FSI-10-TN85

Modeling Sediment Erosion and Deposition with the FLOW-3D

Sedimentation & Scour Model
James Brethour and Jeff Burnham
May 2010

The FLOW-3D 1 sedimentation and scour model has been revamped in Version 9.4, with
new capabilities and approaches. This technical note outlines the new changes to the
model, describes the important equations and approaches, outlines how to use the model
and set up a simulation, and closes with two test cases used to validate the current model.
Overview of the Model
In FLOW-3D Version 8.0, a sediment scour model was introduced to predict the erosion
and deposition of sediment. This model works by emulating both the entrainment of
sediment at the packed bed interface and the drifting and deposition of sediment due to
gravity. When coupled with the three-dimensional fluid dynamics computed by the
equations of mass and momentum conservation, the model is able to simulate the
deposition and entrainment of sand, silt and other non-cohesive sediment. Cohesive
sediments pose additional challenges and are not explicitly considered by the model.
The sedimentation and scour model in FLOW-3D Version 9.4 builds upon the original
model in a number of ways:

Includes non-linear drifting of the suspended phase. This allows more accurate
simulation of the deposition of larger and/or faster drifting sediment.
Includes multiple sediment species, with a maximum of 10 sediment species. This
allows the simultaneous simulation of a system with gravel, coarse sand, sand and
silt, as an example.
Includes a customizable hiding/exposure function to account for varying
entrainment rates of sediment species in mixtures as a function of particle size
relative to the mean.
Includes a bed-load transport model. This approximates the rolling motion that
larger sediment particles experience as they roll over the surface of the packed
bed as opposed to becoming entrained in the bulk flow.

Within each computational cell, each component of sedimentationdrifting, entrainment

and bed-load transportis computed independently and summed together. The fluxes
due to each are described below.
1

FLOW-3D and TruVOF are registered trademarks in the USA and other countries.

Drifting
The drift velocity is computed based on the assumption that the drift of each species can
be considered separate from all of the other species. This is true so long as the particles
do not interact with one another, which is usually true for particles in suspension.
Therefore, one can write momentum balances for each sediment phase i and the mean
flow (i.e., the equation solved by FLOW-3Ds solver):
u s ,i
Ki
1
+ u u s ,i =
P + F
u r ,i
t
f s ,i s ,i
s ,i
u
1
+ u u = P + F
t

(a)
(1)
(b)

Here us,i and u are the velocities of sediment species i and the mean bulk fluid,
respectively, s,i is the microscopic sediment density and the mean mixture density is

n. of species
f s ,i s ,i +1 f s ,i f
i =1

n. of species

i =1

(2)

where f is the fluid density. fs,i is the volume fraction of sediment species i, P is the
pressure, Ki is the drag function for sediment species i, F is the body force, and the
relative velocity, ur,i is
u r ,i = u s ,i u f .

(3)

Subtracting Eq. (1b) from (1a) gives

u drift ,i
t

1
Ki
1
u r ,i .

P
+ u u drift ,i =

f s ,i s ,i
s ,i

(4)

Here the drift velocity udrift,i is

u drift ,i = u s ,i u .

(5)

Assuming that the motion of the sediment is nearly steady at the scale of the
computational time and that the advection term is very small due to small gradients in the
drift velocity, then the result of Eq. 4 is
u r,i =

P
( f ) f s,i .
K i s ,i

(6)

A reasonable choice for the drag function Ki combines form drag and Stokes drag:
2

Ki =

f
3 f s ,i

CD ,i u r, i + 24
f d s ,i
4 d s ,i

(7)

where ds,i and CD,i are the diameter and drag coefficient for sediment species i,
respectively and f is the fluid viscosity.
The mean flow velocity is computed from the velocities of all of the phases:
N
N

u = 1 f s , j u f + f s , j u s , j
j =1
j =1

(8)

so from Eq. 5, we find that the drift velocity of species i is

u drift ,i = (1 f s ,i )u r ,i

N ( i )

f
j =1

s, j

(9)

ur, j

Eqs. 6, 7 and 9 are solved using the quadratic formula since the relative velocity appears
on both sides of the equations.
The drift velocity affects the motion of the sediment separate from its advection with the
fluid. This is quite straightforward in regions away from walls or the packed bed interface
and is exactly what is done in FLOW-3Ds standard drift-flux model. The drift velocity is
essentially an additional velocity in the advection part of the transport equation for the
suspended sediment. However, adjacent to blockages, care must be taken as sediment can
build up in these regions and become packed. This is done by first considering the
additional advection of the suspended sediment into the cell by Eq. 9. Then the
accumulated sediment, is packed at a rate based on the drift velocity. This procedure is
followed in each of the Cartesian 3D directions throughout the domain wherever drifting
occurs.
Optional Calculation of Base Critical Shields Parameter
For non-cohesive sediment, the dimensionless critical Shields parameter cr ,i is used to
define the bed shear stress at which sediment movement begins. A base parameter, cr ,i ,
representing the initiation of motion of a grain size in a flat, horizontal, uniform bed of
identically-sized grains is modified twice: first to adjust for the hiding and exposure
effects of grain-size mixtures and second to adjust for the increased likelihood of erosion
on sloped surfaces. In previous versions of FLOW-3D, the user was required to specify
the base parameter manually. In Version 9.4, the user can now leave the base value
unspecified and a base value will be determined from the Shields-Rouse equation, which
approximates Shields original curve [1,2]. Note that the region of validity for the
Shields-Rouse equation is limited as shown in Fig. 1, and if the sediment is outside of this
region, the base parameter should be user-specified to maintain accuracy.
3

Figure 1. Plot of the result of the Shields-Rouse equation (Eq. 15) as a function of the
dimensionless Rouse Reynolds number, R* (Eq. 14). [2]

The first step to computing the critical Shields number pre the Shields-Rouse equation is
calculating the dimensionless Rouse Reynolds number Ri*:

Ri* = d s ,i

0.1( s ,i f ) f g d s ,i

(14)

and from this, the dimensionless critical Shields parameter is computed from the ShieldsRouse equation [2]

cr ,i =

0.1
*2

Ri 3

Ri*0.52
.
+ 0.054 1 exp
10

(15)

Calculation of Hiding/Exposure Effects in Multiple-Species Simulations

When multiple sediment species are defined, the base critical Shields parameter is
modified to include the effects of hiding and exposure, whereby larger sediment particles
protect finer particles from becoming entrained and are themselves more exposed. The
first step is the calculation of the local mean particle size in each computational cell:
d 50 =

i =1, ns

c d

i i =1, ns

(16)

s ,i

where ci is the mass concentration of sediment species i. The use of a mean particle size
(here defined as d50, also often termed dm) diverges from the literature, which recently has
moved toward the use of the median particle size as d50. The difference in these
approaches is usually well within the bounds of experimental data scatter, and calculating
the mean value is slightly more computationally efficient. The base critical Shields
parameter is modified by the Egiazaroff function [3]:

cr ,i = cr ,i

1.666667
d
log10 19 s ,i
d 50

(17)

Note that from Eq. 17 that if ds,i is far smaller than d50, the denominator will be a small
number and thus enhance the value of cr ,i , because the finer particles are surrounded by
larger particles. Conversely, for values of ds,i much larger than d50, Eq. 17 will serve to
reduce it, because the coarser particles are more exposed when surrounded by finer
particles. The relative effect of the hiding/exposure function for variable particle size in a
typical situation is shown below.
As indicated in Fig. 2, the Egiazaroff function included in FLOW-3D applies to a limited
range of sediment distributions: for very fine sediments mixed with very large sediments,
the relation is unlikely to be ideal. As noted in Buffington and Montgomerys exhaustive
review of hiding/exposure functions [1], there is no single relation that adequately
describes all possible bed mixtures and flow conditions. The hiding/exposure function is
easily customizable in the user-customizable subroutine scour_lift.F.

Figure 2. Plot showing the effect of the hiding/exposure function (Eq. 17) on the critical
particle shear stress as a function of dimensionless sediment diameter d*.

Calculation of Bed Slope Effects

The critical Shields parameter is further modified for sloping surfaces when the angle of
repose is specified. At sloping interfaces, the packed sediment is less stable and is thus
,i [5]:
more easily entrained by the moving fluid. The modification further alters crit
cos sin + cos 2 tan 2 i sin 2 sin 2
,
cr ,i = cr ,i
tan i

(18)

where is the computed angle of the packed interface normal relative to the gravitational
vector g, i is the user-defined angle of repose for sediment species i, and is the angle
between the flow and the upslope direction. Therefore, for flow directly up a slope,
= 0 . The bed angle adjustment can be turned off for a given sediment species by
leaving the value for angle-of-repose blank (unspecified), or, by specifying a negative
value. Other bed-angle relationships can be specified by customizing the scour_lift.F
subroutine.

Entrainment
Entrainment is the picking up and re-suspension of packed sediment. Therefore,
entrainment needs to be computed only at the packed sediment interface. Because it is not
possible to compute the flow dynamics about each individual grain of sediment and it is
often difficult to compute the boundary layer at the interface, an empirical model must be
used. The model used here is based on Mastbergen and Von den Berg [4].
The local Shields parameter at the bed interface is computed based on the local shear
stress, :

i =

.
g d s ,i ( s ,i f )

(19)

Here g is the magnitude of the gravitational vector.

The entrainment lift velocity (volumetric flux) of sediment is then computed as [4]

u lift,i = i n s d*0.3 ( i cr ,i )

1.5

g d s ,i ( s ,i f )

(20)

where d* is the dimensionless mean particle diameter,

f ( s ,i f ) g 3
d * = d 50
,
2

(21)

i is an empirical lifting parameter, whose default value is 0.018 (representative for

submarine sands), and ns is the outward pointing normal to the packed bed interface. ulift,i
is then used to compute the amount of packed sediment that is converted into suspended
sediment, effectively acting as a mass source of suspended sediment at the packed bed
interface. Once converted to suspended sediment, the sediment subsequently advects and
drifts. Other empirical entrainment relations can be specified by customizing the
scour_lift.F subroutine.

Bed-load Transport
Bed-load transport describes the movement of large sediment along the surface of the bed
without being entrained into the bulk fluid flow. As reported in the literature, bed-load
transport may be observed at average shear stresses below the average critical shear stress
reported for the sediment diameter. This indicates that empirical bed-load transport can
be thought of as a way of quantifying otherwise unaccounted turbulent eddies that exceed
the average shear stress of the studied flow region. Therefore, when modeling regions
with relatively large cell sizes or where the bed surface may vary (due to submerged
7

trees, tires, and elevation changes in un-surveyed bed regions, for example), bed-load
transport models can be used to estimate the effect of small-scale turbulent eddies that
would otherwise require an extremely accurate bed survey and extremely small mesh
sizes.
For bed-load transport, the model currently used (and easily customizable in the routine
scrersn.F) is that of Meyer, Peter and Muller [3,5,7]. This model predicts the volumetric
flow rate of sediment per unit width over the surface of the packed bed. The model is
i = i ( i cr ,i )

1.5

(22)

where i is the dimensionless local bed shear described in Eq. 19, cr ,i is from Eq.s 15,
17, and 18 and i is the dimensionless bed-load transport rate and is related to the
volumetric bed-load transport rate per unit width, qi,b by
f
qb ,i = i g s ,i

3 2
d s , i

(23)

where g is the magnitude of the gravitational vector. The value of i in Eq. 22 is a

dimensionless fitting parameter, generally 5.0 - 5.7 for low transport, 8.0 for intermediate
transport, and up to 13.0 for very high transport (for example, sand in sheet flow under waves
and currents). The default value is 8.0, which is also the most common value found in the
literature [5].
Equation 23 computes the bed-load flux in units of volume per bed width per time.
Another piece of information needed an estimate of the bed-load thickness, i.e., the
thickness of the saltating sediment. The relationship [6] chosen to estimate this thickness
is

i
d 50

= 0.3d

0.7
*

cr ,i

0.5

(24)

Therefore, to compute the motion of the sediment in each computational cell, the value of
qi is converted into a velocity by [6]:
ubedload ,i =

qb ,i
.
i fb,i

(25)

Here fb is the volume fraction of the bed-load layer; van Rijn [7] predicts that it is

f b ,i = 0.18

f packed i
.

d* cr ,i

(26)

The direction of the motion is determined from the motion of the liquid adjacent to the
packed bed interface. Therefore,

u bedload,i = ubedload ,i

u
.
u

(27)

Here u u represents the direction of the fluid-sediment mixture adjacent to the packed
interface, and the resulting bed-load velocity, u bedload ,i , is used to transport the packed
sediment.
The mass flux of sediment as computed across computational cell boundaries in FLOW3D is
Qb ,i = u bedload ,i i f b ,i s ,i

(28)

and hence the and fb from Eq. 25 are cancelled out.

The transport rates calculated by the bed-load model and the entrainment model for an
illustrative case are shown in Fig. 3, which indicates the dominance of bed-load transport
for the motion of large particles and the dominance of entrainment for the transport of
small particles. Note that the bed shear stress selected for this case is arbitrarily large.

Figure 3. Plot of the sediment transport velocity (Eq. 25) as a function of the sediment
diameter for a fixed bed shear stress () of 1000 dyne/cm2.
9

Setting Up a Model in FLOW-3D

Number of fluids: the sediment scour model is designed to be used with singlefluid flow. If the two-fluid flow model is used, the model will still work, but the
sediment will not advect or interact with the second fluid.
Turbulence model selection: a turbulence model does not need to be activated if
the flow is laminar. If the flow is turbulent, all of the available models are
compatible with the sediment scour model. The sediment scour model is sensitive
to turbulence modeling, because the turbulence model directly affects the
viscosity as computed in FLOW-3D, and this viscosity is needed to compute the
local shear stress that is used to calculate the rates of entrainment and bed-load
erosion. It is generally recommended that the Renormalized Group model (RNG)
model of turbulence be used. The Large Eddy Simulation (LES) model is also a
viable option for problems where the large eddies can be resolved by the mesh.
Turbulent mixing length: If a turbulence transport model is used (i.e., the one
equation, two equation, or RNG model), it is recommended to not use the
dynamically computed maximum turbulent mixing length model; this model has
shown to give poor results near the interface of the packed sediment. Therefore, it
is best to provide a length scale for the maximum turbulent mixing length a
value of 7% of the hydraulic diameter is a general recommendation. The
hydraulic diameter is the length scale in the simulation that controls turbulence,
and usually corresponds to a pipe diameter, pool depth, or other physical feature.

To activate the sediment scour model within FLOW-3D, navigate to the Physics tab
under Model Setup, and select the Sediment Scour button. The following window appears
(the example below is partially filled in):

Here the user can specify the number of sediment species (a value of 0 turns off the
model) and the maximum packing fraction, which corresponds to fpacked and is the
maximum packing fraction of the sediment in the packed bed. A line in the table appears
for each sediment species. The only value that must be specified by the user is the
sediment diameter. The drag coefficient ( CD ,i ), entrainment coefficient (i), and bed-load
coefficient (i) are all set to typical default values. The density default value is 2.65,
which corresponds to a typical density of silica-based sediment in CGS units. The critical
10

Shields number appears blank: by default, its value is computed using Eqs. 14 and 15. A
user-defined value can be specified, which overrides Eqs. 14 and 15. By default, the
angle of repose model is turned off, which means that Eq. 18 is bypassed during the
calculation. Any value for the angle of repose between 0 and 90 degrees is accepted.
In the Meshing and Geometry panel, the user provides details of the packed sediment
regions by creating first the solid geometry of the regions. Below is pictured a segment of
the Meshing and Geometry window, with a component branch opened:

A packed sediment component is created like any other solid component, except that
Packed Sediment is selected from the Component Type drop-down menu. Also, in the
Packed Sediment Fraction branch, the volume fractions of the various sediments are
specified; here a histogram is plotted indicating the relative amounts of each of the
species. The sum will automatically be 100%, and the total volume fraction will
correspond to the critical packing fraction of the sediment.
Boundary conditions for the sediment scour model relate only to the suspended phase. To
specify the suspended sediment at boundaries, note that in each of the boundary condition
windows, there is a Sediment Concentration button, which opens the following window
when selected:

11

Here the number of entries listed in the Sediment Concentration table corresponds to the
number of sediment species specified. Here a constant concentration can be specified
over time, or the variable button can be selected to specify a time-dependent value of the
sediment concentration at the chosen boundary. Sediment concentrations can only be
specified at boundaries where fluid crosses, and the concentration is specified in units of
mass per unit volume, in the units used during model setup (the same units as density).
The initial conditions for suspended sediment are specified in the Initial tab under Model
Setup. Here suspended sediment can be added globally, or to specified fluid regions. To
provide global sediment concentration, enter the concentrations (again in mass per unit
volume) in the Suspended sediment concentration section of the Initial tab:

For setting sediment concentrations in individual initial fluid regions, create a fluid
region and open the Suspended sediment concentration branch:

12

Here specify the concentration (mass/volume) of the suspended sediment of each species
in the fluid region.

Test Cases
Deposition of bidisperse sediment in a flume
Figure 4 shows the experimental setup of Gladstone et al [7]. The experiment was
performed in a long rectangular tank with one end separated from the remainder by a
lockgate, behind which sediment is held uniformly in suspension until the lockgate is
raised. The experiment began with the raising of the lock gate, and the resulting density
current caused the sediment to distribute along the length of the tank. The distribution of
the sediment was measured, although the ratio of the sediment species in the deposited
bed was not. The experiment was repeated for seven cases, each case with a different
ratio of coarse to fine sediment. The sediments used had a diameter of 25m and 69m
for the fine and coarse sediments, respectively. Both sediments were silicon carbide
particles and had a microscopic density of 3.217g/cm3.
The experiment was simulated with FLOW-3D and began at the moment that the
lockgate was removed. The sedimentation and scour model was used with sediment
diameter and density specified and all other settings left to default. The resulting density
current caused the sediment to distribute along the length of the tank, and the distribution
output was converted via spreadsheet into mass per unit area along the bottom of the
tank. This can be compared directly to the results of Gladstone et al. [7]. Table 1 lists the
cases that had been both conducted experimentally and simulated.

13

Sediment in
suspension

Clear water
Lockgate

0.2m

0.2m

0.2m

5.5m

Figure 4. Experimental tank emulated by the simulation. The sediment in suspension

moves into the remainder of the tank when the lockgate is lifted. The resulting sediment
distribution [7] after settling onto the bottom of the tank was measured and compared
with the FLOW-3D simulation results.

Fine

Coarse

Case A

100%

0%

Case B

80%

20%

Case C

60%

40%

Case D

50%

50%

Case E

40%

60%

Case F

20%

80%

Case G

0%

100%

Table 1: The seven cases that were run using the sedimentation tank described in Fig. 4.
The fine and coarse sediments have diameters of 25m and 69m, respectively.

Figure 5 shows the plots of sediment concentration along the bottom of the tank after
settling has completed. The experimental results [7] were measured after the flow-front
has completed its advance and all sediment had settled to the bottom of the tank. The
time required for settling varied from around 100s (for Case G) to nearly 300s (for Case
F). All simulations were run for a simulation time of 400s to ensure complete settling,
and were computed on a three-dimensional single mesh block with 105,600 uniform
cells, each about a 1.9cm cube. FLOW-3Ds renormalized group (RNG) turbulence
model was used, and the value of the maximum turbulent mixing length is 2.8cm.

14

The trends shown in Fig. 5 predicted by simulation match quite well the trends measured
experimentally. Furthermore, the location of the peak deposit density is well predicted in
each case. Not ideally predicted is the tailing off of the deposits towards the downstream
end of the tank; here FLOW-3Ds model tends to over-predict the deposit density. This
could be due to inadequacies of the turbulence model, as there is not sufficient time for
the turbulence model to develop turbulence, and the model is not designed to deal with
the very transient behavior of the turbulence that would be present in this system.

Figure 5: Plots of deposit density from Gladstone et al [7] (left) and the simulation from
FLOW-3D (right).

Figure 6 shows the evolution of the transport of the fine (25m) sediment for Case D, and
shows the bulk of the sediment settles over the first 120s of the simulation. The packed
sediment contours could not be easily plotted here because they form a layer thinner than
a single computational cell on the bottom of the tank. Therefore, though it appears that
the suspended sediment in Fig. 5 vanishes, it is instead transferring into packed sediment
which is not visible in the figure.
Further adjustments of the parameters, particularly the entrainment parameters, i, and
the coefficient of drag on the sediment particles, CD,i, may provide better results.
However, the goal of this work was to evaluate the performance of the default parameters
as provided to FLOW-3D users.

15

t=0s

t=40s

t=120s
Figure 6: Time evolution of the suspended (left) and packed (right) fine sediment (25m)
for Case D, which has equal proportions of fine and coarse sediment.

Full Dam and Spillway Simulation

A more real-world simulation of a full dam and spillway was simulated with the sediment
scour model. This simulation involves the geometry shown in Fig. 7. The mesh used in
this simulation is composed of four mesh blocks. The mesh block used to define the
spillway (the deep orange in Fig. 8) has a uniform mesh size of 0.34m. The mesh blocks
used to define the regions to the right and left of the spillway (the lighter orange and
yellow in Fig. 8) have a uniform mesh size of 1.02m, while the mesh block defining the
dam itself has a resolution ranging from 0.6m to 0.78m. The total number of cells used is
3,694,508 and the number of active real cells is 2,919,305.
The topography is composed of a packed sediment mixture of three species of sediment
in equal proportions: sand (d=1mm), coarse sand (d=2.5mm) and gravel (d=10mm). All
species have the same microscopic density, 2.65 g/cm3. The critical packing fraction is
0.64, and the default values of the entrainment and bed-load coefficients were used, 0.018
and 8.0, respectively. Also, the critical Shields number was computed automatically
using the Shields-Rouse equation described in Eq. 15.
In the simulation, the water flows through both the spillway and through the outlet pipes
of the dam (to the right of the spillway in Fig. 7). One would expect to see the greatest
rate of erosion in the vicinity of the outlet pipes as these are directly adjacent to the
packed sediment region, while the spillway has a long concrete apron downstream from
the dam to minimize erosion.

16

Figure 7: Initial setup of full dam and spillway simulation showing the topography
(colored region) and the dam structure (grey region). The colors correspond to the surface
elevation of the topography. There is initially no water downstream from the dam. To
give a perspective of scale, the height of the dam from its base to top is 30m.

Figure 8. Mesh setup showing the four mesh blocks used to mesh the problem domain.

17

t=10s

t=50s

t=100s
Figure 9: Time evolution of the water flow in the full dam and spillway simulation. The
isosurface shown is the fluid fraction and the coloring is the velocity magnitude. It is
difficult in these plots to see where the water meets the packed sediment, but in general,
the zero-velocity regions represent packed sediment.
18

Figure 9 shows a time-series plot of the water flow in the simulation. There are two
regions where the water flow is greatest over the spillway and through the outlet pipes.
The spillway flow has the highest flow velocity, but because it flows down the concrete
lined channel, the erosion in this vicinity is minimal. Downstream of the outlet pipes
there is locally high shearing and turbulence; it is in this area the rate of erosion and
entrainment of the packed sediment is greatest.
Figure 10 shows a time-series plot of the packed bed interface, colored by the local
packed sediment height net change. These plots show the expected result where the
greatest rate of sediment erosion occurred immediately downstream of the dams pipe
outlets. Also, there is a region of significant deposition of sediment immediately to the
left of this high-erosion region. Although not clearly visible from the plots in Fig. 8, there
must exist a consistent region of quiescent flow to allow the rate of drifting of sediment
to exceed the rate of the entrainment and erosion due to bed-load transport.
Figure 11 shows the concentrations of the sand, coarse sand and gravel at the end of the
simulation (t=100s). Because the rate of drifting is strongly affected by the sediment
species diameter (Eq. 7)the viscous drag on the sediment increases quadratically with
the inverse of diameter. The critical shear stress is also affected by the sediment species
diameter it rises linearly with the sediment diameter (see Eq. 19), the gravel phase
exhibits a drastically slower rate of net erosion than does the sand. Therefore, since the
original packed bed of sediment is composed of equal parts sand, coarse sand and gravel,
the areas with greatest erosion should see a relatively higher concentration of the coarser
material remaining. And this is indeed what results in the model.

Final Remarks
The sediment scour model from Version 9.3 of FLOW-3D was completely revamped for
Version 9.4. The changes included the addition of multiple sediment species, a bed-load
transport model, a nonlinear drift-flux model and new empirical equations to predict the
entrainment and erosion of sediment. As before, the set of equations are completely
coupled such that the net erosion and/or deposition is a function of all of the equations
working together, and is dependent on the effectiveness of turbulence modeling.
Therefore, although the individual concepts of the model are quite simple, the
combination of the equations into the fully three-dimensional flow model provides a very
powerful tool to the hydraulics engineer. However, it should be noted that it is just that: a
tool. Because of the vast differences in scale involved in sedimentation and scour
problems (the individual particles are millions of times smaller than the overall scale of a
typical domain), the model necessarily requires a lot of empiricism. The empirical
approach means that care must be taken both when setting up problems and when
interpreting results: if one ignores the assumptions made for the empirical models, the
reliability of the results will suffer.

19

t=10s

t=50s

t=100s
Figure 10: Time evolution of the packed bed interface (as if the water were removed)
colored by the packed sediment height net change in the full dam and spillway
simulation. The greatest amount of erosion occurred directly downstream of the outlet
pipes of the damit is here where the local bed shear is the greatest. Also, there is some
deposition of sediment to the left of the outlet pipes as the quiescent flow in this region
allowed the rate of drifting of sediment to exceed the rate of erosion.
20

Sand (d=1mm)

Coarse sand (d=2.5mm)

Gravel (d=10mm)
Figure 11. Plots showing the concentration of the three species of sediment at the packed
interface at the completion of the simulation (t=100s). Of interest here is the lower
concentration of the smaller sediment (i.e., sand & coarse sand) in the high erosion area
downstream of the outlet pipes, contrasted with the relatively higher concentration of the
gravel due to its lower rate of erosion.
21

References
1 Buffington, J.M. and D.R. Montgomery, A systematic analysis of eight decades of incipient motion
studies, with special reference to gravel-bedded rivers, Water Resources Research, 33(8), pp 19932029,
Aug 1997.
2 Guo, J., Hunter Rouse and Shields diagram, Proc 1th IAHR-APD Congress, Singapore, Vol. 2, 10691098, 2002.
3 Kleinhans, M.G., Sort out sand & gravel: sediment transport and deposition in sand-gravel bed rivers,
Ph.D. Thesis, Universitaat Utrecht 2002.
4 Mastbergen, D.R. and J.H. Von den Berg, Breaching in fine sands and the generation of sustained
turbidity currents in submarine canyons, Sedimentology (50) 625-637, 2003.
5 Soulsby, R., Dynamics of Marine Sands, Ch 9: Bedload transport, Thomas Telford Publications, London,
1997.
6 Van Rijn, L. C., Sediment Transport, Part I: Bed load transport, Journal of Hydraulic Engineering
110(10), 1984.
7 Gladstone, C., J.C. Phillips and R.S.J. Sparks, Experiments on bidisperse, constant-volume gravity
currents: propagation and sediment deposition, Sedimentology 45, pp 833-843, 1998.

22