JOURNAL OF HYDRAULIC RESEARCH

2024, VOL. 62, NO. 5, 461–475

https://doi.org/10.1080/00221686.2024.2401903

RESEARCH PAPER

Breaching of sand dikes: assessment of 1D and 2D numerical models against

laboratory experiments

Nathan Delpierrea , Fabian Franzinib , Pierre-Yves Gousenbourgerc , Sylvie Van Emelend , Yves Zeche and
Sandra Soares-Frazao f
a Institute of Mechanics, Materials and Civil Engineering, UCLouvain, Louvain-la-Neuve, Belgium; b Institute of Mechanics, Materials and Civil
Engineering, UCLouvain, Louvain-la-Neuve, Belgium and Fonds National de la Recherche Scientifique, Bruxelles, Belgium; c Reasearcher at
Institute of Mechanics, Materials and Civil Engineering, UCLouvain, Louvain-la-Neuve, Belgium; d Unité Génie Construction et Géomètre,
Haute Ecole Léonard de Vinci, Bruxelles, Belgium; e Institute of Mechanics, Materials and Civil Engineering, UCLouvain, Louvain-la-Neuve,
Belgium; f Institute of Mechanics, Materials and Civil Engineering, UCLouvain, Louvain-la-Neuve, Belgium

ABSTRACT ARTICLE HISTORY

Dike failures often induce highly damaging floods. Improved understanding of the inherent Received 22 September 2023
mechanisms is important to predict and prevent dike failures. This research focuses on sand dike Accepted 3 September 2024
failure by overtopping. During the experiments, the evolution of nine sections was monitored KEYWORDS
using laser-sheet while the water level in the upstream reservoir was captured using ultrasonic Dike failure; erosion
gauges. The impact of the grain size was studied by comparing a fine and a coarser sand. Even processes; laboratory studies;
though the failure mechanisms look similar in both cases, some differences could be observed in laser sheet; one-dimensional
the evolution of the geometry of the dike. Finally, the failure was simulated with 1D and 2D mod- models; two-dimensional
els. The 1D model shows good results for the initial stage of the dike evolution and in predicting models
the evolution of the width of the breach. The 2D model predicts the morphological evolution of
the dike more accurately but slightly overestimates the widening of the breach.

1. Introduction The present research focuses on the failure of non-

Dikes, or earthen embankments, play a key role in cohesive sandy dikes made of uniform material by
flood defence but also in water resource management overtopping. This type of failure has been studied
or energy production. In areas such as the north of extensively over the last decades as non-cohesive fill
Belgium and the Netherlands, millions of people are liv- dikes are the most sensitives to surface erosion. Indeed,
ing under the current sea-level. They are thus greatly Morris et al. (2009) point out that the failure is mostly
dependent on the proper functioning of these struc- led by progressive surface erosion for non-cohesive fill
tures. For the Netherlands only, a 22,000 km long net- material. Visser (1998) first described the failure pro-
work of dikes has been developed, while the country cess for fine sand dikes by clearly identifying the dif-
only counts 880 km of coastline. ferent steps of the breach widening. His theory high-
Therefore, dike failures can greatly worsen flood lights the cohesive behaviour of fine sand that is not
events as they induce important flows in areas intended always representative of non-cohesive material. On the
to be protected. Overtopping is the most common fail- other hand, Coleman et al. (2002) extended the study
ure mechanism for earthen embankments, representing to coarse non-cohesive sand dikes and highlighted the
34% of cases (Foster et al., 2000). Examples include presence of a pivot point, around which the dike rotates
the Hurricane Katrina in New Orleans (2005), where while being eroded. These theories have been further
all dike failures were observed with overtopping (Sills enhanced by Zhang et al. (2017) and Zhou et al. (2019).
et al., 2008), but also the floods that occurred in 2013 Experiments usually focus on the influence of several
along the Elbe River causing multiple failures. parameters, such as the sediment nature, dike height
In such scenarios, it is crucial for decision makers to and discharge (Schmocker & Hager, 2012), the grain
gain a better knowledge of the remaining time before size distribution (Schmocker et al., 2014), the shape and
the dike will break and of the volume of water that will headwater elevation (Müller et al., 2016) or the mate-
pass through it once water begins to overflow the dike. rial constituting an eventual pavement (Pan et al., 2015;
This information can be utilized as an input hydrograph Zhang et al., 2017).
to facilitate the creation of accurate and coherent flood As regards numerical simulations, the complexity
maps. of the ongoing physical processes and the associated

CONTACT Sandra Soares-Frazao sandra.soares-frazao@uclouvain.be

© 2024 International Association for Hydro-Environment Engineering and Research

462 N. DELPIERRE ET AL.

variability make it difficult to model dike breaching in coarse and fine uniform sands. From the experimental
a comprehensive way. Therefore, many authors still rely side, an important and interesting development is the
on parametric breach models using regression equa- extensive use of the laser light sheets, already developed
tions to estimate the evolution of the failure mechanism in prior works (Soares-Frazão et al., 2007), to precisely
(ASCE/EWRI Task Committee on Dam/Levee Breach- catch the morphological evolution of the breach in time
ing, 2011). Based on multiple case studies, laws describ- and therefore to provide valuable data for numerical
ing relations between the overtopping flows and some model validation. In addition to the high quality of the
physical parameters, like the breach width or water measured data, all the parameters required by mod-
inflow, can be derived. Other approaches can be seen as ellers are provided. Another interesting aspect of the
semi-physical as they try to incorporate the geometry experimental results is the comparison between the
of the breach, considered as a rectangle or trapezoidal behaviours of coarse and fine sand. The results show
forms, and to describe its evolution in time considering that the cohesive behaviour of fine sand leads to a
some simplifications (ASCE/EWRI Task Committee on higher peak discharge following lateral block failure.
Dam/Levee Breaching, 2011). The described processes show interesting concordances
Nevertheless, numerical models are very powerful with the observations of Coleman et al. (2002) and
tools to run predictive simulations. When models are Visser (1998), which allow the dataset to be validated.
validated by experiments, they can be used to predict Secondly, we compare this benchmark to one-
behaviour in real life and hence prevent damage or dimensional and two-dimensional finite volume models
hazard. Many numerical models have now been devel- based on the shallow-water-Exner equations aug-
oped to assess the risk of failure for dikes submitted mented with a bank-failure operator. We show that
to overtopping flows with attempts to fully solve the the one-dimensional simulation, although less accu-
physical process. Based on the idea that the breach- rate and physically less appropriate, provides acceptable
ing process is mostly dominated by erosion for non- results while using significantly less simulation time,
cohesive dikes, most of the models consist in solving and is thus a good quick first estimation. The two-
the shallow water equations combined with the Exner dimensional model used in this paper also solves the
equation. These models build on specific transport shallow-water and Exner equations and is enhanced
laws and the resolution schemes assume equilibrium with a feature first presented by Swartenbroekx et al.
bedload transport conditions. That is, the transport (2010) as a bank-failure operator. This feature allows a
rate instantaneously adapts to local flow specifications maximal slope angle to be considered for non-cohesive
(see e.g. Wu et al., 2012, Juez et al., 2014, Barzgaran sand and therefore to correctly represent the local
et al., 2019, Meurice & Soares-Frazão, 2020). Recent mass failures. The one-dimensional model considers
works consider non-equilibrium bedload transport for the morphological evolution of arbitrary cross section
dam breaching (Martínez-Aranda et al., 2023). They based on an augmented Roe approach (Franzini &
account for temporal and spatial delay of the sedi- Soares-Frazão, 2018).
ment transport rate influenced by the flow capacity. The current paper is divided in five sections.
Martínez-Aranda et al. (2021) made a specific compar- Section 2 describes the experimental setup built to run
ison between equilibrium and non-equilibrium condi- the experiments in the laboratory. Section 3 describes
tions and showed that, under appropriate calibration the measurement techniques to capture the bed level
of the parameters, the non-equilibrium approach per- evolution and the water evolution in time. Section 4
formed better in highly transient zones. comments the experimental results concerning the
In such models, the difficulties in properly breaching process and the evolution of the captured
describing the process lie mainly in the variability of the bed level at the different sections of measure. As this
material, such as mean grain diameter, compaction or paper is about experimental and numerical modelling,
water content. Many transport laws have been devel- Section 5 describes the numerical one-dimensional
oped with a particular range of applications, making and two-dimensional models, whose results will be
this choice critical for the quality of the simulation, as compared with the experimental results in Section 6.
highlighted by Van Emelen et al. (2015). Note that the Section 7 is dedicated to the conclusions.
literature provides mostly closure equations for non-
cohesive sands. This is an acceptable approach as soon
2. Experimental setup
as the dike is made of sand coarse enough to be consid-
ered as non-cohesive. However, humidity confers cohe- The breaching experiments were carried out at the
sion and may strongly alter the results. Further work Hydraulics Laboratory of the Institute of Mechanics,
should focus on considering the saturation of water Materials and Civil Engineering (iMMC) of UCLou-
inside the dike, as initiated in Delpierre et al. (2023). vain (Belgium). The set-up is illustrated in Figure 1 and
This work has two goals. First, we provide a new consists of a full-scale dike constructed with an ini-
1:1 experimental benchmark to the community, with tial pilot breach. The flume is 12 m long, 1.2 m wide
results of dike breaching by overtopping obtained with and 0.32 m high. A pump supplies water to a 2-m long,
 JOURNAL OF HYDRAULIC RESEARCH 463

G1 G2 G3 G1 G2 G3

Figure 1. Experimental setup and dimensions of the dike made of (a) coarse sand and (b) fine sand. The only difference relies on the
top of the dike. Distances are in metres.

upstream reservoir. The dike is built 2.88 m down- prevents any local influence of the measurement on the
stream from the reservoir. experiment.
Two dikes were built: one with coarse sand and Three ultrasonic probes are used to determine the
one with fine sand. Coarse sand has median diameter water level evolution at three locations as illustrated
d50 = 1.7 mm, and bed porosity  0 = 45% after com- in : just downstream of the reservoir (x = − 0.75 m),
paction. Fine sand has median diameter d50 = 0.71 mm, at the upstream toe of the dike (x = 0 m) and on the
and bed porosity  0 = 44% after compaction. Both downstream sand layer (x = 2.05 m).
sands have a specific gravity s = 2.615. According to The bed level evolution is captured at nine different
widely used formulas, the Manning friction coefficient positions thanks to three laser-sheets (Soares-Frazão
should range from 0.014 s m−1/3 to 0.0167 s m−1/3 . et al. 2007), as represented in Figure 2. The experiments
Finally, the friction coefficient, n = 0.0167 s m−1/3 , was are run in a completely dark environment: the bed level
determined from uniform flow experiments and cho- is then lit by the laser-sheets, which creates a thin vis-
sen for both sands. A dry friction angle of 85° and ible red line. The red lines are captured by a camera,
a wet friction angle of 30° were considered following as illustrated in Figure 2. We assume that the impact of
Swartenbroekx et al. (2010). refraction is negligible given that the camera is placed in
Geometrically, the dikes are 1.3 m long and use the front of the dike (Soares-Frazao et al, 2007). This effect
whole flume width (1.2 m) with 1:3 slopes, upstream is further limited as the water depth over the dike is very
and downstream. Note that the dike made with coarse small.
sand has a crest of 0.1 m length while the dike made The bed profiles are extracted from the pictures
with fine sand has no crest for mechanical reasons. An using Matlab preprocessing. First, the pictures are
initial trapezoidal breach is dug in each dike to initi- orthorectified to remove deformations due to the cam-
ate the failure process. The breach has a base width of era lenses and position. Two deformations can occur
0.03 m with a 1:1 slope, and a sand level of 0.17 m at its (see Figure 3 for a clear example on a chessboard):
centre. A sand layer of 1 m long and 0.04 m height is set barrel distortion (b) and perspective distortion (c). Sec-
at the downstream toe of the dike to avoid piping effects. ondly, the bed level lines are isolated, selecting the pixels
The upstream part of the dike is initially filled with with the highest contrast. Finally, the results are scaled
0.15 m of water. When the experiment starts, the pump using the ratio pixel/cm.
is activated to provide a constant discharge of 4 l s–1 . To avoid noisy effects due to the proximity of two
This discharge is kept constant during the whole exper- consecutive lasers, we decided to acquire the bed level
iment. at only three positions by trial. The lasers are shifted
after each trial in order to multiply the measurements:
three trials are hence needed in order to obtain the nine
3. Measurement techniques
measurements (see Table 1 for the nine positions and
The important parameters studied in the experiments their associated trial). Therefore, the repeatability of the
are the water and bed level evolution. These are cap- data acquisition protocol is crucial to assess the valid-
tured using two types of measurement techniques: ity of the measurements of the bed level evolution from
ultrasonic probes and laser-sheet. These two tech- one trial to another. Here, the validity has been assessed
niques have the advantage of being non-intrusive. This by performing three experiments where the dike was
464 N. DELPIERRE ET AL.

Laser sheet projections

Z L1
L2
L3

Figure 2. Laser-sheet and camera measurements: (a) indication of the three series of measurements per test (see Table 1 for the
exact positions), (b) result for one series of measurements: the laser appears as a white line.

flush by water. During each experiment, we measured by the additional cohesion generated by the finer sand
the bed level at three different locations. We then com- matrix.
pared the results in Figure 4. The experiments were In stage II, the breach widens by erosion and
performed with fine sand and coarse sand (for a total the width of the crest decreases until it vanished at
of six trials). The results show visually strong similari- t = t 2 . Stage II occurs at t 2 = 60 s for coarse sand and
ties and enhance the confidence in the repeatability of t 2 = 57 s for fine sand. The inflow grows.
the protocol. In stage III, the deepening and widening continues
until the dike in the breach is washed out down to the
base of the dike. Stage III appears to be a steady state
4. Results in our experiment. It occurs at time t = t 3 = 600 s for
This section presents the results from the experiments. both grain sizes.
It is divided in three parts. First, the general process One important difference between the two grain
of failure is presented. Second, the breach evolution sizes is the impact of seepage. As observed in Figure 5, at
is studied more closely. Third, the breach discharge is the toe of the coarse sand dike, a flow is observed. This
estimated using data on geometry and water level evo- flow is due to the seepage through the dike and is not
lution. It is shown, in this section, that the data recorded present in the fine sand dike (Figure 6). However, the
during this experiment respects the expectations and main failure mechanism is still the overtopping over the
follows the well-known models of Visser (1998) and dike through the breach. We will see in the next section
Coleman et al. (2002). Therefore, the dataset can be that both sand types result in the same breach evolution
considered as valid and is shared to the community as a process.
new benchmark to study the breaching and the failure
of sand dikes.
4.2. Evolution of the breach geometry
Figure 7a and b show the evolution of the dike along
4.1. Breaching process
its centreline, in the direction of the flow. The process
The failure mechanism observed for both types of sand clearly shows similarities with the observations pre-
confirms the three first stages of dike erosion observed sented by Coleman et al. (2002, p. 833, figure 6) as
by Visser (1998, §2). Figures 5 (coarse sand) and 6 (fine a pivot point clearly appears at the downstream toe
sand) show pictures of those stages appearing in the in both cases. Coleman characterizes this particular
present experiments. juncture as a pivot due to the continuous erosion that
In stage I, we observe a steepening of the breach causes the dike to revolve around it over time. Unlike
slopes until a critical angle is reached at t = t 1 . This Coleman’s observations, the pivot’s position in both
critical angle is reached at t 1 = 22 s for coarse sand scenarios is consistently located beyond the initial con-
and t 1 = 27 s for fine sand. This might be explained struction of the dike toe. This occurrence is likely
 JOURNAL OF HYDRAULIC RESEARCH 465

Table 1. Cross-sections measurements starting from the

upstream foot of the dike
Cross-section Location Trial
C1 0.60 m A
C2 0.65 m B
C3 0.70 m C
C4 0.80 m A
C5 0.90 m B
C6 1.00 m C
C7 1.10 m A
C8 1.20 m B
C9 1.30 m C

by the laser. Another hypothesis, although less proba-

ble, could be the influence of optical distortion due to
the free surface, whose depth is more important in the
first steps of overtopping. After 15 s, the dike is eroded
with a decreasing erosion rate. The coarse sand dike is
initially eroded faster than the fine one. However, the
fine sand dike then catches up this delay and they both
stabilize at the same level.
Both dikes present similar behaviour to Visser’s
observations (1998) when considering the widening of
the breach, for different cross-sections (Figure 8). Dur-
ing stage I, the channel width at the crest does not
change much, however a narrow channel is dug on
the downstream slope. This channel widens slower in
the case of the fine sand dike, due to a more cohesive
behaviour than the coarse sand one. Conversely, the
vertical digging speed seems similar for both cases. Sub-
sequently, in stage II and up to 120 s, the breach and
the downstream channel undergo progressive widen-
ing and deepening, until the eroded width of the breach
aligns with that of the downstream channel. Finally,
between 120 s and 600 s, corresponding to stage III,
the geometry changes much more slowly before reach-
ing an equilibrium or steady state. It is interesting to
notice that, except for the section in the downstream
slope at 600 s for the fine sand, all cross-sections are
Figure 3. The initial acquired image (a) is subject to distortion.
It is important to post-process the image. The figure shows the symmetrical, which attests the good conditions and the
image after correction of the barrel effect (b) and the perspec- repeatability of the experiments. Globally, the coarse
tive distortion (c). sand dike presents a smoother evolution than the fine
sand dike. This can be explained by an apparent cohe-
attributed to the presence of sand at the downstream toe sion in the fine sand that creates a burst behaviour:
of the dike, as previously described. The presence of an small blocks of fine sand sliding together.
erodible sand layer downstream of the dike is however
more representative of field situations. A final slope is
4.3. Evolution of the water level and discharge
obtained in both cases at around 600 s, corresponding
to a steady state. The fine sand erosion ends up with As mentioned in the introduction, it is of core interest to
an adverse slope but also with a steeper slope than the predict the measured discharge across the breach. The
coarse sand dike. discharge is directly related to the collapse of the dike. It
Figure 7c shows the dike level evolution for the can be deduced from the input discharge (Qin = 4 l s–1 )
cross-section located at the dike crest (x = 0.65 m) but combined with the evolution of the water level in the
similar conclusions can be drawn when studying other reservoir (indeed, the discharge Q through the breach
cross-sections of the dike. For the first 15 s, the bed level corresponds to the evolution of the water level zw in the
increases. One hypothesis to explain this increase is that reservoir multiplied by the horizontal surface Ar of the
a mixture of water and sediment is flowing rather than reservoir, i.e. Q = Qin − Ar dzdtw . This estimation does
clear water, and the exact bed elevation cannot be seen not consider the seepage that was observed in the coarse
466 N. DELPIERRE ET AL.

Coarse Sand Fine Sand

200 200

150 150

Bed level [mm]

Bed level [mm]

100 100

50 50

0 0
0 100 200 300 0 100 200 300
Time [s] Time [s]

x = 0.6 (m) x = 0.8 (m) x = 1.1 (m)

Figure 4. The repeatability of the protocol is assessed. The bed level evolution looks similar for three different trials (solid, dashed,
dash-dotted) at three different positions (blue, grey and black) for both coarse and fine sand.

case, that is assumed to be negligible compared to the Both models are based on the shallow water
ongoing overtopping process. equations (describing water mass and momentum
Figure 9 illustrates the temporal evolution of (i) conservation) augmented with the Exner equations (for
the water level captured by the ultrasonic probe G1 sediment mass conservation). The equations are solved
(Figure 1), and (ii) the temporal evolution of the dis- using an explicit finite-volume numerical scheme
charge in the breach deduced from the measurements Uin+1 = f(Uin , F ∗ ), where the time is represented by the
of the ultrasonic probe G1. superscript n and the exchanges between volumes at the
The water level in the reservoir is expected to grow interfaces are represented by the fluxes F ∗ . After each
until the collapse occurs. The peak of water level occurs temporal update, the bank stability conditions of the
around 41 s (coarse sand) and 46.5 s (fine sand) with dike are verified. If the conditions are no more satisfied,
a maximum level measured at.0.199 m (coarse sand) a bank-failure operator is called to solve the collapse
and 0.203 m (fine sand). The difference between the two (Swartenbroekx et al., 2010). Then, a new time step
sands can be explained by seepage through the dike for can be run on a stabilized dike. The conserved values,
the coarse sand, and by an easier initial erosion due to the numerical schemes for fluxes estimation and the
a lower apparent cohesion. Conversely, the final water bank-failure operator will be explicitly defined for each
level ends up being higher for coarse sand than for fine model in the next subsections. Then, the models will
sand: the final slope of the collapsed dike is steeper for be compared and validated against experimental data
fine sand (as shown in the profiles at Figure 7), which in Section 6.
explains that the final water level will be lower. This also
reflects the fact that coarse sand requires higher shear
stresses to be eroded, when cohesion effects do not play 5.1. One-dimensional model
any strengthening effect any more, as it is the case at the The one-dimensional model considered here was
end of the experiment. developed by Franzini and Soares-Frazão (2018), based
Accordingly, the water discharge is expected to on the two-dimensional model of Murillo and García-
evolve similarly to the water level. Indeed, the peak of Navarro (2010) and including a concept of “energy bal-
water discharge is also delayed between coarse and fine ance” as defined in Murillo and Garcia-Navarro (2014)
sand. It occurs earlier for coarse sand (64 s) than for or in Franzini and Soares-Frazão (2016).
fine sand (67 s); their respective peak value is similar to In this model, the one-dimensional canal is divided
the water level peak: less discharge for the coarse sand into sections of small width. The flow is reduced to
(0.009 m3 s–1 ) than for the fine sand (0.0096 m3 s–1 ). moving in only one direction (the axis of the flume) and
The delay between the highest water level and the high- the values of interest are described as averaged values
est discharge is approximately the same for both sand over the considered cross-section (Figure 10). These
size, though (respectively 20.5 s and 23 s). After the values are the mass of water represented by the area of
peak, the slopes are very similar. water A, the discharge Q, and the mass of sediments
represented by the area Ab in the cross-section. These
areas can evolve with time not only in depth but also
5. Numerical models
in width, which permits the erosion of the dike to be
This section presents two numerical models: a sim- captured.
ple unidimensional finite-volume model and a (more More specifically, the water mass evolution only
complex) bidimensional finite-volume model. depends on the mass exchange at the cell boundaries,
 JOURNAL OF HYDRAULIC RESEARCH 467

Figure 5. The failure evolution of the breach for coarse sand

follows the three first stages of Visser (1998). Figure 6. The failure evolution of the breach for fine sand fol-
lows the three first stages of Visser (1998).

just like the sediment transport. The closure equation

required for the latter is the Meyer-Peter & Müller
formula (1948), for which the range of application is for initial sediment motion (here the value is set at
mostly the non-cohesive particles moving with not 0.047).
much of suspension. The sediments discharge Qs is The evolution of the discharge depends on hydro-
hence: static pressure, the area of sediment evolution, the
longitudinal component of lateral pressure (in case of
  1.5
R Sf expansion or constriction of the flume), and the friction
Qs = 8B g(s − 1)d50 3 − θcr (1) slope. Together they define the exchange of momentum
(s − 1)d50
between adjacent cells that conduct the discharge evo-
where s is the specific gravity of the sediments, d50 is lution. Applying these constraints to the shallow-water
the bed sediment median diameter, R is the hydraulic equations, the temporal dynamics of the conserved
radius (defined as R = A/P), and θ cr is the shear stress variables U = [A, Q, Ab ]T can be simplified under the
468 N. DELPIERRE ET AL.

Bs is the ratio to determine the discretization of the area

of sediment evolution as zb = Ab /Bs (here equal to
A/h, Figure 11). S are the source terms (lateral pressure
and friction):
Coarse sand ⎡ ⎤
0
Bed level [m]

0.2 S = ⎣gI2 − gASf ⎦ (5)

0

For a rigorous definition of the matrices U, F, H and S,

as well as a more complete explanation of the model,
0.0
0.5 1.0 we refer the reader to Franzini and Soares-Frazão
(2018).
Length [m]
Equation (2) is numerically solved via a finite volume
Fine sand scheme applied to this 1D formulation over computa-
tional cells of length dx centred on each cross-section.
Bed level [m]

0.2 Then the (temporal) explicit resolution scheme is given

by:
t
U in+1 = U ni − F ∗L − F ∗R (6)
 x i+1/2 i−1/2
0.0
where the fluxes F ∗R ∗L
i−1/2 and F i+1/2 are the lateral-
0.5 1.0
ized contributions, respectively from the upstream and
Length [m]
downstream sections calculated using the augmented
Bed level measurements Roe scheme with energy balance (Franzini & Soares-
0.2 Frazão, 2018). The source terms (topography and fric-
Bed level [m]

tion) are directly integrated to the fluxes yielding

lateralized values (i.e. values that are different for the
0.1 left and right cells, the difference being the result of the
source term).
0.0 The area of erosion/deposition, Ab , is distributed
0 200 400 600 along the section proportionally to the local water
depth resulting in a non-uniform erosion of the cross-
Time [s]
section. No erosion occurs if the water depth is equal
Figure 7. Validation of the theory of Coleman et al. (2002) Top: to zero. The classical sediment transport theory links
evolution of the bed level along the centreline of the flume: (a) the shear stress at the bottom of the flow to the water
coarse sand, (b) fine sand. Bottom: breach level evolution at the depth. Therefore, linking the sediment transport to the
crest. water depth distribution along the section seems to be
a reasonable assumption.
following equation in vector form: Last but not least, a tilting bank failure operator
is also included. It is activated after each time step
∂U ∂F ∂U
+ +H =S (2) if the stability conditions are not satisfied anymore
∂t ∂x ∂x
(with respect to the dry and wet critical friction angles
where F groups the momentum fluxes and hydrostatic given in Section 2). The general operation of the pro-
pressure contributions: cess is illustrated on Figure 11: section BC is initially
⎡ ⎤ steeper than the friction angle, therefore it is unsta-
Q
⎢ 2 ⎥ ble and is tilted to its stability angle. Segment B’C’ is
F = ⎣ QA + gI1 ⎦ (3)
Qs now stable, but segment AB’ is now unstable and will
1−ε0 be tilted in the next operation. Unlike in Franzini and
where Q is the discharge, A is the area of water, gI1 is Soares-Frazão (2018), the failure operator is here used
the hydrostatic pressure, Qs the sediment discharge both transversally and longitudinally to improve the
and ε0 the sediment’s porosity. H ∂U/∂x captures the capture of sediment mass failures that occur in both
impact of the morphological variation on the momen- directions.
tum, where H is the following matrix:
⎡ ⎤ 5.2. Two-dimensional model
0 0 0
H = ⎣0 0 gA Bs
⎦ (4) The 2D finite-volume model solves the shallow water
0 0 0 and Exner equations on unstructured triangular meshes.
 JOURNAL OF HYDRAULIC RESEARCH 469

Figure 8. Evolution of the width of the breach (left) along time (experiments) and of two cross-sections: at the crest (x = 0.65 m,
laser-sheet C2) with coarse sand (a) and with fine sand (b); and in the slope (x = 0.9 m, laser-sheet C5) with coarse sand (c) and fine
sand (d). The stages identified in Visser’s theory also appear here.

Figure 10. Bank failure operator (Franzini et al, 2018).

Figure 9. Evolution of the reservoir water level (from gauge G1,

dashed line) and of the water discharge (solid line). The figure
below is a zoom between 0 s and 100 s of the figure above.

The flow is now free to move in two directions and

the fluxes are evaluated using a modified LHLLC
scheme (Soares Frazao & Zech, 2011). Like in the
unidimensional model, the sediment transports are also
computed using the Meyer-Peter and Müller (1948)
formula (1).
The conserved variables are now the water depth h,
the discharge qx in the direction of the flow and qy in the Figure 11. Discretization of the erosion.
470 N. DELPIERRE ET AL.

flat perpendicular direction, and the sediments level zb , of 363 cells, evenly distributed along the channel. The
i.e. U = [h, qx , qy , zb ]T . The temporal dynamics can be 2D-mesh is composed of approximately 200,000 ele-
described in vector form as: ments with the smallest ones located at the crest of the
dike with an average edge length of 2.5 mm. Such a
∂U ∂F ∂G
+ + =S (7) refined mesh is needed to capture the widening of the
∂t ∂x ∂y breach better as it has a tremendous impact on the bank
⎡ ⎤
uh failure operator. The codes are developed in C++ and
⎢u2 h + gh2 /2⎥ the 2D code is parallelized with OpenMP. A substantial
F=⎢⎣
⎥
⎦ (8)
uvh difference between the two models lies in the computa-
qsx /(1 − e0 ) tional time and the computational resources needed to
⎡ ⎤ obtain a prediction. It is clear that running a 1D-model
vh
⎢ ⎥ on hundreds of cells is faster and less consuming than
uvh
G=⎢⎣v2 h + gh2 /2⎦
⎥ (9) running a 2D-model on hundreds of thousands of cells.
While the 1D model took less than an hour to simulate
qsy /(1 − ε0 )
⎡ ⎤ the failure, the 2D models took several days. This dif-
0 ference of computation time is balanced by the amount
⎢ ∂zb ⎥
⎢gh − ∂x − Sfx ⎥ of data that we can retrieve from each model. While the
S=⎢⎢ ∂z
⎥
⎥ (10) 1D model is only able to provide information regard-
⎣gh − ∂yb − Sfy ⎦ ing the discharge and the area at a particular x-position,
0 the 2D model can provide precise information regard-
ing the (e.g.) velocities at very local scale, in the x–y
where vectors F and G group the momentum fluxes and
directions. Even if the erosion process is mostly active in
hydrostatic pressure contributions and variation in the
the flow direction, taking the y-direction processes into
x-axis and y-axis directions respectively, also consider-
account is an asset for the 2D model when evaluating
ing transverse fluxes; S are the source terms considering
the breach lateral evolution. The precise knowledge of
slope and friction effects; u is the velocity component
the water depth, directly linked to the shear stress, in the
along the x-axis; v the velocity component along the y-
full area is also of utmost importance when looking for
axis; zb the bed level; qsx the sediment transport along
an optimal design for a given dike. Considering these
the x-axis; qsy the sediment transport along the y-axis;
points, we expect the 2D model to perform better than
Sfx and Sfy the friction slope along the x-axis and the
the 1D model for predicting the dike breaching process.
y-axis, respectively.
The state of the flow is captured at t0 = 0 s, t1 = 46 s,
The finite-volume scheme to numerically solve
and t2 = 120 s.
Equation (4) is as follows:
3
t 6.1. Comparison with the one-dimensional model
U in+1 = U ni − Rj−1 F ∗ Lj + Si t (11)
Ωi j=1 Figures 12–14 show the results obtained using the uni-
dimensional model to simulate the dike breaching with
where Ωi is the cell area and Lj is the interface length; coarse and fine sand. Figures are presented in similar
the fluxes at the interfaces F ∗ are represented consider- way to Section 4 to ease the comparison by the reader:
ing normal and tangential directions obtained through results in Figures 13 and 14 can confirm the theory of
the rotation matrix R to be expressed in a local coor- Visser (1998) and Coleman et al. (2002). For compar-
dinates system attached to the interface, and whose ison purposes, Figure 14 illustrates the water level and
components are computed from the neighbouring cells discharge evolution over time.
using the LHLLC scheme (Soares Frazao & Zech, 2011). Once again, one can observe that the failure process
When the local slope of a cell increases excessively, can be divided into three parts: (i) the first 50 s (illus-
the failure of the bank is simulated by a bank-failure trated here by the results at 46 s), (ii) between 50 and
operator (Swartenbroekx et al., 2010) activated under 120 s (illustrated here by the results at 120 s), and (iii)
the same conditions as in the unidimensional model. between 120 s and 600 s. The 1D model is able to cap-
As the slope of a cell exceeds the considered angle of ture the widening and deepening of the breach during
repose, the algorithm is used to distribute the slope of the first phase as observed in Figure 12 (the solid lines
adjacent cells and check their proper balance. represent the results at 46 s). This phase ends approxi-
mately at the peak of the measured outflow discharge,
that occurs after the peak water-level in the reservoir
6. Discussion of the numerical simulations and
(Figure 14).
the experiments
During the second phase, after the maximum water
In this section, the aforementioned models will be com- level has been reached in the reservoir, erosion accel-
pared to experimental data. The 1D-mesh is composed erates, resulting in an increase of the breached section
 JOURNAL OF HYDRAULIC RESEARCH 471

Coarse Sand (a) (c)

0.5

Bed level [m]

Bed level [m]

0.2 0.2

Width [m]
0.0
0.1 0.1

–0.5 0.0 0.0

0.6 0.8 1.0 –0.25 0.00 0.25 –0.25 0.00 0.25
Length [m] Width [m] Width [m]
Fine Sand (b) (d)
0.5

Bed level [m]

Bed level [m]
0.2 0.2
Width [m]

0.0
0.1 0.1

–0.5 0.0 0.0

0.6 0.8 1.0 –0.25 0.00 0.25 –0.25 0.00 0.25
Length [m] Width [m] Width [m]

Figure 12. Evolution of the width of the breach (left) along time (1D model) and of two cross-sections: at the crest (x = 0.65 m,
laser-sheet C2) with coarse sand (a) and with fine sand (b); and in the slope (x = 0.9 m, laser-sheet C5) with coarse sand (c) and fine
sand (d). The 1D predictive model (black) captures well the widening of the breach but fails at estimating its depth. Experimental
results are presented in grey.

fixed bottom of the channel. Consequently, the com-

puted peak discharge at the breach is much higher than
the experimental one because of the faster erosion of the
Fine sand breach, but ends up at almost the same rate. In contrast,
the widening is well estimated even if the breach banks
0.2
Bed level [m]

are steeper than the real ones.

It can also be observed that the cross-sections
0.1 present a triangular shape in the simulation but not
in the experiment. This difference in shape is created
0.0 by the bank failure operator: the tilting of the bank is
0.25 0.50 0.75 1.00 so important that both banks meet in the middle of
Length [m] the breach. Later, the non-uniform erosion increases
Coarse sand this triangular shape, as the erosion is higher in deeper
water.
0.2 To sum up, the unidimensional model fails at pre-
Bed level [m]

dicting the erosion of the dike in that it overesti-

mates the peak outflow discharge and the erosion
0.1
depth. However, the prediction of the widening of
the breach is satisfying. We can therefore consider
0.0 that one-dimensional model is useful when being used
0.25 0.50 0.75 1.00 as first approach for the design of breach-induced
Length [m]
floods.
Figure 13. Evolution of the bed level along the centreline of the
flume (1D model) for coarse sand (above) and fine sand (below). 6.2. Comparison with the two-dimensional model
As the time lasts, the numerical results (black) differ more and
more from experiments (grey) because the 1D model fails at Figures 15–17 show the results obtained using the two-
predicting the digging of the flow. dimensional simulation.
Similarly to the one-dimensional case, the failure
process can be divided into three parts correspond-
and consequently a decrease in the water level in the ing to the stages previously described: (i) the first 50 s
reservoir as the inflow discharge is not sufficient any (illustrated here by the results at 46 s), (ii) between
more to maintain the level constant (Figures 12 and 13). 50 and 120 s (illustrated here by the results at 120 s),
The erosion process computed in the 1D model is over- and (iii) between 120 s and 600 s. This model gives a
estimated, leading to an overestimation of the peak good prediction of the overall evolution of the breach.
discharge and a lower water level in the reservoir. The The evolution of the bottom profile is relatively close
computed erosion is such that it ends up reaching the to what has been observed during the experiment. In
472 N. DELPIERRE ET AL.

Coarse sand Fine sand

0.25 15 0.25 15
0.20 0.20

Discharge [l s–1]

Discharge [l s–1]
Water level [m]

Water level [m]

(Solid line)

(Solid line)
10 10
(Dashed)

(Dashed)
0.15 0.15

0.10 0.10
5 5
0.05 0.05

0.00 0 0.00 0
0 200 400 600 0 200 400 600
Time [s] Time [s]
0.25 15 0.25 15

Discharge [l s–1]

Water level [m]

Discharge [l s–1]
Water level [m]

(Solid line)

(Solid line)
(Dashed)
0.20 10 0.20 10
(Dashed)

0.15 5 0.15 5

0.10 0 0.10 0
0 50 100 150 0 50 100 150
Time [s] Time [s]

Figure 14. Evolution (1D model) of the water level in the reservoir (dashed lines) and discharge in the breach (solid lines) for coarse
sand (left) and fine sand (right). The model (black) detects the peak correctly but fails at estimating its actual value (grey). The graphs
below are a zoom of the graphs above.

general, Visser’s (1998) prediction is well reproduced:

during the first phase, the model estimates correctly the
widening of the breach although the model yields less
Fine sand symmetrical results than the observed ones (Figure 15).
This could be attributed to the bank failure operator
0.2 that is activated every 50 time steps. Indeed, it was
Bed level [m]

shown in Swartenbroekx et al. (2010) that a converged

0.1
solution can be obtained with the bank failure opera-
tor, but at higher computational costs. Therefore, the
authors recommended to apply it for a limited number
0.0 of iterations within each time step, yielding satisfactory
3.00 3.25 3.50 3.75 4.00
but slightly less accurate results.
Length [m]
During the second phase, the breach widens more
Coarse sand than in the reality. However, the breach level evolution
is correctly captured (Figure 16), as theorised by Cole-
0.2 man et al. (2002), and the model is able to predict the
Bed level [m]

breach shape with the local constriction.

The discharge and reservoir water level are also well
0.1
predicted (Figure 17) despite a slight overestimation of
the peak of discharge. This overestimation explains that
0.0 the model widens the breach too fast.
3.00 3.25 3.50 3.75 4.00 Finally, Figure 18 compares the results obtained with
Length [m]
1D and 2D models. It can be observed that the 1D
Figure 15. Evolution of the bed level along the centreline of model is not able to capture the breach constriction.
the flume (2D model) for coarse sand (above) and fine sand This is not surprising since no lateral forces are really
(below). Compared to Figure 12, the numerical results (black) applied on the banks, in the 1D model. This con-
capture better the deepening of the sand level measured by firms the importance of the y-direction flow in the
experiments (grey). erosive process, as explained previously. However, the
width at the crest (after 0.6 m) is well predicted in both
cases.
 JOURNAL OF HYDRAULIC RESEARCH 473

Coarse Sand (a) (c)

0.5

Bed level [m]

0.2

Bed level [m]

0.2

Width [m]
0.0
0.1 0.1

–0.5 0.0 0.0

0.6 0.8 1.0 –0.25 0.00 0.25 –0.25 0.00 0.25
Length [m] Width [m] Width [m]
Fine Sand (b) (d)
0.5

Bed level [m]

Bed level [m]
0.2 0.2
Width [m]

0.0
0.1 0.1

–0.5 0.0 0.0

0.6 0.8 1.0 –0.25 0.00 0.25 –0.25 0.00 0.25
Length [m] Width [m] Width [m]

Figure 16. Evolution of the width of the breach (left) along time (2D model) and of two cross-section: at the crest (x = 0.65 m, laser-
sheet C2) with coarse sand (a) and with fine sand (b); and in the slope (x = 0.9 m, laser-sheet C5) with coarse sand (c) and fine sand
(d). The 2D predictive model (black) captures well the depth of the breach but fails at estimating its width, which is the opposite than
the 1D model (Figure 11). Experimental results are presented in grey.

Exp
Num

Exp
Discharge (m3 s–1)

Num

Figure 18. Comparison of the breach geometry after 120 s. The

colour bar represents the bed level, for (a) the 1D model and (b)
Figure 17. Evolution of (a) the water level in the reservoir and the 2D model.
(b) the discharge in the breach for coarse sand as calculated
using the 2D model.
widening and deepening of breach in non-cohesive fill
dikes (Coleman et al., 2002; Visser, 1998).
7. Conclusions
First, the fine sand showed an apparent cohesion
Small-scale experiments of dike breaching have been leading to steep slopes, block mass failure and there-
performed using fine and coarse grain sizes. The overall fore irregular widening of the breach. Conversely, the
process demonstrated strong consistency with tradi- width variation was much smoother for the coarse sand
tional observations found in the literature regarding the dike. Second, the fine sand dike failure exhibits a higher
474 N. DELPIERRE ET AL.

and later peak for both the water level in the upstream Notation
reservoir and the breach discharge. It was also observed
A area of water (m2 )
that the delay between the reservoir level peak and the
Ar planar area of the reservoir (m2 )
peak discharge were similar for both sand sizes.
Ab area of sediments (m2 )
Then, the experimental results have been compared
d50 median sediment diameter (mm)
to numerical simulations using 2D and 1D models.
F, F ∗ vector of fluxes
Despite its simplicity, the simple 1D model showed
Gi measurement gauge
good results for the initial stage of the dike failure for all
G fluxes in the y-direction
the monitored parameters even though the width evo-
H alternative flux matrix
lution along the breach was straighter than the exper-
h water depth (m)
iment. However, the results after the peak discharge
Q water discharge (m3 s−1 ), water discharge per
were not as good, as the 1D model tends to overesti-
metre (m2 s−1 )
mate the erosion rate. Nevertheless, the width evolution
qx , qy water discharge per unit width in the x- and
along the breach was correctly estimated even if pre-
y-directions (m3 s−1 )
senting bank slopes that were too straight. However, the
Qs sediment transport (m3 s−1 ), sediment trans-
model was not able to fully evaluate the evolution of
port per metre (m2 s−1 )
the dike in all directions. The fact that the 1D model
qsx , qsy sediment discharge per unit width in the x-
needs less computational cost and still provides inter-
and y-directions (m2 s−1 ).
esting results for the breach width makes it very useful
R hydraulic radius (m)
to determine a first order of magnitude for failure time
R rotation matrix
and discharge in initial stages of design. For further
S vector of source terms
work, it would be interesting to study other ways to dis-
s specific gravity of the sediments (−)
tribute the erosion across a transversal section, see for
Sf friction slope (−)
instance Martínez-Aranda et al. (2019).
T, ti time (s)
The 2D model showed good results in predicting the
U vector of conserved variables
overall evolution of the dike breach. The model was
U ni conserved variables in the element i at time
more accurate than the 1D model in predicting the peak
step n
discharge and water level peak, which are ultimately the
u velocity component along the x-axis
quantities of interest.
v velocity component along the y-axis
The results presented here show the ability of the
x direction of the flume
models to complement each other to get closer to real-
y direction of the width of the flume
ity. However, it should be kept in mind that these mod-
z depth direction
els, as well as most of the models in the literature, are
zb bed level (m)
limited to the evaluation of breaches in dikes made of
zw water level (m)
non-cohesive materials. In the future, efforts must be
0 bed porosity (−)
undertaken to develop numerical models capable of
θ cr critical shear stress (−)
reproducing cohesive effects, due to water content or to
Ω cell area (m2 )
the nature of the materials themselves, through a more
sophisticated consideration of mechanics. The aim is to
be able to assess the resistance to overtopping flows of ORCID
real dikes, more often made of cohesive materials. Sandra Soares-Frazao http://orcid.org/0000-0001-5088-
0278

Acknowledgements References
The authors would like to thank Charles Descantons and ASCE/EWRI Task Committee on Dam/Levee Breaching.
Rémy Dujardin (2014) for their help in designing and run- (2011). Earthen embankment breaching. Journal of
ning the experiments. Hydraulic Engineering 137(12), 1549–1564.
Barzgaran, M., Mahdizadeh, H., & Sharifi, S. (2019). Numeri-
cal simulation of bedload sediment transport with the abil-
ity to model wet/dry interfaces using an augmented Rie-
Data availability statement mann solver. Journal of Hydroinformatics 21(5), 834–850.
https://doi.org/10.2166/hydro.2019.046
The dataset is publicly available in the following repos- Coleman, S. E., Andrews, D. P., & Webby, M. G. (2002). Over-
itory: https://doi.org/10.14428/DVN/R7VH3G topping breaching of noncohesive homogeneous embank-
ments. Journal of Hydraulic Engineering, 128(9), 829–838.
Delpierre, N., Rattez, H., & Soares-Frazao, S. (2023). Finite-
volume coupled surface-subsurface flow modelling in
Disclosure statement earth dikes. Journal of Hydraulic Research, 61(5), 754–763.
No potential conflict of interest was reported by the author(s). https://doi.org/10.1080/00221686.2023.2246936
 JOURNAL OF HYDRAULIC RESEARCH 475

Foster, M., Fell, R., & Spannagle, M. (2000). The statistics strengthened by vegetated HPTRM system. Ocean Engi-
of embankment dam failures and accidents. Canadian neering, 96, 139–148. https://doi.org/10.1016/j.oceaneng.
Geotechnical Journal, 37(5), 1000–1024. https://doi.org/ 2014.12.012
10.1139/t00-030 Schmocker, L., Frank, P.-J., & Hager, W. (2014). Overtop-
Franzini, F., & Soares-Frazão, S. (2016). Efficiency and accu- ping dike-breach: effect of grain size distribution. Jour-
racy of lateralized HLL, HLLS and augmented roe’s scheme nal of Hydraulic Research, 52(4), 559–564. https://doi.org/
with energy balance for river flows in irregular chan- 10.1080/00221686.2013.878403
nels. Applied Mathematical Modelling, 40(17), 7427–7446. Schmocker, L., & Hager, W. H. (2012). Plane dike-breach
https://doi.org/10.1016/j.apm.2016.02.007 due to overtopping: effects of sediment, dike height and
Franzini, F., & Soares-Frazão, S. (2018). Coupled finite- discharge.. Journal of Hydraulic Research, 50(6), 576–586.
volume scheme with adapted augmented Roe scheme https://doi.org/10.1080/00221686.2012.713034
for simulating morphological evolution in arbitrary Sills, G. L., Vroman, N. D., Wahl, R. E., & Schwanz, N. T.
cross-sections. Journal of Hydroinformatics, 20(5), (2008). Overview of New Orleans levee failures: lessons
jh2018109. https://doi.org/10.2166/hydro.2018.109 learned and their impact on national levee design and
Juez, C., Murillo, J., & García-Navarro, P. (2014). A 2D assessment. Journal of Geotechnical and Geoenvironmen-
weakly-coupled and efficient numerical model for tran- tal Engineering, 134(5), 556–565. https://doi.org/10.1061/
sient shallow flow and movable bed. Advances in Water (ASCE)1090-0241(2008)134:5(556)
Resources 71, 93–109. https://doi.org/10.1016/j.advwatres. Soares Frazao, S., & Zech, Y. (2011). HLLC scheme
2014.05.014 with novel wave-speed estimators appropriate for two-
Martínez-Aranda, S., Fernández-Pato, J., & García-Navarro, dimensional shallow-water flow on erodible bed. Interna-
P. (2023). Non-Equilibrium bedload transport model tional Journal for Numerical Methods in Fluids, 66(9), 1019.
applied to erosive overtopping dambreach. Water, 15(17), https://doi.org/10.1002/fld.2300
17. https://doi.org/10.3390/w15173094 Soares-Frazão, S., Le Grelle, N., Spinewine, B., & Zech, Y.
Martínez-Aranda, S., Meurice, R., Soares-Frazão, S., & (2007). Dam-break induced morphological changes in a
García-Navarro, P. (2021). Comparative analysis of HLLC- channel with uniform sediments: measurements by a laser-
and roe-based models for the simulation of a Dam-break sheet imaging technique. Journal of Hydraulic Research,
flow in an erodible channel with a 90◦ bend. Water, 13(13), 45(sup1), 87–95. https://doi.org/10.1080/00221686.2007.
13. https://doi.org/10.3390/w13131840 9521835
Martínez-Aranda, S., Murillo, J., & García-Navarro, P. (2019). Swartenbroekx, C., Soares-Frazão, S., Staquet, R., & Zech,
A 1D numerical model for the simulation of unsteady and Y. (2010). Two-dimensional operator for bank failures
highly erosive flows in rivers. Computers & Fluids, 181, induced by water-level rise in dam-break flows. Jour-
8–34. https://doi.org/10.1016/j.compfluid.2019.01.011 nal of Hydraulic Research, 48(3), 302–314. https://doi.org/
Meurice, R., & Soares-Frazão, S. (2020). A 2D HLL-based 10.1080/00221686.2010.481856
weakly coupled model for transient flows on mobile Van Emelen, S., Zech, Y., & Soares-Frazão, S. (2015).
beds. Journal of Hydroinformatics 22(5), 1351–1369. Impact of sediment transport formulations on breaching
https://doi.org/10.2166/hydro.2020.033 modelling. Journal of Hydraulic Research, 53(1), 60–72.
Meyer-Peter, E., & Müller, R. (1948). Formulas for Bed-Load https://doi.org/10.1080/00221686.2014.939111
transport. 3, 39–64. Visser, P. J. (1998). Breach growth in sand-dikes. https://reposi-
Morris, M., Hassan, M., Kortenhaus, A., & Visser, P. (2009). tory.tudelft.nl/islandora/object/uuid%3A3721e23b-d34c-
Breaching Processes: A state of the art review. FLOOD site 45a9-8b36-e5930462d8e2.
Project Report, T06-06-03. Wu, W., Marsooli, R., & He, Z. (2012). Depth-averaged two-
Murillo, J., & Garcia-Navarro, P. (2014). Accurate numeri- dimensional model of unsteady flow and sediment trans-
cal modeling of 1D flow in channels with arbitrary shape. port due to noncohesive embankment break/breaching.
Application of the energy balanced property. Journal Journal of Hydraulic Engineering 138(6), 503–516. https://
of Computational Physics, 260, 222–248. https://doi.org/ doi.org/10.1061/(ASCE)HY.1943-7900.0000546
10.1016/j.jcp.2013.12.040 Zhang, Y., Chen, G., Hu, J., Chen, X., Yang, W., Tao, A., &
Murillo, J., & García-Navarro, P. (2010). An exner-based Zheng, J. (2017). Experimental study on mechanism of
coupled model for two-dimensional transient flow over sea-dike failure due to wave overtopping. Applied Ocean
erodible bed. Journal of Computational Physics, 229(23), Research, 68, 171–181. https://doi.org/10.1016/j.apor.2017.
8704–8732. https://doi.org/10.1016/j.jcp.2010.08.006 08.009
Müller, C., Frank, P.-J., & Hager, W. (2016). Dyke overtop- Zhou, G. G. D., Zhou, M., Shrestha, M. S., Song, D.,
ping: effects of shape and headwater elevation. Journal of Choi, C. E., Cui, K. F. E., Peng, M., Shi, Z., Zhu,
Hydraulic Research, 54(4), 1–13. https://doi.org/10.1080/ X., & Chen, H. (2019). Experimental investigation on
00221686.2016.1170072 the longitudinal evolution of landslide dam breach-
Pan, Y., Li, L., Amini, F., & Kuang, C. (2015). Overtop- ing and outburst floods. Geomorphology, 334, 29–43.
ping erosion and failure mechanism of earthen levee https://doi.org/10.1016/j.geomorph.2019.02.035