You are on page 1of 20

ORI GI NAL PAPER

A corrected 3-D SPH method for breaking


tsunami wave modelling
Jinsong Xie

Ioan Nistor

Tad Murty
Received: 10 August 2010 / Accepted: 2 May 2011
Springer Science+Business Media B.V. 2011
Abstract A 3-D large eddy simulation model that was rst transformed to smoothed
particle hydrodynamics (LES-SPH)-based model was employed to study breaking tsunami
waves in this paper. LES-SPH is a gridless (or mesh-free), purely Lagrangian particle
approach which is capable of tracking the free surface of violent deformation with frag-
mentation in an easy and accurate way. The Smagorinsky closure model is used to simulate
the turbulence due to its simplicity and effectiveness. The Sub-Particle Scale scheme, plus
the link-list algorithm, is applied to reduce the demand of computational power. The
computational results show that the 3-D LES-SPH model can capture well the breaking
wave characteristics. Spatial evolution features of breaking wave are presented and visu-
alized. The detailed mechanisms of turbulence transport and vorticity dynamics are
demonstrated as well. This application also presents an example to validate the SPH model.
Keywords Smoothing particle hydrodynamics Large eddy simulation
Breaking tsunami waves Turbulence modelling Fragmentation
Lagrangian method Free surface tracking
1 Introduction
Nearshore behaviour of tsunami waves represents one of the most complex physical
phenomena and one of the least understood aspects of this natural disaster. Breaker height,
depth and water particle velocities are of considerable interest for hydrodynamic
description and sediment transport. While it took decades to understand tsunami wave
dynamics, several unknowns still remain, especially with respect to the tsunami wave
breaking mechanism in the breaker zone and subsequent behaviour in the surf zone. Due to
its long wave length/period, it is technically difcult and also very expensive to employ a
physical model to examine such phenomena. It is also challenging to numerically simulate
J. Xie (&) I. Nistor T. Murty
University of Ottawa, 161 Louis Pasteur, Ottawa, ON K1N 6N5, Canada
e-mail: jxie036@uottawa.ca
123
Nat Hazards
DOI 10.1007/s11069-011-9954-x
tsunami wave breaking due to the fact that the numerical model has to include fully 3-D
turbulence modelling and free surface wave tracing techniques.
Tsunami waves, which are long gravity waves, have small amplitudes in deep ocean
conditions, and thus, shallow water equation (SWE) model is the common tool to inves-
tigate the properties of long waves due to the characteristics of their relative water depth
(d/L ( 1) and/or their relative wave steepness (H/L ( 1, where L is wavelength, H is
wave height, and d is water depth). Such a model is computationally efcient and effective
for large domain simulation during wave propagation. More specically, when both of the
above conditions are satised, small amplitude wave theories can be used to investigate
tsunami wave properties. Xie et al. (2007) performed a systematic study of the tsunami
wave propagation using a shallow water equations numerical model. As a tsunami wave
approaches the shore, its wave height increases quickly, while the wave prole becomes
steeper. As documented in the literature, the initiation of breaking starts when the ratio of
breaker height to depth is in the range of 0.61.3. Moreover, the breaking criterion from
solitary wave theory, rst derived by McCowan (1894), later updated by Williams (1981),
is given by:
H
b
0:78d
b
1
in which H
b
is dened as the maximum wave height, and d
b
, breaker depth, is dened as
the still water depth over the beach prole at the location of maximum wave height. In the
process of wave breaking, generally speaking, there are three typical types of wave
breakers in shallow water, namely spilling breaker, plunging breaker and surging breaker,
which depend on the incident wave steepness and beach slope. The classication of various
breakers can be made based on the surf similarity parameter n
b
, which is dened as
(Battjes 1974):
n
b

tan b

H
b
=L
0
_ 2
where b is the beach slope angle, H
b
is the breaker height, and L
0
is the deep water
wavelength of the incident periodic wave. Since the wave length and wave period of a
solitary wave are theoretically innite, Grilli et al. (1997) introduced the following
dimensionless slope parameter for solitary waves:
S
0

sL
0
h
0
1:521
s

H
0
0
_ 3
in which s is the beach slope, h
0
is the offshore water depth, and H
0
0

H
0
h
0
is the dimen-
sionless incident solitary wave height H
0
is the deep water wave height. When S
0
\0.025,
waves break in the form of the spilling breakers which form a series of aerated wave crests
on the beach, a small jet locally generates small eddies in the vicinity of a steep wave crest
(see Fig. 1). When 0.025 B S
0
\0.3, plunging breakers occur. The wave crest curls
downwards and impinges on the wave trough in the front, air bubbles are subsequently
entrained, the horizontal vortex roller dissipates the energy dramatically, and the ow
regime changes from irrotational to rotational, accompanied by the generation of strong
turbulence. Small waves are regenerated and splashed up, and they behave as large and
small rollers or eddies (see Fig. 2). To some extent, the small waves restabilize and
transform into a turbulent bore, with the height of the broken wave then remaining at an
approximate xed proportion to the mean water depth and moving onshore in a bore-like
Nat Hazards
123
Fig. 1 Tsunami wave breaking simulation (spilling breaker)
Nat Hazards
123
Fig. 2 Plunging breaker simulation-breaking, development and formation of plunging jet (Li 2000, the
grey-coloured pictures)
Nat Hazards
123
shape (see Fig. 3). The area from the initiation of wave breaking to this particular point is
dened as breaker zone. The present paper focuses on the breaker zone, somewhat dif-
ferent from the traditional denition of a portion of the surf zone. When 0.3 \S
0
\0.37,
waves break in the form of collapsing or surging breakers, during which most of the wave
energy is reected back from the beach, with a small amount of wave energy lost in the
Fig. 3 A plunging breaker post-breaking and splash-upqualitative comparison with the laboratory
photographs (Li 2000, the grey-coloured pictures)
Nat Hazards
123
breaking process. When the slope is very steep, the surging wave experiences near com-
plete reection, with little breaking after it runs up to a maximum elevation.
Signicant advances have been made in both theoretical and experimental studies on
wave breaking, especially for short waves. Therefore, the literature in short wave breaking
is extensive and detailed. However, the difculty of measuring turbulent velocity due to the
presence of many air bubbles entrained by plunging jets has obstructed experimental
studies on wave breaking. On the other hand, progress on numerical modelling of breaking
waves is still far from complete, and simulation of breaking tsunami waves remains
surprisingly sparse, especially for 3-D modelling. Turbulence modelling and free surface
tracking are two key issues in the study of breaking tsunami waves. Traditionally, Eulerian
methodbased wave breaking model requires the incorporation of two mechanisms to
simulate wave breaking. One is the trigger mechanism for the initiation of wave breaking,
while the other one is the energy dissipation mechanism. A prescribed wave breaking
criterion (when and where the wave breaks) must be established rst so that the dissipation
term can be triggered and added to the momentum equations. Three types of energy
dissipation methods are often employed: the surface roller model (Madsen et al. 1997, the
vorticity model (Svendsen et al. 2000) and the eddy viscosity model (Kennedy et al. 2000).
The difculties in calculating the free surface proles during and after wave breaking are
enormous when one uses the volume of uid (VOF) method (Hirt and Nichols 1981) and
probability density function (PDF) method (Watanabe 1996). In the aerated area, the
conventional assumptions that the free surface is continuous and in the form of a material
surface are no longer valid. Compared with Eulerian methods, Lagrangian methods do not
require the initiation mechanism for wave breaking but focus on the energy dissipation
modelling. Additionally, there is no special need to treat free surface prole, while frag-
mentation can be easily simulated. However, because numerical accuracy of high-order
numerical methods and high spatial and temporal resolution are indispensable to restrict
computational errors and keep computation stable, there have been few papers on 3-D
wave breaking simulation due to the high computational effort required.
Although wave breaking is essentially a 3-D phenomenon, some researchers have
assumed that if the incident wave perpendicularly enters the computational domain,
breaking wave motion can be simplied as a 2-D problem. In the literature, there are three
kinds of models to simulate complicated turbulence ow, namely the Reynolds-averaged
NavierStokes (RANS) model, the large eddy simulation (LES) model and the direct
numerical simulation (DNS) model, enumerated in their increasing order of accuracy and
computational capability demand. Lin and Liu (1998) performed a quantitative comparison
between the spilling and plunging breakers by using an advanced 2-D RANS model; Shao
and Ji (2006) employed an incompressible 2-D SPH-LES model with SPS scheme to study
plunging wave breaking. Their computations were compared with the experimental data of
Ting and Kirby (1995), and good agreement was found between the numerical and
experimental results. However, 2-D SPH-LES model results were shown to be more
effective in describing large free surface deformation and better than those obtained by
using RANS model. Direct numerical simulation (DNS) for 2-D water wave breaking was
carried out by Watanabe and Saeki (1996). Overturning jets, plunging jets and large-scale
eddies were rather accurately simulated. These models are 2-D approximations of the 3-D
ow and hence are less computational demanding than fully 3-D models. However, using
these models, it is difcult to study the details of 3-D ow features such as those observed
during the wave breaking process. Watanabe and Saeki (1999) also carried out a 3-D large
eddy simulation (LES) of breaking waves and found that there was a 3-D coherent eddy
structure comprised of horizontal, helical and vertical eddies following breaking. Helical
Nat Hazards
123
and vertical eddies are generated simultaneously when the overturning jets plunge at rst,
while the positive and negative vorticities are arranged by turns in the lateral direction and
are stretched in the wave direction as the wave front progresses. The fully stretched eddies
are then gradually broken down into small fractions. Although many air bubbles entrained
by breaking waves potentially affect turbulent evolutions in actual wave breaking process,
the effects of bubbles are ignored for the sake of simplicity in the aforementioned papers.
2 A corrected 3-D LES-SPH-SPS model
Meshless methods have been developed to solve the complex surface geometry that cannot
be readily handled by mesh-based methods. Smoothed particle hydrodynamics (SPH) is
one of the earliest gridless methods invented in mid-1970s for the study of astrophysics.
Due to its simplicity and robustness, this numerical method has been extended to the study
of solid mechanics for dynamic phenomena and more recently to complex uid mechanics
problems such as dam breaking and wave propagation. Substantial development and
improvement of meshless methods started in mid-1990s, after which there was a signicant
development of meshless methods in computational physics and mechanics. At present,
SPH is the main meshless method that has been successfully applied in uid ow mod-
elling. In SPH, a computational domain is represented by a set of interpolation points
called particles where the uid medium is discretized by interaction between particles
rather than grid cells. The basic concept of SPH is that continuous media are represented
by discrete particles with volume, density and mass. The particles have a kernel function to
dene their range of interaction, and the hydrodynamic variables are approximated by
integral interpolations. Mesh-based numerical methods limit their applications in many
complex problems such as grid generation, which is not always a straightforward process
and can constitute an expensive task, both in terms of computational time and mathe-
matical complexity. Mesh-free methods are easier to treat large deformations, disconti-
nuities and singularities compared with grid-based methods. In the SPH computations, the
free surface can be easily and accurately tracked by following water particles without
numerical diffusion.
The standard SPH code (SPHysics version 2.2) is an open-source code that has been
released in September 2010, which was jointly developed by researchers at the Johns
Hopkins University (USA), the University of Vigo (Spain), the University of Manchester
(UK) and the University of Rome La Sapienza (Italy). A corrected, improved and opti-
mized 3-D SPH model has been developed specically for free surface hydrodynamics in
the present research.
In SPH, the fundamental principle is to approximate any function in discrete notation at
a particle a by
A
a

b
m
b
A
b
q
b
W
ab
4
where W
ab
W r
!
a
r
!
b
; h
_ _
is the weighting function (or kernel), h is called the
smoothing length, and the mass and density are denoted by m
b
and q
b
, respectively. The
weighting function should satisfy several conditions such as positivity, compact support
and normalization. Also, it must be monotonically decreasing with increasing distance
from particle a and should behave in the form of a Dirac-delta function, as the smoothing
Nat Hazards
123
length h tends to zero (Liu and Liu 2003). Monaghan (1989) obtained a form of the
gradient for a scalar eld through numerical experiments:
rA
a
q
a

b
m
b
A
a
q
2
a

A
b
q
2
b
_ _
rW
ab
5
Divergence of a vector eld can be written in SPH formalism:
r A
a
q
a

b
m
b
A
a
q
2
a

A
b
q
2
b
_ _
r W
ab
6
The above equations are antisymmetric with respect to the a and b subscripts. Conse-
quently, the contribution of particle b to rA
a
(or r A
a
) is identical to the contribution
of particle a to rA
b
(or r A
b
). From a computational point of view, these consider-
ations are important as the calculation time can be reduced by employing them. There is no
need to calculate the contribution of particle a to particle b if the contribution of particle b to
a is already computed. In SPHysics, the user can choose from one of the following four
different kernels: (1) the modied Gaussian (Colagrossi and Landrini 2003), (2) quadratic
(Dalrymple and Rogers 2006), (3) cubic spline (Monaghan 1992) and (4) quintic (Wendland
1995). The commonly used weighting functions in the research project are cubic spline
W r; h a
d
1
3
2
q
2

3
4
q
3
0 q 1
1
4
2 q
3
1 q 2
0 q !2
_
_
_
7
where a
d
10=7ph
2
in 2-D and a
d
1=ph
3
in 3-D, q = r/h. The second derivative of
this kernel is continuous, and the leading truncation error term is O(h
2
). The niteness of
the kernel support means that only a limited number of neighbouring particles b play a role
in all the sums of conservation equations. This is used to reduce the computational time by
building a link list between particles at each time step.
There are various forms in which the NS equations are represented in SPH (Monaghan
1992). The most commonly used form is:
Dq
a
Dt

b
m
b
u
a
u
b
rw
ab
8
(compressible form)
D u
!
a
Dt

1
q
r
!
P H
!
g
!

b
m
b
p
a
q
2
a

p
b
q
2
b
R
ab
_ _
rw
ab
g
!
9
where H
!
refers to the diffusion terms.
As noted in the LES model, one deals with two different scale levelsthe larger eddies,
which represent the mean ow features and the boundary conditions, and the smaller
eddies, which are universal everywhere. The former are much larger than the later ones.
The computational domain must be large enough to represent the energy-containing
motions, while the grid spacing must be small enough to resolve the dissipative scales. In
addition, the time step used to advance the solution is limited by considerations of
numerical accuracy. This requires that the time step in the model is reasonably small;
otherwise, this will make the numerical model unstable and cause large computational
perturbation errors. As a result, huge computational capacity is required, and in some case,
this is extremely computationally intensive. In order to avoid solving the complex
Nat Hazards
123
Poissons equation, which demands signicant computing power, the uid pressure in the
SPH formalism is treated as weakly compressible. The compressibility is adjusted by using
Courant Fredrich Levy (CFL) condition to slow the speed of sound (pressure wave) such
that the time step is reasonable. Monaghan (1989) also pointed out that the sound speed
should be about ten times faster than the maximum uid velocity, thereby keeping density
variations within less than 1%.
P B
q
q
0
_ _
c
1
_ _
10
where c = 7, and B c
2
0
q
0
=c being q
0
1,000 kg m
3
for the reference density and
c
0
c q
0

oP
oq
_ _
_

q
0
the speed of sound at the reference density. For turbulence mod-
elling, several complicated formulations have been proposed to model the sub-grid scale
(SGS) turbulence. The Sub-Particle Scale (SPS) approach is very similar to the concept of
large eddy simulation (LES), which was rst described by Gotoh et al. (2001) to represent
the effects of turbulence in their moving particle simulation (MPS) model. The concept
comes from the fact that large eddies vary from ow to ow, while small eddies are
universal everywhere. The idea is that we only need to compute the resolvable large-scale
turbulence and the mean ow, while the unresolvable small-scale turbulence can be
modelled based on some closure models and can be thus removed from the intensive
calculations. In the recently upgraded SPHysics version 2.2, three different options for
diffusion can be used: (1) articial viscosity (Monaghan 1992), (2) laminar viscosity (Lo
and Shao 2002) and (3) full viscosity (laminar viscosity?Sub-Particle Scale turbulence).
The articial viscosity proposed by Monaghan (1992) has been used very often due to its
simplicity. However, it is a scalar viscosity which cannot take into account the ow
directionality and does not conserve angular momentum. In addition, it has been proved
that it leads to strong dissipation in some case of ow simulation (Dalrymple and Rogers
2006) such as in the case of complex shearing ows where a too large vorticity decay and
unphysical momentum transfer are present. Following Dalrymple and Rogers (2006),
momentum equation with full viscosity can be written as:
d u
!
a
dt

b
m
b
P
b
q
2
b

P
a
q
2
a
_ _
r
!
a
W
ab
g
!

b
m
b
4v
0
r
!
ab
r
!
a
W
ab
q
a
q
b
r
!
ab

2
_ _
u
!
ab

b
m
b
s
b
q
2
b

s
a
q
2
a
_ _
u
!
a
W
ab
11
3 Add source code for solitary wave generation
In order to simulate breaking waves, a source term has to be added into the computational
domain. It should be noted that until now, most of the wave theories were based on velocity
potential formula. A solitary wave (which consists of a single volume of uid propagating
entirely above the undisturbed free surface and is regarded as a weakly non-linear and
dispersive wave) is often used to approximate the leading wave front of a tsunami. While
wave non-linearity tends to make wave front steeper, wave dispersion will counterbalance it.
For this reason, a solitary wave can travel a long distance without signicant shape distortion
and energy loss. For a tsunami generated from far eld, phase dispersion and amplitude
Nat Hazards
123
dispersion lead to wave frequencies separation. As a result, short waves are damping while
the wave travels in deep ocean and only long waves approach the shore. Normally, only the
rst few waves are of concern. For a tsunami generated in the near eld, due to the limited
water depth, group velocity and phase velocity are about equal and, generally, no wave
separation happens. Therefore, it is acceptable to use a single wave in both numerical and
physical models for tsunami wave research in the nearshore region. Also, it is noticeable that
todays computing power still does not allow large area simulation for breaking waves
though our numerical technologies have advanced considerably. Scaling methods can be
used in numerical models to scale down the prototype as it would be in physical models, and
this again brings scaling effects and boundary effects into numerical computation. However,
one can utilize some numerical skills to minimize those impacts on results.
The usual procedure for long wave and more specically solitary wave generation
consists in matching the paddle velocity at each position in time with the vertically
averaged horizontal velocity of the wave. For long waves, it is reasonable to assume that
the horizontal velocities are nearly constant over the depth. Equating the wave board speed
to the depth-averaged particle velocity on the face of the board gives the expression
dX
0
t
dt
uX
0;
t 12
in which X
0
(t) is the wave board location. It can be derived from continuity considerations
that the depth-averaged velocity for shallow waters of permanent form can be written as
(Svendsen 1984):
u x; t
Cgx; t
d gx; t
13
where C is wave celerity, and g(x, t) is sea surface elevation. Substituting into Eq. 12
gives:
dX
0
t
dt

CgX
0
; t
d gX
0
; t
14
Following Eq. 14, Goring and Raichlen (1980) assumed that the sea surface elevation
could be represented as a wave height multiplied by some function of phase angle; Syn-
olakis (1990) recommended integrating the Eq. 14 directly using the RungeKutta
numerical integration method. He stated this technique converges faster, particularly for
cnoidal waves. It is well known for a solitary wave at the position of a piston-type wave
board over the range -?\t \? that the wave prole can be written in the form:
g
s
x; t H sec h
2
h H sec h
2
k Ct X
0
15
where k

3H
4d
3
_
and C

gd H
_
. Substituting Eq. 15 into Eq. 14, one can have an
implicit equation for the wave board time history
X
0
t
H
kd
tan hk Ct X
0
: 16
The motion of the wavemaker was thus programmed to generate a solitary wave by
forcing a (horizontal) velocity eld. The authors adopted a procedure similar to that
introduced by Goring and Raichlen (1980); however, the authors followed the suggestion of
Dalrymple (private correspondence, 2011) that the total stroke duration time be estimated as
Nat Hazards
123
the time it takes for the wave paddle to reach 0.954 times its nal position. The compu-
tational domain, as shown in Fig. 4, has the following dimensions: length 6 m, width 0.2 m
and height 0.55 m with the beach slope 1:13. The length of the at bottom before the mildly
sloping beach is 1.0 m, and the still water depth is 0.3048 m. Table 1 summarizes the
incident wave conditions (one corresponds to a spilling breaker, the other to a plunging
breaker). Since the plunging breaker has stronger turbulence than spilling breaker, more
works are focused on plunging breaker. Based on this, a model of tsunami wave with
relative wave height H/d = 0.3 is produced. The initial condition was set by specifying the
particles motion near the computational entrance (see Fig. 5).
4 Numerical implementation
Four numerical schemes are implemented in resolving SPH: (1) PredictorCorrector
algorithm (Monaghan 1989), (2) Verlet algorithm (Verlet 1967), (3) Symplectic algorithm
(Leimkuhler et al. 1996) and (4) Beeman algorithm (Beeman 1976). The user can choose
one of them according to the needs of application. The Symplectic time integration
algorithm (Leimkuhler et al. 1996) is time reversible in the absence of friction or viscous
effects and hence represents a very attractive option for meshless particle schemes. First,
the values of density and acceleration are calculated at the middle of the time step as:
q
n1=2
a
q
n
a

Dt
2
dq
n
a
dt
; r
n1=2
a
r
n
a

Dt
2
dr
n
a
dt
; P
n1=2
a
B
q
n
1
2
a
q
0
_ _
c
1
_ _
17
where the superscript n denotes time step and t = nDt. In the second stage, momentum
change provides the velocity and hence the positions of particles at the end of each time step:
0 1 2 3 4 5 6 7 8 9
0
0.1
0.2
0.3
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Y(m)
X(m)
Initial Particle Configuration
Z
(
m
)
Fig. 4 Initial particle conguration and computational domain
Table 1 Wave parameters
Water depth (d), m 0.3048 0.1
Wave height (H), m 0.09144 0.015
Wave number (k) 1.4916 3.354
Total stroke duration time (t), s 1.0 1.109
Beach slope (h) 1:13 1:15
Breaker type Plunging Spilling
Nat Hazards
123
V
a
q
a
u
a

n1
V
a
q
a
u
a

n1=2

Dt
2
dV
a
q
a
u
a

n1=2
dt
; r
n1
a
r
n1=2
a

Dt
2
u
n1
a
18
where V is the volume of particle a. At the end of the time step,
dq
n1
a
dt
is calculated using the
updated values of u
n1
a
and r
n1
a
(Monaghan 2005). No matter which scheme is chosen,
time step control is dependent on the CFL condition, the forcing terms and the viscous
diffusion term. According to Monaghan and Kos (1999), the time step is constrained by
dt min 0:4
h
c
0
; 0:25

h
F
_
; 0:125
h
2
t
_ _
19
where c
0
10 U
max
and F are the characteristic velocity and force experienced by a
particle, respectively, U
max
being the maximum ow speed. The particle position at each
time step is expressed as:
dr
i
dt
_ _
a
u
i

a
: 20
For free surface ows, Monaghan (1994) suggests the use of XSPH equation to update
the particle position. This XSPH form is given by:
dr
i
dt
_ _
a
u
i

b
m
b
q
ab
u
i

b
u
i

a
W
ab
21
0 0.5 1 1.5 2 2.5
0
0.2
0.4
0.6
0.8
1
1.2
1.4
u (m/s)
T

(
s
)
wave paddle velocity
0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
0
0.2
0.4
0.6
0.8
1
1.2
1.4
X (m/s)
T

(
s
)
wave paddle trajectory
Fig. 5 Wavemaker moving velocity and trajectory during simulation
Nat Hazards
123
where 0:5 by default. This method moves the particle with a velocity that is close to
the average velocity occurring in its neighbourhood. This form helps maintain an orderly
movement of the particles and prevents penetration of particles associated with different
uids as in a multi-phase ow.
In reality, water waves propagate in a viscous uid over an irregular bottom of varying
permeability. A remarkable fact, however, is that, in most cases, the main body of the uid
motion is nearly irrotational. This is because the viscous effects are usually concentrated in
thin boundary layer near the bottom. The ow region under a water wave train can be
decomposed into two parts, namely the bottom boundary layer, within which the uid
viscous effect is signicant, and the ow outside the boundary layer where the viscous
effect is negligible. The thickness of the bottom boundary layer induced by a wave train
can be estimated by the following simple formula derived from dimensional analysis:
d $O

mT
p
22
where m is the kinematic viscosity of the uid, and T is the wave period. Given the typical
value of m = 1.0 9 10
-6
m
2
/s for water and T = 130 min (from wind waves to tsunami
waves), the corresponding boundary layer thickness ranges from 1.4 mm to 4.5 cm. In the
surf zone where the water depth is from tens of metres to hundreds of metres, the bottom
boundary layer thickness is much smaller than entire water depth and one can treat the
whole ow region as one entity. Three boundary conditions can be applied in SPH:
(1) dynamic boundary conditions (Crespo et al. 2007, Dalrymple and Knio 2000), (2)
repulsive boundary conditions (Monaghan and Kos 1999, Rogers and Dalrymple 2008) and
(3) periodic open boundary conditions. Numerical parameters used in the paper are
summarized in the Table 2 later.
5 Evaluation of the 3-D velocity eld
As mentioned previously, the wave ow is essentially irrotational before the wave breaks.
This can be seen clearly from Fig. 6. If the initial wave is irrotational, the wave will
maintain its original form as it propagates. Only when there are factors such as wave
breaking and wave-structure interaction, the initial wave form changes from irrotational to
Table 2 Numerical model
parameters
Model parameters Value
Order of the kernel 3rd order
Smoothing length ratio 0.03
Fluid particles 45,600
Moving wall particles 290
Wall particles 3,177
Total particle number 49,067
Initial particle spacing dt (m) 0.02
Lateral boundary condition Periodic
Other boundary condition Repulsive
XSPH parameter 0.5
Wall viscosity 8.0e-4
Nat Hazards
123
rotational. Due to the strong directionality of the solitary wave, the horizontal scale is
much larger than the vertical scale. The velocity magnitude in Y-direction, as well as in
Z-direction, is small and, therefore, negligible compared to the horizontal (X-direction)
velocity magnitude. It is evident that uid motion can be simplied into two dimensions
and the ow is irrotational, thus non-breaking wave can be modelled effectively and
efciently by using 2-D models. When the wave breaking initiates, the vertical velocity
obviously becomes signicant (Fig. 7) and the transverse velocity suddenly increases just
when the breaking wave front impinges on the water body in front of the wave. Moreover,
Fig. 6 shows that when wave breaks, the generation of the transverse velocity component
(red colour) has transformed the ow from 2-D to 3-D, while the vertical velocity
Fig. 6 Velocity variation in Y-direction (normal to the page) before and after wave breaking
Fig. 7 Vorticity variation in Y-direction (x
y
)
Nat Hazards
123
magnitude has been increased signicantly. Figure 8 shows that when the breaking wave
jet impinges the bottom of front wave body, the transverse velocity increases in both ways
(in Y and -Y-directions), something that transfers more momentum and energy into the
transverse direction. Figure 9 demonstrates that the momentum and energy are transferred
back and forth with a net transfer in the direction of wave propagation. The breaking
process shows, therefore, a strong transport and turbulent mix of energy in the entire uid,
and all ow variables in 3-D need to be considered at this stage. Finally, the variables of
interest in each direction are at the same order and hence become equally important.
Therefore, 3-D modelling is a must in order to capture the breaking wave process.
Fig. 8 Velocity variation in Y-direction (normal to the page) before and after breaking jet impinges the
bottom of front wave body
Fig. 9 Velocity direction and magnitude at the breaking front
Nat Hazards
123
6 Modelling results of breaking tsunami waves
Figure 2 shows snapshots of wave breaking from the wave initiation to slump down and
then impact the wave front at the beach side. The maximum horizontal velocity is located
at the free surface (top front). When the wave phase velocity surpasses the group velocity,
the wave tends to separate from the wave group to which it initially belonged. Once
separated, the wave moves forward in the way. Once breaking occurs, the overturning jet
hits the forward water surface and rebounds back with a net forward momentum as shown
in Figs. 3, 7, and 8. During this process, air bubbles are entrained and impinged into the
water, causing an energy ux transfer between the fast-running wave front and the slowly
moving water body. The rebounding jet repeatedly plunges and transits to a bore-like front
which moves further inshore. When the wave breaks and reforms several times, the energy
is nally distributed evenly among the whole water body during this period. However, the
horizontal velocity gradually attenuates as the front progresses, while the lateral velocity
and vertical velocity intensify during the course of the wave breaking. It is obvious that the
wave energy has partially transferred from horizontal direction to the lateral and vertical
directions except for the energy dissipation due to viscosity. The boundary friction and
surface resistance have been ignored in the present model but will be considered in the
future development of the model. Figure 10 displays three particles44,869 at the surface,
40,970 at the bottom and 60,660 in the middle depthwhich are chosen to reect their
density and pressure variation with time. The gure represents the random motion of the
water particle during the passage of wave. Figure 11 provides particle pressure and density
variation during simulation time (from t = 0 to t = 5 s ended at time step N = 198) for
the same particle shown in Fig. 10. From these gures, it is clear that the density variation
is quite small and within the assumption range (less than 1%), while the pressure is a
function of density, its perturbation reects the wave propagating process in the compu-
tational domain, with the maximum pressure occurrence at the moment of maximum wave
height. Once the wave passes by, the pressure uctuates around the hydrostatic pressure.
During the wave breaking and splash-up process, a signicant uctuation can be observed
from those gures. Compared with 2-D SPH simulation results (Lo and Shao 2002;
Khayyer et al. 2008), Figs. 2 and 3 demonstrate that the 3-D SPH modelling has very well
reproduced the plunging jet development and the resultant splash-up development, as
observed in the laboratory photographs of Li (2000). Figure 9 shows the velocity direction
and magnitude of the breaking front, and it clearly shows that due to the curl down of the
wave front, the water in the front of the wave moves back to form a portion of the curved
wave body, and then, it swirls up and sweeps down again into the wave body, while portion
of the water particles spins out of the swirl and splashes away from the body. The long
arrow represents the larger speed of water particles. This indicates that the splashed par-
ticles carry lots of energy away from the wave body. That energy is attenuated by the
interaction between the water particles and the air.
7 Examination of the 3-D vortices of the breaking waves
The splash-up is the result of the plunging jet penetration and the associated resulting
momentum and energy exchange between the fast moving overturning jet and the wedge-
shaped water located ahead of the wave. The splash-up is responsible for the generation of
large-scale vortices and plays a signicant role in the momentum transfer and energy
dissipation. Temporal and spatial changes in each vorticity component are investigated to
Nat Hazards
123
visualize the generation and evolution of 3-D large-scale eddies. The vortices are com-
puted as an angular velocity cross-multiplied by particle mass in three directions
x
x
; x
y
; x
z
. The vorticity of each particle is dened as (Monaghan 1992):
x
i

m
j
u
i
u
j
r
i
W
ij
23
Figure 7 shows several vortex pictures: it can be seen that the positive vorticity
(clockwise turning) and negative vorticity (counterclockwise turning) dominate the
rebounding jets. The vortex rollers associated with the transverse eddy are stretched
during propagation of the breaking wave front. The fully stretching eddies are nally cut-
off and decomposed into small fractions due to the instability of the vortex structures,
and splashed particles are successfully reproduced and modelled. Hence, the capacities of
the present SPH model in the simulation of large wave deformations and fragmentation
have been demonstrated through a qualitative comparison with the laboratory photo-
graphs (Li 2000).
Fig. 10 Variation of the three particles (44,869 at the surface, 40,970 at the bottom and 60,660 in the
middle depth) velocity
Nat Hazards
123
Fig. 11 Particle pressure and density variation during simulation time (from t = 0 to t = 5 s, ended at time
step N = 198) for the particles numbered in Fig. 10
Nat Hazards
123
8 Conclusion
The computational results show that the 3-D LES-SPH model can better capture the
breaking wave characteristics compared to other traditional (2-D) methods. The evolution
of the breaking surface prole and large-scale eddies during wave breaking has been
successfully tracked with ease and reliability and was shown to be similar to those
observed during the experiments of Li (2000). The spatial evolution of the breaking wave
is also presented. The detailed mechanisms of turbulence transport and vorticity dynamics
are shown to be well reproduced. It has been conrmed that the 3-D features of breaking
wave can be accurately modelled with a 3-D model. In addition to this work, further
investigations will focus on the accurate free surface tracking of discontinuous waves and
of the air-entrained multi-phased ow. The dissipative ability of Smagorinsky turbulence
closure in combination with rather dissipative SPH model will be further examined, as
several few researchers have shown that such models are highly dissipative.
References
Battjes JA (1974) Surf similarity. In: Proceedings of 14th coastal engineering conference ASCE, vol 1,
pp 466480
Beeman D (1976) Some multistep methods for use in molecular dynamics calculations. J Comput Phys
20(2):130139
Colagrossi A, Landrini M (2003) Numerical simulation of interfacial ows by smoothed particle hydro-
dynamics. J Comput Phys 191:448475
Crespo AJ, Gomes-Gesteira M, Dalrymple RA (2007) Boundary conditions generated by dynamic particles
in SPH methods. CMC Comput Mater Contin 5(3):173184
Dalrymple RA, Knio O (2000) SPH modeling of water waves. In: Proceedings of coastal dynamics. Lund
Dalrymple RA, Rogers BD (2006) Numerical modeling of water waves with the SPH method. Coast Eng
53:141147
Goring D, Raichlen F (1980) The generation of long waves in the laboratory. In: Proceedings of the 17th
coastal engineering conference, American Society of civil engineers, vol 1, pp 763783
Gotoh H, Shibihara T, Sakai T (2001) Sub-particle-scale model for the MPS methodLagrangian ow
model for hydraulic engineering. Comput Fluid Dyn J 9(4):339347
Grilli ST, Sevnden IA, Subramanya R (1997) Breaking criterion and characteristics for solitary waves on
slopes. J Waterway Port Coast Ocean Eng 123(3):102112
Hirt CW, Nichols BD (1981) Volume of uid (VOF) method for the dynamics of free boundaries. J Comput
Phys 39(1):201225
Kennedy AB, Chen Q, Kirby JT, Dalrymple RA (2000) Boussinesq modeling of wave transformation,
breaking and runup I 1D. J Waterway Port Coast Ocean Eng 126:3947
Khayyer A, Gotoh H, Shao SD (2008) Corrected Incompressible SPH method for accurate water-surface
tracking in breaking waves. Coast Eng 55:236250
Leimkuhler B, Reich S, Skeel RD (1996) Integration methods for molecular dynamic. IMA Vol Math Appl
82:161
Li Y (2000) Tsunamis: non-breaking and breaking solitary wave run-up. Rep. KH-R-60, W. M. Keck
laboratory of hydraulics and water resources. California Institute of Technology, Pasadena, CA
Li Y, Raichlen F (2003) Energy balance model for breaking solitary wave runup. J Waterways Ports Coast
Ocean Eng ASCE 129(2):4759
Lin PZ, Liu PLF (1998) A numerical study of breaking waves in the surf zone. J Fluid Mech 359:239264
Liu GR, Liu MB (2003) Smoothed particle hydrodynamics: a meshfree particle method. World Scientic,
Singapore, p 472
Lo E, Shao SD (2002) Simulation of near-shore solitary wave mechanics by an incompressible SPH method.
Appl Ocean Res 24:275286
Madsen PA, Sorensen OR, Schaffer HA (1997) Surf zone dynamics simulated by a Boussinesq model, I.
Model description and cross-shore motion of regular waves. Coast Eng 32:255287
McCowan J (1894) On the highest wave of permanent type. Philos Mag 38:351357
Nat Hazards
123
Monaghan JJ (1989) On the problem of penetration in particle methods. J Comput Phys 82(1):115
Monaghan JJ (1992) Smoothed particle hydrodynamics. Annu Rev Astron Astrophy 30:543574
Monaghan JJ (1994) Simulating free surface ows with SPH. J Comput Phys 110:399406
Monaghan JJ (2005) Smoothed particle hydrodynamics. Rep Prog Phys 68:17031759
Monaghan JJ, Kos A (1999) Solitary waves on a Cretan beach. J Waterway Port Coast Ocean Eng
125(3):145154
Rogers BD, Dalrymple RA (2008) SPH modeling of tsunami waves, advances in Coastal and ocean
engineering, vol 10. Advanced numerical models for tsunami waves and runup. World Scientic,
Singapore
Shao SD, Ji CM (2006) SPH computation of plunging waves using a 2-D sub-particle scale (SPS) turbulence
model. Int J Numer Methods Fluids 51:913936
Svendsen IA (1984) Wave heights and setup in a surfzone. Coast Eng 8:303329
Svendsen IA, Veeramony J, Bakunin J, Kirby JT (2000) Analysis of the ow in weakly turbulent hydraulic
jumps. J Fluid Mech 418:2557
Synolakis CE (1990) Generation of long waves in laboratory. J Waterway Port Coast Ocean Eng Am Soc
Civil Eng 116(2):252266
Ting FC, Kirby JT (1995) Dynamics of surf-zone turbulence in a strong plunging breaker. Coast Eng
24:177204
Verlet L (1967) Computer experiments on classical uids: I. Thermo-dynamical properties of Lennard
Jones molecules. Phys Rev 159:98103
Watanabe Y, Saeki H (1996) Numerical analysis of breaking waves, hydraulic engineering software.
Comput Mech Publ 4:387394
Watanabe Y, Saeki H (1999) Three-dimensional large eddy simulation of breaking waves. Coast Eng J
41(3&4):281301
Wendland H (1995) Computational aspects of radial basis function approximation. Elsevier, Germany
Williams JM (1981) Limiting gravity waves in water of nite depth. Philos Trans R Soc 302:139188
Xie JS, Nistor I, Murty T, Shukla T (2007) Numerical modelling of the Indian Ocean tsunamia review. Int
J Ecol Dev IJED 3(S05):7188
Nat Hazards
123

You might also like