Applied Mathematical Modelling 24 (2000) 457±475

www.elsevier.nl/locate/apm

Numerical solution of the two-dimensional unsteady

dam break
C. Zoppou *, S. Roberts
Center for Mathematics and Its Applications, Australian National University, Canberra, ACT 0200, Australia
Received 13 July 1998; accepted 19 October 1999

Abstract

A numerical model for the solution of the two-dimensional dam break problem is described. The model which is
based on a second-order approximate Riemann solver with a van Leer type limiter is used to solve the shallow water
wave equation on a Cartesian grid. The shallow water equations include source terms which account for resistance to
the ¯ow and the in¯uence of the bed slope. The wetting and drying process is also considered. The model is veri®ed by
comparing the model output with documented results. Very good agreement has been achieved in the veri®cation
phase. A simple example of the collapse of a water supply reservoir in a narrow valley is used to demonstrate the
capability of the model. The model is shown to be stable and robust. It is capable of resolving shocks, handling complex
geometry, including the in¯uence of steep bed slopes and simulating the wetting and drying process. Ó 2000 Elsevier
Science Inc. All rights reserved.

Keywords: Shallow water waves; Dam break; Riemann solvers; Two-dimensional unsteady ¯ows

Nomenclature
Cr Courant number Cr ˆ wDt=Dx
E ¯ux vector in the y-direction
F ¯ux vector in the radial direction
p
Fn Froude number Fn ˆ w= gh
G ¯ux vector in the x-direction
g acceleration due to gravity
H homogeneous solution operator
h water depth
i cell
n the time level
O ordinary di€erential equation operator
q dependent variable
r radial distance

*
Corresponding author. Present address: ACTEW Corporation, G.P.O. Box 366, Canberra, ACT 2601, Australia. Tel.: +61-02-
6279-8641; fax: +61-02-6279-8651.
E-mail address: [email protected] (C. Zoppou).

0307-904X/00/$ - see front matter Ó 2000 Elsevier Science Inc. All rights reserved.
PII: S 0 3 0 7 - 9 0 4 X ( 9 9 ) 0 0 0 5 6 - 6
458 C. Zoppou, S. Roberts / Appl. Math. Modelling 24 (2000) 457±475

ri local ¯ow parameter

S source vector
Sl left shock
Sm contact discontinuity
Sr right shock
Sfx friction slope in the x-direction
Sfy friction slope in the y-direction
S0x bed slope in the x-direction
S0y bed slope in the y-direction
t time
U vector of conserved variables
u velocity in the x-direction
v velocity in the y-direction
w radial velocity
Wk weights
x distance
y distance
Dqlocal local jump in q
Dqupwind upwind jump in q
Dt time step
Dx cell width
g the Manning resistance coecient
ki eigenvalues
m local Courant number
/ total variational diminishing limiter function
w total variational diminishing limiter

1. Introduction

All water impoundment represents a potential risk to life and property. During this century
there have been more than 200 failures of dams greater than 15 m high [1]. They have caused
millions of dollars worth of damage and a loss of more than 8000 lives. These recent disasters
have focused attention on means for assessing public safety and predicting probable damage from
the collapse of a dam. Numerical models have become essential as a predictive tool in evaluating
the risks associated with the failure of the hydraulic structures. In the past, these have been re-
stricted to one-dimensional models [2,3]. Unfortunately, a number of one-dimensional models are
formulated with a non-conservative form of the shallow water wave equations and solutions are
obtained using four-point implicit schemes. The interaction between the main channel ¯ow and
the ¯oodplain ¯ow requires a full two-dimensional model which simulates the drying and wetting
process. In many practical problems ¯ows could change abruptly in time, such as the occurrence
of shock waves when opening sluice gates, or in a dam break, or a breach of a levee. The con-
servative form of the governing equations are more appropriate for problems containing dis-
continuities in the solution. There are techniques for the solution of systems of conservation laws
for problems which involve discontinuous solutions. These schemes have been applied to a wide
class of problems involving the Euler equation of gas dynamics. Only recently have these schemes
been used for the solution of the shallow water wave equations [4±9]. Problems which involve
discontinuities in the solution are known as Riemann problems and numerical schemes essentially
 C. Zoppou, S. Roberts / Appl. Math. Modelling 24 (2000) 457±475 459

solve linearized Riemann problems. To achieve higher-order accuracy and enhanced shock res-
olution, sophisticated techniques are required to obtain physically plausible solutions from the
Riemann solver. This is achieved by limiting the ¯ux or the dependent variables so that oscilla-
tions that are associated with higher order schemes are eliminated. Therefore, models which are
used to simulate rapidly varying unsteady ¯ows must be conservative, must have the ability to
accurately resolve the advancing water front, must include the in¯uence of steep bed slopes and
boundary friction, as well as simulating ¯ow over a dry bed in two-dimensions.
The purpose of this paper is to report on the applicability of a recently developed second-order
weighted average ¯ux scheme [10±14] for the solution of the two-dimensional dam break problem.
The fractional step method is used to solve the shallow water wave equation on a rectangular grid.
This reduces the two-dimensional problem into a sequence of augmented one-dimensional
problems. The second-order approximate Riemann solver is used to solve the homogeneous
operator in the two co-ordinate directions and an adaptive fourth-order Runge±Kutta scheme is
used to solve the ordinary di€erential equation operator. We examine the performance of this
scheme using exact solutions and other numerical results documented in the literature. These test
problems include one- and two-dimensional test cases with varying boundaries and variable
bathymetry and friction. The e€ect of standard limiters in the second-order scheme is also ex-
amined. The main advantage of the proposed method is its ability to simultaneously handle
calculations of slowly varying ¯ows as well as rapidly varying ¯ows containing shocks or dis-
continuities such as those commonly occurring in dam break problems.

2. Shallow water equations

The two-dimensional shallow water wave equation in general conservative form is given by
oU oE oG
‡ ‡ ˆ S; …1†
ot ox oy
where U is the vector of conserved variables, E and G are the ¯ux vectors in the x- and y-
directions and S represents a source vector. The vectors U, E and G can be expressed in terms of
the primary variables, u; v and h as [15]
2 3 2 3 2 3
h uh vh
U ˆ 4 uh 5; E ˆ 4 u2 h ‡ gh2 =2 5 and G ˆ 4 uvh 5;
2 2
vh uvh v h ‡ gh =2
where g is the acceleration due to gravity, h the water depth, u and v are the ¯ow velocity in the x-
and y-directions, respectively. We assume that the system is strictly hyperbolic, that is the matrix
A ˆ oE…U†=oU and B ˆ oG…U†=oU have three real and distinct eigenvalues.
The source term S is given by
2 3
0
S ˆ 4 gh…S0x ÿ Sfx † 5;
gh…S0y ÿ Sfy †
which contains the e€ects of the bed slopes S0x and S0y in the two-Cartesian directions. The bed
friction Sfx and Sfy have been estimated using the Manning resistance law
p p
ug2 u2 ‡ v2 vg2 u2 ‡ v2
Sfx ˆ and Sfy ˆ ;
h4=3 h4=3
in which g is the Manning resistance coecient.
460 C. Zoppou, S. Roberts / Appl. Math. Modelling 24 (2000) 457±475

3. Numerical solution of the two-dimensional shallow water equations

It is the hyperbolic character of the shallow water wave equations that makes ®nding solutions
to these equations dicult. Hyperbolic equations admit discontinuous as well as smooth or
classical solutions. Even for the case in which the initial data is smooth, the nonlinear character
combined with the hyperbolic type of the equations can lead to discontinuous solutions in ®nite
time. The nonlinear character of the equations also means that analytical solutions to these
equations are limited to only very special cases. Numerical methods must be used to obtain so-
lutions to practical problems. Therefore, numerical techniques which admit discontinuities in the
solution, secondary shocks, re¯ections of waves, dry bed and obstructions are necessary for the
solution of the shallow water wave equations. The method of characteristics [16], ®nite di€erences
[6,17±19], ®nite-elements [20,21] and Godunov-type schemes [4,9,13] can be used to solve the
shallow water wave equations.
Initial value problems which involve discontinuities in the solution are known as Riemann
problems. Numerical schemes based on the solution of local Riemann problems are generally
known as Godunov-type methods. Their main advantages are that they are robust and accurately
capture the location of discontinuities such as shocks and contact surfaces. Two-dimensional
Riemann solvers do not appear to have matured enough to be used in the construction of multi-
dimensional schemes. Even if such solvers were available the resulting schemes are likely to be too
complicated for common use [10]. Recently, there have been proposed a number of ecient
approximate Riemann solvers [6,10,11]. These can be used to solve the two-dimensional dam
break problem on a Cartesian grid.
A recent study of a number of numerical schemes for solving the dam break problem under-
taken by the authors has found the approximate Riemann solver developed by Toro [12] to be
robust and ecient. Unlike many other schemes, such as ®nite di€erences and other approximate
Riemann solvers, it avoids the problems associated with sonic points, where a non-physical ex-
pansive shock is produced. Toro's second-order approximate Riemann solver has been used to
solve the dam break problem on a Cartesian grid.

3.1. Cartesian grid

Computation of two-dimensional ¯ows by Godunov-type solvers can be carried out by a se-

quence of augmented one-dimensional operators applied successively in the x- and y-directions.
The standard way to extend these schemes to two and higher space dimensions is via fractional
step or space operator splitting [22]. The simplest space splitting involves replacing (1) by the
sequence of two one-dimensional problems
oU oE
‡ ˆ0 …2†
ot ox
and
oU oG
‡ ˆ0 …3†
ot ox
followed by the solution of the ordinary di€erential equation
dU
ˆ S: …4†
dt
Given initial data Un at time level n for problem (1), the solution Un‡1 at time level
n ‡ 1 is obtained in the following steps; Solve (2) with data Un to obtain an intermediate solution
 C. Zoppou, S. Roberts / Appl. Math. Modelling 24 (2000) 457±475 461

n‡1 n‡1 n‡1

U (x-sweep) followed by solving (3) with the data U to obtain the solution U (y-sweep).
n‡1
Finally, solve the ordinary di€erential equation (4) using U to obtain the complete solution
Un‡1 .
For the x-direction the homogeneous operator is
oU oE
‡ ˆ 0; …5†
ot ox
where
2 3 2 3
h uh
U ˆ 4 uh 5 and E ˆ 4 u2 h ‡ gh2 =2 5;
vh uvh
which has three distinct eigenvalues
p p
k1 ˆ u ÿ gh; k2 ˆ u and k3 ˆ u ‡ gh:
The y-component of the velocity, v is decoupled from the two other equations. The conser-
vative quantity vh is passively advected with the speed u.
Since we are interested in problems with discontinuities, shock waves and contact surfaces, it is
more appropriate to use the integral form of (5)
I
…U dx ÿ E dt† ˆ 0; …6†

which accepts discontinuities in the dependent variables. Evaluating the integral in the control
volume ABCD of dimensions Dx and Dt shown in Fig. 1 produces the conservative scheme
Dt  n 
Un‡1
i ˆ Uni ÿ Ei‡1=2 ÿ Eniÿ1=2 ; …7†
Dx
where Uni denotes the discrete value of U in cell i at time level n and Eni‡1=2 is the intercell ¯ux.
Godunov-type methods utilize the solution of the local Riemann problem with data, Ul ˆ Uni and
Ur ˆ Uni‡1 to de®ne numerically the intercell ¯ux, Eni‡1=2 :
The Riemann problem for (5) is the initial value problem in the domain ÿ1 < x < 1; t > 0
with the data
 n
Ui if x 6 0;
U…x; 0† ˆ …8†
Uni‡1 if x > 0:
For the shallow water wave equation (5), the Riemann problem can be decomposed into three
waves, see Fig. 2. The middle wave is always a contact discontinuity and the left and right waves
can be either shocks or rarefaction waves. Therefore, there are only four possible wave combi-
nations involving discontinuous contact and shocks waves and rarefaction waves, which are

Fig. 1. Control volume for cell i and the intercell ¯uxes.

462 C. Zoppou, S. Roberts / Appl. Math. Modelling 24 (2000) 457±475

Fig. 2. Weights used in the second-order weighted average ¯ux scheme.

continuous. In the region between these waves, h and u are always constant but there is a dis-
continuity in v across the contact discontinuity.
The ®rst-order Godunov method uses the state along the t-axis (x ˆ 0† to de®ne the intercell
¯ux, Eni‡1=2 for the solution of the Riemann problem with data, Uni and Uni‡1 .

3.2. Second-order scheme

Toro [12] developed a second-order accuracy Godunov-type method which uses a solution
averaged over space and time. The averaging takes the form of an integral of the ¯ux, or chosen
variable over some volume.
In (7) Uni is the integral average of U in cell i at time level n and Eni‡1=2 is the integrated nu-
…k†
merical ¯ux. The local Riemann problem (8) consists of three waves with speeds ai‡1=2 separating
four constant states. In Fig. 2 these are denoted by
…k† …1† …4†
Ui‡1=2 ; 1 6 k 6 4 with Ui‡1=2  Uni and Ui‡1=2  Uni‡1 :

The intercell ¯ux, Eni‡1=2 is estimated by integrating along AB in Fig. 2, from x ˆ ÿDxi =2 to
x ˆ Dxi =2 at time t ˆ Dt=2 so that
Z Dxi =2
n 1
Ei‡1=2 ˆ E…U † dx: …9†
Dxi ÿDxi =2

U ˆ U…2x=Dt; Uni ; Uni‡1 † is the solution of the Riemann problem with data Uni ; Uni‡1 evaluated
at the half time level and over an irregular grid, Dxi . The integration could be performed exactly,
but this would be computationally expensive. In practice it is sucient to represent all waves in
the solution as single straight ray waves [10], then (9) becomes a weighted average of the ¯uxes
X
4
…k†
Eni‡1=2 ˆ W …k† Ei‡1=2 ; …10†
kˆ1

…k†
where the coecients Wk or weights are the geometric extents of the constant states. The ¯ux Ei‡1=2
is the value of the ¯ux function E in the region k of the local Riemann problem with data
Uni ; Uni‡1 . The region k ˆ 1 refers to Ul , on the left side of the wave speed Sl in the …x; t† diagram,
k ˆ 2 to the region between wave speeds Sl and Sm and so on up to k ˆ 4, which refers to state Ur .
S …k† is the speed of wave k between states k and k ‡ 1 of the Riemann problem. The weights Wk can
…k† …k† …k†
be written in terms of the Courant numbers mi‡1=2 ˆ Si‡1=2 Dt=Dx, k ˆ 1; 2; 3; 4 with Si‡1=2 the speed
of wave k as follows
 C. Zoppou, S. Roberts / Appl. Math. Modelling 24 (2000) 457±475 463

 .
…k† …kÿ1†
W …k† ˆ mi‡1=2 ÿ mi‡1=2 2 …11†

in which

…0† …4†
X
4
mi‡1=2 ˆ ÿ1; mi‡1=2 ˆ 1; W …k† P 0 8k and W …k† ˆ 1:
kˆ1

Substituting (11) into (10)

1 1X 3
…k† …k†
Ei‡1=2 ˆ …Ei ‡ Ei‡1 † ÿ m DE ;
2 2 kˆ1 i‡1=2 i‡1=2
where
…k† …k‡1† …k†
Ei ˆ E…Uni †; DEi‡1=2 ˆ Ei‡1=2 ÿ Ei‡1=2
is the ¯ux jump across wave k.
These weights produce a scheme that is second-order accurate in space and time and like all
second-order schemes introduces spurious oscillations. In practice, the weights (11) are replaced
by
 .
…k† …kÿ1†
W …k† ˆ /i‡1=2 ÿ /i‡1=2 2

with
…0† …4†
/i‡1=2 ˆ ÿ1; /i‡1=2 ˆ 1:
…k† …k† …k†
The functions /i‡1=2 ˆ sign…Si‡1=2 †wi‡1=2 , where sign…a† ˆ a=jaj are total variation diminishing
limiter functions, used to control the spurious oscillations associated with second-order schemes.
…k†
One possible choice for the limiter, wi‡1=2 is
8 …k† …k†
>
> 1 ÿ 2…1 ÿ jSi‡1=2 j† if ri > 2;
>
>
>
> …k† …k† …k†
< 1 ÿ ri …1 ÿ jSi‡1=2 j† if 1 6 ri 6 2;
…k† …k† …k†
wi‡1=2 ˆ jSi‡1=2 j if 1=2 6 ri 6 1; …12†
>
>
>
> …k† …k† …k†
1 ÿ 2ri …1 ÿ jSi‡1=2 j† if 0 6 ri 6 1=2;
>
>
: …k†
1 if ri 6 0;
which is related to the well-known SUPERBEE [23] limiter. Another choice for a limiter is
8
…k† …k†
>
< jSi‡1=2 j if ri > 1;
…k†
wi‡1=2 ˆ ri…k† …k†
if 0 6 ri 6 1; …13†
>
: …k†
1 if ri 6 0;
which is based on the minimod [23] limiter. The limiter
(
…k† …k† …k† …k†
…k† 1 ÿ 2ri …1 ÿ jSi‡1=2 j†=…1 ‡ ri † if ri P 0;
wi‡1=2 ˆ …k†
…14†
1 if ri 6 0;
is based on the van Leer [23] limiter.
The e€ect of these limiters as well as several other forms that can be found in the literature [11]
are the same. They add sucient dissipation to the scheme to guarantee monotonicity near steep
464 C. Zoppou, S. Roberts / Appl. Math. Modelling 24 (2000) 457±475

gradient, while retaining second-order accuracy in smooth regions. These limiter functions depend
…k†
on the local Courant numbers mi‡1=2 and a local ¯ow parameter
…k† Dqupwind
ri‡1=2 ˆ ;
Dqlocal
where q is some suitable variable. Here, Dqlocal is the local jump in q given by
…k‡1† …k†
Dqlocal ˆ qj‡1=2 ÿ qj‡1=2

and Dqupwind is the upwind jump

( …k‡1† …k† …k†
qiÿ1=2 ÿ qiÿ1=2 if Si‡1=2 P 0;
Dqupwind ˆ …k‡1† …k† …k†
qi‡3=2 ÿ qi‡3=2 if Si‡1=2 < 0:
…k†
If Dqlocal ˆ 0 then ri‡1=2 ˆ 1: For most waves q can be taken as either the water depth or ¯ow
velocity for the shallow water wave equations. Fraccarollo and Toro [11] recommend the use of
q ˆ u as the better choice for two-dimensional shallow water problems. Toro [13] used q ˆ h for
the shocks and q ˆ v for the contact discontinuity. These are the choices used here for the solution
of the two-dimensional shallow water wave equation using the second-order scheme. Note that
…k† …k†
ri‡1=2  1 corresponds to gradients in q that vary slowly in space, and ri‡1=2  0 corresponds to
rapid increase in the magnitude of the gradient q in the direction of wave propagation. In the
…k†
extreme case ri‡1=2 ˆ 0 8k; in which case the second-order intercell ¯ux at boundary i ‡ 1=2 re-
duces to the Godunov ®rst-order intercell ¯ux.
The second-order total variational diminishing intercell ¯ux becomes

1 1X 3
…k† …k† …k†
Ei‡1=2 ˆ …Ei ‡ Ei‡1 † ÿ sign…Si‡1=2 †wi‡1=2 DEi‡1=2 : …15†
2 2 kˆ1

Estimates of the ¯uxes E…2† and E…3† are required by (15), see Fig. 2. Since the solution for h and
u are una€ected by v, then the ¯uxes across the contact discontinuity are identical except for the
¯ux involving the variable v. The ¯uxes across the contact discontinuity are given by
2 3 2 3
…2† …2†
E1 E1
6 7 6 7
E…2† ˆ 4 E…2†
2 5 and E…3† ˆ 4 E…2† 2 5;
…2† …2†
E1 vl E1 v r
…2† …2†
where E1 and E2 are the vector components of the ¯ux calculated by satisfying (6) over the
control volume ABCD shown in Fig. 2 for subcritical ¯ow. At time Dt=2, Dx ˆ Sl Dt and
Dx ˆ Sr Dt. Evaluating (6) over the control volume AH0D gives U ˆ Ul ‡ …E…2† ÿ E…1† †=Sl and
over the control volume HBC0, U ˆ Ur ‡ …E…4† ÿ E…3† †=Sr . Eliminating U yields

Sr E…1† ÿ Sl E…4† ‡ Sl Sr …Ur ÿ Ul †

E…2† ˆ E…3† ˆ ; …16†
Sr ÿ Sl
which is also satis®ed for supercritical ¯ow [11]. These ¯uxes along with Ei and Ei‡1 are used in
(15) to estimate the intercell ¯ux Ei‡1=2 . Toro [12] called this scheme the weighted average ¯ux
scheme. The weighted average ¯ux scheme could also be used to solve the local one-dimensional
Riemann problem in the Finite Volume Method [24,25] for the solution of the two-dimensional
problem using unstructured grids.
 C. Zoppou, S. Roberts / Appl. Math. Modelling 24 (2000) 457±475 465

3.2.1. Boundary conditions

Two ®ctitious cells are required next to each re¯ective or transmissive boundary condition.
These are denoted by i ˆ ÿ1 and i ˆ 0 for the left boundary. For the transmissive boundary
condition at i ˆ ÿ1=2
W0 ˆ W1 ;
Wÿ1 ˆ W2 ;
where W can be the vector of conservative or primitive variables. For the re¯ective boundary
condition at i ‡ 1=2
hi‡1 ˆ hi ; ui‡1 ˆ ÿui ;
hi‡2 ˆ hiÿ1 ; ui‡2 ˆ ÿuiÿ1 :

4. Hypothetical examples

A one-dimensional problem which has a known analytical solution given by Stoker [26] is used
to assess the accuracy of the weighted average ¯ux scheme. The one-dimensional channel 2000 m
in length is horizontal, frictionless, wide and the water is initially stagnant. The water depth h1
upstream of the dam which is located at 1000 m remains constant at 10 m and the downstream
water depth, h0 ˆ 0:1 m. There is an instantaneous breach of a dam. In this problem there is a
transition from subcritical to supercritical ¯ows and it represents a severe test which has high-
lighted problems with a number of numerical schemes. The analytical solution and the solution to
this problem obtained using the second-order weighted average ¯ux scheme is shown in Fig. 3.
There is excellent resolution of both the shock and rarefaction waves. In addition, the entropy
violating solution, which plague many of the other schemes is almost eliminated using this scheme
[27].
The ®rst two-dimensional hypothetical problem is the example used by Fennema and
Chaudhry [18]. In this problem the dam is assumed to fail instantaneously. The water depth
upstream of the dam is 10 m and the downstream water depth is assumed to be either 5, 0:1 m or
dry (0 m). The computational domain consists of a 200 m  200 m region which has been
subdivided into a 41  41 square grid with the dimension of each element as 5 m  5 m. The 200
m long dam wall, which runs parallel to the y-axis is 10 m wide and is centered at x ˆ 100 m. The

Fig. 3. Weighted average ¯ux solution of the one-dimensional dam-break problem where h1 ˆ 10 m and h0 ˆ 0:1 m.
466 C. Zoppou, S. Roberts / Appl. Math. Modelling 24 (2000) 457±475

breach is 75 m wide and centered at 75 m. Along the boundaries, x ˆ 0 and x ˆ 200 m and h are
®xed at the upstream and downstream water depth, respectively. All other boundaries are con-
sidered as re¯ective boundary conditions. The results at t ˆ 5 s were obtained using a time step
Dt ˆ 0:05 s in the numerical scheme with the van Leer type limiter. The original example de-
scribed by Fennema and Chaudhry [18], in which the downstream water depth is 5 m is not a
severe test for the model because the ¯ow is subcritical everywhere. Alcrudo and Garcia-Navarro
[17] suggest that much smaller downstream water depth is a more severe test for the numerical
scheme (cf. Figs. 4±9). Therefore, perspective plots of the water depth for the problems where the
downstream water depth, h0 ˆ 5; 0:1 and 0 m are shown in Figs. 4, 6 and 8, respectively. The
corresponding velocity vectors superimposed on contours of the water depth are shown in Figs.
5,7 and 9. These results are very similar to others found in the literature for this problem [4±6,9].
When there is a ®nite water depth downstream, a shock front always exists, see Figs. 4 and 6.
For the dry bed case, there is no shock traveling downstream, see Fig. 8. There is a signi®cant
di€erence in the velocity vectors between the dry bed and ®nite depth downstream results. In Fig. 9
for example, the velocity vectors occur throughout the computational domain downstream of the

Fig. 4. Weighted average ¯ux solution of a anti-symmetric dam break in a frictionless, horizontal domain with upstream
water depth, h1 ˆ 10 m and downstream water depth h0 ˆ 5 m.

Fig. 5. Velocity vectors and contours for the weighted average ¯ux solution of a anti-symmetric dam break in a
frictionless, horizontal domain with upstream water depth, h1 ˆ 10 m and downstream water depth h0 ˆ 5 m.
 C. Zoppou, S. Roberts / Appl. Math. Modelling 24 (2000) 457±475 467

Fig. 6. Weighted average ¯ux solution of a anti-symmetric dam break in a frictionless, horizontal domain with upstream
water depth, h1 ˆ 10 m and downstream water depth h0 ˆ 0:1 m.

Fig. 7. Velocity vectors and contours for the weighted average ¯ux solution of a anti-symmetric dam break in a
frictionless, horizontal domain with upstream water depth, h1 ˆ 10 m and downstream water depth h0 ˆ 0:1 m.

Fig. 8. Weighted average ¯ux solution of a anti-symmetric dam break in a frictionless, horizontal domain with upstream
water depth, h1 ˆ 10 m and dry bed downstream.
468 C. Zoppou, S. Roberts / Appl. Math. Modelling 24 (2000) 457±475

Fig. 9. Velocity vectors and contours for the weighted average ¯ux solution of a anti-symmetric dam break in a
frictionless, horizontal domain with upstream water depth, h1 ˆ 10 m and dry downstream.

dam. In the ®nite depth case shown in Figs. 5 and 7, although there is a ®nite water depth, the
¯ow velocity vanishes. In the dry bed case, the water depth is extremely small and not zero. In the
numerical scheme, uh and h are the estimated dependent variable. Machine precision will produce
®nite values for the dependent variable u, which is calculated from u ˆ uh=h, even though to
machine precision h is considered as zero. In this case uh, which is a conserved quantity is a better
quantity to examine. In addition, it is the combination of velocity of ¯ow and its associated water
depth that will cause the greatest damage. Therefore, the conservative parameters are more im-
portant than the primary variables.

5. Treatment of the source term

In ®nite di€erence schemes, the introduction of a source term is straightforward [7,18]. Un-
fortunately, these schemes either introduce excessive di€usion or dispersion in the solution. The
source term in schemes which use some form of limiter or arti®cial viscosity contradicts the basic
philosophy of these limiters. Flux limiter and total variational diminishing-type schemes seek to
prevent an increase in the total variation of the numerical solution of the partial di€erential
equations. However, for partial di€erential equations containing source terms, an increase in the
total variational may be the physical consequence of the source term. In addition, these limiters
were designed for problems with weakly nonlinear or non-sti€ source terms. Algorithms for
treating problems containing nonlinear source terms is a new ®eld of research [28]. A standard
approach, which is also used here, is to solve the homogeneous operator followed by the solution
of the non-homogeneous term. In this way the best schemes can be used to solve each sub-
problem. The approach of separating the homogeneous problem and the source term is known as
Strang splitting or fractional step [22] method. For (1) this involves solving the initial value
problems
oU oF
‡ ˆ0
ot ox
with the initial conditions
U…x; 0† ˆ U0 …x† ) U …x; t†
 C. Zoppou, S. Roberts / Appl. Math. Modelling 24 (2000) 457±475 469

for the homogeneous problem, followed by the solution of the ordinary di€erential equation
dU
ˆ S…U; t† …17†
dt
with the initial conditions
S…0† ˆ U …x; t† ) U…x; t†
for the source term. The splitting can be expressed as
Un‡1 ˆ O…Dt† H…Dt† …Un † …18†
in which H…t† is the homogeneous solution operator and O…t† is the ordinary di€erential equation
operator. Any suitable numerical scheme can be used to solve the homogeneous problem. In
certain problems (17) can be evaluated analytically. In many practical problems this will not be
the case and numerical techniques are required. Care must be exercised in solving the ordinary
di€erential equation. If the ordinary di€erential problems contain processes with widely di€erent
time scales this will lead to a sti€ ordinary di€erential equation. These can be resolved using
implicit or adaptive ordinary di€erential equation solvers.
Eq. (18) is only ®rst-order accurate in time, higher-order scheme can developed. For example,
the splitting scheme
Un‡1 ˆ O…Dt=2† H…Dt† O…Dt=2† …Un †
is second-order in time provided that both the homogeneous and ordinary di€erential equation
operators are solved to at least second-order accuracy.
The ®rst test problem considered, which contains a source term is a radially symmetric dam
break centered in a square domain. This two-dimensional problem reduces to a one-dimensional
problem with a geometric source term [10]
oU oF
‡ ˆ S; …19†
ot or
where
     
h wh ÿwh=r
Uˆ ; Fˆ and S ˆ : …20†
wh w2 h ‡ 12 gh2 ÿw2 h=r
p
Here w ˆ u2 ‡ v2 is the radial velocity, and r is the distance from the center of the domain.
The governing equations are solved using time splitting with the homogeneous equations
solved using the weighted average ¯ux scheme. The non-homogeneous problem
dU
ˆ S…U; t†
dt
can be solved analytically to yield
 w  n
Un‡1 ˆ exp ÿ Dt U
r
n
in which U is the solution to the homogeneous operator.
In the solution of (19), 200 computational cells were used, with each cell having a width of
Dx ˆ 1 m. The initial conditions consist of two states separated by a circular discontinuity. The
radius of the circle r ˆ 100 m and it is centered at x ˆ 100 m. Both components of the velocity u
and v are set to zero everywhere and h is set to 10 m within the circle and 1 m outside the circle. In
the computational scheme, Dt ˆ 0:05 s, which satis®es the Courant criterion, Cr ˆ w Dt=Dx 6 1
470 C. Zoppou, S. Roberts / Appl. Math. Modelling 24 (2000) 457±475

and the solution is sought when t ˆ 2 s. The results for the weighted average ¯ux scheme is il-
lustrated in Fig. 10 using (12) as the limiter.pThe
 water depth, h and the absolute value of the
Froude number, which is de®ned as Fn ˆ w= gh have been plotted in this ®gure.
The solutions are indistinguishable in this example. By time t ˆ 2 s a bore has formed. Fluid
drains from the deepest region as a rarefaction wave progresses outwards. Perspective plots of the
solution of the two-dimensional circular dam break with the downstream water depth set to 1, 5
and 0 m are shown in Figs. 11±13, respectively. In all cases the van Leer type limiter, given by (14)
was used. In these examples the computational grid consists of 100  100 cells with each cell 4 m2 .
When the downstream depth is 1 m there is a transition from subcritical to supercritical ¯ow. This
is not the case when the water depth outside the circle is 5 m deep. In this case subcritical ¯ow
exists throughout the computational domain and it does not represent a severe test for a nu-
merical scheme. The second-order scheme produces sharp resolution of abrupt changes in the
primary variables. In all these pro®les, there are small secondary waves which vanish as the pro®le
evolves. These secondary waves are thought to be due to the inability to represent a circle on a
square grid. Therefore, the initial pro®le contains discontinuities in x±y plane and these discon-
tinuities interact producing small secondary waves. Similar results were obtained by Alcrudo and
Garcia-Navarro [4] and Anastasiou and Chan [5].

Fig. 10. One- and two-dimensional solution of the circular dam break problem in a rectangular domain using the
weighted average ¯ux scheme.

Fig. 11. Two-dimensional solution of the circular dam break problem in a rectangular domain with the downstream
water depth set at 1 m using the weighted average ¯ux scheme.
 C. Zoppou, S. Roberts / Appl. Math. Modelling 24 (2000) 457±475 471

Fig. 12. Two-dimensional solution of the circular dam break problem in a rectangular domain with the downstream
water depth set at 5 m using the weighted average ¯ux scheme.

Fig. 13. Two-dimensional solution of the circular dam break problem in a rectangular domain with a dry bed
downstream using the weighted average ¯ux scheme with the van Leer type limiter.

The in¯uence of the slope limiters, (14), (13) and (12) for the simulation of the dry bed problem
in the second-order scheme can be seen by comparing Figs. 13, 15 and 14, respectively. The
SUPERBEE type limiter produces the sharpest resolution of the rarefaction wave. The minimod
type limiter introduces the greatest di€usion, smoothing the pro®le. The van Leer type limiter is a
compromise between the SUPERBEE and the minimod limiter. It does introduce some di€usion
but it is not as di€usive as the minimod limiter. The sacri®ce made in producing a sharp resolution
of the rarefaction wave using the SUPERBEE limiter is that the numerical schemes produced very
high velocities. To satisfy the Courant criterion, a computational time step of only Dt ˆ 0:005 s
was necessary. For the remaining limiters a computational time step of Dt ˆ 0:01 s was required
to produce a stable solution.
The ®nal example is a sudden collapse of a water supply reservoir, which forms part of the
water reticulation system in many cities. These reservoirs are typically 40 m in diameter and 10 m
high and contain approximately 20 Ml of water. They are usually located in elevated positions in
residential areas. Due to their proximity to residential areas, water supply reservoirs also repre-
sent a potential hazard to property and life. The problem consist of a 50 m diameter circular dam
located at x ˆ 75 and y ˆ 100 m in a valley, 200 m  200 m. Initially the water depth in the
472 C. Zoppou, S. Roberts / Appl. Math. Modelling 24 (2000) 457±475

Fig. 14. Two-dimensional solution of the circular dam break problem in a rectangular domain with the downstream
water depth set at 1 m using the weighted average ¯ux scheme with the SUPERBEE type limiter.

Fig. 15. Two-dimensional solution of the circular dam break problem in a rectangular domain with the downstream
water depth set at 5 m using the weighted average ¯ux scheme with the minimod type limiter.

reservoir has an elevation of 10 m, whereas elsewhere the region is dry. The full two-dimensional
shallow water wave equation given by (1) including bed slope and friction terms are solved. In this
problem the source term only a€ects the momentum equation. Therefore, h does not change and
the source term only a€ects the velocities u and v. The second-order weighted average ¯ux scheme
is used with the second-order Strang splitting to solve the two-dimensional problem. An adaptive
fourth-order Runge±Kutta scheme [29] was used to solve the ordinary di€erential equation op-
erator. The computational domain is subdivided into 1600 computational cells, with each cell
Dx ˆ Dy ˆ 5 m and Dt ˆ 0:1 s is the computational time step used in the model. The valley has a
bed slope in the y-direction, S0y ˆ 0:1 and in the x-direction S0x ˆ ÿx=500 ‡ 0:2. This is a para-
bolic valley with a constant slope in the positive y-direction. The Manning resistance coecient is
equal to 0:1. Re¯ective boundary conditions are located on x ˆ 0 m and x ˆ 200 m with trans-
missive boundary condition along y ˆ 0 m and y ˆ 200 m. The solution to this problem at
t ˆ 1; 2; 3; 5; 10 and 20 s are shown in Fig. 16.
By t ˆ 1 s the water has spread radially to approximately twice its original diameter. By t ˆ 2 s,
the wave has attenuated and the peak water depth commences to move down the slope and water
is still ¯owing up the slope. Most of the water is traveling down the slope by t ˆ 3 s. By t ˆ 5 s, the
peak water depth has now progressed to the center of the valley where the highest velocities occur
 C. Zoppou, S. Roberts / Appl. Math. Modelling 24 (2000) 457±475 473

Fig. 16. Solution of the two-dimensional collapse of a water supply reservoir in a valley using the weighted average ¯ux
scheme.

and the wetted area is still increasing. At t ˆ 10 s the ¯ow is directed down the slope and towards
the center of the valley. The wetted area in now receding. The wetted area has contracted to the
center of the valley by t ˆ 20 s. At the bottom of the valley the water depth has increased and
the ¯ow occurs in the direction of the valley. As time progresses all the water ¯ows out of the
computational domain. For this scheme, an empirical strategy was required to obtain meaningful
474 C. Zoppou, S. Roberts / Appl. Math. Modelling 24 (2000) 457±475

results from the model. The simple strategy involved setting the velocities, u and v and the water
ÿ6
depth h to zero if the water depth falls below 1  10 m. This avoided the generation of negative
water depths and very large velocities by the scheme. With this simple strategy the scheme was
found to be robust and stable. The model is capable of handling steep slopes, the in¯uence of a
friction term and simulating the wetting and drying process.

6. Conclusions

A robust model for the solution of the dam break problem in two dimensions has been de-
scribed. Using fractional splitting, the two-dimensional shallow water wave equations can be
treated as augmented homogeneous one-dimensional problems followed by the solution of an
ordinary di€erential equation describing the e€ect of the source terms in the equation. The one-
dimensional equations are solved using a second-order weighted average ¯ux scheme with the van
Leer type limiter used to eliminate oscillations in the solution. The approximate Riemann solver is
robust and boundary conditions are straightforward to incorporate in the scheme as is the wetting
and drying process. An adaptive fourth-order Runge±Kutta scheme is used to solve the ordinary
di€erential equation. A range of limiters have been tested for the second-order scheme. The van
Leer type limiter is not as di€usive as the minimod type limiter and more di€usive than the
SUPERBEE type limiter. The SUPERBEE limiter based scheme produces very high ¯ow ve-
locities. Therefore, it requires much smaller computational time steps than schemes which employ
the other limiters. Although this scheme produces the most accurate results, the improvement in
accuracy does not justify the increase in computational e€ort. In addition, this scheme did not
seem to be as robust as models which employ other limiters.
Veri®cation of the model is made using a number of hypothetical examples. The model is
shown to be conservative, accurate, stable and robust. It is capable of resolving shocks, handling
complex geometries, including the in¯uence of steep bed slopes and frictional resistance and
simulating the wetting and drying process. The high computational eciency in the simulation
and the results indicate that the weighted average ¯ux scheme should be widely exploited in
computational hydraulics.

Acknowledgements

This project was funded by an Australian Research Council Collaborative Research Grant No.
C19700013, Catastrophic Flooding in Urban Areas. A substantial amount of the work associated
with this study was undertaken whilst the ®rst author was a Visiting Research Fellow in the
Department of Mathematics and Statistics, Australian Defence Force Academy. Their hospitality
is also acknowledged.

References

[1] V.P. Singh, Dam Breach Modelling Technology, Kluwer Academic Publishers, Dordrecht, 1996.
[2] D.L. Fread, The NWS/DAMBRK Model: Theoretical Background/User Documentation, Oce of Hydrology,
National Weather Service, Silver Spring, MD, 1988.
[3] K. Mahmood, V. Yevjevich, Unsteady Flow in Open Channels, Water Resources Publications, Fort Collins, CO,
1975.
[4] F. Alcrudo, P. Garcia-Navarro, A high-resolution Godunov-type scheme in ®nite volumes for the 2D shallow
water equations, International Journal for Numerical Methods in Fluids 16 (6) (1993) 489±505.
 C. Zoppou, S. Roberts / Appl. Math. Modelling 24 (2000) 457±475 475

[5] K. Anastasiou, C.T. Chan, Solution of the 2D shallow water equations using the ®nite volume method on
unstructured triangular meshes, International Journal for Numerical Methods in Fluids 24 (11) (1997) 1225±1245.
[6] P. Glaister, Solution of a two-dimensional dam break problem, International Journal of Engineering Science 29
(11) (1991) 1357±1362.
[7] M. Jin, D.L. Fread, Dynamic ¯ood routing with explicit and implicit numerical solution schemes, American
Society of Civil Engineers, Journal of Hydraulic Engineering 123 (3) (1997) 166±173.
[8] J.V. Soulis, Computation of two-dimensional dam-break ¯ood ¯ows, International Journal for Numerical
Methods in Fluids 14 (6) (1992) 631±664.
[9] D.H. Zhao, H.W. Shen, G.Q. Tabios, J.S. Lai, W.Y. Tan, Finite-volume two-dimensional unsteady-¯ow model
for river basins, American Society of Civil Engineers, Journal of Hydraulic Engineering 120 (7) (1993) 863±883.
[10] S.J. Billett, E.F. Toro, On WAF-type schemes for multidimensional hyperbolic conservation laws, Journal of
Computational Physics 130 (1) (1997) 1±24.
[11] L. Fraccarollo, E.F. Toro, Experimental and numerical assessment of the shallow water model for two-
dimensional dam-break type problems, Journal of Hydraulic Research 33 (6) (1995) 843±864.
[12] E.F. Toro, A weighted average ¯ux method for hyperbolic conservation laws, Proceedings of the Royal Society of
London (A) 432 (1989) 401±418.
[13] E.F. Toro, Riemann problems and the WAF method for solving the two-dimensional shallow water equations,
Philosophical Transactions of the Royal Society of London (A) 338 (1992) 43±68.
[14] E.F. Toro, M. Spruce, W. Speares, Restoration of the contact surface in the HLL-Riemann solver, Shock Waves 4
(1994) 25±34.
[15] T. Weiyan, Shallow Water Hydrodynamics, Elsevier Oceanography Series 55, Elsevier, Amsterdam, 1992.
[16] N. Katapodes, T. Strelko€, Computing two-dimensional dam-break ¯ood waves, American Society of Civil
Engineers, Journal of Hydraulic Engineering HY 104 (4) (1978) 1269±1288.
[17] F. Alcrudo, P. Garcia-Navarro, Computing two dimensional ¯ood propagation with a high resolution extension
of McCormack's method, in: P. Molinaro, L. Natale (Eds.), Proceedings on Modelling of Flood Propagation Over
Initially Dry Areas, American Society of Civil Engineers, Milan, Italy, 1994, pp. 3±17.
[18] R.J. Fennema, M.H. Chaudhry, Explicit methods for 2-D transient free-surface ¯ows, American Society of Civil
Engineers, Journal of Hydraulic Engineering 116 (8) (1990) 1013±1034.
[19] A.K. Jha, J. Akiyama, M. Ura, First- and Second-order ¯ux di€erence splitting schemes for dam-break problem,
American Society of Civil Engineers, Journal of Hydraulic Engineering 121 (12) (1995) 877±884.
[20] A.A. Akanbi, N.D. Katapodes, Model for ¯ood propagation on initially dry land, American Society of Civil
Engineers, Journal of Hydraulic Engineering 114 (7) (1988) 689±706.
[21] R.C. Berger, R.L. Stockstill, Finite-element model for high-velocity channels, American Society of Civil
Engineers, Journal of Hydraulic Engineering 121 (10) (1995) 710±716.
[22] G. Strang, On the construction and comparison of ®nite di€erence schemes, SIAM Journal on Numerical Analysis
5 (3) (1968) 506±517.
[23] C. Hirsch, Numerical Computation of Internal and External Flows, vol. 1, Wiley, New York, 1988.
[24] P.W. McDonald, The computation of transonic ¯ow through two-dimensional gas turbine cascades, Paper 71-GT-
89, American Society of Mechanical Engineers, 1971.
[25] R.W. MacCormack, A.J. Paullay, Computational eciency achieved by time splitting of ®nite di€erence
operators, Paper, American Institute of Aeronautics and Astronautics, 1972, pp. 72±154.
[26] J.J. Stoker, Water Waves, The Mathematical Theory with Applications, Interscience, London, 1957.
[27] C. Hirsch, Numerical Computation of Internal and External Flows, Vol. 2: Computational Methods for Inviscid
and Viscous Flows, Wiley, Chitchester, UK, 1990.
[28] A. Lafon, H.C. Yee, Dynamical approach study of spurious steady-state numerical solutions of nonlinear
di€erential equations. Part IV. Stability vs. methods of discretizing nonlinear source terms in reaction±convection
equation, Computational Fluid Dynamics 6 (1996) 89±123.
[29] W.H. Press, S.A. Teukolsky, W.T. Vetterling, B.P. Flannery, Numerical Recipes in Fortran: The Art of Scienti®c
Computing, second ed., Cambridge University Press, Cambridge, 1992.