Downer 2010). It is
recommended that a value of 3.5 is used for this parameter (GSSHAwiki, 2012).
13
DETACH_ERODE, which is labeled as “Rill erodibility coefficient (s/m)” in the soil
erosion map table is an empirical coefficient used in the detachment by surface runoff equation.
While not much data is available for this parameter, it is recommended that this value be taken
from the WEPP model documentation which assigns a value of 0.004 for rangeland and a value
soil erosion map table is an empirical coefficient used in the detachment by surface runoff
equation. It is recommended that a value of 1 is used for this parameter (Flanagan and Nearing,
1995).
SPLASH_K, which is labeled as “Coefficient for detachment by rainfall (1/J)” in the soil
erosion map table is the product of three parameters used in the detachment capacity rate
equation (Downer 2010). The three components of the SPLASH_K parameter are ground cover
factor, the cover management factor, and the soil erodibility factor for detachment by raindrop
impact (Foster, 1982; Wicks and Bathurst, 1996). The soil erodibility factor for detachment by
raindrop impact can be determined by referencing the soil texture to Table 2-2. The ground cover
factor can be found by referencing a land use map, ground cover map, or an aerial photograph. It
is found by subtracting the fraction of ground covered by ground cover from 1.0 (GSSHAwiki,
2012). The cover-management factor can be found by referencing a land use map, canopy cover
map, or a recent aerial photograph. It is found by subtracting the fraction of canopy cover from
14
Table 2-2: Soil erodibility factor for the detachment by rainfall impact (Wicks and
Bathhurst, 1996).
map table is used to calibrate models. Previously this parameter functioned as a product of three
parameters which were the soil erodibility, cropping management factor, and conservation
practice factor. However, because of the difficulty accounting for unknowns, it has recently been
decided by the GSSHA developers that this parameters works better as a calibration dial for both
the Kilinc-Richardson and Engelund-Hansen equations. When using the Soil and Unit Discharge
equation, this parameter is internally hardwired into GSSHA and therefore entering a number in
the mapping table for this parameter will have no effect. As mentioned before, the Engelund-
Hansen equation can be used without calibration, although it is not recommended. In this
circumstance, the SED_K parameter would be given a value of 1. Further discussions about
15
2.4.4 Sediment Fractions
Below the SED_K parameter the sediment fractions for each of the define sediment types
needs to be entered. The total of the sediment fractions should add to 1.0 (GSSHAwiki, 2009b).
While all of the sediment transport equations are sensitive to these values, the Slope and Unit
Discharge equation can be especially sensitive to the sediment fractions. When using the Slope
and Unit Discharge transport capacity equation it may be best to start with generic values for the
sediment fractions if the measured sediment fractions are of questionable accuracy (see Section
3.10.1.).
Parameter sensitivities were performed for each of the relevant parameters for each of the
transport capacity equations on the Goodwin Creek research model. Because the
DETACH_CRIT and DETACH_EXP have recommended values, they were not included in the
parameter sensitivity analysis. For the DETACH_ERODE parameter a sensitivity analysis was
done for values 0.004 to 0.05 which are the values provided from the WEPP for rangeland and
cropland, respectively. For the SPLASH_K parameter, the sensitivity analysis spans a range of
commonly used values. For the SED_K parameter, a range of 0.01-1.0 was used to give a basic
understanding of the sensitivity of SED_K for calibration. The results for the washload discharge
can be seen in Table 2-3 and the results for the sand outlet discharge can be seen in Table 2-4. A
positive percent indicates that as the parameter increased so did the sediment discharge. A
negative percent indicates that as the parameter increased the sediment discharge decreased.
16
Table 2-3: Sensitivity analysis washload discharge results for soil erosion parameters.
Table 2-4: Sensitivity analysis sand discharge results for soil erosion parameters.
By understanding the sensitivities of each of the parameters one can better understand what
parameters need special attention and what parameters are of less significance. Besides the
DETACH_ERODE parameter for the Engelund-Hansen equation and the SED_K calibration
parameter, the parameters do not have a significant influence on the washload discharge,
especially when taking into account the high level of uncertainty associated with sediment
modeling. However, the DETACH_ERODE parameter had significant influence on the sand
outlet discharge. For the parameters that do not significantly affect the solution, a ballpark
estimate from the available sources is adequate. However, when using the DETACH_ERODE
parameter for the Engelund-Hansen equations or when one is specifically interested in the sand
outlet discharge it may be worth the effort to insure accuracy, especially if no calibration data
exists.
17
2.5 Channel Erosion Setup
GSSHA has the ability to model both overland and channel erosion. What has been
discussed so far has been in regards to the parameters for overland soil erosion. As previously
mentioned, only the particles the size of sand are eroded in the channels. The main parameter
involved in channel erosion is the maximum erosion depth (GSSHAwiki, 2009c) which greatly
affects the sand outlet discharge. A sensitivity analysis was performed on the max erosion depth
Dominican Republic. The results can be seen in Table 2-5 for a model with a low density stream
network.
While changes to the maximum erosion depth significantly changed the sand outlet
discharge, as expected there were no significant changes to the washload outlet discharge. If the
sand outlet discharge is of great importance in a model, then it is important that the maximum
erosion depth is accurately defined for each of the arcs in the stream network. However, if the
total volume of sediment discharge is important, then it may not be worth the effort to insure
accuracy in the maximum erosion depth if a low density stream network is used. This is due to
the small proportion of total sediment discharge that the sand composes in models that have low
density stream networks. For example, when the total sand outlet discharge in Table 2-5 is
18
compared to the total wash-load outlet discharge, the additional 50000 m3 of sand outlet
discharge for every meter of increased max erosion depth becomes insignificant. In fact the sand
outlet discharge is only about 1-2% of the washload outlet discharge. However, as the density of
the stream network increases, more and more sand is eroded from the channel bottom. As this
happens, the max channel erosion depth parameter becomes more and more influential in the
Besides the maximum erosion parameter, other parameters involved in channel erosion
are sediment porosity, water temperature, and sand size. A sensitivity analysis was performed to
see the importance of each parameter with the Goodwin Creek research GSSHA model with a
low stream network density. The results can be seen in Table 2-6.
The sediment porosity and temperature parameters had minimal impact on the sand outlet
discharge. The sand size had the greatest impact on the results. This is due to the methods that
GSSHA uses to determine what particles are bedload and what particles are washload. The
defined size of sand indicates the minimum particle size that is considered to be bedload
(Downer, 2010).
19
One of the constraints for channel erosion is that only the erosion of the sand in the bed is
modeled. This means that as the water rises during a flood event the there is no accounting for
the erosion of the fines in the channel bank. Bank erosion is often a major contributor to
sediment runoff. GSSHA may not produce accurate results for streams with unstable channel
banks. Local stream channel conditions should be considered on a case by case basis in order to
Often areas of extremely high erosion can occur in a model as seen in Figure 2-3. While
these areas may represent potential scour holes, often these erosion pits are the result of not using
a high enough stream density or not fixing digital dams. Before calibrating a model, it is good to
go through the initial run to see if any areas have unusually high erosion or deposition. If these
problematic areas exist where they should not, then digital dams should be fixed or stream
networks should be extended to the problematic areas. Since streams are neither a source or sink
for fines (GSSHAwiki, 2009c) then extending the stream network to the problematic area will
remove all sediment erosion and deposition except for sand. If the problem persists, then the
SED_K calibration parameter can be adjusted at the problematic area to decreases the erosion.
Figure 2-3: Example area of high erosion (blue is deposition and red is erosion).
20
All this should be done with consideration of the geomorphic conditions of the watershed
being modeled. Sometimes the net sediment erosion and deposition evens out in these
problematic areas, meaning that most of the eroded soil is deposited in a nearby cell. If one is
only interested in the total sediment discharge for a watershed and if the areas of extremely high
erosion/deposition cancel themselves out, then these problematic areas can be ignored with
reservations. However, while the erosion and deposition may even out in one land use situation,
it does not mean that it will even out in other land use scenarios. A vigilant eye should always be
looking for these problematic areas because they can easily be a source for error in a model.
Since there is often a great deal of uncertainty associated with modeling sediment
transportation, the importance of calibration cannot be stressed enough. Sediment runoff data
exists in a variety of formats, however, when calibrating sediment runoff in GSSHA the most
common target is the total sediment discharge. The total sediment discharge is the sum of the
total sand outlet discharge and the total wash-load outlet discharge, which can both be found in
If a low density stream network is used then the sand outlet discharge often only makes up
a very small percent of the total sediment discharge. In those cases, the model can be calibrated
to the washload outlet discharge only. As mentioned before, the SED_K parameter is used as a
dial when calibrating the sediment transport model. Additionally, the max stream bed erosion
depth can be used to further calibrate a model for sand outlet discharge. Currently, WMS does
21
2.7.1 Using Empirical Equations or Bathymetric Surveys in Calibration
Often measured sediment data is difficult to obtain. When measured sediment flow data
is unavailable, it has been suggested that yields from empirical equations, such as the USLE, be
used as the target during calibration. Then once the base model has been calibrated different
scenarios can be modeled with GSSHA. This would allow GSSHA to show the relative effect in
reservoirs are surveyed to determine how much of the capacity has been lost to sedimentation.
This sediment volume may be used for calibration but it is important to know how much of the
sediment settles in the reservoir and how much passes through. If sand is primarily the only
sediment that settles in the reservoir then only the sand outlet discharge may be calibrated to the
As discussed before, the overland erosion and resulting sediment flow often takes up to a
few months to stabilize. One problem with using the total sediment discharge from the summary
file during calibration is that it includes the sediment discharge that occurred before the sediment
discharge was stabilized. For example, a yearlong simulation may not see sediment discharge for
three months and then it may take another month for it to ramp up to a stable condition. In that
case the total sediment discharge from the summary file would include the sediment runoff for
four months of unstabilized sediment flow. To fix that problem, the total sediment discharge for
only the stable portion of the model can be extracted from the output sedograph. The output
sedograph includes sediment discharge time series data for sediment that originated from
overland erosion for each defined sediment type. It also includes time series data for the sand
22
outlet discharge that initially came from channel erosion. This sediment flow data can be
multiplied out for a discrete period of time to determine the volume of the total sediment
discharge. This resulting total sediment discharge volume can then be temporally normalized for
the stable period of sediment discharge and then compared to measured data for calibration.
When calibrating with the SED_K parameter it can be difficult to know the initial SED_K
values. Because different land uses and soils erode at different rates, the SED_K value should be
different for each of the land use and soil type combinations. A rough estimate for starting
SED_K values to be used in calibration can be determined by finding the product of three
parameters: soil erodibility factor (K), cropping management factor (C), and conservation
practice factor (P). The value of the soil erodibility factor (K) is dependent on the soil texture and
it can be determined by referencing Table 2-7. The value of the cropping management factor (C)
is dependent on land use and it can be determined by referencing Table 2-8. The value of the
conservation practice factor (P) is dependent on land use and it can be determined by referencing
Table 2-9.
Once the three factors are multiplied together, the resulting values can be used as a starting
point for the SED_K calibration dial. At first, when adjusting the SED_K parameters to get a
rough calibration, it is important to maintain the proportional differences between the SED_K
values for the different land use and soil type combinations. Then, once the model results are
close to the measured total sediment discharge, the SED_K values for individual land use and
soil types combinations can be adjusted to match observed conditions. This can be done by
examining the spatial erosion of certain land uses and soil types and adjusting the corresponding
SED_K values as needed until the desired erosion or deposition depth is reached.
23
Table 2-7: Soil erodibility factor (K) (Wanielista, 1978).
24
2.8 Verification
Because of the nature of sediment modeling, verification data is often difficult to obtain.
If the data are available, a common method of verification is to run the calibrated sediment
model for a different period of time to see how the model reacts to different precipitation events.
The resulting sediment runoff can then be compared to measured data to help develop an insight
as to how robust the sediment model is for simulating different periods of time. Anecdotal
evidence from those who have developed GSSHA sediment models suggest that models
calibrated for one set of rainfall events do not always produce accurate results for the same
watershed with a different set of rainfall events. It is recommended that, if possible, calibration
can affect erosion. One of the benefits of modeling sediment transport with GSSHA in WMS is
that the model can easily be modified to account for changes in land use. Furthermore, unlike
popular empirical equations like the USLE, GSSHA can account for spatial variations in land
use.
After developing a calibrated sediment transport base model in GSSHA, one can modify
the existing index maps and mapping tables to account for changes in land use. If the land use for
an area of the model is modified to a type of land use that was already in the model when the
model was calibrated, then no new soil parameters will need to be defined. In that case, only the
index map needs to be updated. However, if a new type of land use is introduced to the model
when modifying land uses, then new index maps and mapping tables will need to be developed
25
to account for the changes. The soil erosion parameters can then be entered in for the new land
The soil erosion parameters that change based on land use or soil type alterations are the
and SPLASH_K parameters can be developed from available literature. However, alterations to
the SED_K parameter based on changes to land use are often difficult because the SED_K
parameter is developed through calibration. One possible way to estimate the SED_K value for
uncalibrated land uses is to interpolate from calibrated SED_K values in the model.
Along with modifying the soil erosion mapping table, it is important that the roughness
mapping table is updated as well. One of the most sensitive parameters on soil erosion is shear
stress. This is calculated from the overland flow velocity which is a function of the overland
roughness. Like the SED_K value, the roughness value is often used as a calibration dial when
calibrating runoff. This often results in the roughness values having unrealistically high values. If
this is the case, then the appropriate proportionality in roughness should be maintained for the
new land uses based on roughness tables in the GSSHA wiki (see APPENDIX).
Another factor to consider when modeling land use changes is whether the new land use
is connected to a channel. If the new land use is not connected to a stream arc, then the
influences of the land use change can be dampened. Once the fines from overland erosion enter a
stream, they cannot be deposited and are ultimately discharged from the watershed.
Models that use the Slope and Unit Discharge equation are often not good predictors of
how land use changes affect sediment discharge. This is likely due to the fact that most
parameters are hard wired into the Slope and Unit Discharge equation. The Kilinc-Richardson
26
and Engelund-Hansen equations are more robust at modeling land use changes because of their
27
28
3 AGUACATE CASE STUDY
3.1 Background
Watersheds in the Dominican Republic are severely affected by sediment erosion and
deposition in reservoirs and in river beds downstream. This is more pronounced due to the
frequent occurrence of hurricanes, tropical storms, and landslides. One single extreme event such
as a category five hurricane like David in 1979 or George in 1998 is capable of exceeding the
average sediment yield predicted by empirical methods like USLE, RUSLE, and MUSLE.
Bathymetric surveys in the major dams reveal that many have volumes of trapped sediments
either close to or above the dead storage. The consequent reduction of the useful life of these
dams stresses the importance of evaluating empirical methods and their predictive accuracy.
The Aguacate reservoir located in central Dominican Republic was constructed in 1992. In
a short period of time, the reservoir capacity dramatically decreased due to sedimentation.
reservoirs in the Dominican Republic (Ramirez et al., 2003). The Aguacate reservoir as seen in
Figure 3-1 is fed by two main tributaries, one from the north and one from the east. The north
tributary is fed by a dam located a few miles upstream. Because the north tributary passes
through a reservoir, it is assumed that it is not a significant source for sediment. A GSSHA
sediment model was developed for the east tributary (the Mahomita River) to help gain some
understanding as to why the Aguacate reservoir filled up with sediment at such a high rate.
29
Figure 3-1: Aguacate dam and reservoir.
WMS was used to create a GSSHA model for the Mahomita River in the Aguacate
Watershed. The catchment area for the watershed is approximately 113 km2. A spatially
distributed GSSHA model having a cell resolution of 200 meters was developed to simulate the
hydrologic behavior of the watershed (see Figure 3-2). This cell size is greater than the
recommended maximum cell size of 150 meters (Rojas, 2008). However the Aguacate watershed
size is substantially greater than the 20 km2 watershed used in the cell size study. With a large
watershed being modeled, a 200 meter cell sizes seemed to sufficiently capture the needed
topographic detail and also allowed the model to run in a reasonable amount of time. While
smaller grid cells may have been preferable, time constraints for this project limited the available
simulation run time. Land cover and soil shapefiles were used to derive the surface
30
characteristics and parameters for the model. Four precipitation stations Novillo, Palo Caja,
Valdesia and Recodo are used to define the precipitation data. The inverse distance weighted
method was used to interpolate the precipitation data across the grid cells. The simulation was
run for a one year period starting 1st Jan, 1983 01:00AM to 31st December, 1983 01:00AM.
Figure 3-2: GSSHA finite difference grid for the Aguacate watershed.
Before the model was used in analyzing advanced watershed processes such as sediment
transport and land use change scenario modeling the hydrologic processes have to be calibrated
with the best available data. Since, the model is required to forecast flow in multiple
precipitation event scenarios, it was necessary that the model be calibrated in a continuous
simulation mode. Similarly, selection of a proper set of calibration parameters should be done so
For hydrodynamic calibrations, GSSHA has an automated tool which uses the Shuffle
Complex Evolution (SCE) algorithm to generate parameters. Several manual calibration runs
31
were done before moving on to the automated calibration, which led to an understanding of
sensitivity of the model to different components and parameters. The automated calibration was
After the Aguacate GSSHA model was calibrated to the measured runoff, the sediment
model was then initialized in the GSSHA Job Control by enabling Soil Erosion. When
initializing soil erosion, GSSHA requires a listing of the sediments that will be modeled and their
associated properties as seen in Figure 3-3. Depending on the model, the soil diameter
parameters can be fairly sensitive so it is worth effort to ensure accuracy. To save time when
running the model, the ADJUST_ELEV card was not enabled. To save time when running the
While in the Overland soil erosion window, the transport capacity equation was selected.
For this case study, two different models were developed. One model was calibrated with the
Kilinc-Richardson transport capacity equation. The other model was not calibrated and it used
the Slope and Unit Discharge equation. Two transport capacity equations were used because the
available measured data for calibration is of questionable accuracy. The Slope and Unit
Discharge equation was used to give us another point of reference and hopefully give us an
32
Figure 3-3: GSSHA Overland soil erosion properties window.
With the soil erosion enabled in GSSHA, the soil erosion mapping table was set up. Land
use and soil type index maps were used to generate the IDs for the soil erosion mapping table.
Besides the SED_K parameter in the soil erosion mapping table, the parameters used for the
Kilinc-Richardson equation and Slope and Unit Discharge equation can be the same. The soil
fractional compositions for each defined sediment type were entered into the soil erosion
mapping table. Using the methods discussed in Section 2.4 for the soil erosion parameters and
Section 2.7 for the erodibility coefficient (SED_K), the rest of the soil erosion mapping table was
completed as seen in Figure 3-4. The erodibility coefficient (SED_K) was set to 1 for the soil
erosion mapping table used in the models that uses the Slope and Unit Discharge equation.
33
Figure 3-4: Soil erosion mapping table Aguacate model.
Next, the channel erosion parameters were setup. The channel erosion parameters are in
the GSSHA Channel Routing Parameters dialog (see Figure 3-5), which can be accessed by
clicking the Edit Parameters… button in the Channel routing computation scheme section of the
GSSHA Job Control window. A sediment porosity of 0.4 and a water temperature of 20 degrees
Celsius were used. As mentioned in Section 2.5, these two parameters are not of much
consequence. However, the sand size can drastically affect the results. We will use a value of
0.45 mm, which means that anything smaller than 0.45 mm will be treated as washload and that
all of the sand in the stream bed has a size of 0.45 mm.
34
Figure 3-5: GSSHA channel routing parameters.
The maximum erosion depth parameter for channel erosion can be entered for each stream
arc. A depth of 0.5 meters was input for the main channel and a depth of 0.2 meters was input for
the branches that feed into the main channel. The maximum erosion depth for the channel was
entered into the stream arc properties window as seen in Figure 3-6. Later on, these values can be
35
Figure 3-6: Max erosion parameter in the arc properties window.
erosion. Running GSSHA with the uncalibrated initial parameters revealed a few areas of
extremely high erosion and deposition as seen in Figure 3-7. All of the problematic areas occur
at locations where channels likely exist, which means that they could be fixed by extending the
stream network to the areas. By examining the amount of sediment that is eroded and deposited
at each location it was determined that there was not a significant amount of net sediment
transferred from these locations. Most of the eroded soil was deposited in a nearby cell. Were
this case study to be used for purposes besides demonstration, we would go through the extra
effort to extend the stream network to the problematic areas. However, for the sake of this study,
we were satisfied with the minimal influence the areas of high erosion and deposition have and
36
Figure 3-7: Areas of high erosion (red) and deposition (blue).
The Aguacate model using the Slope and Unit Discharge transport capacity equation was
ready to run at this point. However, before that was done, the model using the Kilinc-Richardson
transport capacity equation was calibrated. Because of the lack of availability of measured
sediment data for the Aguacate watershed, the target data that was used for calibration was the
sediment yield from the USLE equation for the Aguacate watershed. While measured sediment
data are preferred as the target when calibrating, reliable measured sediment data is often
difficult to find. This will then allow GSSHA to be a tool for more detailed design where
practices like land use change or extreme event analysis can be analyzed. The analysis of the
Aguacate watershed was part of a larger project comparing the GSSHA sediment results to the
37
USLE results and determining how they can be used in tandem for design purposes. The USLE
results for the nearby watersheds Sabana Yegua and Sabaneta were 33,416 tons/km2/year and
33,351 tons/km2/year, respectively. The USLE result for the Aguacate was 10,000 tons/km2/year.
For the case study of interest the USLE yield for the Aguacate watershed was used. Once the
model is calibrated to the yield, then different land use and rainfall scenarios will be modeled to
The output total sediment discharge needed to be temporally and spatially normalized so
that it could be compared with the calibration data. When determining the amount of time that
should be used to normalize the sediment discharge, the amount of stabilized time of sediment
discharge is usually used. The period of stable sediment flow can be determined by examining
the output sedograph from the initial uncalibrated Kilinc-Richardson run as seen in Figure 3-8.
While there is some sediment transport early in the models, most of it is from channel
erosion and not overland erosion. As mentioned before, it often takes a while for the overland
erosion to ramp up in a model. From Figure 3-9 it looks like the model was still building up to
equilibrium in the first few months of the simulation. However, by examining the input
precipitation data, there is not much precipitation during the first few months of the simulation
which would also cause a decrease in overland erosion. It is difficult to tell if the lower sediment
discharge in the first few months is a result of lower rainfall or because the model is building up
to equilibrium or some combination of both. However, since the early months in the simulation
have substantially less precipitation which would result in very little overland erosion anyway,
for the purpose of this study we assumed that the model ramp up time is neglectable. That means
that we did not have to extract the volume of the sediment discharge from the output sedograph
38
time series data for a subset of the simulated time. Instead we just used the sum of the total
washload and sand outlet discharge from the GSSHA summary file when comparing the
10
0.1
Total Sediment Discharge (m3/s)
0.01
0.001
0.0001
Outlet Discharge of Silt from Overland Erosion
Outlet Discharge of Clay from Overland Erosion
0.00001
Outlet Discharge of Sand from Overland Erosion
Outlet Discharge of Sand from Channel Erosion
0.000001
1/1 2/20 4/10 5/30 7/19 9/7 10/27 12/16
Date
With a series of iterative steps the model was calibrated to the target total sediment
discharge within an allowable tolerance. The allowable tolerance for sediment calibration
depends on the uncertainty associated with the model and the available time for calibration. For
39
the Aguacate model using the Kilinc-Richardson equation it was decided that an error tolerance
of 5% would be used. This took five runs and about six hours of simulation time. For each run all
of the SED_K values were multiplied by the same factor until the total sediment discharge was
within the allowable tolerance. The normalized total sediment discharge was calculated by
summing the total washload discharge and the sand discharge in the GSSHA summary file and
then normalizing the results for the watershed area and stable simulated time. The normalized
results for each run can be seen in Table 3-1: Calibration run results.
available or to adjust the sand outlet discharge. The sand outlet discharge is calibrated by
adjusting the channel max erosion depth. For the local erosion/deposition data GSSHA can
export an erosion map that shows the total deposition and erosion at each grid cell for each time
step. If the depth of erosion/deposition is known at a certain location and time, then the SED_K
parameters for individual land use and soil type combinations at the location of known
erosion/deposition can be adjusted in the mapping table. This can be done in increments until the
desired erosion/deposition is reached for the location of interest. Since this data is not available
for the Aguacate model, this additional calibration step was not performed.
40
3.10 Results
The spatially and temporally normalized results for the calibrated Kilinc-Richardson
(KR) and uncalibrated Slope and Unit Discharge (SUD) models can be seen in Table 3-2.
The model that used Slope and Unit Discharge transport capacity equation produced
about fifty times as much sediment discharge as the calibrated Kilinc-Richardson model. Usually
the Slope and Unit Discharge equation is good at predicting the sediment discharge within +/-
200%; however, for the Aguacate model it drastically over predicts the sediment discharge. This
shows the importance of recognizing unrealistic solutions and not completely relying on the
Slope on Unit Discharge equation when modeling sediment discharge. To get an understanding
of the time variation in sediment discharge in each of the GSSHA models, a total sediment
discharge sedograph was developed from the GSSHA results as seen in Figure 3-9.
The sedograph shows that both equations react similarly to the rainfall events. However,
the Slope and Unit Discharge Equation ramps up to a much higher level of sediment discharge
41
100
SUD
10
Total Sediment Discharge (m3/s)
KR
0.1
0.01
0.001
0.0001
1-Jan 20-Feb 10-Apr 30-May 19-Jul 7-Sep 27-Oct 16-Dec
Date
Along with temporal sediment discharge output, GSSHA can also output spatial
distributions of the erosion and deposition as seen in Figure 3-10. The Kilinc-Richardson
GSSHA model seems to produce spots of extremely high erosion, whereas the Slope and Unit
Discharge GSSHA model has more distributed erosion and deposition. Furthermore, the Slope
and Unit Discharge GSSHA model has a great amount of deposition in the channels.
At this point, if additional precipitation and measured sediment flow data were available
42
Kilinc-Richardson Slope and Unit Discharge
Figure 3-10: Spatial distribution of erosion and deposition with both GSSHA models.
3.10.1 Rerunning with the Slope and Unit Discharge and Generic Sediment Fractions
As mentioned in Section 2.4.4, the Slope and Unit Discharge equation is highly sensitive
to the sediment fractions from the soil erosion mapping table. Since the sediment fractions in the
soil erosion mapping table for the Aguacate model were of questionable accuracy because of
limited measured data, the mapping table was modified to have more generic values. After
running GSSHA with the modified mapping table the resulting total sediment discharge was
75,415 tons/km2/year which was 689% higher than the calculated USLE yield. While it is still
43
high, it is much more reasonable than the 534,463 tons/km2/year produced via the measured soil
fractions.
Using the calibrated Kilinc-Richardson model various land use changes were modeled to
explorer the capabilities that GSSHA has for modeling spatial variations in land use.
About half of the Aguacate watershed consists of undeveloped forests and grasslands.
The other half is agricultural which is primarily used for coffee production. One of the valuable
capabilities of GSSHA for the Aguacate watershed is that the effects of deforestation can be
modeled. Using the GSSHA model developed with the Kilinc-Richardson and Slope and Unit
Discharge transport capacity equations, two potential land use changes were simulated. When
modeling the different land use scenarios in the Aguacate watershed the Slope and Unit
Discharge equation did not produce significantly different results. Many of the soil erosion
parameters are hardwired into the Slope and Unit Discharge transport capacity equation, which
means that the Slope and Unit Discharge equation can be inaccurate when predicting the effects
of land use changes. While the Slope and Unit Discharge equation can be sensitive to the
fractional soil composition in the soil erosion mapping table, the land use changes did not require
changes in the soil composition. Because of these problems that the Slope and Unit Discharge
equation has with the Aguacate watershed, this case study will focus on the land use change
The first land use change scenario that was modeled was a complete deforestation of 11.2
km2 in the northeast portion of the watershed, which is about 10% of the total watershed area.
44
For the next land use change scenario the same area would be converted to coffee crop land use.
This common process of land cultivation leaves the land at risk of high amounts of erosion
First, the land use index map and the land use/soil type combination index map were
duplicated. The duplicated index maps were then modified to account for the new land use
scenarios. This was done by assigning an index value of 999 to the cells that overlap the
modified land use in both the land use index map and land use/soil type combination index map
for the deforestation model. The existing and changed land use index maps can be seen in Figure
3-11. Most of the altered land use was forest (red) and grassland (light blue). Because there are
coffee crops in the calibrated base sediment model, the land use IDs for the altered land use area
in the coffee crop scenario model can be given the coffee crop land use ID. The same can be
done for the land use/soil type combination index map in the coffee crop model.
Following the modification of the index maps, the mapping tables for both soil erosion
and roughness were updated to account for the modified index maps in the deforestation model.
The roughness mapping table was changed to reference the modified land use index map and the
soil erosion mapping table was changed to reference the modified combination index map.
To determine the roughness value that should be used for the deforestation scenario, the
suggested roughness value table from the GSSHA wiki was referenced (see APPENDIX).
Sometimes the roughness values in the mapping table are used to calibrate the runoff in a
GSSHA model. This can make the roughness values artificially high as natural storage and other
factors are represented with this parameter. As a result, they may no longer match the suggested
roughness values table in the APPENDIX. In this case it would not make sense to pick a value
from the suggested roughness value table that matches the new land use. However, one can
45
maintain the appropriate proportions as found in the suggested roughness value table when
entering in new roughness values. In this case, the calibrated roughness values are still within
ranges of the suggested roughness values found in the table. Because of that we could use the
roughness value found in the table without further modifications to maintain correct
proportionality. For the deforestation scenario, a bare clay-loam land use was used with the
Altered Area
Figure 3-11: Land use index maps before and after alterations.
To determine the roughness value that should be used for the deforestation scenario, the
suggested roughness value table from the GSSHA wiki was referenced (see APPENDIX).
Sometimes the roughness values in the mapping table are used to calibrate the runoff in a
46
GSSHA model. This can make the roughness values artificially high as natural storage and other
factors are represented with this parameter. As a result, they may no longer match the suggested
roughness values table in the APPENDIX. In this case it would not make sense to pick a value
from the suggested roughness value table that matches the new land use. However, one can
maintain the appropriate proportions as found in the suggested roughness value table when
entering in new roughness values. In this case, the calibrated roughness values are still within
ranges of the suggested roughness values found in the table. Because of that we could use the
roughness value found in the table without further modifications to maintain correct
proportionality. For the deforestation scenario, a bare clay-loam land use was used with the
Because the coffee crop scenario did not introduce any new land use type, the roughness
For the deforestation scenario, the soil erosion mapping table will need to be updated for
the new bare earth on clay soil land use/soil type combination index map. To do this, new values
for SPLASH_K, DETACH_ERODE, and SED_K were developed. While the SPLASH_K and
DETACH_ERODE parameters can be developed with the methods suggested in Section 2.4,
developing the SED_K parameter was more involved. Before sediment calibration the SED_K
parameters were developed by the process discussed in Section 2.7. However, since then the
SED_K parameters have been altered through calibration, the deforestation SED_K factor needs
to be multiplied by the same calibration adjustment factor. The SED_K parameters were
ultimately multiplied by 0.0222 of their original values through the calibration process as seen in
Table 3-1. The new deforested SED_K value for bare earth on clay soil which was developed via
47
the process discussed in Section 2.6 was then multiplied by 0.0222. The resulting new soil
erosion parameters for the bare earth on clay soil are in Figure 3-12.
Because the coffee crop on clay soil land use/soil type combination already exists in the
soil erosion mapping table, no alterations needed to be done to the soil erosion mapping table in
Both modified land use models were then saved and run. The sediment discharge results
While the effects of deforestation increased the total runoff by 38%, the coffee crop
model showed that once the deforested area was agriculturally developed the rate of erosion
would be 18% less than what it was before the deforestation due to erosion prevention methods
48
Table 3-3: Sediment discharge results for land use modification scenarios
The spatial variation of erosion for the area that underwent the land use changes can be
seen in Figure 3-13. The additional erosion in the deforestation scenario is distinct. The existing
condition compared to the coffee crop scenario shows very minor changes.
Net Sediment
Transfer
(m3/grid cell)
Figure 3-13: Spatial variations of net sediment transfer for different LU scenarios.
49
3.11.2 Connecting Land Use Changes to a Stream Arc
If new land uses are not connected to stream arcs then the affects that the land use
changes have on the watershed can be dampened out. To understand the sensitivity that the
model has with hydraulically connecting a land use to a stream arc, two different scenarios were
modeled. First, 5% of the watershed which was connected to a stream arc were given a set of
highly erodible parameters. Second, 5% of the roughly the same area of the watershed but was
not connected to a stream arc was given the same highly erodible parameters. The two different
land use change scenarios can be seen in Figure 3-14. The total sediment discharge results can be
seen in Table 3-4and the spatial variation of results can be seen in Figure 3-15.
Connected Disconnect
Land Use Land Use
Figure 3-14: Connected to stream arc and disconnected to stream LU index maps.
50
Table 3-4: Change in sediment discharge for connected and disconnected scenarios.
Existing
Land Use
Net Sediment
Transfer
(m3/grid cell)
Disconnected
Land Use
Connected
Land Use
51
The changes in total sediment discharge suggest that cells that are disconnected from
stream arcs have a dampened influence on the total sediment discharge. Furthermore, from
looking at the connected land use in Figure 3-15, it can be seen that most of the erosion and
deposition occurs in the cells adjacent to the stream arc. This could be due to the steeper slopes
that occur next to the channel. For the Aguacate model, this means that were the new land uses
are place in regards to the stream arcs can have a large affected on the sediment discharge
results.
One of the benefits of using GSSHA for sediment modeling is the ability to model
different erosion control practices to see how they affect the sediment discharge. One way to
model an erosion practice is to create a land use that uses a buffer zone with low edibility near
the channels. One concern with the Aguacate watershed is that there is a substantial amount of
erosion that occurs near the channels. While GSSHA does not model channel bank erosion, one
can decrease the erodibility of the cells next to the channel to simulate more stable conditions for
the area nearby the channel. For the Aguacate watershed, the buffer was placed around 60% of
the main stream for the watershed. The buffer zone land use index map can be seen in Figure
3-16. The soil erosion and roughness mapping tables were updated for the new index map ID. It
was assumed that the buffer zone would have a low erodibility factor similar to that of a dense
forest.
The resulting total sediment discharge of 3089 tons/km2/year was 68% lower than the
original total sediment discharge which shows the effectiveness of the modeled buffer zone.
52
Figure 3-16: Stream buffer zone land use index map.
In September of 1979 Hurricane David, a category five hurricane, passed through the
Dominican Republic. Because of the torrential rain, many areas in the Dominican Republic
experienced the highest flows on record. These high flows resulted in an increase in sediment
flow and caused many landslides. When developing a dam and reservoir in the Dominican
Republic are of interest because the high sediment flows caused by hurricanes can significantly
affect the storage of a reservoir. Using the Aguacate GSSHA model that was calibrated for
53
To set up the Aguacate GSSHA model for the hurricane simulation two input files had to
be modified. First, the gage input file that stores the precipitation data was changed. Instead of
modeling an entire year for the hurricane simulation, only 50 days were included. The 50 days
include about 40 days before the hurricane and 10 days during and following the hurricane. The
40 days before the hurricane were used to ramp up the model to equilibrium. For the 40 days
preceding the hurricane, measured precipitation values were used in the gage file. However,
during and after the hurricane no precipitation data were available. Instead of using gage
measured precipitation values a SCS Type II 48-hour storm was used to model the hurricane. A
depth value of 300 mm, which was interpolated from a 48-hour Hurricane David depth contour
map, was used for the SCS storm. Once the gage file was updated with this information, the
Next the HMET data were updated. To do this the data was modified from its original
length of one full year in 1983 to only 50 days in 1979 spanning from early August to late
September. Because the simulation spans a much shorter time period, the output hydrograph time
step was then decreased to create a higher temporal resolution hydrograph. The model was then
run.
No discharge data existed to calibrate the model due to problems with the flow gage
during the hurricane. However, the resulting peak flow was about 800 cms which seemed within
reason for an extreme storm over a 113 km2 watershed. A side by side comparison of the net
sediment transfer can be seen in Figure 3-17. The spatially normalized sediment discharge along
with the 1983 spatially and temporally normalized sediment discharge can be seen in Table 3-5.
54
Net Sediment
Transfer
(m3/grid cell) 1983 Hurricane David
Figure 3-17: Comparison of the net sed. transfer for hurricane and 1983 models.
Table 3-5: Normalized sediment discharges for hurricane and 1983 models.
Hurricane David
2
Total wash-load outlet discharge 2708 ton/km
2
Total sand outlet discharge 1782 ton/km
2
Total sediment discharge 4491 ton/km
1983
2
Total sediment discharge 9557 ton/km /year
From examining the spatial erosion and deposition, it looks like the deposition is more
distributed for the hurricane model. The normalized sediment discharge that was modeled for
Hurricane David was 47% of the normalized yearly sediment discharge. While 1983 was a
slightly above average year for precipitation, we can estimate that the sediment runoff from a
large hurricane is about the same as 5-6 months of sediment runoff during a normal water year.
55
56
4 CONCLUSION
The major constraints when using sediment transport in GSSHA include the availability
of sediment calibration data, the amount of simulation time, the accuracy of soil composition
data, the accuracy of particle size data, and the inability of GSSHA to model channel bank
erosion. Furthermore, a calibrated long term GSSHA hydrologic model needs to be constructed
1) Develop a calibrated long term hydrologic GSSHA model with a cell size small
calibration data and the intent of the model. The Kilinc-Richardson equation is best if
sediment calibration data are available and if different land use scenarios will be
modeled. The Slope and Unit Discharge equation is best when no sediment
4) Using parameter estimate guidelines from the literature set up the soil erosion
mapping table. Many parameters do not have much effect on the sediment discharge.
57
discharge. The fractional soil composition in the soil erosion mapping table has a
substantial influence on the washload outlet discharge, especially for models that use
5) Define the channel erosion parameters. The two most important parts for channel
erosion include the sand size and maximum channel erosion depth.
6) Run the model with the initial conditions and look for areas of unusually high
erosion. The problematic area can often be fixed by extending the stream network to
7) For calibration purposes, determine how long it takes the sediment model to ramp up
to equilibrium conditions. Use the sediment discharge volume for the stable period
SED_K parameter for the overland erosion and/or the channel max depth parameter
9) If measured sediment and precipitation data are available for a different period of
time then the model can be run with new precipitation data to see how well the
sediment results match up with the measured sediment data for verification purposes.
10) Once the model is calibrated, various scenarios can be modeled to see how the
events. It is important to remember that cells that are adjacent to cells that contain a
stream arc have greater influence on the sediment discharge results then cells that are
58
The sediment transport modeling capabilities of GSSHA, while still in development, can
be a valuable tool when determining the watershed erosion and sediment runoff. Of particular
interested is GSSHA’s ability to model spatial variations in land use and soil type, and both
spatial and temporal variations in precipitation. Generally there is too much uncertainty involved
in developing all of the physical parameters for the GSSHA sediment transport model to not be
used with calibration data. While there is often a lack of credible observed data the USLE
equations are not a bad way to estimate sediment loads for dam design considerations. In these
situations GSSHA could be a good tool for a more detailed design where land use changes or
59
60
REFERENCES
Downer, C. W., Ogden, F. L., Martin, W., and Harmon, R. S. (2002a). “Theory, Development,
and Applicability of the Surface Water Hydrologic Model CASC2D,” Hydrol. Pro. 16,
255-275.
Downer, C.W. (2010). “Improved Soil Erosion and Sediment Transportation in GSSHA.” ERDC
SWWRP Technical Note.
Flanagan, D. C., and M. A. Nearing. (1995). “USDA Water Erosion Prediction Project Hillslope
Profile and Watershed Model Document.” NSERL Report No. 11. USDA-ARS National Soil
Erosion Research Laboratory.
GSSHA Wiki. (2009c) “Soil Erosion: Channel Sediment Transportation Formula.” USACE.
http://gsshawiki.com/gssha/Soil_Erosion:Channel_Sediment_Transport_Formulation
(Access June 18, 2012)
GSSHA Wiki. (2011) “Surface Water Routing: Overland Flow Routing.” USACE.
http://gsshawiki.com/gssha/Surface_Water_Routing:Overland_Flow_Routing (Access
June 18, 2012)
61
GSSHA Wiki. (2012) “Soil Erosion: Simulations with Soil Erosion.” USACE.
http://gsshawiki.com/gssha/Soil_Erosion:Simulations_with_Soil_Erosion (Access June
18, 2012)
Hunter, S., B. Vieux, F. Ogden, J. Niedzialek, C. Downer, J. Addeigo, J. Daraio, (2003). “A Test
of Two Distributed Hydrologic Models with WSR-88D Radar Precipitation Data Input in
Arizona, Proc. 31st International Conf. on Radar Meteorology”, Am. Met. Soc., 6(12)
Julien, P. Y. (1995). Erosion and Sedimentation, Cambridge University Press, 280 pp.
Johnson, B. E. (2000). “Development of a Particulate Transport Algorithm within the Two
Dimensional Rainfall-Runoff Model (CASC2D).” Vicksburg, MS: U.S. Army Engineer
Research and Development Center.
Kalin, L., and M. M. Hantush. (2006). “Comparative Assessment of Two Distributed Watershed
Models with Application to a Small Watershed.” Hydrol. Proc. 20(11), 2285-2307.
Kilinc, M., and E. V. Richardson. (1973). “Mechanics of Soil Erosion from Overland Flow
Generated by Simulated Rainfall: Hydrology Paper No. 63”, Colorado State University
Ogden, F., and A. Heilig (2001). “Two-Dimensional Watershed-Scale Erosion Modeling with
CASC2D in Landscape Erosion and evolution Modeling.” Landscape Erosion and
Evolution Modeling, Springer.
Rojas, R. et al. , (2008). “Grid Scale Effect on Watershed Soil Erosion Models.” Journal of
Hydrologic Engineering, 13 (9), 793-802.
Wanielista, M.P. (1978). Stormwater Management: Quantity and Quality, Ann Arbor,
Michigan: Ann Arbor Science, Inc., 412 pp.
Wicks, J. M., and J. C. Bathurst. (1996). “SHESED: A Physically Based, Distributed Erosion
and Sediment Component for the SHE Hydrological Modeling System.” J. Hydrol. 175, 213-
238.
Yang et al. (2006). Erosion and Sediment Modeling, Denver Colorado, Denver, CO: Bureau of
Reclamation
62
APPENDIX A. TABLES
63