366 © IWA Publishing 2012 Journal of Hydroinformatics | 14.

2 | 2012

Numerical simulation of flood waves and calculation of

exerted forces on the cylindrical piers in contraction
channels with different cross section profiles
Parviz Ghadimi and Arsham Reisinezhad

ABSTRACT
Parviz Ghadimi (corresponding author)
A numerical model based on two-dimensional shallow water equations is presented. The depth-
Arsham Reisinezhad
averaged velocity components with free-surface elevation have been used as independent Department of Marine Technology,
Amirkabir University of Technology,
variables in the model. The finite element technique is applied to discretize the spatial derivatives. Tehran,
Iran
Triangular elements with quadratic and linear interpolating functions are employed for two E-mail: pghadimi@aut.ac.ir
horizontal velocity components and the free-surface elevation, respectively. The standard Galerkin
method is applied for discretization of the governing equations. Time discretization is performed
using an implicit scheme. The resulting linear system of equations is solved by the GMRES method.
The model is validated using three test cases and the results are compared with an analytical
solution, the result of numerical work and experimental data, respectively. Favorable agreement
was achieved in all three cases. Subsequently, the developed model is applied to simulate
free-surface elevation through a channel contraction. The effects of width of the narrow section as
well as the profile of the cross section of the channel on the wave forces exerted on a circular
cylinder were studied. This was done in a channel with a quartic narrow section. Plots of time
histories of the drag coefficient on the cylinder were produced, demonstrating the effects of the
mentioned parameters.
Key words | cylinder, different cross sections, finite element method, narrow section, shallow water
equation

NOTATION

ρ density ν(x, y, t) average flow velocities along the y direction

~ y; zÞ
Uðx; velocity vector Sfx hydraulic resistance along the x direction
P pressure Sfy hydraulic resistance along the y direction
P0 atmospheric pressure m Manning roughness coefficient
g gravity t time
μ viscosity θ relaxation coefficient
Zs(x, y, t) water surface elevation ~
n normal vector
Zb(x, y) bathymetry profile τ shear stress
Zse(x, y, t) distance between water surface and the center of Δx, Δy mesh size along x and y directions
square elements Δt time step
h(x, y, t) water depth a radius of circular cylinder
H(x, y) still water depth α angle
u(x, y, t) average flow velocities along the x direction Q flux
doi: 10.2166/hydro.2011.062
 367 P. Ghadimi & A. Reisinezhad | Flood waves in contraction channels with different cross section profiles
W channel width
Wg narrow section width
INTRODUCTION
Prediction of the exerted maximum forces on bridge piers
due to floods is one of the most important parameters in
designing bridges. The engineers’ knowledge about this par-
ticular design parameter helps them to reach an optimum
design for the shape of the bridge piers. It also assists
them in selecting the most suitable material for manufactur-
ing the piers in order to obtain the required strength. One of
the other effective parameters in bridge design is the maxi-
mum of the water free-surface height, which hints at the
suitable height of the bridge deck from the still water line.
On the other hand, selection of suitable locations for con-
struction of the bridges is another important consideration
in bridge design. Ordinarily, in view of all the important
limitations for the design and manufacturing of the bridges,
one can assert that a suitable location is where the river is at
its minimum possible width. Although selection of this
location could cause a reduction in the material and cost
of construction of the bridges, careful consideration must
be applied to the reduction in the width of the river which
can increase the water free-surface height in front of the
piers as well as the imposed forces due to the flood wave
propagation. On the other hand, if the width of the river
decreases abruptly over a short length, its shape resembles
that of a contraction channel. In these types of channels,
the influences of the water free-surface height and the
imposed forces on the piers due to the flood wave propa-
gation are considerably different from the rivers whose
width varies over a larger distance.
Because of all the above reasons, in this paper, flood wave
propagation is simulated through a contraction channel with
different cross-sectional profiles depicting the upstream and
downstream of a narrow section in a river. Accordingly, the
behavior of the flood wave is analyzed and the geometrical
effects of these types of channels on the water elevation in
front of the cylindrical piers and the imposed forces on the
piers are examined. Geometries considered as cross-sectional
profiles include rectangular, parabolic and fourth degree
Journal of Hydroinformatics | 14.2 | 2012
polynomials. In order to model the flood wave propagation,
the fluid is considered incompressible and the shallow
water equations (Saint-Venant equations) have been solved
numerically by the finite element method.
Several numerical techniques are available to solve the
two-dimensional shallow water equations for the simulation
of free-surface flows. These include finite volume methods
(FVM) (Zhao et al. ; Valiani et al. ), finite difference
methods (FDM) (Garcia & Kahawitha ; Fennema &
Chaudhry , ; Bellos et al. ). Mynett () has
used the DELFT-FLS SYSTEM to simulate flood in the
polders of Tiel and Culemborg in The Netherlands. The
DELFT-FLS SYSTEM is a two-dimensional hydrodynamic
simulation package especially suited for simulation of the
dynamic behavior of overland flow over initially dry land.
It is based on the full two-dimensional shallow water
equations. These equations are solved following the finite
difference numerical technique on a rectangular staggered
grid. Abida et al. () proposed a numerical method to
model hydraulic flood routing in Medjerda River (Northern
Tunisia). The model is based on 1D Saint-Venant equations
using a four-point implicit finite difference scheme. Tseng
() presented a finite difference implicit MacCormack
scheme for the computation of flood wave propagation in
overland and open channel flows with steep topography.
However, a limited number of models is available in the lit-
erature for simulating a free surface using the finite element
method (FEM). Dhatt et al. () proposed different tri-
angular elements for studying depth-averaged steady and
unsteady flows. The standard Galerkin model was used to
discretize the equations. The algebraic nonlinear system is
solved with an implicite Euler scheme coupled with the
standard Newton–Raphson method. The model is applied
to the velocity and water level in the St. Lawrence River
near the Gentilly area. Aizinger & Dawson () developed
a discontinuous Galerkin finite element method to solve the
shallow water equation. The method was applied to simulate
the water free surface in Galveston Bay and river inflow in
the Gulf of Mexico. Eldho & Sharique () presented a
vertically averaged 2D finite element model for shallow
water flow in coastal areas. The Galerkin method with a
finite difference scheme was used in the spatial and time
domain, respectively. The model was applied to simulate
the hydrodynamic behavior of Bombay harbor. Venutelli
 368 P. Ghadimi & A. Reisinezhad | Flood waves in contraction channels with different cross section profiles Journal of Hydroinformatics | 14.2 | 2012

() presented a Taylor–Galerkin third-order method for 1 @P

 g¼0 ð4Þ
the simulation of free-surface flow through a channel con- ρ @z
striction. The method of characteristics and the Petrov–
Galerkin finite element for modeling of shallow water Integration of Equation (4) and setting the atmospheric
flows have been used by Katopodes & Strelkoff (, pressure P0 at the water surface as the integration constant
) and Katopodes (), respectively. yields
As mentioned earlier, study of the flood waves over cir-
cular cylinders is relevant to the design of the bridge piers P ¼ ρgðZs  ZÞ þ P0 ð5Þ
located in the narrow sections. In this study, the simulated
propagation of a flood wave through a channel contraction where Zs is the water surface elevation (see Figure 1). From
is studied numerically. Accordingly, the effect of a flood Equation (5), assuming that P0 is constant everywhere, we
wave on the forces exerted on a circular cylinder located obtain
in a channel with different cross sections is examined. The
shallow water equations are solved using a finite element @P @Zs
¼ ρg ð6Þ
method to simulate the free-surface elevation in the channel @x @x
and around the cylinder. In order to test the validity of the
numerical model, both one- and two-dimensional dam @P @Zs
¼ ρg ð7Þ
breaks are first simulated and then compared with analytical @y @y
and other numerical findings. Subsequently, the solitary
wave scattering by an isolated circular cylinder is simulated Equations (1)–(3) can be integrated over the water depth (h)
and the result is compared with the experimental data. of a generic cross section. Taking Equations (6) and (7) into
Finally, the model was used for investigation of the effect account with ignoring possible lateral inflows and outflows,
of the narrow section shapes on the free-surface elevation the equations become
and the influences of different cross sections on the wave
forces exerted on the cylinder. @h

!
þ~
u  gradðhÞ þ h  divð~
uÞ ¼ 0 ð8Þ
@t

NUMERICAL METHOD @u

! @Zs

þ~
u  gradðuÞ ¼ g  gSfx ð9Þ
@t @x

Governing equations
@v

! @Zs
þ~
u  gradðvÞ ¼ g  gS fy ð10Þ
The mathematical description of the motion of a viscous @t @y

incompressible fluid, in a nearly horizontal flow, is given

by the Navier–Stokes equations (Hervouet ) as

@U @V @W
þ þ ¼0 ð1Þ
@x @y @z

@U ~ 1 @P 1

!
þ U  divðUÞ ¼  þ div(μgradðUÞ) ð2Þ
@t ρ @x ρ

@V ~ 1 @P 1

!
þ U  divðVÞ ¼  þ div(μgradðVÞ) ð3Þ
@t ρ @y ρ Figure 1 | Calculation domain.
 369 P. Ghadimi & A. Reisinezhad | Flood waves in contraction channels with different cross section profiles Journal of Hydroinformatics | 14.2 | 2012

By applying Manning’s formula for their evaluation, we used simple triangular elements with a linear approximation
have for velocity and water depth. In this early finite element
work, numerical oscillations were observed in water
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
um2 u2 þ v2 height values. This was especially the case when linear tri-
S fx ¼ ð11Þ
h4=3 angular elements were used for studying closed estuary
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi and steady large river flows. Different methods have been
vm2 u2 þ v2 suggested for removing the spurious oscillations generated
Sfy ¼ ð12Þ
h4=3
by the standard Galerkin finite element discretizations.
One of the popular stabilization techniques is the Streamline
Equations (8)–(10) are called shallow water equations
Upwinding/Petrov–Galerkin (SUPG) method when the
(Saint-Venant equations).
linear triangular elements are used for spatial discretization
(Brooks & Hughes ). In other stabilization techniques,
Finite element formulation the use of mixed interpolation is suggested for velocity and
water height in which the water height approximation
The flexible grid system used in the finite element method should be chosen one degree lower than the velocity
enables us to represent any complex topography and bound- approximation. Works done by Hood & Taylor () and
ary conditions, giving the most accurate solution. In this Walters & Cheng () are important among others that
paper, for the purpose of spatial discretization, the standard report non-oscillatory results. Hence, to produce a smooth
Galerkin finite element method is applied. The spatial solution, mixed interpolation (quadratic for velocity com-
domain Ω with boundary Г is discretized by subdivision ponents and linear for free-surface elevation) is adopted in
of the continuum into non-overlapping elements and defin- the present study. Here, for the time-dependent terms, a
ing a number of discrete nodes for each element. The finite difference scheme is used as follows:
finite element approximate solutions of the dependent
variables are
@f f nþ1  f n
¼ ð14Þ
@t Δt
X
N1
uðx; yÞ ¼ uj Nju ðx; yÞ
j¼1
where f is any unknown variable u, v and h. For
X
N1
vðx; yÞ ¼ vj Nju ðx; yÞ ð13Þ other terms, a semi-implicit scheme is applied for the time
j¼1 discretization
X
N2
hðx; yÞ ¼ hj Njh ðx; yÞ
j¼1 f ¼ θf nþ1 þ ð1  θÞf n ð15Þ

where uj, vj and hj are the values of the dependent variables

where θ is called the ‘relaxation coefficient’ and is always
at the nodal points, Nuj and Nhj are the standard basis func-
greater than 0.5. In the present study, for a semi-implicit
tions and N1 and N2 are the total number of unknowns.
scheme, the value of θ is taken to be 0.55 (Hervouet ).
In the shallow water flows, if the viscous effects (e.g. bed
This semi-implicit scheme is accompanied by the nonlinear
friction, wind drag) are small, the problem becomes convec-
term h  divð~
uÞ which brings out a product of the unknowns
tion-dominated. The convection-dominated problems are
h n+1 and ~
unþ1 which cannot be solved simultaneously. In
those associated with the high Peclet numbers as the ratio
order to overcome this difficulty, the following formula is
of convection to diffusion (Brooks & Hughes ). In
adopted:
such a case, one may experience the formation of spurious
modes (oscillations or wiggles) using the standard Galerkin
finite element method. For example, Connor & Wang () uÞ ¼ h prop  divðθ~
h  divð~ unþ1 þ ð1  θÞ~
un Þ ð16Þ
 370 P. Ghadimi & A. Reisinezhad | Flood waves in contraction channels with different cross section profiles Journal of Hydroinformatics | 14.2 | 2012

where hprop is called the ‘height of propagation’. Similarly, Nui , and integration over the domain. As a result of these


!
for advection terms, we write ~u  gradð fÞ as operations, the continuity equations become

!

! 3 hnþ1  hn ZZ

X X
3
~
u  gradð fÞ ¼ ~
uconv  gradðθf nþ1 þ ð1  θÞf n Þ ð17Þ j j
Njh Nih dΩðeÞ þ ðθhnþ1 þ ð1  θÞhnj Þ
j¼1
Δt j¼1
j

where ~
uconv is the advection field. In these equations, hprop ZZ " @Njh h @Njh h
#

and ~
uconv are determined by a sub-iterative method. In × uconv N þ vconv N dΩe
@x i @y i
other words, hprop and ~ uconv are considered equal to h n ZZ
X 6 @Nju h
and ~
un in the first sub-iteration step for every node. How- þ ðθunþ1 þ ð1  θÞu n
Þ hprop N dΩe
j¼1
j j
@x i
ever, in the subsequent sub-iteration steps, these terms are
X 6 ZZ @Nju h
determined for every node as follows: þ ðθvnþ1 þ ð1  θÞv n
Þ hprop N dΩe ¼ 0 ð23Þ
j¼1
j j
@y i

h prop ¼ θh0 nþ1 þ ð1  θÞhn ð18Þ

The momentum equation in the x direction is
~ u0 nþ1 þ ð1  θÞ~
uconv ¼ θ~ un ð19Þ
6 unþ1  un ZZ
X X
6
j j
Nju Niu dΩe þ ðθunþ1 þ ð1  θÞunj Þ
where h0 nþ1 and ~
u0 nþ1 are the current estimations of h nþ1 j¼1
Δt j¼1
j

ZZ  @Nju @Nju

and ~
unþ1 resulting from the previous sub-iterations. Water
× uconv Niu þ vconv Niu dΩe
surface elevation depends on bathymetry profile and water @x @y
X
3 ZZ @N h
depth (Zs ¼ h þ Zb). Accordingly, by applying Equation j
¼ gθðhnþ1  hnj Þ Niu dΩe
(15), the expansion of water surface elevation gradient is j¼1
j
@x
as follows: X
3 ZZ @N h
j
 gZns N u dΩe
j¼1
@x i


!

! qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ggradðZs Þ ¼ ggradðh þ Zb Þ
X
6 gn2 ðunj Þ2 þ ðvnj Þ2 ZZ


!  unþ1 Nju dΩe ð24Þ
¼ ggrad[θhnþ1 þ ð1  θÞhn þ Zb ] ð20Þ j
j¼1 ðhnj Þ4=3

or
The momentum equation in the y direction is

!

!

!

 ggradðZs Þ ¼ gθgradðhnþ1  hn Þ  ggradðZns Þ ð21Þ 6 vnþ1  vn ZZ
X X
6
j j
Nju Niu dΩe þ ðθvnþ1 þ ð1  θÞvnj Þ
j¼1
Δt j¼1
j

Friction terms in Equations (11) and (12) are also non-

ZZ  @Nju u @Nju u

linear, and if we want to avoid undesirable numerical effects × uconv Ni þ vconv Ni dΩe
like totally artificial inversion of velocities under the effect @x @y
X3 ZZ @Nj u
h X 3
of friction, they must be treated in an implicit way. In order
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ¼ gθðhnþ1  hnj Þ Ni dΩe  gZns
to linearize the ~
u u2 þ v2 term, it is discretized as follows: j¼1
j
@y j¼1
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi ZZ @N h X6 gn2 ðunj Þ2 þ ðvnj Þ2 ZZ
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi j
~
u u2 þ v2 ≈ ~ unþ1 ðun Þ2 þ ðvn Þ2 ð22Þ × Niu dΩe  vnþ1 Nju dΩe
@y j¼1
j
ðhnj Þ4=3
ð25Þ
The governing equations are reconfigured into a weak
form by multiplying Equation (8) by the weighting function By a finite element approximation of the dependent vari-
Nhi , and Equations (9) and (10) by the weighting functions ables, the following system of differential/algebraic
 371 P. Ghadimi & A. Reisinezhad | Flood waves in contraction channels with different cross section profiles
equations for the nodal unknown is obtained:
2 38 9 8 9
½A1 ½B1 ½B2 < ½δh = < ½D1 =
4 ½C1 ½A2 ½0 5 ½u ¼ ½D2
: ; : ;
½C2 ½0 ½A3 ½v ½D3
A description of the entries of the matrices and column
vectors in Equation (26) is presented in the appendix.
The integrals in Equations (23)–(25) can be calculated by
natural coordinate. By superimposing the element matrices
for all elements in the proper order, the global matrix for
the entire discretized domain can be obtained in the form
½AfXg ¼ fRg
where [A] is the coefficient matrix. {X} is a column vector con-
sisting of δh, u and v at time instant t þ Δt whereas the column
vector {R} is computed using the values of h, u, v and Zs at time
instant t. The system definition is not complete until the
boundary conditions are implemented in the global matrix
equations. Three types of boundaries are utilized in this
u ~
study; a solid wall boundary ð~ n ¼ 0Þ where n is the
normal to the wall at that point, a certain discharge for the
upstream boundary condition and uniform flow boundary
for the downstream boundary condition. In the present
study, the system of linear equation is solved using the gener-
alized minimal residual (GMRES) method (Saad & Schultz
).
Program steps for the present model
The program takes all the mentioned inputs, allocates the
memories for the desired pointer, calculates the area of
every element and assigns the initial conditions to the vari-
ables u, v and h. Numerical results are improved by using
an iterative method. Accordingly, hprop and ~ uconv are
considered equal to h n and ~
un , respectively, for the first
sub-iteration step. However, for the subsequent sub-iteration
steps, these variables are computed by Equations (18) and
(19). Using these variables and the natural coordinate
system, element matrices are determined and finally
assembled. Boundary conditions are implemented and the
system of equations is solved. This produces the values of
variables at time t þ Δt which are taken as current
Journal of Hydroinformatics | 14.2 | 2012
estimations for the next sub-iteration. The process is
repeated until maximum differences between successive cal-
culated values become less than the maximum permissible
ð26Þ difference. In the present model, the maximum permissible
difference is taken as 0.001. At this stage, the calculated
values of the indicated variables are used for the next time
increment. Figure 2 shows the flow chart of the model devel-
oped in the present study.
Determination of the drag forces
The flood wave force on the circular cylinder can be
obtained by integration of the water pressure over the
ð27Þ wetted body surface as follows:
Z
Fx ¼ Pnx dS ð28Þ
Z
Fy ¼ Pny dS ð29Þ
where P is the pressure, nx and ny are the components of the
unit vector in the normal direction of the boundary in the x
and y directions, respectively, and S represents the wetted
body surface.
The two-dimensional Saint-Venant equations are only
able to specify the height and the velocity components on
the water surface. Then, the wave pressure distribution on
the surface over water depth can be determined by the Ber-
noulli equation (Equation (30)) as in
1
P ¼ ρgZse þ ρgðu2 þ ν 2 Þ ð30Þ
2
In order to calculate the imposed pressure force on
the circular cylinder, the velocity distribution in the x
and y directions and the water free-surface height are
obtained. Firstly, the wetted body surface of the circular
cylinder is divided by the square elements. The value of
the hydrostatic pressure (i.e. the first term of Equation
(30)) is computed at the center of elements by employing
the water free-surface distribution around the circular
cylinder. Also, considering the fact that the velocity com-
ponents are constant over the fluid depth, the value of the
 372 P. Ghadimi & A. Reisinezhad | Flood waves in contraction channels with different cross section profiles Journal of Hydroinformatics | 14.2 | 2012

Figure 2 | Computational flow chart for the present model.

hydrodynamic pressure (i.e. the second term of Equation
(30)) is computed at the center of elements. By multiply-
ing areas of the elements by their corresponding
pressure, the imposed pressure force on every element is
calculated. Lastly, the pressure force components of
elements in the x and y directions are added in order to
calculate the total pressure force in the two directions.
VALIDATION OF THE NUMERICAL SCHEME
To demonstrate the accurate modeling ability of the current
numerical scheme, different cases are considered. First, a 1D
dam break case in a horizontal channel, for which an
analytical solution is known, is presented. Further, a 2D
dam break test case is simulated and the result is compared
 373 P. Ghadimi & A. Reisinezhad | Flood waves in contraction channels with different cross section profiles Journal of Hydroinformatics | 14.2 | 2012

against an available numerical result. Finally, the solitary

wave scattering by a circular cylinder is computed and the
result is compared with existing experimental data.

1D dam break problem

The one-dimensional dam break problem is a common test

for evaluating shock capturing schemes in shallow flows.
Here, a dam is located at the midpoint of a channel whose
length and width are 50 m and 5 m, respectively. The
water depth on the left and right sides of the dam is 0.5
and 0.25 m, respectively (Mohammadian et al. ). The
computational domain was initially divided into 40 triangu-
lar elements with Δx ¼ 2.5 m, Δy ¼ 5.0 m for a coarse mesh
test with the time step being Δt ¼ 0.25 s. However, for the
subsequent tests, 8,000 triangular elements with Δx ¼ Δy ¼
0.25 m were considered for a finer mesh with the time step
being Δt ¼ 0.025 s. The dam is instantaneously removed
across its entire width and the flow conditions are computed
up to time t ¼ 8 s. The results for depth (h) and discharge
(uh) are presented in Figures 3(a) and (b), respectively. As
seen in the figures, the numerical scheme is able to provide
a favorable result relative to the exact solution, without any Figure 3 | Computed values of (a) depth and (b) discharge at time t ¼ 8 s.

numerical oscillations.
In order to determine the effects of the mesh size on the Table 1 | Test specifications to show the influence of mesh size on the accuracy of com-
accuracy of the computational results, five different trials are putational results

performed by the mentioned specifications in Table 1. In

Trial Number of Number of CPU
every trial, the values of depth and discharge are computed number Δx Δy elements nodes Δt time (s)

at five different sections at t ¼ 8.0 s and are compared Trial 1 5.0 2.5 40 105 0.25 6
against the exact solutions (Tables 2 and 3). Sections 1–4 Trial 2 1.0 1.0 500 1,111 0.1 384
are located at distances 10, 20, 30 and 40 m from the Trial 3 0.5 0.5 2,000 4,221 0.05 952
entrance of the computational domain, respectively. Trial 4 0.25 0.25 8,000 16,441 0.025 4,363
Sections 2 and 3 are located at a region where the Trial 5 0.125 0.125 32,000 64,881 0.0125 12,481
unknown variables are approximately nonvarying or con-
stant. Therefore, it can be concluded that any reduction in
the mesh size between Sections 2 and 3 does not cause a sig- conclude that decreasing the mesh size increases the CPU
nificant change in the computed results. On the contrary, as time considerably, but improves the results impressively.
Section 4 is located near the shock region, the accuracy of
the computed results strongly depends on the mesh size. 2D dam break problem
In Section 1, even though the computed result for the
water depth (Table 2) is improved by reduction of the The two-dimensional dam break problem has been used pre-
mesh size, due to the small change in velocity components viously as a benchmark test by Fennema & Chaudhry (,
in this region, an overestimation of Q is observed. Based ). The problem models a partial dam break or rapid
on observation of the indicated results, one can easily opening of a sluice gate. The computational basin has
 374 P. Ghadimi & A. Reisinezhad | Flood waves in contraction channels with different cross section profiles Journal of Hydroinformatics | 14.2 | 2012

Table 2 | Comparison of the computed depth and the exact solution for the 1D dam break
problem

Section 1 2 3 4

Exact 0.4530 0.3662 0.3662 0.3662

Trial 1 0.4488 0.3630 0.3633 0.3246
Trial 2 0.4500 0.3638 0.3637 0.3465
Trial 3 0.4503 0.3641 0.3640 0.3616
Trial 4 0.4510 0.3648 0.3644 0.3668
Trial 5 0.4522 0.3659 0.3653 0.3646

Table 3 | Comparison of the computed discharge and the exact solution for the 1D dam
break problem

Section 1 2 3 4

Exact 0.0974 0.2376 0.2376 0.2376

Figure 5 | 3D view of two-dimensional dam break at t ¼ 7.2 s.
Trial 1 0.0863 0.2331 0.2333 0.1596
Trial 2 0.0990 0.2332 0.2338 0.1980
Trial 3 0.1035 0.2342 0.2346 0.2306
Four line plots of depth (h) at the final time station

Trial 4 0.1059 0.2359 0.2369 0.2344

t ¼ 7.2 s are also presented for assessment at the locations

Trial 5 0.1061 0.2377 0.2379 0.2357

dimensions of 200 m× 200 m with the reservoir headwater
level Hreservoir ¼ 10.0 m and the tail water level Htail ¼ 5.0 m.
A schematic representation of the geometry is shown in
Figure 4. When the dam wall breaks, water is released
through a 75 m wide breach, and a leading bore and
abrupt depression are formed in the basin traveling in the
opposite directions. In the computations, an asymmetric
mesh with 3,100 triangular elements and 6,411 nodes was
used in the computational domain. Figure 5 shows a 3D
view of the water surface elevation at the time t ¼ 7.2 s.
Figure 4 | Partial dam break: computational domain.
y ¼ 130 m, x ¼ 80 m, x ¼ 100 m and x ¼ 120 m (see
Figure 5). Figure 6 shows a comparison between results
of the present model and the characteristic finite element
method produced by Chambers ().
Solitary wave scattering by a circular cylinder
The solitary wave scattering by a circular cylinder is com-
puted to further validate the numerical model. The
computed case in the present study is identical to that
studied experimentally by Yates & Wang (). The exper-
iments were conducted in a wave tank which was 7.6 m in
length and 0.76 m in width. The undisturbed water depth
was H ¼ 4 cm. The cylinder with a radius a ¼ 6.35 cm
was placed at the center of the symmetric line of the chan-
nel. Yates & Wang () measured the time histories of
the wave force and wave surface elevation near the cylin-
der in their experiments. In the present study, a
rectangular computational domain with a length of 3.0 m
(this length has been chosen shorter as at this length
there is no longer any influence of the wave on the compu-
tational results) and a width of 0.76 m is used. The cylinder
of radius a ¼ 6.35 cm is located at the centerline, a distance
0.6 m from the entrance of the computational domain.
The solitary wave is generated at the left boundary of
 375 P. Ghadimi & A. Reisinezhad | Flood waves in contraction channels with different cross section profiles
Figure 6 | Line plot of depth (h) at t ¼ 7.2 s for locations (a) y ¼ 130 m, (b) x ¼ 80 m, (c) x ¼ 100 m and (d) x ¼ 120 m.
the domain. The right boundary is located far enough from
the center of the cylinder (i.e. 2.4 m) so that the wave
propagation becomes completely uniform, making the
application of the uniform flow boundary condition poss-
ible. The computational domain is discretized using six-
node triangular elements as shown in Figure 7. There are
5,508 elements with 11,266 nodes in the computational
domain. Figure 8 shows the comparison of the computed
time histories of the dimensionless surface elevation with
the experimental data along the radius line α ¼ 0 , 180 .
W W
The angle α is measured in the clockwise direction
and the α ¼ 0 line is taken along the upstream centerline
W
of the tank.
Figure 8 shows the comparison of the time histories of
free-surface elevation computed by the present model
against the available experimental data. The observed differ-
ence seen may be attributed to the lack of dispersion terms
in the Saint-Venant equations. In other words, the lack of
dispersion terms causes deformation of the solitary wave
and decreases the wave height during which the wave propa-
gates on the long path. However, even with this difficulty,
relative agreement is displayed between the computed
results and the experimental data.
Journal of Hydroinformatics | 14.2 | 2012
In Figure 9, the computed time histories of the force
on the cylinder are compared with the measured data
produced by Yates & Wang (). Considering the fact
that the solitary wave deforms as it propagates, the numeri-
cal model is seen to predict the maximum force relatively
well.
APPLICATION
One of the most important factors affecting the design of the
river structure is the geometric parameters of the domain.
For instance, the river profile could have a profound effect
on hydrodynamic parameters of the bridge piers. As such,
the designer’s familiarization with the effects of these par-
ameters on the profile of the water free surface can be of
great help in finding an optimized design. On the other
hand, in the studies of the river and actual channels, simpli-
fication of the domain geometry can be used as an approach
for the analysis of the parameters and the influential factors
on the profile of the water free surface and the way it is pro-
pagated. Accordingly, the shape of the channel cross-section
as well as the change of channel width in the longitudinal
 376 P. Ghadimi & A. Reisinezhad | Flood waves in contraction channels with different cross section profiles Journal of Hydroinformatics | 14.2 | 2012

Figure 7 | Top view of the computational domain and zoomed view around the circular cylinder.

direction for the rivers and low width channels can be

defined as two important design parameters.
In this study, a flood wave in the upstream region of the
channel is generated. At first, the effect of the narrow sec-
tion of the channel on how the propagation of this flood
wave occurs has been investigated. Then, the influence of
a narrow section opening on the exerted drag on the circular
cylinder located at the centerline of the channel has been
examined. Finally, the effect of the shape of the channel
cross section on the measure of drag forces exerted on the
mentioned circular cylinder has been studied.

Investigation of narrow section effect on propagation

of the flood wave

Numerical simulation has been performed in a frictionless

Figure 8 | Time histories of the free-surface elevation along the radius:
data; —: experimental data (Yates & Wang 1994).
channel with a length of 68 m and a width of 12 m
(Figure 10). The profiles of the symmetric narrow sections
considered are rectangular (Figure 10(a)) and quartic, i.e.
WWW : calculated
a fourth-degree polynomial (Figure 10(b)) both with length
 377 P. Ghadimi & A. Reisinezhad | Flood waves in contraction channels with different cross section profiles
Figure 9 | Time histories of the forces on the cylinder for various wave amplitudes:
○○○
8 m and width 6 m. The computational domain has also
been discretized using 1,536 triangular elements and 3,245
nodes for a channel with a rectangular narrow section and
1,632 triangular elements and 3,425 nodes for a channel
with a quartic narrow section (Figure 10). In order to
model the advancing flood wave, the initial water depth in
the channel is assumed to be 1 m and the velocity com-
ponents are taken to be zero. The boundary condition at
the upstream section (entrance region) is considered to be
a known flux Q ¼ Q(t) of the transient flood waves
(Figure 11) which could have been generated prior to the
entrance boundary resulting, perhaps, from the sudden
break of a certain column of water. Meanwhile at the down-
stream section (exit), the uniform flow condition is adopted.
For the stability of the numerical code, the relaxation factor
is taken to be 0.55 based on previous experience.
Various contour line plots of flood wave propagation in
a channel with a rectangular narrow section are presented
in Figure 12 and with a quartic narrow section in Figure 13,
both at times t ¼ 6, 9, and 12 s. It is noticeable in these
figures that, due to the asymmetric nature of the mesh,
Journal of Hydroinformatics | 14.2 | 2012
: calculated data; —: experimental data (Yates & Wang 1994).
there are subtle signs of asymmetry in the computational
results. As shown in Figures 12(b) and 13(b), because of
the presence of the narrow section, a portion of the propa-
gating wave has been reflected and the rest, by passing
through the narrow section, has propagated in different
directions. Figures 12(c) and 13(c) also indicate the effects
of channel sides as well as the reflection of the propagating
flood wave towards the central line.
For the purpose of a more careful assessment of the pro-
pagating trend of the flood wave along the channel, time
histories of the free-surface elevation at three different
points A, B and C are determined (see Figure 10) (points
A, B and C are located on the centerline at a distance 26 m,
34 m and 42 m, respectively, from the entrance boundary
of the channel). Time histories of the free-surface elevation
at point A for the channel with a rectangular narrow section
(Figure 14(a)) and with a quartic narrow section (Figure 14(b))
are demonstrated based on three different ratios of channel
width (W ) and narrow section width (Wg). Because of the
presence of the narrow sections, the graphs related to the
narrow sections (W/Wg ¼ 1.5, 2.0) do not follow the same
 378 P. Ghadimi & A. Reisinezhad | Flood waves in contraction channels with different cross section profiles
Figure 10 | Computational domain (top view): (a) complete channel and zoomed view of
the rectangular narrow section and (b) complete channel and zoomed view
of the quartic narrow section.
trend as the time history plots of the free-surface elevation of
the water for the regular channels without the narrow cross
section (W/Wg ¼ 1.0). This is primarily due to the fact that
part of the wave, when passing through the narrow section,
has been reflected and propagated in the opposite direc-
tions. The reflected wave crest causes a peak in the time
history plot of the free-surface elevation for point A and
the height of this peak increases with an increase in the
ratio W/Wg. To continue, with the passing of the reflecting
wave crest and the subsequent arrival of its hollow at
Journal of Hydroinformatics | 14.2 | 2012
Figure 11 | Entering flux to the computation domain at the upstream section.
point A, the time histories of the free-surface elevation
shows a negative value, indicating that the hollow will
further decrease with an increase in the W/Wg ratio. Also,
in contrast to the channel with a rectangular narrow section,
the maximum free-surface elevation for ratios (W/Wg ¼ 1.5,
2.0) in the channel with a quartic narrow section, compared
to that of the channel without the narrow section (W/Wg ¼
1.0), has decreased.
In Figures 15(a, b) time histories of the free-surface
elevation at point B for channels with rectangular and quar-
tic narrow sections are presented, respectively. By
examining these two figures, one can conclude that an
increase in the W/Wg ratio at the narrow section location
is equivalent to a decrease in cross-sectional area of the
channel which itself causes the free-surface elevation to
increase.
Time histories of the free-surface elevation at point C
have also been determined for a channel with a rectangular
narrow section (Figure 16(a)) and with a quartic narrow sec-
tion (Figure 16(b)). As observed in these figures, with an
increase in the W/Wg ratio, the maximum value of the
graph reduces due to the increase of reflective effects.
Also, for the channels with a narrow section, during the pas-
sing of the propagated wave from point C, the free-surface
elevation reduces. But later, due to the reflective waves
from the sides toward the mid-sections of the channel, the
free-surface elevation at that point increases again. Sub-
sequently, with the passing of the propagated and reflected
waves from the sides, the free-surface elevation approaches
zero.
 379 P. Ghadimi & A. Reisinezhad | Flood waves in contraction channels with different cross section profiles Journal of Hydroinformatics | 14.2 | 2012

Figure 12 | Different contour line plots of flood propagation in channel with rectangular narrow section: (a) t ¼ 6.0 s, (b) t ¼ 9.0 s and (c) t ¼ 12.0 s.

In order to assess the effect of the shape of the narrow
section of the channel, plots of time histories of the water
free surface at points A, B and C have been offered for
two types of channels based on a single ratio of (W/Wg ¼
2.0) (Figure 17).
From the examination of these figures, it can be con-
cluded that the shape of the narrow section has caused
a phase shift between the plots of time histories at
points A, B and C. Also, it is quite noticeable that the
reflective effects in the case of a channel with a narrow
rectangular section are more than those of a channel
with a narrow quartic section. Because of this factor,
the maximum of the free-surface elevation curve at
point B for a channel with a narrow quartic section
(Figure 17(b)) is higher than that of a channel with a
narrow rectangular section. In the meantime, the free-sur-
face elevation of the water (Figure 17(c)), due to the
effects of channel sides at point C, experiences the same
trend as point B.
Effects of the width of the opening of the narrow
section and the cross section profile of the channel
on the drag exerted on the cylinder
Although the maximum of the free-surface elevation of the
water at point B for the channel with a rectangular
narrow section is smaller than that of the channel with
a quartic section, because of economic considerations
 380 P. Ghadimi & A. Reisinezhad | Flood waves in contraction channels with different cross section profiles Journal of Hydroinformatics | 14.2 | 2012

Figure 13 | Different contour line plots of flood propagation in channel with quartic narrow section: (a) t ¼ 6.0 s, (b) t ¼ 9.0 s and (c) t ¼ 12.0 s.

and due to the geometric dissimilarity of the rivers with
the rectangular section, a channel with a quartic narrow
section is studied here. Accordingly, the effects of the
change in width of the narrow section opening as well
as the shape of the channel cross section on the drag
force exerted on a circular cylinder (depicting a bridge
pier) have been investigated.
Once again, the numerical modeling is done using a
frictionless channel with a length of 22 m and a width
of 20 m. A circular cylinder of 1 m diameter is held on
the centerline at a distance 11 m from the entrance
region. As mentioned earlier, propagation of the flood
wave as well as the imposed forces on the circular cylin-
der are strongly affected by the changes in the width of
the river or channel. Bearing in mind this fact, and in
order to get closer to the actual geometrical conditions,
the symmetric narrow section considered here is at a
random distance of 5–17 m from the entrance region of
the channel. In order to increase the accuracy of the
results, a combined meshing technique was implemented
for discretization of the computational domain using
4,400 triangular elements with 9,008 nodes (Figure 18).
For numerical modeling of the advancing flood wave,
the initial water depth of the channel is considered to
be 1 m and the velocity components are assumed to be
zero. The boundary condition at the upstream entrance
region is assumed to be a known flux Q ¼ Q(t) (Figure 11),
while at the downstream exit region, a uniform flow
 381 P. Ghadimi & A. Reisinezhad | Flood waves in contraction channels with different cross section profiles Journal of Hydroinformatics | 14.2 | 2012

Figure 14 | Time histories of free-surface elevation at point A at different W/Wg ratios: (a) Figure 16 | Time histories of free-surface elevation at point C at different W/Wg ratios: (a)
channel with rectangular narrow section and (b) channel with quartic narrow channel with rectangular narrow section and (b) channel with quartic narrow
section. section.

condition is applied. For stability of the numerical code, a

relaxation factor of 0.55 was used.
Contour line plots showing how the flood wave is
propagated around a circular cylinder in a channel
with a rectangular cross section and a quartic narrow
section are shown in Figure 19 at times t ¼ 2, 3 and 4 s.
In this figure, the effect of the presence of a circular
cylinder on the free-surface profile of the water is
demonstrated.
In Figure 20, time history plots of the drag force exerted
on the circular cylinder are shown for three different values
of the Wg/W ratio. It is evident from this plot that a change
of the opening width of the quartic narrow section does not
have any considerable influence on the maximum of the
drag force exerted on the cylinder.
Time history plots of the drag force exerted on a circu-
lar cylinder located in a channel with a quartic narrow
section and different cross section profiles (rectangular,
parabolic and fourth-degree polynomial) are illustrated in
Figure 15 | Time histories of free-surface elevation at point B at different W/Wg ratios: (a)
Figure 21. Based on the trends observed in these figures,
channel with rectangular narrow section and (b) channel with quartic narrow
section. we can conclude that the shape of the channel cross section
 382 P. Ghadimi & A. Reisinezhad | Flood waves in contraction channels with different cross section profiles
Figure 17 | Time history plots of free-surface elevation at (a) point A, (b) point B and (c)
point C for channels with different narrow sections on a single ratio of
(W/Wg ¼ 2.0).
has a profound effect on the maximum of the drag force
exerted on the cylinder. It is quite clear that, in comparison
with the channel with a rectangular cross section, the
reduction in cross-sectional area of the channel with para-
bolic and fourth-degree cross sections would cause a
reduction in the maximum exerted drag force by about
10% and 15%, respectively.
Journal of Hydroinformatics | 14.2 | 2012
Figure 18 | Discretization of the computational domain (a coarse mesh) for quartic
narrow section in the presence of a circular cylinder for Wg/W ¼ 0.8.
CONCLUSION
In this paper, a 2D finite element model was introduced for
simulation of the unsteady flows involving the free-surface
phenomenon. Here, the unknown variables which include
the free-surface elevation and velocity components are
approximated by linear and quadratic shape functions,
respectively. Galerkin’s approach is used to spatially discre-
tize the shallow water equations, while a finite difference
scheme in implicit form has been used in the time domain.
The model was then applied to a number of test cases for
validation purposes. 1D and 2D dam breaks were modeled
and their behaviors were compared with analytical and
numerical solutions, respectively, and good agreements
were achieved. Then, the propagation of a solitary wave
was examined around a circular cylinder placed in a rec-
tangular channel. Comparison of the time history plots of
the non-dimensional free-surface elevation of the water
and the exerted non-dimensional forces on the cylinder
with the experimental data show reasonable agreement.
Subsequently, a flood wave at the upstream of the channel
was generated and the effect of the narrow section on the trend
of wave propagation was analyzed. Geometries considered
included channels with rectangular as well as quartic narrow
sections. Based on the results produced, it can be concluded
that a segment of the wave, due to the decrease in width of
 383 P. Ghadimi & A. Reisinezhad | Flood waves in contraction channels with different cross section profiles Journal of Hydroinformatics | 14.2 | 2012

Figure 19 | Different contour line plots of flood propagation in a channel with quartic narrow section in the presence of a circular cylinder: (a) t ¼ 2.0 s, (b) t ¼ 3.0 s and (c) t ¼ 4.0 s.

Figure 20 | Time history plots of exerted drag force on the cylinder for different Wg/W Figure 21 | Time history plots of exerted drag force on the cylinder for different cross
ratios. sections.
 384 P. Ghadimi & A. Reisinezhad | Flood waves in contraction channels with different cross section profiles
the channel, has been longitudinally reflected. One can also
claim that larger wave reflection is directly proportional to
an increase in the ratio of the channel width and the width
of the narrow section. It was also observed that a sudden
change in channel width (depicted by a rectangular narrow
section) has larger effects than the case with a gradual
change of width (depicted by a quartic narrow section).
To continue, the effect of a quartic narrow section open-
ing and the channel cross-section profile over the exerted
non-dimensional forces on a circular cylinder was investi-
gated. From the produced result, it can be expressed that,
in the channel with a gradual change of width, the change
in width of the opening of the narrow section does not
have any effect on the amount of maximum drag force. How-
ever, keeping the width and depth constant and decreasing
the channel cross-sectional surface due to the change in
cross-sectional shape was observed to cause a decrease in
the maximum drag force exerted on the cylinder.
REFERENCES
Abida, H., Ellouze, M. & Mahjoub, M. R.  Flood routing of
regulated flows in Medjerda River, Tunisia. J. Hydroinf. 7 (3),
209–216.
Aizinger, V. & Dawson, C.  A discontinuous Galerkin method
for two-dimensional flow and transport in shallow water.
Adv. Water Res. 25, 67–84.
Bellos, C., Soulis, J. & Sakkas, J.  Computation of two dimensional
dam break induced flow. Adv. Water Res. 14 (1), 31–41.
Brooks, A. N. & Hughes, T. J. R.  Streamline upwind/Petrov–
Galerkin formulations for convection dominated flows with
particular emphasis on the Navier–Stokes equations.
Comput. Meth. Appl. Mech. Engng. 32, 199–259.
Chambers, Z.  A Characteristics Finite Element Algorithm for
Computational Open Channel Flow Analysis. PhD Thesis,
University of Tennessee, Knoxville, TN, USA.
Connor, J. J. & Wang, J.  Finite-element modeling of
hydrodynamic circulation. In Numerical methods in fluid
dynamics. Pentech Press, London, pp. 55–387.
Dhatt, G., Soulaimani, A., Ouellet, Y. & Fortin, M. 
Development of new triangular elements for free surface
flows. Int. J. Numer. Meth. Fluids 6, 895–911.
Eldho, T. I. & Sharique, M. M. A.  Hydrodynamics modelling
of estuaries using an implicit finite element model. J. Inst.
Engrs. India Civil Engng. Div. 8, 15–23.
First received 18 April 2010; accepted in revised form 7 February 2011. Available online 21 June 2011
Journal of Hydroinformatics | 14.2 | 2012
Fennema, R. T. & Chaudhry, M. H.  Implicit methods for two-
dimensional unsteady free surface flows. J. Hydraul. Engng.
27 (3), 321–332.
Fennema, R. T. & Chaudhry, M. H.  Explicit method for 2D
transient free surface flows. J. Hydraul. Engng. 116 (11),
1013–1014.
Garcia, R. & Kahawitha, R.  Numerical solution of the
St. Venant equations with the Mac-Cormack finite
difference scheme. Int. J. Numer. Meth. Fluids 6,
507–527.
Hervouet, J. M.  Hydrodynamic of Free Surface Flows,
Modeling with Finite Element Method. John Wiley & Sons
Inc., New York.
Hood, P. & Taylor, C.  Navier–Stokes equations using
mixed interpolation. In: Finite Element Method in Flow
Problems (J. T. Oden, ed.). UAH Press, Huntsville,
pp. 121–131.
Katopodes, N.  Two dimensional surges and shocks in open
channels. J. Hydraul. Div. 110 (6), 794–812.
Katopodes, N. & Strelkoff, T.  Computing two-
dimensional dam break flood waves. J. Hydraul. Div. 104,
1269–1288.
Katopodes, N. & Strelkoff, T.  Two dimensional shallow water
models. J. Mech. Div. 105, 317–334.
Mohammadian, M., Le Roux, D. Y., Tajrishi, M. & Mazaheri, K.
 A mass conservative scheme for simulating shallow
water flows over variable Topographies using unstructured
grid. Adv. Water Res. 28, 523–539.
Mynett, A. E.  Hydroinformatics and its applications at Delft
Hydraulics. J. Hydroinf. 1 (2), 83–102.
Saad, Y. & Schultz, M. H.  GMRES, A Generalized Minimal
Residual Algorithm for Solving Non-symmetric Linear
Systems. Research report YALEU/RR 257.
Tseng, M. H.  Kinematic wave computation using an efficient
implicit method. J. Hydroinf. 12 (3), 329–338.
Valiani, A., Caleffi, V. & Zanni, A.  Case study: Malpasset
dam-break simulation using a two-dimensional finite volume
method. J. Hydraul. Engng. 128 (5), 460–472.
Venutelli, M.  A third-order Taylor–Galerkin model for the
simulation of two-dimentional unsteady free surface flows.
Appl. Math. Model. 22, 641–656.
Yates, G. T. & Wang, K. H.  Solitary wave scattering by a
vertical cylinder: experimental study. In: Proc. 4th
International Offshore and Polar Engineering Conf., Osaka
Vol 3. International Society of Offshore and Polar Engineers,
Cupertino, CA, pp. 118–124.
Walters, R. A. & Cheng, R. T.  Accuracy of an estuaries
hydrodynamical model using smooth elements. Water
Resour. Res. 16 (1), 187–195.
Zhao, D. H., Shen, H. W., Tabios, G. Q., Lai, J. S. & Tan, W. Y.
 Finite-volume two dimensional unsteady-flow model for
river basins. J. Hydraul. Engng. 120 (7), 863–883.
 385 P. Ghadimi & A. Reisinezhad | Flood waves in contraction channels with different cross section profiles Journal of Hydroinformatics | 14.2 | 2012

APPENDIX The advection matrices:

The matrices in Equation (26) are defined by ZZ @Njh

h
Txij ¼ vconv Nih dΩe i; j ¼ 1; 2; 3
@y
δhnþ1
j ¼ hnþ1
j  hnj j ¼ 1; 2; 3

ZZ @Njh
unþ1
j ; ν nþ1
j j ¼ 1; 2; . . . ; 6 h
Tyij ¼ vconv Nih dΩe i; j ¼ 1; 2; 3
@y

Mh
A1 ¼ þ θðTxh þ Tyh Þ ZZ
Δt @Nju
u
Txij ¼ uconv Niu dΩe i; j ¼ 1; 2; . . . ; 6
@x
Mu
A2 ¼ A3 ¼ þ θðTxu þ Tyu Þ  F u
Δt ZZ @Nju
u
Tyij ¼ uconv Niu dΩe i; j ¼ 1; 2; . . . ; 6
@y
B1 ¼ θBuh
x

The elevation gradient matrices:

B2 ¼ θBuh
y

ZZ @Nju
C1 ¼ gθCxuh xij ¼
Buh h prop Nih dΩe i ¼ 1; 2; 3; j ¼ 1; 2; . . . ; 6
@x

ZZ @Nju
C2 ¼ gθCyuh
yij ¼
Buh hprop Nih dΩe i ¼ 1; 2; 3; j ¼ 1; 2; . . . ; 6
@y

D1 ¼ ðTxh þ Tyh Þhn  ð1  θÞðBuh

x u þ By ν Þ
n uh n

The flux gradient matrices:

 u

M
D2 ¼  ð1  θÞðTxu þ Tyu Þ un  gCxuh Zns
Δt ZZ @N h
j
uh
Cxij ¼ Niu dΩe i ¼ 1; 2; . . . ; 6; j ¼ 1; 2; 3
  @x
Mu
D3 ¼  ð1  θÞðTxu þ Tyu Þ vn  gCyuh Zns
Δt
ZZ @N h
j
uh
Cyij ¼ Niu dΩe i ¼ 1; 2; . . . ; 6; j ¼ 1; 2; 3
The mass matrices: @y

ZZ
The friction matrices:
Mijh ¼ Njh Nih dΩe i; j ¼ 1; 2; 3
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
gn2 ðuni Þ2 þ ðvni Þ2 ZZ
ZZ Fiiu ¼  Niu dΩe i ¼ 1; 2; . . . ; 6
Miju ¼ Nju Niu dΩe i; j ¼ 1; 2; . . . ; 6 ðhni Þ4=3