Available online at www.sciencedirect.

com

ScienceDirect
Journal of Hydro-environment Research 8 (2014) 95e114
www.elsevier.com/locate/jher

Research paper

A physical, movable-bed model for non-uniform sediment transport, fluvial

erosion and bank failure in rivers
Kamal El Kadi Abderrezzak a,b,*, Andres Die Moran a,b, Erik Mosselman c,d,
Jean-Pierre Bouchard b, Helmut Habersack e, Denis Aelbrecht f
a
Université Paris-Est, Saint-Venant Laboratory for Hydraulics, ENPC, EDF R&D, CETMEF, 6 quai Watier, 78401 Chatou, France
b
EDF-R&D, National Laboratory for Hydraulics and Environment, 6 quai Watier, 78401 Chatou, France
c
Deltares, Rotterdamseweg 185, 2629 HD Delft, The Netherlands
d
Delft University of Technology, P.O. Box 5, 2600 AA Delft, The Netherlands
e
University of Natural Resources and Life Sciences, Christian Doppler Laboratory for Advanced Methods in River Monitoring, Modelling and Engineering,
Institute of Water Management, Hydrology and Hydraulic Engineering, Muthgasse 107, 1190 Vienna, Austria
f
EDF, Centre d’Ingénierie Hydraulique, Savoie Technolac, 73373 Le Bourget du Lac, France
Received 20 December 2012; revised 17 September 2013; accepted 18 September 2013

Abstract

Sediment transport processes in rivers continue to pose a challenge when designing movable-bed physical models, particularly for reproducing
the grain sorting and bank erosion (fluvial erosion and mass failure). This paper presents and discusses scale effects of a specific scaling approach for
multi-grain size mixtures that preserves similarity of initial motion for each grain size class and of the bank stability coefficient between the model
and the prototype, but relaxes strict similarity of the Shields and particle Reynolds numbers. This approach is appropriate when bed load transport
near incipient motion conditions is being studied, and allows for larger grain size scales than when full Shields parameter similarity is enforced. As
part of an environmental project to rehabilitate sediment transport through bank erosion, this method has been applied to scale a Froude number
criterion physical model of a reach of the Old Rhine (France). This has resulted in an undistorted scale of 40, and the use of sand as the model bank
material. Each grain size has a different geometrical scale. The time scale for sediment motion is grain size and flow discharge dependent. An
average time scale of 6 has therefore been used (four model hours ¼ one prototype day). A strategy devised for the field case consists of two higher,
larger island groynes that replace the three existing groynes, producing bank erosion for flow rates below the mean annual flow rate. Extrapolation of
model behaviour to the prototype is not a major problem, but the volume of eroded bank material may be underestimated, mainly because of the
relaxation of the Shields number similarity and the apparent cohesive properties of the model bank material.
Ó 2013 International Association for Hydro-environment Engineering and Research, Asia Pacific Division. Published by Elsevier B.V. All rights
reserved.

Keywords: Bank failure; Old Rhine; Physical model; Sediment transport; Scale effects; Similarity

1. Introduction

Several approaches can be used for studying fluvial

* Corresponding author. EDF-R&D, National Laboratory for Hydraulics and
Environment, 6 quai Watier, 78401 Chatou, France. Tel.: þ33 130877911.
E-mail addresses: kamal.el-kadi-abderrezzak@edf.fr, elkadi_kamal@
yahoo.fr (K. El Kadi Abderrezzak), adiemoran@hotmail.com (A. Die
Moran), erik.mosselman@deltares.nl (E. Mosselman), helmut.habersack@
boku.ac.at (H. Habersack), denis.aelbrecht@edf.fr (D. Aelbrecht).
morphology, including fieldwork; theoretical, conceptual and
numerical modelling; or experimental and physical models,
and each approach has its own advantages and limitations.
Physical and numerical models are often combined to provide
solutions to practical problems. Numerical models are widely
used in engineering applications, especially for large-scale

1570-6443/$ - see front matter Ó 2013 International Association for Hydro-environment Engineering and Research, Asia Pacific Division. Published by Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.jher.2013.09.004
96 K. El Kadi Abderrezzak et al. / Journal of Hydro-environment Research 8 (2014) 95e114
problems where scale effects in physical models may become
significant. However, they currently still include empirical
formulations for flow (e.g. turbulence closure equations) and
sediment transport (e.g. sediment sorting, flow-
esedimentestructure interaction) that cannot represent the
details of these processes. For instance, multi-layered bank
failure, bed forms and transport of widely graded sediments
over partially armoured layers are still complex for accurate
description in existing numerical models. The reliability of the
numerical results may therefore be questioned.
Physical modelling can be used as an alternative visual
display tool for investigating complex systems. For-
meprocesseresponse interactions can be replicated intrinsi-
cally, without simplifying the assumptions that have to be
made for numerical models. This is especially true when
studying flow and sediment transport with local three-
dimensional (3-D) features in complex geometries near
structures (e.g. dams, weirs, groynes). The basic aim is to
ensure that the relative magnitudes of all dominant processes
and their associated sediment response are the same in the
model and prototype (Gill and Pugh, 2009). The fluvial pro-
cesses can then be observed in a reduced time-frame, within a
controlled and manageable laboratory environment, while
gathering a comprehensive set of quantitative data that can be
used to calibrate numerical models, thus allowing further,
long-term simulations to be carried out numerically (Peakall
et al., 1996).
1.1. Dimensionless parameters
Similarity relationships for scaling hydraulic phenomena
are well established and thoroughly tested against prototype-
scale data (Heller, 2011). Similarity laws for sediment
movement are far from unanimous. Various approaches based
on the governing equations of flow and sediment dynamics,
dimensionless analysis and semi-empirical concepts have been
published on the topic (e.g. Yalin, 1971; Franco, 1978; Shen,
1991; ASCE, 2000; Ettema et al., 2000; Jain, 2001; Julien,
2002; Gill and Pugh, 2009). A basic dimensional analysis
yields seven independent dimensionless parameters or group-
ings for the flow and sediment transport in the following form
PA (Yalin, 1971; Julien, 2002):
0 1
B C
B V uh ru2 u d u h rs  rC
B  C the prototype sediment diameter according to the vertical
PA ¼ fA B pffiffiffiffiffi ; ; ; ; ; ; C ð1Þ
B gh |{z} y gðrs  rÞd |{z} y ws d r C
@ |ffl{zffl} |fflfflfflfflfflfflffl{zfflfflfflfflfflfflffl} A non-cohesive sand in the model.
F Re q Re
where the dependent variable A in PA might be energy
gradient (i.e. flow resistance), sediment transport, or some
other variable in the channel. Variables and dimensionless
parameters in Eq. (1) are defined in Table 1. In theory all
dimensionless parameters should be matched between the
prototype and model for exact quantitative similarity. How-
ever, this is difficult to obtain without distorting the model and
Table 1
Model bank material scaling according to the geometric length scale Zr, with
qr ¼ 1.
Variables/Parameters Definition
V, h Velocity and flow depth, respectively
g Gravitational acceleration
n Kinematic viscosity of water
r, rs Water and sediment density, respectively
d Sediment diameter
ws Settling velocity
u* ¼ (t/r)1/2 Bed shear velocity with t as the bed shear stress
F ¼ V/( gh)0.5 Froude number
Re ¼ u*h/n Flow Reynolds number
q ¼ ru2 =½gðrs  rÞd Shields number
Re* ¼ u*d/n Particle Reynolds number
u*/ws Relative particle fall velocity
h/d Relative roughness in the absence of bed forms
(rs  r)/r Submerged specific gravity
changing the sediment density rs (ASCE, 2000), so similarity
of some parameters must be relaxed.
1.2. Similarity laws for water flow
Most physical models of rivers are designed to ensure
Froude number F similarity between prototype (subscript p)
and model (subscript m), so that their ratio (subscript r) is
equal to unity, i.e. Fr ¼ Fp/Fm ¼ 1. The flow Reynolds number
Re is relaxed, assuming fully turbulent conditions in both
prototype and model. The similarity of flow resistance is
ensured by matching the relative bed roughness d/h between
the prototype and model.
The Froude number similarity is necessary for local-scale
models for which the velocity field should be replicated
accurately in the model, but can be relaxed for large-scale
cases with small to moderate Froude number (F < 0.8 in
both prototype and model) in order to replicate sediment
transport dynamics accurately (Song and Yang, 1979; De
Vries, 1993; Maynord, 2006; Ho et al., 2010). Similarity of
the relative bed roughness ratio d/h is rarely achieved for sand-
bed rivers, because they are usually associated with low d/h
values (ASCE, 2000; Maynord, 2006). However, Ettema et al.
(2000) have shown a significant scale effect of d/h on bridge
pier scour. Similarity of d/h becomes important in gravel-bed
rivers without bed forms (Young and Warburton, 1996). This
criterion is generally satisfied in undistorted models by scaling
geometrical scale of the model (dr ¼ hr), along with using
1.3. Similarity laws for bed load transport
Scaling sediment transport is based on ensuring similarity
between prototype and model for several key parameters, such
as Shields number q, particle Reynolds number Re* and
relative particle fall velocity ws/u*. Simultaneous similarity
may not be possible for all sediment transport parameters, and
therefore care must be taken to determine when deviating from
 K. El Kadi Abderrezzak et al. / Journal of Hydro-environment Research 8 (2014) 95e114 97

similarity is acceptable. For instance, the particle Reynolds The degree of sophistication for movable-bed scale
number Re* can be relaxed when flow is hydrodynamically modelling depends on the particular objectives of the study, on
rough in both prototype and model, whereas similarity of ws/u* the experience and skill of the modeller who designs the
is generally ignored when the suspended load is not model and interprets the data, and on the availability of the
considered. space within the laboratory. Table 2 lists selected studies of
bed load transport, highlighting the similarity criteria which
Table 2
Summary of selected studies employing movable-bed physical models. Studies are limited to the bed load process.
Reference Scaling criteria Model characteristics Phenomena investigated Model-
prototype
Satisfied Relaxed/ignored
comparison
Song and Yang (1979) V  S/ws F, d/h, q, qc, Re* Distorted. Uniform sand. Froude Maintaining of navigation conditions at a Yes
number exaggerated by a factor of river confluence
1.9
Parent (1988) F, q, qc d/h, Re* Undistorted. Non-uniform sand Morphology and sedimentology of pool- Yes
mixture (truncated to avoid ripple riffle sequences
formation)
Young and Warburton (1996) F, q, d/h qc, Re* Undistorted. Natural material. Flow Morphology and sedimentology in Yes
is rough turbulent in the model braided gravel-bed rivers
Healey (1997) F, q d/h, qc, Re* Undistorted. Mix of sand and gravel. Evaluation of methods for the mitigation Yes
Bank failure not reproduced of embankment (bank) erosion
Davinroy et al. (1999), F, d/h, q, qc, Re* Distorted. Lightweight particle Design of channel-control alternatives Yes
Gaines and Maynord (2001), (plastic) (dikes, bendway weirs, bank line
Rodgers et al. (2003), changes)
Maynord (2006)
Wallerstein et al. (2001) F d/h, q, qc, Re* Distorted. Uniform sand (one grain Geomorphic and hydraulic impact of No
size) large woody debris
Wei et al. (2001) F, q/qc, B  S/h d/h, q, qc, Re* Distorted and undistorted models are Evaluation of bed load and bar formation No
tested. Uniform sand and lightweight following training works
particles (Lapili) are tested
Woidt et al. (2001) F, q/qc d/h, Re* Undistorted. Cohesive loam soil for Sedimentation at a pump intake No
the bank. Fine silica sand for the bed
Waldron (2005) F, q, qc, Re* d/h Distorted. Lightweight particles Efficiency of sediment diversions for No
(synthetic plastics), uniform size rehabilitating degraded wetlands
Marr et al. (2007) F, q d/h, Re*, qc Distorted. Mix of coarse and fine Rate and timing of remobilisation of No
sand stored sediments following dam removal
Bennett et al. (2008) F d/h, q, qc, Re* Distorted. Sand, uniform particle size Use of in-stream woody vegetation for No
restoring meandering pattern
Bromley (2008) F, q Re*, qc Distorted. Mix of sand and gravel. Downstream morphology changes due to Yes
Supercritical flow dam removal
Mefford et al. (2008) F, (q  qc) d/h, q, qc, Re* Undistorted. Mix of sand and gravel. Performance of a high-flow bypass No
Silt and clay material are not spillway in improving bed load transport
represented in the model at a diversion dam structure
Pugh (2008) F, qs/(u*d ), ws d/h, Re*, q, qc Distorted. Uniform sand Design of channel-control alternatives No
for limiting sediment intake at a planned
diversion dam
Weitbrecht and Rüther (2009) F, qc d/h, q, Re* Undistorted. Mix of sand and gravel. Performance of a planned drift wood No
Finer fractions of the model retention concept in an expanding river
sediments are coarsened, fractions of reach
grain sizes <0.2 mm are eliminated
Armanini et al. (2010) F, q, qc d/h, Re* Undistorted. Lightweight particles Design of groynes to improve navigation No
(plastics), uniform particle size. Finer condition
fractions are not reproduced
Ho et al. (2010) q, qc, Re* d/h, F Distorted. Lightweight particles Sediment exclusion at an intake structure No
(crushed coal). Uniform particle size.
Subcritical flow (F < 0.5)
Mefford and Gill (2010) F, qs/(u*d ), ws d/h, q, qc, Re* Distorted. Lightweight particles Evaluation of different restoration works No
(coal), uniform size for creating shallow water habitat along
a stream bend diversion
Simonett and Weitbrecht (2011) F, q/qc d/h, q, qc, Re* Undistorted. Transcritical flow. Mix Design and optimisation of training No
of sand and gravel. Finer fractions works for flood defence
are coarsened, fractions of grain sizes
<0.2 mm are eliminated
Bieri et al. (2012) F d/h, q, qc, Re* Undistorted. Non-uniform sand Design and optimisation process for Yes
mixture sediment flushing operation
98 K. El Kadi Abderrezzak et al. / Journal of Hydro-environment Research 8 (2014) 95e114
are satisfied, relaxed or ignored, the characteristics of the
physical model, the phenomena examined, and whether or not
model-prototype comparison exists.
The most relevant requirement for physical model studies
of bed load is to preserve the similarity of the sediment
mobility expressed in terms of Shields number q, i.e. qr ¼ qp/
qm ¼ 1. Preserving only q is suitable whether the critical
Shields number qc is constant within the different stages of
flow and sediment transport. This may be not the case under
certain conditions, such as non-uniform sediment transport,
variable relative bed roughness ratio and flow regime (laminar
and rough turbulent) (Buffington and Montgomery, 1997).
Einstein and Chien (1956) recommended preserving the
similarity of both q and sediment transport intensity F ¼ qs/
[d1.5( gD)1/2] (qs is the volumetric bed load transport per unit
width and D ¼ (rs  r)/r is the sediment relative mass den-
sity). Pugh (2008) recommended the similarity of the dimen-
sionless unit sediment discharge qs* ¼ qs/(u*d ). However, for
both the Einstein and Chien (1956) and Pugh (2008) ap-
proaches the appropriate sediment transport formula for the
case being modelled must be known.
Kishi et al. (1975) suggested B  S/h and u*/u*c ¼ (q/qc)1/2
similarity criteria (u*c ¼ (tc/r)1/2 is the critical shear velocity
with tc the critical shear stress for incipient sediment motion, B
the channel width and S the overall bed slope). Wei et al. (2001)
showed Kishi et al.’s (1975) approach is suitable whenever the
formation of bars is the dominant process to investigate in the
scale model. According to Song and Yang (1979), the similarity
of the dimensionless unit stream power criterion Pw ¼ V  S/ws
should be respected for studying bar formation.
The methodology applied at the National Laboratory for
Hydraulics and Environment of EDF R&D (Chauvin, 1962)
maintains the similarity of q but also considers the type of
expected bed forms (plane bed, ripples or dunes) when
selecting the geometric scale. Assuming a unique relationship
between the sediment transport intensity F and q, movable-
bed models have been designed at WLjDelftHydraulics
(currently Deltares) by maintaining similarity of q and flow
resistance and relaxing the Froude number criterion
(Struiksma and Klaassen, 1986).
Pugh and Dodge (1991) recommended the difference
(q  qc) should be the same in the model and prototype,
instead of q. This similarity law has been widely used by the
Water Resources Research Laboratory of the U.S. Bureau of
Reclamation (Mefford, 2005; Mefford et al., 2008). More
recently, the United States Army Corps of Engineers (USACE)
proposed a paradigm justifying the use of extremely small-
scale physical models to address river-training issues
(Gaines and Maynord, 2001). However, these models ignore
both Froude and Shields number similarity, and their results
deviate much from the reality in the prototype.
1.4. Limitations of the scaling approaches
Applying these principles should not mask the enormous
complexity that results from scale effects induced by the
incomplete fulfilment of a full set of similarity criteria
associated with the problem under consideration. In fact, not
all dimensionless parameters incorporating grain size can be
modelled correctly at the same time (Julien, 2002). Scaling
effects are reasonably understood for fixed-bed models
(Heller, 2011). On the contrary, they are not so well under-
stood for movable-bed physical model. Examples include
scales effects due to the characteristics of the particles used to
reproduce the prototype bed material, the distortion of
roughness due to eventual bed forms, the presence of sus-
pension while the dominant transport mode in the prototype is
bed load, and the incorrect estimate of the sediment time scale
(Jaeggi, 1986; Ettema et al., 2000).
Most of the above methods have been developed for scaling
physical models under the implicit assumption that the pro-
totype bed material is uniform. When this assumption is not
justified, the modelling becomes even more complicated, since
questions arise concerning the reproduction of the sediment
mixture size distribution in the physical model. Moreover,
these methods are not suitable when investigating bank erosion
processes through fluvial erosion and mass failure. Compared
with fluvial erosion, mass failure mechanisms are discontin-
uous (Rinaldi and Darby, 2007). Bank stability analysis is
therefore essential and the factor of stability, defined as the
ratio between resisting and driving forces, should be main-
tained between prototype and model.
1.5. Objectives of the present work
The purpose of this research is to present and discuss a
specific approach for scaling physical models where non-
uniform sediment transport and bank erosion (i.e. fluvial
erosion and mass failure) are considered to be the dominant
processes. This study is included in an environmental project
aiming at feeding sediments in the Old Rhine (France)
downstream of the Kembs dam by initiating bank erosion. A
pilot site has been selected to guide the rehabilitation work,
and a movable-bed physical model is used as the primary
research tool to find the best way of promoting bank erosion.
The remainder of the paper is organised as follows. Section
2 presents the case study, including a brief background to the
project and the selected pilot site. In Section 3, the scaling
theory is detailed. The model construction and data acquisition
are presented in Section 4. The best strategy to promote bank
erosion is presented in Section 5. The discussion in Section 6
is followed by conclusions in Section 7.
2. Case study: the Old Rhine
2.1. Background
The 50 km reach of the River Rhine between the towns of
Village-Neuf (France) and Breisach (Germany) is divided into
two watercourses: the Grand Canal d’Alsace (GCA) which is a
concrete-bed navigation canal, and the Old Rhine which is a
stable, cobble-bed river channel with protected banks (Fig. 1).
This configuration is the result of engineering works carried out
from the 19th century (e.g. channelisation, in-stream structures)
 K. El Kadi Abderrezzak et al. / Journal of Hydro-environment Research 8 (2014) 95e114
Fig. 1. The Rhine river basin and detail of the Old Rhine area.
up to the late 20th century (GCA, chain of hydropower plants).
The riverbed has undergone strong incision, leading to the for-
mation of an armour layer. The sediment trapping effect of the
numerous dams upstream in Switzerland has resulted in almost
non-existent sediment feed into the Old Rhine. The flow regu-
lation operated by the Kembs dam in order to divert the main
flow of the upper Rhine into the GCA allows only an ecologi-
cally defined minimum flow into the Old Rhine.
2.2. Sediment transport rehabilitation
To recover the dynamics of the Old Rhine and its associated
biodiversity, a project aimed at rehabilitating sediment trans-
port by allowing bank erosion has been initiated (Piégay et al.,
2010). This technique consists of removing embankment
revetment, allowing bank erosion and releasing sediment into
the riverbed. The proposed programme has some inherent
difficulties, however, mostly related to lack of knowledge. In
fact, the crucial challenge is to determine which intervention
to implement, how to manipulate the channel form with
contemporary magnitudes and rates of fluvial processes, while
still yielding morphology processes that promote ecological
functioning. To achieve the objectives of the project, several
reaches of the Old Rhine were selected to assess their suit-
ability for controllable bank erosion. A specific site is studied
in this research paper.
2.3. Pilot site
The selected pilot site, 930 m long, is located between
Rhine-km 191 and 192, on the left (French) bank of the Old
Rhine (Fig. 2). The site includes several groynes, three of
which (referred to as G1eG3) have been taken into account in
the present research. The bed is strongly armoured and made
99
up of uniform gravel with a median diameter d50 ¼ 150 mm.
The homogeneous, non-layered bank material consists of
gravel and non-cohesive sand (Fig. 3), with d50 ¼ 10 mm and
geometric standard deviation s ¼ (d84/d16)0.5 ¼ 8. The Kembs
dam regulates flow, with discharges ranging from 52 m3/s in
winter to 154 m3/s in summer if the flow rate at Basle is below
1400 m3/s. During flood events, the dam diverts a much
greater proportion of the flow into the Old Rhine in order to
maintain the navigability of the GCA. The 1-, 10- and 100-
year return period floods in the Old Rhine are 1350, 2200,
and 3150 m3/s, respectively.
In this stretch, we investigate how the groynes, originally
designed for bank protection and navigation, could be modified
in such a way that the bank is eroded during high-flow events.
3. Design of the physical model
In a model of a complex natural stream, similarity of
sediment transport cannot be achieved at every point in the
model. Therefore, sediment transport is scaled based on sim-
ilarity of average hydraulic conditions.
Here, the scale of interest is related to the grain size dis-
tribution of the bank material. Both fluvial entrainment and
mass failure are considered. The classical scaling approach
based on similarity of the Shields and particle Reynolds
numbers is first presented. This approach is shown to be
inadequate for the objective of the study. A new approach is
thereafter presented.
3.1. Flow dynamics scaling
Dimensional analysis indicates that the dominant forces in
a free-surface flow are inertial, gravitational and frictional.
Assuming the channel is wide (hydraulic radius Rh
100 K. El Kadi Abderrezzak et al. / Journal of Hydro-environment Research 8 (2014) 95e114

Fig. 2. Aerial view of the pilot site.

approximately equal to h), conversion from model to prototype Tr ¼ Xr Vr1 ¼ Xr Zr1=2 ð4Þ
quantities must satisfy the Froude number similarity, i.e.
Qr ¼ Vr Yr Zr ¼ Xr Zr3=2 ð5Þ
Fp Vr
Fr ¼ ¼ pffiffiffiffiffiffiffiffi ¼ Vr Zr1=2 ¼ 1 ð2Þ
Fm gr hr Rer ¼ Vr Zr n1
r ¼ Zr ð6Þ
3=2

where Vr ¼ Vp/Vm is the velocity scale ratio, gr ¼ 1 is the Sr ¼ Xr1 Zr ð7Þ

gravitational-constant ratio, and hr ¼ hp/hm is the flow depth
scale ratio. According to the 1-D momentum equation, the
vertical geometric scale ratio Zr is equal to hr. Defining the
longitudinal and transverse geometric scale ratios by Xr and Yr,
respectively, and setting Xr ¼ Yr, Eq. (2) yields the following
scale ratios for flow velocity V, time T, discharge Q, Reynolds
number Re and longitudinal bed slope S (also called
distortion)
Vr ¼ Zr1=2
where nr ¼ kinematic viscosity ratio. Here, nr ¼ 1 because
water is used as the model fluid. Reynolds number similarity,
Rer ¼ 1, must be relaxed, but similarity of the flow conditions
can be maintained if Re is sufficiently high to establish fully
turbulent conditions. This state is invariably present in the
prototype. In the model, fully turbulent conditions occur if the
Reynolds number Rem is larger than 1400 (Allen, 1947) or
2000 (Gill and Pugh, 2009).
ð3Þ The mean boundary shear stress can be written as t ¼ rghS.
The shear stress ratio reads as
tp
tr ¼ ¼ rr gr Zr Sr ¼ Zr Sr ð8Þ
tm
Considering Darcy’s formula for flow resistance yields

tr ¼ rr fr Vr2 ¼ fr Vr2 ð9Þ

where f ¼ friction factor given by the ColebrookeWhite

equation (Graf and Altinakar, 1993)
 
1 ks 0:85
pffiffiffi ¼ 2 log þ pffiffiffi ð10Þ
f 12Rh Re f

with ks as the surface roughness, assumed equal to d90 in the

present study. For dimensionless homogeneity the Darcy
Fig. 3. Bank material grain size distribution in the prototype and in the model. friction factor is adopted in lieu of Manning’s roughness
 K. El Kadi Abderrezzak et al. / Journal of Hydro-environment Research 8 (2014) 95e114
coefficient. Equation (10) is valid for laminar and turbulent
flows in rivers without bed forms. Assuming fully turbulent
flow conditions, f can be approximated as (Graf and Altinakar,
1993)
sffiffiffi  1=6
8 26 h
¼ pffiffiffi ð11Þ
f g d90
The ratio for the friction coefficient takes the form
 1=3
d90r
fr ¼
Zr
Substituting Eq. (12) into Eq. (9), equating Eqs. (8) and (9)
and using Eq. (3) yields the grain size ratio
d90r ¼ Zr S3r
3.2. Sediment and geometric scaling
3.2.1. Similarity laws for bed load transport
The grain size distribution of the prototype bank material is
approximated using a mixture of four grain sizes in a certain
proportion: 20% (per mass) with dp ¼ d90p ¼ 63 mm (d90 is
grain size for which 90% of sediment is finer by weight), 20%
with dp ¼ 31 mm, 20% with dp ¼ 15 mm, and the remaining
40% with dp ¼ 1.5 mm. This last fraction is likely to be
transported as suspended load. It is therefore considered
irrelevant to the scaling process, since the focus is on the
correct scaling of the bed load. In order to achieve the same
bed state in the model as in the prototype, the Shields and
particle Reynolds numbers should be preserved, i.e.
qp Sr Zr
qr ¼ ¼ ¼1
qm ðrs  rÞr dr
Rep
Rer ¼ ¼ ur dr ¼ tr1=2 dr ¼ Sr1=2 Zr1=2 dr ¼ 1
Rem
The critical Shields number qc should be the same in the
model and prototype, i.e. qcr ¼ 1. According to the Shields
diagram, this can be fulfilled either by imposing Re*r ¼ 1 or
by ensuring Re* is high enough for the flow to remain rough
turbulent in the model and prototype, that is at the point at
which the Shields entrainment function ceases to vary with
grain size Reynolds number.
3.2.2. Similarity law for bank failure
Banks may fail by various mechanisms, which may be of
planar, rotational, cantilever, piping or sapping type. The
occurrence of one or other mechanism depends on several
factors, but the most important are the configuration of ma-
terials in the bank (whether there is one single type of material
or several layers), soil characteristics (cohesive or non-
cohesive), bank shape, and groundwater pressure (Rinaldi
and Darby, 2007). The stability analysis of banks can be car-
ried out by computing the ratio of resisting forces FR to
driving forces FD acting on a unit width of the failure block.
101
Forces FR and FD are expressed as (Amiri-Tokaldany et al.,
2008; Cancienne et al., 2008) (Fig. 4)
FR ¼ cL þ ½W cos a þ P cos ða1  aÞ  Uw  Htw sin atan f
þ Spw tan fb
ð16Þ
FD ¼ W sin a  P sin ða1  aÞ þ Htw cos a ð17Þ
where c ¼ effective cohesion acting along the surface of
ð12Þ failure plane, L ¼ length of the failure plane, a ¼ failure plane
angle (degrees from horizontal), a1 ¼ bank angle (degrees
from horizontal), f ¼ angle of internal friction, fb ¼ angle
expressing the rate of increase in strength relative to the matric
ð13Þ suction, W ¼ weight of a unit width of the failure block,
P ¼ hydrostatic-confining force due to external water level,
Uw ¼ uplift force due to positive pore-water pressure acting on
a unit width of the failure block, Htw ¼ hydrostatic force
exerted by water present in the tension crack on a unit width of
the failure block, Spw ¼ force produced by matric suction on
the unsaturated part of the failure surface (i.e. negative pore-
water pressure). The bank stability coefficient Fs reads
FR
Fs ¼ ð18Þ
FD
Here the bank material is non-cohesive, and only shallow
failures with slip surfaces almost parallel to the bank surface
may occur (Thorne and Abt, 1993). Given that the model
being designed is already subject to complexities due to non-
uniform sediment transport it is prudent to avoid introducing
another source of complexity that would further complicate
the interpretation of the results at the prototype scale. Hence,
ð14Þ saturated conditions are assumed and the tension crack, pore
pressure, matric suction and vegetation effects are ignored.
The primary force which tends to promote bank failure is
ð15Þ therefore the weight of the block. The bank stability coeffi-
cient Fs becomes
W cos a tan f tan f
Fs ¼ ¼ ð19Þ
W sin a tan a
Fig. 4. Sketch of the bank profile (Amiri-Tokaldany et al., 2008). The bank is
deformed by combinations of bed degradation and lateral toe erosion.
GWFS ¼ Groundwater free surface.
102 K. El Kadi Abderrezzak et al. / Journal of Hydro-environment Research 8 (2014) 95e114

Preserving the similarity of bank mass failure implies Fs is Table 3

the same in the model and prototype, i.e. Model bank material scaling according to the geometric length scale Zr, with
qr ¼ 1. Prototype grain sizes are given in bold.
ðtan fÞr ðtan fÞr Prototype Model
Fsr ¼ ¼ Xr Zr1 ðtan fÞr ¼ ¼1 ð20Þ
ðtan aÞr Sr 20 30 40 50 60
d (mm) 15 0.75 0.50 0.40 0.30 0.25
d* 319 19 12.6 9.5 7.6 6.3
3.2.3. Scaling calculations qc 0.047a 0.0305 0.031 0.033 0.038 0.043
qcp/qcm e 1.54 1.51 1.42 1.23 1.09
Similarity of the Shields number implies Eq. (14) should be
d (mm) 31 1.55 1.03 0.78 0.62 0.52
valid for each grain size composing the bank. Using Eq. (14) d* 784 39.2 26.1 19.6 15.7 13.1
with d ¼ d90 to express the grain size scale ratio d90r, and then qc 0.047a 0.038 0.033 0.029 0.030 0.031
equating to Eq. (13) produces qcp/qcm e 1.25 1.40 1.58 1.55 1.52
d (mm) 63 3.15 2.1 1.58 1.26 1.05
1=2
Sr ¼ Zr Xr1 ¼ ðrs  rÞr ð21Þ d* 1594 79.7 53.1 39.8 31.9 26.6
qc 0.047a 0.046 0.041 0.038 0.035 0.034
qcp/qcm e 1.02 1.14 1.24 1.32 1.40
Having established Eqs. (20) and (21), ratios Sr, rsr and fr can a
For consistency with Meyer-Peter and Müller’s (1948) formula,
now be discussed. Bank stability is controlled by the internal qcp ¼ 0.047.
friction angle f, which depends on the grain size distribution
and soil compaction (Buffington et al., 1992). Because the may induce ripple development in the model. This would be
model grain size distribution may not be as well graded as that unrepresentative of the prototype behaviour.
of the prototype, and because the material compaction in the
model may be less than that in the prototype, it is likely that 3.3. New approach: similarity of initial condition for
the friction angle in the model is smaller than that of the sediment motion
prototype (fm  fp). According to Eq. (20), the bed slope
scale must, therefore, satisfy the condition Sr  1. On the other Because critical shear stress is of primary concern in
hand, if the sediment particles in the model are lighter than sediment transport, similarity of the Shields number q must be
those in the prototype (rsm  rsp), the bed slope scale must relaxed as second assumption, while keeping the same
satisfy the condition Sr  1 according to Eq. (21). These entrainment conditions of sediment particles in the model and
conflicting considerations justify that prototype. This assumption means that if the hydraulic con-
ditions in the prototype, Vp and hp (or tp), induce initial mo-
Sr ¼ 1; ðrs  rÞr ¼ 1 and fr ¼ 1 ð22Þ
tion of the prototype grain diameter dp, then the corresponding
hydraulic conditions in the model, Vm ¼ Vp/Z1/2 r and hm ¼ hp/
The model is undistorted. Equalities rsr ¼ 1 and fr ¼ 1 result
Zr (or tm), will also induce initial entrainment of the model
in the same (or similar) sediment material to be used in the
grain diameter dm. Under this assumption, and expressing the
model as in the prototype. Sand is therefore retained as the
shear stress using Darcy’s formula, the prototype and model
model bank material. Equations (14) and (15) become
shear stresses read
qr ¼ Zr dr1 ¼ 1 ð23Þ (
tcp ¼ qcp ðrs  rÞp gdp ¼ 18 rfp Vp2
ð26Þ
Rer ¼ Zr1=2 dr ¼ 1 ð24Þ tcm ¼ qcm ðrs  rÞm gdm ¼ 18 rfm Vm2
Natural material has to be used in order to replicate the
Equations (23) and (24) cannot be satisfied simultaneously. prototype material characteristics, fp and rsp. The model is
Similarity of Re* is, therefore, relaxed as first assumption in non-distorted in order to preserve the similarity of the bank
order to preserve similarity of Shields number and bank sta- stability factor (i.e. Eq. (20)). Introducing Eq. (12) into Eq.
bility factor. The sediment grain size is therefore reduced (26) to express the ratio tcp/tcm and rearranging to solve for dr
according to the geometric scale Zr, i.e. produces

d r ¼ Zr ð25Þ 1=3 qcm

Table 3 summarises the model parameters for prototype
grain sizes dp ¼ 15, 31 and 63 mm, using various geometrical
scales Zr. The commonly used similarity of q would lead to the
distortion of qc between the prototype and the model, because
the dimensionless grain diameter in the model d*m ¼ dm( gD/
n2)1/3 would be lower than the threshold value of 100e150
(beyond which the critical shear number can be assumed
constant (Van Rijn, 1993)). In addition, for dp ¼ 15 mm and
Zr ¼ 30, d*m would be lower than 15 (or Re*m < 25), which
dr ¼ Zr2=3 d90r ð27Þ
qcp
The critical Shields number for prototype grain sizes
dp ¼ 15, 31 and 63 mm ranges between 0.055 and 0.06 ac-
cording to the Shields diagram. However, a constant value
qcp ¼ 0.047 is used here in accordance with the Meyer-Peter
and Müller (1948) formula used for computing the time
scale for initiation of sediment motion, and because the flow in
the prototype is rough (see Section 6). Conversely, the flow
may be not under fully rough conditions in the model; the
 K. El Kadi Abderrezzak et al. / Journal of Hydro-environment Research 8 (2014) 95e114 103

critical Shields number may therefore depend on the grain Table 5

diameter and flow conditions. However, the relation can be Summary of the model bank material scaling, with the geometrical scale
Zr ¼ 40.
cast in explicit form by plotting qc versus the dimensionless
grain diameter d* ¼ d(gD/n2)1/3. In his work, the following Prototype grain size (in mm)
formulation is used (Van Rijn, 1993): 63 31 15
8 Model d (mm) 2 1.15 0.65
>
> q ¼ 0:24d1 1  d  4 d* 50 29 16.4
< c
qc ¼ 0:14d0:64 4  d  10 dr 31.5 27 23
ð28Þ qc
>
> q ¼ 0:044d0:1 10  d  20 0.041 0.035 0.030
: c qcp/qcm 1.14 1.34 1.56
qc ¼ 0:013d0:29
20  d  150
Hiding and exposure are ignored arguing that they may
have only a limited effect at the initiation phase of non- simulating free-surface flow on the 50 km long stretch of the
bimodal sediment mixtures (Kuhnle, 1992; Wilcock and Old Rhine using the open-source MASCARET modelling tool
McArdell, 1993). Equation (27) is first resolved for the pro- (Goutal et al., 2012). Calibration of Strickler’s coefficient was
totype coarsest fraction dp ¼ d90p ¼ 63 mm. The scale d90r is carried out using measured water levels at low, medium and
therefore used for determination of scales of the remaining high-flow discharges. Physical model data points are calcu-
prototype fractions dp ¼ 15, 31 mm. Table 4 summarises the lated using Eq. (23). Also plotted are the threshold conditions
model parameters for dp ¼ 15 mm, using various geometrical for ripple formation proposed by Engelund and Hansen
scales Zr. The geometrical scale 40 satisfies the conditions (1967), the criterion for the threshold between lower-regime
d*m > 15; larger scales are not possible because of the limited plane bed and dunes proposed by Chabert and Chauvin
floor space within the laboratory. Table 5 summarises the re- (1963), and the threshold curve that separates bed load and
sults for the remaining grain sizes, dp ¼ 31 mm and 63 mm. suspension proposed by Garcı́a (1999). Fig. 5 reveals that the
The main implication of this approach is that in the model coarsest model fraction dm ¼ 2 mm would not be transported,
each grain fraction has a different diameter scale and critical whereas discharges larger than 0.138 m3/s (Qp ¼ 1400 m3/s)
Shields number. The scale of each grain size is larger than the mobilise the fraction dm ¼ 1.15 mm. Fraction dm ¼ 0.65 mm is
geometric scale 40. Regarding the finest prototype grain size, in motion for the entire operation flow range, which means it
dp ¼ 1.15 mm, that will likely be transported as suspended will contribute to most of the bed load transport. The finest
load, a model grain size dm ¼ 0.15 mm is retained, since the
use of material smaller than 0.1 mm grain size can facilitate Table 6
cohesive properties that alter sediment transport behaviour. Summary of prototype and model properties.
The resulting model bank material is therefore a mixture of Parameter Prototype Model
four grain diameters: 20% (per mass) with dm ¼ d90m ¼ 2 mm, Froude Number Fr ¼ 1
20% with dm ¼ 1.15 mm, 20% with dm ¼ 0.65 mm, and the Reynolds number Rer > 1
remaining 40% with dm ¼ 0.15 mm. Table 6 summarises the Shields number qr > 1
prototype and model properties. The grain size distribution (1.35 z on average)
Effective shear stress (q  qc)r > 1
used in the model is plotted in Fig. 3. The geometric standard (1.25 z on average)
deviation of the sediment mixture is approximately 3, which is q/qc ratio (q/qc)r > 1 (1.33 z on
quite different from the one in the prototype. This is due to the average)
relaxation of the Shields number similarity. Particle Reynolds number Re*r > 1
Plotting both model and prototype data on a Shields curve Bank stability factor Fsr ¼ 1
Bed Slope Sr ¼ 1
allows visualisation of the theoretical differences in bed forms Channel length (m) 932 23.3
and initiation of sediment transport. Fig. 5 gives plots of the Mean channel width (m) 256 6.4
dimensionless shear stress for each grain size. Prototype data Movable-bed area (m  m) 372  116 9.3  2.9
points represent values of the Shields parameter for a range of Particle size of the bed 150 3.75
flow rates between 400 and 3150 m3/s. For each flow rate, the armour layer (mm)
Diameter of the bank riprap 1500 37.5
average hydraulic variables over the site were obtained by (mm)
Sediment time scale 1 day 4h
Table 4 Flow discharges (m3/s) 400 0.039
Similarity of initial entrainment of sediments applied to dp ¼ 15 mm. Model 600 0.059
bank material parameters according to the geometrical scale. 800 0.079
Prototype Model 1000 0.099
1200 0.119
20 30 40 50 1400 0.138
d (mm) 15 1.0 0.8 0.65 0.50 1600 0.158
dr e 15 18.7 23 30 1800 0.177
d* 380 25 20 16.4 12.6 2000 0.197
qc 0.047 0.033 0.031 0.030 0.031 2200 0.217
qcp/qcm e 1.32 1.52 1.56 1.50 3150 0.311
104 K. El Kadi Abderrezzak et al. / Journal of Hydro-environment Research 8 (2014) 95e114

in motion under the selected flow rates. For dm ¼ 1.15 mm

Fig. 5. Prototype and model values of Shields parameter over a range of flow
discharges. Also shown are the threshold condition for ripple formation pro-
posed by Engelund and Hansen (1967) and that between plane bed and dunes
proposed by Chabert and Chauvin (1963). Curves proposed by Bagnold
(1966), Van Rijn (1984) and Garcı́a (1999) are three different criteria for
the threshold between bed load and suspension.
fraction dm ¼ 0.15 mm should be transported in suspension,
although the Shields values for the corresponding prototype
diameter dp ¼ 1.15 mm are under the suspension limit ac-
cording to Garcı́a’s (1999) criterion. Nevertheless, deter-
mining the flow condition at which initiation of sediment
suspension occurs is difficult (Cheng and Chiew, 1999). Some
criteria have been proposed assuming that suspension occurs
when the upward velocity is higher than the settling velocity
of the sediment particle. Some authors consider transport as
suspension when the saltation length is high enough for par-
ticles to be often out of contact with the bottom. Two specific
threshold curves proposed by Bagnold (1966) and Van Rijn
(1984), respectively, are plotted in Fig. 5, showing that the
prototype fraction dp ¼ 1.15 mm would be transported in
suspension, similarly to its corresponding model diameter.
3.4. Time scale for sediment transport
In general, the time scale for sediment transport Tsr should
be estimated during model calibration by comparing the time
needed to obtain the same well-defined bed morphological
features in prototype and model. In the present study, no field
measurements could be made. Therefore, Tsr is derived theo-
retically from the Exner equation as
(dp ¼ 31 mm), Tsr changes by 45% when Qm changes from
0.0395 m3/s to 0.311 m3/s (400 < Qp < 3150 m3/s). For
dm ¼ 0.65 mm (dp ¼ 15 mm), Tsr changes by 20% when Qm
changes from 0395 m3/s to 0.311 m3/s. However, Tsr varies
little when the flow rate is sufficiently larger than the critical
flow for grain motion. In fact, ignoring the sediment time scale
corresponding to Qm ¼ 0.0395 m3/s (Qp ¼ 400 m3/s) and
dm ¼ 0.65 mm as well as the sediment scale corresponding to
Qm ¼ 0.138 m3/s (Qp ¼ 1400 m3/s) and dm ¼ 1.15 mm, the
average relative variation becomes lower than 5%. A global
sediment time scale is calculated as the average of the grain
motion time scales obtained for dm ¼ 0.65 mm and 1.15 mm
and for Qm between 0.0395 and 0.311 m3/s
(400 < Qp < 3150 m3/s). The average time scale is thus 6.22.
This value is comparable to the one calculated using the mean
diameter of the mixture (Tsr ¼ 6.29). It is of great importance
that Tsr is not much different from the time scale governing the
flow Tr ¼ Z0.5 r ¼ 6.32. For sake of simplicity, Tsr ¼ 6 is
retained, i.e. 4 h of model testing would represent one proto-
type day.
4. Model construction, operation and data collection
4.1. Model construction
The model is 23.3 m long and 6.4 m wide (Fig. 7, Table 6).
A series of twelve cross-sections measured in 2009 was used
to reproduce the geometry of the pilot site. The elevations and
their spacing along each cross-section were scaled, plotted,
and transferred to steel sheets, which were then cut to create
the bed topography at the actual model size. The model was
filled with a coarser material layer, except for a movable-bed
area 9.3 m long and 2.9 m wide (representing an area of
372 m by 116 m at prototype scale) that included the three
groynes and surrounding bank and bed surfaces. A thin con-
crete layer, with sediments of 3.75 mm in diameter (1/40 of
the grain size of the armour layer) glued to it, was placed over
the rigid bed. The model bank sediment was a mixture of the

Xr Zr Zr2
Tsr ¼ ¼ ð29Þ
qsr qsr
Shen (1971) recommended scaling bed load for sand- and
gravel-dominated systems using Meyer-Peter and Müller
(1948). The time scale for grain motion is calculated for a
range of hydraulic conditions and for each grain size class. For
sake of comparison, Tsr is calculated assuming the sediment
mixture is represented by the arithmetic mean diameter. The
prototype and model material bed load have mean diameters
of 25 mm and 0.94 mm, respectively. Results are shown in Fig. 6. Time scale for grain motion for a range of flow rates
Fig. 6. The coarsest fraction dm ¼ 2 mm (dp ¼ 63 mm) is not 0.0395 < Qm < 0.311 m3/s (400 < Qp < 3150 m3/s).
 K. El Kadi Abderrezzak et al. / Journal of Hydro-environment Research 8 (2014) 95e114
Fig. 7. (a) Diagram of the physical model; (b) View showing initial river bank. The red outline marks the limits of the movable-bed and bank area. All dimensions
in m (model). Arrow indicates direction of flow. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this
article.)
four sizes resulting from the scaling process. The bank riprap
was built using angular pebbles with a uniform grain size of
37.5 mm (1/40 of the prototype size), whereas the bed sedi-
ment in the movable area was represented again with a uni-
form grain size of 3.75 mm.
4.2. Experimental runs
Laboratory tests were carried out using constant flow rates
rather than attempting to simulate flow hydrographs. The main
purpose of the model was to evaluate the effect of different
engineering scenarios promoting bank erosion. Thus, it was
necessary to assess this from one or more reference discharge
regimes, regardless of their particular hydrograph shape. Tests
were carried out for flow rates ranging from 0.0395 m3/s to
0.311 m3/s (400e3150 m3/s prototype) over 4 h each (one
prototype day) (Table 6). This duration was suitable for
erosion to occur in the model to allow comparison of various
discharges and scenarios.
At the head of the model, flow entered a headbox where it
passed through a ported pipe diffuser and over a fixed weir.
The sediment supply was zero, because the upstream alternate
bar on the right bank in the prototype does not migrate
downstream (Arnaud et al., 2012). The downstream boundary
condition was defined by a stage-discharge curve, generated
from 1-D numerical modelling of the 50 km long stretch of the
Old Rhine using the 1-D MASCARET tool. The initial
105
movable-bed topography was remade after each experiment.
The bank material was compacted before each test, further
increasing the angle of internal friction, thus approaching the
prototype value (i.e. fm z fp).
4.3. Data acquisition
For each test, measurements included bed elevations using
a 3-D scanner, free-surface velocity using the Large-Scale
Particle Image Velocimetry (LSPIV) technique, water eleva-
tions using a laser-based range instrument with a vertical
precision of 0.1 mm and surface grain size distributions at
selected locations. The surface grain size distributions were
calculated from 2 cm deep samples taken from the final
movable-bed surface, and analysed by sieving.
4.3.1. Bed elevation measurements
A Konica Minolta Vi910Ò 3-D laser triangulation scanner
was used to measure bed elevations at the beginning and end
of each experiment (after draining the model), with simulta-
neous range and colour image acquisition, a range measure-
ment precision (in a plane perpendicular to the range) of
0.008 mm and a point positional accuracy of 0.10 mm.
This portable scanner is composed of a single unit presenting
two circular apertures hosting a laser emitting unit and a
Charge-Coupled Device (CCD) camera (640,480 pixels).
Three lenses provide flexibility in the scan area from
106 K. El Kadi Abderrezzak et al. / Journal of Hydro-environment Research 8 (2014) 95e114
approximately 102 up to 1 m2. The scanner uses a light-stripe
method to acquire the surface geometry by emitting a red laser
beam and converting the reflected light into distance through
an active triangulation approach. The result is a dense set of
640,480 3-D points outlining the part of the bed surface.
To transform the point clouds into an oriented Digital
Elevation Model (DEM), the Geomagic Qualify12Ò software
was used. The measured point clouds were wrapped into an
initial mesh and subsequently cleaned of holes and other
spurious artefacts. Next, the meshed views were registered
(aligned) with one another, first manually using spherical
targets placed on the scanned surface, and then automatically.
The resulting registered views were then merged into a single
unstructured tri-angular mesh. Finally, the merged mesh was
aligned to the common model coordinate system, creating a
DEM of the bed surface. Geomagic Qualify12 was used to
compare initial and final bed elevation DEMs, to compute
areas and volumes of eroded and deposited sediment and to
scale the DEMs to prototype size.
4.3.2. Surface velocity measurements
The time-averaged surface velocity field was measured
using the LSPIV technique. Measurements were performed
for a large spatial domain approximately 5.5 m long
(streamwise) by 1.0 m wide. The free surface was seeded with
5e10 mm diameter floating white cork spheres, coated to
minimise agglomeration effects. Particles were uniformly
seeded using a dispenser located upstream of the imaging
area. It was not possible to measure velocity fields during
active bed evolution, because tracer particles were prone to
beaching near the bank and interacting with the erosion pro-
cess. Images were captured using a high-definition Pan-
asonicÒ HDC-HS9 video camera with a 1920  1080 pixel
image size. Images were processed using the in-house LSPIV
software (El kadi Abderrezzak et al., 2012). The accuracy of
LSPIV measurements was evaluated by comparison with
Acoustic Doppler Velocimetery (ADV) probe measurements,
taken at certain locations of the model. Overall, similar results
were obtained with deviations being less than 6%. Such ac-
curacy can be qualified as acceptable for LSPIV measure-
ments in laboratory flumes, according to the findings of
Kantoush et al. (2011).
Fig. 8. (a) Shape and dimensions of the new island groynes. Dimensions in m (model). (b) New model setup. Distance between groynes is 4.58 m (model).
5. Model results and observations
More than 130 tests and various scenarios were carried out
in the physical model. Detailed results and analyses are
available in Die Moran et al. (2013). The principal criterion
used to judge the efficiency of a given scenario was the bank
erosion volume.
The optimal strategy for creating bank erosion was found in
the form of island groynes that would deflect the flow against
the bank (Fig. 8): the bank riprap is removed over a length of
9.3 m (372 m prototype) and the three initial groynes are
replaced by two larger and higher island groynes, one at the
position of initial groyne G1 and another one at the position of
initial groyne G3. The groyne G2 is removed to create a larger
space between the two island groynes, in which the flow can
recover from the disruption caused by G1. The new groynes
are both perpendicular and disconnected from the bank, sub-
merged for flow rates above 0.0988 m3/s (1000 m3/s proto-
type), leaving a small lateral channel where the flow can
accelerate during low to medium flow discharges.
A series of tests was performed at constant flow rates in the
range 0.0395e0.311 m3/s (400e3150 m3/s prototype) with a
duration of 4 h (1 prototype day) (Figs. 9 and 10). For
0.059 < Qm < 0.0988 m3/s (600 < Qp < 1000 m3/s), the flow
passing through the narrow channel between groyne and bank
impinged on the bank, generating high boundary shear stresses
at the toe, and therefore causing the bank to erode. For larger
discharges, fluvial erosion was mainly caused by water flow-
ing parallel to the bank. In both cases, fluvial entrainment of
sediment from the basal bank area resulted in scour in this
zone. This process continued, increasing bank angle, to a point
at which bank failure occurred. Blocks of bank material fell
into the flow and were then transported downstream of each
groyne. The erosion process reached the top of the bank for
Qm > 0.217 m3/s (2200 m3/s prototype, the 10-year return
period flood), but was not excessive even for Qm ¼ 0.311 m3/s
(3150 m3/s prototype, the 100-year return period flood).
Moreover, observations during the runs clearly showed that the
bed armour layer was not broken. Fig. 11 displays the
resulting volumes of erosion downstream of G1 and G2. The
total eroded material is in the range 0.6e11  102 m3
(380e7040 m3 prototype).
 K. El Kadi Abderrezzak et al. / Journal of Hydro-environment Research 8 (2014) 95e114 107

Fig. 9. Final bed elevation differences after a 4-h test at Qm ¼ 0.0988 m3/s (Qp ¼ 1000 m3/s). (a) around G1, (b) around G3. Dimensions in mm (model). Negative
values indicate degradation. The grey areas were not measured since they were not affected by erosion or deposition, and therefore they are irrelevant for the
results.

The duration of each test was found sufficiently long to
reach a state where no more erosion and sediment transport
occurred. Fig. 12 plot the streamwise velocity field at the end
of the test for Qm ¼ 0.0988 m3/s (Qp ¼ 1000 m3/s); the surface
velocity vectors are shown in Fig. 13. The flow between bank
and groyne is slower than the flow passing around the groyne
on the outer side, but is still redirected towards the bank. The
boundary shear stress is calculated using Darcy’s formulation
with the mean velocity V being estimated from the surface
velocity u as V ¼ 0.9u (Graf and Altinakar, 1993). Results are
shown in Fig. 14, confirming the interruption of sediment
transport, since shear stresses are mostly lower than the critical
value (z0.17 N. 2, estimated from the Shields curve)
necessary for the initiation of motion of the fraction
dm ¼ 0.15 mm.
Values of the prototypeemodel ratios of median diameter
(d50) and standard deviation (s) of the final bed surface with
respect to the same values in the original bank sediment mix

Fig. 10. Final bed elevation differences after a 4-h test at 0.198 m3/s (Qp ¼ 2000 m3/s). Dimensions in mm (model). (a) around G1, (b) around G3. Negative values
indicate degradation. The grey areas were not measured since they were not affected by erosion or deposition, and therefore they are irrelevant for the results.
108 K. El Kadi Abderrezzak et al. / Journal of Hydro-environment Research 8 (2014) 95e114
Fig. 11. Volumes of erosion for each test, downstream of each groyne.
are plotted in Fig. 15 for two selected flow rates. New bed
material does not exhibit a clear downstream fining and an
upwards coarsening of the grain sizes. This finding is not
surprising, because of the highly 3-D effects in place, and the
interaction between the advancing aggradation front and the
continuous sediment inputs generated by bank erosion: ma-
terial which has been transported over a certain distance does
undergo a certain degree of fining, but is continuously mixed
with newly eroded sediments from the bank.
6. Discussion
The bed changes observed in physical models may deviate
from those in the prototype. A standard technique for evalu-
ating the accuracy of a physical model is to assess its ability to
reproduce a well recorded prototype historical event. In the
present case, no historical morphology event with existing data
was monitored.
Although the model is undistorted, which increases its ac-
curacy and facilitates the interpretation of the results, it does
have scale effects due to not respecting part of the similarity
conditions. The most important scale effects are discussed
below, classified into major and minor ones.
Fig. 12. Velocity streamwise field obtained from LSPIV measurements at the end of the test for Qm ¼ 0.0988 m3/s (Qp ¼ 1000 m3/s). (a) G1 (b) G3.
6.1. Major scale effects
6.1.1. Shields parameter
The Shields number was not preserved between the model
and the prototype. From Eqs. (26) and (27), it can be seen that
qm < qp for each grain size. The ratio qp/qm is equal on average
to 1.15, 1.35 and 1.56 for dm ¼ 2 mm (dp ¼ 63 mm), 1.15 mm
(dp ¼ 31 mm) and 0.65 mm (dp ¼ 15 mm), respectively.
Similar values are obtained when comparing the effective
dimensionless shear stress (q  qc) (which controls the bed
erosion rate) between the prototype and the model. It is worth
noting that the fraction dm ¼ 0.65 mm, which contributes to
most of the bed load, entails the highest distortion of the
Shields number.
Comparison of Eq. (25) (i.e. qr ¼ 1 and grain size is
reduced according to Zr) and Eq. (27) shows that relaxing the
Shields number similarity induces an over-coarsening of the
model bank grain sizes. Larger sediment grains are heavier
and hence require a greater shear stress to entrain them, so the
relative amount of erosion observed in the physical model may
be markedly smaller than would occur in the prototype. The
magnitude of this underestimate is not obvious to calculate,
because of the presence of other scale effects that either tend
to compensate or exaggerate further this trend (e.g. cohesion,
vegetation, ripples). If we ignore these additional scale effects,
and in comparison with an “idealized” physical model that
ensures the Shields number similarity at the same geometrical
scale (Zr ¼ 40) and with the same bank material as the actual
model, the magnitude of under-prediction in the prototype
eroded sediment volume could be of the order ðq  qc Þ1:5 r , that
is a factor of 1.4 in average.
6.1.2. Cohesion, pore-water pressure and vegetation
Similarity of the bank failure used in this study relies on the
assumption that bank material in the model is cohesionless.
However, the laboratory observation showed that the materials
used to build the bank material displayed some degree of
cohesiveness, probably associated to the finest fraction
 K. El Kadi Abderrezzak et al. / Journal of Hydro-environment Research 8 (2014) 95e114 109

Fig. 13. Surface velocity vectors obtained from LSPIV measurements at the end of the test for Qm ¼ 0.0988 m3/s (Qp ¼ 1000 m3/s). (a) G1 (b) G3.

dm ¼ 0.15 mm. The bank stability coefficient Fs is therefore
not strictly matched between the prototype and the model. Due
to cohesion, the model bank stability coefficient would be
exaggerated (i.e. Fsm > Fsp), and the model bank erosion
magnitude might thus be reduced in comparison with the
prototype. Furthermore, capillary rise creates negative pore-
water pressure above the water table, resulting in an apparent
cohesion for the bank. Because this phenomenon is a function
of bank material particle size, the extent to which the bank is
affected would be much greater in the model than in the
prototype (where capillary rise would occur only to a limited
extent). However, this apparent cohesion would disappear in
the model, because the bank was fully saturated (Rinaldi and
Casagli, 1999).
Finally, vegetation plays an important role in strengthening
banks, particularly in coarse-grained rivers where cohesion is
largely unimportant. In the context of the present project,
removal of the riparian vegetation is planned, justifying the
non-incorporation of vegetation in the physical model. How-
ever, the remaining roots generally reinforce the soil by
increasing apparent soil cohesion and shear strength (Gray and
Leiser, 1982; Pollen and Simon, 2005). To a lesser extent, they
may create pathways for water to enter the bank, increasing
pore-water pressures, and thus increasing the mass failure risk
(Simon and Collison, 2002). The ability to accurately replicate
root systems of prototype vegetation in a physical model is not
obvious: size, distribution and networking of roots are linked
to the type of vegetation and its age as well as the bank ma-
terial characteristics (Cancienne et al., 2008). It may be
possible to consider the effect of roots by increasing the
cohesion of the surface layer of the model that would contain
vegetation roots in the prototype. Healey (1997) suggested
spreading a thin layer of lime over the model river bank. Such
an “artificial” approach may add complexity to the interpre-
tation of the results, and may be valid where only the general
effect of vegetation on river bank erosion is being investigated
in relation to another process, which is beyond the scope of
this laboratory research work. In any case, due to the absence
of roots in the present model bank, the erosion magnitude may
be somewhat exaggerated in comparison with the prototype.
The degree to which this effect compensates that of cohesion
is difficult to judge.

Fig. 14. Boundary shear stress at the end of the test for Qm ¼ 0.0988 m3/s (Qp ¼ 1000 m3/s). (a) G1 (b) G3.
110 K. El Kadi Abderrezzak et al. / Journal of Hydro-environment Research 8 (2014) 95e114
Fig. 15. Grain size downstream of G1 and G3 for tests at (a) Qm ¼ 0.0988 m3/s
(Qp ¼ 1000 m3/s) and (b) Qm ¼ 0.198 m3/s (Qp ¼ 2000 m3/s). Location of
samples is shown in Fig. 9.
6.2. Minor scale effects
6.2.1. Ripples
It is readily apparent from laboratory observations that
ripples have formed in the model. This happened although the
model scale (Zr ¼ 40) was initially chosen to ensure ripples
would not form. The reasons for ripple formation in the model
are threefold: first the model scaling approach was based on
average hydraulic values over the whole pilot site, whereas the
model had areas where the particle Reynolds number is small
(Re* < 25) so that ripples could occur. The model dimen-
sionless shear stress values for dm ¼ 0.65 mm come near the
threshold condition for ripple formation, as shown in Fig. 5.
Second, as stated by many authors (Raudkivi, 1997; ASCE,
2000; Julien, 2002) using sediment diameters less than
0.7e0.9 mm may produce ripples for flows near or slightly
above the flow associated with incipient bed motion. Finally,
the finest fraction of bank material, dm ¼ 0.15 mm, may not be
washed out completely but rather may travel solely in sus-
pension or may be repeatedly deposited and re-entrained, thus
contributing to ripple formation.
The size of ripples remains too small to significantly alter
the entire flow field. On the other hand, ripple formation may
increase the bed load transport rate, since the vortex moving
across the ripple may mobilise and entrain sediment (Young
and Warburton, 1996). In the present study, the degree to
which sediment transport is over-mobilised within the rippled
bed is reduced, mainly because of the inability of the ripple-
induced vortex to generate large shear stresses.
6.2.2. Time scale for sediment transport
The time scale for sediment transport depends on the
sediment transport formula adopted. A sensitivity analysis of
the initial scaling calculations was undertaken using two other
bed load capacity formulae (Ashida and Michiue, 1972;
Parker, 1979), appropriate for sand-bed and gravel-bed
rivers. High uncertainty in the values of the grain motion
time scale at the stage of initiation of motion was also
apparent. The calculated average time scale is 6.12 and 5.7
using Ashida and Michiue’s (1972) formula and Parker’s
(1979) formula, respectively. Both values do not differ sub-
stantially from that obtained with the Meyer-Peter and Müller
(1948) formula (Tsr ¼ 6.12).
6.2.3. Particle Reynolds number
The particle Reynolds number Re* is a measure of the bed
roughness relative to the thickness of the viscous sublayer. It is
not necessary to preserve Re* between the model and the
prototype provided that the value in the model is high enough
to ensure the flow is hydrodynamically rough. The lower
bound of Re* to ensure such a condition is poorly defined,
mainly because of the data scatter on the Shields diagram,
which makes the placing of a rough flow regime a partially
subjective exercise. ASCE (2000) suggests a value of 400 but
notes that the effects of viscosity become progressively less
important above a value of 60. This is consistent with sug-
gestions of Chanson (1999) and Yalin (1971) who state values
of 100 and 70, respectively. Values as low as 5 and 15
(Ashworth et al., 1994) and as high as 350, 400 and 1000 were
also proposed (Peakall et al., 1996).
When calculating Re* for a non-uniform sediment material,
the question arises which grain size the values should be based
on. Flow conditions in the vicinity of the bed depend upon the
channel bed roughness. The presence of a bed armour layer
and bank sediment sorting imply that a fairly high grain size
will characterise the roughness. In the area surrounding the
eroded bank, it can be argued that a diameter between the bank
material size d90 and the diameter of the bed armour layer
represents near-bed roughness well and disturbs the laminar
sublayer sufficiently to create a hydraulically rough boundary.
Here, d90 is retained as the representative sediment diameter.
For the prototype, the hydraulic variables over the site are
obtained using the calibrated 1-D model MASCARET of the
Old Rhine (because of lack of field data). The flow is found
rough because the particle Reynolds number Re*p is high for
the range of tested flow discharges (i.e. Re*p > 1000). For the
model; the particle Reynolds number Re*m is calculated using
the shear velocity estimated from Darcy’s equation and mean
velocity field and flow depths measured in the laboratory. For
dm ¼ d90m ¼ 2 mm and Qm ¼ 0.0593 m3/s (Qp ¼ 600 m3/s),
Re*m is found to range between 15 and 25, so rough-turbulent
flow might not be achieved for low flow discharges. However,
these values may be exceeded in areas where the roughness is
 K. El Kadi Abderrezzak et al. / Journal of Hydro-environment Research 8 (2014) 95e114
more influenced by the bed armour layer (3.75 mm sediment
diameter).
In any case, with regard to the uncertainty in the definition
of the threshold value of Re* where the flow may be deemed to
be fully rough, it is probable that most laboratory tests were
performed under similar hydraulic conditions to those that
would be likely to occur in the prototype. The scale effect of
not respecting Re* can be qualified as minor.
6.2.4. Relative roughness
Matching the ratio of the grain size to flow depth, d/h,
between the prototype and the model implies that the bank
sediment grain sizes should be scaled directly by the geo-
metric scale, i.e. dr ¼ Zr. In the present work, only the armour
layer and riprap were geometrically scaled.
The similarity of d/h is generally important when consid-
ering surface tension effects. To render these effects negli-
gible, Novak and Cábelka (1981) state that flow velocity
should exceed 0.23 m/s and flow depth should exceed 0.015 m
(0.02 m according to ASCE, 2000). Here, the flow depth and
velocity were high enough in each tested flow discharge. The
average model flow depth near the bank was approximately
2000 times the grain size dm ¼ 2 mm for Qm ¼ 0.0593 m3/s
(Qp ¼ 600 m3/s).
The relative roughness exerts a control on incipient sedi-
ment motion, which can be augmented by other factors such as
bed slope (Recking, 2009). However, for sand mixtures, small
bed slopes and large relative roughness values (d/h > 50), such
control was found to be weak (Garcı́a and Parker, 1991). Thus,
expressing the critical Shields number of the model bank
material using Van Rijn’s (1993) formulation ignoring the
effect of d/h is justified.
6.2.5. Friction coefficient equation
The effect of approximating DarcyeWeisbach’s coefficient
f is assessed by using the original formulation, i.e. Eq. (11).
The grain size scale dr is expressed again as dr ¼ Zrfr/qcr, with
the geometrical scale Zr ¼ 40 and qcp ¼ 0.047. Contrary to Eq.
(27), the grain scale dr now depends on the hydraulic variables
(because of the presence of Re in the original formulation of f,
Eq. (10)). For the coarsest prototype grain size
dp ¼ d90p ¼ 63 mm, the corresponding model grain diameter is
found to range between 1.989 mm for Qp ¼ 400 m3/s and
1.976 mm for Qp ¼ 3150 m3/s. In comparison with the actual
model grain size of 2 mm, the use of Eq. (11) yields a decrease
in the model grain size of approximately 1% on average. For
dp ¼ 31 mm and 15 mm, the decrease in the corresponding
model grain size diameters is 1.7% and 3% on average,
respectively. In summary, the model grain sizes used to
reproduce the prototype material were somewhat coarser than
they should be, but this over-coarsening remains weak.
6.2.6. Reynolds number for flow
Similarity of both the Reynolds and Froude numbers be-
tween model and prototype can only be achieved using a
modelling fluid with a lower viscosity than that of the proto-
type. As water is usually the only modelling fluid available,
111
the only way to reduce the viscosity is by raising the water
temperature which was not feasible in the present study.
The model was scaled based on the similarity of the Froude
number. The flow Reynolds number Re was relaxed with the
proviso that the flow would remain within the fully turbulent
flow regime in both prototype and model. Fully turbulent flow
occurs when Re is larger than a threshold value. Various values
have been proposed: 500 (Peakall et al., 1996), 1400 (Allen,
1947; Bennett et al., 2008) and 2000 (Gill and Pugh, 2009).
The assumption of fully turbulent flow has been verified in
both prototype and physical model for all tested flow rates.
The prototype Reynolds number Rep is obtained from the 1-D
numerical simulation of the Old Rhine using MASCARET,
and shows that the flow is fully turbulent (Rep is higher than
100,000). The model Reynolds number in the area surrounding
the eroded bank is estimated using the recorded flow depths
and mean flow velocity field (V ¼ 0.9u). For the model flow
rate Qm ¼ 0.0593 m3/s (Qp ¼ 600 m3/s), Rem is found to be in
the range between 1000 and 10,000. Therefore, the effects of
viscosity in the model are not exaggerated relative to those of
turbulence.
7. Conclusion
An undistorted, movable-bed physical model was built and
used to investigate non-uniform sediment transport, fluvial
erosion and bank failure in the Old Rhine (France). The model
was not designed for quantitative studies of the riverbed
morphology, but rather to trial various engineering options to
rehabilitate sediment transport through bank erosion.
The model was designed, as closely as physical constraints
would permit, according to similarity of the Froude number,
ColebrookeWhite equation relationship and bank stability
coefficient. This specific scaling approach ensures similarity
of initial motion for each grain size class composing the bank
material, but relaxes the similarity of Shields and particle
Reynolds numbers in order to restrict formation of ripples and
high distortion of the critical Shields number. Sand was
retained as model bank material in order to preserve mass
density and internal friction angle between the model and the
prototype. However, this methodology results in scale effects.
The most important ones are due to the relaxation of the
Shields number and to the fact that the model material shows
cohesive properties. Not respecting the Shields number means
the model grain sizes used to reproduce the prototype material
are somewhat coarser than they should be. The eroded vol-
umes could be underestimated by these larger grains (by as
much as 40%) as well as by the presence of cohesion in the
model bank material allowing the bank to be more stable than
in the prototype. Vegetation roots, which cannot be replicated
in the model and, to a lesser extent, the formation of ripples,
which may increase bed load transport, could partially
compensate the aforementioned scale effects. Consequently,
the eroded bank material volumes in the model are recognised
as not necessarily comparable to the volumes in prototype.
Nonetheless, the physical model should correctly reproduce
the overall magnitude and pattern of erosion and deposition
112 K. El Kadi Abderrezzak et al. / Journal of Hydro-environment Research 8 (2014) 95e114

within the prototype, and provide guidance on the best groyne P hydrostatic-confining force due to external water
modification to produce bank erosion. level
The investigation of the different scenarios showed that the Pw ¼ V  S/ws dimensionless unit stream power
optimal solution consists of replacing the three existing initial qs volumetric bed load transport per unit width
groynes by two higher, larger island groynes, which are qs* ¼ qs/(u*d ) dimensionless unit sediment discharge
perpendicular to and disconnected from the bank, leaving a Q discharge
small lateral channel where the flow can accelerate during low Re Reynolds number
to medium flow discharges. The effectiveness of this config- Re* ¼ u*d/y particle Reynolds number
uration was proven for a range of flow rates up to the 100-year Rh hydraulic radius
return period flood. S longitudinal bed slope
This study has confirmed the utility of physical models in Spw force produced by matric suction on the unsaturated
solving specific engineering problems. Despite their inherent part of the failure (i.e. negative pore-water pressure)
scale effects, they provide insight in the relevant processes and T time
guidance on the best engineering intervention for a specific Ts characteristic time for sediment transport
purpose. Engineering works at the pilot site for implementing u free-surface
pffiffiffiffiffiffiffi
ffi velocity
the described solution are scheduled for late 2012, and will be u ¼ t=r shear velocity
accompanied by an extensive monitoring programme u*c critical shear velocity
(morphology, habitat, biology). Uw uplift force due to positive pore pressure acting on a
unit width of the failure block
V flow velocity
Acknowledgements
ws settling velocity
W weight of a unit width of the failure block
The authors would like to thank two anonymous reviewers
Xr longitudinal geometric scale ratio
for their detailed review and valuable comments, which
Yr transverse geometric scale ratio
greatly help improvement of the original manuscript.
Zr vertical geometric scale ratio
a angle of bank failure plane
a1 bank angle
Notations D ¼ (rs  r)/r relative sediment mass density
q Shields number
qc critical value of Shields number for inception of
sediment motion
B channel width y kinematic viscosity of water
c bank cohesion r water mass density
d sediment diameter rs sediment mass density
d16 grain size for which 16% of sediment is finer by s geometric standard deviation
weight t bed shear stress
d50 median diameter (grain size for which 50% of sedi- tc critical shear stress for incipient sediment motion
ment is finer by weight) f angle of internal friction of sediments
d84 grain size for which 84% of sediment is finer by fb angle expressing the rate of increase in strength
weight relative to the matric suction
d90 grain size for which 90% of sediment is finer by F ¼ qs/[d1.5( gD)0.5] sediment transport intensity
weight
d* ¼ d( gD/y2)1/3 dimensionless grain diameter
f friction factor Subscript
F Froude number p prototype
FD driving forces acting on a unit width of the failure m model
block r scale of a quantity ¼ ratio of the prototype value to
FR resisting forces acting on a unit width of the failure the model value of that quantity
block
Fs bank stability coefficient
g gravitational acceleration
h flow depth References
Htw hydrostatic force exerted by water present in the
tension crack on a unit width of the failure block Allen, J., 1947. Scale Models in Hydraulic Engineering. Longmans, Green &
Co., London.
ks surface roughness Amiri-Tokaldany, E., Samadi, A., Darby, S.E., 2008. Sensitivity analysis of the
l characteristic length effective parameters in the riverbank stability. In: Proc. International
L length of the failure plane Conference on Fluvial Hydraulics RiverFlow 2008 September, Izmir.
 K. El Kadi Abderrezzak et al. / Journal of Hydro-environment Research 8 (2014) 95e114
Armanini, A., Sartori, F., Tomio, G., Cerchia, F., Vergnani, M., 2010. Analysis
of a fluvial groynes system on hydraulic scale model. In: Proc. Interna-
tional Conference on Fluvial Hydraulics RiverFlow 2010, Braunschweig.
Arnaud, F., Schmitt, L., Piégay, H., Beal, D., Rollet, A.J., 2012. Geomorphic
diagnosis (1880e2009) for the restoration of an altered river reach: the Old
Rhine between Kembs and Breisach (France, Germany). In: Proc. IS.Rivers
Integrative Sciences and Sustainable Development of Rivers, Lyon.
ASCE (American Society of Civil Engineers), 2000. Hydraulic Modeling:
Concepts and Practice. In: ASCE Manuals and Reports on Engineering
Practice, Manual, vol. 97. ASCE (American Society of Civil Engineers),
Reston, Virginia.
Ashida, K., Michiue, M., 1972. Study on hydraulic resistance and bed load
transport rate in alluvial streams. Trans. Jpn. Soc. Civil Eng. 26, 59e69.
Ashworth, P.J., Best, J.L., Leddy, J.O., Geehan, G.W., 1994. The physical
modelling of braided rivers and deposition of fine-grained sediment. In:
Kirkby, M.J. (Ed.), Process Models and Theoretical Geomorphology. John
Wiley and Sons Ltd, Chichester.
Bagnold, R.A., 1966. An Approach to the Sediment Transport Problem from
General Physics. U.S. Geological Survey. Professional Paper 422-I.
Bennett, S.J., Wu, W., Alonso, C.V., Wang, S.S.Y., 2008. Modeling fluvial
response to in-stream woody vegetation: implications for stream corridor
restoration. Earth Surf. Process. Land. 33, 890e909.
Bieri, M., Müller, M., Boillat, J., Schleiss, A., 2012. Modeling of sediment
management for the Lavey run-of-river HPP in Switzerland. J. Hydraulic
Eng. 138, 340e347.
Bromley, C., 2008. The Morphodynamics of Sediment Movement through a
Reservoir during Dam Removal. PhD Thesis. University of Nottingham,
Nottingham.
Buffington, J.M., Montgomery, D.R., 1997. A systematic analysis of eight
decades of incipient motion studies, with special reference to gravel-
bedded rivers. Water Resour. Res. 33, 1993e2029.
Buffington, J.M., Dietrich, W.E., Kirchner, J.W., 1992. Friction angle mea-
surements on a naturally formed gravel streambed: implications for critical
boundary shear stress. Water Resour. Res. 28, 411e425.
Cancienne, R.M., Fox, G.A., Simon, A., 2008. Influence of seepage under-
cutting on the stability of root-reinforced streambanks. Earth Surf. Process.
Land. 33, 1769e1786.
Chabert, J., Chauvin, J.L., 1963. Formation des dunes et de rides dans les
modèles fluviaux. Bull. C.R.E.C. 4 (in French).
Chanson, H., 1999. The Hydraulics of Flow in Open Channels. Arnold, London.
Chauvin, J.L., 1962. Similitude des modèles de cours d’eau a fond mobile.
Bull. C.R.E.C. 1, 64e91 (in French).
Cheng, N.S., Chiew, Y.M., 1999. Analysis of initiation of sediment suspension
from bed load. J. Hydraulic Eng. 125, 855e861.
Davinroy, R.D., Gordon, D., Rhoads, A., Abbott, J., 1999. Sedimentation
Study of the Missouri River, Copeland Bend, Miles 569 to 564.5. Hy-
draulic Micro Model Investigation. Technical Report M10. US Army
Corps of Engineers.
De Vries, M., 1993. Use of Models for River Problems. In: Studies and Re-
ports in Hydrology Series, vol. 51. UNESCO.
Die Moran, A., El kadi Abderrezzak, K., Mosselman, E., Habersack, H.,
Lebert, F., Aelbrecht, D., Laperrousaz, E., 2013. Physical model experi-
ments for sediment transport restoration in the Old Rhine through induced
bank erosion. Int. J. Sediment Res. in press.
Einstein, H.A., Chien, N., 1956. Similarity of distorted river models with
movable beds. Trans. ASCE 121, 440e457.
El kadi Abderrezzak, K., Die Moran, A., Jodeau, M., 2012. Optical techniques
for surface velocity and bed elevation measurements in a fluvial physical
scale model. In: Proc. International Conference on Fluvial Hydraulics
RiverFlow 2012, San Jose.
Engelund, F., Hansen, E., 1967. A Monograph on Sediment Transport.
Technisk Forlag, Copenhagen, Denmark.
Ettema, R., Arndt, R., Roberts, P., Wahl, T., 2000. Hydraulic Modeling:
Concepts and Practice. ASCE Publications, Reston VA, USA.
Franco, J.J., 1978. Guidelines for the Design, Adjustment and Operation of
Models for the Study of River Sedimentation Problems. Department of the
Army, Waterways Experiment Station. Hydraulics Laboratory Report H-
78-1.
113
Gaines, R.A., Maynord, S.T., 2001. Microscale loose-bed hydraulic models. J.
Hydraulic Eng. 127, 335e338.
Garcı́a, M.H., 1999. Sedimentation and erosion hydraulics. In: Mays, Larry
(Ed.), Hydraulic Design Handbook.
Garcı́a, M.H., Parker, G., 1991. Entrainment of bed sediment into suspension.
J. Hydraulic Eng. 117, 414e435.
Gill, T.W., Pugh, C.A., 2009. Sediment transport similitude for scaled physical
hydraulic modeling. In: Proc. 33rd IAHR Congress, Vancouver.
Goutal, N., Zaoui, F., Lacombe, J.M., El kadi Abderrezzak, K., 2012. Mas-
caret: a 1-D open-source software for flow hydrodynamic and water
quality in open channel net-works. In: Proc. International Conference on
Fluvial Hydraulics RiverFlow 2012, September 6e8, San Jose.
Graf, W.H., Altinakar, M.S., 1993. Hydraulique fluviale: écoulement perma-
nent uniforme et non uniforme Tome 1. Traité de Génie Civil, Ecole
polytechnique fédérale de Lausanne, Presse polytechnique et universitaire
romanes (in French).
Gray, D.H., Leiser, A.T., 1982. Biotechnical Slope Protection and Erosion
Control. Van Nostrand Reinhold Company, New York.
Healey, M.O., 1997. Investigation of Flood Risk & Erosion Mitigation on the
Rangitata River at Klondyke. PhD Thesis. Lincoln University.
Heller, V., 2011. Scale effects in physical hydraulic engineering models. J.
Hydraulic Res. 49, 293e306.
Ho, J., Coonrod, J., Gill, T., Mefford, B., 2010. Case study: movable bed
model scaling for bed load sediment exclusion at intake structure on Rio
Grande. J. Hydraulic Eng. 136, 247e250.
Jaeggi, M.N.R., 1986. Non distorted models for research on river morphology.
In: Proc. IAHS Symposium on Scale Effects in Modelling Sediment
Transport Phenomena, Toronto.
Jain, S., 2001. Scaling laws for movable-bed river models. In: Proc. 29th
IAHR Congress, Beijing.
Julien, P.Y., 2002. River Mechanics, second ed. Cambridge University Press.
Kantoush, S.A., Schleiss, A., Tetsuya Sumi, T., Murasaki, M., 2011. LSPIV
implementation for environmental flow in various laboratory and field
cases. J. Hydro-environ. Res. 5, 263e276.
Kishi, T., Kuroki, M., Imaizumi, S., 1975. An experimental study on the
meandering flow in straight channel with fixed side bank. In: Proc. 30th
Annual Conference of the Japan Society of Civil Engineers (in
Japanese).
Kuhnle, R.A., 1992. Fractional transport rates of bedload on Goodwin Creek.
In: Billi, P., Hey, R.D., Thorne, C.R., Tacconi, P. (Eds.), Dynamics of
Gravel-bed Rivers, pp. 141e155.
Marr, J.D.G., Hill, C., Johnson, S., Grant, G., Campbell, K., Mohseni, O.,
2007. Physical Model Study of Marmot Dam Removal: Cofferdam Notch
Location and Resulting Fluvial Responses. Report to Portland General
Electric, Portland, Oregon.
Maynord, S.T., 2006. Evaluation of the micromodel: an extremely small-scale
movable bed model. J. Hydraulic Eng. 132, 343e353.
Mefford, B., 2005. Elwha River Surface-water Intake Structure. U.S.
Department of the Interior, Bureau of Reclamation. Hydraulic Laboratory
Report HL-2004-03.
Mefford, B., Gill, T., 2010. Physical Hydraulic Model Proposal for US
Army Corps of Engineers Missouri River Bend Model. U.S. Department
of the Interior, Bureau of Reclamation. Hydraulic Laboratory Report
HL-2010-05.
Mefford, B., Stowell, H., Heinje, C., 2008. Robles Diversion Dam High Flow
and Sediment Bypass Structure, Ventura, CaliforniaePhysical Model
Study. U.S. Department of the Interior, Bureau of Reclamation. Hydraulic
Laboratory Report HL-2008-7.
Meyer-Peter, E., Müller, R., 1948. Formulas for bed-load transport. In: 2nd
IAHR Congress, Stockholm, Sweden.
Novak, P., Cábelka, J., 1981. Models in Hydraulic Engineering e Physical
Principles and Design Applications. Pitman, Boston.
Parent, A.P., 1988. Scale Models of Gravel Bed Rivers. PhD Dissertation.
University of British Columbia.
Parker, G., 1979. Hydraulic geometry of active gravel rivers. J. Hydraulic Div.
ASCE 105, 1185e1201.
Peakall, J., Ashworth, P., Best, J.L., 1996. Physical modelling in fluvial geo-
morphology: principles, applications and unresolved issues. In:
114 K. El Kadi Abderrezzak et al. / Journal of Hydro-environment Research 8 (2014) 95e114
Rhoads, L.B., Thorn, E.C. (Eds.), The Scientific Nature of Geo-
morphology. Wiley & Sons, Chichester, pp. 221e253.
Piégay, H., Aelbrecht, D., Beal, D., 2010. Restauration morphodynamique et
redynamisation de la section court-circuitée du Rhin a l’aval du barrage de
Kembs (Projets Interreg/EDF). Congrès SHF: « Environnement et Hydro-
électricité », Lyon.
Pollen, N., Simon, A., 2005. Estimating the mechanical effects of riparian
vegetation on streambank stability using a fiber bundle model. Water
Resour. Res. 41, W07025. http://dx.doi.org/10.1029/2004WR003801.
Pugh, C.A., 2008. Appendix C: Sediment transport scaling for physical
models. In: Garcı́a, M.H. (Ed.), Sedimentation Engineering: Processes,
Measurements, Modeling, and Practice, pp. 1057e1065.
Pugh, C.A., Dodge, R.A., 1991. Design of sediment models. In: Proc. 5th
Federal Interagency Sedimentation, Federal Energy Regulatory Commis-
sion, Washington, DC.
Raudkivi, A.J., 1997. Ripples on stream bed. J. Hydraulic Eng. 123, 58e64.
Recking, A., 2009. Theoretical development on the effects of changing flow
hydraulics on incipient bedload motion. Water Resour. Res. 45, W04401.
http://dx.doi.org/10.1029/2008WR006826.
Rinaldi, M., Casagli, N., 1999. Stability of stream banks formed in partially
saturated soils and effects of negative pore water pressures: the Sieve River
(Italy). Geomorphology 26, 253e277.
Rinaldi, M., Darby, S.E., 2007. Modelling river-bank erosion processes and
mass failure mechanisms: progress towards fully coupled simulations. In:
Habersack, H., Piégay, H., Rinaldi, M. (Eds.), Gravel-bed Rivers VI: From
Process Understanding to River Restoration. Elsevier, Amsterdam,
pp. 213e239.
Rodgers, M.T., Lamm, D.M., Riiff, E.H., Davinroy, R.D., 2003. Sedimentation
Study of the Middle Mississippi River at Herculaneum, Missouri River
Miles 156.3 to 149.7-Hydraulic Micro Model Investigation. U.S. Army
Corps of Engineers (USACE). Technical Report MXX.
Shen, W.H., 1971. Modeling of fluvial processes. In: Shen, W.H. (Ed.), River
Mechanics, vol. II. Water Resources Publications, Colorado State Uni-
versity, Fort Collins, Colo.
Shen, H.W. (Ed.), 1991. Movable Bed Physical Models. Kluwer Academic,
Boston.
Simon, A., Collison, A.J.C., 2002. Quantifying the mechanical and hydrologic
effects of riparian vegetation on streambank stability. Earth Surf. Process.
Land. 27, 527e546.
Simonett, S., Weitbrecht, V., 2011. Bed load transport in a physical scale
model of two merging mountain streams. In: Rowinski, Pawel, et al.
(Eds.), Experimental Methods in Hydraulic Research, Series: Geoplanet:
Earth and Planetary Sciences. Springer, Heidelberg, Berlin,
pp. 275e286.
Song, C.C.S., Yang, C.T., 1979. Modeling of river with sediment transport. In:
Proc. Special Conference on Conservation and Utilization of Water and
Energy Resources, San Francisco.
Struiksma, N., Klaassen, G.J., 1986. Scale effects in the reproduction of the
overall bed topography in river models. In: IAHR Symposium on Scale
Effects in Modelling Sediment Transport Phenomena, Toronto.
Thorne, C.R., Abt, S.R., 1993. Analysis of riverbank instability due to toe
scour and lateral erosion. Earth Surf. Process. Land. 18, 835e843.
Van Rijn, L.C., 1984. Sediment transport, Part II: Suspended load transport. J.
Hydraulic Eng. 110, 1613e1641.
Van Rijn, L.C., 1993. Principles of Sediment Transport in Rivers, Estuaries
and Coastal Seas. Aqua Publications.
Waldron, R.L., 2005. Physical Modeling of Flow and Sediment Transport
using Distorted Scale Modeling. Master Thesis. Louisiana State
University.
Wallerstein, N.P., Alonso, C.V., Bennett, S.J., Thorne, C.R., 2001. Distorted
Froude scaled flume analysis of large woody debris. Earth Surf. Process.
Land. 26, 1265e1283.
Wei, B-q., Uchijima, K., Hayakawa, H., 2001. A study on similarity laws of a
distorted river model with a movable bed. J. Hydrodyn. Ser. B 13 (1),
86e91.
Weitbrecht, V., Rüther, N., 2009. Laboratory and numerical model study on
sediment transfer processes in an expanding river reach. In: Proc. 33rd
IAHR Congress, Vancouver.
Wilcock, P.R., McArdell, B.W., 1993. Surface-based fractional transport rates:
mobilization thresholds and partial transport of sand-gravel sediment.
Water Resour. Res. 29, 1297e1312.
Woidt, J., Cox, A., Thornton, C., 2001. Evaluation of sedimentation and
erosion trends in the Sacramento river near the M&T/Llano seco pump
station. In: Proc. AGU Hydrology Days, April 2e5. Colorado State Uni-
versity, Fort Collins, CO.
Yalin, M.S., 1971. Theory of Hydraulic Models. Macmillan, London.
Young, W., Warburton, J., 1996. Principles and practice of hydraulic modelling
of braided gravel bed rivers. J. Hydrol. 35, 175e198.