BULLETIN

MATHEMATICAL MODELLING OF SEDIMENT TRANSPORT AND

DEPOSITION IN RESERVOIRS

Guidelines and case studies

2007
TABLE OF CONTENTS

FOREWORD XV

1. INTRODUCTION 1

2. RESERVOIR OPERATION AND SEDIMENT TRANSPORT PROCESSES 5

2.1 Introduction 5

2.2 Sediment transport mechanisms 5

2.3 Modes of reservoir operation and impacts on the sediment balance 7

3 TURBULENT SEDIMENT TRANSPORT 12

3.1 Introduction 12

3.2 Review of selected equilibrium sediment transport equations 16

3.3 Calibration with reservoir data 34

3.4 Non-equilibrium sediment transport 45

3.4.1 Introduction 45
3.4.2 Review of existing theory 49
3.4.3 Modelling of non-equilibrium sediment transport processes : Welbedacht Reservoir (Caledon
River, South Africa) 55
3.4.4 Comparison between calibrations 57

4. DENSITY CURRENTS 58
4.1 Introduction 58

4.2 Occurrence of density currents in reservoirs 59

4.3 Hydraulics of density currents 64

4.3.1 General 64
4.3.2 Velocity distribution 65
4.3.3 Vertical suspended sediment distribution 65
4.3.4 Shear stress distribution 66

4.4 Mathematical description of the velocity distribution and the thickness of a density current 67

4.5 Verification of theory to predict the velocity profile and depth of a density current with laboratory
and field data 74

4.6 Movement of a density current : flow resistance and velocity 75

4.7 Cross-sectional variation in velocity and sediment concentration across a density current in a
reservoir 81

4.8 Motion of the head of a density current 84

4.9 Sediment transport by density currents 89

i
4.10 Density current formation following flushing 91

4.11 Non-equilibrium density current sediment transport 96

4.12 Graded sediment transport by density currents and the sorting process 97

4.13 Formation of a density current 98

4.13.1 Review of theory 98
4.13.2 Prediction by means of minimum stream power principle 105

4.14 Laminar density currents associated with hyperconcentrated sediment transport 108

4.15 Venting of density currents through reservoirs 110

5. MATHEMATICAL MODELS AND CASE STUDIES 112

5.1 One dimensional mathematical models 112
5.1.1 Introduction 112
5.1.2 Reservoir sedimentation model (1D): Mike 11-RFM: Welbedacht Reservoir 113
5.1.3 Reservoir Sedimentation Model: GSTARS: Tarbela Dam, Pakistan (Yang and Simoes, 2003)
125
5.1.4 Reservoir Sedimentation Model RESSASS: Tarbela Dam (TAMS, 1998) 132
5.1.5 River model Mike 11: Lake Roxburgh, New Zealand (Mackay et al., 2000) 135

5.2 Computational modelling of reservoir sedimentation and flushing with a two- dimensional model
138
5.2.1 Background 138
5.2.2 Data 140
5.2.3 Theoretical Background 142
5.2.4 Grid Generation 142
5.2.5 Hydrodynamics 143
5.2.6 Cohesive sediment transport model 144
5.2.7 Calibration of sedimentation during 1973-76 146
5.2.8 Flushing during 1991 148
5.2.9 Flushing with Low Level Outlets 150

5.3 Three-dimensional Mathematical Modelling (turbulent sediment transport) 151

5.3.1 3D Model Equations 151
5.3-2 Three-dimensional Model (case study): Three Gorges Reservoir Project, China (Dou et al.,
2004) 154

5.4 Models of density currents 159

5.4.1 Introduction 159
5.4.1.1 2D vertical mixture models 159
5.4.1.2 Depth-averaged mixture models 160
5.4.1.3 Proposed density current model 160
5.4.2 Case study 1: Laboratory flume and field data, Canada 162
5.4.3 Case study 2: Luzzone Reservoir, Switzerland 166
5.4.4 Conclusion 169

6. CONCLUSIONS AND RECOMMENDATIONS 170

7. REFERENCES 171

ii
LIST OF FIGURES

Figure 1-1 Historical growth in reservoir storage capacity world wide

Figure 2.1-1: Schematic diagram of reservoir sedimentation processes (Graf, 1983)
Figure 2.3-1 Reservoir operation and possible changes in the reservoir outflow
sediment load-discharge relationship

Figure 2.3-2 Sanmenxia Reservoir flushing after reconstruction of the outlets

Figure 2.3-3 Empirical reservoir classification system in terms of storage, runoff and
sediment yield

Figure 3.1-1 Sediment transport relationships for various Chinese rivers

(Delft Hydraulics, 1992)

Figure 3.2-1 Stream power sediment transport relationship at Fenhe Reservoir, China

(Gaun et al., 1991)

Figure 3.2-2 Gravitational power theory calibrated with Chinese data (Wu, 1994b)

Figure 3.2-3 Comparison between stream power equations (Yang and Kong, 1991)

Figure 3.2-4 Verification (with river data) of calibrated new sediment transport
equation (based on flume data)

Figure 3.2-5 Calibration of new sediment transport equation based on flume and river

data

Figure 3.3-1 Sediment concentration versus stream power for two South African

reservoirs (Rooseboom et al., 1986)

Figure 3.3-2 Input stream power versus observed suspended sediment concentration
for South African reservoirs

Figure 3.3-3 Density current data at Welbedacht Dam

Figure 3.3-4 Calibration of new sediment transport equation based on reservoir data

Figure 3.3-5 Calibration of input stream power sediment transport equation using
reservoir data and not allowing for differences in settling velocity.

Figure 3.3-6 Calibration of input stream power sediment transport equation using
reservoir data and allowing for differences in settling velocity.

iii
Figure 3.3-7 Reservoir sedimentation based on stream power sediment transport

equation calibrated with reservoir data

Figure 3.3-8 Reservoir sedimentation based on Engelund-Hansen sediment transport

equation

Figure 3.3-9 Verification of reservoir sediment transport with new sediment transport

equation based on laboratory and river data calibration

Figure 3.4-1 Calibration of dimensionless new sediment transport equation with flume
data
Figure 3.4-2 Reservoir sedimentation profile with uniform sediment size

Figure 3.4-3 Reservoir sedimentation profile with non-uniform sediment size

Figure 3.4-1 Variation of suspended sediment concentration with time

(Partheniades, 1986)

Figure 3.4-2 Relative equilibrium concentration versus bed shear stress parameter

(Mehta and Partheniades, 1973)

Figure 3.4-3 Deposition rates (Mehta and Partheniades 1973)

Figure 3.4-4 Non-equilibrium sediment transport calibration: Welbedacht Reservoir,

1973 to 1976

Figure 4.2-1 Plunge point at Eril Emda Reservoir

Figure 4.2-2 Density current in Guanting Reservoir (Fan, 1986)

Figure 4.2-3 Density current in Sanmenxia Reservoir (Fan, 1986)

Figure 4.3-1 Density current velocity, suspended sediment concentration and shear
stress distributions

Figure 4.4-1 Velocity distribution from channel bed to maximum velocity

(Ashida and Egashira, 1975)

Figure 4.4-2 Relationship between mixing length l o at interface and D2

(Ashida and Egashira, 1975)

Figure 4.4-3 Distribution of velocity, shear stress and sediment concentration

(Ashida and Egashira, 1975)

iv
Figure 4.5-1 Observed versus calculated density current layer depths

Figure 4.6-1 Schematic diagram of a density current

Figure 4.6-2 Friction factor λ m from flume studies (Fan, 1960)

Figure 4.6-3 Friction factor for Guanting Reservoir (Fan, 1960)

Figure 4.6-4 Ratio of density current depth to open channel flow depth in Guanting
Reservoir (Fan, 1960)

Figure 4.7-1 Density current lateral velocity and concentration distribution in

Sanmenxia Reservoir (Fan, 1986)

Figure 4.8-1 Non dimensional density current head shape (Altinakar et al., 1990)

Figure 4.8-2 Density current head (Altinakar et al., 1990)

Figure 4.8-3 Density current head velocity (Altinakar et al., 1990)

Figure 4.8-4 Dimensionless head velocity as function of bed slope (Altinakar et al.,
1990)

Figure 4.8-5 Velocity of density current head using Chezy type equation

Figure 4.9-1 Unsteady density current measured in Guanting Reservoir (Wu, 1994b)

Figure 4.10-1 Density current venting at Welbedacht Reservoir

Figure 4.10-2 Longitudinal profile of Welbedacht Reservoir bed

Figure 4.10-3 Reservoir operation and settling velocity based on observed sediment

characteristics

Figure 4.10-4 Chinese reservoir density current sediment transport relationship (Wu,
1994b)

Figure 4.10-5 Welbedacht Reservoir density current sediment transport relationship

Figure 4.12-1 Graded sediment transport in a density current (Wu, 1994b)

Figure 4.13-1 Schematic diagram of density current formation (Fan, 1960)

Figure 4.13-2 Densimetric Froude number and density current formation (Fan, 1960)

Figure 4.13-3 Densimetric Froude number and density difference ratio (Cao, 1992)

Figure 4.13-4 Densimetric Froude number (Akiyama et al., 1987)

v
Figure 4.13-5 Plunge point characteristics

Figure 4.13-6 Density current formation and minimum stream power

Figure 4.14-1 Typical density current velocity distribution (Cao, 1992)

Figure 4.14-2 Profile of interface of density current (Cao, 1992)

Figure 4.15-1 Density current climbing (Bell, 1942)

Figure 5.1.2-1 Phalaborwa Barrage flood flushing (900 m3/s, February 1996)

Figure 5.1.2-2 Welbedacht Reservoir flushing channel deformation

Figure 5.1.2-3 Welbedacht Dam

Figure 5.1.2-4 Calibrated and observed flushing channel bed profiles at Welbedacht
Reservoir

Figure 5.1.2-5 Welbedacht Reservoir 1995 flushing; cumulative double mass plot
verification of sediment transport

Figure 5.1.2-6 Welbedacht Reservoir 1995 flushing; verification of sediment transport

modelling

Figure 5.1.2-7 Simulation of long-term equilibrium sedimentation at Welbedacht

Reservoir with flood flushing and different outlet configurations

Figure 5.1.3-1 Tarbela Dam and reservoir. The points (+) mark the thalweg and the
locations of the cross sections (Yang and Simoes, 2003)

Figure 5.1.3-2 Hydrology and dam operation for Tarbela Reservoir (1974 to 1996)
(Yang and Simoes, 2003)

Figure 5.1.3-3 Two reservoir cross sections showing uniform sedimentation (Yang and
Simoes, 2003)

Figure 5.1.3-4 Results of the simulation of the Tarbela delta advancement over a period
of 22 years (Yang and Simoes, 2003)

Figure 5.1.3-5 Comparison of measurements and GSTARS3 computation for two cross
sections in the upstream region of Tarbela Reservoir (Yang and Simoes,
2003)

Figure 5.1.3-6 Comparison of measurements and GSTARS3 computation for two

cross sections in the reservoir region of the study reach

vi
Figure 5.1.3-7 Relative error of the thalweg elevation predictions

Figure 5.1.4-1 Tarbela Dam Spillway

Fiugre 5.1.4-2 Simulated and surveyed longitudinal profile of Tarbela Reservoir

Figure 5.2-1 Observed loss in storage capacity due to sedimentation at Welbedacht

Dam

Figure 5.2-2 Longitudinal profile of the Welbedacht Reservoir sedimentation

Figure 5.2-3 Aerial photograph of Welbedacht Reservoir

Figure 5.2-4 . Curvilinear grid used for the 2D model, 250x15 cells

Figure 5.2-5 Time series of observed reservoir inflow

Figure 5.2-6 Time series of observed reservoir water level at dam

Figure. 5.2-7 Initial bathymetry from 1973 (above) and simulated bathymetry in 1976
(below).

Figure. 5.2-8 Simulated sedimentation development from 1973-76, sediment volume

and water volume in the reservoir.

Figure 5.2-9 Welbedacht Reservoir during flushing

Figure 5.2-10 Bathymetry before (left) and simulated after (right) flushing in 1991.

Figure 5.2-11 Observed and simulated minimum bed level as function of chainage
before and after flushing in 1991.

Figure 5.2-12 Flushing simulation with gate invert level at the bed at 1380 m

Figure 5.3-1 Evolution from open channel flow to underflow, with stable plunge
point

Figure 5.3-2 Comparison between computed and measured non-dimensional

velocity distributions

Figure 5.3-3 Simulated turbidity current in Sauenay fjord

Figure 5.3-4 Computed flow and concentration fields for turbidity current in
Saguenay fjord at locations A, B, C

Figure 5.3-5 Station locations and computational grid superimposed on reservoir

bottom

Figure 5.3-6 Velocity variation for stations s11 to s61

vii
LIST OF TABLES

Table 2.2-1 Recorded ratios of outflow to inflow sediment load during density current
venting

Table 2.3-1 Reservoir operation and sediment trapping in Sanmenxia Reservoir

Table 3.1-1 Accuracy ranges of commonly used sediment transport equations

Table 3.2-4 Accuracy of new sediment transport equation in prediction ranges (for
C>0,01%)

Table 3.3-1 Reservoir data

Table 3.3-2 Calibration of stream power relationships

Table 3.3-3 Accuracy ranges of calibrated new sediment transport equation

Table 3.5-1 Suspended sediment size distribution of Welbedacht Reservoir inflow

Table 4.6-1 Interfacial friction factor

Table 4.13-1 Densimetric Froude number (Fp) at plunging

Table 4.13-2 Density current formation (Rooseboom, 1975)

viii
LIST OF SYMBOLS

a adaptation coefficient representing mean settling depth

α dimensionless exponent
α coefficient
A coefficient used in Ackers and White (1973) equation
A integration constant
A channel cross-section area
b constant
B bed width of flushing channel
B top width of flow
Bo buoyancy flux
β constant
βi % of i-th sediment fraction in the bed
c integration constant
c_ coefficient used in Ackers and White (1973) equation
C eq equilibrium concentration
C sediment concentration
C vy average sediment concentration at a distance y above the bed (in percent by volume)

C eq
* relative equilibrium concentration
Cv average sediment concentration
Ct sediment concentration by weight
Cs suspended sediment concentration
Co initial sediment concentration
C fo resistance coefficient
Cd drag coefficient
Ci actual sediment transport per fraction
C ci transport capacity of the i-th class
C1 near bed sediment concentration
Ca reference concentration

C
* equilibrium sediment concentration
CH Chezy coefficient
Cb bed load concentration

C
__ Chezy type coefficient for density current
Co sediment concentration at the bed where
Δz depth of recent deposits
d sediment particle diameter
d 90 sediment size for which 90 % of particles is finer

ix
 d 50 median sediment particle size
d gr sediment particle size
ds representative particle size
δb saltation height
D flow depth
D dispersion coefficient
D* dimensionless particle diameter
D1..4 density current layer depths
eb efficiency coefficient
ε coefficient of turbulence exchange
η parameter as function of sediment particle size, settling velocity and water viscosity
E erosion rate
E applied stream power
E specific energy
Eo empirical erosion rate coefficient
f Lacey's silt factor
f Darcy Weisbach resistance coefficient
fk resistance coefficient with equilibrium sediment transport
fo Darcy Weisbach resistance coefficient where
F mean shear force
Fr Froude number
Fr D densimetric Froude number
F gr mobility number
Fo inflow Froude number
F va shear box power
g gravitational acceleration
g_ gravitational acceleration adjusted for density difference ratio

γ_ specific weight of density current

γ specific weight of water
γd specific weight of deposits
γs specific weight of sediment
G gr sediment transport function
h depth of uniform density current flow
h depth of water above density current

h
_ density current depth
hp density current depth at plunge point
hL maximum climbing height of a density current
h5 density current layer depth
ho density current depth at plunge point

x
hn depth of density current head
H bed form height
H depth of flushing channel
H water depth
I moment of inertia of element around axis 0
Jn water surface slope
Ji density current interface slope
Jo density current bed slope
k roughness coefficient
ks roughness coefficient
k1 dimensionless coefficient
κ Von Kármán coefficient
K1 , K2 constants
KT reservoir capacity - sediment inflow ratio
K3 constant related to sediment concentration
K, K 1 , K 2 coefficient
KW reservoir capacity - inflow ratio
l Prandtl's mixing length
lo mixing length
λ rate of erosion
λm mean friction coefficient of the underflow
λo ; λi coefficients of friction at bed and interface
* adaptation length for each fraction
L i
m coefficient used in Ackers and White (1973) equation
m coefficient related to sediment concentration
m morphological channel stability coefficient
mc stable bank slope
μ dynamic viscosity
η coefficient of rigidity
M constant
M erosion rate
M mass of element
υ kinematic viscosity
n coefficient used in Ackers and White (1973) equation
N coefficient
p total pressure
pc , p m , p s percentages of clay, silt and sand
ψ suspension parameter correction coefficient
∝ angle

xi
∝ empirical erosion rate coefficient
∝ erosion constant
∝ coefficient for non-equilibrium density current sediment transport
∝1 ; ∝2 coefficients
∝3 ; ∝4 constant and coefficient
∝5 ; ∝6 constant and coefficient
P wetted perimeter
P applied power
q discharge per unit width
q so sediment inflow per unit width
q st suspended load transport rate
qt total load transport rate
qbw bed load transport rate
qT total sediment load
q st sediment transport capacity
q d , qe fluxes of deposition and erosion of sediment
qs sediment discharge per unit width
Q flow rate
Qs total sediment discharge
r ratio of predicted to observed sediment transport rate
2 correlation coefficient
r
ρgs v unit input stream power
ρc , ρm , ρs densities of clay, silt and sand fractions
ρT sediment density after T years
ρs sediment density
ρm density of sediment-laden water
ρ2 density in density current
ρ∞ long-term sediment density
ρo initial sediment density
Ro radius of eddies near bed
R hydraulic radius
Rρ hydraulic radius at plunge point
s slope
ss sediment source/sink term
sf energy slope

s
* sediment transport capacity
S2 energy slope for sediment-laden water
S1 energy slope for clear-water
so bed slope

xii
ss water surface slope
tan α ratio of tangential shear to normal force
θ coefficient used in Engelund Hansen equation
τ ce critical shear stress for surface erosion
τ bed shear stress
τ cme critical bed shear stress for mass erosion
τo bed shear stress
τ* shear velocity to the power 2
τo shear stress at bed
dv applied stream power
τ
dy
τi shear resistance at the interface
τB yield shear stress
τo mean bed shear stress
τc critical shear stress
τv stream power
τ oc critical shear stress for incipient motion (Shields, 1936)

τ o_ grain shear stress

T time scale in non-equilibrium sediment transport
T transport stage parameter
T width
u bs particle velocity

ug
_ bed shear velocity
ua effective particle velocity at reference level a

ug
_ effective bed shear
_
u f ,cr Shields critical bed shear velocity
u average velocity in direction
u_, u__ first and second derivations of velocity
U x ,U y fluctuating components of velocity in and transverse directions
Ux mean shear velocity
u_, v_ fluctuating part of velocities in the and directions
v flow velocity
vy flow velocity at distance y above the bed
vmax maximum flow velocity
vp velocity at plunge point
vf front velocity of a laminar density current
v* shear velocity
vo rotation centre point velocity
vn density current head (nose) velocity

xiii
v average flow velocity
vscr critical input stream power for incipient motion
vs average input unit stream power
V reservoir capacity
Vt reservoir capacity after t years of operation
Vo density current flow velocity at plunge point
w particle settling velocity
w distance from centre point of a sphere
w50 settling velocity of median particle diameter
x Einstein correction coefficient
y1 distance from bed where laminar velocity distribution equals that of turbulent flow
yo mathematical distance from bed where flow velocity is zero
z suspension theory coefficient
zd depth of previously deposited sediment
z1 suspension theory coefficient as revised by Rooseboom (1975)
ζ non-equilibrium sediment transport parameter

xiv
FOREWORD

These guidelines have been written by Gerrit Basson, professor in hydraulic engineering and
his predecessor, Albert Rooseboom, both of the University of Stellenbosch, South Africa,
with inputs from members of the ICOLD sedimentation committee.

As reservoir sedimentation has proven to be a serious problem in South Africa, research in

this field has been ongoing for more than 70 years. This publication emanates from extensive
research which has been undertaken over the past 30 years with the support of the South
African Department of Water Affairs and Forestry as well as the South African Water
Research Commission.

A great deal of information has fortunately also been obtained from China, the country with
the most extensive experience in this field.

Given the universal nature of hydraulic formulae it is not surprising, yet gratifying, that
Chinese and South African data generally conform to the same mathematical relationships.
This indicates that these relationships should be applicable in other countries as well. Much
of the information contained here has been condensed from a more comprehensive
publication (Basson and Rooseboom, 1997).

This ICOLD Bulletin follows on Bulletin 115 (ICOLD, 2000): “Dealing with reservoir
sedimentation,” which gave guidelines for management of reservoirs to limit sedimentation.
The guidelines on mathematical modelling of sediment transport dynamics in reservoirs in
this document can be used during the planning and design of new dams, and management of
existing dams.

xv
1. INTRODUCTION

Although there are reservoirs where sedimentation has proven to be a problem within
10 years after construction, sedimentation typically only becomes a significant
problem 50 more years after construction of a dam. With only about 7% of the
existing storage capacity in reservoirs world-wide having been created before 1950
(see figure 1-1), and with the average annual sedimentation rate now being estimated
at 0,8% of the original storage capacities, it is expected that sedimentation will
become a serious problem in many reservoirs during the next 50 years. The average
age of reservoirs is now about 35 years (based on storage capacity) and since many
reservoirs have been designed with a dead storage for sedimentation of about 50 years,
serious sedimentation problems are going to develop with about 50 percent of the live
storage capacity in reservoirs affected within the next 20 years. The effect of loss of
the live storage at hydropower dams is however less of an impact than at
storage/irrigation dams. Most of the existing reservoirs in the world will be completely
silted up in 200 to 300 years from now, with large reaches of river system permanently
modified. The ecological functioning would be completely different and only run-of-
river water diversion or hydropower schemes can be implemented.

In 2006, 21 % of the total storage capacity was filled with sediment, increasing to 42%
by the year 2050.

There has been a dramatic increase in the number and size of the dams being built
after the Second World War, peaking during the 1970’s worldwide (Figure 1-1).
Damming created by a dam results in reduced sediment transport capacity upstream of
the dam and sediment deposition, with the loss of live storage capacity. In many cases
sediment deposition also occurs above the full supply level of the reservoir, sometimes
constituting more than 10 % of the deposited sediment. As sediment deposition
continues, the sediment delta grows higher and eventually flood levels start to rise.
Not only flood levels are affected, but also drainage from agricultural land, bridge
discharge capacity, pump station and hydropower operation and navigation. In semi-
arid climates, the primary effect of reservoir sedimentation is however the loss in
storage capacity for domestic or industrial supply and irrigation.

1
 Global Growth of dams

7000000

6000000

5000000
Storage Capacity (mcm)

4000000
Growth
Loss
3000000

2000000

1000000

0
1900 1910 1920 1930 1940 1950 1960 1970 1980 1990 2000 2010 2020 2030 2040 2050
Year

Figure 1-1: Global growth of reservoir storage capacity and reservoir sedimentation

Reservoir sedimentation occurs worldwide at a rate of about 0.8 percent per year, but
the sedimentation rate in many regions such as Asia is much higher. Using an average
rate, Palmieri (2003) estimated the loss to be approximately 45 km3 per year. The cost
of replacing the lost storage is significant; nearly US$ 13 billion per year would be
needed, even without counting the environmental and social costs associated with new
dams (Palmieri, 2003).

Sediment deposition occurs as a river enters a reservoir and its sediment transport
capacity decreases in the backwater created by a dam. Coarse sediment is typically
deposited first while finer clay and silt fractions are transported much deeper into the
reservoir.

Apart from the obvious fact that sediment build-up within a reservoir leads to
decreasing storage capacity, sediment build-up in specific areas can lead to local
problems. Sediment build-up in a storage reservoir without facilities for preventing
sedimentation typically only approaches equilibrium when the remaining storage
capacity is a few percent of the original storage capacity.

2
Within the reservoir sediment build-up can have serious impacts on diversion and
extraction facilities including turbine penstock inlets and pump station intakes.

In the context of reservoirs sedimentation distinction can be drawn between “small

reservoirs” and”large reservoirs”. Small reservoirs, with capacities in the order of a
few present of their MAR’s (Mean Annual Runoffs), can be designed and operated in
such ways that they retain large proportions of their original capacities over long
periods of time. In the case of small reservoirs the required calculations concentrate
on the shapes and volumes of the equilibrium sediment deposits as well as the sizes of
gates that may be required to sluice (pass incoming sediments through) ant to flush
(scour sediments out). In addition, these calculations serve to predict the associated
sediment concentrations so as to ensure that environmental standards are met with
regard to the releases downstream.

Reservoir sedimentation and the corresponding loss of storage capacity is a common

problem, which has attracted more and more attention in recent years. A number of
different hydraulic measures can contribute to the management of the sedimentation
problem, such as:

• Minimizing deposition in reservoirs through sluicing or venting of density currents

• Removing accumulated sediment from the reservoir through flood flushing
• Diverting sediment-laden flow to bypass the reservoir
• Controlling the location of sediment deposits in the reservoir for later dry
excavation.

The first two mentioned measures above are costly in terms of water, hence effective
reservoir management requires tools such as detailed prediction mathematical models,
which can predict and monitor the effectiveness of such hydraulic measures.

Sediment sluicing and flushing have proven effective in a number of cases, provided
excess water and adequately sized outlets are available or can be installed. Flood
flushing are practiced based on availability of excess water, in two ways: (a) water
level drawdown during the flood season, or (b) in semi-arid climates flushing during

3
single flood events with drawdown to free outflow conditions and filling with clear
water after the flood.

It is generally not practicable to sluice or to flush sediments in the case of storage

reservoirs with capacities well in excess of a few percent of their MAR’s, except in the
case of reservoirs with steep bed slopes where sediment induced density currents
occur. Sediment modelling in the case of large reservoirs therefore concentrates on
predicting the shapes and volumes of sediment deposits as functions of time, and
establishing whether density currents could be sluiced or vented through special gates
in order to limit sediment build-up.

These guidelines serve to provide assistance in performing the calculations which are
necessary for calculating the volumes and shapes of sediment deposits in reservoirs.
More sophisticated models however also consider bed erosion in order to simulate
flushing/sluicing operation.

As it proves to be extremely difficult and costly to get rid of sediment-filled

reservoirs, the impacts of long-term equilibrium sediment build-up should be
considered seriously. Off-channel storage, whereby water is diverted from rivers by
very small weir structures via canals, tunnels or pipelines to reservoirs, created high up
in catchments, should be seriously considered wherever possible as it constitutes one
of a few long-term sustainable development options. Given the complex nature of
sediment transport mechanisms within reservoirs it is not possible to provide simply
recipes for modelling transport within reservoirs. Such modelling should be
undertaken only by persons who have become familiar with the theories in order to
establish which relationships are the most appropriate for their situations.
These guidelines thus contain a number of selected formulae which may be considered
for use. It is recommended that different formulae be used in order to obtain a feeling
for the range of answers that are possible and that as much local field calibration data
be obtained as possible. No specific model is prescribed but it is recommended that
models which have been calibrated with reservoir data, rather than river data, be
preferred.

4
2. RESERVOIR OPERATION AND SEDIMENT TRANSPORT PROCESSES

2.1 Introduction

When a river enters a reservoir, the sediment transport capacity is reduced and the
sediment load is no longer dependent on sediment availability but can be related directly
to hydraulic conditions in the reservoir. While coarser sediment is generally deposited
in the upstream part of a storage reservoir, fine sediments (silt and clay fractions,
d<60 μm), are transported further towards the dam through turbulent suspension, density
currents or colloidal suspension. Deposition occurs in the main (original river) channel
and overbank areas of the reservoir. Typical flow and deposition patterns within large
reservoirs are indicated in Figure 2.1-1 (Graf, 1983).

Figure 2.1-1: Schematic diagram of reservoir sedimentation processes (Graf, 1983)

2.2 Sediment transport mechanisms

Three separate mechanisms of sediment transport can be identified within reservoirs,

namely turbulent suspended sediment transport, density currents and colloidal
suspension.

• Turbulent suspension is the dominant mechanism of sediment transport through

most reservoirs (as in rivers), and is discussed in detail in Chapter 3.

5
• A density current may be defined as the movement under gravity of a stream of
fluid under, through, or over another fluid, the density of which differs by a
small amount from that of the primary fluid. Differences in temperature, salinity
or suspended sediment concentration can provide the mechanism by which a
denser fluid "dives" below the upper fluid. Different types of density currents
have been observed, such as overflow, interflow and underflow, and while the
first two are often more related to temperature or salinity density difference
systems, the latter is the means of sediment transport sometimes found in
reservoirs.

Field data on the ratios of outflowing sediment load to incoming load during
density current venting for different reservoirs are presented in Table 2.2-1
(Bruk, 1985; Delft Hydraulics, 1992; Tan, 1994; Mahmood, 1987).

Table 2.2-1: Recorded ratios of outflow to inflow sediment load during

density current venting

Reservoir Location Storage Reservoir Ratio:

capacity length sediment
(million m3) (km) outflow
to incoming
load
Eril Emda Algeria 160 0,25 - 0,60
Lake Mead USA 38 400 128 0,18 - 0,39
Guanting China 2 270 0,19 - 0,34
Heisonlin China 8,6 2 1,16 - 0,59
Sanmenxia China 9 640 80 (1961) 0,18 - 0,21
Sefid-Rud Iran 1 800 0,20 - 0,37
(original)
Fengjiashan China 398 12 - 14 0,23 - 0,65
Bajiazui China 525 (original) 10 0,46
Liujiaxia China 5 720 0,57 - 0,84
(original)
Nebeur Tunisia 300 0,59 - 0,64

6
 Density currents have been observed in several reservoirs such as Lake Mead,
USA, Sautet Reservoir in France, Metha and Groshnitza Reservoirs in the
former Yugoslavia, and the Nulele Reservoir in the former USSR (Wu, 1994).

The hydraulics of density currents are discussed in Chapter 4.

• Colloidal suspensions are due to electrostatic forces which keep small particles
in suspension, and their transport is therefore not related to the effect of gravity.
Such suspensions not only depend on particle size, but are largely influenced by
water quality. Colloidal sediments fall in the size range of approximately 10-3
to 1 micron, between dissolved particles on the one hand and sediments
suspended by turbulent/laminar flow conditions on the other.

The so-called wash load (normally not considered in laboratory calibrated

standard sediment transport equations), should not be mistaken for colloidal
suspension. The research conducted for this study has shown that the transport
of silt and clay sediment fractions as found in many reservoirs can be described
by relationships which are also valid for coarse sediments.

Colloidal suspensions normally would be present in reservoirs at

concentrations of less than 100 ppm. Due to their estimated low contribution
to total sediment transport through a reservoir (normally less than 3 percent),
such suspensions will not be considered further in this study. Even though they
are important from a water quality perspective.

2.3 Modes of reservoir operation and impacts on the sediment balance

In order to understand how efficiently the above hydraulic measures can deal
with reservoir sedimentation, their respective impacts on sediment loads and
on trapping of sediment need to be considered. Over the long term, a sediment
load-discharge relationship as indicated in Figure 2.3-1 is obtained for a
“natural” river, indicated by data points for the flood season (Sep to Dec) and
the low flow season (Jan to Aug).

7
 100 Observed river sediment load
range before dam construction
2 .2 8 9 2
y = 1E -06x
10

Flood flushing (single y = 2E -06x 1 .9 7 8 6

1 storms)
S ed im en t L oad (m/s)
3

0.1 Seasonal
flushing
0.01
0 Density
0.001 current
venting
0.0001
Sep - Dec
Storage Jan - A ug
0.00001
operation Pow er (Jan - A ug)
Pow er (Sep - Dec )
0.000001
1 10 100 1000 10000
3
D isch arge (m /s)

Figure 2.3-1 Reservoir operation and possible changes in the reservoir outflow sediment
load-discharge relationship

One typical management option but extreme in terms of sediment transport is storage
operation, allowing almost no through-flow of suspended sediment. Almost clear water will
be released from storage operated reservoirs, typically resulting in channel degradation
downstream of the dam. With regular flood hydrograph flushing (normally practised in semi-
arid regions), the operational method is only efficient above a certain discharge. At high
discharges, sediment loads higher than the mean seasonal sediment observed for the pre-dam
scenario can be expected, but sediment equilibrium can be established. Regular flushing with
suitable bottom outlets will ensure that flushed sediment approach the mean background level
of sediment concentrations as for the pre-dam conditions. Sluicing (passing through) is
another method of operation (partial water level drawdown during high inflows) which can
limit the outflowing sediment loads to those for typical natural conditions, but a long-term
equilibrium in sediment inflow and outflow cannot be established, since low inflow
conditions will normally not be sluiced especially in arid conditions in order to avoid risking
failure in water supply. Under such conditions sluicing only delays the rate of reservoir
sedimentation and it needs to be used in conjunction with flood flushing to maintain
substantial long-term reservoir capacity.

8
Figure 2.3-1 shows that reservoir operation with combined sluicing and flushing operation
should in the long term impact least on the sediment balance if practised regularly, coinciding
with high inflow conditions. Rapid changes in water quality with low concentrations of
dissolved oxygen and high suspended sediment loads during flood flushing/sluicing
(uncharacteristic of the natural river) need to be considered, however, in order to protect the
aquatic ecosystem downstream of the dam.

Density current venting of sediment laden flows can also limit the rate of sedimentation, but
very specific boundary conditions are required and therefore the general efficiency is less than
with flushing/sluicing. This is because of a number of reasons such as: only fine sediment is
transported through the reservoir with coarse sediment deposition where the density current
forms, as well as the judicious opening of suitable outlets to vent the density current. It also
requires excess water, but its key benefit is that the reservoir water level does not have to be
lowered.

For flushing/sluicing operation the major constraint is excess water availability, which means
that the reservoir has to be small in relation to the runoff if the water is used consumptively.
In practice, the most efficient passing through of sediment is obtained when the reservoir
capacity is less than 5 % of the mean annual runoff, although larger reservoirs are also sluiced
successfully. The Churchill (1948) and Brune (1953) empirical trap efficiency curves indicate
why reservoirs need to be so small.

The experience gained at Sanmenxia Reservoir, China, (Delft Hydraulics, 1992) with
different reservoir operating rules is further illustrated in Table 2.3-1. It was only after
reconstruction of the outlets that sediment sluicing could be optimized, with much reduced
operating water levels, but with the advantage of maintaining long-term reservoir capacity.
Figure 2.3-1 shows Sanmenxia Dam flushing after reconstruction of the outlets.

9
 Table 2.3-1. Reservoir operation and sediment trapping in Sanmenxia Reservoir

Period Operation* Maximum Minimum water Sediment outflow

water level (m) level (m) as % inflow
9/60 – 3/62 A 332.58 324.04 6.8
4/62 – 7/66 B 325.90 312.81 58
7/66 – 6/70 C 320.13 310.00 83
7/70 – 10/73 D 313.31 298.03 105
11/73 – 10/78 D 317.18 305.60 100
Note* A: Storing water
B: Flood detention and sluicing through 2 tunnels and 4 penstocks
C: Flushing by opening 2 tunnels and 4 penstocks
D: Flushing: 2 tunnels, 4 penstocks and 8 diversion outlets

Figure 2.3-2. Sanmenxia Reservoir flushing after reconstruction of the outlets

When the storage capacity-mean annual runoff (MAR) ratios of reservoirs in the world are
plotted against the capacity-sediment yield ratio, the data plot as shown in Figure 2.3-3.
Most reservoirs have a capacity-MAR ratio of between 0.2 to 3, and a lifespan of 50 to 2000
years when considering reservoir sedimentation.

10
When the capacity-MAR ratio is less than 0.03 especially in semi-arid regions, sediment
sluicing or flushing should be carried out during floods and through large bottom outlets,
preferably with free outflow conditions. Flushing is a sustainable operation and a long-term
equilibrium storage capacity can be reached. Seasonal flushing for say 2 months per year
could be used in regions where the hydrology is less variable with capacity-MAR ratios up to
0.2. Seasonal flushing can also be practised at these relatively high capacity-MAR ratios
when water demands and high sediment loads in the river are out of phase.

When capacity-MAR ratios are however larger than 0.2, not enough excess water is available
for flushing and the typical operational model is storage operation. Density current venting
can be practised at these reservoirs, as well as dredging to recover lost storage capacity.

The operating rules for a reservoir need not be inflexible, but can change with different stages
of storage loss. Storage operation may be continued in reservoirs with large capacities relative
to the sediment loads, while sluicing/flushing operation can be introduced once the loss of
storage capacity reaches a certain stage. These transition zones can be found between the
zones represented in Figure 2.3-3.

100000

100 %
10000
75 %
mean annual sediment yield (year)

1000
50 % dams within
Storage Capacity/

this zone

100 Storage

Flush
10

Flush Storage
0.1
0.001 0.01 0.1 1 10

Storage Capacity/MAR

Figure 2.3-3. Empirical reservoir classification system in terms of storage, runoff and
sediment yield

11
3 TURBULENT SEDIMENT TRANSPORT

3.1 Introduction

Experience in semi-arid climates indicates that in most rivers, the transport rate
of fine sediments is not limited by the hydraulic conditions, but rather by
sediment availability in the catchment. Within a reservoir this changes,
however, and it is possible to quantify sediment transport in terms of hydraulic
transporting capacity.

A large number of sediment transport equations have been derived during the
past century. In most cases the equations were "calibrated" by means of
coefficients with laboratory, and in some cases, field data. There are basically
two groups of equation formats: those that predict bed load and suspended load
transport separately, and those that predict a total sediment load without the
distinction between bed load and suspended load. Most equations have been
tested and calibrated for sand transport only, and the so-called "wash load" (fine
sediments) is not included.

For practical application, specific transport equations are favoured in certain

parts of the world. A number of recent studies have compared the accuracy of
these equations to predict sediment transport. Quite often, though, these
comparisons are biased towards those equations which incorporate some of the
verification data in their calibration. Nevertheless, what is important to note (as
is shown in Table 3.1-1), is the wide range of results which are obtained with
these equations, even under controlled laboratory conditions.

The reason for this is that the understanding of sediment transport processes has
not been developed well enough, even after many years of research in this field.
To name just a few of the complications, neither the interrelationship between
bedforms, associated roughness, hydraulic and sediment transport capacities,
nor the change in velocity profile when sediment are being transported have
been adequately modelled.

12
Due to the abovementioned, the approach in a number of countries is to
calibrate sediment transport equations for site-specific conditions. When "bed
load" is the main component or under other conditions where the sediment
transport capacity is the limiting factor, such an approach can be used
successfully. One set of such relationships is shown in Figure 3.1-1

Figure 3.1 - 1 Sediment transport relationships for various Chinese rivers

(Delft Hydraulics, 1992)

13
 Table 3.1-1 Accuracy ranges of commonly used sediment transport equations

White et al., (1975) compared eight formulae using 1 000 flume and 260 field measurements. The discrepancy ratio X calc / X obs was
plotted against the dimensionless grain size (ρ s qs / ρq ) and the percentages within the 0,5 to 2 range were as follows:
Formula % in 0,5 ≤ X calc / X obs ≤ 2 ranges
Ackers and White (1973) 68
Engelund and Hansen (1967) 63
Rottner (1959) 56
Einstein (1950) (total load) 46
Bishop et al., (1965) 39
Toffaletti (1968) 37
Bagnold (1966) (total load) 22
Meyer-Peter and Müller (1948) 10
The laboratory data include particle sizes from 0,04 to 4,94 mm and field data from 0,095 to 68 mm.
The comparison of formulae by Yang and Molinas (1982) also used laboratory and river data encompassing mean grain sizes from 0,15 to
1,71 mm, channel widths 0,134 to 532 m, flow depths 0,01 to 15,2 m, temperature 0° to 34,3°C, average velocity 0,23 to 1,97 m/s and
slopes from 4,3 x 10-5 to 2,79 x 10-2. The range of data is the same as given by Yang (1973) for the data from which the formula was
derived. The discrepancy ratio, defined as the ratio between computed and measured values, is given as follows:
Formula Data
Lab. River All data
Colby (1964) 0,31 0,61 0,34
Yang (1973) 1,01 1,31 1,03
Yang (1979) 1,02 1,12 1,03
Shen and Hung (1971) 0,91 1,18 0,95
Engelund and Hansen (1967) 0,88 1,51 0,96
Ackers and White (1973) 1,28 1,50 1,31
Maddock (1976) 0,99 0,49 0,92
A different picture is painted by the comparative study carried out by Van Rijn (1984b), also using field and laboratory data. The
discrepancy ratio, r, defined as the ratio of predicted to observed transport rates in per cent were as follows:
Data 0,75 ≤ r ≤ 1,5 0,5 ≤ r ≤ 2 0,33 ≤ r ≤ 3
1 2 3 4 1 2 3 4 1 2 3 4
US Rivers Corps Engrs 53 39 32 6 79 67 61 24 94 87 78 44
Middle Loop River 39 13 37 63 78 37 74 94 96 80 98 100
Indian Canals 30 15 27 3 60 45 48 6 90 73 70 24
Pakistan Canals 23 37 34 13 56 71 71 29 91 94 91 48
Niobrara River 55 13 29 86 95 67 58 98 98 95 98 98
Average of field data 45 32 32 22 76 64 63 39 94 88 84 55
Flumes
Guy et al., (1966) 40 67 56 68 70 89 85 90 91 98 99 98
Oxford 37 20 31 45 84 38 59 89 96 70 81 96
Stein (1973) 54 73 81 56 70 95 97 97 97 97 100 100
Southampton A 64 49 46 49 85 73 79 82 97 91 94 94
Southampton B 18 12 82 91 81 82 96 97 94 97 100 100
Barton-Lin (1955) 35 60 30 40 65 100 50 65 100 100 100 100
Average of laboratory data 41 46 52 59 77 74 77 89 95 89 94 98
Average of all data 43 37 40 36 76 68 68 58 94 88 88 71
In the above table, column 1 lists values obtained by the method of Van Rijn (1984 a & b); 2 by the Engelund-Hansen formula (1967); 3
by the Ackers-White (1973) formula and 4 by the Yang (1973) formula. The result is poor accuracy by the Yang formula for canals in
India and Pakistan, which have the deepest flows of the above data. Since the other formulae produce reasonable results Van Rijn
concludes that "the method of Yang must have serious systematic errors at large flow depth. On the average the predicted values are
much too small".

14
Most sediment transport equations incorporate only parameters which describe the basic
hydraulics and reflect how they are influenced by bedforms and transported sediment in
an indirect way. This is often done by determination of the bed roughness
independently from the sediment transport calculations. Some equations, like those of
Yang (1973), do not contain rules for calculating the velocity.

Instead of comparing the prediction accuracy of sediment transport equations based on

observed data, a more appropriate comparison would be to calculate the main hydraulic
variables from raw data, as was recommended by Van Rijn (1984b). A new approach
based on the interrelationship between hydraulic variables is included here. For
reservoir conditions, it is also important to be able to forecast the transport of fine
sediments (silt and clay), since in many impoundments they form the main sediment
body. The accurate prediction of sediment transport as well as sediment concentrations
is important since deposit shape, resuspension, flushing channel deformation, cohesion,
flocculation, etc., are all dependent on the sediment transporting capacities and particle
size sorting processes.

In practice, the prediction of the transport of fine sediments within reservoirs has been
based on different approaches:

a) Use of the diffusion equation.

b) Use of sediment transport equations which were calibrated originally for sand
fractions adopted to include fine sediments by recalibration with fine sediment
transport data.

c) Combinations of sediment transport equations for sand fractions and diffusion

equations for fine sediments.

Two further complicating matters in sediment transport are:

a) Non-uniform sediment particles. "Real" sediments are not nearly as

homogeneous in size as the sediments used in laboratories. The normal

15
 approach is to assume a representative sediment particle size or to model
different size groups, each with its own sediment transport. Non-uniformity
further leads to hiding/shielding and armouring phenomena involved in the re-
entrainment of sediment.

b) Non-equilibrium sediment transport. Sediment transport equations as derived

from theory and laboratory conditions are valid under steady, uniform flow,
equilibrium conditions only. With fine sediments being transported, however,
due to the low settling velocities, reaction to changing hydraulic conditions is
not immediate.

3.2 Review of selected equilibrium sediment transport equations

Where locally calibrated transport equations exist for reservoirs, they should obviously
receive precedence above formula which have been calibrated elsewhere.

In order to be able to express the sediment transporting capacity of a stream, a formula

needs to contain all the variables what play roles in determining transporting capacity,
particularly as far as hydraulic resistance is concerned.
Comprehensive analysis of variation in hydraulic resistance (Rooseboom & Le Grange,
1999) proves that a number of variables play important roles in determining flow
resistance, which in turn determines sediment transporting capacity.

These variables are:

D = flow depth
g = gravitational acceleration
w = particle settling velocity
d = particle diameter
s = energy gradient
gDs = shear velocity or;
ρgsD = shear stress
ρ = fluid density
ρs = sediment density

16
ks = absolute roughness

ν = kinematic viscosity

As a stream deforms its bed the value of k s can vary widely. Viscosity surprisingly
plays an important role through the development of laminar boundary layer conditions
when equilibrium transport conditions prevail.

Six existing transport equations, which are used for riverine transport calculations and
which are deemed to have merit for application on reservoirs have been selected. They
are discussed below, together with a new formula which includes all the important
variables and has been calibrated with reservoir data.
(i) Engelund and Hansen (1967) equations

By using dimensional analysis, Engelund and Hansen (1967) related input power per
unit area of channel boundary to sediment discharge and proposed the following
relationship:.

5
2GdS
2
φ = 0,10 2 (3.2-1)
V

where

Qs (3.2-2)
φ=
⎛γ s -γ ⎞ 3
⎜⎜ ⎟⎟ gd ;
⎝ γ ⎠

ρgDs
θ= ;
( γ s - γ )d and (3.2-3)

D = flow depth
g = gravitational acceleration
v = flow velocity

17
s = slope
Qs = total sediment discharge
ϒ = specific weight of water
ϒs = specific weight of sediment
d = sediment particle size

The total sediment discharge can therefore be calculated directly by writing Equation
(3.2-1) as
5/ 2
v2 ⎛γ s −γ ⎞ 3 ⎛ ρgDs ⎞
Qs = ⎜⎜ ⎟⎟ gd ⎜⎜ ⎟⎟ (3.2-4)
20 gDs ⎝ γ ⎠ ⎝ (γ s − γ )d ⎠

(ii) Ackers and White (1973)

Ackers and White (1973) used dimensional analysis to derive an equation representing
total sediment discharge in terms of three dimensionless numbers, viz. a sediment
transport function Ggr, a mobility number Fgr and dimensionless sediment particle size
dgr. The parameters are expressed as

n
⎛ F gr ⎞
m
c′D ⎛ gDs ⎞ (3.2-5)
G gr = c′ ⎜ - 1⎟ = ⎜ ⎟
⎝ A ⎠ γ s d/γ ⎜ v ⎟
⎝ ⎠
n 1- n
( gDs) ⎛ v ⎞
F gr = ⎜ ⎟ (3.2-6)
gd ( γ s / γ - 1) ⎝ 32 log 10 D/d ⎟⎠
⎜

1
(3.2-7)
⎛ g( γ s / γ - 1) ⎞ 3
d gr = d ⎜ ⎟
⎝ ⎠
2
v

where ν = kinematic viscosity.

The coefficients c ′ , A, m and n are functions of sediment particle size and have the
following values:

( )
For coarse sediment d gr > 60 : n = 0,0; A = 0,170; m = 1,50; c ′ = 0,025

18
whereas for smaller sizes (60 > d gr > 1) their values are given by:

n = 1 − 0,56 log d gr (3.2-8)

0,23
A= + 0,14 (3.2-9)
d gr

9,66
m= + 1,34
d gr (3.2-10)
(3.2-11)
log c′ = 2,86 log d gr - ( log d gr ) - 3,53
2

(iii) Unit (input) stream power (Yang, 1973, Rooseboom, 1975)

The basic principles of this approach were proven in South Africa in 1974 and
subsequently calibrated with field data from Gariep and Welbedacht Reservoirs. The
stream power principle has been used extensively in South Africa, USA and China
during the past 35 years in the planning and design of reservoirs.

The suspension theory (Rouse, 1937) can be extended to describe both bed load and
suspended load (the case of suspended transport with a relatively large z-value is
equivalent to the bed load condition) as well as the incipient motion criteria, and is
therefore well suited to the analysis of total carrying capacity (Rooseboom, 1975).

Sediment transporting capacity per unit width in terms of flow parameters can be
calculated if it is assumed that sediment particles are transported at the same velocity as
the fluid:

z
⎛ dv ⎞ dv
q s = ∫ C v dy with C ∝ ⎜⎜τ ⎟⎟
D
yo and τ the applied power, (3.2-12)
⎝ dy ⎠ dy

with C = sediment concentration

τ = bed shear stress

19
 w
z = suspension theory coefficient, =
κ gDs
w = settling velocity
κ = Von Kármán coefficient
and y0 the distance from the bed where v = 0 mathematically, which after integration
leads to an equation of the form (Rooseboom, 1975):

⎡ ∝1 ∝ 2 C 0 ⎤
⎢ z1 ⎥ (3.2-13)
(sD 2π gDs ) ⎥
log s = ∝ 1 log v s + log ⎢
q
q ⎢ y0 ⎥
⎢ ⎥
⎣⎢ ⎦⎥

with ∝1; ∝2 = coefficients

z1 = suspension theory coefficient as derived by Rooseboom (1975)
C0 = sediment concentration at the bed where
v = average flow velocity

Yang (1972) found through statistical analysis of available data that Equation 3.2-13
describes sediment transporting capacity particularly well. The analysis by Van Rijn
(1984b) later contradicted this finding, as shown in Table 3.1-1. The equation by Yang
(1972) was, however, only calibrated on laboratory data and as mentioned before, the
study by Van Rijn may have been biased. Yang used a slightly different approach by
including a critical stream power value for incipient motion in this equation (1972):

qs
log = ∝3 + ∝4 log ( v s - v scr ) (3.2-14)
q

with ∝3, ∝4
= constant and coefficient,
vscr
= critical input stream power for incipient motion

and later (Yang, 1973)

20
 qs ⎛ v s - v scr ⎞
log = ∝5 + ∝6 log ⎜⎜ ⎟
⎟ (3.2-15)
q ⎝ w ⎠

with ∝5 ∝6 = constant and coefficient

w = settling velocity

The last term in Equation 3.2-13 was found (Rooseboom, 1975) not to vary much and
can be equated to the ∝5 coefficient of Yang. Yang (1973, 1979) produced two
dimensionless unit stream power equations for sand transport, with and without
incipient motion. In 1984, Yang proposed a gravel transport equation with an incipient
motion term.

The maximum sediment transport capacity of a stream can therefore be determined by

an equation of the type:

qs
log = ∝ log v s + β (3.2-16)
q

where vs represents the average input unit stream power at a cross-section in a reservoir
or river, and ∝ = coefficient and β = constant.

(iv) Stream power theory (Bagnold, 1966)

Bagnold used the stream power per unit bed area, based on general concepts in physics,
to relate the rate of energy dissipation used in transporting materials to the sediment
transport:

⎛γ s -γ ⎞
⎜⎜ ⎟⎟ q bw tan ∝ = τ v e b (3.2-17)
⎝ γ ⎠

with qbw = bed load transport rate

tan∝ = ratio of tangential shear to normal force
v = average flow velocity
eb = efficiency coefficient

21
 τv = stream power

and for suspended load:

(3.2-18)
⎛γ s -γ ⎞ ⎛τ 2 ⎞
⎜⎜ ⎟⎟ q st = 0,01⎜⎜ v ⎟⎟
⎝ γ ⎠ ⎝ w ⎠

with the total load (qt )

(γ s - γ ) ⎛ v⎞
= qbw + q st = τ v ⎜ eb + 0,01 ⎟ (3.2-19)
γ ⎝ tan ∝ w⎠

Based on field data (rivers and reservoirs in China), the sediment carrying capacity =
eb(τv) =ƒ(ϒs qs) has been indicated in Figure 3.2-1 for conditions of deposition and
erosion.

Figure 3.2 - 1 Stream power sediment transport relationship at Fenhe Reservoir, China

(Gaun et al., 1991)

22
 (v) Gravitational power theory (Velikanov, 1954; Dou, 1974; Zhang, 1959)

Velikanov (1954) divided the rate of energy dissipation for sediment transport into two
parts: the power required to overcome flow resistance and the power required to keep
sediment particles in suspension against the gravitational force:

d[(1 - Cvy) U x U y]
ρg(1 - C vy) v y s = ρ v y + g( ρ s - ρ ) Cvy (1 - Cvy) w
dy (3.2-20)
(i) (ii) (iii)

with Cvy = time-averaged sediment concentration at a distance y above the bed in

percent by volume
vY = time-averaged flow velocity at distance y above bed
Ux, Uy = fluctuating part of velocities in the x and y directions

Integration over depth of flow yields

v ρ -ρ
3
g vSD = f o + s (gDw Cv) (3.2-21)
8 ρ

The Darcy Weisbach resistance coefficients are given by:

8gDs (3.2-22)
f= 2
, (Cv ≠ o)
v

and 8gD s0 (3.2-23)

f o= 2
, (Cv = o)
v

with s, so = energy slope with and without sediment.

Assuming that f / f k = a constant, (with f k = resistance coefficient for a saturated

sediment concentration), it follows that

23
(ρ s - ρ ) w C v =
ρ vs a constant. (3.2-24)

From the above, Velikanov's equation can be expressed in the form:

⎛ v3 ⎞ (3.2-25)
C v = K⎜⎜ ⎟⎟
⎝ gDw ⎠

with K = a coefficient to be determined from measured data. Sediment transport

equations of this format have been used extensively by Chinese engineers. Dou (1974)
proposed that the rate of energy dissipation to keep sediment particles suspended should
be equal to that used by sediment particles in suspension:

⎛ γs ⎞ vs
C t = K 1 ⎜⎜ ⎟
⎟w
(3.2-26)
⎝γ s -γ ⎠

with Ct = sediment concentration by weight

K 1 = coefficient

v2
and from Chezy : s= (3.2-27)
C2D

⎛ v3 ⎞
gives : C t = K 2 ⎜⎜ ⎟⎟ (3.2-28)
⎝ gDw ⎠

with K 2 = coefficient

Zhang (1959) assumed that the energy being dissipated in keeping sediment particles
suspended should come from turbulence instead of from the effective power available
from the flow:

24
 Total rate of energy dissipation of clear-water - rate of energy dissipation due to
sediment laden flow = total rate of energy reduction due to the damping effect
(reduction in turbulence due to suspended sediment).

γA v s1 - [γ (1 - Cv) Av s2 + γ s Cv Av s2] = k1 (γ s - γ ) Aw C∝v (3.2-29)

with S 2 , S1 = energy slope of sediment-laden water and clear-water

∝ = dimensionless exponent
Α = channel cross-sectional area
k1 = dimensionless coefficient

When C v is small:

∝
⎡ γ ⎤⎡v ⎤
Cv = ⎢ ⎥ ⎢ (s1 - s2)⎥
⎣ k1 (γ s - γ ) ⎦ ⎣ w ⎦ (3.2-30)

which can be further reduced to:

m
⎛ v3 ⎞ (3.2-31)
C t = K 3 ⎜⎜ ⎟⎟
⎝ gRw ⎠

with K 3 , m = parameters related to sediment concentration, and

RαD = hydraulic radius

Field data plotted according to this relationship are shown in Figure 3.2-2.

25
Figure 3.2 - 2 Gravitational power theory calibrated with Chinese data (Wu, 1994b)

The use of the stream power theory has been well researched and verified with field
data. Three different approaches in the study of sediment transport, based on the
concept that the rate of energy dissipation of flowing water should be related to the rate
of sediment transport, are generally used.

Yang and Kong, (1991) carried out an analysis and comparison of the three stream
power parameters
vs v3
;τv ; and
w gDw

The dimensionless unit stream power correlated best with concentration data by Stein
(1965), indicated in Figure 3.2-3. Velikanov's equation fitted with an S-curve instead
of a straight line on a log-log plot as is required by Equation 3.2-25. Bagnold's

26
 equation is based on general concepts in physics without using fluid mechanics theory
and it is not generally used, and was not reviewed further by Yang and Kong (1991).

Figure 3.2 - 3 Comparison between stream power equations (Yang and Kong, 1991)

Yang and Molinas (1982) showed that bed load, suspended load and total bed-sediment
concentrations can all be expressed by the general form of Equation 3.2-15. Yang and
Kong (1991) illustrated that the three equations based on the gravitational theory can all
be converted to or derived from the basic form of the unit stream power equation, with
differences in coefficients and assumptions to derive coefficients. The assumptions of
Dou, Velikanov and Zhang to obtain v 3 / gRw cannot be supported by laboratory or field
data used by Yang and Kong (1991). The unit stream power theory is therefore best to
use and based on a sound theoretical basis (Yang and Kong, 1991).

27
(vi) Van Rijn (1984a and b)

In the Van Rijn transport model the sediment load is divided between bed load and
suspended load according to the relative magnitudes of the bed shear velocity, and the
particle fall velocity. When the bed shear velocity exceeds the fall velocity, sediment is
transported as both suspended and bed load. Bed load is considered to be transported by
rolling and saltation and the rate is described as a function of saltation height. The
suspended load is determined from the depth-integration of the product of the local
concentration and flow velocity. The reference concentration is determined from the bed
load transport.

The bed load transport rate is computed as the product of particle velocity, ubs, saltation
height, δ b and the bed load concentration, Cb:

qb = u bs .δ b .C b (3.2-32)

Expressions for the particle velocity and saltation height were obtained by numerically
solving the equations of motion applied to a solitary particle. These expressions are
given in terms of two dimensionless parameters which are considered to adequately
describe bed load transport; a dimensionless particle diameter, D*, and a transport stage
parameter, T as defined below:

1/3
⎡ (s - 1) ⎤
D* = d 50 ⎢ 2 g ⎥
⎣ ν ⎦
(3.2-33)

2 2
( u _g ) - ( u _f,cr )
T=
( u _f,cr )2 (3.2-34)

in which u´g is the bed shear velocity, related to grains, u is the mean flow velocity and
u´f,cr is Shields critical bed shear velocity. u´g is the effective bed shear and is so defined

28
in order to eliminate the influence of bed forms since form drag was not considered to
contribute to bed load transport.

Extensive analysis of flume measurements of bed load transport yielded the following
expression for the bed load concentration:

Cb T
= 0.18
Co D* (3.2-35)

in which Co is the maximum bed concentration.

Combining equations for particle mobility, saltation height and Equation 3.2-35 gives
the following expression for bed load transport, valid for particles in the range 0.2 to
2.0 mm:

qb 0.053 T 2.1
= 0.3
(s - 1) gd 350 D* (3.2-36)

The Van Rijn suspended sediment load method is based on the computation of a
reference concentration determined from the bed load transport. Thus the reference
concentration, Ca, is described as a function of the dimensionless particle diameter, D*,
and transport stage parameter, T.

1.5
d 50 T (3.2-37)
C a = 0.015
a D*0.3

The representative particle size of suspended load is generally finer than that of bed
load. Van Rijn relates this particle size, ds to the d50 and geometric standard deviation, σs
of the bed material:

ds
= 1 + 0.011 ( σ s - 1)(T - 25) for T < 25 (3.2-38)
d 50

29
Many factors affect the suspension parameter z, e.g. volume occupied by particles,
reduction of fall velocity and damping of turbulence. These effects are grouped into a
single correction factor, ψ which is used to define a modified suspension number, Z ′
Z ′ = z +ψ (3.2-39)

ψ was found to be a function of the main hydraulic parameters:

0.8
⎡ w ⎤ ⎡C ⎤
0.4

ψ = 2 .5 ⎢ ⎥ ⎢ a ⎥ for 0.1 p1 (3.2-40)

⎣⎢ u f ⎦⎥ ⎣ C o ⎦

where Co is the maximum bed concentration.

By combining the expression describing the velocity and concentration profiles with the
expressions for z and ψ, Van Rijn (1984b) derived the following expression:

q s = FuDC a (3.2-41)

in which F is given by:

z′ 1/ 2
⎡a⎤ ⎡a⎤
⎢⎣ D ⎥⎦ − ⎢⎣ D ⎥⎦
F= z′
(3.2-42)
⎡ a⎤
⎢⎣1 − D ⎥⎦ [1.2 − z ′]

• It is recommended that at least three different formulae be used for

simulation of coarse sediment fractions in reservoir sedimentation
calculations so as to obtain a feel for the variability of the results.

• The stream power concept is powerful in describing the sediment

transport process. Most equations are, however, based on dimensional
analysis and not derived from sound theoretical principles.

• The unit input stream power equation (Yang, 1973), is the only stream

30
 power relationship which can be derived theoretically (Rooseboom,
1975), although some averaging assumptions had to be made. The
proven success of this formulae can be particially attributed to the fact
that it is based on pure hydraulic principles and that it contains a term
representing absolute roughness (yo) in its derived form (equation 3.2-
13).
• The Van Rijn equations are based on several empirical relationships and
is based on separate bed load and suspended transport which cannot be
justified from fundamental theory. The equations do, however, provide
for changes in bed roughness and energy dissipation for different flow
regimes and sediment transport.

• The prediction accuracy of most riverine sediment transport equations

seems to depend on calibration with as extensive a sediment transport
data base as possible (including laboratory and river data). Equations
which have been calibrated with reservoir data are obviously to be
preferred for reservoir modelling.

Sediment transport through reservoirs has traditionally only been described under
conditions of deposition. When management options such as flushing are
incorporated, flow through the reservoir reverts back to river conditions, but with high
sediment transport. A very wide range of sediment transport, hydraulic, bed
roughness and other conditions are therefore encountered.

Basson (1997) has developed a new sediment equation which contains all the important
variables involved in the analysis of sediment transport in rivers and reservoirs and
which could be calibrated for field as well as laboratory data and which was eventually
to be used for calibration with reservoir transport data.

The calibrated new sediment transport equation based on the non-cohesive sediment
river data of Gilbert, Guy et al. and Bagnold is given by:

31
 1,969 2,560
⎛⎛ ρ ⎞ ⎞ - 3,286 ⎛ ks ⎞
0,856
⎛ w ⎞
C = ⎜ ⎜⎜ ⎟ ( gsD )1,5 ⎟ . (0,4 )
-1,146
. . ⎜ ⎟ . ⎜ ⎟ (3.2-43)
⎜ ρ -ρ⎟ ⎟ k s w
⎝ ⎠ ⎜ 0,4 gsD ⎟
⎝⎝ s ⎠ ⎠ D ⎝ ⎠
with C = sediment concentration in % (weight).
The accuracy of the proposed new equation is shown in Table 3.2-4.

Table 3.2-4: Accuracy of new sediment transport equation in prediction ranges

(for C>0,01%)

Calibration based on Percentage of data within accuracy range

Ccalc 0,5 < Ccalc < 2,0 Ccalc

0,67 < < 1,5 0,33 < <3
Cobs Cobs Cobs

Gilbert and Guy flume 86 % 97 % 99 %

data

Bagnold USA river data 48 % 68 % 89 %

Gilbert, Guy and 77 % 92 % 98 %

Bagnold data

Although calibration data have been used in the above verification, the results compared
with other sediment transport relationships are good and even river data are predicted
with relatively good accuracy. Basson (1997) provides comprehensive details on the
verification of the new sediment transport equation against river and field data.
Observed versus calculated sediment transport based on equation 3.2-43 are shown
graphically in figures 3.2-4 and 3.2-5.

32
 Figure 3.3 - 4 Verification (with river data) of calibrated new sediment transport
equation (based on flume data)

Figure 3.3 - 5 Calibration of new sediment transport equation based on flume and river

data

It has been established that good correlations can be obtained with the new sediment
transport equation using flume and river data. The question remains, however, whether

33
 this relationship can also be applied to reservoir data. The reservoir data obtained from
field sampling at selected South African reservoirs, will be analysed here.

3.3 Calibration with reservoir data

There are a number of important differences between reservoir, laboratory and river
data. The latter two sets are normally obtained under uniform and steady (or nearly
steady) flow conditions, and with sediment transport and bed deformation equilibrium.
In a reservoir the two extreme operational situations are full storage operation with non-
uniform, but often near steady conditions in large reservoirs, and drawdown flushing,
normally with highly unsteady and non-uniform flow conditions during the retrogressive
erosion phase.

The particle sizes of sediments being transported through a reservoir furthermore also
differ drastically from the river and flume cases. Along the upper reaches of a reservoir,
coarse sediment is still being transported, representing river conditions, but further
downstream the sorting process soon causes only fine sediment to be transported. This
fine sediment (under storage operation with a large reservoir) can be in the order of
d50 = 2 micron, which is 10 times smaller than the finest sediments typically
encountered in flumes and rivers.

The slope to be used in the sediment transport equation should actually fall somewhere
between the energy slope and bed slope for a full reservoir, while during flushing the
energy slope should approach the bed slope and uniform flow conditions are approached
when equilibrium scour is being reached.

Unsteady flow conditions during flushing can cause over saturation, especially if
erosion takes place close to the dam and suspended sediment samples are taken
immediately downstream of the dam. On the other hand, non-equilibrium sediment
transport also occurs due to the time lag during which fine sediment loads adjust to
changing hydraulic conditions (see discussion in Section 3.5).
Obtaining field data during flushing is problematic. With complete drawdown, it is
impossible and dangerous to obtain concentration samples from the flushing channel,
and hydraulic variables are difficult to monitor within the reservoir.

34
Other processes that can play a role in reservoirs are different mechanisms of sediment
transport and flocculation at high sediment concentrations, which may influence settling
velocities.

All in all, reservoir data should be analysed and selected with great care. It is doubtful
that a relationship which has been calibrated for coarse sediments under uniform, steady
flow conditions can be used at all to predict sediment transport through a reservoir.
Therefore, the approach followed here was to calibrate the proposed sediment transport
relationship with reservoir data and to compare it with the other relationships as
calibrated with river and laboratory data. Under high inflow flushing conditions it is
expected that the reservoir flows will approach river conditions and that the river and
reservoir sediment transport relationships should agree. A sediment transport
relationship based on unit input stream power and storage operation for two reservoirs
of very different sizes as shown in Figure 3.3-1 has been used for many years in South
Africa.

To interpret the reservoir data sediment concentration was plotted against input stream
power (vs) in Figures 3.3-2 and 3.3-3. (Note that settling velocity still needs to be taken
into account to form a comprehensive picture of the observed sediment transport). After
each flushing and closure of the gates at Welbedacht Dam, the stream power decreases
rapidly, although sediment concentrations remain high for some time and only then drop
back to the "normal" vs against C relationship. It was found that the sediments being
transported immediately after closure of the dam outlets were much finer than with free
flow.

35
 Figure 3.3 - 1 Sediment concentration versus stream power for two South African

reservoirs (Rooseboom et al., 1986)

Figure 3.3 - 2 Input stream power versus observed suspended sediment concentration for
South African reservoirs

36
 Figure 3.3 - 3 Density current data at Welbedacht Dam

Reservoir data have been obtained at a number of reservoirs under variable operational
and flood conditions, as indicated in Table 3.3-1

vs
A check whether plotting versus C would not "correct" the relationship as indicated
w
by vs versus C for part of the data proved to be unsuccessful, and everything pointed to
the fact that density currents and not turbulent sediment transport had been present.
This data set was therefore removed from the data used for calibration of the new
transport equation and is discussed and analysed further in Chapter 4.

37
 Table 3.3-1: Reservoir data

No Reservoir Date Description

1a Welbedacht 06/02/94 Drawdown flood flushing
1b Welbedacht 08/02/94 Drawdown flood flushing
1c Welbedacht 19/10/95 Drawdown flood flushing
2a Phalaborwa Barrage 10/01/91 Drawdown flood flushing
2b Phalaborwa Barrage 08/02/85 Drawdown flood flushing
2c Phalaborwa Barrage 21/02/96 Drawdown flood flushing
3a Elandsdrift 04/12/93 Storage operation during minor flood
4a De Mistkraal 04/12/93 Storage operation during minor flood
5a Windsor 05/10//93 Drawdown flushing during low inflows
5b Windsor 03/02/94 Storage operation during flood
6a Mbashe 09/93 Empty reservoir with low inflows
6b Mbashe 25/10/95 Drawdown sluicing with low inflows
6c Mbashe 12/95-01/96 Drawdown flood flushing
7a Nagle 24/04/93 Drawdown flushing with low inflows
8a Gariep 1974 Storage operation during flood

Calibration with the remaining "accurate" selected reliable data gave a correlation
( )
coefficient r 2 of 0,80 using the new transport equation. In a comparable test using

⎛ vs ⎞
input stream power ⎜ ⎟ , similar correlation results were found, as indicated in
⎝ w⎠
Table 3.3-2. This latter equation as calibrated on reservoir data is given as equation 5.3-
1 in Chapter 5, where it is used in mathematical modelling.

Table 3.3-2: Calibration of stream power relationships

Sediment transport equation No of data Correlation coefficient r2

Applied stream power (new
transport equation) 180 0,80 (Figure 3.3-4)
Input stream power (vs/w) 180 0,72 (Figure 3.3-5)
Input (vs;w) 180 0,78 (Figure 3.3-6)

38
The best correlation is obtained with the new applied stream power relationship
(eg. 3.2-43, but with different coefficients). A relationship for input stream power as a
function, with the settling velocity separated, was also used, since it was found in this
research and by Yang (1973) that w is a dominant variable. Testing of accuracy ranges
is indicated in Table 3.3-3 (using the same data used for calibration).

Table 3.3-3: Accuracy ranges of calibrated new sediment transport equation for
reservoir data

Description Ccalc Ccalc Ccalc Ccalc

0,67< <1,5 0,5 < < 2 0,33 < < 3 0,25 < <4
Cobs Cobs Cobs Cobs

New equation 28% 46% 78% 91%

Input (vs/w) 27% 41% 66% 76%

Input = f (vs;w) 29% 49% 76% 86%

New equation 48% 68% 89% -

calibrated with
USA river data
(Bagnold)

The calibration previously obtained from USA river data is also shown in Table 3.3-3
for comparison. The variability in predictions of reservoir suspended sediment
concentrations is much larger than for river or flume data. The accuracy of laboratory
data normally allows a range of 0,5 to 2 of calculated/observed concentrations which
includes almost all of the data, while the comparable range for South African reservoirs
is at least 0,25 to 4. Nevertheless, the reservoir sediment concentration prediction is
better than expected, particularly if the high non-uniformity of flows is considered. (As
sediments are trapped where velocities are low, sediment concentrations in reservoirs
have to be correlated with minimum upstream vs values.)

In modelling, both a sensitivity analysis and reservoir-specific data should be used as far
as possible for the calibration of the sediment transport relationship. It is recommended
that the applied stream power relationship (new equation) be used in future modelling,

39
 preferably re-calibrated with local data. This relationship as calibrated here provides a
relatively good estimate over a wide range of sediment concentrations and hydraulic
conditions for South African reservoir data.

Figure 3.3 - 4 Calibration of new sediment transport equation based on reservoir data

Figure 3.3 - 5 Calibration of input stream power sediment transport equation using
reservoir data and not allowing for differences in settling velocity.

40
Figure 3.3 - 6 Calibration of input stream power sediment transport equation using
reservoir data and allowing for differences in settling velocity.

Mathematical modelling results of reservoir sedimentation (deposition) for the same

hypothetical boundary conditions, but with two different sediment transport equations
are presented in Figures 3.3-7 and 3.3-8, for the stream power sediment transport
equation calibrated on reservoir data, and the Engelund-Hansen equation, (arbitrary
selection as a typical equation used in rivers based on laboratory data calibration),
respectively. (The flow and sediment transport algorithms used in the modelling are
described in Chapter 5. Calibration parameters of the input stream power equation
based on South African Reservoir data are indicated in Chapter 5).

The reservoir data calibrated equation (Figure 3.3-7) shows sediment deposition much
deeper into the reservoir and a much less pronounced delta as in Figure 3.3-8.

Finally, a prediction of the reservoir data was carried out using the previously calibrated
new equation based on flume and river data. This was compared with the observed and
reservoir-calibrated new sediment transport equation. Figure 3.3-9 shows that sediment
transport can be accurately predicted with formulae calibrated on river or flume data
only at high concentrations (C> 3 %) and flow conditions approaching those of a river.

41
 At lower flow velocities and sediment concentrations (storage operation), sediment
transport can be underestimated considerably. This is also the reason why modelling
based on general laboratory/river data calibrations indicates such definite delta
formations with almost no sediment transport downstream of the delta. From the point
of view of reservoir sediment removal, the actual sediment transport through reservoirs
is orders higher than would normally be calculated by means of equations calibrated
with laboratory data.

Figure 3.3 - 7 Reservoir sedimentation based on stream power sediment transport

equation calibrated with reservoir data

42
Figure 3.3 - 8 Reservoir sedimentation based on Engelund-Hansen sediment transport
equation

Figure 3.3 - 9 Verification of reservoir sediment transport with new sediment transport

equation based on laboratory and river data calibration

43
The new sediment transport equation 3.2-43 as calibrated with reservoir data can be
used to predict non-uniform sediment transport, but only within the calibrated particle
size ranges. Reservoir sedimentation simulation based on the stream power equation
(calibrated with reservoir data), has been carried out with uniform and non-uniform
sediment particle size distributions (with the same boundary conditions), and the results
are shown in Figures 3.3-10 and 3.3-11. (Details of the mathematical model flow and
sediment transport algorithms are described in Chapter 5). The differences in the
sedimentation profiles as simulated in Figures 3.3-10 and 3.3-11 emphasize the
importance of accurately predicting the transport of non-uniform sediments. Accurate
determination of sediment sizes is also important when considering sediment density,
consolidation, critical conditions for mass erosion of non-cohesive or cohesive
sediments, etc.

Figure 3.3-10 Reservoir sedimentation profile with uniform sediment size

44
 Figure 3.3-11 Reservoir sedimentation profile with non-uniform sediment size

3.4 Non-equilibrium sediment transport

3.4.1 Introduction

Most mathematical models are based on the assumption that the difference in the
sediment loads between successive cross-sections is deposited (or eroded) within each
reach. A state of sediment equilibrium is therefore reached within each time step of the
calculation. Equilibrium in this case refers to the actual sediment transport being equal
to the transport capacity at a section.

Instantaneous adjustment of the bed profile is a realistic assumption only when coarse
sediments are being transported and without any constraints on sediment availability.
With fine sediments, however, adjustment according to the saturated sediment transport
capacity is not instantaneous and time and distance lags are found with changes in
sediment transport, until equilibrium is reached. This lag, often called "adaptation

45
length", is due to the low settling velocities of the fine sediments. In this adjustment
process the bed roughness, the energy dissipation rate ( κ ) and the sediment transport
capacity change until equilibrium is reached, with minimized stream power.

Two modes of non-equilibrium sediment transport can be identified:

• Undersaturated: Sediment discharge could be availability limited

e.g. due to limited surface erosion, while mass
erosion of the bed can lead to rapid transition
from the under saturated load condition to the full
sediment transporting capacity. Some of the
coarser fractions suspended during mass failure
might be oversaturated and will deposit again.
Processes which involve scour are discussed in
more detail in Basson and Rooseboom (1997).

• Oversaturated: Reservoir sediment deposition processes are

often associated with conditions of over
saturation as the sediment transporting capacity
diminishes through the reservoir. With fine
sediments and a deep reservoir, adaptation
lengths can be longer than the reservoir length.
Different sediment particle sizes will have
different adaptation characteristics.

Using a rotating annular channel-ring system, Mehta and Partheniades, (1973) showed
that suspended sediment concentration diminishes rapidly from an initial value, C o to a

constant value, C eq defined as the "equilibrium concentration", although for the same

flow condition various "equilibrium concentrations" were obtained as Ts shown in

Figure 3.4-1.

46
C eq decreases with decreasing bed shear stress, becoming zero for a threshold value,

τ b min , of the latter. Figure 34-1 shows that for a specific test, the relative equilibrium

C eq
concentration, C eq* = , remains constant and independent of C o (Partheniades,
Co

1986). C eq* was found to depend on τ b as Ts shown in Figure 3.4-2.

Figure 3.4-1 Variation of suspended sediment concentration with time

(Partheniades, 1986)

The time rate of deposition can be represented by C o − C eq the depositable part of the

sediment load. The best fit was found with C * = (C o − C ) / (C o − C eq ) and t / t 50 with

t 50 the time at which C * = 0.50 on a log-normal scale. (See Figure 3.4-3.)

47
Figure 3.4-2 Relative equilibrium concentration versus bed shear stress parameter

(Mehta and Partheniades, 1973)

Figure 3.4-3 Deposition rates (Mehta and Partheniades 1973)

48
 In terms of sediment transport theory, the adjustment to "equilibrium" sediment
transport under the same flow conditions, thereby reducing the suspended sediment load
as shown in Figure 3.4-1, can be attributed to adjustment of the stream power to
minimize energy dissipation. The adjustment is not immediate, due to the relatively low
settling velocities of the cohesive sediments. The "equilibrium" condition that is
reached does not represent equilibrium sediment transport (or maximum transport), but
rather an equilibrium state under conditions of limited sediment availability.

Mathematical descriptions of the non-equilibrium transport process have been given by

Galapatti and Vreugdenhill (1985) and Di Silvio (1995), but the key variables involved
in minimization of stream power were not included. Chinese researchers (Han and He,
1990), have calibrated non-equilibrium equations with field data and have established
criteria for non-equilibrium sediment transport calibration coefficients for rivers and
reservoirs. Most of these equations are of the same format as the steady
advection-dispersion equation.

3.4.2 Review of existing theory

Until recently the "best" equilibrium transport formula was selected among dozens of
such equations in the literature. Sediment transport in real life, especially in reservoir
storage operation, is often not in equilibrium.

Equilibrium formulae based on uniform flow experiments in hydraulic laboratories are

no longer considered as satisfactory components in mobile-bed modelling systems.
Time and space lags between actual and equilibrium transported sediment loads should
therefore be considered (Cunge, 1989).

The basic equation of 2D diffusion of sediment concentration can be written in the form
(Zhang, 1980):

δC δ 2C + w δ C
v =ε (3.4-1)
δx δ z2 δz

with C ( x, t ) = sediment concentration

49
 ∈ = coefficient of turbulent exchange
w = settling velocity of sediment particles
v = flow velocity

An analytical solution of Equation 3.4-1 is possible with the following boundary

conditions for the case of deposition:

δC
at the water surface : z = D, ε + wC = 0
δz

δC w
on the reservoir bed : z = 0, = − Cκ ,0 = constant
δz ε

at entrance : x = o, C = C o f (z )

Zhang (1980) derived the analytical solution :

∞
2 ∝2n
C cp (x) = C Kp + ( C o - C kpo ) ∑ (3.4-2)
n=1 ( ∝n / K 1 + K 1 /4)( ∝n + K 1 /4 + K)
2 2 2 2

which, after differentiation and simplification, reads:

dC cp (x)
= ∝ w[ C Kp - C Cp (x)]/q (3.4-3)
dx

By substituting initial conditions x = 0, C cp ( x ) = C 0 , the final expression of the rate of

change in sediment concentration along the reservoir is:

q s = q st + e (−αwx / q ) (q so − q st ) (3.4-4)

with qs = sediment discharge per unit width at the exit

q st = sediment-carrying capacity

q so = sediment inflow per unit width at the entrance

50
If ζ = e (−αwx / q ) denotes sediment transport under non-equilibrium conditions
(0 < ζ < 1) , when x = 0, ζ = 1,0, q so = q so no net deposition or scouring occurs and

the incoming sediment discharge equals the outflowing discharge. When

x = ∞, ζ = 0, q = q s it means that outflowing sediment discharge equals the sediment-
carrying capacity after self-adjustment along the river course.

Computation of sediment transport under equilibrium conditions is a special case of

non-equilibrium conditions. Generally, the equilibrium can only be re-established over
a long distance (for fine sediments) (Zhang, 1980).

In the case of scouring, similar equations can be derived (Zhang, 1980)

The boundary conditions are: Z = 0, C ko = Const.

δC
Z = D,∈ ± wC = 0
δz

x = o, C = Const.

and the final equation: q s = q st - e(-∝ wx/q) ( q so - q st )

In the case of deposition α 1 = 1 + K 1 / 2 ≈ 1 (3.4-5)

and ζ 1 = e(− wx / q )

while with scouring

∝2 = π / K 1 + K 1/4
2

= π / K1

= K v* π 2 / 6 w

v * = gDs

51
 = K 0,5 g 0,5 Q0,3 s0,5

Final expressions with the inclusion of empirical coefficients (K 3 , K 4 and K 5 ) read:

(
Deposition: ζ 1 = exp K 3 , C K 4 (0,41) − 0,77 / gC x / q ) (3.4-6)

Erosion: ζ 2 = exp (− K 5 Q 0,3 s 0,5 x / q ) (3.4-7)

Soares et al., (1982) derived equations for non-equilibrium suspended sediment

transport, similar to those of Zhang (1980).
The mass balance of sediment of a given size dj is:

⎛δ C j ⎞
Q ⎜⎜ ⎟⎟ _x = (- q d + qe )T_x
j j
(3.4-8)
⎝ δ x ⎠

with q dj , qej = fluxes of deposition and erosion of sediment

Q = flow rate
T = width
δC
is neglected in this equation
δt

Let CT * j = average concentration according to sediment transporting capacity, then:

if C i j > CTi +j1* (deposition):

j j j*
Rate of deposition : q d = - w j ( C - CT ) (3.4-9)

δ Cj
and Equation 3.4-8 becomes = = - w j ( C j - CT j* ) T/Q (3.4-10)
δx

Integrating Equation 3.4-10 between two sections:

C i+1 = ( C i - CT i+1 ) exp (- w j ΔxT/Q)

j j *
(3.4-11)

52
if C i j ≤ CTi +j*l (erosion will occur depending on the availability of sediment of the given

size class at d j on the stream bed)

Rate of erosion: q ej = λ (C bj − C j ) (3.4-12)

The erosion rate is a function of the difference between the availability on the bed and
the concentration carried by the flow. Although erosion is dominant, deposition will
still occur at a rate w j C j and Equation 3.4-8 becomes:

δ Cj λj j T
= -( C j - Cb ) ( λ j + w j ) (3.4-13)
δx λ j + wj Q

Integration of Equation 3.4-13 between sections i and i + 1 yields:

⎛ j λ j ⎞ ⎡ TΔx ⎤ λj
C i+1 = ⎜⎜ C i - C b ⎟⎟ exp ⎢- ( λ j + w j )
j j j
⎥ + Cb (3.4-14)
⎝ λ j + wj ⎠ ⎣ Q ⎦ λ j + wj

When Δx is large, discharge in section i + 1 will approach the transport capacity

:

j* λj j
Therefore CT i+1 = Cb
λ j + wj

⎡ TΔx ⎤
and the final equation: C i+1 = CT i+1 + (C i - CT i+1) exp ⎢ - ( λ j + w j )⎥
j j* j j*
(3.4-15)
⎣ Q ⎦

Equation 3.4-16 is similar to the equation presented by Karanshev (1963):

⎡ TΔx ⎤
C i+1 = CT i+1 + (C i - CT i+1) exp ⎢ -
j j* j j*
(1 + K j ) w j ⎥ (3.4-16)
⎣ Q ⎦

and also similar to that of Zhang (1980):

q s = q st - ( q so - q st ) . exp (- ∝ wx/q) (3.4-17)

53
Sundborg (1964) developed another non-equilibrium equation which does not allow for
erosion:

C i+1 = C i exp [- (TΔx/Q) wφ (w)]

j j
(3.4-18)

Equation 3.4-18 was used by Hurst and Chao (1975) in a model for Tarbela Reservoir.
The equation implies that if in section i the concentration of sediment of size dj is zero,
the same will be true for all downstream sections, which is clearly an error.

While the transport of coarse sediment depends exclusively on local hydrodynamic

conditions, the transport of fine particles also depends on the conditions upstream. Di
Silvio (1995) proposes the use of the following non-equilibrium transport equation:

δ Ci 1 (3.4-19)
= * ( β i C*ci - C i )
δx L i

with Ci = the actual sediment transport per fraction

βi = the % of i-th fraction in the bed composition

L*i = the adaption length for each fraction

C ci* = the transport capacity of the i-th class

The adaptation length L*i can be obtained either experimentally or by an asymptotic

solution of the 2D suspended transport equations (Di Silvio and Armanini, 1981;
Galapatti and Vreugdenhil, 1985.) An evaluation of L*i is given by the following
approximate formula:

Li wi = k s + (1 - k s ) exp ⎛⎜ - 1,5 ⎛ k s ⎞ . w ⎞⎟
* -1/ 6

⎜ ⎜ ⎟
vD D D ⎝ ⎝D⎠ v* ⎟⎠ (3.4-20)

54
 vD
For fine particles, L*i particle falling distance = , which means that the adaptation
wi
lengths for silt and clay may be even larger than the reservoir length.

δβ i C ci
For coarse sediment, L*i → 0 , and C i to β i C ci* and the erosion rate becomes
δx
as with equilibrium transport and instantaneous adaptation.

All the above non-equilibrium sediment transport equations have the same format and
were derived for steady flow conditions.

3.4.3 Modelling of non-equilibrium sediment transport processes : Welbedacht

Reservoir (Caledon River, South Africa)

In reservoirs, inflows, water levels and inflow sediment concentrations are highly
variable and unsteady, and non-uniform flow conditions prevail especially during floods
when high sediment loads are being transported into a reservoir. Testing of the
proposed non-equilibrium sediment transport approach has been carried out for
Welbedacht Reservoir for the period 1973 to 1976. The simulation runs started in
August 1973 when the first fill of the reservoir commenced. Observed instantaneous
inflows and sediment concentrations, and water levels at the dam have been used as
boundary conditions. The simulations continued to October 1976 when the first basin
survey was carried out. During this three-year period major floods occurred in the
Orange River system and 36 million m3 of storage capacity was lost in Welbedacht
Reservoir due to sedimentation.

Using a typical suspended sediment particle size distribution that has been observed at
the upper end of the reservoir, one equilibrium and 2 non-equilibrium fractions have
been modelled as indicated in Table 3.4-1.

The simulated sedimentation profiles for Welbedacht Reservoir are indicated in Figure
3.4-4.

55
 Table 3.4-1: Suspended sediment size distribution of Welbedacht Reservoir inflow

Fraction no Particle size range (mm) Percentage in size range

3 0,106 - 0,25 5

2 0,05 - 0,106 19

1 < 0,05 76

Figure 3.4-4 Non-equilibrium sediment transport calibration: Welbedacht Reservoir,

1973 to 1976

The simulated and observed sedimentation profiles as indicated in Figure 3.4-4 show
1
that an adaptation coefficient for the fine sediments of "a"≥ 2,8 (a = ) gives a reliable
α
sedimentation profile prediction (Also refer to equation 5.3-3, Chapter 5). There are,
however, many uncertainties involved such as:

56
 • unknown sediment density

• cross-sectional deformation, assumed here proportional to flow depth for erosion

and deposition

• changing suspended sediment size distributions of inflows during low and high
flows

• sediment cohesion and consolidation, limiting re-entrainment during floods.

3.4.4 Comparison between calibrations

Han and He (1990) give typical calibrated adaptation coefficient values for Chinese
field data as:

Reservoir sedimentation (deposition) : α = 0,25

Reservoir flushing with fine sediments : α = 1,0
Rivers with coarse, non-uniform sediments : α = varies for each fraction

These adaptation coefficients calibrated with Chinese reservoir data agree very well
with the calibrated flume adaptation coefficients which varied between a = 2,0 and 2,8.
The differences between Chinese and South African calibrated adaptation parameters
can be ascribed to:

- different fraction size ranges

- different sediment characteristics
- different equilibrium sediment transport equations for fine sediments.

57
4. DENSITY CURRENTS

4.1 Introduction

Apart from turbulent suspended sediment transport, which is the dominant mechanism
by which sediment is transported through most reservoirs, density currents in certain
reservoirs provide an additional mechanism for transporting sediments. Density
(gravity) currents are flows driven primarily by a difference in density between the
current itself and its surroundings. The difference in density can be caused by
temperature, chemical species (pollutants) or material phases (sediments).

Figure 4.1-1: Density contours of the head of a gravity current entering a reservoir.

Figure 4.1-2: Density contours of the body and head of a gravity current moving

The monograph of Simpson (1997) provides many examples of natural and man-made
gravity currents. A river entering a reservoir can often form a density current. The
density difference can be created by transporting water that is warmer or cooler, more
saline or less saline, more turbid or less turbid than that of the reservoir. A density
current is not that different from normal open-channel flow: in the latter case air
replaces the overlaying fluid body and the stream can therefore also be seen as a
"density current"; only in the case of air above a water stream, the density difference is
so large that the inertial effects of the air compared with those of water may be
neglected.

58
 Density currents are not only found when liquids of different densities move relative to
each other. The movement of moist air in the form of clouds spilling down a mountain
valley or dust rolling down a slope are common examples of density currents. When
one analysis the behaviour of density currents related to sediment transport through
reservoirs (also known as turbidity currents), a number of questions come to mind:

- Why do density currents occur?

- Under what conditions will density currents form?
- How can the density current movement be described mathematically?
- What is the relationship between sediment transport and the hydraulics of a
density current?
- Can density currents be utilized effectively as a means of passing sediment
through a reservoir without deposition?

4.2 Occurrence of density currents in reservoirs

The formation of density currents was first observed at the beginning of the 20th century
in some of the world's large reservoirs. Along the upper reaches of a reservoir, stable,
floating debris is observed, indicating the so-called "plunge point" where the inflowing
river stream changes into a density current (Figure 4.2-1). The stationary nature of the
debris is caused by the slow upstream movement (or near zero velocity) of the overlying
water mass, just downstream of the point where the sediment-laden inflow dives below
the stored water mass. Apart from the evidence at the plunge point, nothing can be seen
of the density current, except when it exits the reservoir through suitable low-level
outlets provided that the density current reaches the dam. Under unfavourable boundary
conditions, the density current could either be broken up through turbulent mixing in
certain parts of the reservoir or through deposition of sediment, whereby the density
difference which drives the density current, decreases.

59
 Figure 4.2 - 1 Plunge point at Eril Emda Reservoir

Density currents do not only occur in reservoirs on heavily silt-laden rivers such as the
Eril Emda Reservoir in Algeria, Lake Mead in the USA or Sanmenxia Reservoir in
China, but also in reservoirs with flows containing low sediment concentrations such as
the Sautet Reservoir in France.

Density currents may travel long distances through reservoirs, for example over 100 km
in Lake Mead and 80 km (1961) in the Sanmenxia Reservoir.

Typical density current velocity and suspended sediment profiles are shown in Figure
4.2-2 as observed in Gaunting Reservoir, China and in Figure 4.2-3 as observed in
Sanmenxia Reservoir, China (Fan, 1986).

60
 Figure 4.2 - 2 Density current in Guanting Reservoir (Fan, 1986)

Figure 4.2 - 3 Density current in Sanmenxia Reservoir (Fan, 1986)

It was found in practice that a density current moves at a slow speed, typically of the
order of 0,1 to 0,3 m/s (higher in some cases), and yet is able to transport high sediment
loads for distances of over 100 km through a reservoir. As will be shown later, these
low velocities are related to the density difference between the lower denser fluid and
the upper "clear" water, which effectively reduces the effective gravitational
acceleration by 100 to 1 000 times. The effect of this is that the flow of the density
current can be described as if in "slow motion". In flume experiments it was noticed

61
that a density current can "jump" over obstacles in its path, due to the low relative
"gravity" (g). Around bends the surface level increase is often so much that fluid spills
out of the main channel and at the downstream outlet the density current can flow
vertically upwards provided that the outlet is correctly sized and is at a certain height.

Density currents have often been associated with delta formation in reservoirs. It was
and is still believed that a delta provides the ideal boundary conditions for a density
current to form. The authors, however, believe that in fact a density current can lead to
the formation of a delta. As will be shown later, more favourable conditions for the
formation of a density current exist without a delta.

Graf (1983) stated that density currents in reservoirs typically transport sediment
particles smaller than 20 micron in diameter with settling velocities of less than
0,03 cm/s, which implies that a density current velocity of 0,03 cm/s is required to
maintain suspension of the sediments. He stated that density currents will be either
depositing, eroding or steady-state, with equilibrium between rates of bed erosion and
bed deposition existing in the latter case.

It will be shown in this chapter that density current sediment suspension is similar to
that in open channel flow. Deposition of coarse sediments normally occurs at the
plunge point due to insufficient transport capacity of the density current, while the
further movement of the density current can be associated with deposition of sediment
(at underwater obstacles), or with steeper slopes erosion may occur depending on the
critical shear stress and sediment availability.

At the plunge point, transition occurs from a normal turbulent stream to a density
current, with continuity in discharge being maintained. Not all incoming sediment is
however removed by the density current, as the larger particle sizes are immediately
deposited, and invariably in reservoirs all over the world it has been found that only the
finest sediment fractions of silt and clay are transported by density currents through the
reservoirs. This means that although the settling velocities of the particles are much
reduced through the reduction in effective gravity, the transport capacity of a density
current is still not sufficient for the transport of all incoming sediments.

62
Past experimental work on turbidity currents include:

Parker et al. (1987) investigated steady, supercritical turbidity current flow over an erodible
bed. One of objectives of the experiments was to create a self-accelerating turbidity current.
The flume used was 0.7 m wide, 1.7 m deep, 20 m long having slopes of 0.05 and 0.08.
Measurements were made using micro propellers and siphons. Their results include self-
similar profiles of mean velocity and concentration, as well as relations for water entrainment,
sediment entrainment and bed resistance. These relations are typically required for closing
depth-averaged numerical models. They concluded that self-acceleration is possible, even
though they were not able to produce such a flow.

Garcia and Parker (1993) also investigated sediment entrainment by a steady supercritical
saline underflow. The flume used was 0.3 m wide, 0.78 m deep, 12.8 m long having a slope of
0.08 for the first 5 m. Their results include profiles of mean velocity and concentration, as
well as relations for sediment entrainment and boundary shear stress. They concluded that in
addition to purely sand-driven currents, currents of silty mud could also entrain substantial
amounts of sand and carry it to deep waters.

Garcia (1993) also investigated hydraulic jumps of steady supercritical turbidity and saline
currents. The flume used was 0.3 m wide, 0.78 m deep, 12 m long having a slope of 0.08 for
the first 5m. Measurements were made using micro propellers and an optical probe for
concentration. Results include self-similar profiles of mean velocity and concentration, as
well as sediment deposition profiles. The results showed that the vertical structures of saline
and fine-grained turbidity currents, having similar inlet conditions, are approximately the
same before and after a hydraulic jump. It was also observed that the greatest amount of water
entrainment occurred in the supercritical flow region.

Lee and Yu (1997) investigated steady turbidity and saline currents. The flume used was 0.2m
wide, 0.6 m deep, 20 m long with a slope of 0.02. Measurements were made using a magnetic
current meter and siphons. Their results include self-similar profiles of mean velocity and
concentration, as well as criteria for stable plunge point prediction.

Choux (2005) investigated the spatial and temporal evolution of a quasi-steady turbidity
current having an initial volumetric concentration of 14%. The density difference of the

63
turbidity current was 231kg/m3. The flume used was 0.3 m wide, 0.3 m deep and 10 m long.
Measurements were made using ultrasonic doppler velocity profilers and siphons. The current
was quasi steady, because a fixed amount of mixture was allowed to drain from an overhead
tank into the flume. The results include: the spatial evolution of the mean velocity profile
within the head, body and tail; the spatial evolution of the rms velocity profile within the
head, body and tail; the spatial evolution of the mean grains size profile within the head, body
and tail; the spatial evolution of the mean concentration profile within the head, body and tail.
The results suggest that self-similarity holds for turbidity currents having a mean
concentration less than 7%. The rms velocity measurements show that the turbulence intensity
is greatest within the head and at the base of the flow, while being the weakest at the velocity
maximum and tail. It was also found that grain sizes are well mixed within the head, while
normal grading occurs within the body and normal to inverse grading occurs within the tail.

4.3 Hydraulics of density currents

4.3.1 General

In the previous sections, density currents and possible reasons/explanations for some of
the associated phenomena were given in general terms. It is necessary to obtain a
proper mathematical understanding of the movement of a density current and also to
review some of the theories proposed by other researchers in the past, as no generally
accepted theory exists.

The vertical velocity, suspended sediment concentration and shear stress distribution in
density current flow, as indicated in Figure 4.3-1, will form the basis of discussions in
the following sections.

64
 Figure 4.3 - 1 Density current velocity, suspended sediment concentration and shear
stress distributions

4.3.2 Velocity distribution

In laboratory and field studies, the vertical velocity distribution has been found to be
increasing logarithmically in layer 1 from the bed to the maximum velocity level
(between layers 1 and 2) (Figure 4.3-1). (Ashida and Egashira, 1975; Chikita, 1989)
From this point upwards in layer 2, the profile again seems to be decreasing
logarithmically until a sudden change in slope at the so-called "inflection point" between
layers 2 and 3 is reached. Through layer 3 the velocity decreases upwardly until it
reaches zero at the point of contact where the velocity becomes negative due to
upstream flow in the upper reservoir layer. In layer 4 the velocity is often zero or
slightly negative with a typical open-channel velocity distribution.

4.3.3 Vertical suspended sediment distribution

The sediment distribution should be seen in relationship to the velocity distribution.

(Figure 4.3-1). In the lower layer 1, the sediment distribution is similar to that found
under turbulent flow conditions. Normally the sediment is also uniformly distributed
due to the small particle sizes. The reduction in the z value is due to the low settling
velocity and related change in distribution as explained in Chapter 3, with

65
 w
z=
κ gD1 s

Just above the maximum velocity level, the sediment concentration decreases rapidly to
almost zero near the inflection point in the velocity profile, which is also normally
regarded as the upper boundary of the density current. Only a small quantity of
sediment is diffused into layer 3 and is transported in this layer.

In terms of our knowledge of turbulent suspended sediment transport, it is easy to

explain the rapid reduction in sediment concentration in layer 2. The relative suspended
concentration for turbulent open channel flow is given by:

x
⎛ dv ⎞
⎜τ ⎟
C ⎜⎝ dy ⎟⎠
= z
Co ⎛ dv ⎞
τ ⎜⎜ ⎟⎟ (3.8-1)
⎝ dy ⎠ o
z
⎛ dv ⎞
∴ Cα ⎜⎜τ ⎟⎟
⎝ dy ⎠

dv
which means that as the maximum velocity is reached at the top of layer 1, → 0 and
dy
τ → 0 and therefore C → 0 . It is thus only the turbulence along the boundary of the
density current which carries any suspended sediment above layer 1. In theory the
suspended sediment transport in layer 2 should be zero, caused by the "barrier" of
dv
τ → 0 at v = v maximum between layers 1 and 2.
dy

4.3.4 Shear stress distribution

Within the lower two layers (1 and 2), the shear stress varies linearly (Figure 4.3-1)
(Ashida and Egashira, 1975). In layer 3 a non-linear shear stress distribution is found,
while at the top in layer 4, the distribution is linear as in open channel flow.

66
 Although layers 1 and 2 are normally described as the density current, it is believed that
at least layers 1, 2 and 3 should be considered in the mathematical description of a
density current. From continuity:

q = v1 D1 + v2 D2 + v3 D3 (4.3-1)

and

D1 D2 D3
q c = ∫ v1C1 dy + ∫ v 2 C 2 dy + ∫ v3 C 3 dy
yo D1 D2
D1
(4.3-2)
= ∫ v1C1 dy
yo

with y measured from the bed and if it is assumed that layers 2 and 3 make insignificant
contributions to the sediment load. Therefore all three layers form part of the density
current.

4.4 Mathematical description of the velocity distribution and the thickness of a density
current

Within the lower layer with thickness D1 , Ashida and Egashira (1975) have found a
logarithmic velocity distribution in laboratory tests (see Figure 4.4-1) and therefore
adopted a velocity distribution equation of the form:

v ⎛ v* . k s ⎞ 1 y (4.4-1)
= Ar ⎜ ⎟ + ln
Δρ g D1 s ⎝ ν ⎠ κ 1 ks
ρ

67
Figure 4.4 - 1 Velocity distribution from channel bed to maximum velocity

(Ashida and Egashira, 1975)

In layer two, at the interface, Ashida and Egashira related the mixing length (lo ) to the
layer thickness (D2 ) and found the relationship indicated in Figure 4.4-2.

Figure 4.4 - 2 Relationship between mixing length l o at interface and D2

(Ashida and Egashira, 1975)

68
 With l = lo + κ 2 (D1 + D2 − y ) at the interface, Ashida and Egashira again assumed a
logarithmic velocity distribution for layer 2. By making further assumptions regarding
the shear stress distribution in layer 3 (see Figure 4.4-3), they derived a velocity
distribution relationship for layer 3 and together with assumed sediment concentrations,
the discharge and sediment transport continuity equations could be solved
mathematically.

Ashida and Egashira were some of the first authors to describe the velocity, shear stress
and suspended sediment distribution in a density current. Some of the assumptions
made by them can, however, be criticized. The assumed relationship between l o and
D2 is not based on theory and need therefore not be true for all data, as is found with a
different data set in Figure 4.4-2. The mixing length (or R) at the interface should
rather be calculated theoretically by using boundary layer theory as will be shown in the
next section.

Figure 4.4 - 3 Distribution of velocity, shear stress and sediment concentration

(Ashida and Egashira, 1975)

In density current flow, three boundaries can be observed: at the bed, at the top of layer
2 where the suspended sediment concentration → 0 and at the level where v → 0.

69
With certain assumptions, it should be possible to describe the velocity distribution and
layer thicknesses in a density current mathematically. The knowledge of laminar and
turbulent boundary conditions can be used effectively to describe the mathematical
relationships. Unlike the case of flow, the velocity distribution relationship for a density
current cannot be established without considering the influence of sediment transport.

The suspended sediment transport in a density current affects the energy dissipation rate
as represented by Kappa (κ ) as follows: (Refer to Figure 4.3-1)

a) At the bed in layer 1, the highest sediment concentration is found with a related
reduction in κ to estimated values of as low as 0,2 (typical of that found in open
⎛ dv ⎞
channel flow). A reduction in κ leads to an increased velocity gradient ⎜⎜ ⎟⎟ in
⎝ dy ⎠
layer 1 near the bed.

b) At the top of layer 3, v → 0 and the suspended sediment concentration

approaches zero. Therefore κ → 0,4 , the standard value in turbulent open
channel flow with no sediment transport.

c) As the sediment concentration rapidly decreases in layer 2, the inflection point in

the velocity distribution (or the interface) is probably caused by the sudden
change in the value of κ at the interface.

The velocity profile of a density current can be described by assuming the

following:

i) A laminar boundary layer at the top of layer 3 where v → 0

ii) Turbulent flows, corrected for density difference, in layers 3 and 2.

Laminar flow above the interface (above layer 2) would at first seem to
be more probable, but such an assumption cannot explain the relatively
great depth of layer 3 and the velocity distribution. A correction to the

70
 density difference is also required, since some sediment is diffused
upward and transported in layer 3.

iii) Turbulent flow, corrected for density difference, in layer 1, derived from
the same principles as those applicable to rough bed conditions.

iv) Logarithmic velocity distributions in layers 1, 2, and 3 for turbulent

conditions, adjusted with correct κ values as these values change with
depth in each layer.

For layers 2 and 3

At the top of layer 3, with y measured downward as in Figure 4.3-1, in the laminar
layer:

dv
τ=μ = Δρ gs(D2+3 - y)
dy

d v Δρ gs(D2+3 - y)
∴ = (4.4-2)
dy μ

In the rest of layer 3, the flow is turbulent and the shear stress:

2
ρ 2⎛ d v⎞
τ= R ⎜ ⎟ = Δρ gs(D2+3 - y) (4.4-3)
2π ⎜⎝ dy ⎟⎠

Following the same derivation as with turbulent open channel flow, it is possible to
show that (Basson and Rooseboom, 1997):

dv 2πΔρ g D2+3 s
∴ = (4.4-4)
dy ρy

71
Assuming that at the boundary between the laminar and turbulent density current flow at
the top of layer 3, the applied power values are equal i.e.:

dv dv
τ laminar = τ turbulent
dy dy

τ la min ar = τ turbulent , then

dv dv
laminar = turbulent
dy dy

( - y) 2πg ′ D 2+3 s Δρ
Δρgs D 2+3 = , with g′ = .g (4.4-5)
μ y ρ

Solving Equation (4.4-4) now for y = h5 , the thickness of the boundary layer, to be
used as integration constant when deriving the velocity:

2πg ′ D 2+3 s. μ
∴ h5 =
Δρgs( D 2+3 - h5 )

2π μ
= (4.4-6)
ρΔρg( D(2+3) - h5 )s

Integration of Equation 4.4-4 yields:

v(2+3) = 2π g′ D(2+3) sl n y+ A (4.4-7)

with A an integration constant.

At the top of layer 3 at the boundary layer:

Δρ gs
v(2+3) = (2 Dy- y2 . h 5) = 2π g′Ds l n h 5 + A (4.4-8)
2μ

Δρgs Dh5
∴ A= - 2πg_Ds ln h5
μ

72
 and Equation 4.4-7 becomes:

⎛ y ⎞ ΔρgsD2 + 3h5 D2 + 3
v(2+3) = 2πg ' D( 2 + 3) sln ⎜⎜ ⎟⎟ + (4.4-9)
⎝ h5 ⎠ μ h5

Substituting Equation 4.4-6 in Equation 4.4-9 gives:

⎛ y gsD ρΔρ ⎞ Δρ gsD 2π .μ

v2+3 = 2π g′ D(2+3) sl n ⎜⎜ ⎟+
⎟ μ . gsD ρΔρ
⎝ 2π μ ⎠

⎛ y gsD ρΔρ ⎞
= 2πg ′ D(2+3) s ln ⎜ ⎟ + Δρ 2π gsD (4.4-10)
⎜ 2π μ ⎟ ρ
⎝ ⎠

In layer 1, the velocity at any level can be derived, but with the density difference
adjustment.
Therefore

⎛ D1 ⎞
v1 = 2π g′ D1 sl n ⎜⎜ ⎟⎟ (4.4-11)
⎝ yo ⎠

Equations 4.4-10 and 4.4-11 can now be solved together by using the fact that the
maximum velocities in both equations are common:

Equation 4.4-10 becomes

⎛ D gD(2+3)s ρΔρ ⎞ Δρ
v(2+3) (max) = 2π g ′ D(2+3) s l n ⎜ ⎟ + 2π gsD (4.4-12)
⎜ 2π μ ⎟ (2 + 3)
ρ
⎝ ⎠

and Equation 4.4-11:

⎛ D1 ⎞
v(1) max = 2π g′ D1 sl n ⎜⎜ ⎟⎟ (4.4-13)
⎝ yo ⎠

73
 ks
with yo =
30

Equations 4.4-12 and 4.4-13, however, contain 3 unknowns: D1 , D2+3 and vmax. A third

equation is therefore required which expresses D1 in terms of D2+3 . This third equation
is found when combined translation and rotation in turbulent flows are considered.
dv
Turbulent flow translates at relative velocity v 0 = y o . In a density current the same
dy
principle applies near the bed as well as at the top of layer 3, so that the translation
velocity at the top and bottom of the density current velocity profile should be
equivalent:

dv dv
y1 = y3
dy10 dy30 = constant if acceleration = 0 (4.4-14)

g ′ D1 s g ′ D(2+3) s
∴ = (4.4-15)
κ1 κ 2+3

∴ D1 = κ2 1
2

D(2+3) κ (2+3)

with typical expected values of κ 1 → 0,2 to 0,3 and κ ( 2+3) = 0,3 to 0,4

Equations 4.4-12, 4.4-13 and 4.4-15 can now be applied simultaneously to solve for D1
and D2+3 (Basson and Rooseboom, 1997).

4.5 Verification of theory to predict the velocity profile and depth of a density current
with laboratory and field data

Using the data of Ashida and Egashira (1975) and Chikita (1989) as sources of
laboratory and field data respectively, it is possible to calculate the upper layer depths
D2+3 from observed D1-values as a first check on the validity of Equations 4.4-12 and

74
 4.4-13. Predicted D(2+3) values can, however, only be determined accurately if correct

values of κ 1 and κ (2+3) are used. Although κ 1 values have been determined by Ashida

and Chikita, no data for κ 2+3 -values are available, and realistic values had to be
assumed.

Assuming κ ( 2+3) = 0,4 (as for no sediment transport) and κ 1 = 0,2 at the bed, the

predicted D2+3 versus observed D2+3 data are indicated in Figure 4.5-1. It is clear that
the assumptions made in the derivation of Equations 4.4-12 and 4.4-13 give consistent
upper layer depths for both laboratory and field data. Predicted versus observed D(2+3)

depths however still need to be compared using real Kappa values.

Unfortunately D3 data were not published by Ashida, and for only 3 experiments were
the ratios of D(2+3) )/ D2 being indicated as varying from 2,1 to 2,5. Using

κ 1 = 0,25 and κ 2+3 = 0,32 provides realistic observed κ 1 data as well as solutions in
terms of Equation 4.4-15 for assumed D1 and D2+3 = 2,5 x D2 values for Ashida's

data. With κ 1 , κ 2+3 and D1 known, D2+3 can be calculated from Equations 4.4-12 and

4.4-13 and compared with "observed" D(2+3) data = 2,5 x D2 as shown in Figure 4.5-1.

Using the same approach, reservoir data of Chikita (1989), are also used to verify the
mathematical assumptions, as shown in Figure 4.5-1. In this case D(2+3) = 3 x D2 had

to be used, as inferred from the data of Chikita. Although a number of assumptions had
to be made owing to the lack of certain observed variables in the laboratory and field
data, it is believed that the basis of the theory used in deriving Equations 4.4-12, 4.4-13
and 4.4-15 is sound, as was provisionally proven in Figure 4.5-1. Future detailed
laboratory tests will be required to establish κ 1 , κ 2 , κ 3 , D1 , D2 and D3 values with
different channel slopes, discharges, sediment loads and bed roughnesses.

4.6 Movement of a density current : flow resistance and velocity

It is possible to derive theoretically, from equations of sediment transport and discharge

continuity, a relationship for the average velocity of a density current by integrating the
velocity distribution with depth and by using the theory as derived in the previous

75
sections. The many unknowns in such a relationship make it difficult to apply in
practice and therefore most researchers have resorted to Chezy type equations with all
the unknowns incorporated into a single empirical constant CH in the equation:

v = CH . g′Ds (4.6-1)

with s = so, the bed slope, assuming a uniform density current

and D = depth of layers 1 and 2.

Figure 4.5 - 1 Observed versus calculated density current layer depths

Studies have been undertaken by Raynaud (1951), Bata and Knezevich, (1953),
Blancket (1954), Michon et al. (1955), and Levy (1958) on the value of CH in laminar
and turbulent regimes.

Fan (1960) described the movement of a density current in terms of the equation of
motion. Consider a fluid element in a density current:

1 dp dy 1 dτ v dv 1 dv
+ + + + =0 (4.6-2)
γ ′ d x d x γ ′ dy g d x g dt

with p = total pressure = p1 (due to lighter fluid) + Δp (due to density difference)

76
 γ ′ = specific weight of the density current
y = distance from horizontal reference line

If the depth of the density current is h´ and the depth of the overlying water h, then the
total depth is H = h + h1 (see Figure 4.6-1).

Assuming homogeneous fluids, above and below the interface

(4.6-3)
p + γ ′y = γH + ( γ ′ - γ )h′ + γ y o

dp dy dH d 1h dy
and +y = +y + y1 0
dx dx dx dx dx

with γ = specific weight of overlying water

Fan further assumed that τ o , the mean bed shear stress, and τ i ,the shear resistance at
the interface, are proportional to the square of the mean velocity:

λ o ρ ′ v2 ; = λ i ρ v2
τ o= τ1 (4.6-4)
4 2 4 2

Figure 4.6 - 1 Schematic diagram of a density current

77
Where λo and λi are the corresponding coefficients of friction.
Considering a segment of the density current with width b, bed slope Jo and interface
slope Ji, the equation of mean motion of the density current reads:

1 dH _γ dh_ dy o ⎛ λ o v 2 λ i v 2 ⎞ ⎛ v dv 1 dv ⎞
+ + +⎜ + ⎟+⎜ + ⎟=0 (4.6-5)
γ ′ dx γ ′ dx dx ⎜⎝ 8 gh1 8 gh1 ⎟⎠ ⎜⎝ g dx g dt ⎟⎠

in which:
dH
= J0 - Jn
dx
dy o
= J0
dx
Δγ = γ ′ - γ
and Jn = negative slope at water surface.

The frictional stress exerted along the interface is

λ i p′ v 2 = - γ bh
τi= . Jn
4 2 b + 2h

Substituting the above equations into Equation 4.6-4 yields:

y⎛ dh1 ⎞ v 2 d ' h λmv 2 1

⎜ J 0 − ⎟+ − − =0 (4.6-6)
y' ⎝ dx ⎠ g ' h dx 8 gR g

with λ m = mean friction coefficient of the underflow

⎡ b h′ (b + 2h) ⎤
= λo + λi ⎢ + ⎥
⎣ 2h′ + b h (b + 2h′) ⎦ (4.6-7)

For unsteady, non-uniform density currents, Fan used Equation 4.6-5 to predict depths.
For steady density currents λ m = 0,02 to 0,03, but Fan used λ m = 0,05 for unsteady
conditions.

78
For steady, non-uniform density currents:

⎛ ⎞
⎜ 2 ⎟
λm v ⎟
Jo - ⎜ _γ
⎜ ⎟
⎜ 8 gR ⎟
d ′h
= ⎝ γ′ ⎠
dx ⎛ ⎞
⎜ 2 ⎟
1- ⎜ v ⎟
⎜ _γ ⎟
⎜ γ g′h ⎟ (4.6-8)
⎝ ⎠

with R = hydraulic radius =b′h/(2h′ + b)

dh′
and for uniform flow = 0 then
dx

8 Δγ
v= ⋅ ⋅ gRJ o (4.6-9)
λm γ′

8
= . g ′ RJ 0
λm

which is similar to a Chezy type equation for uniform open channel flow.

Fan found the mean friction factor λ m (Equation 4.6-6) from flume studies to be
independent of the Reynolds number and to remain almost constant as shown in Figure
4.6-2, when the flow is turbulent.

For a bed friction factor of λo = 0,02 , Fan found the interfacial friction factor

λi = 0,005 as shown in Table 4.6-1.

79
 Table 4.6-1: Interfacial friction factor

Jo
Average h′ (cm) Average λ i
0,0005 14,4 0,0047
0,005 17,8 0,0051

Figure 4.6 - 2 Friction factor λ m from flume studies (Fan, 1960)

In the Guanting Reservoir, China, Fan found the same constant mean friction factor as
in the flume tests (with the same sediment), with λ m = 0,02 to 0,03 as shown in Figure
4.6-3.

Fan (1960) also found that supercritical and subcritical conditions can prevail in the
density current, and that the transition from one condition to another is possible in the
plunging phase, with submerged hydraulic jumps being formed.

Figure 4.6 - 3 Friction factor for Guanting Reservoir (Fan, 1960)

80
 A relationship between typical depths of flow and density current thicknesses for the
Guanting Reservoir as well as flume data is shown in Figure 4.6-4.

Such a relationship will be site-specific and will depend on a number of specific

boundary conditions such as bed slopes, discharge, sediment characteristics, etc.

So far, only one-dimensional density currents have been considered. The influence of
flow width on a density current is, however, also important in order to quantify total
sediment transport.

4.7 Cross-sectional variation in velocity and sediment concentration across a density

current in a reservoir

At the early stages of impoundment, a density current usually moves along the original
river channel. Later, as the main channel is filled through sedimentation, the
topographical width exceeds the width of the density current. Examples of observed
lateral and vertical velocity and concentration distributions in Sanmenxia Reservoir,
1961, are depicted in Figure 4.7-1 (Fan, 1986).

Figure 4.6 - 4 Ratio of density current depth to open channel flow depth in Guanting
Reservoir (Fan, 1960)

81
Figure 4.7-1 Density current lateral velocity and concentration distribution in
Sanmenxia Reservoir (Fan, 1986)

The general validity of a density current uniform flow velocity equation of the Chézy
type has been demonstrated by experimental studies and also through successful
application.

Harleman (1961) used:

8
v= g ′ hs o (4.7-1)
f (1 + α )

with f = Darcy-Weisbach friction factor

82
and α = factor describing shear distribution at interface as
function of bed shear, with τ i = ατ o (4.7-2)

Harleman proposed the use of α = 0,43 for turbulent density current flow and f-values
from the Moody diagram for pipe flow.

For laminar flow (Re < 1000), Harleman proposed

v = 0,375 Re g′ hso (4.7-3)

which was determined theoretically.

Harleman (1961) applied equation 4.7-1 to Lake Mead, USA and obtained consistent
results with actual measurements. Middleton (1966) experimentally confirmed
Equation 4.7-1 for density currents of salt solutions and clay suspensions, with
Δρ < 1 and s o < 0,03 .

Keunen (1952) obtained a simple relationship from flume studies:

v = C ′′ g ′′ ho s o (4.7-4)

g ( ρ1 − ρ 2 )
with g ′′ =
ρ2

Δρ
in contrast to g ′ =
ρ1

with ρ 2 density in density current,

Keunen proposed C" = 280 cm1 / 2 / s in reservoirs for s o > 20°

83
 Hinze (1960) showed that the empirical relationship used by Keunen (1952) can be
derived theoretically. Using boundary layer theory, Hinze obtained Equation 4.7-4 with
C"=f(suspension density distribution, friction coefficient at interface) with
280<C"<560cm1/2/s.

Bagnold (1962) proposed an auto-suspension model as the criterion for the continued
self-maintenance of a density current, which can be rewritten in the form:

v = C g ′ h sin θ (4.7-5)

⎛ w⎞
but the term sin θ is replaced by ⎜ sin θ − ⎟ , with w = settling velocity of particles.
⎝ v⎠

Bagnold used an energy balance principle:

Layer integrated turbulent energy production = work done against negative buoyancy
force + turbulent energy dissipation.

δu
∫ o ( u′ v′) . dy = ∫ o bvcosθ dy + ∫ o Edy
H H H
(4.7-6)
δy

∴ s 2 BHU( sin θ - vf cosθ ) = ∫ o Edy

H

The energy dissipation term cannot be evaluated properly without velocity and density
profiles. It is therefore difficult to verify Bagnold's concept (Middleton, 1966).

4.8 Motion of the head of a density current

Every uniform flow region of a density current is preceded by an initial head also known
as the nose. Hinze (1960) referred to it as the spreading-out phenomenon, during which
the density current displaces the fluid which it enters. Altinakar et al. (1990) plotted the
shape of the head non-dimensionally as shown in Figure 4.8-1.

84
Figure 4.8-1 Non dimensional density current head shape (Altinakar et al., 1990)

Middleton (1966) observed that the head had a well-defined shape, with a depth twice
the uniform flow depth, while the velocity is less than that of the uniform flow. It is also
a region of intensive erosion. He found from flume studies that the motion of the head
in the density current was closely described by the laws developed by Keulegan (1958)
for saline surges. It was found that the velocity of the density current head on slopes up
to 4% is adequately expressed by Keulegan's formula:

vn = 0,75 g ′hn (4.8-1)

with hn = thickness of the head.

Turner (1973, 1979) proposed Equation 4.8-2 for the velocity of the head:

vn = 2 g ′h n (4.8-2)

with hn = depth of uniform density current flow.

When considering the three-dimensional motion and slope influence, Turner proposed:

85
v n = 0,75 g ′hn (4.8-3)

which is the same equation as proposed by Keulegan (1958).

Density currents are characterized by a distinctive raised head, followed by a quasi-

uniform flow region called the body. (Figure 4.8-2)

Figure 4.8-2 Density current head (Altinakar et al., 1990)

The head constitutes a region of intense mixing and wave breaking, with a highly
irregular front.

Altinakar et al. (1990) compared his flume data with those of Middleton and Turner and
proposed a smaller "Chezy" coefficient value of 0,63 in the equation:

vn = 0,63 g′ h n (4.8-4)

which is shown graphically in Figure 4.8-3.

86
 Figure 4.8-3 Density current head velocity (Altinakar et al., 1990)

The data used in Figure 4.8-3 show considerable scatter, but this is understandable since
the channel slope was not considered. The data in the Altinakar experiments were
derived from bed slopes < 3%.

Altinakar et al. further considered the head velocity as a function of initial buoyancy
flux Bo = + g o qo and slope ( s 0 ):

Uf = ( g 0 q ) f (s 0 )
1/ 2
(4.8-5)

as is shown graphically in Figure 4.8-4.

87
Figure 4.8-4 Dimensionless head velocity as function of bed slope (Altinakar et al., 1990)

The data show considerable scatter, which Altinakar et al., (1990) attributed to
experimental errors and different drag coefficients.

Although the movement of the density current head differs from uniform flow owing to
the displacement of stagnant water at the nose, it is proposed here that since
v a = f (g , H f s o ) a Chezy equation as for the uniform density current with a different C

coefficient may be used:

v n = C n g ′H f s o (4.8-6)

The data of Altinakar are shown in Figure 4.8-5 and it is clear that Equation 4.8-6 does
indeed provide an accurate predictor of the density current head velocity.

88
 Figure 4.8-5 Velocity of density current head using Chezy type equation

4.9 Sediment transport by density currents

Bell (1940) expresses a very interesting view on density currents: "In ordinary streams,
water propels suspended sediment. In turbid density currents, sediment propels water.
Once the significance of this paradox has been grasped, a deal of mystery that surrounds
density currents is stripped away".

From our current knowledge of hydraulics, we know that the density difference
associated with sediment transport is required for density current movement. Both open
channel flows and density current flows will however transport sediment (if available),
since the sediment provides the mechanism through which the bed can be deformed and
whereby the applied energy can be minimized. It is therefore not the water that propels
the suspended sediment or vice versa in streams or density currents, but rather an
intricate interrelationship between sediment and clear-water transport in a process of
minimization of energy.

When a density current develops from turbulent inflow into a reservoir, coarser
sediment is immediately deposited (in the delta area) while the fine sediments, up to the

89
sediment transport capacity of the density current, can be carried through the reservoir.
The physics involved in the sediment transport process is almost the same as with
turbulent open channel flow.

dv
The vertical suspension of sediment is limited by the applied power τ → o at the
dy
maximum velocity level between layers 1 and 2. Although some sediment is diffused
into layers 2 and 3, the amount is negligible compared with the load in layer 1.

The sediment being transported in a density current consists of the silt and clay
fractions, with typical particle sizes and concentration distributions shown in Figures
4.9-1. It is clear that in layer 1 a constant suspended sediment concentration and density
can be assumed in most cases, and therefore the sediment transport can be given by:

D1

q c = ∫ Cv dy
yo (4.9-1)
=Cv

with

z
⎛ d v⎞
⎜⎜τ ⎟
D1 ⎝ dy ⎟⎠
C = C o ∫ yo z
dy
⎛ dv ⎞
⎜⎜τ ⎟⎟
⎝ dy ⎠o

Density current movement is unsteady due to the nature of floods, especially in arid
areas. Measurements at the Guanting Reservoir clearly illustrate the characteristic of an
unsteady density current with a reduction in sediment transport and with finer sediment
being transported by the density current than by the inflow (Figure 4.9-1).

90
 Figure 4.9-1 Unsteady density current measured in Guanting Reservoir (Wu, 1994b)

Fan (1960) investigated the sediment transport capacity of density currents in terms of
the limiting particle size being transported. From flume tests he established that coarse
sediment settles almost immediately while finer particles are transported at a constant
gradation d90 = 0,008 to 0,018 mm and d50 = 0,002 to 0,003 mm, with plunge point
velocities of 40-80 mm/s.

In the Guanting Reservoir the suspended sediment in the density current was found to be
much coarser with d90 = 0,01 to 0,13 mm and d50 = 0,002 to 0,003 mm, with a plunge
point velocity of 0,2 m/s resulting in greater transport capacity and the related larger
particles that were transported compared to flume studies.

4.10 Density current formation following flushing

While Lewis (1936) stated that density currents had been observed in Lake Arthur, no
quantitative data or information was published. No other density currents (turbidity
driven) had been reported in the country until recently. Although the sediment loads in
many local rivers are high enough to create density currents, and suspended sediment

91
loads include enough fine particles, the main limiting factor must be the small bed
slopes of reservoirs, especially after some time of deposition, which prevent the
progression of density currents through reservoirs.

During recent research density currents were observed after each flushing of
Welbedacht Reservoir during filling of the reservoir. While the 5 gates in the dam were
closed to cause filling of the empty reservoir, river outlets remained open which allowed
venting of sediment transported towards the dam.

While suspended sediment samples taken near the surface upstream of the dam
indicated low concentrations, the suspended concentrations of outflows sampled just
downstream of the dam continued to be high for many hours after closure of the main
gates, as shown in Figure 4.10-1.

Figure 4.10-1 Density current venting at Welbedacht Reservoir

The streampower relationship proved that the high observed concentrations cannot be
related to turbulent suspended sediment transport in the reservoir and that some other
transport mechanism, which could only be density currents, had to be responsible for the
high concentrations being vented from the reservoir. Flushing with water level draw-
down of course provided high turbulent sediment transport capacity and related high

92
concentrations. Once the main outlets were closed, however, the coarse sediments were
immediately deposited, whereas high concentrations of fines created density differences,
resulting in density currents moving along the bed as the reservoir filled. The steep
front set slope of the sediment deposit within Welbedacht Reservoir provided enough
streampower for the density current to move through the reservoir down to the dam
(Figure 4.10-2). The duration of density current flow is limited, however, by the
limited availability of sediment after turbulent scouring action had taken place during
the flushing process. After some time, the outflow concentration decreased to become
equal to the inflow sediment concentration.

Figure 4.10-2 Longitudinal profile of Welbedacht Reservoir bed

Using the scoured profile of the main channel, it is possible to calculate the mean
velocity of the density current observed in Welbedacht Reservoir. Suspended sediment
concentration can then be related to the input streampower function for the density
current and compared with similar reservoir data from China.

Unfortunately only the averaged suspended sediment particle size distribution was
determined for the 1994 flushings at Welbedacht Dam, and information on the sediment

93
 size transported by the density current as compared to turbulent transport is not
available. During 1995, however, the reservoir was flushed again under similar flow
conditions as in 1994 and this time more detailed sampling and grading analyses were
carried out. Figure 4.10-3 shows the effective settling velocity changes from the full
reservoir before flushing, during flushing and during filling with density current venting.

Figure 4.10-3 Reservoir operation and settling velocity based on observed sediment

characteristics
Wu (1994b) derived a sediment transport equation for a density current based on similar
assumptions as those which were used to obtain the Velikanov (1954) sediment transport
equation for open channel flow sediment transport:

3
v
C = K1
gh w 50 (4.10-1)

Δλ
with v = K 2 gh
γ

94
 and K1 , K 2 = constants
h = height of density current
w50 = settling velocity of median particle diameter

Field data of Guanting, Sanmenxia and Hongshan reservoirs with sediment

concentrations < 100 kg/m3 were used to calibrate Equation 4.10-1 as indicated in
Figure 4.10-4 and Equation 4.10-2.

0,285
⎛ v3 ⎞ (4.10-2)
C = 12,75 ⎜⎜ ⎟
⎟
⎝ ghw 50 ⎠

Figure 4.10-4 Chinese reservoir density current sediment transport relationship (Wu,
1994b)

Calculation of the input streampower versus sediment transport of the Welbedacht

Reservoir density current shows a direct relationship (Figure 4.10-5) and also agrees
with Chinese reservoir data (Figure 4.10-4).

95
 Figure 4.10-5 Welbedacht Reservoir density current sediment transport relationship

4.11 Non-equilibrium density current sediment transport

Wu (1994b) proposed the use of a non-equilibrium relationship in predicting/modelling

sediment transport by density current through a reservoir, similar to the approach
followed for turbulent fine sediment transport in reservoirs in China and also proposed
in this study:

dC - ∝ w
= (C- C*) (4.11-1)
dx vh

with α = coefficient
C* = equilibrium sediment concentration
v = density current velocity
h = density current depth

Equation 4.11-1 which has been calibrated for turbulent open channel flow conditions
in this study (Chapter 3), will probably also be applicable in the case of density current
sediment transport. Much more data collection and analyses of equilibrium sediment

96
 transport in density currents need to be carried out, however, before a proper theoretical
understanding of non-equilibrium sediment transport would be possible.

4.12 Graded sediment transport by density currents and the sorting process

Field data from China have been used to establish an empirical relationship between
sediment size and transport:

Cx dx90
% = 1,1. % - 12 (4.12-1)
Co dD90

with Co = mean suspended sediment concentration at entrance (plunge point),

D90 = size of particles for which 90% of the material is finer, from the frequency
distribution curve for the appropriate sample.

This relationship is shown graphically in Figure 4.12-1 (Wu, 1994b).

Non-uniform sediment transport in a density current can be predicted as for

homogeneous open channel flow when the equilibrium sediment transport equation is
calibrated for various particle sizes suspended in the density current.

Figure 4.12-1 Graded sediment transport in a density current (Wu, 1994b)

97
4.13 Formation of a density current

4.13.1 Review of theory

Many attempts have been made to predict the formation of a density current:

Dequennois (1956) suggested (γ ′ − 1)q as a criterion for the turbulent flow to form a
density current.

Levy (1958) analysed the minimum silt concentration below which a density current will
not occur.
Fan (1960) concluded that in the transition from open channel flow to density current
dh′
flow, increases rapidly at the interface, and → ∞ . Equation 4.6-9 can therefore be
dx
rewritten with

v2
=1 (4.13-1)
Δγ
gh′
γ′

From Figure 4.13-1, it is evident that the velocity decreases as the depth increases in the
transition zone and that the velocity reaches a minimum value and the water depth a
maximum value at the plunge point. Fan therefore proposed that the specific energy

2
v
E = h′ +
Δγ
2 g′h
γ

reached a minimum value at the plunge point, or

2
dE q
= 1- =0
d ′h Δγ 3
gh
γ o

98
 v0
=1
Δγ
gho
γ′

This parameter is known as the densimetric Froude number (FrD ) .

Figure 4.13-1 Schematic diagram of density current formation (Fan, 1960)

By conducting flume tests, Fan established a relationship for the densimetric FrD as
given in Figure 4.13-2.

with (4.13-2)
vo
= 0,78(< 1)
Δγ
gh o
γ

Fan gave the reason for the experimental FrD value being < 1 as "a point of inflection in
the streamlines at the interface, which is located downstream of the plunge point,
where FrD = 1 ."

99
Figure 4.13-2 Densimetric Froude number and density current formation (Fan, 1960)

Fan (1986) found that the formation of density currents can be forecast by means of
Equation (4.13-2) for concentrations < 100 kg/m3. It was, however, found that the
densimetric Froude number also depends on the inflow sediment concentration and
decreases with increasing concentration. This explains why different researchers have
found such a wide range of densimetric Froude numbers which predict density current
formation from their experiments.

Cao (1992) published a relationship between FrD and the density difference (Figure
4.13-3). It appears that when sediment concentrations are < 40 kg/m3, FrD ≈ 0,78 . As
the sediment concentration increases the density current changes from turbulent to
transitional to laminar with hyperconcentrations.

100
Figure 4.13-3 Densimetric Froude number and density difference ratio (Cao, 1992)

Akiyama et al., (1987) found an approximate average densimetric Froude number of

0,68 at plunging. The FrD was only weakly related to the inflow densimetric Froude
number (Fo) and the angle of channel divergence as shown in Figure 4.13-4.

Figure 4.13-4 Densimetric Froude number (Akiyama et al., 1987)

101
Savage and Brimberg (1975) considered the plunging phenomenon in terms of
equations of motion for gradually varying two-layer flow. The pressure distribution was
considered to be hydrostatic and mixing was neglected. Different solutions for the
interface profile were obtained depending on the bed slope and the relative importance
of the two boundary shear stresses.

Denton et al. (1981) carried out experiments on temperature-induced density currents

and found that for mild bed slopes, interfacial and bed shear stresses are important, but
for steep slopes (S o > 0,02) , momentum balance conditions at the plunge point dictate a
maximum densimetric Froude number criterion of 0,67.

Singh and Shah (1971) and Philpott (1978)also conducted flume studies and obtained
FrD ≈ 0,7 for the plunge point, although a range of 0,3<Fr D<0,7 was observed.

The criteria for plunging used by the above researchers indicate a general variation of
observed values 0,1<FrD<0,7, which shows the unreliability of prediction of density
current formation.

Previous studies have dealt with sloping channels, rectangular cross-sections of constant
width, or triangular shapes that represent reservoir geometry. Investigations of the
formation of a density current based on the densimetric Froude number show
considerable variation, as is shown in Table 4.13-1.

Table 4.13-1: Densimetric Froude number (Fp) at plunging

Reference Fp
Ford and Johnson (1980) 0,1 to 0,7
Itakura and Kishi (1979) 0,54 to 0,69
Singh and Shah (1971) 0,30 to 0,80
Kan and Tamai (1981) 0,45 to 0,92
Fukuoka and Fukushima (1980) 0,40 to 0,72
Farrell and Stefan (1986) 0,66 to 0,70
Akiyama et al. (1987) 0,56 to 0,89

102
Akiyama et al. (1987) showed that the angle of divergence also plays a role in estimating
Fp, but the scatter of data is still considerable.

Rooseboom (1975) also proposed a method to predict the formation of a density current.
Consider a fluid element with density ρ + Δρ underneath a non-uniform fluid layer of
density ρ. Under normal turbulent flow conditions, the element with length Δ x is

pushed forward owing to a pressure difference which exists across the element:

Δp = ρ gs s .Δx (4.13-3)

with s s = slope of surface line

ρ = mass density of upper fluid

In situations where density currents are important, the value of ss is small and ≈ s f , the

energy slope.

Due to the existing density difference, an additional pressure difference is generated

across the element = Δρgs o Δx (so = reservoir bed slope).

The ratio

" density" pressure Δρ . s o

= >1 (4.13-4)
" turbulent" pressure ρ . s f
indicates the relative importance of density differences in the forward propulsion of
sediments through reservoirs.

Using the Chezy equation, Equation 4.13-4 can be rewritten:

Δρ so C 2 R
ρ v2 (4.13-5)

with R = hydraulic radius, C = Chezy coefficient.

103
Density difference will therefore play an important role relative to turbulent suspension
if the value of Equation 4.13-5 is high, i.e. in cases of :

• large flow depths

• large density differences
• steep reservoir bed slopes
• low flow velocities.

Δρ
Assuming ≈ constant in all reservoirs, R = average water depth at full supply level,
ρ
and taking the cross-sectional area as D2,
the following dimensionless parameter was obtained:

2 5
s o .C . D
Q
2 (4.13-6)

with Q = discharge, with sufficient sediment transport capacity through reservoir.

The data in Table 4.13-2 were used to calibrate Equation 4.13-6

With the limited data it is difficult to obtain a definite relationship. Conditions in Treska
Reservoir represent a poorly developed density current and therefore a value of > 10000
for the dimensionless parameter was assumed (Rooseboom, 1975).

Table 4.13-2 Density current formation (Rooseboom, 1975)

Reservoir so D(m) Q(m3/s) C(m1/2/s) ⎛ So.C 2 . D5 ⎞

⎜
⎜ 2
⎝ Q ⎠
2
Lake Mead 0,00074 82 500 70 54 000
Sautet 0,018 50 200 63 56 000
Treska 0,004 10 9,5 55 13 000
Sengari 0,0049 16 1 59 18 000 000

104
4.13.2 Prediction by means of minimum stream power principle

A flowing stream of water with a movable bed will always try to change the boundary
conditions imposed on it until equilibrium conditions are reached. At equilibrium the
energy dissipation per unit volume is minimized. In terms of applied stream power, this
means that the flow can adjust the values of κ, k2 D and sf in Equation 4.13-7.

dv gD sf
τ = 30 ρ g sf D (4.13-7)
dy κ . ks

Minimization of streampower will also be the reason why a turbulent stream dives to the
reservoir bed to progress as density current through the reservoir. At the plunge point
(as indicated in Figure 4.13-5) input streampower for the density current should
therefore become equal to or less than that of the turbulent inflow, assuming that the
streampower in the upper layer is approximately zero as v → o in the upper layer.

At the plunge point from streampower continuity:

Streampower of turbulent inflow = streampower of density current

D d

∫ ρgv s
yo
f = ∫ Δρgv s o
yo
(4.13-8)

Therefore, for a density current to form it is expected that:

ΔρgQ s o
≤1
ρgQ s f and from continuity this simplifies to (4.13-9)

Δρs o
≤1 (4.13-10)
ρs f

105
 Figure 4.13-5 Plunge point characteristics

Laboratory data on the plunge phenomenon are rather scarce since most experiments
have concentrated on uniform density current movement. As a preliminary test of
Equation (4.13-10), laboratory data of Akiyama et al. (1987), in which the density
difference was created by temperature differences, was used (Figure 4.13-6).

Figure 4.13-6 Density current formation and minimum stream power

106
Verification of Equation 4.13-10 to test for density current formation does indeed show
that less streampower is used by the density current than by turbulent open channel
inflow. Furthermore, Figure 4.13-5 shows that the densimetric Froude number does not
provide an accurate means of predicting density current formation due to the large
scatter in values. The general trend indicates a lower density current stream power
value as the densimetric Froude number increases.

Even if the stream power in the layer above the density current is added to that of the
density current streampower, the difference will be negligible due to the small slope
(due to the reservoir depth) and low velocity in the upper layer.

Equation 4.13-10 seemingly contradicts the Equation 4.13-4 proposed by Rooseboom

(1975), who stated that larger density differences, larger slope so, and smaller slope sf
(of reservoir) will provide better conditions for density current formation. A steep bed
slope in the reservoir and large density difference will indeed help to ensure the
continuous movement of the density current through a reservoir, but what Equation
4.13-10 in fact says is that at the plunge point a steeper approaching turbulent flow
slope sf will create favourable conditions for density current formation as the river meets
the water mass in the reservoir. The high momentum of the inflowing river enables it to
dive underneath the water mass, and once near the bed the density difference creates the
density current flow and prevents a "hydraulic jump" of the submerged flow, with
consequent turbulent suspended sediment transport through the reservoir. This is shown
schematically in Figure 4.13-5.

Field and laboratory studies indicate that the density differences between
riverflow/density current and reservoir water can be very small. The dominant variables
therefore in the formation of a density current (given that there is some density
difference), are the river and density current slope differences. On the other hand, the
bed slope in the reservoir still needs to be steep enough for the propagation of the
density current after formation.

Equation 4.13-10 also shows that for a given inflow river slope, the reservoir bed slope
at the plunge point has an upper limit at which turbulent mixing can prevent a density
current from forming.

107
4.14 Laminar density currents associated with hyperconcentrated sediment transport

So far only the characteristics of turbulent density currents have been addressed.
Laminar density currents also occur and are associated with hyperconcentrated sediment
transport.

First, consider the differences between vertical velocity distribution in laminar and
turbulent density currents as shown in Figure 4.14-1 (Cao, 1992).

Figures 4.14-1a and 4.14-1b show typical laminar conditions in the density current. At
Reynolds number (Ren) < 43 (Figure 4.14-1a), the boundary condition at the interface is
laminar, the resistance at the interface is relatively small, and the profile of velocity
distribution of the density current is similar to that of open channel flow. The resistance
at the interface is, however, not negligible as in homogeneous flow.

From flume studies, Cao (1992) established three types of plunging patterns (Figure
4.14-2):

Type A: Supercritical open channel flow with hydraulic jump transition at the
plunge point.
Type B: Subcritical open channel flow with mild undulation transition to density
current.
Type C: Hyperconcentrated homogeneous laminar open channel flow with still
transition to density current. Two types C can be further observed:
J m = J o (Figure 4.14-2C) and J m > J o (Figure 4.14-2C1). In the latter
the plunging point is not stationary but keeps progressing until reaching
the point where a non-uniform density current forms with the stable
gradient J m = J o when the plunging point becomes stationary.

The flow velocity, sediment transport and prediction of formation of laminar density
currents can be derived following similar approaches as with turbulent density current
flow in the previous sections.

108
Figure 4.14 - 1 Typical density current velocity distribution (Cao, 1992)

109
 Figure 4.14 - 2 Profile of interface of density current (Cao, 1992)

4.15 Venting of density currents through reservoirs

When a density current reaches the dam, it can either be vented through bottom outlets
or will "climb" up the dam and fall back to form a muddy pool. The climbing nature of
the density current can be utilized to vent the density current at a high level. By either
providing a siphon type outlet or curtain at the dam, the density current can be aided to
flow out at even higher levels above the bed. This approach will be especially useful at
existing reservoirs without bottom outlets and where density current flows are
experienced.

Bell (1942) indicated the possibility of using a curtain near the dam in flume studies as
shown in Figure 4.15-1.

Fan (1960) investigated the maximum height up to which a density current can "climb"
to be vented and found that openings directly in the path of a density current were the
most efficient.

110
Laboratory results indicated that the maximum height (hL) to which a density current
can climb to be vented through a circular outlet (Wu, 1994b):

1/ 5
⎛ 0,154λQ 2 ⎞ ⎛ γ oq2 ⎞
hL = ⎜⎜ ⎟⎟ + ⎜⎜ ⎟
2 ⎟
(4.15-1)
⎝ Δγg ⎠ ⎝ 2Δγgh ⎠

with q = unit width density current discharge (m2/s)

h = height of density current (m)
Q = outlet capacity (m3/s)
γ = specific weight of clear-water (kg/m3)
γo = specific weight of turbid water

The discharge capacity of outlets should equal the density current flow for maximal
discharge of transported suspended sediments. Secondly the outlets should not be too
high above the bed, but at a maximum elevation given by the potential energy of the
density current, represented in the last term of Equation 4.15-1.

Figure 4.15-1 Density current climbing (Bell, 1942)

111
5. MATHEMATICAL MODELS AND CASE STUDIES

5.1 One dimensional mathematical models

5.1.1 Introduction

The theory derived and discussed in previously chapters can now be used to predict
long-term equilibrium reservoir sedimentation. Not only sediment deposition during
storage operation, but also flood flushing with retrogressive erosion have to be evaluated
in order to make accurate predictions of long-term sedimentation. In this chapter one
dimensional and state-of-the-art two dimensional (2D plan) models are described with
representative case studies of turbulent (1D) sediment transport. Mathematical
modelling of density current sediment transport is also described in 2D and 3D models.

It is not the idea to promote specific models in this Chapter, but rather to indicate the
key characteristics such a model should have. The following turbulent sediment
transport models and case studies are presented in this chapter:

ƒ Mike 11 – Reservoir Flushing Model (1D) – Welbedacht Dam, South Africa.

Deposition and erosion model designed to simulate deposition and flushing in
cohesive and non-cohesive sediments.
ƒ GSTARS – Tarbela Dam, Pakistan. Quasi 2D using stream tubes and equilibrium
non-uniform sediment transport equation.
ƒ RESSASS (1D) – Tarbela Dam, Pakistan. Coarse and fine sediment fractions.
ƒ Mike 11 (1D) – Roxburgh Dam, New Zealand River model used on reservoir. Non-
cohesive sediments.
ƒ Mike 21 (2D) (quasi 3D) – Welbedacht Dam, South Africa.
Modified river model to be applicable to shallow reservoirs with fine cohesive
sediment transport.
ƒ 3D model with Three Gorges project case study, China.

112
5.1.2 Reservoir sedimentation model (1D): Mike 11-RFM: Welbedacht Reservoir

5.1.2.1 Model description

Several mathematical models have been developed in the past to simulate reservoir
sedimentation deposition processes, but only a handful have been designed to simulate
sluicing/flushing with reservoir water level drawdown (Han, et al., 1973, 1990;
Thomas and Prasuhn, 1977; Chollet and Cunge, 1980; Sanchez, 1982; Zhang, et al.,
1983; Pitt and Thompson, 1984; Holly and Rahuel, 1990; Yang, 1992; Vasiliev et al.,
1993; Wang, 1993;, etc). Although most "river" mathematical models can simulate the
latter condition, which are close to river flow conditions, there are factors such as fine
sediment transport, cohesive sediment deposits, consolidation of the bed, as well as
different modes of re-entrainment of sediment which also need to be accounted for.

For modelling of reservoir sedimentation processes, a model should be able to simulate

both short-term (flushing) and long-term deposition events, with non-cohesive and
cohesive sediments. It should be borne in mind that the "ideal" mathematical model is
only as good as the theory it is based on.

The mathematical model MIKE 11 of DHI Water and Environment has been modified
to incorporate the theory for reservoir sedimentation processes, which has subsequently
been calibrated and verified with reservoir data. The reservoir flushing model which
has been developed has the following characteristics:

a) One-dimensional (1D) Model

One-dimensional models are mostly used in river and reservoir applications around the
world, although computationally "heavy", two-dimensional (2D) and even 3D models
have been developed. The main constraints in using a 2D model with sediment
transport is often a lack of data for calibration of the model and the long simulation run
times required. For flushing application, a 1D model could be adequate due to the
river-like flow conditions which prevail in the narrow channel which is formed when
reservoir flushing is practised.

Two-dimensional models can be of specific benefit when considering:

113
 • deposition outside the main channel across the wide-open reservoir basins often
encountered;

• sediment build-up at a specific position in a reservoir, such as at a tunnel or

hydropower intake; and

• modelling of flushing when sediment transport conditions vary across the main
channel.

b) St Venant equations for hydrodynamic simulation

A fully hydrodynamic approach is required to describe the rapidly changing flow and
bed level conditions during flushing. The model must also be able to simulate the
supercritical flow conditions which can be encountered during drawdown flushing.
Typical outflow conditions during flood flushing are indicated in Figure 5.1.2-1 as
observed at Phalaborwa Barrage, 1996.

Figure 5.1.2-1 Phalaborwa Barrage flood flushing (900 m3/s, February 1996)

114
c) Sediment transport

The unit (input) stream power equation (Yang, 1973) was implemented in the MIKE 11
model by using user-defined input parameters M and N in Equation 5.3-1:

⎛ vs ⎞
log(C) = M + N log⎜⎜ ⎟⎟ (5.3-1)
⎝ wi ⎠

with C = sediment concentration (ppm)

M,N = values defined by user
v = flow velocity
s = energy slope
wi = settling velocity of fraction i.

Equation 5.3-1 has been calibrated with data from a number of South African reservoirs
for flood flushing and storage operation conditions as described in Chapter 3, with
M = 4,31 and N = 0,343.

Several other well-known sediment transport equations are also available in the MIKE
11 model to simulate transport of coarse fractions.

d) Coupled solution of flow and sediment equations, with sediment continuity

A coupled solution is required due to the rapidly changing hydrodynamic and sediment
transport conditions during flushing and during floods when sediment is deposited.

e) Non-uniform sediment modelling

It is essential in reservoir modelling that sediment transport calculations are carried out
per size fraction in order to model the sorting process and related non-cohesive and
cohesive deposits through the reservoir.

115
f) Non-equilibrium transport of fine sediments

Non-equilibrium transport of fine sediments occurs, which means that transition to

saturated sediment transport capacity conditions is not instantaneous as for coarser
fractions, but a time and distance lag is involved. The multifraction model (MIKE 11)
can operate with a traditional equilibrium transport equation for some fractions and a
non-equilibrium formula for others.

The non-equilibrium modelling is based on solution of the unsteady

advection-dispersion equation, with a source/sink term (ss) describing erosion and
deposition:

δC δC δ 2 C = ss (5.3-2)
+u -D
δt δx δ x2

with D = the dispersion coefficient,

u = flow velocity
t = time
x = distance in direction of flow

In the case of non-cohesive sediments the source/sink term (ss) is represented by:

ss =
(C *
−C ) (5.3-3)
aT

with C* = the equilibrium sediment transport calculated with a sediment

transport formula,
"T" = the time scale defined as settling time (water depth divided by
settling velocity),
"a" = a calibration parameter as described in Chapter 3 and can be
interpreted as a mean (relative) settling depth. In the model
implementation, the coefficient "a" is different for various given
size fractions, for erosion and deposition.

116
 For non-cohesive sediments, incipient motion is determined according to the sediment
transport model used. For cohesive sediment (fractions) the source/sink term should be
represented by:

ss = 0 ; τ < τ ce (No erosion) (5.3-4)

ss = E( τ - τ ce ) ; τ ce < τ < τ cme (Surface erosion) (5.3-5)

ss =
(C *
−C )
; τ > τ cme (Mass erosion, no lag) (5.3-6)
Δt

with τ = bed shear stress,

τ ce = critical shear stress for surface erosion,
τ cme = critical shear stress for mass erosion,
E = erosion rate.

g) Calculation of erosion

The model combines theory for non-cohesive and cohesive sediment to model three
cases of erosion: cohesive sediment, non-cohesive sediment and a mixture of the two.
In the latter case, a linear combination of the cohesive and non-cohesive relationship is
used. The erosion rate, assuming both cohesive and non-cohesive sediments, is
calculated, and the actual rate determined via linear interpolation.

For surface erosion, the erosion rate is checked against sediment transporting capacity
in order not to exceed the capacity. Mass erosion modelling is based on the assumption
that (instantaneous) erosion will take place to satisfy the sediment transporting capacity.

The availability of sediment for erosion is checked. If the calculated erosion rate
exceeds the available sediment supply, the rate is reduced to reflect the amount of
available sediment.

117
h) Cross-section deformation

Solution of the bed continuity equation determines whether erosion or deposition will
occur.
When deposition occurs it is assumed that:

dz = aDb (5.3-7)

with dz the change in bed level, a and b input calibration parameters and D the flow
depth.

During erosion phases it is assumed that the cross-section will be of a trapezoidal shape
described by regime type equations (relating width and depth to the discharge). Bank
slopes of the trapezoid are required as input parameters. The difference between the
actual bed level and that of the trapezoidal section is integrated to determine what
changes are required.

The following special cases have been identified:

- The trapezoidal section is "smaller" than the actual section: Equation 5.3-7 is
used for changing the bed level

- Erosion goes below the original (bedrock) section : increase width if bed is non-
erodible and lower bed if sides are non-erodible; otherwise erosion is limited.

Typical surveyed cross-sections of Welbedacht Reservoir are shown in Figure 5.1.2-2.

118
 Figure 5.1.2-2 Welbedacht Reservoir flushing channel deformation

i) Consolidation of sediments

Modelling of consolidation is important when critical conditions for mass erosion are
linked to sediment characteristics and density. A consolidation methodology as
proposed by Miller (1953) which relates sediment density changes with time to reservoir
operation and sediment characteristics was incorporated in the model, with certain
changes:

- The consolidation constant used by Miller (1953) was calibrated with South
African reservoir data based on reservoir basin surveys. This was necessary
because it was found that sedimentation and changing water demands lead to
changes in reservoir operation patterns in time.
- Erosion and deposition rates are not constant with time, and the integration
method adopted by Miller had to be changed by incorporating an effective time
approach which accounts for previous deposits as well as recent changes in the
bed surface.

119
5.1.2.2 Calibration and verification of the 1D model (RFM) with South African reservoir
data

a) General

A number of South African reservoirs have been monitored during flood flushing events
since 1993. The data for Welbedacht Reservoir are used here to calibrate the 1D model.
Welbedacht Dam (Figure 5.1.2-3) was completed in 1973 on the Caledon River and is
located in a high sediment yield region. During the first three years of operation, it had
lost 36 million m3 of its original 114 million m3 capacity due to sedimentation. By 1991
the full supply capacity was only 17 million m3.

Figure 5.1.2-3 Welbedacht Dam

During 1991 it was decided to use flood flushing in order to regain some of the storage
capacity. The dam is equipped with five gates, but their elevation is 15 m above the
original river bed. Two flushings during October 1991 managed to scour a channel with
a bed width of approximately 50 m, and a longitudinal bed profile as indicated in Figure
5.1.2-4. During subsequent flood flushings during the rainy seasons of 1994 and 1995,
additional field data were gathered with which the 1D model was calibrated and
verified.

120
5.1.2.3 Calibration of the 1D model

The bed shear stress required for mass erosion was determined for two flood flushings
during October 1991, using a trapezoidal flushing channel profile in the model and the
1991 sedimentation levels. The scoured bed profile as surveyed in 1992 after the
flushings was compared with simulated profiles and a critical mass erosion bed shear
stress at the end of flushing was determined.

Unfortunately, sediment transport loads were not measured during the 1991 flushings
and the model could therefore only be calibrated in terms of bed shear stress to achieve
the observed scoured bed profile. A sediment transport equation calibrated in this
bulletin using existing data from a number of South African reservoirs (with flushing
and storage operation) was used. The observed and calibrated bed profiles are shown in
Figure 5.1.2-4. Observed inflows and water levels at the dam, with inflow suspended
sediment concentrations based on a sediment load-discharge rating curve were used as
boundary conditions in the model. The surveyed 1991 reservoir bed was used as the
starting condition, and calibration was based on the surveyed 1992 reservoir bed profile.

Figure 5.1.2-4 Calibrated and observed flushing channel bed profiles at Welbedacht
Reservoir

121
5.1.2.4 Verification of the 1D model

Using exactly the same model setup which had been used to determine shear stresses
(1991 flushings), it was possible to simulate a 1995 flushing at Welbedacht Dam while
using as input the stream power relationship which had been calibrated with South
African reservoir data. The simulated curve of cumulative water discharge versus
cumulative flushed sediment is shown in Figure 5.1.2-5, and excellent agreement with
observed data was obtained. The observed and predicted sediment transport versus time
relationship is shown in Figure 5.1.2-6. The 1D model prediction reliability has
therefore been proven by the 1995 flushing event.

Although not calibrated for a wide range of flood events, the calibrated model for
Welbedacht Reservoir might be used to investigate changes in the outlet configuration
and operation to improve sluicing/flushing efficiency and to predict sustainable long-
term equilibrium reservoir storage capacities.

Figure 5.1.2-5 Welbedacht Reservoir 1995 flushing; cumulative double mass plot
verification of sediment transport

122
 Figure 5.1.2-6 Welbedacht Reservoir 1995 flushing; verification of sediment transport
modelling

5.1.2.5 Simulation of long-term Welbedacht Reservoir capacity

Using the same model configuration as calibrated and verified in the previous sections,
simulation of the long-term reservoir capacity which could be achieved through flood
flushing was carried out with the current outlets (at a high elevation), but also a scenario
with hypothetical gates located at the original river bed.

With the estimated 1 in 2-year flood peak discharge at Welbedacht Dam being equal to
800 m3/s and the present reservoir capacity being less than 1 percent of the mean annual
runoff, sufficient excess water with high erosive power is available for flushing. Taking
the dominant inflow at 500 m3/s during floods (conservatively low) and inflow
suspended sediment concentrations based on the sediment load-discharge rating curve,
the long-term equilibrium bed profiles which have been found by simulation are
indicated in Figure 5.1.2-7.

123
Figure 5.1.2-7 Simulation of long-term equilibrium sedimentation at Welbedacht
Reservoir with flood flushing and different outlet configurations

The current high elevation of the outlets prevents retrogressive erosion far upstream
from the dam. In the scenario with bottom outlets and water level drawdown, flushing
will be much more efficient. According to Figure 5.1.2-7 it looks as if a major part of
the original 114 million m3 reservoir capacity can be restored, but only the flushing
channel will be scoured down to the original bed level with bottom outlets provided.
The flushing channel shape determines the long-term storage capacity of the reservoir,
and overbank sediment deposits will not be resuspended during flushing (unless
flushing duration is long and meandering of the flushing channel develops). The long-
term equilibrium storage capacities that can be maintained, simulated with high outlet
(current) and bottom outlets, are 10 million m3 and 30 million m3, respectively.

The Mike 11 – RFM considers deposition and erosion processes. Simulation results with
this model for a 1973-1976 deposition period (used for calibration) is presented in
Chapter 3.

124
5.1.3 Reservoir Sedimentation Model: GSTARS: Tarbela Dam, Pakistan (Yang and
Simoes, 2003)

5.1.3.1 Background

Tarbela Dam, located in northern Pakistan along the Indus River, is the largest earth-fill
dam in the world. The reservoir, with a gross storage capacity of 14340 million m3, is a
38 km long run of the river type of reservoir with two major tributaries, the Siran and
the Brandu. Tarbela’s main function is provision of water for irrigation. Additionally,
the hydropower capacity of the power station totals 3478 MW and produces 32% of
Pakistan’s needs. The reservoir’s storage capacity has been continuously depleted since
the dam has been built in 1974, with an annual inflow rate of 265 million tons of
sediment, most of which in the silt and clay range. This loss in capacity threatens the
resources and revenue associated with the dam and reservoir.

5.1.3.2 Mathematical model description

GSTARS3 (Generalized Sediment Transport model for Alluvial River Simulation is the
most recent version of a series of numerical models for simulating the flow of water and
sediment transport in alluvial rivers developed at the Sedimentation and River
Hydraulics Group of the Technical Service Center, U.S. Bureau of Reclamation,
Denver.

GSTARS consists of four major parts. The first part uses both the energy and the
momentum equations for the backwater computations. This feature allows the program
to compute the water surface profiles through combinations of subcritical and
supercritical flows. In these computations, GSTARS3 can handle irregular cross
sections regardless of whether single channel or multiple channels are separated by
small islands or sand bars.

The second part is the use of the stream tube concept, which is used in the sediment
routing computations. Hydraulic parameters and sediment routing are computed for
each stream tube, thereby providing a transverse variation in the cross section in a semi-
two-dimensional manner. Although no flow can be transported across the boundary of a
stream tube, transverse bed slope and secondary flows are phenomena accounted for in

125
 GSTARS3 that contribute to the exchange of sediments between stream tubes. The
position and width of each stream tube may change after each step of computation. The
scour or deposition computed in each stream tube give the variation of channel
geometry in the vertical (or lateral) direction. The water surface profiles are computed
first. The channel is then divided into a selected number of stream tubes with the
following characteristics: (1) the total discharge carried by the channel is distributed
equally among the stream tubes; (2) stream tubes are bounded by channel boundaries
and by imaginary vertical walls; (3) the discharge along a stream tube is constant (i.e.,
there is no exchange of water through stream tube boundaries).

The third part is the use of the theory of minimum energy dissipation rate (Yang, 1971,
1976; Yang and Song, 1979, 1986) in its simplified version of minimum total stream
power to compute channel width and depth adjustments. The use of this theory allows
the channel width to be treated as an unknown variable, which is one of the most
important capabilities of GSTARS3. Whether a channel width or depth is adjusted at a
given cross section and at a given time step depends on which condition results in less
total stream power.

The fourth part is the inclusion of a channel bank side stability criteria based on the
angle of repose of bank materials and sediment continuity.

GSTARS3 is a general numerical model developed for a personal computer to simulate

and predict river and reservoir morphological changes caused by natural and
engineering events. Although GSTARS3 is intended to be used as a general engineering
tool for solving fluvial hydraulic problems, it does have the following limitations from a
theoretical point of view (Yang and Simoes, 2003):

a) GSTARS3 is a quasi-steady flow model. Water discharge hydrographs are

approximated by bursts of constant discharges. Consequently, GSTARS3 should not be
applied to rapid, varied, unsteady flow conditions.

b) GSTARS3 is a semi-two-dimensional model for flow simulation and a semi-three-

dimensional model for simulation of channel geometry change. It should not be applied

126
 to situations where a truly two-dimensional or truly three-dimensional model is needed
for detailed simulation of local conditions

c) GSTARS3 is based on the stream tube concept. Secondary currents are empirically
accounted for. The phenomena of diffusion, and superelevation are ignored.

d) Many of the methods and concepts used in GSTARS3 are simplified approximations of
real phenomena. Those approximations and their limits of validity are, therefore,
embedded in the model.

5.1.3.3 Model setup

In this example, GSTARS3 is used to simulate 22 years of reservoir sedimentation

(from 1974 through 1996) for a reach that spans nearly 93 km upstream from the
Tarbela Dam (see figure 5.1.3-1). The hydrology of the system is given in figure
5.1.3-2, together with the dam operation. The tributaries have a relatively small
contribution when compared with the main stem discharge, therefore they are not
included in figure 5.1.3-2 (but they are included in the computations).

There is a large percentage of silt and clay in the sediments in transport, but there is no
data to simulate them using the Krone/Ariathurai methods. Secondly, analysis of the
1996 cross-sectional data suggests that deposition occurs in the form of a horizontal fill
as can be observed in figure 5.1.3-3 below. Finally, there exists a rating curve but
during the calibration runs it was observed that the amount of deposition was severely
underpredicted. That rating curve was adjusted to correctly reproduce the reservoir
deposition volume. The distribution of the incoming sediment load was also a subject of
calibration.

127
 Figure 5.1.3-1 Tarbela dam and reservoir. The points (+) mark the thalweg and the
locations of the cross sections used in this study (Yang and Simoes, 2003)

Figure 5.1.3-2 Hydrology and dam operation for Tarbela Reservoir (1974 to 1996) (Yang
and Simoes, 2003)

The Tarbela Reservoir’s bathymetry was discretized using existing surveyed cross
sections, which are marked in figure 5.1.3-1. The horizontal sediment deposition
observed in figure 5.1.3-3 indicates that there is not much transverse variation in the
sedimentation processes, therefore GSTARS3 simulations were carried out using one
stream tube.

128
 Figure 5.1.3-3 Two reservoir cross sections showing uniform sedimentation (Yang and
Simoes, 2003)

Yang’s (1973) equation was used for this study. Because there is no information
concerning the deposition characteristics of the silt and clay fractions, it is difficult to
use the Krone/Ariathurai methods effectively. Instead, the Yang (1973) was
extrapolated for these size ranges. (The particle size distributions are in the range of
0.002 to 2.0 mm). This approach can sometimes yield good results, especially in
mainly depositional processes such as those occurring in large reservoirs. However,
care should be exercised, and the results of the simulations should always be
confirmed by careful validation using field data.

Daily time steps for the hydraulic computations and 4.8 hours for sediment routing
computations (8 040 time steps for hydraulics, 40 200 times steps for sediment) were
used.

5.1.3.4 Simulation results

The results of a 1996 survey carried out in Tarbela Reservoir, corresponding to the
end of the period of the simulation, are used here. Note that this example does not
constitute an exhaustive and definitive study of the sedimentation processes in Tarbela
reservoir for the 1974-1996 period.

129
 The simulation results for the thalwegs are shown in figure 5.1.3-4. They are in good
agreement with measurements, especially as in what concerns the location of the
frontset of the delta and its slope. This is important to determine capacity loss, the
useful life of the reservoir, and the impact that dam operations have on the reservoir’s
deposition pattern.

Cross-sectional geometries were better predicted in the downstream reservoir region

than in the upstream region. That is because the upstream part has mostly riverine
characteristics, and horizontal deposition is not the most appropriate technique for this
circumstance. However, this region is limited to the first 21 cross sections, which
represents less than one fourth of the entire simulated reach. Even then, deposition
volumes are in general well predicted, even if thalweg elevations are not very
accurate. Two representative cross section in this region are shown in figure 5.1.3-5
for comparison purposes.

Figure 5.1.3-4 Results of the simulation of the Tarbela delta advancement over a period of
22 years (Yang and Simoes, 2003)

130
 Figure 5.1.3-5 Comparison of measurements and GSTARS3 computation for two cross
sections in the upstream region of Tarbela reservoir (Yang and Simoes, 2003)

In the reservoir region, which constitutes the focus of the study, deposition volumes are well
predicted, in spite of a small tendency to overpredict the thalweg elevations. Four
representative cross sections are shown in figure 5.1.3-6. These cross sections were taken
from the full width of the reservoir region.

Figure 5.1.3-6 Comparison of measurements and GSTARS3 computation for two cross
sections in the reservoir region of the study reach

131
The error associated with the predicted thalwegs is shown in figure 5.1.3-7. Most data are
inside the 20% error band, which is a very good result for this type of simulation (22 years of
sedimentation spanning a 36 km reach). However, this study could easily be improved with
the use of little additional data. Such data would comprise more accurate bed-sediment size
distributions, as well as more information about the inflowing sediment sizes traveling
through the Indus. Further improvements would be attained from a study of the cohesive-
sediment fraction properties within the reach (Yang and Simoes, 2003).

Figure 5.1.3-7 Relative error of the thalweg elevation predictions

5.1.4 Reservoir Sedimentation Model RESSASS: Tarbela Dam (TAMS, 1998)

a) Background

Tarbela Dam was built between 1970 and 1975 and is located on the Indus. The dam
is vital to Pakistan’s economy as it provides 40 per cent of the country’s water storage,
crucially important for irrigation during the dry season, and 35 per cent of the
country’s energy requirements from hydroelectric generation. About 25% of the
original live storage had been lost and the sediment delta was approaching the dam by
1997.

b) Description of the model

The numerical model RESSASS of HR Wallingford was used to simulate reservoir

sedimentation in Tarbela Reservoir. The model is one-dimensional in the sense that
all hydraulic variables are considered constant at any cross-section at any particular

132
 time. The model is based on the equations that describe flow and sediment movement.
Within the model, sediment transport calculations are carried out for a range of
different sediment sizes. The model is a time-stepping one, that is, initial conditions
are input to equations which predict water levels and bed levels a short time later,
typically one day. These predictions then provide the input conditions for the next
time step. The process is repeated many times to make predictions over the required
time period. Cross-sections are used to specify the geometry of the reservoir. The
inflow of water and sediment at the upstream end of the reservoir are described using
a discharge-time sequence and a discharge-sediment concentration relationship. The
model then predicts revised bed levels and the location and composition of sediment
deposits.

The model runs on a 60 year sequence of sediment and water inflows. The sequence
consists of the 1967 to 1996 inflows to Tarbela repeated once to give the required 60
year sequence. The inflows are based on the flow records at Besham Qila. The
sediment inflows were calculated using the sediment rating equation on these inflows
together with the Ackers and White equation. For the model verification runs, the
model water levels were matched to the observed levels since the first impoundment
of the reservoir.

The model operates with the following boundary conditions:

• Variations of incoming flow with time;

• Corresponding variations of incoming sediments with time;
• Variations of water level at the dam as defined by the operating policy.

This means that, at any particular time, the model computes the outflow through the
dam (the model does not take account of where that outflow occurs or whether the
infrastructure at Tarbela is capable of delivering that discharge). It also computes the
outflow of sediments at any particular time based on flow conditions immediately
upstream of the dam.

133
c) Verification of the model performance

The model was verified by simulating the observed sediment deposition from the date
of impoundment to 1996. The observed flow sequence at Besham Qila from 1975 was
used together with the appropriate sediment rating curve. Observed historic reservoir
water levels were used for the downstream boundary conditions. Based on field
observations, the model was set up with two sand sizes, 0.155 mm and 0.200 mm, and
five silt sizes giving a range of settling velocities between 0.002 mm/s and 3.6 mm/s.
With these realistic parameters the model gave good agreement between observed and
predicted bed levels and volumes of deposition. Figure 5.1.4-1 and Figure 5.1.4-2
show the predicted and observed longitudinal profiles and the predicted and observed
stage-storage curves for 1996, respectively. It can be seen that the agreement between
the model and the observed values is good. The agreement between the observations
and predictions is particularly good in the neighborhood of the pivot point and
immediately in front of the dam. One can thus have confidence that the model is
representing accurately the processes of sediment deposition within Tarbela reservoir.

Figure 5.1.4-1 Tarbela Dam spillway

134
 Figure 5.1.4-2 Simulated and surveyed longitudinal profile of Tarbela Reservoir

5.1.5 River model Mike 11: Lake Roxburgh, New Zealand (Mackay et al., 2000)

In this study particular attention was given to determining the effect on Alexandra of sediment
redistribution in Lake Roxburgh, New Zealand. The relationship between water level and
flow at Alexandra for given headwater level conditions at Roxburgh Dam was determined
based on the latest cross-section survey data. The latest rating curve was also compared to
previous curves.

The sediment redistribution which has occurred in Lake Roxburgh and the practice of
lowering the reservoir in advance of a flood event have combined to reduce flood levels at
Alexandra by 1.7 m from January 1994 to December 1999.

Overall a total of 11.5 million m3 of sediment has been eroded from the middle and upper
reaches of Lake Roxburgh since the January 1994 flood event. The total sediment moved
since the full bed surveys completed in February 1994 and December 1999 is about 10
million m3. An additional 1.5 million m3 of material was moved between the partial bed
survey in January 1994 and the February 1994 bed survey.

135
Water levels at Alexandra during the November 1999 flood would have peaked at about 143.8
m if there had been no sediment redistribution in the lake since 1994, and the lake had not
been drawn down at the flood peak. This level is 1.5 m higher than the measured peak water
level of 142.3 m in November 1999.

A sediment transport model was calibrated that gave a good comparison between actual
measured and computed sediment volume changes in Lake Roxburgh. The key aspects of the
model (Mike 11) selected were:

ƒ Ackers and White sediment transport equation used since reservoir sediment is sand and
gravel.
ƒ One-dimensional since the reservoir is long and narrow.

The model was calibrated by adjusting the sediment grain size from upstream to downstream
until a satisfactory match in bed volume changes was achieved. Calibration using sediment
grain size implicitly allows for the reduced erodibility of fine cohesive sediments without
having to be concerned with the details of the erosion process.

This calibrated model was used to simulate the effects of continuing the flood drawdown
strategy (to promote sediment redistribution within Lake Roxburgh) using either 850 m3/s or
1100 m3/s as the threshold flow for drawing the lake down.

For a flood drawdown threshold of 850 m3/s, the model predicted further erosion of 1.3
million m3 of bed material over 5 years and 2.1 million m3 over 10 years. For flows between
2400 and 3600 m3/s, this would reduce water levels at Alexandra Bridge for a given
headwater level at Roxburgh Dam by a further 0.7 – 0.8 m in 5 years time and by 1.1 m in 10
years time (Figure 5.1.5-1).

136
Figure 5.1.5-1 Predicted Future Bed Level and 3400 m3/s-1 Flood Profiles at Year 0.5
and 10 (850 m3/s-1 F.D.D. threshold)

137
5.2 Computational modelling of reservoir sedimentation and flushing with a two-
dimensional model

5.2.1 Background

This section of the paper describes a mathematical model, which is applicable to shallow
reservoirs, and its application to the Welbedacht Reservoir in South Africa. The mathematical
model is an adaptation of an existing river morphological model, which has been used in
numerous studies throughout the world. The key characteristics of the model are:

• Solution of the fully dynamic 2D St. Venant equations

• Parameterised description of the vertical distribution of main flow, helical flow and
• Suspended sediment transport
• Representation of time and space lag of suspended sediment
• Simulation of dynamic development of bed level, alluvial resistance and bank erosion.

Computational river models can be used to simulate reservoir sediment flushing, but with
special consideration of the following:
• Sediment transport into the reservoir has to consider the total load, including the fine
washload
• Deposited sediment in the reservoir is often cohesive and critical conditions for re-
entrainment should be incorporated
• During flushing retrogressive erosion is dominant, but both erosion and deposition
processes are active
• During flushing near supercritical flow conditions can be reached

The research and application have been carried out as a joint effort between University of
Stellenbosch, South Africa, and DHI Water & Environment, Denmark. The Mike 21C model
was used.

The Welbedacht Dam was completed in 1973 on the Caledon River, South Africa, which is
located in a high sediment yield region with a sediment yield of about 3000 t/km2.a. During
the first three years of operation the reservoir had lost 36 million m3 of its original 114 million
m3 storage capacity due to sedimentation (Figure 5.2-1). A longitudinal profile of the

138
reservoir sedimentation is shown in Figure 5.2-2. As a verification case the model was used
to simulate this three year period (1973 to 1976). Subsequently, alternative operation
strategies were tested in the model with the objective to sluice the sediment through the
reservoir and flush out accumulated sediment.

120

100
Storage Capacity (Mcm)

80

60

40

20

0
1970 1975 1980 1985 1990 1995 2000 2005
Year

Figure 5.2-1. Observed loss in storage capacity due to sedimentation at Welbedacht Dam

1415

1410 Bed level 1998

survey
Full supply level1402.9 m
1405

1400
Elevation (m)

Sediment
Sediment Dam

1395
105 Mcm lost = 92 %
1390 Original
riverbed level
1973
1385

1380

1375
0 10 20 30 40 50 60
Distance (km)

Figure 5.2-2. Longitudinal profile of the Welbedacht Reservoir sedimentation

139
5.2.2 Data

The geometry of the reservoir was taken from aerial photography (Figure 5.2-3), and
mapped with a curvilinear grid, as shown in Figure 5.2-4. The grid contains 250 cells
in the longitudinal direction, and 15 points across, totaling 3750 grid cells fully
flooded. The model was formulated with an upstream inflow and a downstream
reservoir water level. Time-series were available for 1973-76 both for inflow
discharge and reservoir level, indicated in Figures 5.2-5 and 5.2-6.

Dam

Figure 5.2-3. Aerial photograph of Welbedacht Reservoir

Dam

Figure 5.2-4 . Curvilinear grid used for the 2D model, 250x15 cells

140
 2500
2000
1500
m3/s

1000
500
0
Aug-73 Nov-73 M ar-74 Jul-74 Nov-74 M ar-75 Jul-75 Nov-75 M ar-76 Jul-76

Figure 5.2-5. Time series of observed reservoir inflow

1405
Water level (m)

1400
1395
1390
1385
1380
Aug-73 Nov-73 M ar-74 Jul-74 Nov-74 M ar-75 Jul-75 Nov-75 M ar-76 Jul-76

Figure 5.2-6. Time series of observed reservoir water level at dam

Basson and Rooseboom, (1997) gives the following sediment characteristics for the
Welbedacht inflow:

• Fraction 1: d50<0.05 mm, 76% (mud)

• Fraction 2: 0.05<d50<0.106, 19% (silt)
• Fraction 3: 0.106<d50<0.25 mm, 5% (fine sand)

For this study the following relationship between concentration and discharge was found from
observed data:

C = 793.32Q 0.664 5.2-1

141
 Where C is in g/m3 and Q in m3/s. All the sediment is basically taken as being the
same cohesive type.
The relationship between discharge and concentration is however poor due to
sediment availability limitation from the catchment.

The settling velocity seems to be in the order of magnitude 1 mm/s. It is noted that this
parameter does not play a completely decisive role in the cohesive sediment transport
model. The settling velocity only influences the rate of deposition, which may sound
critical. However, the system tends to stabilise itself because a lower settling velocity
will cause slower deposition, which causes the sediment concentration to rise, again
causing higher deposition. The sensitivity towards the settling velocity was tested and
a change of about 10% was found by doubling the value.

By using the hydrograph from 1 August 1973 to 30 September 1976 it was found that
the total inflow to the reservoir was 4925 million m3 of water. The sediment
concentration expression yields that along with the water, 53 million m3 of sediment
entered the reservoir (assuming 2650 kg/m3 density for the sediment). According to
Basson et al (1996), the sedimentation during the period was 36 mill m3. This suggests
an average trapping efficiency of 68% for the Welbedacht reservoir.

5.2.3 Theoretical Background

The modelling tool is described in the following, including grid generation,

hydrodynamic, cohesive sediment transport and bathymetry development.

5.2.4 Grid Generation

MIKE 21C is designed to function on any orthogonal curvilinear grid. Here we

employ a grid generated with an anisotropic conformal mapping method, which is
much more flexible than conformal mapping, as there is no constraint on the cell
aspect ratio. It is often preferable to use cells that are elongated in the flow direction
for this type of models. Anisotropic conformal mapping is described by the
anisotropic Laplace equations:

142
 ( gxξ ) ξ + ( g −1 xη )η = 0
5.2-2
( gyξ ) ξ + ( g −1 yη )η = 0

Where:

g Grid weight
x,y Horizontal coordinates (m)
ξ,η Transformed integer coordinate system

The grid weight, equal to the aspect ratio of the grid, is given by:

xξ2 + yξ2
g= 5.2-3
xη2 + yη2

Orthogonality is enforced in all boundary points through the non-linear condition:

xξ xη + yξ yη = 0
5.2-4
f ( x, y ) = 0

Which is stating that the grid line is orthogonal to the boundary (zero dot product) and that the
grid line follows the boundary line (second condition). The elliptic equations are solved
implicitly with continuous update of the boundary condition that is handled by Newton
iteration. Stone's Strongly Implicit Procedure (Stone, 1968) is applied for the elliptic
equations.

5.2.5 Hydrodynamics

The hydrodynamics are described by the 2-dimensional Saint-Venant equations:

∂p ∂ ( p 2 / h) ∂ ( pq / h) gp p + q ∂ ⎛ ∂ ( p / h) ⎞
2 2
∂s ∂ ⎛ ∂ ( p / h) ⎞
+ + + + gh = h ⎜E ⎟ + h ⎜⎜ E ⎟
∂t ∂x ∂y 2
C h 2 ∂x ∂x ⎝ ∂x ⎠ ∂y ⎝ ∂y ⎟⎠

∂q ∂ ( pq / h) ∂ (q 2 / h) gq p + q ∂ ⎛ ∂ ( q / h) ⎞
2 2
∂s ∂ ⎛ ∂ ( q / h) ⎞
+ + + + gh = h ⎜E ⎟ + h ⎜⎜ E ⎟ 5.2-5
∂t ∂x ∂y 2
C h 2 ∂y ∂x ⎝ ∂x ⎠ ∂y ⎝ ∂y ⎟⎠
∂h ∂p ∂q
+ + =0
∂t ∂x ∂y

143
Where:

p,q Flux field (m2/s)

h Water depth (m)
t Time (s)
g Acceleration of gravity (9.81 m/s2)
C Chezy number (m1/2/s)
s Surface elevation (m)
E Eddy viscosity (m2/s)

The Saint-Venant equations are solved on the curvilinear grid with a finite-volume
scheme. State-of-the-art methods from CFD are applied for the solution, which
incorporates the use of a Cartesian base for the velocity field, non-staggered allocation
with momentum interpolation (Majumdar et al, 1992). The SIMPLER method
(Patankar, 1980) is applied for the continuity equation.

5.2.6 Cohesive sediment transport model

The transport of the cohesive sediment is described by an advection-dispersion

equation:

∂hc ∂p ' c ∂q ' c ∂ ⎛ ∂c ⎞ ∂ ⎛ ∂c ⎞

+ + = ⎜ hD xx ⎟ + ⎜⎜ hD yy ⎟+E−D 5.2-6
∂t ∂x ∂y ∂x ⎝ ∂x ⎠ ∂y ⎝ ∂y ⎟⎠

Where:

p',q' Modified flux field (m2/s)

c Concentration (g/m3)
Dxx Dispersion in the x-direction (m2/s)
Dyy Dispersion in the y-direction (m2/s)
E Erosion function
D Deposition function

144
 The modified flux field that transport the suspended sediment is derived from the
depth integrated flux field in the manner:

⎛ p' ⎞ ⎛ p⎞ h ⎛− q⎞
⎜⎜ ⎟⎟ = α 01 ⎜⎜ ⎟⎟ + α 02 ⎜⎜ ⎟⎟ 5.2-7
⎝ q' ⎠ ⎝q⎠ R⎝ p ⎠

Where a01 and a02 are functions of the distribution of momentum and sediment over
the water column. The term h/R is the water depth divided by the streamline radius of
curvature; the latter derived from the flow field. The modified flux field arises from
the 3-dimensional character of the problem.

α01 modifies the streamwise convection, and represents the fact that the sediment

concentration rises towards the bed, while the velocity rises towards the surface. The
streamwise convection of the sediment is hence not as effective, as α01=1 would imply.
A value of α01=1 is found for uniformly distributed sediment, i.e. very fine material.
α01 is calculated from the logarithmic velocity profile and the distribution of sediment.

α02 represents the impact of secondary flow, and produces convection across the

streamlines. α02 is calculated from the helical flow taken from standard theory (see
e.g. Rozowskii, 1957), and the distribution of sediment.

The α01 and α02 parameters are calculated from local values of the flow velocity, flow
resistance and settling velocity. The calculation is done on each morphological time-
step in describing the change in sediment concentration and sedimentation in time.

An implicit scheme is applied for the AD equation in which the local availability of
sediment is accounted for by limiting the erosion to not surpass the available mud in
the cell. The implicit solution of the AD equation furthermore allows for implicit
updating of the local mud layer thickness, which is done through the source/sink terms
of the equation (sediment entering the water column comes from the bed, and vice
versa). The implicit AD scheme is unconditionally stable for any choice of the time-
step.

145
 The dispersion in the equation originates from the profile functions, while additional
dispersion can be added. Refer to DHI (2003) for details about the dispersion terms.
Herein only the dispersion associated with the profile functions were applied
(particularly important across streamlines). No additional dispersion has been added.

A standard cohesive model gives the erosion and deposition functions (E and D):

m
⎛ τ ⎞
E = E 0 ⎜⎜ − 1⎟⎟ , τ > t ce
⎝ τ ce ⎠ 5.2-8
⎛ τ ⎞
D = w s c⎜⎜1 − ⎟, τ < t cd
⎟
⎝ τ cd ⎠

Where:

ws Settling velocity, ws ~ 1 mm/s

τce Critical shear stress for erosion, τce ~0.2 N/m2 for fluid mud and ~0.6 N/m2 for mud
τce Critical shear stress for deposition, τce ~ 0.05 N/m2
E0 Erosion constant, E0 ~0.1 g/m2/s
m Exponent (non-linearity) of the erosion, m ~ 1-3

The mentioned value ranges are standard, though the variation from situation to situation can
be substantial.

5.2.7 Calibration of sedimentation during 1973-76

Calibration/validation is carried out by comparing observed and simulated

sedimentation rates for the initial three years of operation from 1973-76.

The present model has 82 million m3 of initial water storage calculated by assuming
1402.9 m (full supply level) water level everywhere in the model. According to
Basson et al (1996) the initial storage capacity was 114 mill m3, which is substantially
higher. The reason for this is that the present model does not cover the total area
counted in the 114 mill m3, e.g. tributaries etc.

146
 The initial bathymetry from 1973 (based on cross section data) and the simulated
bathymetry in 1976 are shown in Figure 5.2-7. The following cohesive model
parameters were used in the simulation:

ws Settling velocity 1 mm/s

τce Critical shear stress for erosion 0.6 N/m2
τce Critical shear stress for deposition 0.05 N/m2
E0 Erosion constant 0.1 g/m2/s
m Exponent (non-linearity) of the erosion 2
n Porosity 0.5
ρ Density 2650 kg/m3

1973 Simulated 1976

Figure. 5.2-7. Initial bathymetry from 1973 (left) and simulated bathymetry in 1976
(right).

The simulated bathymetry development from 1973-1976 was integrated into volumes, as
shown in Figure 5.2-7. The simulated sedimentation was 25 million m3, while the observed
was 36 mill m3. The reason for this is that the model does not cover the whole reservoir. The
ratio between the simulated and observed sedimentation matches the ratio between the full
initial reservoir volume and the volume in the model.

The simulated sedimentation is fairly sensitive to the model parameters. Of particular

importance is the critical erosion shear stress. In the present simulations there is actually some
erosion during the flood peaks, which can be seen also from Figure 5.2-8.

147
 100
Water volum e
90
Sedim ent volum e
80
70
Volume (mill m3)

60
50
40
30
20
10
0
Aug-73 Nov-73 Mar-74 Jul-74 Nov-74 Mar-75 Jul-75 Nov-75 Mar-76 Jul-76
Date (1973-76)

Figure. 5.2-8. Simulated sedimentation development from 1973-76, sediment volume and
water volume in the reservoir.

5.2.8 Flushing during 1991

Flushing during floods has been carried out since 1991 at Welbedacht Dam. A typical
flushing viewed from the right bank at the dam looking upstream is shown in Figure
5.2-9.

According to Basson et al. (1996) the reservoir full supply capacity volume had
dropped to 17 million m3 by 1991. The flushing operation was simulated in the model
by linearly decreasing the reservoir water level from full supply level to 1392 m
during a period of 5 hours.

Figure 5.2-9. Welbedacht Reservoir during flushing

148
The simulations yielded an incised channel being cut into the reservoir. The channel was cut
from downstream moving upstream. As the erosion migrates upstream into a wider section, it
will start dividing into more channels, as shown in Figure 5.2-10. A comparison with
observed minimum bed levels is given in Figure 5.2-11. The agreement is satisfactory,
considering that the survey was not carried out immediately after the flushing.

Figure 5.2-10. Bathymetry before (left) and simulated after (right) flushing in 1991.

1405

1400
Minimum bed level (m)

1395

1390

1385
Simulated minimum (after flushing)
1380 Minimum (before flushing)
Observed after flushing
1375
40.4 41.4 42.4 43.4 44.4 45.4 46.4
Chainage (km )

Figure 5.2-11. Observed and simulated minimum bed level as function of chainage before
and after flushing in 1991.

149
5.2.9 Flushing with Low Level Outlets

The gates at Welbedacht Dam are not placed at the original river bed level, but 15 m
above the bed. If the gates were located at the bed level, flushing should be more
effective. This scenario was simulated with the model for a one month duration
flushing at 600 m3/s (Figure 5.2-12) and it is clear it has a significant impact on the
long-term storage capacity.

Dam

Figure 5.2-12. Flushing simulation with gate invert level at the bed at 1380 m

5.2.10 Conclusions

Flushing of sediment from a reservoir requires: excess water, suitable large low level
outlets, a steep and narrow reservoir basin and judicious operation. Flushing could be
carried out on a seasonal basis, or during specific flood events (usually in semi-and
conditions).

A basic cohesive transport model was incorporated into a 2D curvilinear modelling

system. This model was originally developed for sand transport, and has a pseudo 3D
description included accounting for the 3-dimensionality of the flow and sediment.
The model solves the Saint-Venant equations on any orthogonal grid combined with
an advection-dispersion equation that includes the 3-dimensionality through profile
functions. The inclusion of the 3D effects results in the transporting flux field being
different from the depth-integrated flux field. In the streamwise direction the

150
 convection is reduced due to the fact that most of the sediment is located in the lower
part of the water column. The 3-dimensionality also results in depth-integrated
convection across the main streamline due to streamline curvature driven secondary
flow.

A calibration of the model was carried out for the Welbedacht reservoir. The default
parameters for the cohesive model gave satisfactory results, though there is room for
improvement. The model was then applied for simulation of flushing of the same
reservoir in 1991. It was demonstrated that the model is capable of capturing the
incised channel that is cut into the reservoir bed during such flushing operations, and
that the model can also handle the 2-dimensionality of the channel formation. The
model can be used as management tool to determine the effectiveness of flushing
operation under various flood and outlet conditions.

5.3 Three-dimensional Mathematical Modelling (turbulent sediment transport)

5.3.1 3D Model Equations

Generally, the 3D flow field is determined by the following Reynolds-averaged continuity

and Navier-Stokes equations (Wang and Wu, 2004):

∂ui
=0 (5.3-1)
∂xi

∂ui ∂ (ui u j ) 1 ∂pi 1 ∂τ ij

+ = Fi − + (5.3-2)
∂t ∂x j ρ ∂xi ρ ∂x j

where:

ui-= Velocity components (i = 1, 2, 3)

Fi= External forces including gravity per unit volume
p = Pressure
τ ij = Turbulent stresses determined by a turbulence model

For shallow water flow, the pressure can be assumed to be hydrostatic and all the vertical
components of fluid acceleration can be ignored, thus yielding the quasi-3D governing
equation as (Wang and Wu, 2004):

151
∂u ∂v ∂w
+ + =0 (5.3-3)
∂x ∂y ∂z

∂u ∂ (uu ) ∂ (vu ) ∂ ( wu ) ∂z 1 ∂τ xx 1 ∂τ xy 1 ∂τ xz
+ + + = −g s + + + + fv (5.3-4)
∂t ∂x ∂y ∂z ∂x ρ ∂x ρ ∂y ρ ∂z

∂v ∂ (uv) ∂ (vv) ∂ ( wv) ∂z 1 ∂τ yx 1 ∂τ yy 1 ∂τ yz

+ + + = −g s + + + − fu (5.3-5)
∂t ∂x ∂y ∂z ∂y ρ ∂x ρ ∂y ρ ∂z

where f is the Coriolis coefficient.

The hydrostatic pressure assumption brings significant simplification to the full 3D problem
of equations 5.3-1 and 5.3-2. However, this assumption is valid only for the gradually varying
open-channel flows. A full 3D model without this assumption should be used in regions of
rapidly varying flows.

Sediment transport is divided into suspended load and bed load and hence the flow domain is
divided into a bed-load layer with a thickness δ and the suspended load layer above it with a
thickness of h - δ . The exchange of sediment between the two layers is through deposition
(downward sediment flux) at a rate of Dbk and the entrainment from the bed load layer

(upward flux) at a rate of Ebk . The distribution of the sediment concentration in the upper
layer is governed by the following advection-dispersion equation:

∂ck ∂[(u j − wsk δ j 3 )ck ] ∂ ⎛ vt ∂ck ⎞

+ = ⎜⎜ ⎟⎟ (5.3-6)
∂t ∂x j ∂x j ⎝ σ c ∂x j ⎠

where ck is the local concentration of the k-th size class of suspended sediment load; δ j 3 is

the Kronecker delta with j=3 indicating the vertical direction. At the free surface, the vertical
sediment flux is zero and hence the condition applied is:

vt ∂ck
+ wsk ck = 0 (5.3-7)
σ c ∂z

At the lower boundary of the suspended sediment layer, the deposition rate is Dbk = wsk cbk ,

while the entrainment rate, Ebk is:

152
 vt ∂ck
Ebk = − = wsk cb*k (5.3-8)
σ c ∂z

where cb*k is the equilibrium concentration at the reference level z = zb + δ , which needs to
be determined using an empirical relation. In the 3D model, the bed load transport is
simulated by using the equation:

∂ (δ b cbk ) ∂ (α bx qbk ) ∂ (α by qbk ) 1

+ + + (qbk − qb*k ) = 0 (k = 1, 2,..., N ) (5.3-9)
∂t ∂x ∂y L

where cbk is the average concentration of bed load at the bed load zone, qbk is the bed load

transport rate of the k-th size class and qb*k is the corresponding bed load transport capacity.

α bx and α by are the direction cosines of the bed shear stresses, known from the flow
calculation.

The bed change can be determined by either the exchange equation:

⎛ ∂z ⎞ 1
(1 − p 'm ) ⎜ b ⎟ = Dbk − Ebk + (qbk − qb*k ) (5.3-10)
⎝ ∂t ⎠ k Ls

or the overall sediment mass-balance equation integrated over the water depth h:

⎛ ∂z ⎞ ∂ (hCtk ) ∂qtkx ∂qtky

(1 − p 'm ) ⎜ b ⎟ = + + =0 (5.3-11)
⎝ ∂t ⎠ k ∂t ∂x ∂y

where Ctk is the depth-averaged sediment concentration and qtkx and qtky are the components

of the total load sediment transport in the x- and y- directions:

h
⎛ v ∂c ⎞
h
⎛ v ∂c ⎞
qtkx = α bx qbk + ∫ ⎜ uck − t k ⎟∂z; qtky = α by qbk + ∫ ⎜ vck − t k ⎟∂z (5.3-12)
δ ⎝ σ c ∂x ⎠ δ ⎝ σ c ∂x ⎠

153
5.3-2 Three-dimensional Model (case study): Three Gorges Reservoir Project, China
(Dou et al., 2004)

This study applies a 3D mathematical model for suspended load motion in turbulent flows,
which is based on the following new methods:

• A new stress model based on the stochastic theory of turbulent flows by Dou
(1980)
• Refined wall function for treatment of solid walls
• Introduction of traditional equations of suspended load motion and sorting of bed
material into a 3D model
• Orthogonal curvilinear grid system, with horizontal layers employed over the
water depth

The equation system is solved by using the SIMPLE–C algorithmic. The above model is first
validated using hydrological data collected before and after the construction of the existing
Gezhouba Dam. The model is then applied to the sedimentation problem of the much larger
Three Gorges Project (located upstream of the Gezhouba Dam) and predictions are made of the
sediment deposition pattern, the size distribution of the deposits and the flow fields.

a) Background

The dam area of the Three Gorges Project is located within Xiling Gorge on the Yangtze River.
A 16km long river reach was simulated between Miaohe and the dam axis at Sandouping, a
deep valley of heavily weathered anticline diorite and granite, with low mountains and hills on
both sides. The width of the river in this section is usually 600m to 700m during a flooding
period. The maximum width of 1 400m is found near Sandouping. The river reach under
investigation is slightly curved. The inlet is Miaohe with a width of only about 500m. Figure
5.3-1 shows the river regime in the dam area. The riverbed consists of gravel and cobble. The
slope is steep with rapid flows. The water surface gradient during dry and flooding periods
varies between 0.375% and 0.506%. This reach is amongst the worst in terms of navigation
conditions in the Sichuan province. After the impoundment of Gezhouba Reservoir, the present
reach belongs to the perennial backwater area: during dry periods the water level rises 15m to
20m, during flooding periods it rises 2m to 4m, and the gradient becomes 0.016% to 0.297%.

154
The Yangtze River has a high runoff volume, and the sediment discharge is also large. The
average annual runoff is 4.39x1012m3, and the average annual suspended load discharge is 526
million tons.

1 3
2

Figure 5.3-1: River Regime in the Dam Area of Three Gorges Project
(Dou et al., 2004)

b) Generation of Computational Grids

A total of 163 layers of mesh were arranged along the longitudinal direction, and 81 layers
across the river width. At the maximum water depth there were 15 layers in the vertical
direction. Orthogonal curve grids were generated numerically and the total number of grid
points is 163 x 81 x 15.

c) Verification of Deposition in the Three Gorges Dam Site Area Due to Gezhouba
Reservoir

A period of three years was selected for the reservoir sedimentation verification (from June
1981, after the impoundment of the existing Gezhouba Reservoir to December 1984)
Calculation time step was 3 to 4 days on average. Calculation of sediment concentration of
suspended load was conducted for 7 grain size groups, and the average values during 1981 to
1984 were used regarding the percentage of each grain size.

155
Table 5.3-1 is a comparison between the calculated and measured amounts of deposition for the
reaches numbered 1 to 4 on Figure 5.3-1. It can be seen from Table 5.3-1 that the predictions
were quite close to the measured values.

Table 5.3-1: Comparison between the calculated and measured deposition amount for
various reaches (104 m3) (Dou et al., 2004)
Reach 1 2 3 4 Total
Calculated 271.2 231.9 300.4 432.6 1236.1
Measured 310.5 224.6 240.4 418 1193.5
Error (%) -12.7 3.2 25 3.5 3.6

a) Calculation Conditions for Predictions of the Three Gorges Project

The incoming hydrographs of runoff and graded suspended load into the Three Gorges Project
dam area are provided by the Yangtze River Scientific Research Institute by use of a 1D model
calculation.

During the initial operation of the reservoir both the amount of suspended load and their grain
sizes will be small. After 90 year of operation, most suspended load will enter the dam area and
the median diameter will be increased to 0.027mm, close to the average annual median diameter
of natural suspended load, i.e. 0.031mm, For large discharges, the median diameter could be
larger than 0.031mm.

Though the grain size distribution of suspended sediment entering the dam area may vary with
time and discharges, it is assumed that the size distribution is related to d50. In this model study,
the following equation by Zhou (1997) was used. It is based on statistical analysis of data from
Yangtze River Scientific Research Institute:

⎛D ⎞
−0.639⎜ L ⎟
PL = 1 − e ⎝ d50 ⎠
(5.3-13)

where DL is the grain size of sediment group (mm), PL the percentage by weight of the sediment
group in question. The above equation is used for the size distribution of suspended load. A
total of 7 size groups for suspended load were used in the calculation.

156
e) Hydrological Series in the Simulation

The hydrological series adopted for the calculation is a 10-year series, based on observed data
during 1961 to 1970. The average incoming water and sediment loads during the 10 years is
close to the annual averaged amounts, and is considered highly representative. In order to take
into consideration the impact of extra-large flood and sediment years during long-term
simulation, it has been suggested that after 30 years of operation of the reservoir an extra high-
flood-high-sediment year, 1954, be inserted into the hydrological series. In total, a period of 76
years was simulated.

f) Sediment Deposition at the Upper Reach of the Dam Area

Figure 5.3-2 shows deposition development results of the entire 16 km reach compared to that
of the 6.8 km near dam reach.
Deposition Amount (10 m )

1200
3
6

1050
900
Entire 16km
750
reach
600
6.8km Near
450
Dam Reach
300
150
0
0 10 20 30 40 50 60 70 80

Year of Operation (yr)

Figure 5.3-2: Development of the deposition of the entire and near dam reaches
(Dou et al., 2004)

The simulated results indicate that during the initial 32 year period, the regime is mainly
characterised by a large cross-sectional area and small flow velocity. Though sediment
concentration is small and the grain size is fine at the inlet reach of the dam area, it still
maintains a saturated state. The entire river reach appears to have uniform and unidirectional
deposition. The accumulated deposition in the entire river reach and near-dam reach will be 589

157
million m3 and 312 million m3 respectively. The average thickness in the upper reach will be
33.3m and that of the near-dam reach is 24.3m.

After 60 to 70 years of operation, the upper reach of the dam area will reach a state of
equilibrium, i.e. both erosion and deposition will occur in the dam area without unidirectional
deposition. After 76 years , the accumulated amount of deposition in the entire reach will be a
steady 1.208 million m3 and that of the near dam reach will be 714 million m3, with the average
thickness in the upper reach 56.3 m and that in the near-dam reach 56.0m, which are very close
to one another.

The results of the Three Gorges model demonstrate that within 30 years' reservoir operation, the
cumulative sediment deposition will have no serious impact on the project's normal operation
both in the dam area and in the fluctuating backwater region. But several decades later,
sediment deposition in the fluctuating backwater region might affect the navigation and harbour
operation during extremely dry years when the reservoir's water level drops down to the lowest
level.

g) Sediment Concentration Field

The largest difference between 3D and 2D models is that 2D models can only simulate depth-
averaged flow fields and sediment concentration fields, while 3D models can predict flow fields
and sediment concentration fields at different water depths or elevations. The 3D simulated
results however show that the distribution of sediment concentrations over the depth is
relatively uniform. Sediment concentration at the riverbed is slightly larger than that in the
middle and near the surface.

158
5.4 Models of density currents

5.4.1 Introduction

Numerical models used in simulating turbidity currents can be divided into two
groups: Depth-averaged models and 2D vertical mixture Navier-Stokes models. Both
these models have their advantages and disadvantages. In order to practically solve the
Navier-Stokes model, some sort of turbulence model has to be assumed (with the k-ε
turbulence model being the most widely used). However, unlike depth-averaged
models the Navier-Stokes models are able to provide information on the vertical
structure of a turbidity current.

5.4.1.1 2D vertical mixture models

2D vertical mixture models make no assumptions about the vertical profiles of the flow
(temperature, salinity, concentration and velocity) and are therefore quite general. The main
challenges with these models are computational cost (a 2D mesh is required) and the selection
of a proper turbulence model.

A turbulence model is required to describe the turbulent transport of momentum and mass.
However, turbulence model selection is complicated by the fact that most turbulence models
and wall functions are based on shear flows that do not account for buoyancy effects. It is
well known that stable stratification attenuates turbulence, while unstable stratification
enhances turbulence.

Choi and Garcia (2002) modeled a gravity current by using a 2D vertical mixture model,
together with an extended k-epsilon turbulence model. Their results were compared with that
of a depth-averaged mixture model. They concluded that the self-similarity assumption used
in depth-averaged models is justified.

Huang and Imran (2005) used a 2D vertical mixture model, together with an extended k-
epsilon turbulence model to simulate a turbidity current. The model allowed the bed boundary
to dynamically evolve by sediment deposition and entrainment. Their results compared well
with experimental data and they recommended using a turbulent Schmidt number of 1.3 for
the turbulent flux of sediment.

159
Brors and Eidsvik (1992) used a 2D vertical mixture model, together with a Reynolds stress
turbulence model to simulate a turbidity current. They argue that k-epsilon models
underpredict the turbulence level at the current’s maximum velocity. Their Reynolds stress
model gave a higher turbulence level at the current’s velocity maximum.

5.4.1.2 Depth-averaged mixture models

In contrast to 2D vertical mixture models, depth averaged models assume some sort of self-
similar vertical profile (temperature, salinity, concentration and velocity) and is therefore less
general, but computationally less expensive (a 1D mesh is required). The main challenge with
these models are selecting empirical relations for water entrainment and shear velocity to
close the governing equations.

Parker et al. (1986) used a three and a four equation depth-averaged mixture model to
investigate the possibility of self-acceleration of turbidity currents. Self-acceleration occurs
when a turbidity current reaches a critical speed whereon sediment entrainment increases, the
concentration and density difference increases and subsequently the current velocity
increases. This leads to even more sediment entrainment. A self-reinforcing cycle of sediment
entrainment and increasing velocity develops. Their numerical results showed that self-
acceleration is possible.

Winslow (2001) investigated the sensitivity of a turbidity current (depth-averaged mixture

model) to various initial conditions, channel properties, closure relationships and fluid
mixture properties. The results showed that initial conditions (current height, velocity and
sediment concentration) only have a short lived effect on model predictions. However,
channel properties (slope and bed friction) and fluid mixture properties (kinematic viscosity
and sediment grain size) control the long-term evolution of turbidity currents.

5.4.1.3 Proposed density current model

It is important to note that density differences can occur due to temperature, material phases
with different densities or chemical species with different densities. Hence, in order to
properly model gravity currents the mechanism responsible for the difference in density has to
be identified. In the case of turbidity currents, the density difference is created by the
multiphase mixture of sediments and water.

160
Manninen (1996) recommends different models for multiphase flows, depending on the
strength of the coupling between the different phases.

a) For drag dominated flows in which the phases are strongly coupled (their velocities
equalize over short length scales) a homogeneous flow model is recommended.
b) For flows where gravity or centrifugal forces create velocity differences (e.g.
sedimentation of sand in water), but the velocities of the phases also equalize over
short length scales, the mixture model is recommended. Most liquid-particle mixtures
fall within this category.
c) The phases of gas-particle mixtures are generally weakly coupled (their velocities do
not equalize over short length scales) and therefore a full multiphase model is
recommended.

For turbidity currents the mixture model is should be used. It consists of the following
equations:

Density equation:
ρ = cρ s + (1 − c )ρ w

where ρ is the density of the mixture; c is the volume fraction of sediment; ρs is the density of
the sediment and ρw is the density of water.

Mixture continuity equation:

∂ρ ∂ρu i
+ =0
∂t ∂xi
where t is time; xi corresponds to the i’th spatial coordinate axis and ui is the Reynolds
averaged velocity of the mixture.

Mixture momentum equation:

∂ρu i ∂ρu i u j ∂P ∂ ⎛ ∂u i ⎞
+ =− + ⎜μ − ρ u í′u ′j ⎟ + (ρ s − ρ w )cg i
∂t ∂x j ∂xi ∂x j ⎜ ∂x ⎟
⎝ j ⎠
where P is the Reynolds averaged pressure; µ is the dynamic viscosity of the mixture; ρ u í′u ′j

is the Reynolds stress and gi is the gravity acceleration in the i’th direction.

161
Sediment continuity equation:

∂ρc ∂ρc(u j − ws δ j 2 ) ∂ ⎛⎜ μ t ∂c ⎞⎟
+ =
∂t ∂x j ∂x j ⎜⎝ Sct ∂x j ⎟⎠

where ws is the settling velocity of sediment (calculated from an algebraic force equilibrium
equation); δi2 is the kronecker delta (for the vertical direction); µt is the turbulent viscosity and
Sct is the turbulent Schmidt number.

In addition to the above equations a turbulence model is required to model the Reynolds
stresses occurring in the mixture momentum equation.

5.4.2 Case study 1: Laboratory flume and field data, Canada

5.4.2.1 Background

Kassem and Imran (2001) have used a Navier-Stokes model (2D vertical) to simulate
a turbidity current. Their results compare favourably with the laboratory observations
of Lee and Yu (1997), as well as Garcia and Parker (1989).

The governing equations used in their model are the Reynolds-averaged,

incompressible Navier-Stokes equations for fluid and momentum conservation along
with a sediment-conservation equation for each grain size. Furthermore, in order to
close the turbulent stress terms in the Navier-Stokes equations a k-ε turbulence model
was introduced. Although the turbidity current can be viewed as incompressible, its
density varies due to the continual change in mixture composition of the grain sizes.
To properly model this behaviour the conservation form of the governing equations
were used.

An implicit finite volume method was used to discretize the above mentioned
differential equations in space and time. The resulting non-linear, coupled algebraic
equations were then linearized and solved iteratively using a scheme such as Newton-
Raphson.

Kassem and Imran's Navier-Stokes model was used to simulate a turbidity current in a
laboratory flume, as well as a turbidity current in the Saguenay fjord, Canada. The

162
 results compared favourable and indicated that the Navier-Stokes model can be
successfully applied to small scale and field scale problems.

5.4.2.2 Laboratory scale application

The numerical model was used to simulate the turbidity current movement for the
flume experiments of Lee and Yu. The flume dimensions was as follows: 20 m long,
0.2 m wide and 0.6 m deep. The suspended material was kaolin. The computational
grid consisted of 10000 grid points (10 x 1000), with a time-step of 0.1 s.

Figure 5.4-1: A-E shows that the numerical model computes the detail of the
plunging movement. This confirms Lee and Yu's observation that the plunge point
location is time dependent.

Figure 5.4-1: Evolution from open channel flow to underflow, with stable plunge point

Figure 5.4-2 shows that the typical vertical structure of the turbidity current agrees
well with the experimental observations of Lee and Yu.

163
 Figure 5.4-2: Comparison between computed and measured non-dimensional velocity
distributions

5.4.2.3 Field scale application

The numerical model was also used to simulate the movement of a turbidity current along a
40 km section in the Saguenay fjord, Canada. The fjord has an approximately constant width.
Using the data synthesized by Mulder (1998) for the 1663 flood, caused by a massive land-
slide a computational grid with 100 000 grid points (4000 x 25) was selected with a time-step
of 10 to 40 s.

Figure 5.4-3 shows that after 9 hours the turbidity current has traveled nearly 30 km. It shows
how the thickness of the turbidity current increases, due to the entrainment of the ambient
water surrounding the current. It also reveals the formation of an internal hydraulic jump
where the slope changes at 20 km. This is in qualitative agreement with the laboratory flume
observations made by Garcia and Parker (1989).

164
 Figure 5.4-3: Simulated turbidity current in Sauenay fjord

Figure 5.4-4 A, B and C are the vertical structure views corresponding to locations A, B, C in
Figure 5.4-3. Figure 5.4-4 A shows the plunge point, whose location and depth agrees
closely with the semi-empirical formula described by Akiyama and Stefan (1984). Figure
5.4-4 B reveals how the thickness increases due to entrainment of the ambient water. It shows
how this entrainment increases the discharge of the turbidity current and correspondingly
decreases the sediment concentration. Figure 5.4-4 C reveals the backward flow over the
head due to the displacement of the ambient water.

165
Figure 5.4-4: Computed flow and concentration fields for turbidity current in Saguenay
fjord at locations A, B, C

5.4.3 Case study 2: Luzzone Reservoir, Switzerland

De Cesare et al. (2001) have also used a Reynolds-averaged Navier-Stokes model

(3D) to simulate the movement of a turbidity current in the Luzzone Reservoir in the
Swiss Alps.

5.4.3.1 Governing equations

The incompressible Navier-Stokes equations used by De Cesare et al. are:

166
Fluid conservation:
∂ui
=0 (5.4-1)
∂xi

Momentum conservation:
∂ui ∂ (ui u j ) Δρ ∂σ ij
+ = gi + (5.4-2)
∂t ∂x j ρ ∂x j

where

−p ⎛ ∂ui ∂u j ⎞
σ ij = δ ij + ν eff ⎜⎜ + ⎟⎟ (5.4-3)
ρ ⎝ ∂xi ∂xi ⎠

Sediment conservation with allowance for depositional behaviour:

∂cs ∂ (cs ⋅ ui ) ν eff ⎛⎜ ∂ui ∂u j ⎞⎟ ∂ ⎛ ρf ⎞

+ = cs + − vsi ⎜ cs ⎟
∂t ∂xi σ c ⎜⎝ ∂x j ∂xi ⎟⎠ ∂xi ⎜⎝ ρ ⎟⎠

where:
• ui is the velocity vector component
• xi is the spatial vector component
• t is time
• ∆ρ is the density difference between the particle and the fluid
• g is the gravity acceleration constant
• σij is the stress tensor component on plane i in direction j
• p is pressure
• νeff is the effective kinematic viscosity
• cs is the sediment concentration
• vsi is the particle sinking velocity
• ρ is the mixture density
• ρf is the fluid density

Furthermore, a k-ε turbulence model (modified to account for the buoyancy effect)
was used to provide closure for the stress terms appearing in the momentum equation.
A current-bed interaction model was also added to account for sediment deposition
and erosion. This current-bed interaction model is based on the ideas of Parker et al.

167
 (1986,1987). The above mentioned differential equations are then discretized in space
and time using the finite volume method.

5.4.3.2 Model application

a) Field measurements

During the summers of 1995 and 1996 measuring campaigns took place in the
Luzzone reservoir to quantify the variations in discharge, temperature and turbidity. A
1000 yr flood was synthesized from the obtained data and used as input for the
numerical model. The 1000 yr flood had a peak discharge of 137 m3/s and the
maximum sediment concentration (d50= 0.02 mm), based on in situ measurements,
was set at 265 g/l.

b) Numerical model

Figure 5.4-5 shows the computational grid used to represent the topography of the
reservoir. The topography was obtained from bathymetric measurements done in
1994. The computational grid consisted of 36 000 cells (100 along the main reservoir
axis, 20 laterally and 18 vertically), with a typical bottom-center cell representing the
physical dimensions of 25x5x2 m3. Although a current-bed interaction model was
used alongside the Navier-Stokes equations, the computational grid (and hence the
reservoir topography) was not allowed to evolve with time.

Figure 5.4-5: Station locations and computational grid superimposed on reservoir

bottom

168
c) Results

Figure 5.4-5 indicates the locations of six stations where the flow details were
extracted from the numerical model. Figure 5.4-6 gives the time variation of the flow-
velocities for the six stations. The mean river inflow velocity is only 0.09 m/s for the
design flood. The turbidity current however attains a maximum of 2.5 m/s in the
narrow canyon just below the inlet of station s11 and maintains a velocity in excess of
1.5 m/s in the larger part of the reservoir, from stations s31 to s61. After about 40
minutes, the turbidity current head arrives at the dam wall. It is reflected and returns
upstream, interacting with the still downstream moving body of the turbidity current.
This returning current travels upstream over a distance of about two-thirds of the total
reservoir length. It reaches a velocity of 0.3 m/s. The global motion inside the lake
becomes insignificant after approximately four hours, while the sediment inflow
stopped already after 1.5 hours. A sediment laden underwater "muddy lake" is formed,
which will then settle its granular material over several hours or even days.

Figure 5.4-6: Velocity variation for stations s11 to s61

5.4.4 Conclusion

The results of the Navier-Stokes model indicates that the plunge point location and
depth, as well as the vertical structure of a density current event can be successfully
simulated. However, further research is still needed on combining the model with bed-
current interaction models, as well as investigating the behavior of the model in flow
conditions which are not conducive to density current formation.

169
6. CONCLUSIONS AND RECOMMENDATIONS

A great deal of information is available on sediment transport, deposition and erosion

in reservoirs. Mathematical modelling however needs to be undertaken with great
care as sediment transport in reservoirs tends to vary greatly both in time and in space.
As sedimentation problems in reservoirs are very difficult and costly to remedy
afterwards, dam designers need to use a conservative approach in making allowance
for sedimentation problems in dam designs.

Suitable mathematical models developed for reservoir sedimentation, describing the

physical processes involved as accurately as possible, should be used for reliable
simulations. Accurate prediction of the long-term sedimentation yield is however as
important as modeling of sediment deposition patterns in a reservoir. Where
turbulence is the main sediment transporting mechanism, one-dimensional
mathematical models are still often used for long-term sediment deposition
predictions, but for detailed studies two dimensional or quasi-3D models are
recommended for use in future, incorporating a fully hydrodynamic approach (quasi-
steady for long-term simulations), modules for erosion and deposition in cohesive and
non-cohesive sediments, to be able to simulate storage, sluicing or flushing reservoir
operation.

Where density currents form in a reservoir, Navier-Stokes 2D vertical or 3D models

should be used to describe the formation, movement and sediment transport of the
density current.

These models are calibrated on local reservoir field data, especially when dealing with
cohesive sediments.

170
7. REFERENCES

Ackers, P. and White, W.R. (1973). Sediment Transport: New Approach and
Analysis. JHD, Proc. Am. Soc. Civ. Engrs, 99 (HY-11).

Akiyama and Stephan, J. and Stefan, H.G.(1987). Onset of Underflow in Slightly

Diverging Channels. J. Hydr. Engrg., Am. Soc. Civ. Engs, 113 (7), pp. 825-844.

Altinakar, M.S., Graf, W.H. and Hopfinger, E.J. (1990). Weakly Depositing Turbidity
Current on Small Slopes. J. Hydr. Res., 28 (1), NL.

Annandale, G.W. (1984). Deposition of Sediment in Reservoir Basins. (In Afrikaans).

D. Eng. Thesis, University of Pretoria, South Africa.

Ashida, K. and Egashira, S. (1975). Basic Study on Turbidity Currents. Japan Soc. of
Civil Engrg., 237, pp 37-50.

Bagnold, R.A. (1962). Auto-suspension of Transported Sediment: Turbidity

Currents. Proc. Royal Soc. of London, Ser. A., 265 pp. 315-319.

Bagnold, R.A. (1966). An Approach to the Sediment Transport Problem from General
Physics. Geol. Survey professional paper 422-1.

Basson, G.R. (1997). Development of a New Sediment Transport equation based on

applied stream power ICHE, Hong Kong.

Basson, G.R. and Rooseboom, A (1997). Dealing with Reservoir Sedimentation.

South African Water Research Commission Publication.
Basson, G.R., Rooseboom, A., Olesen, K.W. (1996). "Mathematical Modelling of
Reservoir Flushing", Proceedings of the International Conference on Reservoir
Sedimentation, Fort Collins, USA, pp 559-570, 1996
Basson, G.R. and Rooseboom, A. (1997). Dealing with Reservoir Sedimentation.
South African Water Research Commission publication.

Bata, G. and Knezevich, B (1953). Some Observations on Density Currents in the

Laboratory and in the Field. Proc. Minnesota Int. Hydraul. Convtn.

Bell, H.S. (1040). Density Currents as Agents for Transporting Sediments. J. Geol.
L.5.

171
Bell, H.S. (1942). Stratified Flow in Reservoirs and its Use in Prevention of Silting.
Misc. Publication No 491, US Dep of Agriculture.

Blancket (1954). In Delft Hydraulics, 1992.

Brors, B. and Eidsvik, K. (1992). “Dynamic Reynolds modeling of turbidity currents.”

Journal of Geophysical Research.

Bruk, S. (ed) (1985). Methods of Computing Sedimentation in Lakes and Reservoirs.

Unesco, Paris, IHP-11 Project.

Brune, G.M. (1953). Trap efficiency of reservoirs. Trans. Am. Geoph. Union, 34 (3).

Cao, R. (192). Experimental Study on Density Current with Hyper concentration of

Sediment. Int. J Research, 8 (1).

Chikita, K. (1989). A Field Study on Turbidity Currents Initiated from Spring

Runoffs. Wat. Resour. Res., 25 pp. 257-271.

Chollet, J.P. and Cunge J.A. (1980). Simulation of Unsteady Flow in Alluvial
Streams. Appl. Mathematical Modelling, 4.

Choi, S. and Garcia, M. H. (2002). “k-_ turbulence model of density currents

developing two-dimensionally on a slope.” Journal of Hydraulic Engineering.
Choux, C., Baas, J., Mccaffrey, W., and Haughton, P. (2005). “Comparison of spatio-
temporal evolution of experimental particulate gravity flows at two different initial
concentrations, based on velocity, grain size and density data.” Sedimentary Geology.
Churchill, M.A. (1948). Analysis and use of reservoir sedimentation data. Discussion
by L.C. Gottschalk, Proc. Federal Inter-Agency Sedimentation Conference,
Washington DC., Jan. 1948, pp.139-140.

Cunge (1989). In Delft Hydraulics, 1992.

De Cesare G, Schleiss, A, and Hermann, F. (2001). ASCE Journal of Hydraulic

Engineering V127 No. 1; pp. 6-16.

Delft Hydraulics (192). The Control of Reservoir Sedimentation – A Literature

Review. Ministry of Development Co-operation, Government of the Netherlands,
Delft Hydraulics.

172
Delft Hydraulics (1992). The Control of Reservoir Sedimentation- A Literature
review. Ministry of Development Co-operation, Government of the Netherlands, Delft
Hydraulics.

Denton, R.A., Faust, K.M. and Plate, E.J. (1981). Aspects of Stratified Flow in Man-
made Reservoirs. Research Report ET-203, Sonderforschungs-bereich 80, Univ. of
Karlsrühe, Germany.

Dequennois, H. (1956). New Methods of Sediment Control in Reservoirs. Water

Power, pp. 174-180.

DHI. (2003). "MIKE 21C River Hydrodynamics and Morphology", User Guide, DHI
Water & Environment, 2003

Di Silvio, G. and Armanini, A. (1981). Influence of the Upstream Boundary

Conditions of the Erosion-Deposition Processes in Open Channel. XIX Int. Ass. Hyd.
Res. Congress, Paper 22, sub. A(a), New Delhi, India, 1981.

Di Silcion, G. (1995). River modelling. UNESCO IHP-IV Project H-1-2, Working

Group on Erosion, Riverbed Deformation and Sediment Transport in River Basins as
related to National and Manmade changes.

Dou, G. (1974). Similarity Theory and its Application to the Design of Total
Sediment Transport Model. Research Bulletin of Nanjing Hydraulic Research Inst.,
Nanjing, China.

Engelund, F. and Hanse, E. (1967). A Monograph on Sediment Transport in Alluvial

Streams. Teknisk Forlag, Denmark.

Fan, J. (1960). Experimental Studies on Density Currents. Scientia Sinica, 9(2), pp

275-303.

Fan, J. (1986). Turbid Density Currents in Reservoirs. Wat. Int. 11(3), pp. 107 – 116.

Fernandez, R. and Imberger, J. (2006). “Bed roughness induced entrainment in a high

Richardson number underflow.” Journal of Hydraulic Research.

Galapatti, G. and Vreugdenhil, C.B. (1985). A Depth-integrated Model for Suspended

Sediment Transport. J. Hydr. Res., Int. Ass. Hyd. Res., 23 (4), pp. 359 – 377.

Garcia, M and Parker, G. (1989). Experiments on hydraulic jumps in turbidity

currents near a canyon-fan transition: Science V245, pp. 393-396.

173
Garcia, M. and Parker, G. (1993). “Experiments on the entrainment of sediment into
suspension by a dense bottom current.” Journal of Geophysical Research.
Garcia, M. (1993). “Hydraulic jumps in sediment-driven bottom currents.” J. Hydraul.
Res.

Gilbert, G.K. (1914). The Transportation of Debris by Running Water, Based on

Experiments made with the Assistance of E.C. Murphy. U.S. Geol. Survey Prof. Paper
86,263 p.

Graf, W.H. (1971) Hydraulics of Sediment Transport. McGraw-Hill.

Graf, W.H. (1983). The Hydraulics of Reservoir Sedimentation. Int. Water Power
and Dam Construction. 35, (4), pp. 45 – 52, April.

Guan, Y., Rong, F., Wang, J., Yin, L. and Wang, H. (1991). A Numerical Model for
Sedimentation in Fenhe Reservoir and the Adjoining Reaches. Int. Journal of
Sediment Research, 6 (1), IRTCES.

Guy, H.P., Simons, D.B. and Richardson, E.V. (1966). Summary of Alluvial Channel
Data from Flume Experiments, 1956 to 1961. Sediment Transport in alluvial
channels. Geological Survey Professional Paper 462 – 1.

Han, Q., et al. (1973). Non-equilibrium Transportation of Sediment in Reservoirs.

Collection of Reports on Reservoir Sedimentation, Yellow River Conservancy
Commission (in Chinese).

Han, Q and He, M. (1990). A Mathematical Model for Reservoir Sedimentation and
Fluvial Porcesses. Int. J. Sed. Res., IRTCES, 5 (9), April, pp. 43-84.

Harleman, D. (1961). Stratified flow. In: Handbook of Fluid Dynamics, Ed. V.

Streeter, McGraw Hill, New York, USA.

Hinze, J.O. (1960). Turbulence. McGraw-Hill Book Company, Inc., New York, N.Y.

Holly, F.M. and Rahuel, J.L. (190). New Numerical/Physical Framework for Mobile-
bed Modelling.: J. Hydr. Res., Int. Ass. Hyd. Res., 28 (4-5).

Huang, H., Imran, J., and Pirmez, C. (2005) “Numerical modeling of turbidity currents
with a deforming bottom boundary.” Journal of Hydraulic Engineering.

Hurst and Chao (1975). In Delft Hydraulics, 1992.

ICOLD (2000). Dealing with Reservoir Sedimentation, ICOLD Bulletin 115.

174
Kassem, A and Imran, J. (2001). Geology; V29 No. 7, pp. 655-658.

Keunen, P.H. (1952). Estimated Size of the Grand Banks Turbidity Current. An. J.
Sci.

Lee, H and Yu, W. (1997). ASCE Journal of Hydraulic Engineering V123 No. 6, pp.
520-528.

Lee, H. and Yu, W.(1997). “Experimental study of reservoir turbidity current.”

Journal of Hydraulic Engineering.

Levi, I.I. (1958). Injenernaia Ghidrologhia. Moscow.

Lewis, A.D. (1936). Silting of Four Large Reservoirs in South Africa.

Communication No 5, 2nd Congress on Large Dams, Washington.

Mackay, G.A. Webby, M.G., and Walsh, J.M. (2000). Lake Roxburgh Hydraulic
Modelling. Contact Energy Study Brief CLU#5, New Zealand.

Mahmood, K. (1987). Reservoir Sedimentation: Impact, Extent and Mitigation.

World Bank Technical Paper, No. 71, The World Bank, Washington D.C.

Majumdar, S., Rodi, W., Zhu, J. "Three-dimensional finite-volume method for

incompressible flows with complex boundaries", Journal of Fluids Engineering, Vol.
114, pp 496-503, 1992
Manninen, M. and Taivassalo, V.(1996). On the mixture model for multiphase flow.
VTT Publications.

Mehta, A.J. and Partheniades, E.(1973). Effects if Ohysico-chemical Properties of

Fine Suspended Sediment on the Degree of Deposition. Proc. Of the International
Symp. On River Mechanics, Bangkok, 1, Paper A – 41, pp. 465-476.

Michon, X., Goddet, J. and Bonnefille, R. (1955). Etude Thèorique et Experimentale

de Courants de Densite. Tome 1 et 2, Lab. Nat. d’Hydraulique, Chatou, France.

Middleton, G.V. (1966). Experiments on Density and Turbidity Currents. Can. J.

Earth Sci., 3 (4).

Miller, C.R., (1953). Determination of the Unit Weight of Sediment for use in
Sediment Volume Computations. Memorandum, Bureau of Reclamation, U.S. Dep.
of the Interior, Denver, Colorado.

175
Mulder, T, Syvitski, J and Skene, K. (1998). Modelling of erosion and deposition by
turbidity currents generated at river months. Journal of Sedimentary Research V68,
pp. 124-137.

Palmieri, A. (2003). Social and economic aspects of reservoir conservation. World

Water Forum, Kyoto, Japan.
Parker, G, Fukushima, Y and Pantin, H. (1986). Self accelerating turbidity currents.
Journal of Fluid Mechanics V. 171, pp. 145-181.
Parker, G et al. (1987). Experiments on turbidity currents over an erodilde bed.
Journal of Hydraulic Research, Delft, The Netherlands, V25(1), pp. 123-147.

Partheniades, E (1986). The Present State of Knowledge and Needs for Future
Research on Cohesive Sediment Dynamics. In: Wang, S.Y., Shen, H.W. and Ding,
L.Z. (ed) River Sedimentation III, Proc. Third Int. Symp. On River Sedimentation,
Univ. if Mississippi, Mississippi, USA, pp. 3 – 25.

Patankar, S.V. "Numerical Heat Transfer and Fluid Flow", McGraw-Hill, 1980

Philpott, W. (1978). The Plunging of Density Currents. Research Report, Dept. of

Civil Eng., Univ. of Canterbury, Christchurch, New Zealand.

Pitt, J.P. and Thompson, G. (1984). The Impact of Sedimentation on Reservoir Life.
Proc. Symp. On Challenges in African Hydrology and Water Resources, IAHS,
Harare, Zimbabwe, Publ. No 144.

Raynaud (1951). In Delft Hydraulics, 1992.

Rooseboom, A. (1975). Sediment Transport in River and Reservoirs. D.Eng

dissertation, University of Pretoria, (In Afrikaans). (Later published by Water
Research Commission, South Africa, Report No 297/1/92, 1992 English).

Rooseboom, A and Mulke, F.J. (1982). Erosion Initiation. Proc. Symp. On recent
developments in the explanation and prediction of erosion and sediment yield, IAHS
Publication No. 137.

Rooseboom, A, et al. 1986. Welbedacht Reservoir – The Effect of Different

Operating Rules on the Sedimentation Rate and Reservoir Yield with Reference to
Off-channel Storage in the Proposed Knelpoort Reservoir. South African Department
of Water Affairs and Forestry.

176
Rooseboom, A. (1992). Sediment Transport in Rivers and Reservoirs – A South
African perspective. South African Water Research Commission, Report No 297/1/29,
South Africa.

Rooseboom, A and Le Grange, A. (2000). The Hydraulic Resistance of Sand

Streambeds under Steady Flow Conditions. Journ. Hydr. Research Vol. 38, 2000 No
1.

Rozowskii, I.L. "Flow of water in bends of open channels", English translation: Israel
Progr. For Scientific Transl. Jerusalem, 1961

Sanchez, J.C. (1982). Mathematical Model for Simulation of Delta Formation and
Erosion Downstream of a Reservoir. Proc. 14th Int. Congress on Large Dams, ICOLD,
Rio de Janeiro, Brasil, III (Q54, R8), pp. 117-129.

Savage, S.B. and Brimberg, J. (1975). Analysis of Plunging Phenomena in Water

Reservoirs. J. Hydr. Res., Int. Ass. Hyd. Res., 13 (2) pp. 187-204.

Simpson, J. (1997). Gravity currents in the environment and the laboratory.

Cambridge University Press.

Singh, B. and Shah, C.R. (1971). Plunging Phenomenon of Density Currents in

Reservoirs. La Houille Blanceh, 26 (1) pp. 59 –64.

Soares et al. (1982). In Delft Hydraulics, 1992.

Stein, R.A. (1965). Laboratory Studies of Total Load and Apparent Bed-load. J. of
Geoph. Res., 70 (8), pp. 1831 – 1842.

Stone, H.L. " Iterative solution of implicit approximations of multidimensional partial

differential equations", SIAM Journal Num. Analysis, Vol. 5, No 3, pp 530-559, 1968

Sundborg, A. (1964). The Importance of the Sediment Problem in the Technical and
Economic Development of River Basins. Ann. Acac. Reg. Sci. Uppsala, 8, pp. 3 –52.

TAMS. (1998). Tarbela Dam Sediment Management Study. Pakistan Water and
Power Development Authority.

Tan, Y. (1994). Reservoir Design and Management to Control Sediment. Topic VI,
Russia.

177
Thomas, W.A. and Prashuhn, A.L. (1977). Mathematical Modelling of Sediment
Transport, Scour and Deposition in River Channels. 17th Congress Int. Ass. Hyd.
Res., Germany.

Turner, J.S. (1973) Buoyancy Effects in Fluids. Cambridge University Press,

Cambridge, England.

Turner, J.S. (1979). Buoyancy Effects in Fluids. Cambridge Univ. Press, London,
England.

Van Rijn, L.C. (1984a). Sediment Transport, Part I: Bed Load Transport, J. Hyd.
Engrg., 110 (10).

Van Rijn, L.C. (1984b). Sediment Transport, Part II: Suspended Sediment Load
Transport, J. Hydr. Engrg., 110 (11).

Vasiliev, O.F., et al. (193). Mathematical Modelling of Sedimentation in Deept

Reservoir. Proc. Of 1st ICHE, I, Part A, USA.

Velikanov, M.A. (1954). Principle of the Gravitational Theory of the Movement of

Sediments. Acad. Of Sci. Bul., USSR, Geophy Series, No. 4, pp. 349 – 359.

Wang, S.S.Y. (193). Advances in Hydro-Science and Engineering. Proc. Ist ICHE,
USA.

White et al. (1975). Sediment Transport Equation Accuracies. In: Raudkivi, A.J.
Loose Boundary Hydraulics, 1992, Publisher R Maxwell.

Winslow, K.(2001). PhD Dissertation: Sediment transport by turbidity currents. UMI

Dissertation Services.

Wu, D. (1994a). Personal Communications, IRTCES, China.

Wu, D (ed.) (1994b). Movement of Density Current in Reservoir. International

Research and Training Center on Erosion and Sedimentation.

Yang, C.T. (1972). Unit Stream Power and Sediment Transport. Proc. Am. Soc. Civ.
Engrs., 98 (10).

Yang, C.T. (1973). Incipient Motion and Sediment Transport. Proc. Am. Soc. Engrs.,
99 (HY10).

Yang, C.T. (1979). Unit Stream Power Equations for Total Load. J. Hydr., 40, pp.
123 – 138.

178
Yang, C.T. and Molinas, A. (1982). Sediment Transport and Unit Stream Power
Function. J. Hydr. Div. Am. Soc. Civ. Engrs., 108 (HY6), June, pp. 774 – 793.

Yang, C.T. (1984). Unit Stream Power Equation for Gravel. J. Hydraul. Div. Am.
Soc. Civ. Engrs., 110 (HY12).

Yang, C.T. and Kong, X. (1991), Energy Dissipation Rate and Sediment Transport.
J. Hydraul. Res. , 29 (4).

Yang, G. (1992). Mathematical Modelling of Alluvial Rivers. (in Chinese)

Yang, C.T., and Simoes. (2000)

Yang, C.T., and Simoes. (2003). GSTARS3 Reference Manual.

Zhang, Q and Long, Y. (1980). Sediment Problems of Sanmenxia Reservoir, Proc.

Intern. Symp. On River Sedimentation, 2 China.

Zhang, Q., et al (1983). A Mathematical Model for the Prediction of the

Sedimentation Process in Rivers. Proc. 2nd ISRS, China.

Zhang, R (1959). A Study of the Sediment Transport Capacity of the Middle and
lower Yangtze River, J, Sediment Research, 4 (2), Beijing, China (In Chinese).

Zhou, Z (1995). Preservation of Reservoir Storage Capacity – Experience of China.

Proc. 195 Int. Workshop on Reservoir Sedimentation, San Francisco, California.

179