International Sym posium on W ater City W ater Forum

Incheon, K orea 14 ~ 22 of A ugust, 2009

ON THE FINITE ELEMENT TECHNIQUE
FOR THE SHALLOW WATER EQUATIONS
APPLICATION TO URBAN FLOODING

NGUYEN The Hung
(1)
, TO Thuy Nga
(2)
, LE Hung
(3)
, LE Van Thao
(4)


(1)
Faculty of water resources engineering, Danang University of Technology, 54 Street Nguyen
Luong Bang, Lien Chieu Districts, Da Nang City, VietNam,
Phone: (84).905233440 ; e-mail: profhungthenguyen@gmail.com ; hungnt@udn.vn
(2)
Faculty of water resources engineering, Danang University of Technology, 54 Street Nguyen
Luong Bang, Lien Chieu Districts, Da Nang City, VietNam,
Phone: (84).905586568 ; e-mail: ngoclan0472@gmail.com
(3)
Faculty of water resources engineering, Danang University of Technology, 54 Street Nguyen
Luong Bang, Lien Chieu Districts, Da Nang City, VietNam,
Phone: (84).905888950 ; e-mail: lehungtk3@gmail.com
(4)
Faculty of water resources engineering, Danang University of Technology, 54 Street Nguyen
Luong Bang, Lien Chieu Districts, Da Nang City, VietNam,
Phone: (84).973496313 ; e-mail: quangngaiBK@gmail.com


ABSTRACT
A non standard Galerkin method for the solution of the shallow water equations in
conservative form with using the characteristic-based split algorithm is presented. The main
advantage of the model are high accuracy and ability solving the time discontinuous shock
waves, the supercritical and subcritical regimes as well as the transitions between the regimes;
the domain is discritization by triangular elements, so it is capable of handling complex
geometry. The results are verified by comparison with measurement data; a case study on
urban flooding, downstream of Han river, is showed.

Keywords: Finite element method, shallow water equations, characteristic-based split
algorithm

1 INTRODUCTION

The two dimensional horizontal flow model or shallow water equations system are very
popular for treatment of flow of lakes, rivers, oceans.
Many difficults issues for solving the two dimensional horizontal flow model are non-self
adjoint operator in governing equations and how to treat the nonlinear advective terms
(Cullen and Morton, 1980; Zienkiewicz and Taylor, 2002).
The finite element method when applied to hydrodynamic problems gives an accurate
solution. This method is conservative and therefore avoids aliasing errors associated with
nonlinear terms.
Moreover, it has the advantage over the finite difference method of being flexible in the
treatment of irregular domains and to allow a variable resolution, thus permitting a focus on
regions of interest.
Navon (1982, 1983) introduced the Numerov Galerkin finite element method for the
shallow water equations with an Augmented Lagrangian constrained optimization method to
enforce integral invariants conservation. Similar work was done by Zienkiewicz and
International Sym posium on W ater City W ater Forum
Incheon, K orea 14 ~ 22 of A ugust, 2009
Heinrich (1979) with a finite element penalty method, and by Zienkiewicz and others (1984).
Navon (1987 ), N. T. Hung (2004) applied a two-stage finite element Galerkin method
combined with a high-accuracy compact approximation to the first derivative for solving the
shallow water equations.
In this paper, Author develops the characteristic- based split (CBS) algorithm for the
temporal discretization and a non- standard Galerkin finite element method for space solving
the full shallow water equation.

2. THE CHARACTERISTIC BASED SPLIT ALGORITHM FOR THE SHALLOW
EQUATION
2.1. Governing equation
The shallow water equations in conservative form are as follow ( 7 ):
Where Q, F
x
, F
y
, H : are vectors, and

1
1
1
1
1
1
1
]
1

yx
xx
x
h
h
pq
h
gh
h
p
p
F
2
2
2
1 (3)
1
1
1
1
1
1
1
1
1
1
]
1

yy
xy
y
h
gh
h
q
h
h
pq
q
F
2
2
2
1
(4)
1
1
1
1
1
1
1
1
]
1

+
+

+
+

3 / 7 2
0
2 2 2
0
3 / 7 2
0
2 2 2
0
0
h C
q p q n
g
y
z
gh
h C
q p p n
g
x
z
gh H
(5)
where:
p = u.h, u = Being the depth-averaged x-direction component of velocity
q = v.h, v = Being the depth-averaged y-direction component of velocity
h = water depth
g = Celeration due to gravity

xx
,
xy
,
yx
,
yy
= Reynolds stresses where the first subscript indicates the axis
) 1 ( 0 +

H
y
F
x
F
t
Q
y
x
) 2 (

,
_

q
p
h
Q
International Sym posium on W ater City W ater Forum
Incheon, K orea 14 ~ 22 of A ugust, 2009
direction normal to the face on which the stress acts
= Fluid density
z
0
= Channel invert elevation
n = Mannings roughness coefficient
C
0
= Dimensional constant (C
0
= 1 for SI units and C
0
= 1.486 for English units)
The Reynolds stresses are determined using the Boussinesq approach relating stress to
the gradient in the mean currents:

y
v
v
x
v
y
u
v
x
u
v
t yy
t x y xy
t xx

2
) (
2
(6)
where
t
= is the eddy viscosity, which varies spatially and is solved empirically as a function
of local flow variables (Rodi 1980):

6 / 1
0
2 2
8 .
h C
q p g n C
t
+
(7)
Where C = is a coefficient that varies between 0.1 1.0

2.2 The split - temporal discretization
Equation (1) can rewrite as follow:

+

n y
x
H
y
F
x
F
t
Q
(8)
With H
n+
being treated as a known quantity evaluated at t = t
n
+ .t in a time
increment t; and [0,1].
Equation (8) can discritize in time using the characteristic - Galerkin process
(Zienkiewicz and Taylor, 2002)

n
y x
k
k
n n
y
n
x
n n
H F
y
F
x x
u
t
H F
y
F
x
t Q Q
1
]
1

+ +
)) ) ( ) ( ( . .
2
( ) ( ) (
1
(9)
where k = 1,2 ; u
1
= u ; u
2
= v
At this stage we have to introduce the split in which we substitute a suitable
approximation for H.
First, we remove the term H from Eq. (9), and the next step we introduce an auxiliary
variable Q
*
such that

n
y x
k
k y x
n
F
y
F
x x
u
t
F
y
F
x
t Q
Q Q Q
1
]
1

)) ( ) ( ( . .
2
) ( ) (
*
* *
(10)
Equation (10) will be solved subsequently by an explicit time step applied to the
discritized form. The correction step is as follow

k
n
k
n n
x
H
u
t
Q Q Q Q

+
. .
2
2
* 1
(11)
The above equation is fully self-adjoint in the variable Q which is the unknown; so a
Galerkin type procedure can be optimally used for spatial approximation.
International Sym posium on W ater City W ater Forum
Incheon, K orea 14 ~ 22 of A ugust, 2009

2.3 Finite element algorithm
The spatial discritization of the finite element non-Galerkin method for the equation
(11) is as follow:

global
i i
m
elements
i
dxdy y x Q dxdy y x Q y x Q ). , ( ). , ( ), , ( , i=i,j,k (12)
This notation defines the inner product when a function is multiplied by a trial function
i
(12):

i
= V
i
.I +
i
(13)
V
i
= Galerkin part of the trial function
I = Identity matrix

i
= Non- Galerkin part of the trial function
Linear piecewise polynomials on triangular elements is used where for a given
triangular element each variable is represented as a linear sum of interpolating functions, for
example

k j i j
j j
y x V t Q
, ,
e
) , ( ). ( Q (14)
where
Q
j
(t) = Represents the scalar nodal value of the variable Q at the node of three vertices
of the triangular element
V
j
= Basis function also the trial function
The basis function can be expressed in natural coordinate as follows:
V
i
= ) (
2
1
i i i
c x b y a
A
+ + (15)
where
A = Area of the triangle
b
i
= y
j
- y
k
, a
i
= x
k
- x
j
, c
i
= x
j
.y
k
- x
k
.y
j

i, j, k = Cyclically permuted ( i, j, k = 1, 2 , 3 ) (16)
In temporal discretization, we chose parameter = 0.66, so the schemes are semi-
implicit form. The time limitations are as follows

2
2
h
t t and
u
h
t t
i
(17)
W here
= Is the kinematic viscosity
h = Element sizes

2.4 Test problem
The test problem used is a truncation of rectangular channel with regular grid domain
(5 x 5). The time and space increments are: x = 100 m, y = 50 m, t = 60 sec,
coriolis force f = 2.w.sin
where w = 7.29 x 10
-5
sec
-1
, = 15
0
,

s
x
=
s
y
= 0, Manning coefficient n = 0.035,
acceleration of gravity g = 10 m.s
-2
.
International Sym posium on W ater City W ater Forum
Incheon, K orea 14 ~ 22 of A ugust, 2009




































Boundary conditions
The boundary conditions are chosen as follows: Upstream u = u(0,y,t) = 0.500 - 1.000
m/s , v = v(0,y,t) = 0.0 m/s, downstream h = 4.00 + 0.5sin(t/21600) m, and rigid boundary
conditions v = v(x,100,t) = 0.0 m/s , v = v(x,300,t) = 0.0 m/s
Results:
Figures 1 and 2 show the calculated water surface elevation and the velocity distribution
at the nodes of the grid domain using the characteristic- based split algorithm for the temporal
discretization and standard Galerkin method for the spatial discritization. The solutions are
verified by comparison with measurement data and analytical solutions showing a good agree:
max error of velocity < 0.03 m/s
max error of water depth < 0.01 m
The non-standard Galerkin method test problem will present in another paper.

3 APPLICATION TO URBAN FLOODING
The focused urban flooding region is downstream of Han River of Danang City, Vietnam.
The upstream boundary conditions of region are imposed discharge, and the downstream ones
are tidal water level.
Figure 1 : Calculus time step at nt = 4320 of five day forecast
contour of water surface elevation (h)

Figure 2: Calculus time step at nt = 4320 of five day forecast
Contour of velocity distribution (u)
International Sym posium on W ater City W ater Forum
Incheon, K orea 14 ~ 22 of A ugust, 2009

Figure 3: Inundations map of Downstream of Han River

According to statistic in the past, the frequency of discharge in Thanh My and Nong Son is
difference and causes flooding damage in the downstream of Han river when there is great
rainfall at Tuy Loan basin and the dangerous flood shapes are double crest (the flood shapes
are similar in the past: November 1999, the discharge of Bung river Q
Bung
= 25% Q
ThanhMy
,
Q
Con
= 5% Q
ThanhMy
. A combination of flood discharge, at upstream, that is calculated to
frequency at Thanh My, Nong Son and Tuy Loan, may cause heavy flood, these forecasting
problems are showed as follows (NGUYEN The Hung & LE V an H oi, 2001):
Upstream boundary condition
Scenario A
+ Discharge at Thanh My, with frequency 2% (Q
max2%
= 7400 m
3
/s), Nong Son 5% (Q
max5%
=
12300 m
3
/s), Tuy Loan 1% (Q
max1%
= 1650 m
3
/s) (scenario A, from A.1.1, A1.2 to A.3.3.2).
Scenario B
+ Discharge at Thanh My, with frequency 5% (Q
max5%
= 6200 m
3
/s), Nong Son 2% (Q
max2%
=
15000 m
3
/s), Tuy Loan 1% (Q
max1%
= 1650 m
3
/s) (scenario B, from B.1.1, B1.2 to B.3.3.2).
Downstream boundary condition (Tien Sa station)
Tidal shape happens in 7 days (heavy flood is usually 7 days), with frequency 1%, 5%,
10% and the case study as follows:
+ Normally (h = 0,00 m);
+ tidal surge by storm with frequency 1% ( h = 2,1 m ), 5%( h = 1,25 m );
+ tidal surge by seasonal wind ( h = 30 cm );

Abbrivation of forecasting scenarios are given the labels as follows:
letter A corresponding to flood discharge at Thanh My 2%, Nong Son 5%, Tuy Loan 1%
letter B corresponding flood discharge at Thanh My 5%, Nong Son 2%, Tuy Loan 1%
first numbers (the numbers after the letters: A
1
, B
1
)
- Number 1 shows tidal water level boundary at Tien Sa station with frequency 1%
International Sym posium on W ater City W ater Forum
Incheon, K orea 14 ~ 22 of A ugust, 2009


Figure 4: Upstream boundary condition of scenario A ! Discharge Q " t


Figure 5: Upstream boundary condition of scenario B - Discharge Q " t


Figure 6: Downstream boundary condition ! Tidal water level H " t
International Sym posium on W ater City W ater Forum
Incheon, K orea 14 ~ 22 of A ugust, 2009

- Number 2 shows tidal water level boundary at Tien Sa station with frequency 5%
- Number 3 shows tidal water level boundary at Tien Sa station with frequency 10%
+ next numbers (the numbers following the first numbers: A.1.1 ; B.1.2
- Number 1 shows non-tidal surge
- Number 2 shows tidal surge by seasonal wind
- Number 3 tidal surge by storm: 3.1 storm with frequency 5%;
3.2 storm with frequency 1%
Results of flooding forecast:
Results of flooding forecast corresponding with many different scenario, combination of
flooding frequency and different boundary conditions are given in 2 tables follows.
a/ Flooding discharge at Thanh My with frequency Q
ThanhMy
2%, Nong Son Q
NongSon
5%,
Tuy Loan Q
TuyLoan
1% (scenario A, from A.1.1 to A.3.3.2)
Table 2: Collective results of forecasting scenario
Q
ThanhMy
2%, Q
NongSon
5%, Q
TuyLoan
1%


Water elevation at station (m) Maximum velocity
(m/s)
scenario NHS47 Cam Le Cau Do NHS15 NB47
A.1.1 3.91 4.55 4.8 5 5.79 2.24
A.1.2 3.96 4.57 4.85 5 5.8 2.17
A.1.3.1 4.24 4.67 4.92 5.04 5.82 1.92
A.1.3.2 4.45 4.86 5.06 5.16 5.89 1.65
A.2.1 3.9 4.56 4.83 4.97 5.79 2.2
A.2.2 3.94 4.58 4.85 4.98 5.8 2.13
A.2.3.1 4.14 4.69 4.94 5.05 5.83 1.87
A.2.3.2 4.45 4.89 5.09 5.19 5.9 1.61
A.3.1 3.88 4.57 4.84 4.97 5.79 2.27
A.3.2 3.91 4.59 4.85 4.99 5.8 2.22
A.3.3.1 4.1 4.7 4.95 5.06 5.83 2,00
A.3.3.2 4.5 4.9 5.11 5.21 5.9 1,78

b/ Flooding discharge at Thanh My with frequency 5%, Nong Son 2%, Tuy Loan 1%
(scenario B, from B.1.1 to B.2.3.2)

Table 3: Collective results of forecasting scenario
Q
ThanhMy
5%, Q
NongSon
2%, Q
TuyLoan
1%


Water elevation
(m)
at station Maximum
velocity (m/s)
Scenario NHS47 Cam Le Cau Do NHS15 NB47
B.1.1 3.83 4.47 4.73 4.9 5.7 2.12
B.1.2 4.87 4.49 4.75 4.97 5.7 2.15
B.1.3.1 4.06 4.59 4.83 4.97 5.73 1.88
B.1.3.2 4.4 4.78 4.97 5.09 5.79 1.55
B.2.1 3.82 4.48 4.74 4.9 5.7 2.17
B.2.2 3.87 4.5 4.75 4.91 5.7 2.1
B.2.3.1 4.08 4.51 4.85 4.98 5.73 1.82
B.2.3.2 4.38 4.82 5.01 5.12 5.8 1.56
International Sym posium on W ater City W ater Forum
Incheon, K orea 14 ~ 22 of A ugust, 2009
4 CONCLUSION
In comparison with the Galerkin finite element method , it is observed that the
characteristic - non Galerkin method with using the characteristic-based split algorithm for the
solution of the shallow water equations in conservative form is more accuracy and ability to
solve the time discontinuous shock waves, the supercritical and subcritical regimes as well as
the transitions between the regimes because of self-adjoint operator of governing equations
and using special test function. This model is good applying in urban flooding.

REFERENCES

Cullen, M. J. P., and Morton, K., K., 1980, Analysis of evolutionary error in finite element
and other methods, Jour. Comput. Phys., v.34, p. 245-267.
Navon, I. M., 1982, A Numerov-Galerkin technique applied to a finite element shallow
water equations model with exact conservation of integral invariants, in Kaway, T., ed.
Finite element flow analysis, Univ. Tokyo Press, Tokyo, p. 75-86.
Navon, I. M., 1983, A Numerov-Galerkin technique applied to a finite element shallow water
equations model with enforced conservation of integral invariants and selective lumping,
Jour. Comput. Phys., v.52, N
0
.2, p. 313-339.
Navon, I. M., 1987, A two stage, high accuracy, finite element Fortran program for solving
shallow water equations, Computers & Geosciences, v.13, n
0
. 3, p. 255-285.
NGUYEN The Hung, Mathematical model of the two dimensional vertical flow, Journal of
Vietnam National Science & Technology, N
0
7+8, Hanoi 1990.
NGUYEN The Hung, Mathematical modeling of sediment transport two dimensional
horizontal, Proceedings of International Conference on Engineering Mechanics Today,
Vol. 1, p. 541-548, Hanoi 1995.
NGUYEN The Hung, 2004, Finite element method in flow problems, Construction Pub.
House, Hanoi.
NGUYEN The Hung, 2005, A two stage, high accuracy, finite element technique of the two
dimensional horizontal flow model, Modeling, Simulation and Optimization of Complex
Processes: Springer- Verlag Berlin Heidelberg, p. 255-233.
NGUYEN The Hung & LE Van Hoi (2001), Applying mathematical models forecast
inundation of Han river, Danang City (Project: Treatement of pollution environment causing
flooded and intensity of capacity to cope of flooding damage problems in Danang City).
Zienkiewicz, O. C., and Heinrich, J. C., 1979, A unified treatment of steady stage shallow
water equations and two dimensional Navier-Stocks equations a finite element penalty
function approach : Computer Math. Appl. Mech. and Eng., v. 17/18, p. 673-688.
Zienkiewicz, O. C., Vilotte, J. P., Nakazawa, S., and Toyoshima, S., 1984, Iterative methods
for constrained and mixed approximation : an inexpensive improvement of F.E.M.
performance : Inst. Numerical methods in Engineering Report C/R/489/84, Swansea,
United Kingdom, 20p.
Zienkiewicz, O. C., and Taylor R. L., The Finite element method, 5
th
edition Butterworth
Heinemann 2000.