Nonlinear Coupled 2D Boussinesq Type

Equations for Shallow Water Waves using Finite

Volume Finite Difference Methods
Vinita1 and Prashant Kumar2*††
1,2* Department of Applied Sciences, National Institute of Technology,
Delhi, India.

*Corresponding author(s). E-mail(s): [email protected];

Contributing authors: [email protected] ;
† These authors contributed equally to this work.

Abstract
In practice, nonlinear coupled Boussinesq type equations for shallow water waves
are used to solve the wave resonance at a port. This work focuses on the utilization
of nonlinear processes over varying bathometry in Nwogu’s 2D depth integrated
Nonlinear Coupled Boussinesq type equations (NCBTEs) for shallow water wave
propagation in the coastal zone. The combined finite volume and finite difference
method is used to solve the 2D depth integrated NCBTEs of Nwogu. The conser-
vative flux form of the 2D Nwogu’s NCBTEs is solved by the finite volume method
(FVM), while the remaining dispersive terms are solved by the finite difference
method (FDM). The system uses an approximation Riemann solver using Roe’s
technique for advection and the well-balanced topography source term upwinding
for advection fluxes. Analytical rectangular solutions with both linear and non-
linear harbor oscillations are used to validate the mode. The current numerical
scheme’s accuracy is verified and confirmed through preceding numerical study.
The practical ramifications of ocean engineering with the wave motion problem
on the feasible Paradip port, Odisha, India, can be addressed with this model
technique.

Keywords: Nonlinear coupled Nwogu’s Boussinesq type equation, coupled hybrid finite
volume finite difference method, solitary waves, regular waves, water wave propagation

1
1 Introduction
Introduction The mathematical representation of When shallow water is present in the
coastal zone due to certain factors, such as earthquakes, storms, water mass movement,
and high rainfall, water waves propagate as a result. Accurate simulations of accu-
rately depicted nonlinear and dispersive water waves are crucial to coastal and ocean
engineering, which also accurately represent shore wave phenomena such as diffraction,
refraction, harmonic interaction, and shoaling and runup. First, [1] describes water
wave phenomena. These equations are called standard Boussinesq equations because
they are expressed in terms of the depth averaged velocity. They are only applicable to
relatively shallow water because they become unstable at depths more than one-fifth
of the same wavelength in deep water . Consequently, numerous substitute formula-
tions possessing enhanced dispersive qualities have been presented. Recent years have
seen a lot of interest in the extended Boussinesq formulation of [2] for simulating trans-
mission of waves from deep to shallow water . In order to minimize wave propagation
errors generated from linear theory, Nwogu’s formulas are written in terms of the hor-
izontal velocity and hauteur of the surface at a specific depth.
Other popular sets of equations with similar dispersion features have been obtained
by [11] and [12] assuming modest nonlinearity and dispersion . A lot of work has been
done recently to improve the nonlinear and dispersive properties of Boussinesq-type
models by adding high order nonlinear and dispersive terms. These extended models,
also known as slow-order augmented Boussinesq-type equations, offer a more realis-
tic representation of the phase and the group velocities in intermediate water with a
water depth to wavelength ratio up to 1/2. Examples of these include Bingham et al.
[13], Gobi et al. [14], Lynett and Liu [16], and Madsen et al. [17, 18, 19]. The numeri-
cal integration of these terms requires a considerable increase in computational effort
since they are more challenging to integrate. From a numerical perspective, to solve
Boussinesq-type equations, the finite difference (FD) method was the most popular
approach until recently. For examples, see Beji and Nadaoka [12], Furman and Bing-
ham [20], Furman and Madsen [21], Madsen and Sorenson [11], Madsen et al. [17],
Nwgou [2], and Wei and Kirby [24]. . The FD approach is quite popular because it
makes approximating higher-order derivatives easy and produces well-structured lin-
ear systems that are easy to solve (such tri-diagonal ones).However, over the past
ten years, the use of finite element (FE) methods to solve extended Boussinesq-type
models has increased; for examples, see [25], [26], [27], [28], [29], [30], [31], [32], and
[33]. The primary drawback of the Finite Element approach is that, typically, even
in low-order extended Boussinesq equations, higher-order spatial derivatives must be
minimized. While within the FE framework relatively efficient algorithms with higher-
order precision can be developed, major complications and stability problems can also
occur, particularly on arbitrary grids.
The FV approach is well acknowledged for its advantages in numerically simulat-
ing the NSWE with respect to topography and wet/dry front treatment, for further
studies, see [34], [36], and their references. For example, [37], [38], [40], [34], Delis et
al. [35], [41], Hubbard and Dodd [42], Li and Raichlen [44], Marche et al. [45], Nikolos
and Delis [46], Titov and Synolakis [47], and )Toro [48] are just a few examples of the
many applications to which The NSWE has been resolved by the use of FV schemes.

2
Specifically, the advantage of Godunov-type FV schemes based on Riemann solvers
is that they are completely conservative schemes with inherent shock-capturing capa-
bilities, can precisely account for the bed topography, and can solve the integral form
of the nonlinear equations. Hubbard and Dodd [42], LeVeque [49], Marche et al. [45],
Toro [48], Toro and Garcia [36], Brocchini et al. [38], Delis et al. [34]. Significant
progress has been made in computing the topography source term within the FV
framework. As a result, a number of mathematical and numerical techniques have
been put forth to balance the source term and the flux gradient in order to compute
stationary or nearly stationary solutions. Referencing [34] and its references, we
may say that research on this property—known as well balancing—is currently rife.
Another significant issue is the emergence of dry regions, which might be caused by
the water’s velocity or by the starting circumstances. Therefore, managing the wet-
ting and drying of shifting borders (such as shoreline motion) [50] is a problem that
has garnered a lot of interest. Using the NSWE equations [47], a number of strategies
have been put forth in various models and numerical schemes. For Boussinesq type
equations, [21],[22] have employed an extrapolation technique to enable the models to
handle moving boundaries. Other methods have been devised to avoid dealing with
wet/dry interfaces, such as modeling the shore as porous or slotted )[43], artificially
wetting dry cells [42], and eliminating the dry cells from the computational domain
[38, 39]. Classical FV schemes (of the Godunov type) have been updated only recently,
and in one spatial dimension, to solve augmented Boussinesq type equations.

With these modifications, the FV approach is used to solve the nonlinear shallow
water portion of the equations, while FD schemes are used to discretize the dispersive
terms, creating hybrid FV/FD schemes (see, for instance, ([51],[52, 53], [54], [55], [56],
[57], [22], Ma et al. [58], Roeber and Cheung [59], Shi et al. [60], Shiach and Ming-
ham [61], Soares-Frazao and Guinot [62]). The hybrid schemes have been established
for Boussinesq-type equations to integrate the flexibility and shock-capturing capabil-
ities of the FV approach into dispersive wave models. These schemes combine the FV
and FD methodology. This technique is particularly useful for short and long wave
interactions as the solution may be readily transformed into entirely FV solution of
the NSWE by deleting the higher order Boussinesq elements, if desired. Furthermore,
two space dimensions have been added to this hybrid technique, although for uniform
structured grids Kim and Lee [63],Tonelli and Petti [64]. Similar to the FD approach,
hybrid FV/FD schemes can significantly limit modeling when dealing with 2D irreg-
ular geometries, despite being relatively easy to build for structured grids. A method
to lessen the impact of a structured grid was introduced in [65], where a Godunov-
type second order FV scheme was utilized to solve the 2D by [11] equations using the
cut-cell approach.
This study computes the wave-induced oscillation by the nonlinear periodic long waves
in the irregular bounded domain with changing bathometry using a connected hybrid
model based on FVM and FDM. The time stepping of the first-order spatial deriva-
tives and the fourth-order precise central difference method. Furthermore, FVM is
connected with an analytical method based on Roe’s scheme and Upwinding scheme
for a limitless area with a steady water depth . The periodic waves in this instance

3
roughly correspond to the non-linear periodic long waves. Additionally, the present
numerical model takes into account dispersion, convective nonlinearities, and viscous
dissipation. Furthermore, by assuming a time-dependent radiative barrier, the bound-
ary condition is also considered. Comparing the simulation findings with earlier studies
[5, 9, 10] and [3] validates the validity of the new numerical model. Convergence anal-
ysis is also performed to confirm the accuracy of the existing numerical model. The
wave elevation can be increased or decreased, mostly depending on the computational
domain. The present Bousinesq’s model can be used to six key points in the realistic
Paradip port in Odisha, India, to investigate the impact of incident wave angle. To
identify the safe spot within the Paradip port, a discussion of the wave height profile
at each key region has been carried out. All things considered, the suggested numerical
method offers an effective tool to comprehend the wave elevation profile correspond-
ing to various resonant modes in the complex domain of changing bathymetry when
viscous dissipation, dispersion, and convective nonlinearity effects are added.

2 Formulation of Mathematical Model For Shallow

Water
Examining a rectangular domain [xL , xR ]×[yL , yR ], we find that it has top and bottom
borders, as seen in Fig.1. z = b(x, y) provides the bottom boundary, and z = b(x, y) +
h(x, y, t) = S(x, y, t) provides the top boundary. In this case, h(x, y, t) = G(x, y, t) +
h(x, y, t), η(x, y, t) = h(x, y, t) + zb (x, y), where zb = −0.531h(x, y, t) and η represent
the total water depth, h(x, y, t) is the variable water depth, and G(x, y, t) is the surface
dimension from the still water depth.

Fig. 1 Sketch of the figurer of water wave propagation in shallow water.

4
 The system’s dispersion properties were most accurately approximated by linear
wave theory when an optimal value of zb =−0.531h was employed. This allowed the
equations to be more broadly applicable than the traditional Boussinesq equations,
covering a greater range of water depths. To proceed further consider the following
Nwogu’s 2-D depth integrated nonlinear coupled Boussinesq equation are given as

∂ zb2 h2
    
∂η ∂ ∂ h
+ (h + η)u + [(h + η)v]y + ( − )h (uxx + vxy ) + zb +
∂t ∂x ∂y ∂x 2 6 2
 2
h2
   
∂ zb h
h ((hu)xx + (hv)xy ) + ( − )h (vyy + uxy ) + zb + h ((hv)yy + (hu)xy ) = 0,
∂y 2 6 2
(1)

∂u ∂u ∂u ∂η z2
+u +v +g + b (uxxt + uxyt ) + zb ((hut )xx + (hut )xy ) = 0, (2)
∂t ∂x ∂y ∂x 2
∂v ∂v ∂v ∂η zb2
+v +u +g + (vyyt + uxyt ) + zb ((hvt )yy + (hut )xy ) = 0. (3)
∂t ∂y ∂x ∂y 2
Equations ((1)-(3)) can be rewritten by letting ηt = (H −h)t = ∂H/∂), with ∂h/∂t = 0
we have
 2
h2
  
∂H ∂ ∂ z h
+ [(Hu)] + [(Hv)] + ( b − )h (ux + vy )x + zb + h ((hu)x + (hv)y )x
∂t ∂x ∂y 2 6 2 x
 2 2
  
zb h h
+ ( − )h (ux + vy )y + zb + h ((hu)x + (hv)y )y = 0,
2 6 2 y
(4)

and for the momentum equations (2) and (3), after multiplication of these equation
with H (total water depth) and using the relation gH∇η = gH∇(H − η) = gh∇H −
gH∇ h = (gH 2 /2)x + gH∇zb we get

∂u ∂u ∂u ∂η z2
H + Hu + Hv + gH + H b (uxxt + uxyt ) + Hzb ((hut )xx + (hut )xy ) = 0,
∂t ∂x ∂y ∂x 2

or we can write in more compact form as

  2   
z ∂ 1
(Hu)t + H b (uxx + vxy ) + zb ((hu)xx + (hv)xy ) + Hu2 + gH 2
2 t ∂x 2
(5)
∂u ∂zb
+ Hv = −gH − uΨc + ΨM x ,
∂y ∂x

5
similarly the other momentum equation for v takes the form

zb2
    
∂ 2 1 2
(Hv)t + H (vyyt + vxyt ) + zb ((hv)yy + (hu)xy ) + Hv + gH
2 t ∂y 2
(6)
∂u ∂zb
+ Hv = −gH − uΨc + ΨM y ,
∂x ∂y

thus equations (4)-(6)represent the conservative form as

∂
U + ∇. (F (U ∗ ) + G(U ∗ )) = S (U ∗ ) , Ω × [0, t] , ⊂ R2 × R+ (7)
∂t

where S(U ) = Sb + Sd is the source term, which contains the dispersive term Sd and
bottom slope Sb , and F and G are the nonlinear flux vectors.


    
H Hu Hv
U = Hu , F (U ∗ ) = Hu2 + 21 gH 2  G(U ∗ ) =  Huv 
Hv Huv Hv 2 + 21 gH 2
   
0 −Ψc
Sb = −gH ∂z∂x
b
Sd =  −uΨc + ΨM x 
−gH ∂z
∂y
b
−vΨc + ΨM y ,

Due to computational constraints, the aforementioned equations are used in this report
under the equivalent forms H = q1 , Hu = q2 , and Hv = q3 .
   
q2 q3
   
H q1 q3 q2
 2
U (q) = Hu = q2  , F (q) =  qq21 + 12 gq12  , G(q) = 
  
q1 
q q q32 1
Hv q3 3 2
q1 newline q1 + 2 gq12
    (8)
0 −Ψc
Sb = −gq1 ∂z
∂x
b
Sd =  −uΨc + ΨM x 
∂zb
−gq1 ∂y −vΨc + ΨM y ,

zb2
 
P = Hu + H (uxx + vyx ) + zb ((hu)xx + (hv)yx ) ,
2
 2 
zb
Φ = Hv + H (uxy + vyy ) + zb ((hu)xy + (hv)yy ) ,
2
( 
2
h2
       
zb h
Ψc = u − h u x + vy + zb + h (hu)x + (hv)y
2 6 x 2 x x
 !  )
 2
h2
      
zb h
 − h ux + vy + zb + h (hu)x + (hv)y  ,
2 6 y 2
y y

6
  
H
U ∗ = Hu
Hv
 2 
zb
ΨM y = Ht (uxy + vyy ) + zb ((hu)xy + (hv)yy ) ,
2
 2 
zb
ΨM x = Ht (uxx + vxy ) + zb ((hu)xx + (hv)xy ) ,
2

An ideal value of zb = −0.531h was proposed by Nwogu (1993) for modeling wind
generated waves by matching the ceterity from the linearized governing equations
and Airy wave theory over the sea depth parameter kh < π, where k represents the
wave number. The governing equation of the Boussinesq type describes the flow and
dispersion process using hyperbolic and parabolic components. In a given application,
the reference depth zb is a free parameter that is used in the parabolic term to optimize
the dispersion properties.
Z Z I
∇.F dA = F.n̂dΥ. (9)
Ω Υ

Next apply the Gauss divergence to Eq.(7) over the computational domain Ω and its
boundaries Υ we obtained
Z Z I Z Z
d ∗
S (U ∗ ) dΩ, Ω × [0, t] , ⊂ R2 × R+ .


U dΩ + H̄.n̂ dΥ = (10)
∂t Ω Υ Ω

T
In this case, Υ represents the borders of region Ω, where H̄ = [F, G].n̂ = [n̂x , n̂y ] is
the non-linear flux vector. The unit normal vector is pointing outward toward the Ω
region. In a control volume with computational cell Cp ⊂ Ω, a discrete approximation
to Eq. (10) will be used. This way, the surface integral will represent the overall flow
through the volume boundaries, while the volume integral will represent the integral
over the area Cp . At a given time, the average values of the conserved values over the
control volume are shown by
Z Z
1
Up = U ∗ dΩ, (11)
|Cp | Cp
the conservative form of the Eq.(10) can be written as

∂Up∗
I Z Z
1 1
+ (F nˆx + Gnˆy ) dΥ = SdΩ, (12)
∂t |Cp | Υp |Cp| Cp

3 Numerical formulation
3.1 State Interpolation
The numerical solutions of q̂ at x 12 , x 32 ,..., xM + 21 are required in order to evaluate the
fluxes F 12 , F 32 ,..., FM + 21 in our proposed finite volume equation. These solutions are

7
denoted by q̂ 12 , q̂ 32 ,..., q̂M + 32 . However, since the numerical solutions of q̂i are constant
in space in each cell, each cell Ii is probably going to have a different numerical solution
than its nearby cells. Because the numerical solution of q̂i is often discontinuous at
the boundary xi− 12 and xi+ 12 of each cell Ii , the values q̂i+ 21 and q̂i− 12 are therefore
not simple. To address this, a broad strategy is implemented by setting

Fi+ 12 = F(q̂li+ 1 , q̂ri+ 1 ), (13)

2 2

where the numerical solutions to the left of xi+ 21 are q̂li+ 1 , while the solutions to
2
the right of xi+ 12 are q̂ri+ 1 , and the flux evaluation function is F(q̂li+ 1 , q̂ri+ 1 ).
2 2 2

3.2 First-order Upwind

The simplest choice for q̂li+ 1 . Let q̂ri+ 1 and is the setup
2 2

q̂li+ 1 = q̂i , q̂ri+ 1 = q̂i+1 , (14)

2 2

this represent the first order accurate in space.

3.3 Second-order Upwind

Van Leer offered a more complex solution, which is taking
1+κ 1−κ
q̂li+ 1 = q̂i + (q̂i+1 − q̂i ) + (q̂i − q̂i−1 ), (15)
2 4 4
1+κ 1−κ
q̂ri+ 1 = q̂i+1 + (q̂i − q̂i+1 ) + (q̂i+1 − q̂i+2 ), (16)
2 4 4
for which the parameter κ must be selected from the range [−1, 1]. For all values of
κ, it is shown that this method is second-order accurate in space. This approach uses
the average of q̂i and q̂i+1 to become a central interpolation for κ = 1. It reduces to a
completely one-sided extrapolation for κ = −1. For κ = 1/3, it can even be third-order
accurate in space. . The displayed model is
∂Up∗
I Z Z
1 1
+ (F nˆx + Gnˆy ) dΥ = SdΩ, (17)
∂t |Cp | Υp |Cp | Cp
For the finite volume method, we need to interpolate the state variables at the cell
interfaces.

3.4 First-order Upwind Scheme

Using the first-order upwind scheme, the interpolated values at the cell interfaces are:
∗
Up,l = Up∗ , Up,r
∗ ∗
= Up+1 (18)
The flux at the interface is then evaluated using these interpolated values:
∗ ∗ ∗ ∗
Fp+ 12 = F (Up,l , Up,r ), Gp+ 12 = G(Up,l , Up,r ). (19)

8
3.5 Second-order Upwind Scheme
Using the second-order upwind scheme, the interpolated values at the cell interfaces
are:

∗ 1+κ ∗ 1−κ ∗
Up,l = Up∗ + (Up+1 − Up∗ ) + ∗
(Up − Up−1 ), (20)
4 4
∗ ∗ 1+κ ∗ ∗ 1−κ ∗ ∗
Up,r = Up+1 + (Up − Up+1 )+ (Up+1 − Up+2 ), (21)
4 4
for which the parameter κ must be selected from the range [−1, 1]
The flux at the interface is then evaluated using these interpolated values:
∗ ∗ ∗ ∗
Fp+ 21 = F (Up,l , Up,r ), Gp+ 12 = G(Up,l , Up,r ). (22)
The domain is separated into a finite number of control volumes in order to apply the
finite volume approach. The integrals are approximated using the proper numerical
methods, and the temporal derivative term is estimated using a finite difference
method.

Step 1:The time derivative can be approximate by the forward difference scheme.

∂Up∗ Up∗,n+1 − Up∗,n

≈ , (23)
∂t ∆t
where ∆t is the time step, and Up∗,n , Up∗,n+1 are the values of Up∗ at time levels n and
n + 1, respectively.
Step 2: Discretize the Flux Terms The flux integrals over the boundary Υp can be
approximated as a sum over the edges of the control volume. Let e denote an edge of
the control volume, and ∆Υe be the length of the edge. The flux terms can be written
as: I X
(F nˆx + Gnˆy ) dΥ ≈ (Fe nˆx + Ge nˆy ) ∆Υe , (24)
Υp e
where Fe and Ge are the fluxes through edge e.

Step 3: The source term S = Sb + Sd in integral form can be approximated as

Z Z
SdΩ ≈ Sp |Cp |, (25)
Cp
where Sp is the average value of the source term in the control volume Cp .

Step 4: Combine All Terms Combining the discretized terms, we get:

Up∗,n+1 − Up∗,n 1 X
+ (Fe nˆx + Ge nˆy ) ∆Υe = Sp . (26)
∆t |Cp | e

9
 Step 5: Solve for the New Time Level Rearranging the equation to solve for Up∗,n+1 ,
we get
∆t X
Up∗,n+1 = Up∗,n − (Fe nˆx + Ge nˆy ) ∆Υe + ∆tSp . (27)
|Cp | e
This is the outcome of using the Boussinesq model with the finite volume approach.
Every time step, the solution Up∗ can be updated using the resulting equation. The
Appendix section 6 discusses the flux evaluation through each interface and the flux
vector splitting using Roe’s scheme.
The governing equations are spatially discretized using a hybrid finite volume and
finite difference method. The finite-differencing procedure for the source words is
straightforward and is not repeated here. Our primary area of interest is the numerical
application of the convective flux term within the FV method’s framework. To satisfy
the requirements for numerical stability, the time increment ∆t must be limited by
the CFL condition.
!
∆x ∆y
∆t = Cr min p , p (28)
i,j | ui,j | + gHi,j | vi,j | + gHi,j
the Courant number, Cr, is fixed at 0.5 for each simulation.

3.6 System of equations for evaluation of velocities

The explicit time-stepping approach integrates the hyperbolic and source terms of
the governing equations to produce the evolution variables H, P, and Q at the cell
centroid.The parabolic parts of the governing equations that are limited to expression
in spatial derivatives are included in the source term. The evolution variables P and
Q contain the remaining dispersion terms, which are x, y, and xy derivatives of u and
v.Discreteizing the dispersion terms in Eqs. (22) and (23) yields two sets of equations
in terms of u and v, respectively. Using a second-order central difference approach,
this is accomplished.
  
(zα )i,j ui+1,j − 2ui,j + ui−1,j
Pi,j = (Hu)i,j + Hi,j (zα )i,j
2 ∆x2
 (29)
(hu)i+1,j − 2(hu)i,j + (hu)i−1,j
+ (vi,j )xy + + ((hv)i,j )xy
∆x2
  
(zα )i,j vi,j+1 − 2vi,j + vi,j−1
Qi,j = (Hv)i,j + Hi,j (zα )i,j
2 ∆y 2
 (30)
(hv)i,j+1 − 2(hv)i,j + (hv)i,j−1
+ (ui,j )xy + + ((hu)i,j )xy
∆y 2

10
 By solving the velocity components u and v independently for each dimension, the
implicit solution for velocity results from the effects of the parabolic terms within the
governing equations.

    
Hi,j C1,j Hi,j R1,j ... ... ... u1,j P1,j
Hi,j L2,j Hi,j C2,j Hi,j R2,j ... ...   u2,j   P2,j 
    
 ... Hi,j L3,j Hi,j C3,j ... ...   u3,j   P3,j 
  ..  =  ..  (31)
    
 .. .. .. .. ..

 . . . . .  .   . 
   
 ... ... ... ... Hi,j RN −1,j  uN −1,j  PN −1,j 
... ... ... Hi,j LN,j Hi,j CN,j uN,j PN,j

Similarly, for the velocity v:

    
Hi,j Ci,1 Hi,j Ti,1 ... ... ... vi,1 Qi,1
Hi,j Bi,2 Hi,j Ci,2 Hi,j Ti,2 ... ...   vi,2   Qi,2 
    
 ... Hi,j Bi,3 Hi,j Ci,3 ... ...   vi,3   Qi,3 
  ..  =  ..  (32)
    
 .. .. .. .. ..

 . . . . .  .   . 
   
 ... ... ... ... Hi,j TN −1,j  vi,N −1  Qi,N −1 
... ... ... Hi,j Bi,N Hi,j Ci,N vi,N Qi,N

The terms for Ci,j , Li,j , Ri,j , Bi,j , and Ti,j are given as follows:

(zα )2i,j
 
1 (zα )i,j 2hi,j
Ci,j = (zα )i,j − − + (vi,j )xy + (zα )i,j (hv)i,j (33)
(zα )i,j ∆x2 ∆x2 2
   
(zα )i,j hi−1,j (zα )i,j 2hi+1,j
Li,j = (zα )i,j + , Ri,j = (z )
α i,j + (34)
∆x2 ∆x2 ∆x2 ∆x2
   
(zα )i,j hi,j−1 (zα )i,j hi,j+1
Bi,j = (zα )i,j − , Ti,j = (zα )i,j + (35)
∆y 2 ∆y 2 ∆y 2 ∆y 2
Only the adjacent cells in the x or y directions are taken into account by the
central difference method, and the tri-diagonal matrices are symmetric and positive
definite. A Thomas algorithm was employed by Ferziger and Peric to determine the
flow velocity (u,v).

3.7 Boundary conditions

Appropriate boundary conditions are required for numerical models; in this work, three
different types of boundary conditions were used: (1) Reflected boundary condition;
(2) Internal generating waves; and (3) Absorbing boundary condition.

11
3.7.1 Reflective boundaries
The wave will be fully reflected when it hits a solid wall. The boundary requirements
for a general reflecting barrier with an outward normal vector ~n should be

~u · ~n = 0, (36)

∇η · ~n = 0. (37)
where the fluid domain is Ω, the boundary is ∂Ω, the outward normal vector is n, and
the position inside the boundary is x. As the tangent to the border of the velocity
component u.T notation, we need

∂(u.T )
= 0, (38)
∂n
This boundary condition does not conflict with the in-viscous flow under consideration
because it imposes a no-shear flow condition along the boundary walls.

3.7.2 Initial waves generation

The internal wave on constant water depth proposed by Wei et al. [24] requires the
addition of a source term S(x, y, t) in the continuity equation. This is achieved by using
a source term function appended. The source function was obtained by applying the
Fourier transform and Green’s functions to the computational domain wave generating
approach, which involved solving the linearized and non-homogeneous equations of
Nwogu. The current model uses this source function wave-making technique to let the
reflected waves easily depart through the wave generator at each (simulation) time
step, a source function S(x,y,t) is introduced to the mass conservation equation to
produce the desired oscillation signal in the wave producing area.

80
S(x, t) = Dexp σ(x − xs )2 sin(θ),


with σ= , (39)
δ 2 L2
where θ = λy −ωt is the wave incident angle, ω is the wave frequency, L is the wave
length, and xs is the position of the wave-making area. The wave number is denoted in
the y direction by the expression λ(= ky = ksinθ). The wave generator area’s width,
W = δL/2, is determined by the parameter δ, However, for a monochromatic wave,
the amplitude of the source function, D, is defined as
√
2 σA0 cos(θ(ω 2 − α1 gh3 k 4 )
D= √ , (40)
ωk πexp(−l2 /4σ(1 − α(kh)2 )
where η is the surface evaluation, α = −0.531, l(= kx = kcos(θ)). In the x-direction,
α1 = α+1/3 is the wave number. The linear superposition of component waves is what
we employ for irregular waves. According to the irregular wave notion, the surface
evaluation η is defined as
∞
X
η(t) = ai cos(ωi t + i ), (41)
i=1

12
where i denotes the beginning phase of the waves, and ai and ωi denote the amplitude
and frequency of the component wave, respectively. The source function now takes the
following form using this idea
M

X
S(x, y, t) = exp σ(x − xs )2 Di sin(λi y − ωi t + i ), (42)
i=1
the source amplitude D takes the form
√
2 σAi cos(θi (ωi2 − α1 gh3 ki4 )
Di = √ , (43)
ωki πexp(−li2 /4σ(1 − α(ki h)2 )
where li = ki sin(θi ). The maximum wavelength between the components is used to
determine the wave-making area W.

3.7.3 Absorbing boundaries

An ideal absorbing boundary simulates open boundary without wave reflection,
absorbing all wave energy and producing no reflection, with defined wave phase speed
and propagation direction. The radiation condition is now defined as

∂η ∂η
+ c cosθ = 0, (44)
∂t ∂x
The estimated radiation condition for two-dimensional applications is expressed as

∂2η ∂ 2 η c2 ∂ 2 η
+c − = 0, (45)
∂tt ∂xt 2 ∂yy

A damping layer will be applied in front of the lateral radiation borders of the compu-
tational domain utilizing a number of tried-and-true methods in order to approximate
the absorbing boundaries in this study. In this layer, the momentum equations are
supplemented with the damping terms. For instance, our primary equation is changed
as follows in the horizontal direction:

Ut = F (η, u) − w1 (x)u − w2 (x)uxx , (46)

In the context of the Navier-Stokes equation, the second-order derivative term
aligns with the linear viscous terms [15]. The damping coefficients, w1 (x) and w2 (x),
are defined as follows:
(
0, for x < xs ,
w1 (x) = (47)
α1 ωf (x), for x > xs ,
(
0, for x < xs ,
w2 (x) = (48)
α2 ωf (x), for x > xs ,
where ω denotes the viscous coefficient, and xs refers to the starting position of the
damping layer within the computational domain, which extends from x = 0 to x = x1 .

13
The constants α1 and α2 are determined for each specific scenario, and f (x) is given
by the following expression:
 n 
exp xx−x s
1 −xs
−1
f (x) = . (49)
exp(1) − 1
Numerical experiments show that incorporating a damping layer alongside a radi-
ation boundary condition results in more effective wave damping than using radiation
boundary conditions alone. The constant n is chosen depending on the specific case,
and the damping layer is typically designed to have a width equal to two to three
times the wavelength of the waves.

4 Validation
4.1 Convergence Analysis
In this case, the error norm for N1 , N2 , and N3 is found after discretizing the rectangle
harbor into N1 = 2000,N2 = 3000, and N3 = 4000 number of elements.The least
squares formula is used to express the number of discrete elements. FThe results of
the convergence analysis are displayed for a rectangular domain in Fig. 2, where the
ordinate (right panel) represents the error norm and the abscissa (left panel) represents
the total number of discretized elements for FVM; the abscissa logarithm of the total
number of discrete elements. The FVM’s order of convergence for the rectangular
domain is 1.96, and the error norm rapidly decreases as the element count increases.
This illustrates the convergence and validates the current numerical scheme.

× 10-7 Error Norm vs. Domain Division (Ni) Logarithmic Error Norm vs. Logarithm Value of Domain Elements
4 -6.4
FVM FVM

3.5 -6.5

-6.6
3
Logarithmic Error Norm

-6.7
Error Norm

2.5
-6.8
2
-6.9

1.5
-7

1 -7.1

0.5 -7.2
0 1 2 3 4 5 6 7 8 9 10 2 2.5 3 3.5 4 4.5 5 5.5 6
Domain Division (Ni) × 105 Logarithm Value of Domain Elements (log Ni)

Fig. 2 By applying current numerical technique, the error norm plot between error norm and total
number of discretized elements (left panel) and logarithmic error norm and total number of discretized
elements’ logarithm value (right panel) are obtained.

14
4.2 Bichromatic waves groups
When a wave breaking model for random waves is combined with a runup method,
Mase’s laboratory measurements—which include shoaling, breaking, and swash
motion—can offer helpful test cases. On a beach with a slight inclination, the run-up,
breaking, and transformation of bichromatic wave groups is the goal of the bichro-
matic waves. The test wave flume was 27 meters in length and 0.5 meters in width.
The irregular wave generator was located at the left end of the flume, while a lit-
tle beach with a 1 : 20 slope was located at the right end. Ten meters from the
beach’s toe was where the wave generator was placed. A wave-flume with dimensions
(x, y) ∈ [−5, 25] × [0, 2.5]m and an undisturbed sea depth (h = 0.47 m) at x = 0 m is
assumed in our numerical experiment. Information obtained from four specially cho-
sen wave gauges (WG) along the flume, namely those labeled 1 (at x = 10 m at the toe
of the sloping beach), 8 (at x = 16.9 m), 10 (at x = 17.9 m), and 12 (at x = 18.9 m)
were selected for comparison in the present study, as shown in Fig. 3, and the width
of the wave generator region, determined by the parameter δ in 39, was set at 0.5. A
γ value of 0.6 and a CFL value of 0.3 were used in the wave breaking criterion. To
absorb wave energy and prevent non-physical reflections from the closed boundary, a
4 m sponge layer was placed to the left end of the domain.
The following equations were used to describe the resulting bichromatic wave trains
 
1
η(t) = A cos(2πf1 t) + A cos(2πf2 t), f1,2 = fm 1 ± , (50)
20
the wave amplitude is denoted by A and fm is the mean frequency. Three distinct
mean frequency values,fm = 1, 0.6, and0.3Hz and A = 0.015m have been used in
this experiment.A shift in the wave’s mean frequency causes its properties to change.
The steepness of deep ocean waves reduces with decreasing FM, and the likelihood
of plunging breakers occurring instead of overflowing breakers increases at low mean
frequencies. Using the surf similarity parameter established by Battjes [6], it is
explained in Tonelli and Petti [7]. The Spilling breaking of waves occurs in the first
and second test cases, or those with fm = 1Hz and 0.6 Hz, respectively, whereas the
wave breakers in the third test case, which utilizes fm = 0.3 Hz, are mostly of the
plunging type. In these test situations, bed friction was disregarded, and 10( − 6) was
chosen as the wet/dry wd tolerance parameter.
We had to determine the frequency components of the original wave signal because
the numerical wave trains were generated internally using linear theory and the
bichromatic wave pattern provided in Eq. (50). However, it took multiple attempts to
discover the optimum match with the measured waves due to a discrepancy between
the intended and actual experimental frequencies and amplitudes . It is mentioned
in Tonelli and Petti [7] that as there was no active absorption of reflected waves and
spurious harmonic generation was not adjusted at the generator in the trials, it is
challenging to precisely replicate the laboratory settings. After multiple attempts,
we discovered that we could replicate the original signal with mean frequencies of
fm = 1.03, 0.605, and 0.31 Hz, respectively, for each example.

15
 M. Kazolea et al. (2018)
0.05 present numerical scheme
Finite volume scheme by Mase(1995)

η (m)
0

-0.05
60 62 64 66 68 70 72 74 76 78 80

0.05

η (m)
0

-0.05
60 62 64 66 68 70 72 74 76 78 80

0.05
η (m)

-0.05
60 62 64 66 68 70 72 74 76 78 80
t (s)

Fig. 3 Time series of the surface elevation for Bichromatic waves with fm = 1 Hz in wave gauges
WG1, WG2, and WG3 (top to bottom) are compared between Mase’s experimental data(red dot-
ted)Mase [5], our numerical solution (black dotted line) and Kazolea analytical solution (blue line)
Kazolea and Delis [10].

M. Kazolea et al. (2018)

0.1 present numerical scheme
Finite volume scheme by Mase(1995)
η (m)

-0.1
60 70 80 90 100 110 120 130
0.05
η (m)

-0.05
60 70 80 90 100 110 120 130
0.05
η (m)

-0.05
60 70 80 90 100 110 120 130
0.05
η (m)

-0.05
60 70 80 90 100 110 120 130
t (s)

Fig. 4 Time series of the surface elevation for Bichromatic waves with fm = 0.6 Hz in wave gauges
WG1, WG2, and WG3 (top to bottom) are compared between Mase’s experimental data(red dot-
ted)Mase [5], our numerical solution (black dotted line) and Kazolea analytical solution (blue line)
Kazolea and Delis [10].

16
 M. Kazolea et al. (2018)
0.04
present numerical scheme

0.02 Finite volume scheme by Mase(1995)

η (m)
0

-0.02

-0.04
60 62 64 66 68 70 72 74 76 78 80

0.04

0.02
η (m)

-0.02

-0.04
60 65 70 75 80 85 90 95
t (s)

Fig. 5 Comparison of the time series of the shoreline elevation for bichromatic wave trains produced
with different fm values are compared between Mase’s experimental data(red dotted)Mase [5], our
numerical solution (black dotted line) and Kazolea analytical solution (blue line) Kazolea and Delis
[10].

Figs. 3 and 4 compare the time series of the surface elevation measured at the
different wave gauges for the different cases, both experimentally and numerically. In
the same way, Fig. 6 contradict the time series of the vertical shoreline displace-
ment during the runup and run-down in the three test scenarios that were recorded
experimentally and numerically. The beach is where wave breaking occurs, and the
run-up and run-down patterns that were recorded in each sample show a wide range
of breaking regime differences. As seen in Fig. 5, the region’s swash motion is domi-
nated by short frequency oscillations (0.3Hz and 0.6Hz), with plunging breaking waves
predominating. This happens when the waves must run-up independently because the
incident wave period is longer than the swash cycle. But things are obviously different
at higher frequencies. In this instance, lengthy waves predominate in the swash action,
while breakingwaves are of the spilling variety. As a result, individual swashes become
less evident, and a group-induced low-frequency wave takes center stage in the swash
motion.

17
 M. Kazolea et al. (2018)
0.05
present numerical scheme
Finite volume scheme by Mase(1995)

η (m)
0

-0.05
60 62 64 66 68 70 72 74 76 78 80
0.05

η (m)
0

-0.05
60 65 70 75 80 85 90 95

0.05
η (m)

-0.05
60 65 70 75 80 85 90 95
t (s)

Fig. 6 Time series of the surface elevation for Bichromatic waves with fm = 0.3 Hz in wave gauges
WG1, WG2, and WG3 (top to bottom) are compared between Mase’s experimental data(red dot-
ted)Mase [5], our numerical solution (black dotted line) and Kazolea analytical solution (blue line)
Kazolea and Delis [10].

Despite a few small discrepancies, Overall, there is excellent agreement between

the experimental results and the projected surface elevation. Furthermore, there is a
strong correlation between the observations and the generated swash motions. This
good agreement shows that wave shattering, breaking, and swash oscillation can be
simulated in a range of wave situations using the wave breaking technique of the
current numerical model.

18
 Wave gauge WG1 Wave gauge WG2
5 5
Simulated Simulated
4.5 Experimental by Beji(1993) 4.5 Experimental by Beji(1993)

4 4

Normalized S 3.5 3.5

Normalized S
3 3

2.5 2.5

2 2

1.5 1.5

1 1

0.5 0.5

0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
f (Hz) f (Hz)

Wave gauge WG3 Wave gauge WG4

5 5
Simulated Simulated
4.5 Experimental by Beji(1993) 4.5 Experimental by Beji(1993)

4 4

3.5 3.5
Normalized S

Normalized S
3 3

2.5 2.5

2 2

1.5 1.5

1 1

0.5 0.5

0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
f (Hz) f (Hz)

Fig. 7 Comparisons of normalized spectra for both our current analytical result (solid blue lines)
and experiment result by Beji and Battjes [9] at the Gauges (stations) WG1, WG2, WG3, and WG4.

The normalized energy spectrum for both our current result and experimental
result for four located stations is shown in Fig.(7). Following the experimental results
from Beji and Battjes [9], In the whole region beneath each spectrum, four wave gauges
along the submerged bar equal one.The carrier frequency on the lee side of the bar
releases the higher harmonic bound that developed along the front slope as long as
the water depth frequency kh naturally rises. The density of spectral energy at high
frequencies is somewhat overestimated by the numerical findings, especially at the
gauges positioned toward the bar’s lee side. In this case, the peak energy is due to
the overshoaling of the numerical waves on the relative Steepen front slope of the bar.
Nwogu’s weakly nonlinear dispersive BT model is well known for this characteristic.
At the lee side of the bare, where de-shoaling takes place when the energy rises at the
fundamental harmonic, the top and front of the bar are carried over. Nevertheless, as
the numerical model correctly illustrates, the overall results from both experimental
and analytical methods are satisfactory. This is because de-shoaling causes the wave
breakdown and redistribution of the total energy among the fundamental harmonic
and other harmonics to primarily occur in the area behind the bar where the water
depth rises again.

19
4.3 Solitary wave interaction for rectangular domain
To evaluate the numerical model in 2D instances, we now look at the wave evolution
in a closed basin of size Lx, Ly, where the origin is defined at the basin’s left-bottom
corner. The initial free surface displacement, which has a Gaussian distribution about
the basin center (t = 0.0 s), provides the initial surface removal from the still water
depth.

  
Lx 2 Ly 2
G0 (x, y, 0) = h0 exp −β (x − ) + (y − ) (51)
2 2

20
Fig. 8 3D Surface Elevation for different time values using standard Boussinesq Equations.

where Lx and Ly both have taken as 10 and h0 = 0.1, β = −0.4. The extended
Boussinesq equation is applied to irregular model wave propagation over a gradually
changing bathymetry in shallower to deeper water. Here, the model is restated in
the form of a conservation rule, and as such, a conservative unstructured FV scheme
can numerically approximate it Nwgou [2]. Instead of using the widely used depth-
averaged velocity, they developed a system of equations that used the velocity u = ua
at any distance zb from a still water level h as the velocity variable.
The 3D plots shown in Figs.(8) generated at various time steps t = 0 s, 20 s, 40 s, 80 s
illustrate the evolution of the surface elevation η(x, y, t) of the shallow water domain

21
governed by the Boussinesq equations. These results are crucial for understanding
wave propagation dynamics, non-linear wave interactions, and the dispersive effects
in shallow water environments.
At t = 0, the Gaussian initial condition is clearly observed, centered at (Lx /2, Ly /2),
which represents a localized disturbance in the water surface. This disturbance could
correspond to an initial wave packet or a localized energy release, simulating, for
example, the effects of a sudden wave generation or a localized wind disturbance.
As time progresses, the 3D plots for t = 20 s and t = 40 s show the gradual spreading
of the initial wave in both the x- and y-directions. The dispersive nature of the
Boussinesq model becomes evident in these plots, where the wave spreads while main-
taining its central peak due to the balance between non-linear effects and dispersion.
The wave fronts are almost symmetric, indicating uniform wave propagation from the
disturbance center across the domain, which is characteristic of shallow water wave
dynamics when depth is relatively constant across the domain.
The Boussinesq model incorporates both non-linearity and dispersion, allowing for a
more realistic representation of shallow water waves compared to simpler models such
as the shallow water equations (SWE). The spreading wave fronts in the plots indi-
cate how the dispersive effects slow down the wave propagation speed in comparison
to the linear wave theory, where dispersion is not accounted for. This is particularly
evident in the attenuation of the wave amplitude over time (e.g., at t = 40 s and
t = 80 s), where the wave fronts continue to expand, but the peak elevation decreases
due to the dispersive spreading of wave energy.
By t = 80 s, the surface elevation has evolved significantly. The disturbance has
spread across most of the domain, and the 3D plot shows a broader wave pattern with
a diminished peak at the center. This time step illustrates the long-term evolution
of the wave packet, where non-linear effects continue to influence the wave shape,
but dispersion dominates, leading to a wider, more diffused wavefront. The decrease
in amplitude suggests energy dispersion, a key feature of the Boussinesq model that
accounts for the finite-depth effects in shallow water. This long-term dispersion of
energy is crucial in coastal engineering applications, particularly in understanding
how waves dissipate over time and distance. Although not explicitly modeled with
reflective boundaries in the current setup, the behavior of the waves near the domain
edges (especially in later time steps) can give insights into potential boundary reflec-
tions and their influence on the interior solution. In practical applications, such as
in coastal regions or harbor simulations, the accurate representation of boundary
reflections is critical. The current simulation does not show significant reflections
due to the sufficiently large domain and relatively short simulation time, but in a
longer simulation, such effects would likely become more pronounced. This suggests
that future work could incorporate absorbing boundary conditions or simulate wave
interactions with obstacles to capture more realistic scenarios.

The observed wave spreading and attenuation in the 3D plots provide valuable
insights into wave dynamics in coastal or harbor areas. The Boussinesq model’s abil-
ity to capture both dispersive and non-linear effects makes it an essential tool for
modeling wave behavior in these regions. The initial peak in surface elevation could

22
correspond to a wave generated by a ship or a localized wind disturbance, and the sub-
sequent spreading represents how such waves would propagate through shallow water
environments. The numerical results highlight how waves lose energy as they propa-
gate, which is critical in designing coastal defenses, understanding sediment transport,
and predicting the effects of waves on structures such as breakwaters or piers.
The 3D plots at different time intervals provide a clear visualization of wave prop-
agation in shallow water under the Boussinesq model. The results illustrate the
importance of both dispersive and non-linear effects, showing how waves evolve from a
localized disturbance, propagate through a shallow water domain, and eventually dis-
perse energy over time. Our comprehension of wave dynamics in a variety of shallow
water scenarios, from man-made disturbances in coastal zones to natural events like
tsunamis, can be enhanced by these numerical results. Future studies can extend this
work by incorporating complex boundary conditions, bathymetry variations, and inter-
actions with structures to further enhance the applicability of the model in real-world
scenarios.

t=0 t = 10
2 2

1.8 1.8

1.6 1.6

1.4 1.4

1.2 1.2
H(m)

H(m)

1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
-50 0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
x (m) x (m)

t = 15 t = 20
2 2

1.8 1.8

1.6 1.6

1.4 1.4

1.2 1.2
H(m)

U(m)

1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0
0 50 100 150 200 250 300 350 0 50 100 150 200 250 300 350
x (m) x (m)

Fig. 9 Constant solitary wave profiles for different time at t=0, 10, 15, 20.

The analysis of solitary wave profiles generated using the 2D Boussinesq equations
for shallow water reveals insightful dynamics over time at constant amplitudes (see
figs (9. At the initial time t=0, the solitary wave exhibits a pronounced peak at its
crest, which is indicative of the initial wave energy being highly localized. This profile
serves as a benchmark for subsequent time steps. By t=10 seconds, the wave begins to

23
propagate, showing a slight dispersion that results from the balance between nonlin-
earity and dispersion inherent in the Boussinesq model. As the wave progresses to t=15
seconds, the peak slightly broadens, and the amplitude remain constant marginally,
indicating energy distribution over a broader spatial region. Finally, at t=20 seconds,
the solitary wave continues to maintain its shape and amplitude.

4.4 2-D run up of solitary profile at the Paradip port region

Deepwater natural port located in Paradip on India’s east coast is called Paradip
Port. The port of Paradip is accessed by a dredged canal and features an artificial
lagoon-style harbor that is shielded by two ”Break Waters” made of rubble mounds.
On the port’s northeastern edge, the North Break Water measures 538 meters, while
the south break water measures 1217 meters. Large-scale laboratory experiments were
conducted to study solitary wave runup on a conical island, and the results were
published in (briggs et al. [4]). The main impetus for the research was the discovery
of surprisingly high run-up heights on the lee side of tiny islands during many large-
wave tsunami incidents in the 1990s. The water surface height was measured at the
following locations using three wave gauges: W G1 = (4.4, 0.275), W G2 = (4.5, 0.170),
W G3 = (4.6, 0.045).,W G4 = (4.7, 0.035).,W G5 = (4.8, 0.025).,W G6 = (4.9, 0.015).

Fig. 10 Google mapping image of the Paradip port showing the six record stations that were chosen
and labeled as WG1 through WG6, each of which is located in a different critical area within the port.

The right sktich shown in Fig.10 depicts a schematic representation of water wave
interaction in a harbor. The open sea region is denoted by Ω, which is the larger,
unbounded area where incident waves originate and travel towards the harbor. The
harbor itself is represented by the bounded subregion Cp ⊂ Ω. The harbor is protected
and may have complex geometries as indicated by the irregular boundary of Cp .
The pseudo boundary, denoted by Γp , is a dashed circular arc within Ω, representing a
boundary where certain conditions or measurements might be applied. Incident waves
are shown coming from the open sea towards the harbor, as indicated by arrows on

24
both sides of the image. Point O is the origin or a reference point located within the
harbor. Line AB crosses through the harbor’s entrance from point A to point B, and
line DE represents another line segment that another entrance or boundary line. The
radius RU extends from the origin to the pseudo boundary Γp , while the radius RB
extends from the origin to another significant point or boundary within the geometry.
Additionally, point F serves as another reference point within the open sea region.
This sketch is crucial for studying wave dynamics, harbor protection, and coastal engi-
neering applications, particularly for ports like the Paradip port in Odisha, India. In
this section, the wave evaluation η(x, y, t) is carried out for TT-Shaped harbours.The
total domain of length of 2000cm with width 1500cm and depth 60cm is considered.
The image displays the model geometry of the TT-Shaped harbor, normalized by the
parameter λ. The boundary is located at a 0.6-cm dimensional radius, the y-axis is
parallel to the coastal area, the x-axis is oriented toward the ocean, and the wave pro-
file is measured at wall B. The complete domain is discretized into 6808 triangular
tiny elements with normal displayed as red and total nodal points of 2393.

(a) (b)
1 1
η (m)

0 η (m) 0

-1 -1
0 10 20 30 40 50 60 0 10 20 30 40 50 60
1 1
η (m)

η (m)

0 0

-1 -1
0 10 20 30 40 50 60 0 10 20 30 40 50 60
0.05 0.1
η (m)

η (m)

0 0

-0.05 -0.1
0 10 20 30 40 50 60 0 10 20 30 40 50 60
Time (sec) Time (sec)

Fig. 11 Surface Elevation η for TT-Shaped harbour at different h0 values using standard Boussinesq
Equations for (a) h0 =1,5,7 and (b) h0 =10,15,20.

Figure shows how the wave profiles (η) change over time (t) for various hight
regions (11). In the left panel (a) we taken the smaller values of the water hight (1cm
,5cm,7cm ) and at the right panel we taken some higher values for the height of the
harbour (10cm,15cm, 20cm). The amplitudes and frequency of the wave profiles are
different in all the cases. It has been observed that in the harbor with a TT form,
it can be seen that the wave height reach to a maximum values at the lowest point
in the area. From here one can conclude that the TT shaped geometry can be used
for artificial harbor at the shallow water. As a result, the harbor’s form is crucial in

25
reducing wave height.
Two breakwaters that form an artificial lagoon-style harbor protect Paradip Port, a
deep-water port on India’s east coast. We use the Boussinesq model to investigate
wave propagation in this port, taking into consideration non-linear and dispersive
effects. The domain is discretized with a grid spacing of 10 meters and a time step of
0.1 seconds. The domain is roughly two kilometers long and one and a half kilometers
wide. Absorbing boundary conditions with a sponge layer width of 3 meters are used
to prevent reflections at the domain boundaries.
Wave gauges are placed at coordinates corresponding to WG1 (400 m, 200 m),WG2
(600 m, 300 m),WG3 (800 m, 400 m) WG4 (800 m, 400 m), WG5 (1700 m, 1100 m)
and WG6 (2000 m, 1000 m). The initial surface elevation is modeled as a Gaussian
pulse centered in the domain. The average water depth is assumed to be 15 meters.
The Boussinesq equations are solved using a finite difference method with explicit
time stepping. Surface elevation data is collected at each wave gauge over the simu-
lation period, and synthetic experimental data is generated by adding noise to the
numerical results.

26
 Surface Elevation at Wave Gauge 1 Surface Elevation at Wave Gauge 2
0.15 0.25

0.2
0.1
0.15

0.1
0.05
0.05
η at WG 1

η at WG 2
0 0

-0.05
-0.05
-0.1

-0.15
-0.1
-0.2

-0.15 -0.25
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
x(m) x(m)

Surface Elevation at Wave Gauge 3 Surface Elevation at Wave Gauge 4

0.15 0.1

0.08

0.1 0.06

0.04

0.05 0.02
η at WG 3

η at WG 4

0 -0.02

-0.04

-0.05 -0.06

-0.08

-0.1 -0.1
0 10 20 30 40 50 60 70 80 90 100 0 10 20 30 40 50 60 70 80 90 100
x(m) x(m)

Fig. 12 A perspective view of the surface elevation along the centerline at different gauges and the
free surface (top) demonstrates how regular waves are shaded.

The numerical simulation of 2-D wave propagation is shown ing Fig. 12 using the
Boussinesq equations models the behavior of surface water waves in a 2000m × 1500m
domain over a 100s period with a Gaussian pulse as the initial disturbance. With spa-
tial and temporal resolutions of 10m and 0.1s, respectively, the finite difference method
is employed to update horizontal velocity components and surface elevations itera-
tively. Wave gauges positioned at (800, 400), (1000, 500), (1300, 700), and (1700, 1100)
meters capture the time series of surface elevation, revealing wave dispersion and inter-
actions with domain boundaries. The results show smooth wave profiles initially, with

27
increasing complexity and reduced amplitudes as waves propagate and interact. The
simulation highlights the importance of spatial and temporal resolutions, boundary
effects, and initial conditions in accurately capturing wave dynamics, offering insights
for refining models with varying bathymetry and enhanced boundary treatments.

5 Conclusion
In this study, we consider a Nwogu’s 2-D depth integrated nonlinear-Boussinesq type
equation for shallow water wave propagation with using nonlinear-phenomena over the
varying bathymetry in the costal region. We use the finite volume technique (FVM)
and the finite difference method (FDM) to solve the conservative version of the math-
ematical model. The 2-D nonlinear Nwogu’s pair Boussinesq type problem is solved
in conservative flux form using the finite volume approach, and the dispersive terms
in the model are solved using the finite difference method. The well balance topogra-
phy source term S upwinding is used in conjunction with the approximate first and
second order Upwind scheme for the advection fluxes F and G in the scheme to carry
out the time integration. Accuracy and the existing numerical technique are examined
using convergence analysis for the rectangular domain. The effect of geometry, con-
sidering amplification, is investigated using the existing numerical method on harbors
with TT shapes. The model is used to build and remodel an artificial port in a coastal
location with varying bathymetry, taking into account the effects of partial reflection,
dissipation, and convective non-linearities. The numerical scheme is implemented on
the realistic Paradip port, Odisha, India.

6 Appendix
7 Fluxes Evaluation
r
The evaluation functions A(li+ 1 ,j , q̂i+ 1
l
) and G(q̂i+ 1
r
, q̂i+ 1 ) are considering in the
2 2 ,j 2 ,j 2 ,j
following subsection.

7.1 Splitting of vectors fluxes

To talk about the 2-D example of flux vector splitting, one can check to see if the
fluxes satisfy the following homogeneity property:

A(q̂)q̂ = F(q̂), B(q̂)q̂ = G(q̂), (52)

where F(q̂) and B(q̂) are given in Eq.(8). A(q̂) and B(q̂) are the jacobian of F(q̂) and
G(q̂) respectively and are given as
   
0 1 0 0 0 1
q̂22 gq2 2q̂2  − q̂2 q̂23 q̂3 q̂2 
A(q̂) = − + 0, B(q̂) = (53)

q̂12 q̂1 q̂1  2 q̂1 q̂1 q̂1 
q̂
− q̂q̂3 q̂22 q̂3 q̂2
q̂1 q̂1 − q̂32 + gq21 0 2q̂
q̂1
3
1 1

28
The decomposition of the jacobian A(q̂) and B(q̂) are then

A∗ = RΛ∗ R−1 , B ∗ = SΣ∗ S −1 , (54)

The diagonal matrix Λ∗ has diagonal components that are the eigenvalues of A∗ , the
matrix whose columns are the eigenvalues of R, and R−1 is the inverse of R. B ∗ ’s
eigenvalues are the diagonal elements of the diagonal matrix Σ∗ . Similarly, S is the
matrix whose columns are B ∗ ’s eigenvalues, and S −1 is S’s inverse. These are offered as
 ∗   
λ1 0 0 1 0 1
Λ∗ (q̂) =  0 λ∗2 0  , R(q̂) = λ∗1 0 λ∗3  (55)
q̂3 q̂3
0 0 λ∗3 q̂1 1 q̂1
where
q̂2 p q̂2 q̂2 p
λ∗1 (q̂) = − g q̂1 , λ∗2 (q̂) = , λ∗3 (q̂) = + g q̂1 (56)
q̂1 q̂1 q̂1
and
σ1∗ 0 0
   
1 0 1
Σ∗ (q̂) =  0 σ2∗ 0  , S(q̂) =  q̂q̂21 1 q̂q̂21  (57)
0 0 σ3∗ σ1∗ 0 σ3∗
and
q̂3 p q̂3 q̂3 p
σ1∗ (q̂) = − g q̂1 , σ2∗ (q̂) = , σ3∗ (q̂) = + g q̂1 (58)
q̂1 q̂1 q̂1
It is important to note that the wave speeds λ∗1 , λ∗3 , σ1∗ , and σ3∗ are not separate from
the eigenvalues of matrices A and B and are not physically correct. Consequently, the
matrices A∗ and B ∗ must be divided into their positive and negative portions as

A∗+ = RΛ∗+ R−1 , A∗− = RΛ∗− R−1 , ∗

B+ = SΣ∗+ S −1 , ∗
B− = SΣ∗− S −1 (59)

where

max(λ∗1 , 0) min(λ∗1 , 0)
   
0 0 0 0
Λ∗+ =  0 max(λ∗2 , 0) 0 , Λ∗− =  0 min(λ∗2 , 0) 0 
∗ ∗
0 0 max(λ3 , 0) 0 0 min(λ3 , 0)
(60)
and

max(σ1∗ , 0) min(σ1∗ , 0)
   
0 0 0 0
Σ∗+ =  0 max(σ2∗ , 0) 0 , Σ∗− =  0 min(σ2∗ , 0) 0 
0 0 max(σ3∗ , 0) 0 0 min(σ3∗ , 0)
(61)
then the following conditions must hold

Λ∗ = Λ∗+ + Λ∗− , A∗ = A∗+ + A∗− Σ∗ = Σ∗+ + Σ∗ −, B ∗ = B+

∗ ∗
+ B− (62)

29
And finally we have the following flux evaluation functions
l r ∗ l l ∗ r r
FF V S (q̂i+ 1
,j , q̂i+ 1 ,j ) = A+ (q̂i+ 1 ,j )q̂i+ 1 ,j + A− (q̂i+ 1 ,j )q̂i+ 1 ,j (63)
2 2 2 2 2 2

l r ∗ l l ∗ r r
GF V S (q̂i,j+ 1 , q̂
i,j+ 1 ) = B+ (q̂i,j+ 1 )q̂i,j+ 1 + B− (q̂i,j+ 1 )q̂i,j+ 1 (64)
2 2 2 2 2 2

7.2 Flux difference splitting

Flux difference splitting scheme is similar to the flux evaluation scheme except the
following flux differences of the flux vectors of the form

1  1  
F ∗ (q̂i+
l
1
r
,j , q̂i+ 1 ,j ) =
l
F(q̂i+ 1 , q̂ r
1 ) − dF (q̂ l r
i+ 2 ,j i+ 2 ,j ,
1 )q̂ 1 (65)
2 2 2 2 ,j i+ 2 ,j 2

1  1  
G ∗ (q̂i+
l
1
r
,j , q̂i+ 1 ,j ) =
l
G(q̂i+ 1 , q̂ r
1 ) − d G (q̂ l r
i+ 2 ,j i+ 2 ,j ,
1 )q̂ 1 (66)
2 2 2 2 ,j i+ 2 ,j 2
where
   
l r l r r r
dF (q̂i+ 1 )q̂ 1 = Q F (q̂ 1 ), q̂ 1 q̂ 1 ) − q̂ i+ 2 ,j ,
1
2 ,j i+ 2 ,j i+ 2 ,j i+ 2 ,j i+ 2 ,j
    (67)
l r l r r r
dG (q̂i+ 1 )q̂
,j i+ ,j 1 = Q (q̂
G i+ ,j 1 ), q̂ 1
i+ ,j q̂ 1
i+ ,j ) − q̂ i+ ,j ,
1
2 2 2 2 2 2

7.3 Roe’s Scheme

In 2-D case the Roe’s scheme can be used by considering

0 1 0
 
2 2
l
QF ,Roe (q̂i+ 1
r
, q̂i+ 1
∗
) = A(q̂i+ 1 ) = −ui+ 12 ,j + ci+ 12 ,j 2ui+ 12 ,j 0 , (68)
2 ,j 2 ,j 2 ,j
−ui+ 21 ,j vi+ 12 ,j vi+ 12 ,j ui+ 21 ,j

where

hli+ 1 ,j hri+ 1 ,j
     
2 2
h̃i+ 21 ,j
l  l l  r  r r  ∗
q̂i+ = hi+ 12 ,j ui+ 12 ,j  , q̂i+ = hi+ 12 ,j ui+ 12 ,j  , q̂i+ = h̃i+ 12 ,j ũi+ 12 ,j 
 
1 1 1
,j
2 2 ,j 2 ,j
hli+ 1 ,j vi+
l
1 hri+ 1 ,j vi+
r
1 h̃i+ 12 ,j ṽi+ 12 ,j
2 2 ,j 2 2 ,j
(69)
and
q q q q
hli + 21 , juli + 12 , j + hri + 12 , juri + 12 , j hli + 12 , juli + 12 , j + hri + 21 , juri + 12 , j
ũi+ 12 ,j := q q , ṽi+ 12 ,j := q q
hli + 12 , j + hri + 21 , j hli + 12 , j + hri + 12 , j
(70)

30
we can also chose

0 1 0
 
l
QG,Roe (q̂i+ 1
r
, q̂i+ 1
∗
) = B(q̂i+ 1 ) =  −ui,j+ 21 vi,j+ 12 vi,j+ 21 ui,j+ 12  , (71)
2 ,j 2 ,j 2 ,j
−ui,j+ 12 + c2i,j+ 1 0 2vi,j+ 21
2

where

hli+ 1 ,j hri+ 1 ,j
     
2 2
h̃ i+ 1
2 ,j
l  l l  r  r r  ∗
q̂i+ = hi+ 12 ,j ui+ 12 ,j  , q̂i+ = hi+ 12 ,j ui+ 21 ,j  , q̂i+ = h̃i+ 21 ,j ũi+ 21 ,j 
 
1 1 1
2,j 2 ,j 2 ,j
l l r r
hi+ 1 ,j vi+ 1 ,j hi+ 1 ,j vi+ 1 ,j h̃i+ 12 ,j ṽi+ 12 ,j
2 2 2 2
(72)
and q q
hli + 12 , juli + 12 , j + hri + 21 , juri + 12 , j
ũi+ 12 ,j := q q , (73)
hli + 21 , j + hri + 12 , j
we can also discretize the source term involved in the model as
Z x 1Z y 1
1 i+
2
j+
2 g(bi+1,j − bi−1,j )  r 
Si,j2 = −gh(x, y, t)bx (x, y)dxdy = − hi− 1 ,j + hli+ 1 ,j ,
∆x∆y x 1 y 1 2∆x 2 2
i− j−
Z x 21 Z y 21
1 i+
2
j+
2 g(bi,,j+1 − bi,j−1 )  r 
Si,j3 = −gh(x, y, t)by (x, y)dxdy = − hi,j− 1 + hli,j+ 1
∆x∆y x 1 y 1 2∆y 2 2
i− j−
2 2
(74)

31