You are on page 1of 16

Available online at www.sciencedirect.

com

Coastal Engineering 55 (2008) 506 – 521


www.elsevier.com/locate/coastaleng

Alternative forms of the higher-order Boussinesq equations:


Derivations and validations
Z.L. Zou ⁎, K.Z. Fang
State Key Laboratory of Coastal and Offshore Engineering, Dalian University of Technology, Dalian 116024, PR China
Received 18 August 2006; received in revised form 8 January 2008; accepted 13 February 2008
Available online 2 April 2008

Abstract

An alternative form of the Boussinesq equations is developed, creating a model which is fully nonlinear up to O(μ4) (μ is the ratio of water
depth to wavelength) and has dispersion accurate to the Padé [4,4] approximation. No limitation is imposed on the bottom slope; the variable
distance between free surface and sea bottom is accounted for by a σ-transformation. Two reduced forms of the model are also presented, which
simplify O(μ4) terms using the assumption ε = O(μ2/3) (ε is the ratio of wave height to water depth). These can be seen as extensions of Serre's
equations, with dispersions given by the Padé [2,2] and Padé [4,4] approximations. The third-order nonlinear characteristics of these three models
are discussed using Fourier analysis, and compared to other high-order formulations of the Boussinesq equations. The models are validated against
experimental measurements of wave propagation over a submerged breakwater. Finally, the nonlinear evolution of wave groups along a horizontal
flume is simulated and compared to experimental data in order to investigate the effects of the amplitude dispersion and the four-wave resonant
interaction.
© 2008 Elsevier B.V. All rights reserved.

Keywords: Boussinesq equations; Wave propagation; Nonlinear waves

1. Introduction and Schäffer (1998) (hereafter referred to as MS98) used the


fourth-order polynomial in z of velocity potential to derive the
When water waves propagate towards the coast, they present fourth-order Boussinesq equations. Although these higher-order
stronger nonlinear behavior and very complicated motions. velocity distributions in vertical direction are necessary for the
Two-dimensional Boussinesq equations are commonly used to prediction of fluid kinematics (Gobbi et al., 2000), the
deal with these problems, since they take wave nonlinearity into derivation of the equations actually does not necessarily need
account and demand relatively little computation. In the effort to the same order accuracy of the velocity distribution as that of the
make Boussinesq-type equations more accurate and efficient, a derived equations. For example, Serre (1953) assumed a
variety of new forms have been developed over the past decade uniform horizontal velocity (i.e., accurate to O(μ0)) and linear
(e.g. Madsen and Sørensen, 1992; Nwogu, 1993; Schäffer and vertical velocity (accurate to O (μ2)), obtaining a Boussinesq
Madsen, 1995; Wei et al., 1995; Madsen and Schäffer, 1998; model with dispersion accurate to O(μ2) and full nonlinearity
Agnon et al., 1999; Gobbi et al., 2000; Zou 1999, 2000; Madsen up to O(μ2). The equations of Green and Naghdi (1976)
et al., 2002, 2003; Lynett and Liu, 2004). (referred to hereafter as the G–N equations) can be obtained
The derivation of Boussinesq-type equations usually starts using the same velocity profile, and have the same level of
with the establishment of a vertical distribution of velocity or accuracy. It may be expected that extending this velocity
velocity potential. For example, Gobbi et al. (2000) and Madsen distribution to the next higher order would give equations with
dispersion accurate to O(μ4) and full nonlinearity up to O(μ4).
This is the goal of the present study, which will use the
⁎ Corresponding author. Tel.: +86 411 4706053; fax: +86 411 4708526. σ-transformation to develop an alternative form of the fully
E-mail address: zlzou@dlut.edu.cn (Z.L. Zou). nonlinear Boussinesq equations accurate to O(μ 4 ).
0378-3839/$ - see front matter © 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.coastaleng.2008.02.001
Z.L. Zou, K.Z. Fang / Coastal Engineering 55 (2008) 506–521 507

Numerical simulations of nonlinear wave motions depend on 8 validate the new Boussinesq models against experimental
the accuracy of linear dispersion, shoaling, and the nonlinear data.
characteristics of the Boussinesq model. In the existing
Boussinesq models, linear dispersion and shoaling have been 2. The vertical velocity distributions
greatly improved by deriving a set of higher-order Boussinesq
equations (e.g. Wei et al., 1995; Gobbi et al., 2000), by applying Cartesian coordinates (x, y, z) are used, with the origin at the
enhancement techniques (Schäffer and Madsen, 1995; Madsen still water level and the z-axis pointing upwards. The following
and Schäffer, 1998), or by using a computation velocity (Zou, dimensionless variables are introduced:
2000). The second- and third-order transfer functions and the
pffiffiffiffiffiffiffi
amplitude dispersion, however, are still difficult to enhance for x y z gh0
higher values of μ. MS98 presented enhancements to different xV ¼ ; yV ¼ ; zV ¼ ; tV ¼ t;
L0 L0 h0 L0
forms of the Boussinesq model, including a fourth-order (μ4),
g 1 1
fully nonlinear set of Boussinesq equations expressed in terms gV ¼ ; uV ¼ ðu V; vVÞ ¼ pffiffiffiffiffiffiffiðu; vÞ; w V ¼ pffiffiffiffiffiffiffiw; ð2:1Þ
of the velocity at an arbitrary vertical level. Their results show A0 e gh0 Ae gh0
that the amplitude dispersion and the third-order transfer h p
hV ¼ ; pV ¼
function of the surface elevation have good accuracy only for h0 qgA0
kh b 0.5. The range of the corresponding second-order transfer
function is a bit larger, with good accuracy up to kh = 1. The where a prime indicates the dimensionless version of a physical
Boussinesq equations developed by Gobbi et al. (2000), which variable. t is the time, u and w are the horizontal and vertical
are fully nonlinear up to O( μ 4 ) and use a Padé [4,4] velocity components of the fluid, η is the free surface elevation,
approximation for the linear dispersion, have similar properties. g is the gravitational acceleration, ρ is the fluid density, p is the
This kind of limitation on the range of accuracy for higher-order fluid pressure, and finally h0, A0, and L0 denote a characteristic
transfer functions and amplitude dispersions has inspired efforts water depth, wave amplitude, and wavelength respectively.
to develop other formulations, for example that of Agnon et al. μ = h0 /L0 is the dispersion parameter, and ε = A0 / h0 is the
(1999) and Madsen et al. (2002, 2003). This model consists of nonlinearity parameter. In the present study, μ is assumedpffiffiffiffiffiffiffito be
four partial differential equations from the dynamic and small but
pffiffiffiffiffiffiffi the value of ε is arbitrary. The quantities e gh0 and
kinematic surface and bottom boundary conditions with Ae gh0 in Eq. (2.1) are the horizontal and vertical speeds u′ and
truncated solutions of Laplace equation. A sophisticated w′ of water waves in shallow water, so the dimensionless
numerical scheme is generally needed for this model, especially velocities u and w are of O(1) for the pure wave case. As the
for the horizontal 2D case (Fuhrman et al., 2004). As these following derivations make use of dimensionless variables
improvements to the accuracy of the Boussinesq formulation exclusively, the prime will now be dropped for simplicity of
are gradually made, more complicated and interesting physical notation.
phenomena have been considered. Examples include surf beat We will use the continuity equation
and net circulation (Madsen et al., 1997), side-band instability
Aw
(Madsen et al., 2002), and crescent waves (Fuhrman et al., þ j  u ¼ 0;  h b z b eg ð2:2Þ
2004). The equations developed in the present paper still belong Az
to the type of Wei et al. (1995), Madsen and Schäffer (1998) and and the zero horizontal vorticity equation
Gobbi et al. (2000), but are in different form and presented with
further investigations, validations and applications. The fourth- Au
order, fully nonlinear Boussinesq equations derived in the  A2 jw ¼ 0;  h b z b eg ð2:3Þ
Az
present study and their reduced forms are used to simulate the
amplitude dispersion effect and the four-wave resonant to derive the vertical velocity distributions. Note that Eqs. (2.2)
 
interaction by considering the nonlinear evolution of wave and (2.3) use the horizontal gradient operator j ¼ Ax A A
; Ay .
groups in a wave flume. The enhanced equations in terms of The boundary conditions for these two equations are
depth-averaged velocity presented by MS98 are also used for
this simulation in order to investigate the effect of different Ag
nonlinearity accuracies on the numerical results. w¼ þ eu  jg; z ¼ eg and ð2:4Þ
At
The paper is organized as follows. The vertical distributions
of velocity components are given in Section 2. Section 3 derives
an alternative form of the fully nonlinear Boussinesq equations w ¼ jh  u; z ¼ h ð2:5Þ
up to O(μ4) by using the σ-transformation. Section 4 presents a
reduced form of the equations derived in Section 3, which have at the free surface and bottom respectively.
the Padé [2,2] dispersion and fully nonlinearity to O(μ2), In order to deal with the variable free surface elevation and
Section 5 develops two sets of the enhanced equations with water depth, we introduce the σ-transformation:
Padé [4,4] dispersion and fully nonlinearity to O( μ2) and O
(μ4), respectively. The characteristics of the resulting equations z  eg
X ¼ x; Y ¼ y; Z¼ ; T ¼ t; ð2:6Þ
are investigated in Section 6 by Fourier analysis. Sections 7 and d
508 Z.L. Zou, K.Z. Fang / Coastal Engineering 55 (2008) 506–521

where d = h + εη. In these new coordinates, the fluid domain Continuing the cycle, we substitute this expression into
− h b z b εη becomes fixed: − 1 b Z ≤ 0. The relations between Eq. (2.8) and integrate with respect to Z. This gives us an
derivatives in the two coordinate systems are expression for w accurate to O(μ2):
A A 1 A  
¼  ðeg þ ZdT Þ ; w ¼ wð0Þ þ A2 wð2Þ þ O A4 ; ð2:16aÞ
At AT d T AZ
P 1 P P  A
j ¼ j  e jg þ Z jd ; and ð2:7Þ in which
d AZ
A 1 A P P P P P P

¼ ; wð0Þ ¼  j  ðh uÞ þ Zd j  u þ eg j  u and ð2:16bÞ


Az d AZ
P A A  
where j ¼ AX ; AY . The subscript T indicates a partial 1P 1P   P
derivative with respect to T. In the new coordinates, Eqs. (2.2) wð2Þ ¼ Zd j  ðdG Þ  j  d 2 K  e jg  G
and (2.3) read 2 6 
1 2 2 P P  1 3 3P
þ Z d j  G  e jg  K þ Z d j  K
2 6
1 Aw P 1 P P  Au
P
þ j  u  e jg þ Z jd ¼ 0;  1bZb0  e jg 
1 1 2
dG  d K :
d AZ d AZ 2 6
(2.16c)
ð2:8Þ
ð2:16cÞ
 
1 Au P 1 P P  Aw
¼ A2 jw  e jg þ Z jd ;  1bZb0: ð2:9Þ Expressions (2.15a), (2.15b) and (2.16a), (2.16b), (2.16c) are
d AZ d AZ
the vertical distributions of the horizontal and vertical velocities
The depth-averaged horizontal velocity can be expressed in u and w, accurate to O(μ2). In the next section, we will find that
the new coordinates as velocity expressions of this order can be used to formulate a
governing equation accurate to O(μ4).
Z 0
P
u¼ udZ: ð2:10Þ
1 3. Derivation of the governing equations fully nonlinear to
O(μ4)
Integrating Eq. (2.9) with respect to Z gives
  The dimensional forms of the momentum equations are
u ¼ U þ O A2 ; ð2:11Þ
Au Au jp
þ ðu  jÞu þ w ¼ ð3:1Þ
where U = U(X,Y,T) is an unknown function. Applying integrals At Az q
(2.10) to (2.11) yields U = ū + O(μ2), and substituting this result
back into Eq. (2.11) gives u = ū + O(μ2). Substitution of this Aw Aw 1 Ap
þ ðu  jÞw þ w ¼  g: ð3:2Þ
result into Eq. (2.8) and integration with respect to Z from − 1 to At Az q Az
Z, while applying the bottom boundary condition (2.5), yields
P
  These can be expressed using dimensionless variables (2.1)
P P
w ¼  j  ðhPuÞ þ dZ j  P
u þ eg j  Pu þ O A2 : ð2:12Þ and the new coordinates shown in Eq. (2.6) as

This can now be used to specify Eq. (2.9) more clearly. We  P


uT þ e u  j u
introduce the notation  
P P P P P P P P P
e P Z P 
C ¼ j j  ðh u Þ þ eg j j  u; K ¼ jj  u ð2:13Þ þ w  gT  eu  jg  dT þ eu  jd uZ
d e
and integrate with respect to Z, which gives P 1 P P 
¼  jp þ e jg þ Z jd pZ ð3:3Þ
d
1
u ¼ U T ð X ; Y ; T Þ  A2 ZdC þ Z 2 d 2 K : ð2:14Þ
2  P
A2 wT þ eA2 u  j w
To determine U⁎(X,Y,T), we integrate Eq. (2.14) with respect to  
Z from −1 to 0. This gives an expression for u accurate to O(μ2): eA2 P Z P 
þ w  gT  eu  jg  dT þ eu  jd wZ
P   d e
u ¼ u þ A2 uð2Þ þ O A4 ; ð2:15aÞ
1 1 ð3:4Þ
¼  pZ 
in which d e
 
ð2Þ 1 1 1 2
u ¼  Z þ dG þ Z  d K :
2
ð2:15bÞ where the subscripts T and Z denote partial derivatives with
2 2 3 respect to T and Z.
Z.L. Zou, K.Z. Fang / Coastal Engineering 55 (2008) 506–521 509


In the new coordinate system, the dimensionless depth-
integrated continuity equation reads
pð4Þ ¼ dD f
1 P
2
Z jg A  B
1
3
 
1 2 1P 1P P A
 Z d j  A  j  B  e jg 
Ag P P 2 2 6 d
þ j  ðd uÞ ¼ 0: ð3:5Þ  
AT
A
Defining the derivative operator D ¼ AT þeP
u  j, Eq. (3.5)
P
1 3 2 P
 Z d j
6
A
d
P
 e jg  2
B
d
1 P
 Z 4d3 j  2
24 d
B
g
   
can be written as
P P P

f
1 1
e Zd ðZ þ 1ÞA þ Z 2  1 B 
2 3
 A
d
P P P
þ e jg j  u
P    
Dg ¼  j  ðh u Þ þ eg j  u ð3:6Þ 1 1 1 1 1 B P P P
þZ 2 d Zþ Aþ Z2  B  þ jd j  u
3 4 4 2 3 d
or  

Dd ¼ ed j  u:
P P
ð3:7Þ
1 2
 Z d
2
1
3

1 P
2
j  ðdAÞ þ
1 1 2
6 2
P P P
Z  1 j  ðdBÞ j  u g
ð3:12Þ
From Eqs. (3.6) and (3.7), w(0) given by Eq. (2.16b) can be
written as with using the notations
P P P P P P

Z A ¼ dG ¼ d j j  ðh u Þ þ eg j j  u ð3:13Þ
wð0Þ ¼ Dg þ Dd: ð3:8Þ
e P P P
B ¼ d 2 K ¼ d 2 j j  u: ð3:14Þ
As will be shown in the following derivation, this expression
is helpful in simplifying Eqs. (3.3) and (3.4). In particular, after The governing equation can now be obtained by substituting
using this new expression for w we will see that the terms in the expressions (2.15a), (2.15b), (2.16a), (2.16b), (2.16c) and
brackets of Eqs. (3.3) and (3.4) will be of an order higher than (3.10a), (3.1b,c), (3.10d) into (3.3) and integrating the equation
εμ4. with respect to Z from − 1 to 0. During this process, the
In order to derive the governing equations, we first need an following relations will be useful. For the pressure terms, we
expression for the fluid pressure. This can be obtained by have
substituting the velocity expressions derived in Section 2 into
Eq. (3.4), and applying Eq. (3.8) as follows: Z  
0
P 1 P P 
 jp  e jg þ Z jd pZ dZ
1 d
1 1 P Z 0
 pZ ¼ A2 Dw þ þ eA4 uð2Þ  jwð0Þ P 1  P P 
Z¼0
d e ¼ j pdZ þ e jg þ Z jd p Z¼1
h  P i
P  1 Awð0Þ 1Z d
þ eA w  uð2Þ  e jg þ Z jd
4 ð2Þ
1P 0
  d AZ  jd  pdZ
þ O A6 : ð3:9Þ d 1
Z 0 P
1P jh
¼ j d pdZ þ pjZ¼1 : ð3:15Þ
Integrating with respect to Z from 0 to Z gives d 1 d
 
p ¼ pð0Þ þ A2 pð2Þ þ A4 pð4Þ þ O A6 ; ð3:10aÞ The boundary condition p = 0 (Z = 0) has been used in this
in which derivation. For the convection terms, we have

Z Z 0
0 P P P P
p ð0Þ 1
¼ dZ; p ð2Þ
¼ dD ð0Þ
w dZ ; and ð3:10b; cÞ u  j udZ ¼ u  j u
e 1 Z 0 
Z
P  
þA4 uð2Þ  j uð2Þ dZ þ O A6 :
Z 0 Z 0 1
P
p ð4Þ
¼ dD ð 2Þ
w dZ þ ed fu
ð2Þ
 jw ð0Þ ð3:16Þ
Z Z
h  P i
P  P P
 wð2Þ  uð2Þ  e jg þ Z jd j  u gdZ:
Using Eqs. (3.13) and (3.14), the last integral of Eq. (3.16)
can be expressed as
ð3:10dÞ
Z 0  1  P
Substituting the expressions for w(0), u(2), and w(2) into uð2Þ  j uð2Þ dZ ¼ A j A
Eqs. (3.10a), (3.10b,c), (3.10d) yields 1 12
1  P  P

 A j Bþ B j A
 24
ð2Þ
P P P P
1 2 P P 1  P
p ¼ dD Z j  ðh u Þ þ eg j  u þ Z d j  u þ B j B
2 45
ð3:11Þ ð3:17Þ
510 Z.L. Zou, K.Z. Fang / Coastal Engineering 55 (2008) 506–521

P
The other convection term can be written as In the above expressions, j has been replaced by ▿ for the
P
Z   terms which are independent of Z, since we have j ¼ j from
1 0
P 1  P 
Eq. (2.7) when a variable is independent of Z and its derivative
w  gT  eu  jg  Z dT þ eu  jd uZ dZ
d 1 e with respect to Z is zero. Eqs. (3.20) and (3.21a), (3.21b),
 Z¼0
1 P 1  P  (3.21c), (3.21d) are fourth-order Boussinesq equations which
¼ u w  gT  eu  jg  Z dT þ eu  jd
d e Z¼1 are fully nonlinear up to O(μ4) and have embedded linear
Z 0  dispersion accurate to O( μ4). MS98 derived a set of higher-

1
d 1
u
Aw Au

AZ AZ
 P P  1
 e jg þ Z jd  dT þ eu  jd
e
P 
dZ g order Boussinesq equations accurate to O(εμ4) in terms of
Z 0  the depth-averaged velocity (the Eqs. (3.2) and (3.7) in their
¼
1 P
u j  ðduÞdZ  P
P
u j  ðd P
uÞ ð3:18Þ paper), in which only the O(εμ4) nonlinear terms at O(μ4) are
d 1
Z retained under the assumption of ε = O(μ). The present
1 0 ð2Þ    
¼ A4 u j  duð2Þ dZ þ O A6 : formulation for P(4) (3.21d) includes all of the O(μ4) terms,
d 1 and is written in a compact form. The expansion of P(2) (3.21c)
is also written in compact form, its expansion in terms of ε was
Similarly to Eq. (3.17), we can express the last integral as
given by MS98 (see (3.7) in their paper). A similar expansion of
Z   P(4) would be too lengthy, so we will not pursue this result here.
0
P 1 P 1 P P

uð2Þ j  duð2Þ dZ ¼ A j  ðdAÞ  A j ðdBÞ þ B j  ðdAÞ


1 12 24 If the O(μ4) terms are omitted, these equations reduce to the
1 P ð3:19Þ G–N equations or to Serre's equations for a horizontal bottom.
þ B j  ðdBÞ:
45 To see this, note that P(2) can be expressed as:
Using results (3.10a), (3.10b,c), (3.10d) to (3.19), we can  
express the governing equations in terms of the depth-averaged
horizontal velocity ū:
f 1 P 1 P
P ð2Þ ¼  dD jðdj  u Þ þ jðjh  u Þ
3 3
 
P
gt þ j  ð d u Þ ¼ 0 ð3:20Þ
1 P 2 P P
þ jhDðdj  u Þ þ jgD dj  u þ jh  u
6 3 g
P P P   ð3:22Þ
ut þ eð u  jÞ u þ jg þ eA4 M ð4Þ ¼ A2 P ð2Þ þ A4 P ð4Þ þ O A6 ;
ð3:21aÞ or alternatively
where
1

ð4Þ 1 1 1 P ð2Þ ¼ fdj D2 ð2g  hÞ


M ¼ ðA  jÞA  ½ðA  jÞB þ ðB  jÞA þ ðB  jÞB
12 24 45
6 ð3:23Þ
1 1 1 þjð4g þ hÞD g  jð2g  hÞD hg
2 2
þ Aj  ðdAÞ  ½AjðdBÞ þ Bj  ðdAÞ þ Bj  ðdBÞ
12d 24d 45d
ð3:21bÞ by using Eqs. (3.8) and (3.9). Thus, Eqs. (3.20) and (3.21a),
  (3.21b), (3.21c), (3.21d) can be seen as an extension of the G–N
1 2 P P equations to order O(μ4). Notice that this extension is produced
P ð2Þ ¼ j d 2 D dj  u þ jh  u
2d 3 by a perturbation of the horizontal velocity from the uniform
1 P P vertical distribution to a second-order distribution in μ2 (2.15a).
 jhD dj  u þ jh  u ð3:21cÞ
2 So although the new equations go up to the fourth order, the

f ½
1 1 1 A 1 velocity distributions needed to derive them are only second-
P ð4Þ ¼ j d 2 D dj  A  d 2 j  þ ejg  A order (see Eqs. (2.15a), (2.15b) and (2.16a), (2.16b), (2.16c)).
d 12 24 d 12

g
1 1 3 B 1 A weak point of Eqs. (3.20) and (3.21a), (3.21b), (3.21c),
 dj  B þ d j  2  ejg  B (3.21d), as pointed out by MS98, is that the embedded linear
36 120 d 24

½ 
1 1 A 1 1 B dispersion has a singularity at the dimensionless wave number
jhD dj  A  d 2 j   dj  B þ d 3 j  2 kh ≈ 4.2. This produces a numerical instability in the equations.
4 6 d 12 24 d


1 1 A In the following two sections, we solve this problem by
e j d2 A  B  þ ejgj  P u ð3:21dÞ presenting two working models which are obtained by im-
12d 2 d

g
1 B 1 1 B proving the above model.
þ B þ jdj  Pu þ e jh A  B  þ jdj  P
u
30 d 12 2 d
   4. Equations with Padé [2,2] dispersion and fully nonlinear
1 1 7 P
e j d 2 j  ðdAÞ  j  ðdBÞ j  u to O(μ2)
12d 2 30
 
1 1 P
þe jh j  ðdAÞ  j  ðdBÞ j  u One way to remove the singularity in dispersion relation of
12 2
Eqs. (3.20) and (3.21a), (3.21b), (3.21c), (3.21d) is to use the
in which dispersion enhancement technique proposed by MS98, which
will be described in the following section. As an alternative, we
A ¼ d ½jj  ðh P
u Þ þ egjj  P
u ; B ¼ d 2 jj  P
u; d ¼ h þ eg: here solve this problem for a reduced vision of the Eqs. (3.20)
Z.L. Zou, K.Z. Fang / Coastal Engineering 55 (2008) 506–521 511

and (3.21a), (3.21b), (3.21c), (3.21d) by introducing the linear dispersion embedded in the equations is improved by
assumption of ε = O( μ 2/3 ) to omit the higher-order terms of using the velocity at an arbitrary z level. Their numerical results
O(εμ4 ): the term M (4) in Eq. (3.21a) and the nonlinear terms of for solitary wave propagation showed the justification of this
P(4) . Then, P(4) is reduced to the linear form kind of models, i.e. the fully nonlinearity up to O(μ2) can make
the model be applicable for the strong nonlinear wave motions
f 1 3 P 1
P ð4Þ ¼  h jj  jj  ðh ut Þ þ h2 jj near breaking point. Compared to models that are fully
24 12 nonlinear up to O(μ4), this kind of model is rather simple in
P 1 4 P
 ½hjj  ðh ut Þ þ h jj  jj  ut form. As it does not contain fourth-order spatial derivatives, it is
120
1 P
also easier to deal with in numerical simulations.
 h2 jj  h2 jj  ut : ð4:1Þ
36
5. Equations with Padé [4,4] dispersion and fully nonlinear
The singularity discussed above can now be removed by to O( μ4) or to O( μ2)
using the following lower-order equation in Eq. (4.1):
Eqs. (3.20) and (3.21a), (3.21b), (3.21c), (3.21d) have an
P P P 1 P
ut þ eð u  jÞ u þ jg ¼ A2 j½j  ðh ut Þ embedded linear dispersion relation accurate to O( μ4), this
2 accuracy is not good enough for moderate water depths. In this
1 P
 A2 h2 jj  ut section, we improve the accuracy of the dispersion equations
6 
þ O eA2 ; A4 : ð4:2Þ by applying the technique of MS98. The following terms of
order O( μ6) containing four free parameters (α1,α2,β1,β2) of
Eq. (4.2) is obtained by retaining the leading nonlinear and order O(1) are first obtained by applying the operator ▿(▿·) to
dispersion terms in Eqs. (3.21a), (3.21b), (3.21c), (3.21d). Then Eq. (3.21a) and multiplying the result by (α2 − α1)μ2 h2 ,
the fourth-order spatial derivatives in Eq. (4.1) can be applying the operator ▿(▿ · h) to Eq. (3.21a) and multiplying
transformed into third-order spatial derivatives, and we obtain the result by − α2μ2h2, applying the operator ▿(▿2(▿)) to
the following expression for P(4): Eq. (3.21a) and multiplying the result by (β1 − β2)μ4h4, applying
the operator ▿(▿2(▿ · h)) to Eq. (3.21a) and multiplying the
1 1 C 1 result by β2μ4h3, respectively.
P̂ ð4Þ ¼ h2 jj  C  h3 jj  þ h4 jj
6 8 h 40 n o
P P P
C 1 ða2  a1 ÞA2 h2 jj  ut þ jg þ eð u  jÞ u  A2 P ð2Þ ð5:1Þ
 2 þ h3 jj  
h 48 ¼ O A6
P P
1
 jðjh  ut Þ þ jhj  ut  h4 jj n h io

80 P P P
a2 A2 hjj  h ut þ jg þ eð u  jÞ u  A2 P ð2Þ ð5:2Þ
1 P P

 jðjh  ut Þ þ jhj  ut ; ð4:3aÞ  


h ¼ O A6
where
ðb1  b2 ÞA4 h4 jj2 j  f P P P
ut þ jg þ eð u  jÞ u g ð5:3Þ
P P P  6
C ¼ ut þ eð u  jÞ u þ jg: ð4:3bÞ ¼O A

P P P  
The resulting momentum equation together with the b2 A4 h3 jj2 j  fh½ ut þ jg þ eð u  jÞ u g ¼ O A6 : ð5:4Þ
continuity Eq. (3.20) forms a new set of equations
Adding them to the momentum Eq. (3.21a) yields the
P
gt þ j  ð d u Þ ¼ 0 ð4:4Þ following set of the enhanced equations
P
ð4Þ gt þ j  ð d u Þ ¼ 0 ð5:5Þ
ut þ eð u  jÞ u þ jg ¼ A2 P ð2Þ þ A4 P ̂
P P P
 
þ O eA4 ; A6 : ð4:5Þ P P P
ut þ eð u  jÞ u þ jg þ eA4 M ð4Þ
ð4Þ
 
In which P ̂ is given by Eq. (4.3a). This set of equations ¼ A2 P Vð2Þ þ A4 P Vð4Þ þ O A6 ; ð5:6aÞ
does not have singularity in the dispersion relation, this can be
seen from its following dispersion relation where
 
x2 1 þ k 2 h2 =15 P Vð2Þ ¼ 1 þ A2 L1 P ð2Þ  L1 C; ð5:6b; cÞ
¼ : ð4:6Þ
hk 2 1 þ 2k 2 h2 =5
P Vð4Þ ¼ P ð4Þ  L2 C
Hereafter, Eqs. (4.4) and (4.5) will be referred to as
BouN2D2 (N2 denotes fully nonlinear up to O(μ2) and D2 L1 uða2  a1 Þh2 jj  a2 hjj  h ð5:7aÞ
denotes Padé [2,2] dispersion). Wei et al. (1995) developed a
model with the same nonlinearity and dispersion, for which the L2 uðb1  b2 Þh4 jj2 j  þb2 h3 jj2 j  h ð5:7bÞ
512 Z.L. Zou, K.Z. Fang / Coastal Engineering 55 (2008) 506–521

and C is given by Eq. (4.3b). The parameters (α1,β1,α2,β2) are (1995) also have this nonlinearity, but less accurate dispersion
determined by matching the linear dispersion and shoaling (Padé [2,2]). Due to this similarity in nonlinearity, the above four
properties of the equations to reference solutions. Eqs. (5.5) and models (BouN2D4, MS98-2, BouN2D2, and the model of Wei
(5.6a), (5.6b,c) comprise the set of enhanced, higher-order et al.) can be referred to as Serre-type equations. Improvements in
Boussinesq equations central to the present study. Hereafter, we the dispersion will have a positive effect on the nonlinear
will refer to the above model as BouN4D4 (N4 denotes fully characteristics of these Serre-type equations (see Madsen and
nonlinear up to O(μ4), and D4 denotes Padé [4,4] dispersion). Schäffer, 1998; Zou, 2000). Dingemans (1994) investigated
BouN4D4 can reduce to the enhanced equations in terms of the applicability of the Serre-type equations (the Serre equation
the depth-averaged velocity presented by MS98 by using the with extended linear dispersion). Similarly, Section 8 will
assumption μ = O(ε) to omit terms of higher order than εμ4. also compare models BouN2D2, BouN2D4, BouN4D4, and
This model will be referred to as MS98-1. The main difference MS98-1 in numerical simulations. The effects of O(μ4) nonlinearity
between BouN4D4 and MS98-1 is the inclusion of these will be seen by comparing BouN2D4, BouN4D4, and MS98-1,
nonlinear terms. As we will see in Section 6 and subsection 8.3, while the effects of dispersion accuracy will be seen by comparing
the O( μ4) terms make a significant contribution to the third- BouN2D2 and BouN2D4. The amplitude dispersions embedded in
order nonlinearity, especially in the amplitude dispersion. To aid these models will also be examined. Before doing this, we first
in this investigation, we make use of a reduced version of make the Fourier analysis of the equations in the next section.
BouN4D4 where P(4) is given by the linearized expression (4.1)
and M(4) is omitted. As for Eq. (4.1), this is done by introducing 6. Fourier analysis of the equations
the assumption ε = O(μ2/3) and omitting O(εμ4) terms. Then,
the resulting equations are of the form In this section we use Fourier analysis to derive the third-
order transfer function and amplitude dispersion of all four
P
gt þ j  ðd uÞ ¼ 0 ð5:8Þ models, assuming a horizontal bottom. This result will allow us
to examine their higher-order nonlinear characteristics on a
P P P   theoretical basis. The following Stokes-type expansions are
ut þ eð u  jÞ u þ jg ¼ A2 PWð2Þ þ A4 PWð4Þ þ O A6 ð5:9aÞ
introduced for this purpose.
where
  g ¼ A1 cos h þ eðA20 þ A22 Þcos 2h þ e2 ðA31 cos h þ A33 cos3 hÞ
ð2Þ ð2Þ
PW ¼P þ L1 μ2 G  C ; ð5:9b; cÞ ð6:1Þ
ð4Þ ð4Þ
PW ¼ P̃  L2 C̃
P
u ¼ U1 cos h þ eU22 cos 2h þ e2 ðU31 cos h þ U33 cos 3hÞ ð6:2Þ
1 P 1 P
G ¼ hj½j  ðh ut Þ  h2 jðj  ut Þ; ð5:9d; eÞ
2 6
P x ¼ x0 þ e2 x3 : ð6:3Þ
C̃ ¼ ut þ jg
In these equations we use the phase θ = kx − ωt. The case of
ð4Þ
in which P̃ is given by Eq. (4.1), L1 and L2 are given by A20 = 0 is considered here. To obtain governing equations for the
Eqs. (5.7a), (5.7b). We will refer to this model as BouN2D4 (N2 first-, second-, and third-order problems, we substitute these
denotes fully nonlinear up to O(μ2), and D4 denotes Padé [4,4] expansions into the models (BouN4D4, BouN2D4, or
dispersion). For μ bb 1, the assumption ε = O( μ2/3) means that BouN2D2) and collect terms of like order up to ε2. Solving
water waves are more strongly nonlinear than in the classical the first-order problem yields the dispersion relation of the
Boussinesq equations, which were based on the assumption model, which is a Padé [4,4] expansion for BouN4D4 and
ε = O(μ2), and they are also stronger than those considered by BouN2D4 (α1 = 1/9,β1 = 1/945) and a Padé [2,2] expansion for
MS98-1, which is based on the assumption ε = O(μ). BouN2D2. Solving the second-order problem gives the second-
In terms of formal order, BouN2D4 is similar to the enhanced order transfer functions A22 and U22. Up to second order the
Boussinesq equations in terms of the velocity at an arbitrary z level Fourier analysis results for BouN4D4 and BouN2D4 will be the
given by MS98 (Eqs. (6.3a) and (6.3b) in their paper). Hereafter we same as those of MS98-1, so we will not repeat these results
refer to this model as MS98-2. This model is obtained by enhancing here. The third-order solutions (A31,A33,U31,U33 and ω3), how-
the (μ2,ε3μ2) Boussinesq equations of Wei et al. (1995) to the Padé Stokes
ever, will be different. Figs. 1 and 2 present the ratios A33/A33
[4,4] dispersion through applying the enhancement technique of and ω3/ω3 Stokes
for models BouN4D4, BouN2D4 and BouN2D2,
Schäffer and Madsen (1995) Although the two models have the where A33 Stokes
and ω3Stokes are the third-order surface elevation
same order of dispersion and nonlinearity, the nonlinear character- transfer function and amplitude dispersion of the Stokes equations
istics of the two models will be different (see details in Section 6) respectively (e.g. Skjelbreia and Hendrickson, 1961). The results
since the different velocity variables are used for them. The order of for model MS98-1 are given in these figures as well. Fig. 2 also
nonlinearity for these models is the same as that of the Serre's displays the result of model FN4 by Gobbi et al. (2000) for
equations (Serre, 1953), so the MS98-2 and BouN2D4 models can comparison. It can be seen that BouN4D4 gives better results than
be seen as extensions of Serre's equations with improved MS98-1, because the former retains all O(μ4) nonlinear terms
dispersion. Model BouN2D2 and the equations of Wei et al. while the latter retains only O(εμ4) nonlinear terms (i.e., the
Z.L. Zou, K.Z. Fang / Coastal Engineering 55 (2008) 506–521 513

Fig. 1. Ratio of third harmonic A33/AStokes


33 . Fig. 3. Ratio of the second and third harmonics of BouN2D4 and MS98-2.

leading O(μ4) nonlinear terms). Similarly, MS98-1 gives better MS98). In MS98, ω3/ω3Stokes was actually predicted more ac-
results than BouN2D4, which neglects all O(μ4) nonlinear terms. curately by the original equations than by the enhanced equations
These comparisons show that the O( μ4) nonlinear terms (see Fig. 14 in MS98). In addition to the above comparisons,
make a significant contribution to the third-order nonlinear Fig. 3 presents the comparison of BouN2D4 and MS98-2 for the
Stokes Stokes
characteristics of the model, especially the amplitude dispersion results of A22/A22 and A33/A33 . Since the two models have
ω3. The inclusion of all O( μ4 ) nonlinear terms (model the same order of accuracy but different velocity variables (the
BouN4D4) gives a much better result than only retaining the former model is in terms of depth-averaged velocity and the latter
O(εμ4) nonlinear terms (model MS98-1). The latter model in terms of velocity at an arbitrary vertical level), the comparison
only produces a small improvement compared to neglecting all can show us the effect of different velocity variables. It is seen
O( μ 4 ) nonlinear terms (model BouN2D4). As will be seen in from the figure that MS98-2 gives a better result but the difference
subsection 8.3, the amplitude dispersion embedded in MS98-1 is small.
and BouN2D4 does not reproduce known dispersion effects in The parameters α2 and β2 are determined by optimizing the
the nonlinear evolution of wave groups adequately. In Fig. 2, shoaling property of the equations, as done by MS98. Their
the results of models FN4 and BouN4D4 are of similar accuracy results for these values can also be adopted directly in
within the kh range used by Gobbi et al. (2000), probably because Eqs. (5.1)–(5.4), except that an additional factor of 1/4 should
they have the same orders of nonlinearity and dispersion. It is be applied to β2 due to differences in expressions (5.3) and (5.4)
worthwhile to notice that BouN2D2 actually provides better for the higher-order equations. (This difference does not affect
results than BouN2D4 in two cases: for ω3/ω3Stokes when kh N 1.0, the analysis of the model's shoaling properties.) Thus, we have
and for A33/A33Stokes
when kh N 3.2. This shows that an improve- α2 = 0.146488 and β2 = 0.00798359/4.
ment in the dispersion relation of BouN2D4 does not lead to a
corresponding improvement in the amplitude dispersion. Similar 7. Numerical implementation
results were obtained by MS98 in their comparison of the Wei
et al. (1995) model and its enhancement (Eqs. (6.3) and (6.3b) of In this section we describe a numerical scheme for implementing
BouN4D4 in the one-dimensional case. Similar methods were used
for models BouN2D4 and BouN2D2. After some manipulation, we
can obtain the one-dimensional governing equations of BouN4D4
with a source function and sponge layer:
P
gt þ ðd u Þx ¼ fs ð x; t Þ  g fd ð xÞ ð7:1Þ

P P
Ut ¼ e u  ux  gx  eA4 M ð4Þ þ A2 P Vð2Þ þ A4 P Vð4Þ
 
þ S þ O A6 ; ð7:2Þ

where

P P 1 2PP a1 1 4P
U ¼ u þ L 1 u þ L2 u  h u þ  h u xxxx
3 xx 3 45

1 2P a1 1 4P
S ¼  h uxxt þ  h uxxxxt : ð7:3Þ
Fig. 2. Ratio of amplitude dispersion ω3/ωStokes
3 . 3 3 45
514 Z.L. Zou, K.Z. Fang / Coastal Engineering 55 (2008) 506–521

Fig. 4. Setups of Luth et al. (1994) (up) and Ohyama et al. (1994) (down) experiments. All dimensions in (m).

In Eq. (7.1), fs(x,t) is the source function and − η fd(x) (2001), who studied the nonlinear evolution of wave groups in a
represents wave dissipation by the sponge layer. M(4),P′(2),P′(2), flume. The first two experiments were also used in GK99 for
L1,L2 are one-dimensional versions of the quantities defined in validation of model FN4 (fully nonlinear up to O(μ4)); this test
Eqs. (3.21b) and (5.5) through (5.7a), (5.7b). Similarly, ū is a case was chosen to ease comparison between their model and
one-dimensional version of the depth-averaged velocity. Since our own. The present model BouN4D4 is the same in dispersion
we have moved ut to the left-hand side of the equation, the and nonlinearity as FN4 (both are accurate to a Padé [4,4]
variable C given by Eq. (4.3b) is redefined as expansion in dispersion, and fully nonlinear up to O( μ4)), but
possesses a different formulation, so the comparison between
PP
C ¼ e u ux þ gx : ð7:4Þ them can show the effect of different forms of the Boussinesq
models on the numerical results.
One critical problem in the numerical solution of a higher- The experiment of Shemer et al. (2001) provides a good
order Boussinesq model is the dispersion error introduced by basis for checking for the third-order nonlinearity of the
discretization. This error can sometime overwhelm the actual Boussinesq models. According to Stokes wave theory, the third-
dispersive terms in the equations. The problem can be avoided order nonlinearity makes the amplitude contribute to the
by choosing a numerical scheme in which the dispersion errors dispersion relation, and thus waves in envelope with higher
are of higher order than the dispersion relation embedded in the amplitude travel faster than those with lower amplitude. As a
equations. To this end, we adopt a numerical scheme similar to result, wave envelops become skew, i.e., the front part of a wave
that of Gobbi and Kirby (1999) (hereafter GK99), except as group is pressed while the back part is stretched. New free-wave
noted below. components will also be generated during the process due to
All spatial derivatives in our simulation are approximated by nonlinear resonant interaction and the wave spectrum is
a seven-point difference formula, which simplifies the coding expected to become wider during wave group evolution. The
process. This was not the case in GK99, where the choice of simulations for all these nonlinear properties call for high
difference formula is dependent on the order of the term. accuracy in nonlinearity and dispersion of a mathematical wave
A fourth-order filter was used every Nf time steps, where model.
40 b Nf b 300 for the cases studied here. This is necessary to The three models of the present study (BouN4D4, BouN2D4,
remove the high-frequency noise caused by sharp changes of and BouN2D2) are used to simulate the above cases in order to
bathymetry, nonlinear wave motions, and discretization errors. investigate the effects of nonlinearity and dispersion on the
When the number of grid points per wavelength is greater than numerical results. The effects of nonlinearity in the equations are
40, the filter will remove Nyquist frequency oscillations without observed by comparing model BouN4D4 with model BouN2D4,
significantly affecting the solution.

8. Model validation Table 1


Incident wave characteristics for Luth and Ohyama experiments
In this section we compare our numerical results to two Luth experiments Ohyama experiments
datasets. The first is taken from the experiments of Luth et al.
Case (a) Case (c) Case (2) Case (4) Case (6)
(1994) and Ohyama et al. (1994), who investigated the propa-
gation of non-breaking waves over a submerged bar. The Wave amplitude (m) 0.01 0.0205 0.025 0.025 0.025
Wave period (s) 2.02 1.01 1.341 2.012 2.683
second dataset comes from the experiment of Shemer et al.
Z.L. Zou, K.Z. Fang / Coastal Engineering 55 (2008) 506–521 515

which neglects O(μ4) nonlinear terms. The effects of dispersion 8.1. Comparison to the Luth et al. (1994) experiment
are observed by comparing model BouN2D4 with model
BouN2D2, which are both fully nonlinear up to O(μ4) but have The experimental setup and incident wave characteristics of
different dispersion relations. Luth's data are given in Fig. 4 and Table 1. The difference between

Fig. 5. Wave surface elevations of Luth's experiment (circle) and the numerical results of BouN4D4 (solid), BouN2D4 (dash), BouN2D2 (dash–dot) and FN4 (gray).
516 Z.L. Zou, K.Z. Fang / Coastal Engineering 55 (2008) 506–521

our numerical results and the experimental data is very small for time to form. In order to model this situation correctly, the
gauges before the bar and over the bar's upstream slope. Thus, simulation needs to be terminated at an appropriate moment.
only those gauges over the top of the bar and downstream of the The time step of the simulation is determined by other factors,
bar are published here. however, so it is not possible to exactly match the duration of
Fig. 5 plots the surface elevations of models BouN4D4, wave propagation. The remaining discrepancy leads to a phase
BouN2D4, BouN2D2, and FN4 in addition to the data of Luth difference between the numerical simulation and the experi-
et al. (1994). It can be seen that for case (a) BouN4D4 and FN4 mental measurements (see GK99 for a more detailed discussion
provide similar results, and that both perform quite well for of this problem). The comparison of the results of BouN4D4
most of the gauges. There are slight phase differences between and FN4 shows that FN4 presents better results than BouN4D4,
the two model's results at the last two gauges, far downstream especially for wave phase, this may be caused by the difference
of the bar. in the forms of governing equations.
In the case with a shorter period (c), some differences appear BouN2D4 and BouN4D4 give almost the same results in case
between BouN4D4 and the experimental data. This can be (a), even though the former lacks O(μ4) nonlinear terms. It is an
explained as follows. Since only the first 46 s are available from expected result since O(μ4) nonlinear effects are not very
the experimental record, a stable periodic wave may not have important for longer period waves. For the shorter period waves

Fig. 6. The spatial variation of the Fourier components of the wave elevations of Luth's experiment (circle) and the numerical results of BouN4D4 (solid), BouN2D4
(dash), BouN2D2 (gray dash–dot) and FN4 (gray dash).
Z.L. Zou, K.Z. Fang / Coastal Engineering 55 (2008) 506–521 517

of case (c), however, there are some phase differences between the 8.2. Comparison to the Ohyama et al. (1994) experiment
two models. This error reflects the fact that O(μ4) (and higher)
nonlinear terms may affect the phase velocity. As expected, The experimental setup and incident wave characteristics of
BouN2D2 gives the worst performance due to the low accuracy of Ohyama's data are given in Fig. 4 and Table 1. Fig. 7 compares
its dispersion relation. the numerical results of BouN4D4, BouN2D4, BouN2D2, and
Fig. 6 gives the Fourier amplitudes of the surface elevations FN4 with Ohyama's data for cases (2), (4), and (6). The rapid
by Fourier analysis. We can see that BouN4D4 and FN4 give change of water depth over the bar can sometimes lead to
similar results, although the latter predicts smaller peaks at the numerical instability, which will require the use of more
third and fourth harmonics than those of the measurement. In numerical filtering. Models BouN4D4 and FN4 give the results
case (c) FN4 behaves better than BouN4D4 but overestimates which are much closer to the measurements at gauge 3 in all
the first harmonic downstream of the bar, and the other three cases, and have quite similar performance at gauge 5 expect for
models do not. In case (a) BouN4D4, BouN2D4, BouN2D2 the discrepancy in phase speed. Compared with experimental
give similar results for the first and second harmonics, but there data, it seems that model BouN4D4 predicts faster waves while
are some differences in the higher-order harmonics. The same FN4 predicts slower waves. The phase discrepancy between
trend can be seen in case (c), except that differences begin to BouN4D4 and FN4 decreases gradually from case (2) to case
occur even for the second harmonic. (6), following a gradual increase in the incident wavelength.

Fig. 7. Wave surface elevations of Ohyama's experiment (circle) and the numerical results of BouN4D4 (solid), BouN2D4 (dash), BouN2D2 (gray dash–dot) and FN4
(gray dash).
518 Z.L. Zou, K.Z. Fang / Coastal Engineering 55 (2008) 506–521

The reason for the observed differences between BouN4D4,


FN4, and the experimental measurement is not very clear.
BouN2D2 exhibits even larger errors in the phase velocity,
which however decrease with increasing wave period from case
(2) to case (6). Model BouN2D4 gives much better results.
Since these two models have the same degree of nonlinearity,
the difference must be due to the lower accuracy of dispersion in
model BouN2D2.

8.3. Comparison to the Shemer et al. (2001) experiment

In the experiment conducted by Shemer et al. (2001) on the


nonlinear evolution of wave groups, a wave maker is driven by
one of the following two signals:

sðt Þ ¼ s0 cosðXt Þcosðϖ0 t Þ ðX ¼ ϖ0 =20Þ or ð8:1Þ

sðt Þ ¼ s0 jcosðXt Þjcosðϖ0 t Þ ðX ¼ ϖ0 =20Þ; ð8:2Þ

where s0 is the forcing amplitude, ϖ0 is the carrier frequency,


and Ω is the modulation frequency. These two signals produce
Fig. 9. Experimental data (circle) of wave surface elevations of wave groups for
identical wave group envelopes, but with different spectra. The
driving signals (8.2) and the numerical results of BouN4D4 (solid), BouN2D4
spectrum of the first signal is bimodal, having two distinct peaks (dash) and MS98-1 (dash–dot).
of identical amplitude at the frequencies ϖ0 ± ϖ0/20. The
spectrum of the second signal has its maximum at the carrier
frequency ϖ0, plus several smaller peaks at discrete frequencies In the corresponding simulations, we set the spatial and
spaced by ϖ0/10 with three dominant modes. The wave maker temporal steps to 0.02 m and 0.01 s and collected data until a
is positioned at x = 0.0 m in a wave flume with water depth steady state is reached. The models BouN4D4, BouN2D4, and
0.6 m. The periods of the carrier waves are 0.9 s for both signals, MS98-1 are chosen for these simulations. The latter two are
and the corresponding value of kh is 3.0. Thus, both wave reduced versions of BouN4D4, with the same dispersion but
motions are within the deep water regime. different degrees of nonlinearity. The purpose of this compar-
ison is thus to demonstrate the effect of nonlinearity degree on
the nonlinear evolution of wave groups. Figs. 8 and 9 compare
the wave surface elevations of the experimental data and models
BouN4D4, BouN2D4, and MS98-1, for driving signals (8.1)
and (8.2) respectively. It can be seen that close to the wave
maker (x = 0.24 m), the three numerical results are almost the
same and agree with the data. As the wave propagates along the
flume, the results of the three models diverge: the envelopes
given by the BouN2D4 are basically symmetric, the envelopes
given by the MS98-1 show a skewness feature but have bigger
difference with the data while the envelopes given by the
BouN4D4 show the skewness feature which are similar to the
data. So only BouN4D4 can predict the wave elevations well.
Since the reason for the skewness of wave envelope is the
amplitude dispersion, these differences can be attributed to the
different amplitude dispersions ω3 of the three models, as
shown in Fig. 2. In the figure, BouN4D4 has the most accurate
amplitude dispersion at kh = 3.0 (corresponding to this experi-
mental case), so it succeeds in reproducing the skewness
of the wave envelope, but BouN2D4 and MS98-1 have lower
accuracy of amplitude dispersion for the same range of kh, so
they cannot reproduce the amplitude dispersion effect present in
the measurement adequately.
Fig. 8. Experimental data (circle) of wave surface elevations of wave groups for Figs. 10 and 11 show Fourier amplitude spectra of the wave
driving signals (8.1) and the numerical results of BouN4D4 (solid), BouN2D4 group's surface elevation at x = 0.24 m and 9.47 m for driving
(dash) and MS98-1 (dash–dot). signals (8.1) and (8.2) respectively. In the experimental spectra,
Z.L. Zou, K.Z. Fang / Coastal Engineering 55 (2008) 506–521 519

Fig. 10. Amplitude spectra at x = 0.24 m, 9.47 m for wave groups in Fig. 8.

amplitudes have increased on the high-frequency side when the propagate along the tank. Shemer et al. (2001) used Zakharov
wave reaches x = 9.47 m and the spectrum has become equations to simulate this process in which 10 free-wave modes
asymmetric. BouN4D4 produces amplitude spectra similar to were chosen to calculate the evolution of wave spectrum along
those of the experiment, and BouN2D4 produces spectra which the tank. Reference can be made to this paper for detailed
change very little by the time they reach the end of the flume, discussions. The cubic Schrödinger equation was also used by
which shows that the wave components given by this model them for this simulation, but it is shown that the appearance of
remain unchanged along the flume, while MS98-1 presents asymmetric wave groups and the generation of free-wave
asymmetric feature and behaves better than BouN2D4, but modes could not be adequately described by this equation due to
worse than BouN4D4. The asymmetric growth of the spectrum its symmetric properties and the lack of nonlinear resonant
shown in the figures presents another nonlinear characteristic of interactions of different wave components in the wave
nonlinear wave group evolution apart from the amplitude spectrum.
dispersion, i.e. the generation of new free-wave components due Besides the free-wave modes in the wave spectrum, there are
to resonant wave–wave interaction. This is a third-order (O(ε3)) also bound wave components, such as the waves with the sum
nonlinear effect which causes the energy transfer from one and difference frequencies, but these components do not grow
frequency to another in a wave spectrum and is referred to as the significantly along the tank as seen from the corresponding
four-wave resonant interaction. This phenomenon was first energy at these frequencies in the spectra of Figs. 10 and 11. So
investigated by Phillips (1960), Longuet-Higgins (1962), the asymmetric form of the wave spectra at location x = 9.47 m is
Benny (1962) and Hasselmann (1962), they showed the mainly owing to the growing of new free mode waves along the
significance of wave resonant interaction for the growth of a tank. In Figs. 10 and 11, this asymmetric feature has been
wave spectrum in deep water. For the present cases, although produced pronouncedly by BouN4D4, but lost by BouN2D4.
there are only two free-wave modes for the signal (8.1) or three MS98-1 gives the results between the two. This shows the
free-wave modes for the signal (8.2) near the wave maker contribution of fully nonlinear terms at O(μ4), which are omitted
(x = 0.24 in Figs. 10 and 11), the four-wave resonant interaction by BouN2D4 and partly by MS98-1. These two models are
will inspire other new free-wave modes to grow as waves based on the assumptions ε = O(μ2/3) and ε = O(μ) respectively.

Fig. 11. Amplitude spectra at x = 0.24m, 9.47m for wave groups in Fig. 9.
520 Z.L. Zou, K.Z. Fang / Coastal Engineering 55 (2008) 506–521

The above numerical results show that these nonlinearity by Shemer et al. (2001). In this case BouN4D4 gives much
accuracies are not appropriate for the description of amplitude better results than BouN2D4 and MS98-1, showing a marked
dispersion and the four-wave resonant interaction, and the amplitude dispersion effect and the free-wave generation. The
inclusion of all O(μ4) nonlinear terms is necessary to simulate simulation produces a skewed wave envelope and generates
this kind of nonlinear wave motions. high-frequency free-wave components, agreeing with the
experimental data. The other two models, however, produce
9. Conclusions basically symmetric wave envelopes and amplitude spectra.
Since the three models use the same dispersion approximation
This paper has developed an alternative form of the two- (the Padé [4,4] expansion), this result can be entirely attributed
dimensional, fourth-order Boussinesq equations (model to the more accurate nonlinearity of BouN4D4.
BouN4D4), creating a model which is fully nonlinear up to O(μ4)
and approximates the dispersion relation by a Padé [4,4] expansion. Acknowledgments
These equations are derived from the depth-averaged momentum
and continuity equations, and expressed in terms of the depth- The authors would like to thank the financial support by the
averaged horizontal velocity. A σ-transformation is used in the National Natural Science Foundation of China, under Grant
derivation to deal with arbitrary free surface and water bottom Nos. 50479053 and 59979002 and the support by Program for
variations, and no limitation is imposed on the bottom slope. In Changjiang Scholars and Innovative Research Team in
order to improve the linear dispersion characteristics of the University and the Special Funding for Ph.D Education in
equations, we employed the enhancement technique introduced University (Grant No. 20040141030).
by MS98. This improves the linear dispersion to a Padé [4,4]
expansion in kh of the squared celerity, according to Stokes linear
References
theory. The equations reduce to the G–N equations, on the other
hand, if O(μ4) terms are omitted and this enhancement is not
Agnon, Y., Madsen, P.A., Schäffer, H.A., 1999. A new approach to higher order
applied.
Boussinesq model. J. Fluid Mech. 399, 319–333.
Two reduced sets of the new equations are introduced by Benny, D.J., 1962. Nonlinear gravity wave interaction. J. Fluid Mech. 14,
using the assumption ε = O( μ2/3) to linearize some of the O(μ4) 577–584.
terms. Both sets retain all nonlinear terms up to O(μ2). The first Dingemans, M.W., 1994. Comparison of computations with Boussinesq-like
reduced set (BouN2D2) has a Padé [2,2] dispersion approxima- models and laboratory measurements. MAST-G8M note, H1684. Delft
tion, and the second (BouN2D4) has Padé [4,4] dispersion. Hydraulics. 32 pp.
Fuhrman, D.R., Madsen, P.A., Bingham, H.B., 2004. A numerical study of
These can be seen as Serre-type equations with extended dis- crescent waves. J. Fluid Mech. 513, 309–341.
persion relations. Gobbi, M.F., Kirby, J.T., 1999. Wave evolution over submerged sills: tests of a
Fourier analysis of waves propagating for a horizontal high-order Boussinesq model. Coast. Eng. 37, 57–96.
bottom shows that including all O(μ4) nonlinear terms (model Gobbi, M.F., Kirby, J.T., Wei, G., 2000. A fully nonlinear Boussinesq model for
BouN4D4) greatly improves the amplitude dispersion ω3 of the surface waves. Part 2. Extension to O(kh)4. J. Fluid Mech., 405, 181–210.
Green, A.E., Naghdi, P.M., 1976. A derivation of equations for wave
model. Retaining only the O(εμ4) nonlinear terms (as in model propagation in water of variable depth. J. Fluid Mech. 78, 237–246.
MS98-1) produces only a small improvement compared to the Hasselmann, K., 1962. On the nonlinear energy transfer in a gravity wave
second-order nonlinear model BouN2D4. Although BouN2D2 spectrum. Part 1. J. Fluid Mech. 12, 481–500.
has less accurate dispersion characteristics, it can actually pro- Longuet-Higgins, M.S., 1962. Resonant interaction between two trains of
gravity waves. J. Fluid Mech. 12, 333–336.
duce a better amplitude dispersion ω3than model BouN2D4 in
Luth, H.R., Klopman, G., Kitou, N., 1994. Kinematics of waves breaking
certain regimes. This confirms the findings of MS98, in which partially on an offshore bar; LDV measurements of waves with and without
the equations of Wei et al. (1995) with Padé [2,2] dispersion a net onshore current. Report H-1573. Delft Hydraulics. 40 pp.
give a better result for ω3 than the corresponding enhanced Lynett, P., Liu, P.L.F., 2004. A two-layer approach to wave modeling. Proc.
equations with Padé [4,4] dispersion. R. Soc. Lond., A 460, 2637–2669.
The new models were validated against experiments of non- Madsen, P.A., Sørensen, O.R., 1992. A new form of the Boussinesq equations
with improved linear dispersion characteristics. Part 2. A slowly-varying
breaking wave propagation over a submerged bar (Luth et al., 1994; bathymetry. Coast. Eng. 18, 183–204.
Ohyama et al., 1994). The new models are also compared to FN4 Madsen, P.A., Sørensen, O.R., Schäffer, H.A., 1997. Surf zone dynamics
(Gobbi et al., 2000). It is shown that BouN4D4 and FN4 perform simulated by a Boussinesq type model. Part 2: surf beat and swash
best among the four models. The differences between their results oscillations for wave groups and irregular waves. Coast. Eng. 32, 289–320.
and that of BouN2D4 is not very significant, however, which shows Madsen, P.A., Schäffer, H.A., 1998. Higher-order Boussinesq-type equations for
surface gravity waves: derivation and analysis. Phil. Trans. R. Soc. Lond. A 356,
that O(μ4) nonlinear terms do not have a significant effect on wave 3123–3184.
propagation in the tested cases. As expected, BouN2D2 gives the Madsen, P.A., Bingham, H.B., Liu, H., 2002. A new Boussinesq method for
worst results. Furthermore, there are still some unexplained dif- fully nonlinear waves from shallow to deep water. J. Fluid Mech. 462,
ferences between the numerical results of BouN4D4 and FN4 and 1–30.
Madsen, P.A., Bingham, H.B., Schäffer, H.A., 2003. Boussinesq-type formula-
the experimental data of Ohyama et al. (1994).
tions for fully nonlinear and extremely dispersive water waves: derivation
In order to investigate the effects of various nonlinear and analysis. Proc. R. Soc. Lond., A 459, 1075–1104.
approximation levels, we also simulated the nonlinear evolution Nwogu, O., 1993. Alternative form of Boussinesq equations for nearshore wave
of wave groups and compared the results to a flume experiment propagation. J. Waterw. Port Coast. Ocean Eng., ASCE 119 (6), 618–638.
Z.L. Zou, K.Z. Fang / Coastal Engineering 55 (2008) 506–521 521

Ohyama, T., Kiota, W., Tada, A., 1994. Applicability of numerical models to Skjelbreia, Hendrickson, 1961. Fifth order gravity wave theory. Proc. 7th Conf.
nonlinear dispersive waves. Coast. Eng. 24, 213–297. Coast. Eng., pp. 184–196.
Phillips, O.M., 1960. On the dynamics of unsteady gravity waves on finite Wei, G., Kirby, J.T., Grilli, S.T., Subramanya, R., 1995. A fully nonlinear
amplitude. Part 1. The elementary interaction. J. Fluid Mech. 9, 193–217. Boussinesq model for surface waves. Part 1. Highly nonlinear unsteady
Serre, P.F., 1953. Contribution to the Study of Permanent and Non-permanent waves. J. Fluid Mech. 294, 71–92.
Flows in Channel. La Houille Blanche, vol. 8, pp. 374–388. Zou, Z.L., 1999. Higher order Boussinesq equations. Ocean Eng. 26, 767–792.
Schäffer, H.A., Madsen, P.A., 1995. Further enhancements of Boussinesq-type Zou, Z.L., 2000. A new form of high-order Boussinesq equations. Ocean Eng. 27,
equations. Coast. Eng. 26, 1–14. 557–575.
Shemer, L., Jiao, H.Y., Kit, E., Agnon, Y., 2001. Evolution of a nonlinear wave
field along a tank: experiments and numerical simulations based on the
spatial Zakharov equation. J. Fluid Mech. 427, 107–129.

You might also like