Computers Math. Applic. Vol. 24, No. 7, pp. 69--75, 1992 0097-4943/92 $5.00 + 0.

00
Printed in Great Britain. All rights reserved Copyright(~ 1992 Pergamon Press Ltd

P R E D I C T I O N OF S U P E R C R I T I C A L F L O W
IN O P E N C H A N N E L S
P. GLAISTER
Department of Mathematics
P.O. Box 220, University of Reading
Whiteirn|ghts, Reading, RG6 2AX, United Kingdom
(Received J~l~ 1991)

A.lmtract--A finite difference~.hemebased on flux differencesplitting is presented for the solution

of the one-dimcmslonalshallow water equations in open channels. A linearisod problem, analogous
t o that of Riemann for gM dynamics, is defined and a scheme, based on numerical characteristic
decomposition, is pre~nted for obtainln~ approximate solutions to the linearlsed problem. The
method of upwind dlifm~mcingis treed for the resulting scalar problems, tosether with a flux limiter
for obtaining a second order scheme whichavoids non-physical,spurious oo~;llations. The schemeis
applied to 8 problem of flowin a river whose geometryinduces a region of supercriticedflow.

1. I N T R O D U C T I O N
The flow of water in a channel with (local) rectangular cross section and smoothly varying
bottom surface is governed by the one-dimensional shallow water equations for open channels.
The assumption of hydrostatic pressure distribution is used in deriving these equations [1]. Since
analytical solutions of these equations are not generally available, they axe solved numerically.
Computational models of river flow based on the so-called 'St. Venant' equations of open channel
flow axe a well established tool in engineering practice. Such models can provide quantitative
information on discharge, velocity and level for a variety of purposes including flood defence
design, navigation, flood forecasting, flood plain zoning, dam-break analysis and irrigation scheme
control.
Several explicit and implicit finite difference methods have been used to solve the shallow water
equations [2-4]. One feature of this set of hyperbolic equations is the formation of bores, i.e.,
discontinuous solutions, which can be dii~cult to represent accurately even if a shock-capturing
method is used. The most popular scheme in the hydraulic engineering community is the im-
plicit method of Preissman [5]. However, this scheme does not perform well in the presence
of discontinuities. Finite element methods have also been applied to these equations (see, for
example, [6]).
In the field of unsteady gas dynamics governed by the Euler equations, where shocks are
frequently present, some authors have designed finite difference schemes that have good shock-
capturing properties (see, for example, [7]). These schemes solve linearised Riemann problems
using upwind differencing and flux limiters to obtain shocks that are spread over two or three
mesh points. An alternative approach to flux difference splitting was proposed by Vila [8] for
the equations of isentropic gas dynamics and has been applied with ~uccess to the shallow water
equations [9]. The scheme of Godunov [10] solves Riemann problems exactly using an iterative
procedure. Vila simplifies this iteration using approximate Riemann invariants, and achieves
second order accuracy by considering generalised Riemann problems, i.e., ones where the data
is assumed to be piecewise linear discontinuous. The scheme in [7] applies upwind differencing
to a specially constructed set of scalar problems. Second order accuracy is then achieved using
classical second order scalar schemes, limited to avoid non-physical oscillations in the solution.
In this paper, a new scheme is presented for the open channel flow equations that incorporates
the ideas mentioned earlier for the Euler equations. Although the derivation of this scheme is

Typeset by .AA~oTFfl
69
70 P. GLAISTKR

detailed, its implimentation is straightforward. The resulting algorithm is efficient and produces
satisfactory results for a test problem of geometry-induced supercritical flow in a river.
2. GOVERNING EQUATIONS
The St. Venant equations governing the rough-turbulent flow of water in an open channel can
be written as
_W, + G_, - g, (2.1)
where
_W -- (A, Q)T, (2.2a)

_G(W) = Q, , (2.2b)

and

o_(_w)= ( o,-oA ( .

The quantities A - A(z,t) and Q = Q(z,t) = Au(z,t) represent the c ~ t i o n a l mrea and
ms.flow, respectively, at a general position z measured along the channel, and at time t, where
u is the fluid velocity. The gravitational constant is represented by g. We consider the ease where
the channel is locally rectangular, so that
A = d(z,z),
where B is the breadth and d the depth. The height is then h = h(z,f) = d(z,t) + z(z),
where z is the height of the river bed. Also K - AIM (hydraulic radius) 2/s, where M ~ n g ' s
constant M is taken u 0.03, and the hydraulic radius = A/wetted perimeter = Bd/(2d + B).
Equations (2.1)-(2.2c) can then be written, after some manipulation, as
_w,+C.=L (2.3)
where
_w= (~b,~ u) T , (2.4a)
T

and

B'fz) B'(z) (, ~bBfz) '~-4/s~ "r (2.4c)

.fCw_)= -u4, B(z) gC,z'(z)-0.000O#,/,ulul \ 2 ~ + g B ( z ) , , / ,] '

with u = u(z, t) as before, and ~ = ~(z, t) = g d(z, t) = g (h(z,t) - z(z)), as the dependent
variables which we solve for. [N.B. ~ = gA/B, ~u = gQ/B.]
We consider first the solution of Equations (2.3)-(2.4c) with f(_w) - 0_, i.e., with frictionleas,
straight, parallel channel walls and flat surface bottom, and then extend the scheme to include
the case where the source terms f_(w_)~ 0.
3. LINEARISED RIEMANN P R O B L E M
If the approximate solution of Equations (2.3)-(2.4c) is sought using a finite difference method,
then the solution is known at a set of discrete mesh points (z,t) = ( z j , t , ) at any time t = tn.
Following Godunov [10], the approximate solution w_~ to w_ at (zj,tn) can be considered as a
set of piecewise constants w = w? for z E (zj - Az/2, zj + Az/2) at time tn where ~ z =
z$ - z$-1 is a constant mesh spacing. A Riemann problem is now present at each interface
z~-1/2 -- 1/2 (zj_x + z j ) separating adjacent states wn_j-x, _w~_. Iftbe shallow water equations, are
linearised by considering the 3acobian matrix of the flux function F to be constant in each interval
(z~_x, zj), the resulting equations can be solved approximately using explicit time stepping. The
time step At is restricted so that the solutions of adjacent Riemann problems do not interact.
The scalar problems that result from this analysis can be solved by upwind differencing; however,
an approximate Jacobian matrix needs to be constructed in each interval, so that shock-capturing
is automatic.
 Prediction of supercritical flow 71

4. A P P R O X I M A T E RIEMANN SOLVER
OF
Consider first Equations (2.3)-(2.4c) with f(w) - 0. The Jacobian matrix A = ~w of the flux
function F(_w) has eigenvalues Ai with corresponding eigenvectors _ei, i = 1,2, given-by

~, = ~ + v ~ , ~_, = (1,. + v T ) T, (4.1a)

As - u - vf~, e_2 - (1,u - vf¢) T. (4.1b)

This information can be used to develop approximate solutions of the Riemann problem of
Section 3.
Consider two adjacent states _wr., _wR (left and right) given at either end of the cell (xL, zR) on
an z-coordinate line,and consider also the algebraicproblem of findingan approximate Jacobian
- -4(_wL,_wR) in this cellsuch that
2 A_w -- A_F, (4.2)
where A(. ) -- (.)~ -- (.)L. A solution to this problem, for arbitrary jumps A_w, can be used to
obtain a conservative scheme with good shock-capturing properties. Equivalently, approximate
eigenvalues Ai and corresponding eigenvectors ~i of A can be sought, such that
2
A_w = Z 6~~i, (4.3a)
i=1

and
2
AF = ~ A, &, ~_i, (4.3b)
i=1

where &i are wavestrengths prescribed in terms of the arbitrary jump A w.

By initially considering small jumps A w, so that Equations (4.3a-b) are satisfied to
within O(A2), a solution of Equations (4.3a-b) can be determined [II]. The required approximate
eigenvalues, eigenvectors and wsvestrengths are

AI = ~ + 5, e_'1= (1,~ + ~)T, (4.4a)

A2 = ~ - tb, ~ = (1, ~ - tb)T, (4.4b)
1 1~
~ = ~ A~ + ~ ~ Au, (4.5a)

1 1¢
52 = ~ A¢ - ~ ~ Au, (4.5b)

where the approximations to u, ¢ and ~f~ in (zL,zR) are given by

~ = vf~uR + vf~-uL
4-~+ ~ ' (4.~a)
-- ~'R~--L, (4.6b)

and

(4.~c)

respectively. Thus, using (4.3b), Equation (2.3) with _f(_w)- 0 can be approximated by

At A~ = 0, (4.7)
CAMWA24:7-F
72 P. GLAISTER

where Ax and At represent the mesh spacing in the z and t directions, respectively, and the
point P may be L or R. Upwind differencing now applied to (4.7) gives the following first order
algorithm for the solution of Equations (2.3)-(2.4c), with Lf(w_)- 0:
add At ~i &i e_.-i to wR, when ~i > 0, or,
Az
(4.8)
add At Ai ~i es to wL, when ~i < 0.
A x - -

Thus, we note the direction of flow of information given by the approximate eigenvalues ~i
and use this information to update the solution consistent with the theory of characteristics of
Equation (2.3). In addition, second order transfers of these first order increments can be made
to achieve higher accuracy, providing they are limited to maintain monotonicity [12]. The use of
these "flux-limiters" improves accuracy without introducing non-physical spurious oscillations,
especially at bores. To allow depression waves to be treated correctly, the first order increment
can be considered as two separate increments, being sent to either end of the cell [13].

5. E X T E N S I O N T O V A R I A B L E B R E A D T H AND D E P T H , AND F R I C T I O N
In the case where the breadth and depth of the channel varies, a 'source-term' f ~ 0 is present,
given by (2.4c). This term, however, contains no derivatives of flow variables, and therefore the
scheme of Section 4 can be retained for shock-capturing. It has been shown that for linear systems
the source term f should be upwinded in the same way as Fx (see [14]). This has been used
successfully for compressible flows in ducts of smoothly varying cross section [15]. Specifically,
approximating f in the interval (ZL,zR) by

g A"x'x-0"0009g' lal , (5.1)

where ~ = Vr~--~ - ,/~ = ~/B(zr.)B(za), and projecting

1 2
.__
= - (5.2)
i----1
enables Equations (2.3)-(2.4c) to be solved approximately. The first order algorithm can be
written as in (4.8), where the &~ are replaced by modified wavestrengths 7i = &i + ~i. (N.B. ~z
and ~2 are easily determined from (5.2) in terms of the components of f.)
6. A T E S T P R O B L E M
We consider the flow in a channel with a smooth constriction and a sloping bottom sur-
face, which shelves. This geometry induces a flow which becomes supercritical. The channel is
10,000 metres long, and the breadth, B, varies from 10 metres to 5 metres to 10 metres (see
Figure 1). The bed slope is taken to be a constant value, except between 4500 and 5500 metres
where twice this value is taken (see Figure 2). Only one boundary condition needs to be applied
to each end of the channel. At the left-end the mass flow, Q, is specified, and at the right-hand
end the depth, d, is specified by extrapolation from the interior.

7. N U M E R I C A L RESULTS
Figures 3, 4 and 5 show the results for a slope of 0.01, and the mass flow, Q, the depth, d, and
the Froude number, F - u/V~-J, are all displayed. Figures 6, 7 and 8 show the corresponding
results for a slope of 0.02. Both sets of results are for 100 mesh points, a massflow of 20 cubic
metres per second and after a time of 3000 seconds. We see that the flow has reached a steady
state, with a good prediction of the flow, even though it is supercritical in some, or all, of the
channel.
The explicit finite difference scheme of Section 4 and Section 5 is computationally efficient
and the c.p.u, time taken to compute the results given above was as follows. Using an Amdahi
V7 with 100 mesh points takes 0.0096 c.p.u, seconds to compute one time step and a total of
2.88 c.p.u, seconds to reach a real time of 3000 seconds, using 300 time steps, each of 10 seconds.
 Prediction of supercritical flow 73
”

s--

2--

I--

O-
1CCG m 3mO 6mo 5wo bow ?cm 0cca woo low
-I --

-2. -

A--

Figure 1. Chanuel breadth.

0 Iwo 2ooo 3ooo4ooosooo6ooo~~~ 1OMC

Figure 2. Channel cross-section.

d Q

5.00 -. 30 0 --

3.75 -- 22.5 -.
. . .-. . . . . -- ..“. . . . . . . . . . . . . . . . . . . . . . . . . ._ . . . . . . . . . . .

2.50 -. IS.0 ._

I.25.. 7.5 --
... . . .. . . .. .
.. . . .. .... . - .. .. . . . . .

aa u)(x) 7500 Icow x 25M so00 MO IMoa x

-1.25 -. -7.5 ..

-2.50 -- -15.0 .-

-3.7s -. -22.5 --

-5.w -. -30.0 --

Figure 3. A plot of the depth, d, for a bed- Figure 4. A plot of the mcmsflow, Q, for a
slope of 0.01. bed-slope of 0.01.

8. CONCLUSIONS
A conservative finite difference scheme is presented for the solution of the shallow water equa-
tions in open channels based on flux difference splitting. By considering linearised Riemann
problems, and solving these approximately using upwind differencing, enables the geometry-
?4 P. GLAISTgR
F d

2.0 5.00

1.5

I.O 2.50.

0.5 I.~

I I I I I I I I
2500 SO00 ?~OO 100~ x 25OO 50GO ?500 10000 x

-O.S -I.~

-1.0 -2.50

-I.S -3.75 •

-2.0 .$.00.

Fisure 5. A plot of the Froude number, F, for Fisure 0. A plot of the depth, d, for s bed-
a bed-slope d 0.01. • lope of 0.02.
Q F
30.0. 2.0

..~-..
~.S 1.5
..°.. . . . . . . . . . . . . . - ............

15.0. I°0

7.5 0.5

I ! I I | I ! I
2S00 5O00 ?SO0 1OOOO 25O0 5OO0 ~ 10000

-7.5 -0,5

-15.0 - I .0

-22.5 -I.5

-30.0 -2.0

Figure 7. A plot of the m a n flow, Q, for a Figure 8. A plot of the Proude ~ , F , for
bed-slope of 0.02. a bed-slope of 0.02.

induced supercriticsl flow in an open channel to be predicted sstisfsetorily. The resulting scheme
is computstionslly efficient and can be used with confidence to predict accurate solutions to open
channel flow problems.

REFERENCES

1. J.J. Stoker, Wiler WsveJ, Interscience Publishers, Wiley and Seres, New York, (195T).
2. M.B. Abbott, Compststiomsl HId~slie.: E/cinema, o] the TAeorl o.f Ftee-$sryaee FIst~, Pitman, London,
(m~).
3. J. Cuap, F.M. Holley and A. Vorwey, Pmcfical AJpec~. o] Gompstsfionsl Ri~er HydmuKcs, Pitman,
~ , (1980).
4. R.J. Feanenm and M.H. Clumdry, S ~ of one.dimmslmal dam-break flows, Jour. H~d. Rein. 25 (1),
41-01 (i~,7).
5. A. ~ Pmpqatiea des imummcmcl ~ . lee cmmzx et rivikm, Premnted at the Fir.t Go~lreu
s! ~te YrneA'Asss¢iatls. /or Oompsi.tlo=, Grmoble, Prince, (April, 1901).
6. J. Goumeimlle and F. Lepelatm, A method to m l w Che one.dha~slel~l shal~w water equsticm uahq;
chsmcteg~ics and flux qdlttialg, In Prec. o] IAe /I,/. Co#]. H1dn.dte asd gs~/roamests/Modellint o.t
 Prediction of supercritic~d flow 75

Coast"i, Emtsari=e a~d River WaterJ, Bradford, United Kingdom (Edited by R.A. Falconer, P. Goodwin
and R.G.S. Matthew), Gow-m"Tedmicsl, 506-515, (1989).
7. P. Glaist~, An apprm~mate l i n ~ Riemann solver for the tree-dimensional Euler equations with real
8asu, J. Comp~t. Pkls. T4 (2), 382-4G8 (1988).
8. J.P. Vila, Shnp]/fied Godunov schemes for 2 × 2 systems of conservstiom laws, SlAM J. Numer. Anal. 28
(6), UTS-1192 (1986).
9. J.P. Vila, Schemas numm-[ques en hydrsulique des ecotdmnente &vec discontinuities, In Pros. XII Congncu
IAHR, Lstmmme, (Edited by J ~ . Cunge and P. Achers), IAMR, Delft, The Netherlands, (1987).
10. S.K. Godunov, A dltference method for the numericsl computation of couti~uons solutions d hydrodynamlc
equations, Tramflsted as JPRS ?225 by U.S. Dept. of Comme~e, 1960, Mst. $6ornik 4Y (3), 271--306 (1959).
11. P. Glaister, Dilfercnce schemes for the shallow water equations, Numerical Analysis Report 9/8?, University

12. P.K. Sweby, High resolution schemes using flux llmlters fc~ hyperbolic consez~tion laws, SIAM J. N1,mer.
Ansi. 21 (5), ~5-1011 (1984).
13. P.K. Sweby, A modification of Roe's scheme for entropy satisfying solutions of scalar nora-linear conserv~icm
laws, Numerical Analysis Report 6/82, University of lq,u~l|n~, (1982).
14. P.L. Roe, Upwind differencing so.heroism,hyperbolic conservation laws with source terms, In First Int. Co=-
I r e u on Hyperbolic Problems, St. Etienne, (Edited by C. Oar~uu~ and D. Serre), Springcs~-Verlag, Berlin,
pp. 41-51, (1986).
15. P. Ghdste~r, Flux difference splitting for the Euler equations in one spatial coordinate with area wwiation,
Int. J. N~mer. Meth. Fluids 9 (1), 97-119 (1988).