Shape optimization of geotextile tubes for sandy

beach protection
Damien Isebe†,∗, Pascal Azerad† , Frederic Bouchette♭ ,
Benjamin Ivorra♮ and Bijan Mohammadi† .
†
Institut de Mathématiques et de Modélisation de Montpellier, Université de Montpellier II, 34095 Montpellier, France
♭
Montpellier Geosciences, Université de Montpellier II, 34095 Montpellier, France
♮
Departamento de Matemática Aplicada, Facultad de Ciencias Matemáticas, 28040 Madrid, Spain

Draft for IJNME

Abstract
The paper describes how to tackle new challenging coastal engineering
problems related to beach erosion with a shape optimization approach.
The method modifies the shape of the sea bottom in order to reduce beach
erosion effects. Global optimization is shown to be necessary as the related
functionals have several local minima. We describe the physical model
used, the proposed protection devices against beach erosion and real case
applications.

Keywords. Shape design, global optimization, water wave propagation, scat-

tering, coastal engineering, beach erosion.

1 Introduction
Beach erosion problems bring increasing engineering demand. Indeed, about
70% of world beaches are crossing an erosion phase, 20% are stable and 10%
show signs of fattening [1, 2]. Obviously, this has major economical and envi-
ronmental impacts. Groins, breakwaters and other coastal structures are used
to decrease water wave energy or to control sediment flows. The shape of these
devices are usually determined using simple hydrodynamical assumptions, struc-
tural strength laws or empirical considerations. However, as we will see, these are
∗
Email: [email protected]

1
not fully satisfactory because of secondary effects. Our aim is to take advantage
of shape optimization techniques, mainly used in aeronautics [3, 4], to propose
new solutions to tackle this problem. This approach is fully innovative in coastal
engineering.
Also, efficient global optimization algorithms are necessary to avoid the design
to converge to local minima. Indeed, we will see that the related functionals have
several local minima. Moreover, the search space is often non-connected. We use
a semi-deterministic algorithm to allow global optimization of systems governed
by PDEs with a low calculation complexity [5, 6].
The paper is structured as follows. In section 2, we recall the state of the art
on geotextile tubes and we describe their parameterization. Section 3 presents
the flow model used for the water waves propagation. Section 4 is dedicated
to the description of the minimization problem. Finally, section 5 displays and
discusses optimization results for two beach protection studies in Northwestern
Mediterranean sea with the aim of reducing the energy available for sediment
transport. In the Appendix, we briefly expose the optimization method used.

2 Geotextile tubes and erosion process

Water waves propagating toward the coast are characterized by their height H,
their period T and their direction θ. The period does not change during the
propagation but the direction and the height may vary when approaching emerged
or submerged structures. This is mainly due to scattering phenomenon.
Knowing the height H of the wave is crucial for the study of an erosion
problem. Indeed, the suspension of sediments, produced by water wave action, is
the main mechanism of erosion process. The suspension of sediments is essentially
linked to the associated water wave mechanical energy E = 81 ρgH 2 where ρ is the
water density and g the gravity acceleration [7, 8, 9].
Oceanographer observations of erosion show that the water waves can be
roughly sorted in two categories according to their height H, below or above
a critical value Hlim . Basically, those above Hlim , mainly present during storms,
are erosive. They generate large mechanical energy. On the other hand, when
H < Hlim , waves foster the reconstruction of eroded beaches. In that follows, the
first class of water waves will be called erosive and the second class constructive.
In order to decrease water waves impact along the coastline many structures
have been proposed [10, 11]. Until recently the most used are emerged break-
waters or groins built with rocks or concrete. However, these techniques are
expensive and only short-term solution for the beach protection because they
mainly transfer in space the erosion process (figure 1 shows the negative impact
of emerged groins on erosion. Accretion occurs only upstream from the longshore

2
drift, whereas erosion is amplified downstream. Hence, it is soon necessary to
build another groin further downstream... and so on).
Currently, interest goes to a new generation of soft structures having less im-
pact on coastal hydro-sedimentary system [12, 13, 14]. These devices are geotex-
tile tubes, also called geotubes (figure 2-Up). These geotubes are long cylinders
made of synthetic textile and filled up with sand.
This paper discusses shape and location optimizations of geotubes for two
sites both located on the Northwestern Mediterranean French coast.
The first analysis concerns the protection of a 2.4 km beach located between
Sète and Agde [15] (figure 2-Down-(Left)). This is an large-scale industrial
project under strict feasibility constraints. The Bas-Rhône Languedoc Company
(BRL) is in charge of the device layout and installation. This company has a great
experience of land-use management and development in Languedoc-Roussillon
(a french area) in order to perform engineering studies and advisory services.
Concerning the studied site, the coastal zone is characterized by a very low tidal
excursion and moderate waves. This beach is subjected to severe erosion and the
coastline has recorded a shoreward displacement of about 50 m since 1967 with
a rate of shore retreat of about 1 m/year since 10 years [16]. In this spot, the
critical wave height Hlim is about 2 m.
The second situation concerns the protection of the Le Grau du Roi Le Bou-
canet beach (figure 2-Down-(Right)). This problem leads to a higher dimension
design problem, as we shall see below.

2.1 Geotube parameterization

Along the barred beach between Sète and Agde, the coastal topobathymetric
profile includes two natural sandbars, created under the water wave action. An
engineering preliminary study suggested first to restructure the beach and the two
natural sandbars by sand recharging and secondly to place two geotubes side-by-
side behind the second natural sandbar in order to protect the new beach [15]. In
the sequel, we model the two geotubes as a single one twice larger. Figure 3 shows
a cross-section of the initial topobathymetry, the suggested sand recharging and
the location of the two geotubes.
The initial topobathymetric data available for this study consists of series of
echosounding data obtained from numerous monitorings since 2000 [16, 17]. So,
we recreate exactly the real morphodynamic of the spot.
Geotubes will merely be represented by local modification of the topobathym-
etry. More precisely, in each node of the discrete domain we have a given value
for the initial topobathymetry. To add a geotube arbitrarily in the domain,
we parameterize its position using a series of N control points in the domain.
Cubic splines are used to connect these points. Once the location known, the

3
shape (cross-section) of the geotube is given by a Gaussian function of the form
2
f (d) = He−sd where d is the normal distance to the spline. This adds two addi-
tional parameters (s, H) for the control of the height and width of the geotubes.
Thus, we have in each node a modification of the topobathymetry (see figure 4)
which accounts for the presence of the geotubes.

3 Flow model
The propagation of water waves over linear irregular bottom bathymetry and
around islands involves many processes such that shoaling, refraction, energy
dissipation and diffraction.
To compute the water wave propagation, we use the REF/DIF code [18, 19,
20]. REF/DIF is a open source software designed for modelling wave propaga-
tion over a weakly varying seabottom. It takes into account both refraction and
diffraction phenomena or wave attenuation. However, this model does not ac-
count for the reflection phenomenon. This appears, for instance, in water wave
propagation in a harbor with vertical emergent structures. Our applications only
concern propagation toward sandy beaches. The model has been validated on
various experimental test cases [21, 22, 23].
REF/DIF is based on a parabolic weakly nonlinear combined refraction and
diffraction model which incorporates all of the effects mentioned above [24, 20].
This model combines the essential features of the two specific approaches, a mild
slope model [25, 26, 27, 28] and a diffraction model [29, 30]. In this section, we
briefly present the nonlinear combined refraction/diffraction model.

3.1 Nonlinear combined refraction/diffraction model

Kirby & Dalrymple, at the Center for Applied Coastal Research (University of
Delaware, US), developed a general formulation governing waves in a domain with
slow but arbitrary depth variations [24, 31]. The following parabolic approxima-
tion for refraction/diffraction is obtained,

∂A ∂ (kCCg )
2ikCCg + 2k (k − k0 ) (CCg ) A + iA (1)
∂x ∂x
 
∂ CCg ∂A
∂y
+ − k (CCg ) K ′ |A|2 A = 0
∂y
where r
g
C= tanh kh (phase speed), (2)
k

4
 2kh
(1 + sinh 2kh
)
Cg = C (group velocity), (3)
2
C cosh 4kh + 8 − 2 tanh2 kh
 
′ 3
K =k , (4)
Cg 8 sinh4 kh
h(x, y) is the local water depth and g the gravity. The local wave number, k(x, y),
is related to the angular frequency of the waves, ω, and the water depth h by the
nonlinear dispersion relationship

ω 2 = gk tanh kh. (5)

k0 is a reference wave number related to the incoming condition. Equation (1) is

valid under the mild slope assumption ∇h/kh << 1 [24].
Equations (1) and (5) are numerically solved by an implicit finite difference
method [19, 20] in a domain of size [0, mr] × [0, nr]. We define xi = (i − 1) ∗ ∆x
and yj = (j − 1) ∗ ∆y where ∆x and ∆y are the space steps on each direc-
tions. The axis x = 0 is the offshore boundary where the incoming condition is
prescribed. The water wave propagation is computed for the increasing i. The
lateral boundaries y = 0 and y = nr have open boundary conditions. Concern-
ing the topobathymetry, at each node (i, j) of the grid, we have a positive value
h(i, j) corresponding to the water depth.
In the model (1), A = A(x, y) is the complex amplitude related to the water
surface displacement by

η(x, y) = ℜ(A(x, y)eikx ). (6)

4 Cost function
An optimization problem refers to the minimization of a cost function J, also
called objective function (see the Appendix ). In this study, our aim is to optimize
the shape of a given geotube and its distance to the coast in order to minimize
the energy available for sediment transport in the near-shore zone. We have seen
in section 2 that this cost function can be expressed as a function of the water
wave mechanical energy E = 81 ρgH 2 where ρ is the water density and H = 2A
[7, 8]. This energy is crucial in the erosion process as it fosters the motion of
sediments.
Another important issue for the modified geometry (i.e. after addition of the
geotube) is that the geotube should not increase the sea bottom fluid particle ve-
locity1 shoreward. For a water wave propagation in the x-direction, the stationary
1
Also called bottom orbital velocity

5
bottom orbital velocity Uorb is given by [30]
Agk
Uorb = cos k · x (7)
ω
with x = (x, y) and k = k(cos θ, sin θ) where k is the wave number and θ the
wave direction.
However, section 2 and Table 1 suggest two main categories of water waves:
the constructive and the erosive ones which, as we said, correspond to waves with
heights below and above the critical water height Hlim = 2 meters. Hence, for a
given direction of propagation θ, the following cost function is considered
R
EH>Hlim dS
Z Z
D initial initial
Jθ = R + (kUorb k − kUorb k)+ + ( EH<Hlim dS − EH<H dS)+
E
D H<Hlim
dS D D
lim

(8)
where (x)+ = max(x, 0) is any regularized max operator in order to have Jθ
differentiable and initial is related to the initial unmodified sea bottom. D is the
area where the minimization is desirable. The first term of J aims to maximize
the constructive waves and minimize destructive ones (as a low-pass filter for
waves). The second (resp. third) term penalizes orbital velocity energy (resp.
energy produced by constructive waves) greater than their initial values (i.e. for
the unperturbed region).
As we are interested by multi-point optimization, we consider the following
weighted combination X
J= p θ Jθ (9)
where pθ is the observation frequency for a given wave direction. This last point
again brings the importance of global minimization tools.
This is also clear from figure 5 which shows a sampling of the functional J
along one dimension for a situation where the only parameter is the distance
of the geotube to the beach. The area D where functional J is computed lies
between 100 and 250 meters far from the coastline. The height of the geotube
has been set to 3 m, its width to 12 m and the offshore distance sampled between
100 and 750 m seaward for a simple straight geotube. The minimum is obtained
for a geotube located at around 350 m from the coastline (we come back to this
simulation in section 5).

5 Application to beach erosion control

In this section we apply the optimization algorithm (briefly described in the
Appendix) to PDE based shape optimization problems arising in the design of
beach protection device. We consider two different sites where the beach is either
straight or curved (see figure 2).

6
5.1 Straight geotube
We first design a straight geotube. This problem arises in the protection of a beach
located between Sète and Agde (Northwestern Mediterranean sea, Languedoc-
Roussillon, France) where industrial constraints reduce the number of design
parameters to two: the distance from the coast and the height of the tube. The
computational domain for the flow is 2400 m longshore and 1200 m shoreward.
The mesh size is 1m cross-shore and 5m longshore. The tube is of the same
length as the studied costal zone. In that case, as the tube was already pre-
designed, the width is fixed to 12 m. The propagation is performed for water
waves data available at 1.2 km of the coastline, stemming from French National
Center Archive for In Situ Wave Data (e.g. from table 1). Recreational sailing
and tourism navigation constraints due to local topobathymetry imply that the
search space [100 m, 200 m]∪[300 m, 850 m] is disconnected. This because no tube
should be placed on the second natural sandy bar (see figure 3) to keep a minimum
depth of 2 m. Coastal engineering suggests to build devices to control the energy
generated in the area between 100 and 250 meters far from the coastline (denoted
by D in the cost function definition, see section 4). This zone corresponds to the
gap between the first and the second natural sand bar.
Coastal engineering first guess would suggest to set the geotextile tube around
550 m from the coast to recreate the third natural sand bar which is missing in
this site. Available geotubes are 3 m high which leaves an acceptable depth of
water to float ships. The optimization procedure locates the geotube at 353
m far from the coast with a height of 2.5 m. This result is confirmed by the
sampling of figure 5. For this reduced number of parameters, which was imposed
by industrial constraints, we could have manually searched the design space. But
we preferred to test our optimizer on this case for which the cost function J is
clearly non-convex and possesses several local minima (see figure 5).
Table 2 shows that the optimized geotube is inactive for the constructive
water waves and reduces the erosive ones. In addition, because of the multi-
point problem considered, this is true for all studied propagation directions : the
geotube plays its low-pass filter role.
Global functional histories for the current and best element found are shown
in figure 6. We plot the output of the core minimization algorithm for different
initializations provided by our multi-layer construction. Several local attraction
basins have been visited. The core minimization algorithm has been called about
50 times. Overall the optimization has required about 700 functional evaluations.
Each state evaluation requires about 15 min on a (3 GHz - 1 GB) personal com-
puter. Figure 7 shows the water wave height for the protected and unprotected
beaches for erosive SSE condition (See Table 1).
This shows that a geotube set immediately before the second natural sand bar

7
makes it possible to break the water waves, therefore attenuating their energy in
the nearshore zone. More precisely, this study reveals that the geotube must
break the water waves sufficiently far away from the coastline but not too far,
otherwise the wave recreates itself. In addition, figure 8 ensures that this opti-
mized configuration does not increase the bottom orbital velocity compared to the
initial configuration, which guarantees that bottom shear-stress is not amplified
(see section 4).

5.2 Curved geotube

The topobathymetry is the one of Le Grau du Roi Le Boucanet beach (North-
western Mediterranean sea, Languedoc-Roussillon, France). The studied area
corresponds to a 3 km longshore and 4.4 km seaward zone with a mesh size of
5 m. We consider the design of one geotube defined by 8 control parameters. A
spline is defined passing by 3 nodes (this makes 6 parameters). The two remain-
ing parameters are height and width of the tubes, which are let free as well. The
length of the tube is required to remain between 500 m and 2000 m.
The cost function (8) is minimized in the region D, which lies between 100
et 250 m offshore, depicted in figure 9-(Left). For computation time reasons, we
only consider a mono-directional south incident water wave with a period T= 8
s and an amplitude A= 3 m.
The optimized geotextile tube is 1.5 km long with a height of 13 m and a
width of 70 m. Its location is represented in figure 9-(Right). This optimized
configuration reduces drastically the energy available for sediment transport. In
this case, the cost function decreases by more than 65% compared to the un-
modified topobathymetry. In practice, it is obvious that putting this optimized
geotube into construction won’t be easy, owing to its size. But in this section,
the objective is to highlight the efficiency of the global optimization algorithm.
We display in figure 10 the water height H in the whole domain for the two
configurations. We observe that, in the region D (close to the east coast), the
water height is reduced from 2 m to 1 m by the use of the optimized geotextile
tube. Also, we observe an important decrease of the bottom orbital velocity in
the optimized configuration (see figure 11). This is important to reduce sand
suspension and displacement by water waves. Finally, free surface elevation is
clearly reduced in region D after geotube introduction (see figure 12).

6 Conclusion
A coastal engineering problem has been modelled and optimal shape design per-
formed for sandy beach protection. Results have shown that geotubes can be de-
signed to reduce beach erosion, under the constraint of minimum water draught

8
for coastal navigation. These devices permit reduction of wave energy, orbital
velocity and free surface elevation. Finally, complementary studies have shown
that despite not being accounted for during design, the generated currents [32]
are also lower after the introduction of the protection device.

7 Acknowledgements
This work is part of the COPTER research project (2006-2009) NT05 - 2-42253,
supported by the French National Research Agency. The authors would like to
thank the Bas-Rhône Languedoc Engineering Co. for their feedbacks during this
work realization.

References
[1] R. Paskoff. Côtes en danger. Ed. Masson., 1993.

[2] R. Paskoff. Les plages vont-elles disparaître? Le Pommier, 2006.

[3] B. Mohammadi and O. Pironneau. Applied shape optimisation for fluids.

Oxford University Press, 2001.

[4] A. Jameson, F. Austin, M.J. Rossi, W. Van Nostrand, and Knowles G. Static
shape control for adaptive wings. AIAA Journal, 32(9):1895–1901, 1994.

[5] B. Ivorra. Optimisation globale semi-déterministe et applications indus-

trielles. PhD Thesis. Université de Montpellier 2, 2006.

[6] B. Ivorra, B. Mohammadi, D.E. Hertzog, and J.G. Santiago. Semi-

deterministic and genetic algorithms for global optimization of microfluidic
protein-folding devices. International Journal for Numerical Methods in En-
gineering, 66(2):319–333, 2006.

[7] C.C. Mei. The applied dynamics of ocean surface waves. Vol I. World Sci-
entific Publishing, 1989.

[8] R.G. Dean and R.A. Dalrymple. Water Wave Mechanics for Engineers and
Scientists. Vol 2. World Scientific Publishing, 1991.

[9] R.G. Dean and R.A. Dalrymple. Coastal processes with Engineering Appli-
cations. Cambridge University Press, 2004.

9
[10] du Territoire et de l’Environnement Ministère de l’Aménagement. La défense
des côtes contre l’érosion marine. Pour une approche globale et environ-
nementale. 1997.

[11] Ramoge. La défense des plages contre l’érosion. Guide à l’attention des
maîtres d’ouvrages. www.ramoge.org, 2002.

[12] R. Tomlinson, J. McGrath, L.A. Jackson, and I. Tuner. Monitoring of a

multi-functional submerged geotextile reef breakwater. Proc. 28th Interna-
tional Conference on Coastal Engineering, pages 1923–1935, 2002.

[13] R. Ranasinghe and I.L Turner. Shoreline response to submerged structures:

a review. Coastal Engineering, pages 536:65–79, 2006.

[14] R. Archetti, M. Kramer, D. Paphitis, C. Mosso, A. Lamberti, and M. Di

Risio. European experience of low crested structures for coastal management.
Coastal Engineering, pages 52:841–866, 2005.

[15] N. Fraysse. Aménagement du lido de sète à marseillan. protection de la partie

nord du lido: étude comparative de solution. Technical report, Bas-Rhône
Languedoc Ingénierie, 2006.

[16] R. Certain. Morphodynamique d’une côte sableuse microtidale à barres: le

golfe du lion (languedoc-roussillon). PhD Thesis, Université de Perpignan,
pages 1–209, 2002.

[17] S. Meulé, V. Rey, R. Certain, and C. Pinazo. Wave transformation on a mi-

crotidal barred beach (sète,france). Journal of Marine System, pages 38:19–
34, 2005.

[18] J. T. Kirby and R. A. Dalrymple. Combined refraction/diffraction model

ref/dif 1, user’s manual. Coastal and Offshore Engineering and Research,
Inc., Newark, DE., January, 1985. (Revised June, 1986).

[19] J. T. Kirby and R. A. Dalrymple. Combined refraction/diffraction model

ref/dif 1, version 2.5. documentation and user’s manual. Research Report
No. CACR-94-22,Center for Applied Coastal Research, Department of Civil
Engineering, University of Delaware, Newark., 1994.

[20] J. T. Kirby and R. A. Dalrymple. Refdif1 v2.5 refraction diffraction model.

Technical report, CACR. Short A.D 1999. Beach and Shoreface Morphody-
namics., Wiley:380pp, 1995.

10
[21] P.A. Work and J.M. Kaihatu. Wave transformation at pensacola pass,
florida. Journal of Waterway, Port, Coastal and Ocean Engineering, 123:314–
321, 1997.

[22] A. Chawla, H.T. Ozkan-Haller, and J. T. Kirby. Spectral model for wave
transformation over irregular bathymetry. Journal of Waterway, Port,
Coastal and Ocean Engineering, 124:189–198, 1998.

[23] A. W. Raichle. Numerical predictions of surfing conditions at mavericks,

california. Shore and Beach, 66(2):26–30, 1998.

[24] J. T. Kirby and R. A. Dalrymple. A parabolic equation for the combined

refraction diffraction of stokes waves by mildly varying topography. J. Fluid.
Mechanics., 136:443–466, 1983.

[25] J. C. W. Berkhoff. Computation of combined refraction-diffraction. In

ASCE., editor, Proc. 13th Coastal Eng. Conf., Vancouver, pages 471–490,
1972.

[26] A. C. Radder. On the parabolic equation method for water-wave propaga-

tion. Journal of Fluid Mechanics, 95(1):159–176, 1979.

[27] I.G. Jonsson and O. Skovgaard. A mild-slope wave equation and its appli-
cation to tsunami calculations. Mar. Geodesy, 2:41–58, 1979.

[28] N. Booij. A note on the accuracy of the mild slope equation. Coastal Engi-
neering, 7:191–203, 1983.

[29] C.C. Mei and D.K.P. Yue. Forward diffraction of stokes waves by a thin
wedge. J. Fluid Mech., 99(1):33–52, 1980.

[30] C.C. Mei and E.O. Tuck. Forward scattering by long thin bodies. SIAM J.
Appl. Math., 39(1):178–191, 1980.

[31] J. T. Kirby and R. A. Dalrymple. Verification of a parabolic equation for

propagation of weakly-nonlinear waves. Coastal Engineering., 8:219–232,
1984.

[32] D. Isèbe, F. Bouchette, P. Azerad, B. Ivorra, B. Mohammadi, and R. Certain.

An introduction to the design of beach defence structures by numerical shape
optimization. In preparation, 2006.

[33] B. Mohammadi and J-H. Saiac. Pratique de la simulation numérique. Dunod,

2003.

11
[34] B. Ivorra, B. Mohammadi, L. Dumas, P. Redont, and O. Durand. Semi-
deterministic vs. genetic algorithms for global optimization of multichannel
optical filters. International Journal of Computational Science and Engi-
neering, 2(3):170–178, 2006.

[35] L. Debiane, B. Ivorra, B. Mohammadi, F. Nicoud, T. Poinsot, A. Ern, and

H. Pitsch. A low-complexity global optimization algorithm for temperature
and pollution control of a complex chemistry flame. International Journal
of Computational Fluid Dynamics, 20(2):93–98, 2006.

[36] B. Ivorra, A.M. Ramos Del Olmo, and B. Mohammadi. A semi-deterministic

global optimization method. application to a control problem of the burger
equation, comparing with other methods. Journal of Optimization Theory
and Applications, 135(1), 2007.

A Optimization method
In this appendix, we briefly present the global optimization method used for the
applications.
We consider the following minimization problem

min J(x) (10)

x∈Ω

where J : Ω → IR is the cost function, x is the optimization parameter belonging

to an admissible space Ω ⊂ IRN , with N ∈ IN. Replacing J by J − Jmin , where
Jmin is the minimum of J, we can assume that minx∈Ω J(x) = 0.
In this section, we give a short presentation of an original global optimization
method used to solve (10) (a full description can be found in [5]):
Consider any optimization algorithm (gradient, Newton, etc...) to solve (10).
This we call core optimization algorithm. It has an output denoted by A0 (v0 , N0 , ε) ∈
Ω where v0 ∈ Ω is one of its initial condition, N0 ∈ IN+ is the number of iterations
and ε ∈ IR defining the stopping criterion. The parameters N0 and ε are fixed at
the beginning of the optimization process. In the sequel, we denote A0 (v0 , N0 , ε)
by A0 (v0 ).
We assume that there exists an initial condition vε ∈ Ω such that A0 (vε )
approximate the solution of (10) with a precision ε ∈ IR+ . The general idea
of the optimization method presented here is to improve the efficiency of any
particular core optimization algorithm A0 by making it global. So, we consider
that the minimization problem is solved if and only if the initial condition v0 lies
in the global minimum attraction basin of J. To find such an initial condition, we

12
consider v0 = s as a new variable and solve the following optimization problem,
which is a reformulation of (10),

min J(A0 (s)). (11)

s∈Ω

To perform the minimization (11), we propose to use a two-layer semi-deterministic

algorithm (here, we use the simplified notation SDA) based on the secant method
[33].
The first layer algorithm A1 (v1 , N1 , ε), where the number of iterations N1 ∈
+
IN is fixed, requests the algorithm A0 . It reads:

• Input: v1 , N1 , ε
• v2 chosen randomly
For i going from 1 to N1
• oi = A0 (vi )
• oi+1 = A0 (vi+1 )
• If J(oi ) = J(oi+1 ) EndFor
• If min{J(ok )k=1,...,i } < ε
EndFor
• vi+2 = vi+1 − J(oi+1 ) J(ovi+1
i+1 −vi
)−J(oi )
EndFor
• Output: A1 (v1 , N1 , ε) := argmin{J(ok ), k = 1, ..., i}

Note that the step vi+2 = vi+1 − J(oi+1 ) J(ovi+1

i+1 −vi
)−J(oi )
is the secant method
applied to the new functional v 7→ J(A0 (v)).
However, as this line search minimization algorithm might fail, an external
level A2 is added to the algorithm A1 to have a multidimensional search. Hence,
v1 = w is seen as a new variable in A1 and solve the following optimization
problem, which is a reformulation of (11),

min J(A1 (w)). (12)

w∈Ω

To perform the minimization (12), we consider a two-layer algorithm, with

an output called A2 (w1 , N2 , ε), where the number of iterations N2 ∈ IN+ is fixed,
which calls algorithm A1 (which in turn calls A0 ). It reads:

• Input: w1 , N2 , ε
• w2 chosen randomly
For i going from 1 to N2
• pi = A1 (wi )
• pi+1 = A1 (wi+1 )

13
 • If J(pi ) = J(pi+1 ) EndFor
• If min{J(pk )k=1,...,i } < ε
EndFor
• wi+2 = wi+1 − J(pi+1 ) J(pwi+1
i+1 −wi
)−J(pi )
EndFor
• Output: A2 (w1 , N2 , ε) := argmin{J(pk ), k = 1, ..., i}

Again, the step wi+2 = wi+1 −J(pi+1 ) J(pwi+1

i+1 −wi
)−J(pi )
is the secant method applied
to the new functional w 7→ J(A1 (w)).
In practice, we randomly choose the initial condition w1 ∈ Ω and we con-
sider (N0 , N1 , N2 ) = (10, 5, 5). These values give a good compromise between
computation complexity and result accuracy. This algorithm was first described
in [33], along with mathematical background. It has been validated on various
benchmark test cases [5] and industrial applications [6, 34, 35, 36].

South South South East East South East East

constructive Hs = 0.76m Hs = 0.85m Hs = 0.85m Hs = 0.66m
water Ts = 4.96s Ts = 5.21s Ts = 5.21s Ts = 4.99s
waves p = 24.66% p = 22.75% p = 22.75% p = 17.5%
erosive Hs = 2.91m Hs = 3.233m Hs = 3.233m Hs = 3.55m
water Ts = 7.54s Ts = 7.78s Ts = 7.78s Ts = 8.03s
waves p = 2.84% p = 3.25% p = 3.25% p = 2.5%

Table 1: Typical hydrodynamic data used in our simulations . Significative height

Hs , mean period Ts and mean frequency of observation p for four significant
directions of propagation.

14
 South South South East East South East East
constructive
water == == == ==
waves
weights 24.66% 22.75% 22.75% 17.5%
erosive
water 15% gain 30% gain 16% gain 15% gain
waves
weights 2.84% 3.25% 3.25% 2.5%
overall gain 23%

Table 2: Multi-point optimization: variations of constructive and destructive

energies for various incidences θ. (==) means that the quantity is almost un-
changed, less than 0.1% variation.

Figure 1: Secondary effect of emerged groins: erosion has been amplified on one
side of the groins (downstream from the longshore drift).

15
Figure 2: Up- A geotextile tube before submersion (source: BRL engineering);
Down- Left: the barred beach between Sète and Agde, Right: the beach of Le
Grau du Roi Le Boucanet (source: Google Earth).

16
 Figure 3: The improvement of the considered site (source: BRL engineering).

modification of
the topobathymetry A ε
due to the geotube

the initial
−ε topobathymetry

Figure 4: Left: The profile parameterization for the geotube in an academic

linear topography; Right: Up- The initial topobathymetry of the barred beach,
Down- Implementation of a geotube in the topobathymetry. Note that both
ends are smoothed by a suitable function.

17
 1.15

Cost function value J

1.05

0.95

0.85

0.75
100 200 300 400 500 600 700 800
Distance Coastline/Geotextile (m)

Figure 5: Cost function evolution w.r.t to the geotube position. This is difficult
to solve with a local minimization algorithm.

0.9
Best convergence
History of convergence
0.88

0.86
value of J

0.84

0.82

0.8

0.78
0 10 20 30 40 50
call to core minimization algorithm

Figure 6: Functional values history for the best element found by the core min-
imization algorithm and the current value of the functional. One sees that the
optimization algorithm is not a descent method.

18
 4m

2m

0m
height

−2m

−4m

−6m D
0m 100m 200m 300m 400m 500m 600m
offshore distance

Figure 7: Bottom curve: cross-shore topobathymetry. Solid top curve: water

height compared to the original height (dashed line).

2m

0m
orbital velocity

−2m

−4m

−6m D
0m 100m 200m 300m 400m 500m
offshore distance

Figure 8: Bottom curve: cross-shore topobathymetry. Solid top curve: orbital

velocity compared to the velocity on the unprotected beach (dashed line).

19
Figure 9: (Left): The initial topobathymetry where the white box represents the
region D for the cost function computation; (Right): The modified topobathym-
etry with the optimized geotube.

Figure 10: The water height H in the whole domain: (Left) for the initial con-
figuration, (Right) for the optimized configuration.

20
Figure 11: The bottom orbital velocity in the domain: (Left) for the initial
configuration, (Right) for the optimized configuration.

Figure 12: The free surface elevation in the cost function region D: (Left) for the
initial configuration, (Right) for the optimized configuration.

21