You are on page 1of 33

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/268210250

Explicit schemes for dam-break simulations

Article in Journal of Hydraulic Engineering · January 2003

CITATIONS READS

12 1,072

2 authors:

Stephen G Roberts Christopher Zoppou


Australian National University Australian National University
138 PUBLICATIONS 2,423 CITATIONS 74 PUBLICATIONS 2,050 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Stephen G Roberts on 13 November 2014.

The user has requested enhancement of the downloaded file.


Explicit Schemes for Dam-Break Simulations∗
Christopher Zoppou† Stephen Roberts‡

April 30, 2004

Abstract
Dam-break problems involve the formation of shocks and rarefaction fans. The performance
of twenty explicit numerical schemes used to solve the shallow water wave equations for sim-
ulating the dam-break problems is examined. Results from these schemes have been compared
with analytical solutions to the dam-break problem. Most of the numerical schemes produce rea-
sonable results for subcritical flows. Their performance for problems where there is a transition
between subcritical and supercritical flows is mixed. Although many numerical schemes satisfy
the Rankine-Hugoniot condition, some produce solutions which do not satisfy the entropy condi-
tion, producing non-physical solutions. This was the case for the majority of first-order schemes
examined. Numerical schemes which consider sonic points in the solution are guaranteed to
produce entropy satisfying solutions. Second-order schemes avoid the generation of expansive
shocks, however some form of flux or slope limiter must be used to eliminate oscillations that
are associated with these schemes. These limiters increase the complexity and the computational
effort required, however they are generally more accurate than their first-order counterparts. The
limiters employed by these second-order schemes will produce monotone or total variational di-
minishing solutions for scalar equations. Some limiters do not exhibit these properties when they
are applied to the non-linear shallow water wave equations. This comparative study shows that
there are a variety of shock-fitting numerical schemes that are efficient, accurate and robust and
suitable for solving the shallow water wave equations when discontinuities are encountered in the
problem.
2000 Mathematics Subject Classification: primary 76M20; secondary 65M99, 76M12
Keywords: Shallow water waves, dam-break, finite differences, Riemann solvers, conservative
schemes

1 Introduction
Mathematical models based on the shallow water wave equations have become an accepted tool for
modelling environmental problems involving unsteady flows in waterways. The shallow water wave
equations have their origins in the 19th century work of the French mathematician Barrè de Saint
Venant (de St. Venant 1871). Although these equations are simplified descriptions of a complex phe-
nomenon, they incorporate the salient features which influence unsteady fluid motion in open chan-
nels. With the advent of modern computers and the development of efficient numerical techniques,
solutions of these equations are now tractable.

Submitted to ASCE J. Hydraulic Engineering

Principal Research Scientist, CSIRO, Land and Water Technical Specialist, Water Division, ACTEW Corporation. 4
Hammond Close, Oxley, Canberra, ACT, Australia, 2903. Tel: +61 2 2646 5709, Fax: +61 2 6246 5853, Email: christo-
pher.zoppou@cbr.clw.csiro.au.

Senior Lecturer, School of Mathematical Sciences, Australian National University, Canberra, ACT, Australia, 0200.

1
The most challenging feature of the shallow water wave equations is that they admit discontinuities
and smooth solutions. Even the case in which the initial data is smooth can lead to discontinuous
solutions in finite time. The non-linear character of the equations also means that analytical solutions
to these equations are limited to only very special cases. Consequently, numerical methods must be
used to obtain solutions to practical problems which include discontinuities in the solution.
The problems associated with solving the shallow water wave equations with standard techniques
such as the popular Preissmann scheme, have been demonstrated by Meselhe and Holly (1997). A
variety of numerical schemes which exploit the vast knowledge accumulated on the behaviour of
hyperbolic systems, has been developed to sharply resolve discontinuities and produce accurate so-
lutions in smooth regions. These schemes have been developed for general hyperbolic conservation
laws, such as the Euler equations for gas dynamics. More recently, these techniques have been applied
to the solution of the shallow water wave equations.
In this paper twenty explicit conservative schemes are examined for the solution of the one-
dimensional homogeneous shallow water wave equations, where there are discontinuities in the so-
lution. These include naive finite difference schemes, first-order upwind schemes, second-order
schemes which employ flux limiters and Godunov-type schemes which employ exact, first, sec-
ond, third-order approximate Riemann solvers and a smooth particle hydrodynamic solver. All these
schemes have been recently used to solve the shallow water wave equations. The recently developed
biased averaging procedure (Choi and Liu 1989), which has not previously been used to solve the
shallow water wave equations is also included in this review. These models have been ranked, based
on the values reported in Tables 2 and 3 which were obtained from their performance using two hypo-
thetical problems with known analytical solutions. The hypothetical problems, the models and their
performance are described in this paper.
In shock-capturing models, explicit schemes rather than implicit schemes are typically used. Even
though explicit schemes are subject to the usual Courant restriction (Hirsch 1988), there is a substantial
reduction in the computing time required. Conservative schemes are considered because they have
a number of properties essential to obtain physically realistic solutions. One-dimensional numerical
schemes can be employed to solve two-dimensional problems using the fractional step method (Strang
1968). Non-homogeneous equations can be readily accommodated using operator splitting (Strang
1968).
Well known analytical solutions to the dam-break problem were used to assess the performance
of these shock-capturing schemes. Much of the mathematical theory is omitted. However, when
appropriate, sufficient references to applications and theoretical papers have been provided.

2 Two-Dimensional Unsteady Flow


Truly one-dimensional flow does not exist in nature. For example, the interaction between the main
channel and the flood plain flow requires a fully two-dimensional model. Unfortunately, there are
very few truly two-dimensional unsteady flow models (Alcrudo and Garcia-Navarro 1993). Two-
dimensional problems are generally solved on a Cartesian co-ordinate system using fractional step or
space operator splitting (see for example, Glaister 1991, 1993, Jha et al. 2000 and Zoppou and Roberts
2000) or they can be solved using the finite volume method (Hirsch 1988, 1990) on an unstructured
grid (see for example, Alcrudo et al. 1993, Anastasiou and Chan 1997, Bradford and Katapodes 1999,
Causon et al. 1999, Mingham and Causon 1998, Zhao et al. 1994, Zhao et al. 1996 and Zoppou and
Roberts 1999). In both cases the problem reduces to the solution of a local one-dimensional problem
normal to the cell boundary. Flows in open channels, which are generally confined to a well defined

2
channel and are essentially along the length of the channel, are approximated by one-dimensional flow.
In some practical problems, flows could change abruptly in time and space, such as the occurrence
of shock waves when sluice gates are opened suddenly in a dam-break, or the breach of a levee.
Therefore, there is a need for efficient, accurate and robust one-dimensional solvers for the solution
of both one- and two-dimensional problems which can handle smooth and discontinuous solutions.
In two space dimensions, the conservative shallow water wave equations take the form (Weiyan
1992)
ht + (hu)x + (hv)y = 0
 
2 1 2
(uh)t + u h + gh + (uhv)y = −gh(S0x − Sfx )
2 x
 
1
(vh)t + (uvh)x + v 2 h + gh2 = −gh(S0y − Sfy )
2 y

in which h is the water depth, u is the water velocity in the x co-ordinate direction, v is the water
velocity in the y co-ordinate direction, g is the acceleration due to gravity, t is the time, S0x , S0y is
the bed slope in the x and y directions respectively and Sfx , Sfy are respectively the friction slopes in
the x and y Cartesian directions.
For simplicity, consider the application of the finite volume method to a two-dimensional Carte-
sian grid. Utilizing the rotational invariance property of the shallow water wave equations, the two-
dimensional equations reduce to the solution of an augmented one-dimensional problem normal to
each interface between grid cells. The one-dimensional equations solved are

ht + (uh)x = 0 (1)
 
2 1 2
(uh)t + u h + gh = −gh(S0x − Sf ) (2)
2 x
(vh)t + (uvh)x = 0. (3)
The first two equations are just the one-dimensional shallow water wave equations with source terms
and the third equation defines the contact discontinuity. The one-dimensional shallow water wave
equations are de-coupled from the third equation, which is simply an advection equation for vh,
which propagates with speed u.
Using operator splitting, the problem can be treated as two augmented problems involving the
solution of an homogeneous operator and an ordinary differential equation operator sequentially. The
subsequent sections deal with the solution of the homogeneous operator only. Equation (3) can be
solved using a suitable upwinding scheme, for the two-dimensional problem. A number of authors
(for example, Causon et al. 1999 and Molls and Molls 1998) either ignore or neglect the contact
discontinuity. However, this equation is required for improved stability (Le Veque 1998 and Toro et
al. 1994) and to obtain full accuracy in a second- or higher-order scheme (Le Veque 1998).

3 One-Dimensional Unsteady Flow Equations


The governing equations for one-dimensional open channel flow in a wide rectangular channel can be
expressed as
Ut + Fx + S = 0 (4)

3
where      
h uh 0
U= ,F = ,S = .
uh u h + 12 gh2
2 −gh(S0 − Sf )
The friction slope Sf is assumed to be given by the Manning equation

η2u | u |
Sf =
h4/3
where η is the empirical Manning resistance coefficient.
An equivalent but non-divergent form of (4) is

Ut + A(U)Ux = S (5)

where  
0 1
A(U) = 2 (6)
−u + gh 2u

√ vector F, Ai,j = ∂Fi /∂Uj . There are two distinct eigenvalues, λ1 =


is the√Jacobian of the flux
u + gh and λ2 = u − gh, therefore the shallow water equations are hyperbolic and exhibit wave
like behaviour (Smoller 1983).

4 Hypothetical Example
The performance of a numerical scheme for the solution of the shallow water wave equations will be
assessed by considering two dam-break problems in a horizontal, wide and frictionless channel. The
two test cases involve the instantaneous breach of a dam where the flow is initially stagnant either side
of the dam.
The analytical solution consists of a combination of shocks, S and rarefaction waves or fans, R
if the downstream water depth is non-zero, otherwise there is only one rarefaction fan. For example,
in Figure 1 a shock, where λ(Ul ) < λ(Ur ) and rarefaction fan, where λ(Ul ) > λ(Ur ) separate
the left; l, and right states, r. The region between these states is called the intermediate state m.
The state to the right of a discontinuity in the solution, Ur can be connected to state, Ul on the left
of the discontinuity by either a 1-shock, S1 where λ1 (Ul ) < S1 < λ1 (Ur ), a 2-shock, S2 where
λ2 (Ur ) < S2 < λ2 (Ul ), a 1-rarefaction fan, R1 or a 2-rarefaction fan, R2 via the intermediate state
m. Because S1 and R1 represent back waves and S2 and R2 are front waves, a discontinuity in the
solution will decompose into either an S1 or R1 and either an S2 or R2 wave (Smoller 1983).
For an arbitrary flux function F(U) the following relationship between the shock speed, S and
the states Ul and Ur across the shock, called the Rankine-Hugoniot jump condition must be satisfied
(Le Veque 1992)
F(Ul ) − F(Ur ) = S(Ul − Ur ). (7)
The set of solutions which satisfy the Rankine-Hugonoit jump condition is known as weak solutions to
the governing equations. However, solutions to conservation laws, which contain shocks and contact
discontinuities are not uniquely determined by their initial values. The physically relevant solution
from the set of weak solutions must also satisfy the entropy conditions (Smoller 1983)
for the 1-shock
S1 < λ1 (Ul ) λ1 (Ur ) < S1 < λ2 (Ur ), (8)
while for the 2-shock
λ1 (Ur ) < S2 λ1 (Ul ) < S2 < λ2 (Ul ) (9)

4
which are also known as Lax shock inequalities. The entropy condition simply ensures that shocks
are only formed when characteristics converge. Violation of these conditions will result in the non-
physical rarefaction shock or an expansion shock.
The water depth upstream of the dam h1 remains constant at 10m, however the downstream
water depth h0 is either 5m or 0.1m. This problem has a known analytical solution given by Stoker
(1957)(see also Wu et al. 1999). It has been noted by Alcrudo and Garcia-Navarro (1994) that when
h0 = 5m, subcritical flow only exists. This problem does not represent a severe test case and it
may not highlight problems with a numerical scheme, therefore a more severe test is required. The
second example, where the downstream water depth is set at h0 = 0.1m is a severe test case and it
has highlighted problems with a number of numerical schemes. In the two hypothetical case studies,
the computational domain [0, L = 2000m] was subdivided into 200 cells each 10m in length. The
upstream and downstream boundary conditions are set to the initial water depth for all time and the
solution is obtained at time t = 50s using a time step of ∆t = 0.1s.

5 Numerical Schemes
Since the governing equations are written in conservative form, it would seem reasonable that the
numerical scheme should also possess this property. Finite differences written in the form of
∆x n
Un+1
j = Unj − (f n
− fj−1/2 ) (10)
∆t j+1/2
are considered to be in the desirable conservative form because it is a discretized form of the integral
form of the conservation law, which simply states that the rate of change in U in a cell is equal to the
difference in the fluxes f entering the cell. It is the integral form of the governing equations which
n
allows discontinuities in the solution. Methods differ in the way the intercell flux fj±1/2 is estimated,
see Figure 2.

5.1 Naive Finite Difference Schemes


There are numerous finite difference schemes which have been used to solve partial differential equa-
tions. Many of these schemes do not explicitly consider the properties of the governing equations.
These are considered as naive finite difference schemes. A naive approximation of the intercell flux
n
fj+1/2 from the known cell values at time t, uses simple averaging between the nodal flux values,
Fj+1 and Fj so that fj+1/2 = 12 (Fj+1 + Fj ). Unfortunately, this averaging results in an unstable
scheme. Other schemes attempt to remedy this by approximating higher derivatives.
A well known first-order scheme is the Lax-Friedrichs scheme where the flux (Lax 1957)
1 ∆x
fj+1/2 = (Fj+1 + Fj ) − (Uj+1 − Uj ) (11)
2 2∆t
consists of a diffusive term approximating ∆x2 /2/∆tUxx added to the unstable centered flux term.
The solution to the dam-break problem using the Lax-Friedrichs scheme is shown in Figure 4(a). It
is obvious from this figure that this is a highly diffusive scheme. The diffusion is so large that it has
completely dispersed the shock. This scheme is therefore not useful for monitoring the progress of the
shock and rarefaction fans. However, it can be used as the foundation to higher-order schemes which
are described later (Yee 1997). In general, first-order schemes are very diffusive.

5
To obtain a higher resolution of shocks and rarefaction fans, higher-order finite difference schemes
can be employed. If downwind differences are used to approximate Ux this will result in the second-
order Lax-Wendroff scheme with the flux written as (Lax and Wendroff 1960)

n+1 1 n ∆t  n 
Fj+1 + Fnj−1 − Aj+1/2 Fnj+1 − Fnj

fj+1/2 = (12)
2 2∆x
in which Anj+1/2 = (Anj + Anj+1 )/2. Upwind differences result in the Beam-Warming scheme
(Warming and Beam 1976).
The results for the dam-break problem using the Lax-Wendroff scheme is shown in Figure 4(b).
These results are typical of the results that second-order numerical schemes, such as the Beam-
Warming scheme will produce. The dissipative scheme introduces oscillations in the solution. In the
second example, the numerical scheme became unstable because it produced negative water depths.
The second-order MacCormack scheme is a two step predictor-corrector scheme given by (Mc-
Cormack 1969)
∆t
Ūn+1 = Unj − Fnj+1 − Fnj

j
∆x
and for the second step
1 n  ∆t  n+1 
Un+1
j = Uj + Ūn+1
j − F̄ j − F̄ n+1
j−1 .
2 2∆x
It is identical to the Lax-Wendroff scheme for linear problems. The McCormack solution to the
dam-break problem is illustrated in Figure 4(c). These results are similar to those produced by the
Lax-Wendroff scheme. Oscillations occur in the vicinity of abrupt changes in the dependent variables.
This scheme also failed to produce a solution to the second example.
In general, methods based on naive finite-difference approximations may behave well for smooth
solutions but can give disastrous results when discontinuities are present. Naive discretizations of
the governing equations will generally result in either very smeared solutions or oscillations adjacent
to discontinuities. This is because finite-differences produce poor approximations to discontinuous
functions.
To eliminate the oscillations introduced by naive second-order schemes, some form of dissipation
or limiter is used. A number of limiting and dissipation strategies have been developed for the Lax-
Wendroff and McCormack schemes.
Dissipation involves the addition of an artificial viscosity which approximates Ux to the basic flux.
For example, the Lax-Wendroff flux given by (12) is replaced with fj+1/2 = fj+1/2LW − ε
j+1/2 (Uj+1 −
Uj ) where εj+1/2 is at least proportional to ∆x to maintain second-order accuracy (Hirsch 1990).
There are numerous approaches for estimating the artificial viscosity that is added to a numerical
scheme. A number of these are described and demonstrated using the hypothetical examples.
Oscillations in the second-order MacCormack scheme can be reduced by using artificial vis-
cosity or dissipation applied to the conserved variables at the predictor-corrector steps, Un+1 j =
Uj M ac M ac M ac
+ εj+1/2 (Uj+1 − Uj ) − εj−1/2 (Uj M ac M ac
− Uj−1 ). This approach has been used by Fen-
nema and Chaudhry (1986), Fennema and Chaudhry (1990), Gharangik and Chaudhry (1991) and
Rahman and Chaudhry (1998). The diffusion coefficient εj+1/2 is computed from a normalized form
of the gradients of one of the dependent variables and it also depends on an empirical parameter
K, (0.5 ≤ K ≤ 3) which is used to regulate the amount of dissipation (Jameson et al. 1981). Third-
order artificial viscosity can also be used (see, for example Hirsch 1988, p. 279). The unusual feature
of this scheme is that the viscosity is applied after the solution has been calculated.

6
Results for the MacCormack scheme with artificial viscosity are shown in Figure 4(d), in which
εj+1/2 was computed using the dependant variable h. Most of the oscillations observed in Figure
4(c) have been eliminated. This has been achieved with K = 3 which is at the upper limit of the
recommended range for this parameter. Since the viscosity is applied after the solution has been
obtained, this approach is susceptible to instabilities which may produce negative values and the
scheme could fail. This was the situation for the second hypothetical example. The scheme failed
to produce meaningful results. The greatest disadvantage with these schemes is that the dissipation
terms are usually chosen in an ad-hoc fashion.
Alternative strategies have been developed so that only sufficient diffusion is applied to naive
second-order schemes in an attempt to eliminate the numerical oscillations. Central to this strategy
and to the application of flux and slope limiters is how much limiting is necessary to avoid numeri-
cal oscillations in the solution. Slope limiters impose constraints on the gradients of the dependent
variables. Flux limiters impose constraints on the gradients of the flux P functions. A measure of the
oscillations in a solution is provided by the total variation T V (U ) = ∞
n n n
j=−∞ |Uj − Uj−1 | of the
grid values Uj at time level n. Oscillations are avoided if the total variation does not increase with
time. Many higher-order schemes are constructed so that T V (Un ) ≤ T V (Un+1 ). It is possible to use
limiters to produce TVD MacCormack and Lax-Wendroff schemes (see for example Garcia-Navarro
et al. 1992, Jha et al. 2000, Nakatani and Komura 1993, Rider 1993, and Yee 1989).
Davis (1987) describes a strategy for introducing artificial dissipation in a numerical scheme so
that the scheme is total variational diminishing. The result is a dissipation term which does not contain
free parameters. The dissipative terms are added to the solution of the MacCormack scheme so that
the solution is Un+1
j = UM j
ac +(D
j+1/2 −Dj−1/2 ) where Uj
M ac is the MacCormack scheme solution

and Dj±1/2 is artificial dissipation introduced to produce a total variational diminishing scheme. The
nonlinear dissipation term depends on the Courant number and is limited so that oscillations in the
solution are suppressed.
The results from the use of the MacCormack scheme with dissipative terms are shown in Figure
4(e) where the minmod limiter has been used. The oscillations that are observed in the MacCormack
scheme, shown in Figure 4(c) have been eliminated. The scheme has resulted in the excellent res-
olution of the shock and rarefaction fan. For the second example, shown in Figure 5(a), reasonable
results were also obtained using Davis’s total variational diminishing scheme.
The theory which is used to establish a total variational diminishing scheme is based on scalar
equations. Their application to non-linear systems does not ensure that the solution will be total
variational diminishing. This is the case here, where the solution contains some small oscillations.
However, it has produced excellent resolution of the shocks and rarefaction fan and most of the os-
cillations associated with second-order schemes have been suppressed. In addition, this scheme was
able to produce a solution for the second example, which was not the case for the second-order Lax-
Wendroff and the MacCormack naive finite difference scheme, with and without artificial viscosity.
Most of the numerical schemes examined in this paper produce similar results to those shown in
Figure 4(e). This was not necessarily the case for the more severe example shown in Figure 5.
The flux corrected transport scheme originally developed by Boris and Book (1973) and extended
later by Zalasek (1979) is a two-step flux hybrid scheme consisting of a combination of a low-order
total variational diminishing scheme and a higher-order scheme. The objective is to create a higher-
order scheme in smooth regions which reduces to a more robust first-order algorithm near disconti-
nuities. A suitable low-order solution could be obtained by an upwinding scheme. The solution from
the upwind scheme is corrected by adding some anti-diffusion flux. The amount of anti-diffusion is
given by Aj+1/2 = fj+1/2LW − f U P , where, f LW is the flux calculated using a higher-order scheme
j+1/2

7
such as the second-order Lax-Wendroff scheme and f U P is the flux calculated using a first-order up-
wind scheme, such as the Steger and Warming upwind scheme (Steger and Warming 1981). This
flux is limited so that neither new extrema are created nor existing extrema accentuated, producing a
corrected flux Acj+1/2 . This is used to update the conserved quantities calculated using the upwind
scheme, UU j
P using Un UP c c
j+1 = Uj+1 − ∆t/∆x(Aj+1/2 − Aj−1/2 ). In regions where the variables
vary smoothly, the high-order scheme is employed. In regions where the variables vary rapidly, the
low-order scheme is employed. For the test example, the flux corrected transport produced the results
shown in Figure 5(f). There is excellent resolution of the shocks and rarefaction fan by this scheme.
Unfortunately, the limiter employed in this scheme does not suppress all the oscillations associated
with the second-order scheme. In addition, the scheme has produced an entropy violating solution at
the sonic point (Hirsch 1990) which is located at the dam site x = 1000m. There is a discontinuity
in the simulated water depth, h and velocity, u which occurs when critical flow, λ = 0m/s occurs
and there is a transition from subcritical to supercritical flow or visa versa. In the example where
h0 = 5m, flow is always subcritical, sonic points are not encountered and the solution is continuous
along the rarefaction fan. In the second example, where h0 = 0.1m a transition between subcritical
and supercritical flow occurs. In this case the sonic point occurs at the original location of the dam.
Sweby (1984) used the underlying philosophy of the flux-corrected transport scheme of Boris and
Book (1973) to develop a generalization of the total variational diminishing Lax-Wendroff schemes.
It simply adds a dissipation term to the Lax-Wendroff scheme which ensures that the final scheme is
total variational diminishing. Yee (1987) extended the work of Sweby and developed symmetric total
variational diminishing schemes. The flux is given by
 
UP
fj+1/2 = fj+1/2 + φ(β + , β − ) fj+1/2
LW UP
− fj+1/2

UP
where, fj+1/2 LW is the second-order conservative Lax-
is the first-order conservative upwind flux, fj+1/2
Wendroff flux, and φ(β + , β − ) is a limiter function which ensures that the scheme is total variational
diminishing. Examples of total variational diminishing (TVD) limiters include minmod (13), super-
bee (14), van Leer and MUSCL (16) limiters given by Yang and Prezekwas (1992) and Chang (1995)
as
φ(β + , β − ) = minmod(1, β + , β − ) (13)

φ(β + , β − ) = max(0, min(1, 2β − ), min(2, β − )) (14)


+ +
+ max(0, min(1, 2β ), min(2, β )) − 1

β + + |β + |ω β − + |β − |ω
φ(β + , β − ) = + −1 (15)
1 + |β + |ω 1 + |β − |ω
φ(β + , β − ) = max(0, min(2, 2β + , 2β − , (β + + β − )/2)) (16)
in which β + = ∆j−1/2 /∆j+1/2 , β − = ∆j+3/2 /∆j+1/2 , minmod(1, β ± ) = min(1, β ± ) if β ± > 0
otherwise it is equal to 0 and ∆j+1/2 = Uj+1 − Uj . When ω = 1, (15) is the well known van Leer
limiter (van Leer 1977), and when ω = 2 (15) is the van Albada limiter (van Albada 1982) which has
been used by Jha et al. (2000). The ratio β ± is a measure of the smoothness of the data near xj . For
smooth data β ± ≈ 1 and is far from 1 near a discontinuity.
Results for the test problem using the Steger and Warming first-order upwind scheme (Steger and
Warming 1981) are illustrated in Figures 5(b), 5(c), 5(d), and 5(e) using the minmod, monotonic,
Monotonic Upstream Schemes for Conservation Laws (MUSCL) (van Leer 1973) and superbee lim-
iters respectively. For scalar equations, all these limiters add sufficient diffusion in the scheme to

8
produce a monotone preserving scheme. This is not necessarily the case for a non-linear system of
equations. For the shallow water wave equations, the minmod and MUSCL limiters have not intro-
duced sufficient diffusion to completely eliminate the oscillations in the solution. They have however,
produced better resolution of the shocks and rarefaction fan than the schemes which use the monotonic

β + + |β + | β − + |β − |
φ(β + , β − ) = + −2 (17)
1 + β+ 1 + β−
and superbee limiters. In these schemes, the spurious oscillations in the solution have been suppressed
by the additional diffusion they have introduced. The price paid for suppressing oscillations in the
solution, is that these schemes are also diffusive. In addition, the Steger and Warming scheme, like
many other first-order upwind schemes, produce entropy violation solutions at the original location
of the dam. This is demonstrated in the next section. In the symmetric total variational diminishing
scheme, the overall performance of the scheme is very dependent on the performance of the monotone
first-order scheme.

5.2 Upwind Schemes


A great deal is known about the mathematical structure of the governing equations and their solution
(Stoker 1957). One family of schemes which attempt to exploit these properties is upwind schemes.
In these schemes, space discretizations are consistent with the propagation properties of the problem.
The space differencing depends on the sign of the eigenvalues. A number of first-order flux-vector
splitting upwind schemes have been considered. These are by Steger and Warming (1981), Bermudez
and Vazquez (1994), Yang et al. (1993) and Harten (1983).
In the flux-vector splitting method attributed to Steger and Warming (1981), F = F+ + F− . Then
the governing equation becomes Ut + (F+ )x + (F− )x = 0. To explicitly consider the direction of
the flow, (F+ )x is discretized using backward differencing and (F− )x is discretized using forward
differencing so that F− = (F − |F|) /2 and F+ = (F + |F|) /2. The intercell flux becomes
1 1 
fj+1/2 = (Fj+1 + Fj ) + |F|j+1 − |F|j . (18)
2 2
Assuming that F = EU (Steger and Warming 1981), then |F| = |E|U, where |E| = M|Λ|M−1
with    
1 1 −1 u+c 0
M= and Λ = .
2c u + c −u + c 0 u−c
The results using this scheme are illustrated in Figure 5(g) and are typical of those produced by first-
order upwind schemes, producing an entropy violating solution at the sonic point. This is due to the
inability of upwind schemes in defining the upwind direction. At a sonic point the eigenvalues of the
Jacobian vanish.
Equation (18) assumes that F = EU. This is valid if F(U) is a homogeneous function of degree
one. The shallow water wave equations is a homogeneous function of order two, so that F 6= EU. The
form of flux splitting used by Steger and Warming is not appropriate here (see, for example Hirsch
1988, p. 417). It is possible to construct a matrix, G that does satisfy F = GU (Bermudez and
Vazquez 1994). For (4)
   
0 1 1 √ 1 √
G = PΛP−1 = , P =
−u2 + gh/2 2u u + c/ 2 u − c/ 2

9
and  √ 
u + c/ 2 0 √
Λ=
0 u − c/ 2
which are used in (18) to estimate |F| = |G|U.
Results for this scheme are illustrated in Figure 5(h). Similar results were obtained by Bermudez
and Vazquez (1992). This scheme has also produced an entropy violating solution at the sonic point,
that is, at the transition from subcritical to supercritical flows. The error is more dramatic than the
results shown in Figure 5(g). This is due to greater diffusion introduced by the Steger and Warming
scheme, than by the Bermudez and Vazquez scheme. There is a fix for the entropy violating solution.
The modification involves adding a small value to the eigenvalues (Buning and Steger 1982). This is
reported to provide a smooth transition at the sonic point at the expense of introducing a small error.
Alternatively, the flux can be decomposed into two components such that (Yang et al. 1993)

Ut + A+ + A− Fx = 0


where A+ + A− = I. With the intercell flux given by

fj+1/2 = A+ −
j−1/2 Fj+1 + Aj+1/2 Fj

where A± = MΛ± M−1 , Λ± = diag(λ± ±


i ) and λi = (1 ± sgn(λi ))/2 for i = 1, 2. This scheme has
been used recently by Jin and Fread (1997). The solution of the test problems using this scheme is
illustrated in Figure 5(i). This scheme also produces an entropy violating solution.
Harten (1983) used the mean-value theorem to estimate the Jacobian (6)

F(Unj−1 ) − F(Unj ) n
= A(U), Unj−1 < Uj < Unj (19)
Unj−1 − Unj

which satisfies the Rankine-Hugoniot condition, to establish whether the characteristics emanate from
the left or the right. Defining an indicator function δ(Uj−1 , Uj ) = 1 if A(U) ≥ 0, otherwise it is
equal to 0. Then the upwind intercell flux is given by
 
fj+1/2 = Fj + 1 − δj−1/2 (Fj − Fj−1 ) .

Results obtained from this scheme are illustrated in Figure 4(f). Although this scheme has produced
reasonable results for the subcritical flow problem, it became unstable for the second example and is
therefore not robust.

5.3 Godunov-Type Schemes


An alternative approach to solving the shallow water wave equations for problems containing dis-
continuous solutions, is by solving the Riemann problem in a Godunov-type scheme. The Riemann
problem for a general hyperbolic system is the following initial-value problem

Ut + F(U)x = 0

with 
Ul x < 0
U(x, 0) =
Ur x > 0.

10
In the traditional
 Godunov scheme, the numerical solution is considered piecewise constant in each
mesh cell xj−1/2 , xj+1/2 . The cell interface at xj+1/2 separates two constant states Ul and Ur
which form the Riemann problem. The steps involved in a Godunov-type scheme are; (i) aver-
aging, (ii) solving the Riemann problem and (iii) re-averaging to obtain the nodal dependent vari-
ables. For each cell interface xj+1/2 Ul and Ur are estimated from the reconstruction polynomial
Uj (x) = Pj (x). The polynomial used to estimate the states Ul and Ur determine the accuracy of
the scheme. For Godunov’s first-order method, the profile is approximated by piecewise constant
Pj (x) = Uj , therefore Ul = Uj and Ur = Uj+1 . The intercell flux fj+1/2 is calculated using
an exact or approximate solution to the Riemann problem given the initial states Ul and Ur . The
intercell flux is then used to update the conserved quantities at the nodes. A second-order scheme is
produced if linear interpolation Pj (x) = Uj + sj (x − xj ) is used to estimate the left and right states.
Other well-known schemes using higher-order reconstruction of the solution are van Leer’s MUSCL,
third-order accuracy Piecewise Parabolic Method (PPM) Collela and Woodward (1984) and arbitrary
high-order Essentially Non-Oscillatory (ENO) (Harten 1989) schemes. Ul and Ur at the cell interface
are limited in all the higher-order schemes to avoid oscillations in the solution near discontinuities.
With these values, the Riemann problem is solved using an exact or approximate first-order Riemann
solver. The gain in accuracy is achieved by using higher-order averaging to estimate the left and right
states in the Riemann problem.

5.3.1 Exact Riemann Solver


Equations (7), (8), (9) and the Riemann invariants
p p p p
um + 2 ghm = ul + 2 ghl and um − 2 ghm = ur − 2 ghr (20)
provide the necessary conditions to solve all possible Riemann problems that could be encountered
for the shallow water wave equations. There are 16 possible combinations to connect state Ul with
state Ur and therefore to calculate the intercell flux. Most of these will involve the solution of a
non-linear equation. The intercell flux is used in (10), which is the re-averaging step to advance the
solution in time. The results of the exact Riemann solver for the test problems are shown in Figure
6(a). Although an exact Riemann solver is used, the numerical scheme still produces a discontinuity
in the solution at the original location of the dam. At this point entropy is not satisfied.

5.3.2 Approximate Riemann Solvers


Due to the complexity of the exact Riemann solver, there have been attempts to develop approxi-
mate Riemann solvers. The first-order approximate Riemann solvers attributed to Roe (1981), which
have been used recently by Jha et al. (1995), Osher’s scheme (Engquist and Osher 1981 and Osher
and Solomon 1982) and the Harten, Lax and van Leer (HLL) (Harten et al. 1983) scheme as used
by Fraccarollo and Toro’s (1995) are described. The second-order Weighted Average Flux (WAF)
(Fraccarollo and Toro 1995) and the third-order Piecewise Parabolic Method are also evaluated.
Roe’s approach replaces the Jacobian in (5) by a constant Jacobian matrix A=A (Ur , Ul ) which
satisfies the Rankine-Hugoniot condition
A (Ur , Ul ) (Ur − Ul ) = F(Ur ) − F(Ul ) (21)
for any finite value of Ur − Ul . This results in the following constant Jacobian matrix
 
0 1
A (Ur , Ul ) =
c2 − u2 2u

11
in which
√ √ s  
hl ul + hr ur hl + hr p
u= √ √ ,c = g and h = hl hr .
hl + hr 2
This is by far the most common method of approximating the Jacobian. The eigenvalues and right
eigenvectors of the Jacobian A (Ur , Ul ) are given by

λ1 = u − c, λ1 = u − c, e1 = [1, u − c]T and e2 = [1, u + c]T .

The structure of the solution of the Riemann problem consists of two shocks, one for each eigenvalue
λ1 and λ2 . Since the eigenvectors are independent, then
2
X
Ur − Ul = αi ei (22)
i=1

which is solved for the wave strengths αi , to give the following O(∆2 ) approximation
   
1 ur − ul 1 ur − ul
α1 = hr − hl − h and α2 = hr − hl + h .
2 c 2 c

Substituting (22) into (21) gives (Roe 1986)


2
X
F(Ur ) − F(Ul ) = λi α i e i . (23)
i=1

Equation (23) can be applied across negative or positive waves to calculate the intercell flux. For the
negative waves X
fj+1/2 = F(Ul ) + λi αi ei , (24)
λi ≤0
P
where λi ≤0 denotes the sum of shocks with eigenvalues less than zero. This expression can be used
in (10) to update Uj .
Results for the hypothetical example using this scheme are illustrated in Figure 6(b).
A non-physical stationary shock, in the region of the rarefaction fan has been produced by this
scheme. Roe’s approximate Riemann solver can be modified so as to avoid entropy violating solution.
Entropy fixes can be found in Harten and Hyman (1983) and Roe and Pike (1984). This fix was
successfully employed by Jha et al. (2000) in their implementation of Roe’s method for estimating
the interface flux.
Osher’s approach is based on a flux splitting approach (Osher and Solomon 1982), where the
intercell flux can be expressed as the one-sided flux
Z Ur
fj+1/2 = F(Ul ) + A− (U)dU. (25)
Ul

This approach is similar to the intercell fluxes estimated in the Roe scheme given by (24). It differs
however, in the integration path that is used. It is assumed that the integration path consists of two
sub-curves, Il and Ir which pass through sonic points Sl where λl = 0m/s and Sr where λr = 0m/s.
These integration paths are curves associated with the set of right eigenvectors (Toro 1997).

12
All physically realistic estimates for Osher’s intercell flux, fj+1/2 for the one-dimensional shallow
water wave equation using P -ordering, are given in Table 1 where FSl and FSr are the fluxes at the
sonic points Sl and Sr respectively.
The quantity, Um = (hm , um )T is obtained using the Riemann invariants (20) which can be
solved for p p
um = (ul + ur )/2 + ghl − ghr (26)
and √ √ 2
ul + 2 ghl − ur + 2 ghr
hm = . (27)
16g
The Riemann invariants, (20) are also used to obtain a solution at the sonic points. In Osher’s
scheme, sonic points Sl , and Sr are explicitly used in the approximate Riemann solver and therefore
this scheme excludes expansion shocks. Osher’s P scheme has been used recently by Zhao et al.
(1994, 1996) to solve the two-dimensional shallow water wave equations.
The results for the test case using Osher’s physical ordering or P scheme is shown in Figure 6(c).
Although this first-order scheme has introduced some diffusion it has produced better results than any
of the other first-order schemes. Despite the similarity to Roe’s approximate Riemann solver, this
scheme has overcome the entropy violating solution.
Total variational diminishing schemes have at best, first-order accuracy at extrema of the solution
and genuine extrema can only decrease with time. There is some question as to whether this is desir-
able for conservative laws which contain dispersive or source terms. To remedy this and to produce
globally higher-order accuracy, Harten and Osher (1987) developed the essentially non-oscillatory
schemes (ENO) which allow the loss of amplitude at one time step to be regained at another (Harten
et al. 1986).
In the ENO scheme described by Nujic (1995) and used by Singh and Bhallamudi (1998), the
numerical flux fj+1/2 is approximated up to the required accuracy using a moving stencil. Second-
order accurate numerical fluxes are obtained for fj+1/2 in the Lax-Friedrichs type scheme using

1 α
fj+1/2 = (Fl + Fr ) − (Ur − Ul ). (28)
2 2
MUSCL is used to obtain second-order accurate estimates of the states on the left and right of the cell
interface, see Figure 3 so that
δUj δUj+1
Ul = Uj + and Ur = Uj+1 +
2 2
where the quantity δUj is used to limit Uj to avoid the introduction of spurious oscillations in the
numerical solution. In addition, the coefficient

α ≥ max |λi |, i = 1, 2 and j = 1, ..., N


j

where N is the number of computational nodes in the problem. This is no more than satisfying the
Courant condition.
Results for the two hypothetical problems are shown in Figures 4(h) and 6(h) using the minmod
limiter. The scheme is slightly diffusive, however it does not produce oscillations in the solution,
nor does it produce an entropy violating solution. The Lax-Friedrichs implementation of the ENO
scheme is accurate, robust and very simple to implement. Although higher-order ENO schemes can
be developed (see for example Harten 1989, Shu and Osher 1988, Shu and Osher 1989 and Yang et al.

13
1993), these schemes require larger computational stencils. This may not pose a serious problem for
problems solved on Cartesian grids. It does pose a problem for unstructured grids, where it is difficult
to develop higher-order averaging along each cell interface.
Causon et al. (1999) use the van Leer limiter, MUSCL averaging with the first-order approxi-
mate Riemann solver described by Toro (1992). The first-order HLL scheme is based on the scheme
described by Harten et al. (1983). The intercell flux, fj+1/2 is evaluated by solving the approximate
Riemann problem which involves only two shocks, although two rarefaction fans could be considered,
separating three states; l, r and m. The intercell flux is chosen from

 F(Ul ) if S1 ≥ 0
fj+1/2 = F(Um ) if S1 < 0 < S2 (29)
F(Ur ) if S2 ≤ 0.

For the shallow water wave equations, the intermediate state variables, um and hm are given by (26)
and (27). These are used to estimate the shock speeds
 p p 
S1 = min ul − ghl , um − ghm and (30)
 p p 
S2 = max ur + ghr , um + ghm (31)

which are estimates of the largest and smallest propagation speed and satisfy the entropy condition.
Although this may overestimate the true wave speed, Fraccarollo and Toro (1995) suggest that this en-
hances the stability of the scheme. These wave speeds are used to solve for the flux in the intermediate
state
S2 F(Ul ) − S1 F(Ur ) + S1 S2 (Ur − Ul )
F(Um ) = (32)
S2 − S1
by satisfying the integral form of the conservation law over a control volume. This expression is also
valid for S2 ≤ 0 and S1 ≥ 0. This is all the information required to estimate the intercell flux, fj+1/2
from (29).
For the test problem, the results from this first-order approximate Riemann solver is shown in
Figure 6(d). There is reasonable resolution of the shock and rarefaction fans. More importantly, the
problems with the inability to satisfy the entropy condition at the dam, common to many of the other
schemes, has almost been eliminated in this scheme. This first-order scheme has been used by Ming-
ham and Causon (1998), Causon et al. (1999) and Zoppou and Roberts (1999) as the approximate
Riemann solver in the simulation of two-dimensional flow using the shallow water wave equations.
The Piecewise Parabolic Method is a higher-order extension of Godunov’s method. Quadrtic
curves are used to interpolate the dependent variables on the edges of piecewise cells. In the averaging
step, parabolic interpolation is used to obtain left and right states Ul and Ur in the Riemann problem.
These values are constrained so that new extrema are not produced. The domain of dependence for
the cell interface is determined and the piecewise monotone parabolic function within the domain of
dependence is integrated. The left and right states at a cell interface are simply the averages of the
integrated quantities over the domain of dependence. The Riemann problem is solved using the left
and right states at a cell interface. If the cells are equally spaced, then the scheme is fourth-order
accurate, otherwise it is third-order. The Piecewise Parabolic Method has been used by Savic and
Holly (1993).
Results using the Piecewise Parabolic Method for the test problem are shown in Figure 6(f). In this
problem, the first-order HLL approximate Riemann solver described by Fraccarollo and Toro (1995),

14
(26)-(32) was used to solve the Riemann problem. There is excellent resolution of both the rarefaction
fan and the shock, with the shock resolved over the fewest computational nodes.
Choi and Liu (1998) use the Biased Averaging Procedure (BAP), which play a similar role to flux
limiters to limit a second-order approximation of the upwind fluxes in the Steger and Warming flux
splitting scheme. The slope sj in the linear reconstruction polynomial for the backward and forward
fluxes F± ±
j (x) = F (Uj ) + sj (x − xj ), x ∈ (xj−1/2 , xj+1/2 ) is estimated using
 
−1 B(sl ) + B(sr )
sj = B (33)
2
in which, sl = (F+ (Uj ) − F+ (Uj−1 ))/∆x and sr = (F+ (Uj+1 ) − F+ (Uj ))/∆x are evaluated at
xj and the negative flux F− , is approximated at xj+1 . The biased average (33) plays a similar role to
limiters, preventing possible oscillations near discontinuities. The intercell flux is computed from
fj+1/2 = F+ −
j (x)|xj+1/2 + Fj+1 (x)|xj+1/2 .

Although Choi and Liu (1998) provide four examples of the biased function √ for the shallow water
2 −1
√ equations, the only successful biased function is given by B(x) = x/ 1 + x and B (x) =
wave
x/ 1 − x2 . The results shown in Figure 4(i) and 6(i) for this explicit scheme were produced using
+
the Lax-Fredrichs scheme, (28) to estimate the fluxes F± j where Fj = (fj + αUj )/2 and Fj =

(fj − αUj )/2.


These results suggest that the biased function has not eliminated all the oscillations that are com-
monly associated with second-order schemes. Oscillations are still present at abrupt changes in the
dependent variables. These oscillations are more pronounced for u than h. In the more severe prob-
lem, very large negative values of velocity were obtained. In some instances, negative simulated
values may affect the stability of the scheme. Therefore, this scheme is not very robust. However,
this is the first attempt to use this scheme to solve the shallow water wave equations. The selection of
alternative biased functions may eliminate these problems. If this can be achieved, it may prove to be
a very accurate and efficient scheme.
Instead of using higher-order averaging to improve the accuracy of Godunov-type schemes, Toro
(1989) and Billett and Toro (1997) describe a second-order approximate Riemann solver, called the
weighted average flux scheme. In this approach, the flux, fj+1/2 is given by
2
F(Ul ) + F(Ur ) 1 X (k) (k) (k)
fj+1/2 = − Aj+1/2 νj+1/2 ∆f (Uj+1/2 )
2 2
k=1

(k) (k+1) (k)


in which ∆f (Uj+1/2 ) = f (Uj+1/2 ) − f (Uj+1/2 ) represents the change in flux across each shock,
(k) (k)
νj+1/2 = ∆tSk /∆x, Aj+1/2 is a flux limiter and k is the shock with speed Sk . The flux in the
intermediate state, m is given by (32) with their shock speeds S1 and S2 estimated using (30) and (31)
respectively. The dependent variables, um and hm are approximated using (26) and (27). The flux
(k)
limiter, Aj+1/2 is defined for each wave. Appropriate flux limiters are given by Fraccarollo and Toro
(1995). This scheme has been used by Fraccarollo and Toro (1995) and Zoppou and Roberts (2000)
to solve the shallow water wave equations.
The use of this scheme with a superbee type limiter for the test problem produced the results shown
in Figure 6(e). There is excellent resolution of both the shock and the rarefaction fan. The problem
with the non-physical solution at the dam has been eliminated in this scheme. It is not surprising
that this scheme produces excellent resolution of the rarefaction fan. In this scheme, the intermediate
states are established by assuming that the Riemann problem consists of two rarefaction fans.

15
5.4 Other Schemes
There are numerous schemes that could be employed to solve the shallow water wave equations. Two
schemes which have been recently used to solve the dam-break problem have also been included in
this review. These are the Space-time conservative scheme developed by Chang (1995) and used by
Molls and Molls (1998) and the Lagrangian Discrete Parcel Method used by Wang and Shen (1999).
Molls and Molls (1998) have adapted the numerical scheme developed by Chang (1995) for the
solution of the shallow water wave equations. The explicit scheme is conservative in both space and
time on a staggered grid. It differs from other schemes which treat the spatial and temporal operators
separately. By defining conservation elements wisely and considering the spatial derivatives of the
dependent variables as dependent variables, flux evaluation at a cell interface is carried out without
averaging and re-averaging. The functions U(x, t) and F(x, t) are assumed to be approximated by
first-order Taylor series expansion in x and t

f ∗ = fjn + (fx )nj (x − xj ) + (ft )nj (t − tn ) + O(∆x2 , ∆t2 ).

It is assumed that U∗ and F∗ satisfy the original governingH equations. The flux over non-overlapping
elements must also satisfy the integral conservation law −F∗ dt + U∗ dx = 0. Substituting for U∗
and F∗ , integrating and simplifying results in the following discrete form of the approximation
1  n+1/2 n+1/2 n+1/2 n+1/2

Un+1
j = Uj−1/2 + Uj+1/2 + Fj−1/2 − Fj+1/2
2
in which
n+1/2 ∆x n+1/2 ∆t n+1/2 ∆t2 n+1/2
Fj±1/2 = (Ux )j±1/2 + Fj±1/2 + (Ft )j±1/2
4 ∆x 4∆x
and  2
∂F ∂U
Ft = − = −A2 (Ux ).
∂U ∂x
This is the first step in the marching scheme. The final step involves estimating Ux , using central
differencing
n+1
Un+1 n+1
j+1/2 − Uj−1/2
(Ux )j =
2
where Un+1j±1/2 are approximated using the Taylor series over the time step (n + 1/2)∆t to n∆t. Un-
fortunately, central differencing will produce oscillations in the solution near a discontinuity. To
overcome this limitation Chang (1995) replaced the central differencing by a weighted average of
forward and backward differencing given by (15) in which β ± = ±2(Uj±1/2 − Uj )/∆x. This is
equivalent to slope-limiting U.
The use of the space-time scheme with ω = 1.0 is shown in Figures 4(g) and 6(g). This is one
of the most accurate and robust schemes examined, providing excellent resolution of the shock and
rarefaction fan. Like other second and higher-order schemes, the space-time conservation scheme also
produces some small oscillations near the shock.
The discrete parcel method is a Lagrangian scheme based on the Smoothed Particle Hydrodynamic
scheme (Monaghan 1985) developed for astrophysical fluid dynamics. In this method a continuum
can be represented by a large number of parcels that carry mass, momentum and energy. The ma-
jor advantage of adopting a Lagrangian framework is that the advective derivative in the continuity

16
equation is eliminated. Assuming that the water depth is approximated by a kernel function, the water
depth of parcel j is given by
j+p
X
hn+1
j = hnj + Vk (xj − xk ) [W (xj − xk , lj ) + W (xj − xk , lk )]
k=j−p+1

and the velocity of flow is evaluated using


j+p
X
un+1 = unj + g∆t Vk (xj − xk ) W (xj − xk , lj )/lj2 + W (xj − xk , lk )/lk2
 
j
k=j−p+1

in which W is a normalized interpolating kernel function, 2p is the number of parcels over which the
interpolating kernel is applied, Vj is the volume of parcel j, lj = l0 hj /h0 is a smoothing length and
l0 and h0 are the initial smoothing length and water depth respectively. The location of the parcel of
water is given by xn+1
j = xnj +uj ∆t where u = (un+1 +un )/2. The symmetric Gaussian interpolating
kernel function
1
W (xk − xj , l) = √ exp(−((xk − xj )/l)2 )
l π
ensures that momentum is conserved (Wang and Shen 1999). The unusual feature of this scheme is
that the length of the discrete parcels of water varies with changing flow conditions. The discrete
parcel method has been recently used to solve the one-dimensional dam-break problem by Wang and
Shen (1999). They incorporate a viscosity term to eliminate oscillations in the solution. It was found
that the viscosity term had little influence on the solution and has been omitted here.
For the two hypothetical problems, l0 = 50m, h0 = 10m and p = 15. These values were obtained
by trial and error by minimizing the difference between the analytical and simulated solution for the
second hypothetical example. These values are not selected automatically and are not dependent on
the behaviour of the transient being modelled. Using these values for the simulation of the dam-break
problem produced the results shown in Figures 5(j) and 6(j). This scheme is very diffusive and is
very sensitive to the selection of the initial smoothing length, water depth and interpolating window,
2m. The results indicate that this approach is reasonable in resolving the shock, but is has introduced
significant phase errors in the simulation of the rarefaction fan. Considering all these factors, this
scheme is not robust.

6 Comparison of Schemes
To quantify the performance of the various schemes for solving the shallow water wave equations, the
normalized L1 norm between the exact and simulated water depth and flow velocity were calculated
for the two hypothetical examples. The normalized L1 norm is defined as
Pk
j=1 |cj − C(xj )|
L1 = Pk
j=1 |C(xj )|

in which the first moment of the approximate solution, cj , evaluated using the solution obtained at
the end of the computational period, is calculated using all the computational nodes in the domain,
j = 1, ..., k. The first moment is normalized by the corresponding first moment of the exact solution,
C(xj ). The relative computational effort for each scheme was also established. These values are

17
shown in Tables 2 and 3. The L1 norm and the computational time has been standardized using the
Lax-Friedrichs scheme results. The smaller the L1 norm, the more accurate the scheme. The greater
the relative computational time, the greater the computer resources required. The L1 norm for the Lax-
Friedrichs scheme for the water depth and velocity is 0.917272 × 10−1 and 0.423581 respectively for
the dam-break problem where h0 = 0.1m. The L1 norm for the Lax-Friedrichs scheme for the water
depth and velocity is 0.489452 × 10−1 and 0.360070 respectively for the dam-break problem where
h0 = 5m.
Although the first-order Lax Friedrichs scheme is the most efficient scheme, it is not very useful
for tracking the progress of a shock, because it introduces excessive diffusion smothering the shock
and rarefaction fan. In the first-order flux-vector splitting schemes, calculating |F| in the Bermudez
and Vazquez scheme, is more expensive than calculating |F| in the Steger and Warming scheme.
Ensuring that F = GU has not significantly affected the accuracy of the scheme, but it has increased
the computational effort required. Yang’s scheme is the most accurate first-order upwind scheme and
computationally more expensive than the Steger and Warming scheme and less expensive than the
Bermudez and Valduez. The Harten estimate of the Jacobian is very efficient but this scheme is not
robust, failing for the h0 = 0.1m test case. Although the MacCormack scheme is a two-step scheme,
it is very efficient. Unfortunately, it is not very robust, also failing for the h0 = 0.1m problem. The
first-order approximate Riemann solvers of Roe, Osher and the HLL scheme are very efficient. Roe’s
approximate Riemann solver produces an entropy violating solution which makes this scheme one of
the least accurate. Fortunately, there is a fix for this problem. In contrast, Osher’s P scheme and the
HLL scheme are robust and consistently produce accurate results for both primary variables, h and u.
Second-order naive finite difference schemes are efficient and accurate for subcritical flows, but
failed to produce solutions for the more severe problem. These schemes are not suitable for simulating
the dam-break problem. The addition of artificial viscosity improves the accuracy of the MacCormack
scheme. However, since the viscosity is applied after the solution has been calculated, the artificial
viscosity does not improve the stability properties of the scheme. The use of total variational dimin-
ishing flux limiters in the MacCormack scheme improved the stability properties of the scheme as
well as its accuracy. A stable and robust scheme comes at a price, an increase in computational effort.
This is also the case with the use of flux limiters in the Lax-Wendroff scheme. The Lax-Wendroff
scheme, incorporating some form of flux limiter, has produced a more robust scheme. The minmod
and MUSCL limiters produce similar results which are more accurate than the results produced with
the use of the monotone and superbee limiters. Schemes which use some form of limiter will gen-
erally produce a more robust and accurate scheme. They also avoid the entropy violating solution
common to first-order schemes. However, the computational effort required by these schemes will
be approximately twice that required by the underlying scheme alone. The second-order approximate
Riemann solver of Toro (WAF) is robust and consistently produces very accurate results. The flux
corrected transport scheme is computationally expensive and does not provide any additional benefit
over other second-order schemes examined which also employ a flux or slope limiter. The piecewise
parabolic method is the most accurate scheme and avoids the entropy violating solution, but it is not
as efficient as either the first- or second-order approximate Riemann solvers. The computational effort
required by this scheme and the exact Riemann solver would exclude the use of these schemes in two-
dimensional models. In addition when these schemes are compared with other schemes examined,
they are complicated to implement.
With the exception of the very diffusive Lax-Freidrichs scheme, the discrete parcel method is the
least accurate scheme. Due to the integrating window, which is executed for every computational
node, this is a very expensive scheme. There are a number of parameters that must be estimated in
the discrete parcel method. The performance of this scheme is dependent on these parameters which

18
are not automatically selected or which depend on the behaviour of the problem. These observations
are consistent with the conclusions found by Monaghan (1997). In summary, there are more suitable
schemes for solving the shallow water wave equations.
The space-time conservative scheme is very accurate, efficient and simple to implement. Unfor-
tunately, its performance for the subcritical problem is not as impressive as its performance for the
second problem. This inconsistency in accuracy requires further investigation. Otherwise, this scheme
is a viable alternative to second-order approximate Riemann solvers. The biased averaging procedure
is also a very accurate scheme but not as efficient as other second-order schemes. Negative values were
produced near abrupt changes in the dependent variables, which may result in an unstable scheme.
There is also a problem with choosing an appropriate biased function. Only one of the four biased
functions provided by Choi and Liu (1998) provided meaningful results when it was used to solve the
shallow water wave equations. This scheme could be a competitive scheme if alternative more robust
biased functions could be found. The ENO is not as efficient as other second-order schemes, however
it was the simplest scheme to implement and it is robust. A major advantage with ENO schemes is
that it is relatively straight forward to extend the second-order scheme to higher-order accuracy by
increasing the size of the computational stencil.

7 Conclusions and Recommendations


A number of numerical schemes for solving the one-dimensional shallow water wave equations ap-
plied to problems containing discontinuities in the solution have been examined. These include naive
finite difference schemes, first-order upwind schemes, second-order schemes which employ flux lim-
iters and Godunov-type schemes which employ exact, first, second, third-order approximate Riemann
solvers and a smooth particle hydrodynamic solver. The recently developed biased averaging proce-
dure, which has not previously been used to solve the shallow water wave equations is also included
in this review.
With the exception of naive second-order schemes, all schemes produce reasonable results for
subcritical flows. However, for supercritical flows, many of the first-order schemes produce non
physical entropy violating solutions. There are fixes for entropy violating schemes. Yang et al. (1993)
conservative first-order upwind scheme produces the best results in comparison to the other naive
first-order finite difference schemes. In general, second-order schemes avoid the entropy violating
solutions associated with first-order schemes. However, they introduce oscillations in the solution.
For linear equations, these oscillations can be suppressed using some form of slope or flux limiter.
Unfortunately, this is not generally the case for the non-linear shallow water wave equation. The
superbee is the most comprehensive limiter while the minmod limiter is the least compressive or
the most diffusive. The total variational diminishing second-order scheme due to Davis is the best
second-order finite difference scheme.
Physically based approximate Riemann solvers perform better than finite difference schemes. Ap-
proximate Riemann solvers are more accurate and robust than finite difference schemes. The only
exception to this is Roe’s approximate Riemann solver. It produces solutions that are similar to those
produced by first-order upwind schemes. The solution contains an entropy violating shock. The ap-
proximate Riemann solvers of Osher and the HLL scheme produce similar solutions. These first-order
schemes are robust and produce excellent results for a wide range of flow conditions and avoid the
entropy violating solution that plagues other first-order schemes. They also explicitly include the con-
tact discontinuity which occurs in the one-dimensional subset of the two-dimensional problem, which
is ignored in other Riemann solvers. However, the HLL scheme is much simpler to implement. More

19
research into biased functions suitable for solving the shallow water wave equations is required before
the BAP scheme is competitive with other second-order schemes such as the ENO scheme. The ENO
is not as accurate as some of the other second-order schemes, however it is robust and very simple to
implement.
First-order schemes are generally one-half as computationally expensive as second-order schemes
but only half as accurate. However, efficient first-order schemes can be used to produce higher-order
accurate solutions. The third-order PPM is based on a first-order Riemann solver which has been
shown to be accurate at a price. Alternatively, higher-order native schemes can be developed. The
WAF, space-time and DPM are examples of higher-order native schemes. The second-order WAF
approximate Riemann solver is accurate, robust and efficient. It avoids the entropy violating solution
and includes the contact discontinuity. The DMP scheme is a poor choice for the solution of the
shallow water wave equation. It is very inefficient and dependent on arbitrarily selected parameters.
The exact Riemann solver is also very inefficient.
In summary, the most accurate scheme is the third-order piecewise parabolic method, followed by
the WAF, BAP, ENO, space time, Davis TVD scheme and the second-order schemes of Yee. For ease
of implementation, efficiency and robustness, the HLL or Osher’s P scheme are the recommended
first-order schemes. Suggestions for second-order schemes, in order of preference, are the WAF
scheme space-time scheme or the use of MUSCL averaging with an efficient first-order scheme. The
flexibility of the arbitrary order ENO scheme should not be overlooked.

8 References
Alcrudo, F., and Garcia-Navarro, P. (1993). “A high-resolution Godunov-type scheme in finite vol-
umes for the 2D shallow water equations.” International Journal for Numerical Methods in Fluids,
16(6), 489-505.
Alcrudo, F., and Garcia-Navarro, P. (1994). “Computing two dimensional flood propagation with a
high resolution extension of McCormack’s method.” Proceedings on Modelling of Flood Propagation
Over Initially Dry Areas, American Society of Civil Engineers, P. Molinaro and L. Natale Eds., Milan,
Italy, 3-17.
Anastasiou, K., and Chan, C.T. (1997). “Solution of the 2D shallow water equation using the finite
volume method on unstructured triangular meshes.” International Journal for Numerical Methods in
Fluids, 24(11), 1225-1245.
Bermudez, A., and Vazquez, M.E. (1992). “Flux-vector and flux-difference splitting methods
for the shallow water equations in a domain with variable depth.” Computer Modelling of Seas and
Coastal Regions. Ed. P.W. Partridge, Computational Mechanics Publications, Elsevier Applied Sci-
ence, London, 256-267.
Bermudez. A., and Vazquez, M.E. (1994). “Upwind methods for hyperbolic conservation laws
with source terms.” Computers Fluids, 23(8), 1049-1071.
Billett, S.J., and Toro, E.F. (1997). “On WAF-type schemes for multidimensional hyperbolic
conservation laws.” Journal of Computational Physics, 130(1), 1-24.
Boris, J.P., and Book, D.L. (1973). “Flux-corrected transport, SHASTA: A fluid transport algo-
rithm that works.” Journal of Computational Physics, 11(1), 38-69.
Bradford, S.F., and Katapodes, N.D. (1999). “Hydrodynamics of turbid underflows. I: Formula-
tion and numerical analysis.” Journal of Hydraulic Engineering, American Society of Civil Engineers,
125(10), 1006-1015.
Buning, P.G., and Steger, J.L. (1982). “Solution of the two-dimensional Euler equations with

20
generalized coordinate transformations using flux vector splitting.” American Institute of Aeronautics
and Astronautics, Paper 82–0971.
Causon, D.M., Mingham, C.G., and Ingram, D.M. (1999). “Advances in calculation methods for
supercritical flow in spillway channels.” Journal of Hydraulic Engineering, American Society of Civil
Engineers, 125(10), 1039-1050.
Chang, S.C. (1995). “The method of space-time conservation element and solution element: A
new approach for solving the Navier-Stokes and Euler equations.” Journal of Computational Physics,
119(2), 295-324.
Choi, H., and Liu, J.G. (1998). “The reconstruction of upwind fluxes for conservation laws: Its
behaviour in dynamic and steady state calculations.” Journal of Computational Physics, 144(2), 237-
256.
Colella, P., and Woodward, P. (1984). “The piecewise-parabolic method (PPM) for gas dynamical
simulations.” Journal of Computational Physics, 54(1), 174-201.
Davis, S.F. (1987). “A simplified TVD finite difference scheme via artificial viscosity.” Society
for Industrial and Applied Mathematics, Journal for Scientific and Statistical Computing, 8(1), 1-18.
de Saint Venant, B. (1871). “Théorie de mouvement Non-permanant des eaux avec application
aux crues des rivières et à l’introduction des marées dans leur lit.” Acad. Sci. Comptes rendus, 73,
148-154, 237-240.
Engquist, B., and Osher, S. (1981). “One sided difference approximations for nonlinear conserva-
tion laws.” Mathematics of Computation, 36(154), 321-351.
Fennema, R.J., and Chaudhry, M.H. (1986). “Explicit numerical schemes for unsteady free-
surface flows with shocks.” Water Resources Research, 22(13), 1923-1930.
Fennema, R.J., and Chaudhry, M.H. (1990). “Explicit methods for 2-D transient free-surface
flows.” Journal of Hydraulic Engineering, American Society of Civil Engineers, 116(8), 1013-1034.
Fraccarollo, L., and Toro, E.F. (1995). “Experimental and numerical assessment of the shallow
water model for two-dimensional dam-break type.” Journal of Computational Physics, 33(6), 843-
864.
Garcia-Navarro, P., Alcrudo, F., and Saviron, J.M. (1992). “1-D open-channel flow simulation
using TVD-McCormack scheme.” Journal of Hydraulic Engineering, American Society of Civil En-
gineers, 118(10), 1359-1372.
Gharangik, A.M., and Chaudhry, M.H. (1991). “Numerical simulation of hydraulic jump.” Jour-
nal of Hydraulic Engineering, American Society of Civil Engineers, 117(9), 1195-1211.
Glaister, P. (1991). “Solutions of a two-dimensional dam break problem.” International Journal
on Engineering Science, 29(11), 1357-1362.
Glaister, P. (1993). “Flux difference splitting for open-channel flows.” International Journal for
Numerical Methods in Fluids, 16(7), 629-654.
Harten, A. (1983). “High resolution schemers for hyperbolic conservation laws.” Journal Com-
putational Physics, 49(3), 357-393.
Harten, A. (1989). “ENO schemes with subcell resolution.” Journal of Computational Physics,
83(1), 148-184.
Harten, A., and Hyman, J.M. (1983). “Self adjusting grid methods for one-dimensional hyperbolic
conservation laws.” Journal of Computational Physics, 50(2) 235-269.
Harten, A., and Osher, S. (1987). “Uniformly high-order accurate nonoscillatory schemes, I.”
Society for Industrial and Applied Mathematics, Journal for Numerical Analysis, 24(2), 279-309.
Harten, A., Osher, S., Engquist, B., and Chakravarthy, S.R. (1986). “Some results on uniformly
high-order accurate essentially nonoscillatory schemes.” Applied Numerical Mathematics, 2(3/5),
347-377.

21
Harten, A., Lax, P., and van Leer, A. (1983). “On upstream differencing and Godunov-type
scheme for hyperbolic conservation laws.” Society for Industrial and Applied Mathematics, Review,
25(1), 35-61.
Hirsch, C. (1988). Numerical Computation of Internal and External Flows, Volume 1: Funda-
mentals of Numerical Discretization, John Wiley and Sons, Chichester.
Hirsch, (1990). Numerical Computation of Internal and External Flows, Volume 2: Computa-
tional Methods for Inviscid and Viscous Flows, John Wiley and Sons, Chichester.
Jameson, A., Schmidt, W., and Turkel, E. (1981). “Numerical solutions of the Euler equations by
finite volume methods using Runge-Kutta time-stepping schemes.” American Institute of Aeronautics
and Astronautics, Fluid and Plasma Dynamics Conference, Palo Alto, California, AIAA-81-1259.
Jha, A.K., Akiyama, J., and Ura, M. (1995). “First- and second-order flux difference splitting
scheme for dam-break problem.” Journal of Hydraulic Engineering, American Society of Civil Engi-
neers, 121(12), 877-884.
Jha, A.K., Akiyama, J., and Ura, M. (2000). “Flux-splitting schemes for 2D flood flows.” Journal
of Hydraulic Engineering, American Society of Civil Engineers, 126(1), 33-42
Jin, M., and Fread, D.L. (1997). “Dynamic flood routing with explicit and implicit numerical
solution schemes.” Journal of Hydraulic Engineering, American Society of Civil Engineers, 123(3),
166-173.
Lax, P. (1954). “Weak solutions of non linear hyperbolic equations and their numerical computa-
tion.” Communications in Pure and Applied Mathematics, 7, 159-193.
Lax, P., and Wendroff, B. (1960). “System of conservation laws.” Communications on Pure and
Applied Mathematics, 13(2), 217-237.
LeVeque, R.J. (1992). Numerical Methods for Conservation Laws, Birkhauser Verlag, Basel.
LeVeque, R.J. (1998). “Balancing source terms and flux gradients in high-resolution Godunov
methods: The quasi-steady wave-propagation algorithm.” Journal of Computational Physics, 146(1),
346-365, 1998.
MacCormack, R.W. (1969). “The effect of viscosity in hypervelocity impact cratering.” American
Institute Aeronautics and Astronautics, Paper 69-354.
Meselhe, E.A., and Holly, F.M. (1997). “Invalidity of Preissmann scheme for transcritical flow.”
Journal of Hydraulic Engineering, American Society of Civil Engineers, 123(7), 652-655.
Mingham, C.G., and Causon, D.M. (1998). “High-resolution finite-volume method for shallow
water flows.” Journal of Hydraulic Engineering, American Society of Civil Engineers, 124(6), 605-
614.
Molls, T., and Molls, F. (1998). “Space-time conservation method applied to Saint Venant equa-
tions.” Journal of Hydraulic Engineering, American Society of Civil Engineering, 124(5), 501-508.
Monaghan, J.J. (1985). “Particle methods for hydrodynamics.” Comp. Phys. Rep., 3, 71-124.
Monaghan, J.J. (1997). “SPH and Riemann solvers.” Journal of Computational Physics, 136(2),
298-307.
Nakatani, T., and Komura, S. (1993). “A numerical simulation of flow with hydraulic jump us-
ing TVD-MacCormack scheme.” Proceedings of XXV Congress of International Association for
Hydraulic Research, 1, 9-13.
Nujic, M. (1995). “Efficient implementation of non-oscillatory schemes for the computation of
free-surface flows.” Journal of Hydraulic Research, 33(1), 100-111.
Osher, S., and Solomon, F. (1982). “Upwind difference schemes for hyperbolic conservation
laws.” Mathematics of Computation, 38(158), 339-374.
Rahman, M., and Chaudhry, M.H. (1989). “Simulation of dam-break flow with grid adaptation.”
Advances in Water Resources, 21(1), 1-9.

22
Rider, W.J. (1993). “Methods for extending high-resolution schemes to non-linear systems of
hyperbolic conservation laws.” International Journal for Numerical Methods in Fluids, 17(10), 861-
885.
Roe, P.L. (1981). “Approximate Riemann solvers, parameter vectors, and difference schemes.”
Journal of Computational Physics, 43(2), 357-372.
Roe, P.L., and Pike, J. (1984). “Efficient construction and utilization of approximate Riemann
solutions.” Computing Methods in Applied Science and Engineering, R. Glowinski and J.L. Lions,
Eds., Amsterdam, North-Holland.
Savic. L.J., and Holly, F.M. (1993). “Dambreak flood waves computed by modified Godunov
method.” Journal of Hydraulic Research, 31(2), 187-204.
Shu, C.W., and Osher, S. (1988). “Efficient implementation of essentially non-oscillatory shock-
capturing schemes.” Journal of Computational Physics, 77(2), 439-471.
Shu, C.W., and Osher, S. (1989). “Implementation of essentially non-oscillatory shock-capturing
schemes, II.” Journal of Computational Physics, 83(1), 32-78.
Singh, V., and Bhallamundi, S.M. (1998). “Conjunctive surface-subsurface modeling of overland
flow.” Advances in Water Resources, 21(7), 567-579.
Smoller, J. (1983). Shock waves and reaction-diffusion equations, Springer-Verlag, New York.
Steger, J.L., and Warming, R.F. (1981). “Flux vector splitting of the inviscid gas dynamic equa-
tions with application to finite-difference methods.” Journal of Computational Physics, 40(2), 263-
293.
Stoker, J.J. (1957). Water Waves, The Mathematical Theory with Applications, Interscience, Lon-
don.
Strang, G. (1968). “On the construction and comparison of finite difference schemes.” Society for
Industrial and Applied Mathematics, Journal for Numerical Analysis, 5(3), 506-517.
Sweby, P.K. (1984). “High resolution schemes using flux limiters for hyperbolic conservation
laws.” Society for Industrial and Applied Mathematics, Journal of Numerical Analysis, 21(5), 995-
1011.
Toro, E.F. (1989). “A weighted average flux method for hyperbolic conservation laws.” Proceed-
ings of the Royal Society, Series A, 423, 401-418.
Toro, E.F. (1992). “Riemann problems and the WAF method for solving the two-dimensional
shallow water equations.” Philosophical Transactions of the Royal Society, London, Series A, 338,
43-68.
Toro, E.F. (1997). Riemann solvers and numerical methods for fluid dynamics. Springer-Verlag,
Berlin.
Toro, E.F., Spruce, M., and Speares, W. (1994). “Restoration of the contact surface in the HLL-
Riemann solver.” Shock Waves, 4 , 25-34.
van Albada, G.D., van Leer, B., and Roberts, W.W. (1982). “A comparative study of computational
methods in cosmic gas dynamics.” Astronomy and Astrophysics, 108(1), 76-84.
van Leer, B. (1973). “Towards the ultimate conservative difference scheme, I. The quest of mono-
tonicity.” Lecture Notes in Physics, 18, 163-168.
van Leer, B. (1977). “Towards the ultimate conservative difference scheme. IV A new approach
to numerical convection.” Journal of Computational Physics, 23(3), 276-298.
Wang, Z., and Shen, H.T. (1999). “Lagrangian simulation of one-dimensional dam-break flow.”
Journal of Hydraulic Engineering, American Society of Civil Engineers, 125(11), 1217-1220.
Warming, R.F., and Beam, R.M. (1976). “Upwind second-order difference schemes and applica-
tions in aerodynamic flow.” American Institute of Aeronautics and Astronautics, 14(9), 1241-1249.

23
Weiyan, T. (1992). Shallow Water Hydrodynamics: Mathematical Theory and Numerical Solution
for Two-Dimensional System of Shallow Water Equations, Elsevier Science Publishers.
Wu, C., Huang, G., and Zheng, Y. (1999). “Theoretical solution of dam-break shock wave.”
Journal of Hydraulic Engineering, American Society of Civil Engineers, 125(11), 1210-1215.
Yang, J.Y., Hsu, C.A., and Chang, S.H. (1993). “Computations of free surface flows part 1: One-
dimensional dam-break flow.” Journal of Hydraulic Research, 31(1), 19-34.
Yang, H.Q., and Prezekwas, A.J. (1992). “A comparative study of advanced shock-capturing
schemes applied to Burger’s equation.” Journal of Computational Physics, 102(1), 139-159, 1992.
Yee, H.C. (1987). “Construction of explicit and implicit symmetric TVD schemes and their ap-
plications.” Journal of Computational Physics, 68(1), 151-179.
Yee, H.C. (1989). “A class of high-resolution explicit and implicit shock-capturing methods.”
National Aeronautics and Space Administration, Technical Memorandum 101088, Ames Research
Center, Moffett Field, California.
Yee, H.C. (1997). “Explicit and implicit multidimensional compact high-resolution shock-capturing
methods: formulation.” Journal of Computational Physics, 131(1), 216-232.
Zalesak, S.T. (1979). “Fully multidimensional flux-corrected transport algorithms for fluids.”
Journal of Computational Physics, 31(3), 335-362.
Zhao, D.H., Shen, H.W., Lai, J.S., and Tabios, G.Q. (1996). “Approximate Riemann solvers in
FVM for 2D hydraulic shock wave modeling.” Journal of Hydraulic Engineering, American Society
of Civil Engineers, 122(12), 692-702.
Zhao, D.H., Shen, H.W., Tabios, G.Q., Lai, J.S., and Tan, W.Y. (1994). “Finite-volume two-
dimensional unsteady-flow model for river basins.” Journal of Hydraulic Engineering, American
Society of Civil Engineers, 120(7), 863-883.
Zoppou, C., and Roberts, S. (1999). “Catastrophic collapse of water supply reservoirs in urban
areas.” Journal of Hydraulic Engineering, American Society of Civil Engineers, 125(7), 686-695.
Zoppou, C., and Roberts, S. (2000). “Numerical solution of the two-dimensional unsteady dam
break.” Applied Mathematical Modelling, 24(7)457-475.

24
Table 1: Intercell flux, Fj+1/2 estimated using Osher’s P -ordering approximate Riemann solver for
physically realistic flow conditions.

Conditions ul − cl ≥ 0 ul − cl ≥ 0 ul − cl ≤ 0 ur + cr ≥ 0
ur + cr ≥ 0 ur + cr ≤ 0 ur + cr ≤ 0 ul − cl ≤ 0
um − cm ≥0 Fl F l + F r − F Sl F Sl + F r − F Sr F Sl
um − cm ≤0 Fl + Fm F l + F m − F Sl Fm + Fr Fm
um + cm ≥0 −FSl +Fr − FSr −FSr
um + cm ≤0 F l + F Sr − F Sl F l + F r − F Sl Fr F Sr

Table 2: Normalized L1 norm between the simulated water depth and velocity for the dam break
problem with h0 = 5m and h1 = 10m using various numerical schemes and their relative execution
time (CPU).

Numerical Scheme L1 L1 Relative


Water Depth h Velocity u CPU Time

Lax-Friedrichs 1.00 1.00 1.0


Bermudez and Vazquez 0.24 0.23 7.4
Steger and Warming 0.31 0.29 3.4
Yang et al. first-order scheme 0.23 0.22 5.9
Harten 0.10 0.08 3.6
Roe 4.64 4.90 3.2
Osher’s P scheme 0.22 0.20 3.5
HLL 0.22 0.21 3.6
Lax Wendroff 0.24 0.23 4.0
MacCormack 0.24 0.23 2.2
MacCormack with
artificial viscosity 0.13 0.12 3.0
Davis 0.10 0.11 5.4
Yee (Minmod limiter) 0.12 0.06 7.1
Yee (Monotone Limiter) 0.32 0.30 7.1
Yee (MUSCL limiter) 0.08 0.08 6.8
Yee (Superbee limiter) 0.31 0.29 7.6
Space-time 0.14 0.18 5.1
ENO 0.09 0.09 16.5
BAP 0.07 0.07 8.9
DPM 0.45 0.38 296.9
WAF 0.10 0.06 6.4
FCT 0.06 0.06 7.8
PPM 0.06 0.07 9.5
Exact Riemann solver 0.39 0.18 11.3

25
Table 3: Normalized L1 norm between the simulated water depth and velocity for the dam break
problem with h0 = 0.1m and h1 = 10m using various numerical schemes and their relative execution
time (CPU).

Numerical Scheme L1 L1 Relative


Water Depth h Velocity u CPU Time

Lax-Friedrichs 1.00 1.00 1.0


Bermudez and Vazquez 0.31 0.17 7.1
Steger and Warming 0.28 0.15 3.4
Yang et al. first-order scheme 0.19 0.14 5.7
Harten N/A N/A N/A
Roe 0.48 0.30 3.1
Osher’s P scheme 0.21 0.15 3.5
HLL 0.21 0.14 3.3
Lax Wendroff N/A N/A N/A
MacCormack N/A N/A N/A
MacCormack with
artificial viscosity N/A N/A N/A
Davis 0.12 0.07 5.2
Yee (Minmod limiter) 0.15 0.09 6.8
Yee (Monotonic Limiter) 0.31 0.17 6.7
Yee (MUSCL limiter) 0.14 0.10 6.6
Yee (Superbee limiter) 0.29 0.15 7.3
Space-time 0.10 0.07 5.1
ENO 0.12 0.08 16.4
BAP 0.10 0.09 8.9
DPM 0.50 0.25 297.6
WAF 0.06 0.07 6.4
FCT 0.12 0.09 7.5
PPM 0.08 0.06 9.2
Exact Riemann solver 0.11 0.17 10.3

26
t
n
fa
n S2

y
tio

uit
c

tin
fa R
re

on
1
ra

k
isc

oc
m

sh
td
tac
con
r
l
x
Figure 1: Shocks and rarefaction fan associated with the dam break problem with finite water depth
everywhere.

fj-1/2 fj+1/2

j-1 j-1/2 j j+1/2 j+1


Figure 2: Intercell fluxes for cell j.

Figure 3: Piecewise linear MUSCL reconstruction using three consecutive cells.

27
12 4 12 4

9 3 9 3

6 2 6 2

u(m/s)

u(m/s)
h(m)

h(m)
3 1 3 1

0 0 0 0
0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000
(a) x(m) x(m) (b) x(m) x(m)

12 4 12 4

9 3 9 3

6 2 6 2
u(m/s)

u(m/s)
h(m)

h(m)
3 1 3 1

0 0 0 0
0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000
(c) x(m) x(m) (d) x(m) x(m)

12 4 12 4

9 3 9 3

6 2 6 2
u(m/s)

u(m/s)
h(m)

h(m)

3 1 3 1

0 0 0 0
0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000
(e) x(m) x(m) (f) x(m) x(m)

12 4 12 4

9 3 9 3

6 2 6 2
u(m/s)
h(m)
u(m/s)
h(m)

3 1 3 1

0 0 0 0
0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000
(g) x(m) x(m) (h) x(m) x(m)

12 4 12 4

9 3 9 3

6 2 6 2
u(m/s)
h(m)

u(m/s)
h(m)

3 1 3 1

0 0 0 0
0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000
(i) x(m) x(m) (j) x(m) x(m)

Figure 4: Solution of the dam-break problem where h1 = 10m and h0 = 5m using (a) Lax-Friedrich
scheme, (b) Lax-Wendroff scheme (c) MacCormack scheme, (d) MacCormack scheme with artificial
viscosity, (e) Davis total variational diminishing scheme, (f) Hartens conservative upwind scheme,
(g) space-time scheme, (h) essentially non-oscillatory scheme (ENO), (i) biased averaging procedure
(BAP) and (j) discrete parcel method (DPM).

28
12 12 12 12

9 9 9 9

6 6 6 6

u(m/s)

u(m/s)
h(m)

h(m)
3 3 3 3

0 0 0 0
0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000
(a) x(m) x(m) (b) x(m) x(m)

12 12 12 12

9 9 9 9

6 6 6 6
u(m/s)

u(m/s)
h(m)

h(m)
3 3 3 3

0 0 0 0
0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000
(c) x(m) x(m) (d) x(m) x(m)

12 12 12 12

9 9 9 9

6 6 6 6
u(m/s)

u(m/s)
h(m)

h(m)

3 3 3 3

0 0 0 0
0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000
(e) x(m) x(m) (f) x(m) x(m)

12 12 12 12

9 9 9 9

6 6 6 6
u(m/s)

u(m/s)
h(m)

h(m)

3 3 3 3

0 0 0 0
0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000
(g) x(m) x(m) (h) x(m) x(m)
12 12

9 9

6 6
u(m/s)
h(m)

3 3

0 0
0 500 1000 1500 2000 0 500 1000 1500 2000
(i) x(m) x(m)

Figure 5: Solution of the dam-break problem where h1 = 10m and h0 = 0.1m using (a) Davis
total variational diminishing scheme, Lax-Wendroff/Upwind scheme with (b) minmod limiter, (c)
monotonic limiter, (d) MUSCL limiter, (e) superbee limiter, (f) flux corrected transport (FCT), (g)
Steger and Warming upwind scheme, (h) Bermudez and Vazquez conservative upwind scheme and (i)
Yang et al. conservative upwind scheme.

29
12 12 12 12

9 9 9 9

6 6 6 6

u(m/s)
u(m/s)

h(m)
h(m)

3 3 3 3

0 0 0 0
0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000
(a) x(m) x(m) (b) x(m) x(m)

12 12 12 12

9 9 9 9

6 6 6 6
u(m/s)

u(m/s)
h(m)

h(m)
3 3 3 3

0 0 0 0
0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000
(c) x(m) x(m) (d) x(m) x(m)

12 12 12 12

9 9 9 9

6 6 6 6

u(m/s)
u(m/s)

h(m)
h(m)

3 3 3 3

0 0 0 0
0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000 0 500 1000 1500 2000
(e) x(m) x(m) (f) x(m) x(m)

12 12

9 9

6 6
u(m/s)
h(m)

3 3

0 0
0 500 1000 1500 2000 0 500 1000 1500 2000
(g) x(m) x(m) (h)

12 12

9 9

6 6
u(m/s)
h(m)

3 3

0 0
0 500 1000 1500 2000 0 500 1000 1500 2000
(i) (j) x(m) x(m)

Figure 6: Solution of the dam-break problem where h1 = 10m and h0 = 0.1m using (a) Godunov
scheme with an exact Riemann solver, and approximate Riemann solvers of (b) Roe (c) Osher’s P
scheme, (d) HLL scheme, (e) weighted average flux (WAF), (f) piecewise parabolic method (PPM),
(g) space-time scheme, (h) essentially non-oscillatory scheme (ENO) (i) biased averaging procedure
(BAP) and (j) discrete parcel method (DPM).

30
9 Notation
A(U) = Jacobian of the flux vector;
Aj = Jacobian of the flux vector at node j;
Acj+1/2 = corrected flux;
(k)
Aj+1/2 = flux limiter applied across the kth shock;
B(x) = biased functions;
Cj =√exact solution at node j;
c = gh celerity;
cj = approximate solution at node j;
cl,m,r = celerity in the left, intermediate and right states respectively of the Riemann problem;
Dj+1/2 = artificial dissipation;
e1,2 = right eigenvectors of A;
F(U) = vector of flux values;
Fj = flux vector at node j;
Fj = flux vector at node j in the pedictor step;
n
fj+1/2 = intercell flux;
G = matrix which satisfies F = GU;
g = acceleration due to gravity;
h = water depth;
hj = water depth at node j;
hl,m,r = water depth in the left, intermediate and right states respectively of the Riemann problem;
h0 = water depth downstream of a reservoir;
h1 = water depth upstream of a reservoir;
I = identity matrix;
K = empirical coefficient;
L = length of the computational domain;
L1 = normalized sum norm;
lj = smoothing length of parcel j;
l0 = initial smoothing length;
M = matrix of right eigenvectors of A;
p = half width of the interpolating window;
R1,2 = 1 or 2-rarefaction fan;
S = vector of source and sink terms;
S = shock speed;
Sf = friction slope;
Sfx = friction slope in the x co-ordinate direction;
Sfy = friction slope in the y co-ordinate direction;
Sl,r = left and right respectively sonic points;
S0x = bed slope in the x co-ordinate direction;
S0y = bed slope in the y co-ordinate direction;
S1,2 = 1 or 2-shock;
sj = slope in the linear reconstruction polynomial;
sl,r = left and right flux spatial gradients;
T V (Un ) = total variational;
t = time;

31
U = vector of conservative quantities;
Uj = vector of conservative quantities at node j;
Uj = vector of conservative quantities at node j in the pedictor step;
Ul = vector of conservative quantities left of a discontinuity;
Ur = vector of conservative quantities right of a discontinuity;
Vk = volume of parcel k;
u = water velocity in the x co-ordinate direction;
ul,m,r = water velocity in the left, intermediate and right states respectively of the Riemann problem;
v = water velocity in the y co-ordinate direction;
W (xk − xj , l) = normalized interpolating kernel function;
x = co-ordinate direction;
xj = distance to node j;
y = co-ordinate direction;
α = max |λi |;
j
αi = wave strengths;
β ± = ∆j+1/2∓1 /∆j+1/2 ;
δj+1/2 = indicator function;
π = 3.1416;
(k)
νj+1/2 = change of flux across the kth shock;
∆j+1/2 = Uj+1 − Uj ;
∆x = computational distance step;
∆t = computational time step;
η = the Manning resistance coefficient;
εj+1/2 = artificial viscosity coefficient;
Λ = diagonal matrix containing the eigenvalues of A;
Λ± = diag(λ± );
λi = eigenvalues;
λ±i = (1±sgn(λi ))/2;
φ(β + , β − ) = flux limiter; and
ω = exponent in the superbee limiter.
Superscript
LW = Lax-Wendroff scheme;
M ac = MaCormack scheme; and
U P = Upwind scheme.
Subscripts
i = eigenvalue number;
j = computational node;
l = left state;
n = computational time level;
m = intermediate state; and
r = right state.

32

View publication stats

You might also like