WORKSHOP ON After a brief introduction of general aspects of Interested students and staff will be considered
tsunami generation and propagation, the main aim for participation provided they can indicate that
TSUNAMI MODELLING & is to investigate a new Variational Boussinesq the topic is of direct or future interest for them.
Development of a new Method & Code which should become a major As an indication, additional participants can be:
new tool in simulating tsunamis in the Indonesian
VARIATIONAL BOUSSINESQ CODE area. The theory, the numerical implementation Junior or senior scientists (PhD holders) who
and selection of case studies will be investigated are experienced in modelling (theory) or
in detail. simulation (numerics) of tsunami and/or ocean
waves.
PROGRAMME LAY-OUT
Students (at least with S1) who are
The programme is very flexible and requires interested in the topic and want to be
intense participation. In the mornings joint considered for further PhD-study in this area.
activities are scheduled: some lecturing about
major aspects, literature study and discussions. HOW TO APPLY
The afternoons will mainly be devoted to
reworking the information and starting to Please write a motivation, addressing these
develop various numerical implementations. aspects, to the address given below.
There will be occasional lectures by specialists If accepted for participation, workshop material
about related topics. and coffee/tea and lunches will be provided free
of charge. For students additional funding may be
PARTICIPATION available.
The workshop marks the start of the execution of DEADLINE FOR SUBMISSION: 9 JANUARY 2006
a recently funded KNAW-Mobility Programme (SPIN NOTIFICATION FOR ACCEPTANCE: 16 JANUARY 2006
23 27 January 2006 II, 05MP08), with participants:
Organization and application for participation:
INSTITUT TEKNOLOGI BANDUNG 1. Dr. Nining Sari Ningsih (ITB, Bandung)
2. Dr. Andonowati (ITB, Bandung) The workshop is organised by Laboratorium
3. Dr. Agus Suryanto (UniBraw, Malang) Matematika Indonesia (LabMath-Indonesia) in
Organised by 4. Dr. M. Nurhuda (UniBraw, Malang) collaboration with Pusat Pemodelan Matematika
Laboratorium Matematika Indonesia 5. Prof. Dr. E. (Brenny) van Groesen (UTwente, dan Simulasi (P2MS) ITB.
(LabMath-Indonesia) Netherlands)
in collaboration with 6. Ir. Gert Klopman (UTwente, Netherlands) All correspondence should be directed to:
Pusat Pemodelan Matematika dan Simulasi (P2MS) 7. Dr. Onno Bokhove (UTwente, Netherlands) admin@labmath-indonesia.or.id
Institut Teknologi Bandung
They will be responsible for the detailed
programme and for most of the lecturing. LABMATH-INDONESIA
Participation by others is very restricted to secure Office: Jl. Cigadung Raya Barat 7A Kav. A-2, Bandung
the workshop atmosphere of the meeting. Phone: +62 (0)22 9127 1863
admin@labmath-indonesia.or.id , aantrav@attglobal.net
www.labmath-indonesia.or.id
05MP08
Report of Workshop
Introduction
The workshop marked the beginning of the execution of the KNAW- Mobility
Programme. All executers of the MP were present (except Klopman) and acted as the
`coordinators of the workshop. The first steps in the execution of the research were
made, and participants were introduced to the phenomenon and the specifics of the
code to be developed in the MP.
The workshop (funded outside KNAW-MP) gave ample opportunity to discuss the
details of execution the MP, making plans and arrangements. Besides that, the
participants could be evaluated for possible further training and educational
programmes.
Execution
For schedule, see Annex A.1, for composition of workgroups, see Annex A.2; for
workshop material see Annex D
Participants
The workshop was announced well in advance. The response was large (approx. 40
people applied to participate) and a severe selection was made to allow in total 16
participants besides the 5 coordinators and 2 senior staff from Malyasia.
At the end of the week the coordinators selected participants who performed best.
The first price was for Mauludi (Physics UniBraw Malang), a three month stay at one
of 3 groups (Suryanto-MathUniBraw, Nining ITB-Oceanography, Andonowati
LabMath-Indonesia, Bandung). Four other prices (500.000 Rp to buy books of their
choice) were awarded to: Herry (S Dharma, Yogya), Mashuri (UNSOED), Rima
(OSA-ITB) and Al Azhar (OSA-ITB).
For a complete list of participants, see Annex B
Results
The last day Friday the groups presented their results.
Group 1 and 2 (spectral, resp FE VBM) had a programme running for the simplest
(linear) case, a good result in such a short time.
Groups 3 and 4 presented tsunami simulations with Delft3D for three cases studies:
tsunami approaching Bali-Lombok area form the North and from the South, and
tsunami entrance in Malacca street. The results, although not yet completely reliable,
showed characteristic features and an unexpected noice in the plotting programme.
Group 5 showed some theoretical results about tsunami generation.
1
05MP08
In the schedule below only the plenary activities are listed. Before and after these plenary
meetings, time was used for self-work and group-work.
Work-groups were formed, each consisting of 3 to 4 people, to work on a specific topic. They
reported on their results with an oral presentation on Friday afternoon. Award winners are
noted italic.
2
PARTICIPANTS
OSA, ITB: Nining Sari Ningsih
MA, ITB: Andonowati
MA, UniBraw: Agus Suryanto
Physics, UniBraw: M. Nurhuda
AAMP, UTwente: E. (Brenny) van Groesen
LABMATH-INDONESIA
Office: Jl. Cigadung Raya Barat 7A Kav. A-2, Bandung
Phone: +62 (0)22 9127 1863
admin@labmath-indonesia.or.id
www.labmath-indonesia.or.id
government funding requires more and more Services/Consulting: decision support systems
usefulness ITC, financial institutions
Natural Hazards (water, earthquakes, pollution)
1
MOTIVATION STRUCTURE
INDUSTRIES POLICIES
GOVERNMENT THEMATIC LABORATORIUM
accountability
transparency
RESEARCH PARK DATA INDONESIA
efficiency
LABORATORIUM MATEMATIKA INDONESIA
Mu Commerces
actions.
PIP
Leasure
Re
sid
Spin-off consultancy companies may result from
the activities and commercial activities will be
en
tia
l
ME
accommodated on a fair basis.
RI = Research Institute SME
RPB = Research Park Building
MoPIP = Mono-Pre-Industrial Plant
MuPIP = Multiple-Pre-Industrial Plant
SME = Small and Middlesized Enterprises
ME = Middlesized Enterprises
2
BULETIN PEMODELAN
DAFTAR ISI
Sepatah kata dari redaksi ......... 1
MATEMATIKA
Pengumuman lomba pemodelan
Nomor 1, Volume 1, September 2005
matematika ............................. 1
Tsunami dan gelombang air ... 2, 3 Sepatah kata dari redaksi
Petunjuk untuk guru ................. 4 Buletin ini diterbitkan untuk menumbuhkan minat para pelajar
dan guru SMA pada bidang pemodelan matematika. Model
ies
IKUTILAH KOMPETISI matematika sebenarnya telah dipelajari oleh para pelajar sejak
PEMODELAN MATEMATIKA ! mereka duduk di sekolah dasar, misalnya dalam menuangkan
er
Kompetisi ini terbuka bagi seluruh
soal-soal cerita ke dalam perumusan matematika. Bahkan dalam
S
pelajar SMA dan diikuti secara
kehidupan sehari-hari, misalnya ketika mengatur strategi
t
berkelompok yang terdiri dari 2 s/d
4 orang. Setiap kelompok dipimpin berbelanja dengan jumlah uang terbatas, kita secara tidak sadar
r
po
oleh seorang ketua dan dapat ... telah melakukan pemodelan matematika.
....
R e
l
ica
Diterbitkan oleh Dewan Redaksi
n
LabMath-Indonesia Ketua: Dr. R. Simanjuntak
bekerja-sama dengan Anggota:
c h
Te
Pusat Pemodelan Matematika dan Dr. Gerard Jeurnink
Simulasi (P2MS) ITB Dr. Wono Setyabudhi
dan Jurusan Matematika ITB Dr. Andonowati
Penanggung jawab Prof. E. van Groesen
Pimpinan LabMath Indonesia
www.labmath-indonesia.or.id
FACILITATE:
VISIT US !!!
provide hardware: building, computer-infrastructure www.labmath-indonesia.or.id
provide support: secretarial, grant-searching,
proposal writing, links to industry- society,
advice / guidance / support for starting own company
Office: Jl. Cigadung Raya Barat 7A Kav. A-2, Bandung, INDONESIA Office: Jl. Cigadung Raya Barat 7A Kav. A-2, Bandung, INDONESIA
Phone: +62 +22 9127 1863, admin@labmath-indonesia.or.id Phone: +62 +22 9127 1863, admin@labmath-indonesia.or.id
INVITATION
WORKSHOP DINNER
WEDNESDAY, 25 JANUARY 2006
at SIERRA COFFE & LOUNGE
19:00 -- finish
DRINK & SNACK
will be served in our house prior the dinner
leaving ITB at 17:00; transportation is provided
3
05MP08
At the start:
G. Klopman, M.W. Dingemans and B. van Groesen, A variational model for fully non-
linear water waves of Boussinesq type; in Proceedings IWWWFB 2005.
E. van Groesen & G. Klopman, Dispersive effects in tsunami generation; Proceedings
Rogue Waves 2004, Brest France, 2005.
G. Klopman, M.W. Dingemans and B. van Groesen, Long-distance propagation of wave
groups over bathymetry using a variational Boussinesq model; submitted to Proceedings
IWWWFB 2006.
On request:
Many (recent) publications in the literature, about tsunami modelling and Boussinesq-
type of models are available and can be consulted/copied. A limited number of books is
available.
E. van Groesen & J. Molenaar, Advanced Modelling in Science, Wiley, to be published
2006; Chapter 5 (Variational Modelling and Methods) and Section 6.3 (Variational
surface wave theory).
A variational model for fully non-linear water waves of Boussinesq type
Gert Klopman1 , Maarten W. Dingemans2 and Brenny van Groesen3
Abstract.
Using a variational principle and a parabolic approximation to the vertical structure of the velocity potential, the equations
of motion for surface gravity waves over mildly sloping bathymetry are derived. No approximations are made concerning the
non-linearity of the waves. The resulting model equations conserve mass, momentum and positive-definite energy. They are shown
to have improved frequency-dispersion characteristics, as compared to classical Boussinesq-type of wave equations.
1. Introduction. Classical Boussinesq-type of models suffer from the introduction of high-order mixed
spatial and temporal derivatives. Further, many Boussinesq-type models are not derived from variational
principles and do not satisfy energy conservation. We want to obtain a model for non-linear waves which does
not have these drawbacks.
High-order non-linear models, like Dommermuth & Yue (1987) and Agnon et al. (1999), solve free-surface
evolution equations derived from a Hamiltonian under the constraint that the Laplace equation is satisfied
exactly in the interior of the fluid domain. However, these models have to relate free-surface quantities to those
at some fixed level. This is done by using truncated Taylor-series expansions, thus destroying the exact solution
of the Laplace equation in the interior of the domain and with that the conservation of energy. Conservation of
energy is important, since it can be related to high wave-number instabilities of the model.
Dingemans (1997) describes several methods for constructing Boussinesq-type models with positive-definite
Hamiltonian, but these methods are quite tedious and have certain ambiguities regarding the order of certain
operators, see also Broer (1974, 1975) and Broer et al. (1976). The described models are weakly non-linear.
The present method is easier and unambiguous, leads to a positive-definite Hamiltonian and can be fully
non-linear if desired. Besides the fully non-linear form we also give a simpler weakly non-linear form. The
drawback of the present model is, that instead of higher-order spatial and/or mixed spatial-temporal derivatives,
an additional elliptic equation in the horizontal plane has to be solved (which is also the case for Agnon et al.
1999).
In the following, a parabolic approximation is used for the vertical distribution of horizontal velocity, since
this is simple and eases the derivations. This parabolic approximation already gives improved linear dispersion
characteristics as compared to classical Boussinesq-type models.
However, better performance can be achieved by choosing a power-series expansion in the vertical direction
of the velocity potential. For each additional term in the power series an additional elliptic equation has to be
solved, without increasing the order of the spatial derivatives in the model equations. Additional terms result
in further improvement of the frequency-dispersion characteristics as well as non-linear behaviour of the model.
However, the description of this model will be postponed for the moment.
2. Fully non-linear model. We start from the variational principle for irrotational water waves in the
form as given by Miles (1977) (see also Milder 1977):
ZZ
(2.1) 0 = L = L dx dt,
where (x, t) is the surface elevation, (x, z, t) is the velocity potential and the energy density H(, x , z ; x, t)
is given by the sum of kinetic and potential energy densities:
Z
1 1
(x )2 + (z )2 dz + g 2,
(2.3) H =
h 2 2
while the mass density is taken to be constant and equal to one. Further h(x) is the still-water depth
and g is the gravitational acceleration. This Lagrangian variational principle is equivalent to the Hamiltonian
approach, as shown by Miles (1977). Note that the Hamiltonian H(, x , z ) itself is the spatial integral of H:
1 Appl. Analysis and Math. Physics, University of Twente, The Netherlands, e-mail: g.klopman@math.utwente.nl
2 Boomkensdiep 11, 8303KW Emmeloord, The Netherlands, e-mail: maarten.dingemans@wldelft.nl
3 Appl. Analysis and Math. Physics, University of Twente, The Netherlands, e-mail: groesen@math.utwente.nl
1
R
H = H dx . Now we make the following Ansatz for the potential (x, z, t), corresponding with a parabolic
behaviour over depth with z = 0 at the bed and = at the free surface:
(2.4) (x, z, t) = (x, t) + f (z; h, ) (x, t), with
1 2h + z +
(2.5) f (z; h, ) = ( z ) .
2 h+
We take this choice, because we only want time derivatives of (x, t) and (x, t) to appear in the Euler-Lagrange
equations, and because we know that for a horizontal bottom we have z = 0 at z = h and therefor expect
parabolic behaviour at leading ordera .
Under the assumption of a mildly sloping bottom, i.e. neglecting spatial derivatives of h(x), the velocity
components become:
" 2 #
1 h+z h+z
(2.6) x = x 1 + x + f (z; h, ) x and z = .
2 h+ h+
Note that (x, t) is the vertical velocity z at z = (x, t). From these, we find for the energy density H:
2 2
1 2 1 1
(2.7) H = (h + ) x x (h + ) x + (h + ) x (h + ) x
2 3 3 90
1 1
+ (h + ) 2 + g 2 .
6 2
b
Now by taking variations of L with respect to , and we get from L = 0:
2 1
(2.8) t + x (h + ) x x (h + ) x = 0,
3 3
2 2
1 2 2 1
(2.9) t + g + x x (h + ) x x + (h + ) x
2 3 3 45
1 1 2 7 1
+ 1 + (x )2 2 + x (h + ) x x (h + ) x = 0,
6 5 3 15 5
1 7 2 1
(2.10) (h + ) + (x )2 (h + ) x (h + )2 x x
3 15 3 5
1 1 2
+ x (h + )2 x (h + )2 x (h + )3 x = 0.
3 5 15
We introduce u x , and note from (2.8) that the discharge q(x, t) and depth-averaged velocity U (x, t) are:
2 1
(2.11) q (h + ) U, and U = u x (h + ) x .
3 3
Then the system of equations to be solved can be written as:
(2.12) t + x ( (h + ) U ) = 0,
( 2 2
1 1 1
(2.13) t u + x g + U (h + ) x x + (h + ) x
2 3 45
1 1 2 2 2 7 1
+ 1 + (x ) + x (h + ) u x (h + ) x = 0,
6 5 3 15 5
1 7 2 1
(2.14) (h + ) + (x )2 (h + ) u (h + )2 x x
3 15 3 5
1 1 2
+ x (h + )2 u (h + )2 x (h + )3 x = 0.
3 5 15
So we have to solve two time-evolution equations for (x, t) and u(x, t), as well as an elliptic equationc for
(x, t). Further it can be observed that, for given (x, t) and (x, t), equation (2.14) is a linear equation in
(x, t).
a Higher-order
PM
performance can be obtained by chosing (x, z, t) = (x, t) + m=1
(z )m m . Chosing M = 2 also gives
a parabolic approximation to (x, z, t), but results in better linear dispersion than the present model. The cost is that two elliptic
equations for 1 (x, t) and 2 (x, t) have to be solved, instead of one elliptic
RR equation for (x, t) in the
present model. L
b With L = H (, , , , , ; x, t) we have L =
t x x x L
+ L
L
+ dx dt and e.g. = t
H H
x ( .
x )
c We talk already of elliptic in anticipation to the two-dimensional extension we are planning.
2
3. Weakly non-linear model. If we assume that also the free-surface slope x is small and can be
neglected in the Hamiltonian density (2.7), we get after taking the variations the following system of equations:
1
(3.1) t + x (h + ) u (h + ) x = 0,
3
( 2 )
1 2 1 1
(3.2) t u + x g + u (h + ) x (h + )2 (x )2 + 2 = 0,
2 3 45 6
2
(3.3) (h + ) + x (h + )2 u (h + )3 x = 0.
5
These are somewhat simpler in appearance as the fully non-linear system (2.12)(2.14). Note that this also has
a positive-definite Hamiltonian.
4. Linear dispersion. When linearizing the equations for a horizontal bed, i.e. the still-water depth h is
constant, we get:
1 2 2 2 3 2
(4.1) t + h x u h x = 0, t u + g x = 0 and h + h2 x u h x = 0.
3 5
We look for linear wave solutions propagating as b(x, t) = b exp[ i (kx t) ], where k is the wave number and
is the angular frequency. We find for the above linearized Boussinesq-type equations, with (, u, ) denoting
the complex-valued amplitudes of (, u, ) respectively:
1 2 2 2
(4.2) i + i k h u + k h = 0, i u + i gk = 0 and h + i k h2 u + k 2 h3 = 0.
3 5
Non-trivial solutions exist only if the following dispersion relationship is satisfied:
1 2
2 h 1 + 15 (kh)
(4.3) = (kh)2 2 2
.
g 1 + 5 (kh)
The first terms of a Taylor-series expansion around kh = 0 are:
2 h 1 2 4
= (kh)2 (kh)4 + (kh)6 (kh)8 + O (kh)10 .
(4.4)
g 3 15 75
This dispersion relation can be compared with the exact linear dispersion relation:
(4.5) 2 = g k tanh kh, which has the Taylor-series expansion
2
h 1 2 17
= (kh)2 (kh)4 + (kh)6 (kh)8 + O (kh)10 .
(4.6)
g 3 15 315
So they start differing with the term proportional to (kh)8 . Both dispersion curves are compared in Figure 5.1(a).
They differ less than 1% for kh < 2.3 and less than 2.8% for kh < .
5. Numerical examples. In order to test the models some pre-
liminary computations on a horizontal bed are performed. As initial T [s] H [m] h/ H/h
condition we use periodic waves computed with the method of Rie- 10 2.0 0.0695 0.40
necker and Fenton (1981). We both test the fully non-linear (2.12) 6 1.8 0.1280 0.36
(2.14)) and the weakly non-linear (3.1)(3.3) model. 4 1.5 0.2208 0.30
In all cases we use a constant water depth h = 5 m and g = 9.81 Table 5.1
m/s2 . We consider three cases as given in Table 5. A pseudo-spectral Wave conditions for numerical examples.
method has been used with 100 points per wave length and also 100 points per wave period. A periodic spatial
domain has been used. The elliptic equation for has been solved by means of a conjugate gradient method
(the bi-CGSTAB method), see Quarteroni and Valli (1997). The time-integration has been performed using a
four-stage Runge-Kutta integration method. No numerical damping and smoothing have been applied.
The results of the computations are shown after a simulation time of five wave periods, see Figure 5.1. During
these computations no instabilities occurred. In a similar model using Agnon et al. (1999) we often encountered
numerical instabilities. We think the present model performs well because of the positive definiteness of the
Hamiltonian density, which guarantees good dynamical behaviour of the approximate equations. A check on
the numerical accuracy has been performed. Averaging over one wave length shows that the absolute errors
of the fully non-linear model and T = 4 s are as follows after five wave periods: for the Hamiltonian density
2 105 with hHi = 2.6504, for the error is 3 1017 and for the free-surface potential gradient we find an
error of 2 1016 . Also for the weakly non-linear model the errors are of similar magnitude.
As is obvious from Figures 5.1(b-d) the weakly non-linear model performs not good enough for practical
purposes. The fully non-linear model performs very well.
3
2
1.8
1.5
1.6
1.4
1
1.2
( h / g )1/2
0.5
0.8
0.6 0
0.4
0.2 0.5
0
0 0.5 1 1.5 2 2.5 3 30 20 10 0 10 20 30
kh x
(a) Linear dispersion curves (b) T = 10 s.
0.8
1
0.6
0.4
0.5
0.2
0
0
0.2
0.4
0.5
0.6
0.8
15 10 5 0 5 10 15 10 8 6 4 2 0 2 4 6 8 10
x x
(c) T = 6 s. (d) T = 4 s.
p
Fig. 5.1. (a): Linear dispersion curves h/g as a function of kh for the Boussinesq model (4.3) (solid line) and the exact
linear dispersion (4.5) (dash-dash line). (b)(d): Snapshots of the free-surface elevation after 5 wave periods (b)(d). of the fully
non-linear model (dash-dash line), the weakly non-linear model (dash-dot line) and the Rienecker and Fenton solution (solid line).
6. Conclusion. We have presented a relatively easy derivation of Boussinesq-like equations from a vari-
ational principle having a positive definite Hamiltonian. The resulting mass and momentum equations have
conservative form. Preliminary computations show promising behaviour of the fully non-linear model. At the
conference we intend to show results compared with measurements of waves over varying bathymetry and with
other models.
References.
Agnon, Y., Madsen, P.A. & Schaffer, H.A. 1999. A new approach to high-order Boussinesq models.
J. Fluid Mech. 399, 319333.
Broer, L.J.F. 1974. On the Hamiltonian theory of surface waves. Appl. Sci. Res. 29, 430446.
Broer, L.J.F. 1975. Approximate equations for long wave equations. Appl. Sci. Res. 31 (5), 377395.
Broer, L.J.F., van Groesen, E.W.C. & Timmers, J.M.W. 1976. Stable model equations for long water
waves. Appl. Sci. Res. 32 (6), 619636.
Dingemans, M.W. 1997. Water wave propagation over uneven bottoms, Adv. Ser. on Ocean Eng. 13, World
Scientific, Singapore, 967 pp.
Dommermuth, D.G. & Yue, D.K.P. 1987. A high-order spectral method for the study of nonlinear gravity
waves. J. Fluid Mech. 184, 267288.
Milder, D.M. 1977. A note on: On Hamiltons principle for surface waves. J. Fluid Mech. 83(1), 159161.
Miles, J.W. 1977. On Hamiltons principle for surface waves. J. Fluid Mech. 83(1), 153158.
Quarteroni, A and Valli, A. (1997). Numerical Approximation of Partial Differential Equations, second
corrected printing, Springer Series in Comp. Math. 23, Springer Verlag, Berlin etc., 543 pp.
Rienecker, M.M. and J.D. Fenton (1981). A Fourier approximation method for steady water waves. J.
Fluid Mech. 104, 119-137.
4
DISPERSIVE EFFECTS IN TSUNAMI GENERATION
E. van Groesen1, G. Klopman1
1. Applied Analysis & Mathematical Physics, University of Twente, The Netherlands,
(email: groesen@math.utwente.nl).
Abstract: This paper investigates the effect of dispersion in the initial phase of the generation
of tsunamis. An approximate explicit formula is given for the surface elevation and for an
approximation of the wave length that takes the extent of the event region into account. A
Boussinesq model is described; a numerical implementation makes it possible to perform
efficient and accurate simulations that include dispersive effects; several results with bottom
motions that may be characteristic for motions of tectonic plates are presented.
1. Introduction
The precise process of the generation of waves by seismic events is still rather unclear.
This paper will describe the dependence of the generated waves for a certain event on the
effects of nonlinearity and dispersion. Often, for propagation of tsunamis, the SWE (Shallow
Water Equations) are used. These equations contain nonlinear effects but neglect the
dependence of the fluid velocity on the depth with the argument that this will be a realistic
assumption as long as the wavelength is much larger than the depth. It is known that for
propagation over long distances, the small effects of dispersion may nevertheless become
important enough to be incorporated. This is illustrated in Fig. 1, where in a uni-directional
model it is seen that dispersion causes the initial single wave to split, leading to smaller wave
elevations and, consequently, smaller amplified wave heights during run-up.
Fig. 1. Illustration of the propagation of a wave in the position time plane (horizontal and vertical). The
wave is generated at the bottom middle and runs to the left above a flat bottom over a distance of 1350 km. The
right traveling wave meets an underwater mountain with shallowest part at 700 km, extending from 350 to 1050
km. At the left is shown a simulation based on SWE, at the right including dispersive effects.
Dispersion can also be important for interaction of successive tsunamis and may lead to
much larger surface elevations than expected (see [1]). In the rest of this paper we will
concentrate on the initial phase of the generation, and take bottom motions that may represent
the motion of tectonic plates. We show the dispersive and nonlinear effects for a model with
one horizontal direction; this excludes diffraction and may be justified for very directional
events over small distances. Linear non-dispersive theory can be dealt with in a good
approximation theoretically and we give explicit formulas in Subsection 2.1. To incorporate
nonlinearity and dispersion, we show simulations with a Boussinesq model in Section 3.
One of the problems that are encountered by studying the generation phase is the lack of
models or data of the seismic process itself. On the one, hand a common approximation for
the resulting wavelength is = c0Tevent where is the wavelength, c0 = gh0 the velocity
of long waves above a depth h0 , and Tevent the duration of the seismic event. This duration is
largely unknown; in fact, the above relation seems often be used to find the duration from
observed or (inversely) calculated wavelengths of the tsunami. In using this formula, the
extension of the moving region is neglected, which may be an important contribution; we
adjust the expression for the wavelength in Subsection 2.1. But in any case, the event duration
must be substantial; for instance for a tsunami with wavelength 60 km, with moving region of
40 km length at a depth of 4000 m (with velocity 200 m/s), this leads to Tevent = 100 s with the
new formula, and even 300 s based on the formula above. From recent surveys of the bottom
in the region of the 26 December 2004 North Sumatra earth quake, the maximal vertical and
horizontal displacements are rather small, of the order of 4 and 10 m respectively. For such
events it may be argued that dispersive effects can be neglected, since the wavelength is an
order of magnitude larger than the depth. Yet we will see that because of rather steep
gradients, dispersive effects are already present close to the generation region.
With , u, h denoting the surface elevation, horizontal fluid velocity and depth
respectively, the usual SWE model is given by
( )
t ( + h ) + x (( + h )u ) = 0, t (u ) + x g + u 2 2 = 0
We will linearize this model, and look for analytic approximations, after which we briefly
describe the Boussinesq model that includes dispersive effects.
x + c0t = constant
x c0t = constant ( x, t )
( x, t ) Trise
x W x
Fig 2. This figure illustrates the characteristic method in the ( x, t ) -plane, the region of influence of the
generation region, and gives at the right a picturesque interpretation of the resulting wavelength. For the essential
understanding it is sufficient to recognize that each point in the generation region acts as a source which
generates two pulses that propagate along the characteristic lines.
x
3. Simulations
Below we present some simulations for various bottom motions. Although nonlinear effects
are largely negligible, all calculations did take nonlinearity into account. Bottom motions of
amplitude a = 3 m are considered, and the half-width of the moving region around x=0 is in all
cases set to W = 20 km . The time dynamics of the bottom change is smooth, given during the
event by 1 + cos ( t Trise ) . Monotone bottom changes are given by a (1 + cos ( x / W ) ) / 2 ,
and for dipolar bottom motions, this function is shifted and continued skew-symmetric in x
to generate a positive front wave running to the right and a negative front wave to the left. The
initial bottom depth is 4000 m.
In the plots below we compare the dispersive and the non-dispersive evolutions for various
rise times Trise . The effect of the dispersion is clearly visible already after a short propagation
distance, although immediately after the generation, the shapes are quite similar.
Observe that the amplitude amplification of the (bipolar) wave running over the underwater
mountain differs somewhat from the usually accepted formula that the wave height scales
with the depth according to = h 1/ 4 which would give a factor of 201/ 4 2.1 ; note also the
decreased travel distance because of the depth changes at the right.
1.5 1.5
dispersive dispersive
nondisp nondisp
1 1
[m]
[m]
0.5 0.5
0 0
0.5 0.5
50 100 150 200 250 50 100 150 200 250
x [ km ] x [ km ]
1.2 3
dispersive dispersive
nondisp nondisp
0.8 2
[m]
[m]
0.4 1
0 0
60 40 20 0 20 40 60 30 20 10 0 10 20 30
x [ km ] x [ km ]
Fig 3. Surface elevation generated by bottom rise with Trise = 200 (left) and 20 s (right) , corresponding to
Trise = 2t*, and t * / 5 respectively; on top: the right running wave after 1000 s, in lower plots, the surface
elevation at time equal to Trise .
1.5 2
dispersive
nondisp 1.5
1
1
0.5
0.5
0
[m]
[m]
0 0.5
0.5
1.5
2
1
2.5
1.5 3
0 50 100 150 200 250 600 400 200 0 200 400 600
x [ km ] x [ km ]
Fig 4. Evolution generated by a bipolar bottom motion with Trise = 100 s . In the left plot the effects of dispersion
after 1000 s are shown. In the right plot, with dispersive calculation, the left running wave propagates over an
even bottom. The right running wave meets an underwater mountain with top at x=400 km, and height 3800 m
(so 200 m water depth), leading to amplitude amplification and reduction in propagation speed.
References
1. Mirchina N. & E. Pelinovsky (2001) Dispersive intensification of tsunami waves, ITS
2001 Proceedings, pg 789-794
2. Luke, J.C. (1967), A Variational principle for fluid with a free surface, J. Fluid Mech. 27
395.
3. Klopman, G., M. W. Dingemans & E. van Groesen (2005), A variational model for fully
non-linear water waves of Boussinesq type, Proceedings 20th International Workshop on
Water Waves and Floating Bodies, ed. J. Grue, Spitsbergen, pg 129-132.
Long-distance propagation of wave groups over bathymetry
using a variational Boussinesq model
Gert Klopman1 , Maarten W. Dingemans2 and Brenny van Groesen3
Introduction
Large vessels and floating structures moored with soft springs can be sensitive to large motions due to a
resonant excitation at low frequencies by wave groups. In deep water, this sub-harmonic forcing is mainly
due to a direct variation of the wave radiation stresses on the floating structure by the time variation of the
wave amplitude within the wave groups.
However, in shallow water two other excitation mechanism are possible: first by sub-harmonic bound
waves, which are negligible in deep water, and second by sub-harmonic free waves generated by the trans-
formation of the wave groups over the bathymetry. Here we will study the transformation of wave groups
over bathymetry and the associated generation of sub-harmonic free waves.
Since in general the sea bottom slopes are small, the wave groups have to travel over long distances, i.e.
hundreds or thousands of wave lengths, from deep water to the location of the moored floating structure.
Therefor, for practical computations, often approximate models of non-linear Schrodinger (nls) type are
used which allow for large space and time steps of the order of the wave group scales, see e.g. Dingemans &
Otta (2001) for a review. Liu & Dingemans (1989) presented such as model, subsequently used for numerical
modelling by Dingemans et al. (1991) and Kennedy & Kirby (2003).
However, here we use the variational Boussinesq model of Klopman et al. (2005) to compute wave-group
propagation in one spatial dimension, requiring space and time steps much smaller than the wave length
and period, respectively. First, the model is verified against the results of a fully non-linear potential model
(Westhuis, 2001). Subsequently, computations are made for soliton propagation over an underwater bar.
References
Dingemans, M.W. 1997. Water wave propagation over uneven bottoms, Adv. Ser. on Ocean Eng. 13, World
Scientific, Singapore, 967 pp.
Dingemans, M.W. & Otta, A.K. 2001. Nonlinear modulation of water waves, Adv. in Coastal and Ocean Eng.,
7, Ed. P. L.-F. Liu, World Scientific, Singapore, 176.
Dingemans, M.W., Petit, H.A.H., Meijer, Th.J.G.P. & Kostense, J.K. 1991. Numerical evaluation of the
third-order evolution equations for weakly nonlinear water waves propagating over uneven bottoms. Proc. Comp.
Mod. in Ocean Eng. 91, Eds. A.S. Arcilla et al., Balkema, Rotterdam, 361370.
Kennedy A.B. & Kirby, J.T. 2003. An unsteady wave driver for narrowbanded waves: modeling nearshore
circulation driven by wave groups, Coastal Eng. 48(4), 257275.
Klopman, G., Dingemans, M.W. & van Groesen, E. 2005. A variational model for fully non-linear water waves
of Boussinesq type, Proc. 20th IWWWFB, Longyearbyen, Spitsbergen, Norway, May 2005.
Liu, P.L.-F. & Dingemans, M.W. (1989). Derivation of the third-order evolution equations for weakly nonlinear
water waves propagating over uneven bottoms. Wave Motion, 11(1), 4164.
Miles, J.W. 1977. On Hamiltons principle for surface waves. J. Fluid Mech. 83(1), 153158.
Westhuis, J.-H. 2001. The numerical simulation of nonlinear waves in a hydrodynamic model test basin, University
of Twente, Enschede, The Netherlands, 243 pp.
Westhuis, J., van Groesen, E. & Huijsmans, R. 2001. Experiments and numerics of bichromatic wave groups,
J. Waterway, Port, Coastal and Ocean Eng. 127(6), pp. 334342.
Soliton over a horizontal bed at t = 0 and 900 [s] surface elevation ( time = 900.00 s )
1
1 bottom
0 free surface
0.5
1
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
[m]
1 0
0 0.5
1
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000 1
0.02
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
0
0.02
4500 4600 4700 4800 4900 5000 5100 5200 5300 5400 5500 bottom
0.01 0.1 long wave
0
L [m]
0.01 0
4500 4000 3500 3000 2500
0.01
0
0.1
0.01
2500 2400 2300 2200 2100 2000 1900 1800 1700 1600 1500 0 1000 2000 3000 4000 5000 6000 7000 8000 9000 10000
x [m] x [m]
(a) Soliton over a flat bed. (b) Soliton propagating into shallow water.
t = 900 [s]
1
Hubris
present model
0.8
0.6
0.4
0.2
[m]
0.2
0.4
0.6
0.8
1
3500 4000 4500 5000 5500 6000 6500 7000 7500 8000 8500 9000
x [m]
(c) Comparison of present model results (red line) with Westhuis (2001) (blue line).
Figure 1: Soliton propagating on a slope of 0.012 connecting a region of 12 m depth with a 6 m deep part.
surface elevation ( time = 0.00 s ) surface elevation ( time = 1200.00 s )
bottom bottom
1 free surface 1 free surface
0.5 0.5
[m]
[m]
0 0
0.5 0.5
1 1
L [m]
0 0
0.1 0.1
bottom bottom
1 free surface 1 free surface
0.5 0.5
[m]
[m]
0 0
0.5 0.5
1 1
L [m]
0 0
0.1 0.1
0.5
[m]
0.5
0.8
0.6
0.4
0.2
[m]
0.2
0.4
0.6
0.8
1
1.8 2 2.2 2.4 2.6 2.8 3 3.2
x [m] x 10
4
0.8
0.6
0.4
0.2
[m]
0.2
0.4
0.6
0.8
1
1.8 2 2.2 2.4 2.6 2.8 3 3.2
x [m] x 10
4
1 2
23-27 Jan 2006
Subduction Process
:
Conclusions
5 6
1
7 8
9 10
INDONESIA IS LOCATED IN
AN ACTIVE SEISMIC REGION
SEISMICITY OF INDONESIA
Acknowledgments DEC 26, 04 SUMATRA-ANDAMAN
Hamzah Latief 9.0 EQ GENERATED TSUNAMI
Natawijaya MORE THAN 280,000 DEATHS
Mark van der Meijde MARCH 26,05: NIAS 8.7 EQ
Jan Sopaheluwakan
Scott-expedition APRIL: SIBERUT 6.7 EQ
e.a
11 12
2
Tsunami inundations area in Banda Aceh POST TSUNAMI SURVEY (Latief, 2005 in ITST)
Banda Aceh
7
4.2
3.5
2.3
2.3
3.5
3.2
1.5
1.5
Lhoknga 25-34m 1
1
2.7
13 14
17.6 m
Direction of
tsunami
propagation
Direction of
tsunami
propagation Velocity = 3-7m/s
15 16
LHOKNGA
29.6
17.5
3.9m
View
Lhoknga Lhoknga 6
33
Direction of
tsunami 15.6
propagation
3.9
3.0 3.0 m
23.8
17 18
3
Cement Plan of 23.8
23.8
31.2
25.6
18.6
24.7
22.0
Arah aliran
gelombang
tsunami
19 20
LHOKNGA
INTERTIDAL REEFS
WAS UPLIFTED AND
DIED
Courtesy of Danny HN Danny Hilman Natawijaya
23 24
4
WEST SIMELUE
Damage Analysis from Satelite images
IKONOS
10 January 2003
29 December 2004 Quickbird
23 Juni 2004
28 December 2004
DLR Digital
Globe
27 28
IKONOS IKONOS
DLR DLR
29 30
5
Analysis satelite images Analysis satelite images
Flooded land
Forest
31 32
10 January 2003
30 December 2004
(SUBANDONO, 2005)
33 34
35 36
6
Seismic Wave Propagation
Receiver Array
(seismometers)
Source
(Earthquake)
37 38
Earthquake Epicenter
Fig. 1. Regional map showing earthquakes with magnitudes >5.0 from 1965 to 25 December 2004 from
the earthquake catalog of the National Earthquake Information Center (NEIC)
39 40
Fig. 2. Map showing aftershock locations for the first 13 weeks after the 26 December 2004 earthquake
from the NEIC (yellow dots, with radii proportional to seismic magnitude) Fig. 3. Vertical-component ground displacements for periods
41 42
7
Fig. 6. Constraints on the tsunami source area obtained from the timing of tsunami arrivals at various DATA: Vertical bed motion ????
locations around the Indian Ocean
43 44
Fig. 7. (Left) Tsunami model at a time of 1 hour 55 min after earthquake initiation, computed for a
composite slip model with fast slip (50-s rise time) in the southern portion of the rupture and slow slip Fig. 8. Summary rupture scenario for the 2004 Sumatra-Andaman earthquake
(3500-s rise time) in the north
T. Lay et al., Science 308, 1127 -1133 (2005) T. Lay et al., Science 308, 1127 -1133 (2005)
45 46
8
Ammon e.a.,
May 2005, Science
Ammon e.a.,
May 2005, Science Animation of the vertical
velocity wave field in the
Global movie of the source region. The
vertical velocity wave computation includes
field. The computation periods of 12 seconds
includes periods of 20 and longer with a total
seconds and longer duration of about 13
minutes. As the rupture
and shows a total
front propagates
duration of 3 hours. northward the wave-field
The largest gets compressed and
amplitudes seen in amplified in the north and
this movie are the drawn out to the south.
Rayleigh waves The radiation from
traveling around the patches of large slip
globe. Global seismic shows up as circles that
are offset from each
stations are shown as
other due to the rupture
yellow triangles. propagation (a Doppler-
like effect).
49 50
51 52
Slide scar
Ridges from thrust faulting propagating southwards Slide
Blocks
Evidence
of slope
failure
Abyssal
Plain
Processed by Caris
Processed by Caris
Canyon Accretionary wedge Plate Boundary
Plate boundary (Approx. 40km across)
53 54
9
Historical Large Earthquake and
MAXIMUM TSUNAMI RUN-UP (ITST,2005) Subduction Boundary
SRILANKA
INDONESIA
Feb, 1861
ANIMATION of
The 2004 Aceh tsunami
Nov-24, 1833
10
Earthquake epicenters of Sumatra (Natawidjaja, 2002)
2004
04
20
1861
1935
5,2 cm/year
1833
1883
Cyclic recurrent of
Mentawai earthquakes
Bathymetry around Sumatera around 200 years 6 cm/year
63 64
before after
Aceh 2004
Toli-
Toli-toli 1996 Banggai 2000 Biak 1996
N Sumatera 1935
W Sumatera 1833
Krakatau 1883
East Java 1994 Flores 1992 British Admiralty Charts, 4th edition, Verbeek in 1886
year 1868
11
The 1833 West Sumatra Tsunami
The 1883 Krakatau Volcano Tsunami
1833
10 mnt 30 mnt
20 mnt 40 mnt
Aditya&Latief, 2002
67 68
69 70
Tools
Control
Digital map
Layer
Control
Legendda
Batimetri
Database
Search function
Navigator Basis
Data
71 72
12
Qualitative / Quantitative
Tsunami Generation Modelling
Quantitative Tsunami Modelling
Generation
Qualitative understanding increases Generation mechanisms
Quantitative modelling: still poor
In particular: Free propagation over bathymetry
9 Detection tsunamigenic areas, prediction, timing Presence of currents
9 Earthquake ? Rupture ?
? vertical bottom excitation ?
Land-sea boundaries
? tsunami ? Closed sea areas (Eastern Indonesia)
Tectonic modelling
Crust as elasto-plastic plate: hor. Displacement Okada vert.
Flooding, wetting & drying
displ. Numerical codes
(no dynamics)
Earthquake-generated
tsunamis
Seafloor
deformed by
earthquake
Wave initiated
with small
amplitude and
high speed in
deep water
Speed decreases
and amplitude
increases in
shallow water
76
Questions Contents
Generation region
Can we design scenarios from
1. Intro Tsunami
seismology ??
most likely place of eruptions
/ plate dynamics 2. Qualitative Tsunami Modelling
up- down- sideward motion Excitation
direction
Generation
time scales
total energy input in water Propagation
What is resulting wave motion? Land-sea boundary: wetting & drying, flooding
13
Contents
1 2
23-27 Jan 2006
3 4
5 6
1
Tsunami Generation due to EQ
Ammon e.a., May
2005,
ScienceEvolution of
uplift and subsidence
above the megathrust
with time. The
duration of the rupture
is 550 seconds. This
movie shows the
history of the uplift at
each point around the
fault and, as a result,
the dynamic part of
the motion is visible
(as wiggling contour
lines). The simulation
includes periods of 12
s and longer. The final
frame of the movie
shows the static field.
7 8
Quanitative Tsunami
Modelling Scenarios for Energy conserved water flow. Energy dissipating
seismic events; Long nondispersive (???) run-up and flooding
Overview: Modelling level 0 consequences for waves changing to dispersive, inland from
initiation of water breaking waves in coastal shoreline
Free propagation over bathymetry motion zone
Quantities
u Quanitative Tsunami
h0 , c0 , 0 , 0 h( x ) : depth
Strong dispersive
effects
n
Modelling
Tevent : duration c ( x ) : wave speed energy conserving
-
u
( x ) : wave amplitude smooth waves Overview: Modelling level 0
(x ) : wave length
p
energy dissipating
breaking waves
Free propagation over bathymetry
Relations (de-) focussing Generation mechanisms
c0 = gh0 c( x ) = gh( x ) diffraction & Approximate surface wave equations
0 = c0Tevent (x ) = Tevent c ( x ) refraction
(SWE)
0 = ?? E = g 2 / 2 (Energy)
Landing tsunami: Variational theory of surface waves
E - conservation
initial tsunami
profile resembles 4 ( x )h( x ) = 04 h0
Undular bore? Boussinesq type of approximations
vertical Soliton? Variational Boussinesq Model (VBM)
displacement
11 12
2
Surface Gravity wave Models Generation region Non-dispersive arguments! ??
Provided lambda /h >>1
Basic quantities
h( x , t ) : bottom variations SWE Linearised SWE
( x, t ) surface elevation ( x , t ) : surface elevation
u ( x, t ) surface fluid velocity t ( + h) + x (( + h)u ) = 0 t ( + h) + x (hu ) = 0
u( x , t ) : horizontal velocity
(
t (u ) + x g + u 2 2 = 0 ) t (u ) + x ( g ) = 0
c0 = gh0 : wave speed above h0
Dynamics: dispersion & nonlinearity & energy conservation
dispersion: dependence of velocity on wavelength Formula for surface elevation (generated waves)
arbitrary bottom variations, near generation region (depth h0)
Approximate Equations t t
( x, t ) = t h( x c0 t + c0 , )d t h( x + c0 t c0 , )d
1 1
Linear, Non-dispersive 2 0 20
t ( + h ) = x (hu ) if t h = 0 (
t2 ( ) = x c 2 x ,) t
x + c 0 t = constant
t
t (u ) = x ( g ) with c = gh 2 x c 0 t = constant
(x , t )
(x , t )
Shallow Water Equation (SWE), non-dispersive
t ( + h ) = x ((h + )u )
bottom
motion
(
t (u ) = x g + u 2 / 2 ) x x
13 14
Trise
Example Event
(Realistic??)
h0 = 4000 m; c0 = 200 m/s;
W = 20 km; Trise = 10 s ( = t * 10);
= 40 + 2 km x
W
Inverse modelling Mansinha - Smylie method 72:
15 vertical displacement = initial tsunami profile 16
LabMathindonesia
Boussinesq model & code (& Klopman 05) Boussinesq model & code (& Klopman 05)
Variational formulation of full nonlinear free surface wave equation
Variational formulation of full nonlinear free surface wave with arbitrary bed motions; inviscid, irrotational.
equation with arbitrary bed motions; inviscid, irrotational. Exact energy conservation, momentum accurate.
Exact energy conservation, momentum accurate. For SWE (shallow water equations) the vertical fluid motion is ignored
For SWE (shallow water equations) the vertical fluid motion is completely;
ignored completely; Dispersion is neglected.
Dispersion is neglected. For BOUS (Boussinesq) vertical fluid motion is approximated:
Dispersion is almost exact, 8th order.
For BOUS (Boussinesq) vertical fluid motion is approximated:
Dispersion is almost exact, 8th order. 1
c0 = gh0
P
Dispersion: h 0,8
which is an additional function on the at surface and bottom Maximal Amplitude decreases e
surface: dimension reduction, at cost e 0,2
condition for prescribed d
of addl elliptic eqn. excitation. Non-dispersive arguments! ?? 0
0 1 2 3 4 5 6
Provided lambda /h >>1
kh0
17 18
3
Boussinesq model & code (& Klopman 05) Boussinesq model & code (& Klopman 05)
Variational formulation of full nonlinear free surface wave equation Illustration of Dispersive effects for undisturbed left-running
with arbitrary bed motions; inviscid, irrotational. wave and right wave running over underwater mountain
Exact energy conservation, momentum accurate.
For SWE (shallow water equations) the vertical fluid motion is ignored SWE BOUS
completely;
Dispersion is neglected. t
For BOUS (Boussinesq) vertical fluid motion is approximated: i
Dispersion is almost exact, 8th order. m
e
19 20
a = 3m
1.2
t = Trise
dispersive dispersive
1
1
[m]
[m]
0.5
0.4 1
0.5
t = 1000 sec 0
[m]
[m]
0.5
0
0 0 0.5
1.5
60 40 20 0 20 40 60 30 20 10 0 10 20 30 2
x [ km ] x [ km ]
1
1.5
1.5
dispersive 2.5
dispersive nondisp
0.5
0.5
0.5
0.5
50 21 100 150 200 250
50 100 150 200 250 22
x [ km ] x [ km ]
Technical Remarks
Non-dispersive arguments! ??
Dispersion: Provided lambda /h >>1
Can be largely neglected, but:
small effects will influence large-distance calculations, leading to
reduced wave and run-up heights
for fast events on small regions (volcano eruptions) addition of
dispersion leads to essentially different behaviour;
interaction of successive tsunamis can produce much higher waves
(Pelinovsky ea 2004)
Multi-directionality:
diffraction and refraction (waves around islands)
extension Bous-code immediate, yet needs good transparent
boundary -conditions
Sensitivity analysis from theory and simulations:
generating mechanisms (absent Nias-tsunami)
scaling factors, critical cases, scenarios
23
4
TECHNICAL REPORT
Laboratorium Matematika Indonesia (LabMath-Indonesia)
ab
E. van Groesen
a Dept. Applied Math, University of Twente, The Netherlands
b LabMath-Indonesia, Bandung, Indonesia
EXECUTIVE SUMMARY
This part 1, in a series of papers on a model for tsunami simulations, gives a description of the variational
derivation and formulation of the underlying Variational Boussinesq Model in Cartesian coordinates.
The following concise description of the Variational Boussinesq model will include time-dependent bottoms.
Based on this model, a numerical implementation can be used to simulate tsunami generation by tectonic
plate motions, earthquakes and underwater landslides.
The first section describes the general variational frame work for the 3D surface wave equations including the
vertical direction. Then the next section provides a reformulation in which the vertical dimension is formally
removed. Although still exact, the kinetic energy functional of this formulation has to be approximated for
practical use. This is shown in Section 3, first for simple cases (shallow water equations and fully dispersive-
linear), and then in the framework of the new Boussinesq model. Remarks and comments are given in the
final section.
Other aspects of the model (such as formulation in Earth coordinates), additions (such as land-sea boundary
conditions that are necessary for flood models - wetting and drying-, and to simulate accurately reflections of
waves), and numerical implementations will be dealt with in other parts.
Abstract
This part 1, in a series of papers on a model for tsunami simulations, gives a de-
scription of the variational derivation and formulation of the underlying Variational
Boussinesq Model in Cartesian coordinates.
The following concise 1 description of the Variational Boussinesq model (pre-
sented for fixed bottom in [3]) will include time-dependent bottoms. Based on this
model, a numerical implementation can be used to simulate tsunami generation by
tectonic plate motions, earthquakes and underwater landslides.
The first section describes the general variational frame work for the 3D surface
wave equations including the vertical direction 2 . Then the next section provides a
reformulation in which the vertical dimension is formally removed. Although still
exact, the kinetic energy functional of this formulation has to be approximated
for practical use. This is shown in Section 3, first for simple cases (shallow water
equations and fully dispersive-linear), and then in the framework of the new Boussi-
nesq model. Remarks and comments are given in the final section. Other aspects of
the model (such as formulation in Earth coordinates), additions (such as land-sea
boundary conditions that are necessary for flood models - wetting and drying-, and
to simulate accurately reflections of waves), and numerical implementations will be
dealt with in other parts.
horizontal dimensions); confusion may arise because often the number of horizontal dimensions are
used in surface wave theory.
Corresponding author.
Email address: groesen@math.utwente.nl (E. van Groesen).
Furthermore we will restrict to irrotational flows, which is the common restriction when talking
about waves .
Notation
The vertical z-axis points opposite to the direction of gravity; the gravitational constant of
acceleration is denoted by g, which is assumed to be constant. Together with two horizontal di-
rections this forms a Cartesian coordinate system. The surface elevation is denoted by (x, y, t)
and measured from some reference level z = 0. The depth is given by h (x, y, t) so that the
bathymetry is described by z = h (x, y, t). At several places we will use the total height of the
watercolumn at a specific position and time: D = h + .
Lukes variational formulation ([5]) The basic equations for the gravity driven irrotational
motion of a layer of incompressible fluid with a free surface follow from the dynamic varia-
tional principle
Z
Crit, P (, ) dt
where Z Z
1
P (, ) = dx t + ||2 + gz dz .
h 2
Note that this is a pressure principle since the integrand denotes the pressure in the fluid,
according to Bernouillis formulation of the Euler equations for irrotational fluid.
To derive the equations from the variational principle, we first note that the vanishing of the
first variation of the functional with respect to variations in leads to
Z Z Z
dt dx {t () + ()} dz = 0.
h
Rewriting
Z Z
t () dz = t () dz ()z= t ()z=h t h
h h
the vanishing for all variations leads to Laplace equation in the interior fluid, and to one
2
bottom and one free surface boundary condition, explicitly given by
The equation in the interior layer is the continuity equation, since it can be interpreted as the
equation for incompressible fluid: u = 0, with u = . The equation at the bottom is
recognized as the impermeability of the bottom (no flow through the bottom); similarly, the
equation at the free surface is the kinematic surface condition that no water flows through the
water surface.
1
t + ||2 + g (x, t) = 0 at z = (x, t.) (2)
2
This is the dynamic free surface equation: it is Bernoullis equation stating that the pressure
at the surface of the water should vanish, which is the assumed pressure condition.
The above problem is a formulation for the 2D fluid layer. Formally speaking we can reduce it
to a problem in one space variable only. This reduction of dimension from two to one spatial
dimension is an essential improvement, however at the cost that a functional (the kinetic energy)
has to be introduced that is not easily expressed explicitly in the variables to be introduced.
We reason as follows to obtain a formulation in terms of two basic quantities. These basic
quantities will be the (canonical) variables: the surface elevation (x, t) and the free surface
potential
(x, t) := (x, z = (x, t) , t) .
The choice of this last variable is motivated by the fact that prescribing the surface potential, the
solution of the interior Laplace problem is uniquely determined. Hence, assuming this interior
problem is solved for arbitrary surface potential, the reduction is achieved.
We start to introduce the kinetic energy K (, , t) as a functional of the basic quantities. The
kinetic energy is found as the value function of the following minimization problem
K (, , h) = min { K (, , h) | = at z = } (3)
Observe that the minimizing function satisfies the linear problem for the Laplace equation
3
with the boundary conditions:
R R
Then, using as before h t dz = t h dz t B t h, the functional P (, ) can be
rewritten as
Z Z 1
P (, ) = dx t + ||2 + gz dz .
h 2 Z
1 2
Z Z Z
2
= t dx K (, , h) g h dx + t dx dz.
2 h
The governing equations, obtained by variations with respect to and in the action principle
are the so-called Hamilton equations (in potential form), explicitly given by:
t = H (, , h) ,
t = H (, , h) .
4
Using the form of the Hamiltonian, this can be rewritten as
t = H (, , h) = K (, , h)
t = [ K (, , h) + g] ,
The above equations are in the form of a system of equations of first order in time:
0 1 H (u, , h)
t = . (6)
1 0 H (u, , h)
This is the form in which more general Hamiltonian systems are described.
Reflection
Recapitulating the results so far, we observe that, at least formally, the problem is now well
formulated as a problem in basic variables that depend only on x and t but not on z anymore.
The price that is paid is that we deal with the rather difficult functional K. We will see the
advantage in the next section.
2.1 Additionals
This subsection contains some additional material that is not absolutely needed for the main line of
the description of the Boussinesq model in the next section.
We have derived the Hamilton equations from manipulations in the variational formulation, which
proves their validity in a direct way. Yet it is useful to write these equations in full, which should
coincide with the original equations. To do this, recall the Hamilton equations, in potential form:
t = H (, , h) = K (, , h)
t = [ K (, , h) + g] .
The variational derivative K (, , h) is found from the relation for variations and corresponding
variations of the potential in the interior which satisfies the equations (4) < K (, , h) , >=
[ ] dz B t h]. Using the properties of this reduces to < K (, , h) , >=
R R
dx [
dx [ NS ]z= and hence
R
K (, , h) = [ NS ]z=
d
t = (x, (x, t) , t) = [t ]z= + [z ]z= t .
dt
To calculate K (, , h) we have to realize that we vary the boundary of the integration domain,
and that at this boundary the value of has to remain prescribed to be . Hence, we get a correction
5
when just differentiating with respect to , a correction that is reflected in a change of at the surface
from a change given by S = . [z ]z= . Therefore
1
K (, , h) = ||2 K. [z ]z=
2 S
From this it indeed follows that t = [ K (, , h) + g] is the correct dynamic surface condition.
Any dynamical system with state variable U (possibly a vector function) is called a Hamiltonian
system if it evolves according to an equation of the form
t U = U H (U, t) (7)
where is a skew-symmetric operator (the so-called structure map 3 ). One of the most significant
properties of a Hamiltonian system is that it describes systems for which the Hamiltonian (the energy)
is conserved during the evolution if the Hamiltonian is autonomous, i.e. does not depend explicitly on
time. Indeed,
d
H (U, t) = < H (U, t) , t U > + H (U, t)
dt t
= < H (U, t) , U H (U ) > + H (U, t)
t
= H (U, t)
t
where the vanishing of < H (U, t) , U H (U ) > is a consequence of the skew-symmetry of the
operator . Hence
d
H (U, t) = 0 if H (U, t) = 0. (8)
dt t
For the surface wave problem under consideration,this property implies that if there are no bottom
motions, i.e. h (x, t) = h (x) is time independent, the Hamiltonian is autonomous and the total energy
is conserved during the evolution. When t h 6= 0, energy will be inserted or subtracted from the fluid
motion in the layer.
It may be observed that, even when h is time dependent, the functional K (, , t), and hence the
total Hamiltonian, is invariant for adding a constant to . This is expected since not the potential
but the velocity that is obtained from its spatial derivatives, are the physical variables. This motivates
to introduce a velocity-type of variable u (x, t) := x and to consider u, as the basic pair of
variables, instead of , . Writing the Hamiltonian then as H (u, , t), there is the relation between
3 More precisely, it is only called a Hamiltonian system if is also invertible; if not, it is called a
Poisson system. It is even possible to allow that itself depends on the state variable, = (U ); then
much of the basic properties still hold provided that satisfies the so-called Jacobi conditions.; see
e.g. Van Groesen & De Jager 1994.
6
the variational derivatives given by H = x u H , and the equations become
0 x H (u, , t)
t
= . (9)
u x 0 u H (u, , t)
Observe that this is again of the form of a Hamiltonian system, where now the matrix differential op-
erator, which is skew symmetric, is the structure map . The equations are in this way in conservation
form, expressing the time derivatives as spatial derivatives of corresponding fluxes:
t = x [u H (u, , h)]
t u = x [ H (u, , h)] .
Conservation properties
In dynamical systems, symmetries that are present often correspond to conservation properties. This
is also the case in the present surface wave problem. In particular, suppose that the bottom is flat:
h (x, t) = h constant. Then it is clear that the problem exhibits a translation symmetry. The corre-
sponding functional that is a constant of the motion (invariant integral)is the horizontal momentum
I
d Z
I (u, ) = 0, with I (u, ) = udx.
dt
The reformulation of Lukes principle led to a formulation that, at least formally, is described
in Hamiltonian form as a problem in basic variables that depend only on x and t but not on
z anymore. The Hamiltonian contains the kinetic energy functional K, defined by (), which
cannot be expressed explicitly in the basic variables. This is the essential problem of surface
wave theory.
All (variationally consistent) approximate models for surface waves can be seen as a choice of
approximation of this kinetic energy:
Inserting this approximation into the Hamiltonian, then leads to the approximate equations
which still have the same Hamiltonian form.
The approximations can be of various kind, but all are based on the idea to restrict the class
of wave motions to a class that contains the wavemotions one is interested in, accepting that
other tpe of motions will not be well represented, and accepting that spurious wave motions may
be introduced. Verificatio afterwards of a specific approximate motion satisfies the pre-defined
7
assumptions is therefore essential and replaces in practise the cumbersome - often impossible -
mathematical justification of the approximation.
Shallow water assumption, where one assumes that the wavelenght is much larger than the
depth of the layer, thereby ignoring vertical fluid motions; this is often used as a model for
tsunami propagation. All dispersive effects are ignored, and formally (in the full nonlinear
formulation), the equations are a hyperbolic system, and waves will break.
Linear wave theory, where it is assumed that bottom variations and surface elevations are
small compared to other dimensions. This leads to linear equations, that take vertical fluid
motion into (approximate) account. The resulting linear equations describe dispersive prop-
erties of small amlitude waves (above nearly horizontal bottom).
Boussinesq type of approximations (many variants are around) are meant to describe rather
long, rather small waves, and incorporates both nonlinear and dispersive effects in a certain
approximation. In the next subsections we will briefly describe the two extreme cases and in
more detail the new Boussinesq equation.
We will now derive an approximation for the kinetic energy in the so-called shallow water
approximation. This means that the wave length of the waves under consideration are much
larger than the depth of the layer. This specifies the meaning of shallow, relating the depth
to the type of waves under consideration: kh 0 long waves k 0 on any depth, or arbitrary
wave lengths on a shallow layer h 0.
The assumption leads to the idea to approximate the fluid potential at each depth by its value
at the free surface: (x, z, t) (x, t) independent of z. Taking this approximation in the
kinetic energy integral, but taking into account the integration till the free surface, leads to the
shallow water approximation given by
1Z Z
KSW E = ( + h) (x )2 dx t hdx
2
The non-quadratic term (x )2 in the integrand is the only term that leads to nonlinearity in
the Hamilton equations, which read in potential form
t = x [(h + ) x ] t h
t = g u2 /2
Introducing a velocity type of variable as before, u = x , the first equation can be rewritten
like t (h + ) = x [(h + ) u] , which is clearly the continuity equation expressing mass con-
servation. The second equation is Bernoullis equation at the surface with approximation of the
squared velocity. Together this is the SWE in conservation form
t (h + ) = x [(h + ) u]
h i
t u = x g + u2 /2 (10a)
Linearised SWE
8
Restricting to waves of small amplitudes, we arrive at the linearised equations, which read in
conservation form
t u = x [g] (11a)
t (h + ) = x [hu]
Note that this corresponds to neglect all higher than quadratic term in the kinetic energy:
1Z 2
Z
KlinSW E = h (x ) dx t hdx.
2
1Z 1Z
K (, , h) = [ NS ]z= dx B t hdx
2 2
directly in , is possible in the so-called linear approximation.
For small amplitude waves and small bottom variations, we replace the free surface and the bottom by
a flat surface, i.e. we replace the fluid domain by a horizontal strip, h < z < 0 where h is spatially
constant, an effective, possibly time-dependent, depth. Then the kinetic energy is given by
1Z 1Z
Klin , , h = [z ]z= dx B t hdx.
2 2
We can find an explicit expression by solving the Laplace problem in the strip:
Using Fourier transformation in x, denoting by a hat the Fourier transform of a function, is given
by (suppressing the time dependence for the moment)
Z cosh k z + h
(x, z) = (k) eikx dk.
cosh k h
9
where we introduced functions D (k) and B (k) given by
1
D (k) = k tanh k h , B (k) =
cosh k h
Formally we can rewrite the functional in terms of the surface potential by introducing pseudo-
differential operators D and B which are, by definition, such that after Fourier transformation the
operator becomes multiplication; so for instance for D (k) :
D
d (k) = D (k) (k) .
Such an operator is called a pseudo-differential operator and D (k) is called the symbol of the operator
(reducing to an ordinary differential operator when the symbol is a polynomial) . The operators D
and B are symmetric and positive, corresponding to the fact that the symbol is real and positive. The
the kinetic energy gets the form
1Z
Klin , , h = .Ddx
2
and the governing linear Hamilton equations become
t = D Bt h, t = g.
This can be written as second order equation for the surface elevation
t2 = gD.
t2 + 2 (ix ) = 0.
The fact that is a nonlinear function of k means that surface waves have dispersion: the monochro-
matic mode exp i (kx (k) t) travels with the phase velocity given by
r
cp (k) = (k) /k = g tanh k h /k,
which depends on k . Since is a concave function of k , waves of different (smaller) wavelength travel
with different (smaller) velocity, so that waves of different wavelength starting at the same position
become separated, they are dispersed (NOT dissipated, they do not vanish after some time). Note
that infinitely long waves have the largest phase speed which is given by
q
c = gh.
4 The dispersion relation is obtained as the necessary condition such that b(x, t) = b exp[ i (kx t) ],
where k is the wave number and is the angular frequency, is a solution; this requires a certain relation
between and k, say = (k). The solution exp[ i (kx (k) t) ] is then called a mono-chromatic
mode.
10
A Taylor expansion of the dispersion relation (k) near k = 0 leads to
2 h 1 2 17
= (kh)2 (kh)4 + (kh)6 (kh)8 + O (kh)10
g 3 15 315
2 h h2
!
1
(kh)2 (kh)4 h2 x2 1 + x2 (unstable)
g 3 3
!1
(kh)2 2 2 h2 2
h x 1 (stable).
1 + (kh)2 /3 3 x
The SWEquations miss the vertical fluid motion, and therefore the essential dispersive proper-
ties. We now describe a method to include dispersion by using the variational formulation to
insert in a consistent way vertical fluid motion.
For the vertical fluid motion (the fluid potential) we will now take an approximation. The idea
is to take a parameterized family of potentials, say (p) in the expression for the kinetic energy,
and then, instead of minimizing over all potentials that satisfy the surface Dirichlet condition,
only over the parameterized family (i.e. minimizing over de parameters):
K (, , h) = min { K (, , h) | = at z = }
o
restricted min
p
{ K (, , h) | { (p)}p
n o
= min K (p, , , h) = Kappr (, , h) (12)
p
where as before 1
Z Z
K (, , t) := dx ||2 dz B t h
h 2
which when restricted to the family becomes:
K (p, , , t) = K ( (p) , , t)
The equations that result by taking variations with respect to the parameters of the family are:
<< K, p >>= 0
p
where <<, >> denotes the integration over x and z.
11
We will now motivate the Ansatz for a choice of a parameterized family of potentials. The
kinetic energy can be rewritten if the potential satisfies the Laplace equation in the interior like
1Z n o Z
K (, , h) = [ NS ]z= + B [ NB ]z=h dx B t hdx
2
The normal surface derivative in the first integrand is given by
[ NS ]z= = [x x + z ]z= = x x + [z ]z= 1 + x2
W (x) := [z ]z=
which we are not able to express explicitly in terms of , . The second integral contains the
unknown bottom potential B : variations led to the required bottom condition. Therefore we
take these two functions of x as the parameters in the family of potentials, and will take
variations wrt these quantities.
= and z = W at z = ,
= B at z = h
Functions with simple z-dependence are taken. If we restrict to polynomials (which may, in
fact not be the best choice), the lowest order polynomial in z that can satisfy these conditions
is a parabola, explicitly given by
B + W (h + )
= 2 (z )2 + W (z ) +
(h + )
Actually, the formula can be simplified somewhat. Since B is not prescribed, it is sufficient to
make sure that the potential at the bottom can have arbitrary variations; hence we can take as
family:
1 (z )2
(A, W, ) := A + W (z ) +
2 h+
where now A and W are the parameters to be varied. (We introduced the additional denomena-
tor in the first term to obtain a better scaling for the parameter A; this term could be absorbed
in A).
12
Remark 1 A somewhat different reasoning to find a class of potentials is as follows. At the
bottom, the boundary condition reads
t h = Nb = [z + hx x ]z=h
We make the approximation that the spatial bottom variation is small, and appoximate
t h [z ]z=h .
This will be the essential approximation in the derivation! So now we specify the vertical fluid
velocity at the bottom. Then the potential is uniquely defined and given by
1 W + t h
= (z )2 + W (z ) + ;
2 h+
note that the potential at the bottom is then found to be
1
B = (h + ) (t h W ) +
2
as in [3]. Note that in this way there is only one parameter to be varied in this family, namely
W , which is easier than two. However the price is that the bottom conditon is approximated. It
has to be seen how good this last approximation is, compared to the 2-parameter formulation.
Having given an approximation for the fluid potential, we now insert this in the Hamiltonian
to find the explicit expressions for the equations. The potential above depends (in a rather
difficult way) on the surface elevation , the depth and its time derivative, and on the additional
parametersW, A.
Formally, inserting this potential in the expression for the kinetic energy functional leads to a
functional
n o
KBous (, , h) = min K (A, W, , , h) = min { K ((A, W ), , , h) }
A,W A,W
t = HBous (, , h) = K,
t = HBous (, , h) = g K,
13
and two additional equations from variations with respect to W and A in the definition of the
kinetic energy:
0 = W K (A, W, , , h) (13)
0 = A K (A, W, , , h)
The above method is an example of variational restriction. We can actually see beforehand
what this restriction implies: If we denote the optimal solutions with an overbar, and similarly
the interior potential, the last equations imply
Z Z Z
[.] dzdx [t hB ] dx = 0
for all possible variations within the class. That is, = A. (z )2 + W. (z ) for arbitrary
A, W . Hence we find that the class of potentials does not exactly satisfy the Laplace equa-
tion, but rather the projection into the two (Ritz-Galerkin) directions determined from these
variations (which is as good as can be expected in a two-parameter family).
Z Z h i h i
W (x) . (z ) dz .NB t h . (h + ) = 0,
Z Z h i h i
2 2
A (x) . (z ) dz + .NB t h . (h + ) = 0.
Since the variations in W, A may depend on x in an arbitrary way, we find that the potential
satisfies at each position x the projections
Z h i h i
. (z ) dz .NB t h . (h + ) = 0, (14)
Z h i h i
. (z )2 dz .NB t h . (h + )2 = 0.
These are two equations from which, in principle, the parameters A, W can be solved in terms
of the other quantities. We will formulate this a bit different in the following.
t = Ns at z = ,
Let us now derive the Boussinesq kinetic energy functional in detail. To that end we substitute
the Ansatz of the potential to get an explicit expression in the parameters:
1Z Z 2
Z
K (A, W, , , h) = || dzdx t h.B dx
2
14
and
(z )
z = A +W
h+ ! " #
1 A 2 A
x = (z ) + Wx (z ) + x + x (z ) W
2 h+ x h+
= B z x
we arrive at
1Z Z Z
K (A, W, , , h) = ||2 dzdx t h.B dx
2Z Z
1 n o
= 2z 1 + x2 2Bx z + B 2 dzdx
Z2
1
t h. A W (h + ) + dx
2
Despite this complicated form, it can be observed beforehand that no higher that first order
derivatives of the parameters A, W enter the integrands, namely in the term B. Hence variations
with respect to these parameters, required to find the kinetic energy:
n o
KBous (, , h) = min K (A, W, , , h) ,
A,W
will lead to second order (elliptic) equations for A and W , in agreement with the previous result
(14).
Below we provide some severely simplified versions of the equations (which can be used for
preconditing of the solution of th efull elliptic equations).
Special cases
For A = W = 0 we get
1Z n o
K (0, 0, , , h) = dx 2x D t h. ;
2
15
which is the case of SWE for which (x, z) = (x) independent of z.
For A = 0 (potential linear in z) we get
1Z n o
K (A, W, , , h) = dx 2x D t h.
2
1
h i
+ 1+ x2 D W 2
2x Wx W D2 + x W D
2
1
+ + Wx2 D3 Wx x D2 + 2t h.W D;
3
variations wrt W lead to
2 3
n h i h io h i
x D Wx + 2 1 + x2 D x x D2 W 2x x D + x x D2 + 2t h.D = 0
3
2
n o
for mild bottom slope x D3 Wx + 2D xx D2 W + xx D2 + 2t h.D = 0
3
for mild bottom slope, and lowest (second) order in D/L :
2 2
x D Wx + 2W + 2t h = 0
3
from which we find that in the area where bottom motions take place we have approximately
W = t h
which should be extended with exponentially decaying tails outside the excitation area.
The lowest order result in the excitation area can also be obtained by supposing we neglect
spatial derivatives of A and W , and also in D. Then
1Z
K (A, W, , , h) = dx (16)
2
1 2 1
h i
1 + x2 A AW + W 2 2x x AD2 + x W D
D
3 2
1
h i
+ 2x D 2t h. AW D+ ;
2
Variations wrt A lead to
h i 2
1+ x2 AW = x x + t h.
3
and variations wrt W to
h i
1 + x2 (A + 2W ) = 2x x + 2t h.
which leads to lowest order result (for mild surface elevations) as above: W = t h.
16
4 Conclusions and remarks
Examples of simulations, comparison with other schemes, and an investigation of the perfor-
mance of the resulting scheme will be reported in a successive report.
For simplicity of exposition we have given all the formulas above for the case of one horizontal
direction; the extension to two horizonal directions is immediate.
References
[1] E. van Groesen & G. Klopman, Dispersive effects in tsunami generation; Proceedings Rogue Waves
2004, Brest France.
[2] E. van Groesen & J. Molenaar, Advanced Modelling in Science, Wiley, to be published 2006.
[3] G. Klopman, M.W. Dingemans and B. van Groesen, A variational model for fully non-linear water
waves of Boussinesq type; in Proceedings IWWWFB 2005.
[4] G. Klopman, M.W. Dingemans and B. van Groesen, Long-distance propagation of wave groups
over bathymetry using a variational Boussinesq model; submitted to Proceedings IWWWFB 2006.
[5] J.C. Luke, A variational principle for fluids with a free surface, J. Fluid Mech. 27 (1967), 395
[6] J.W. Miles, On Hamiltons principle for surface waves, J. Fluid Mech. 83 (1977), 153-158
17