You are on page 1of 26

Alternative Hydraulics Paper 5, June 11, 2012

Long waves in open channels their nature, equations,


approximations, and numerical simulation
John D. Fenton
Institute of Hydraulic and Water Resources Engineering
Vienna University of Technology
Karlsplatz 13/222, 1040 Vienna, Austria
http://johndfenton.com/Alternative-Hydraulics.html
johndfenton@gmail.com
Abstract
This is a document that is not ready. It is intended for referees of a paper submitted to Agricultural
Water Management so that they can judge some of the results. The paper should be ready by the
middle of June.
Table of Contents
1. Introduction . . . . . . . . . . . . . . . . . . . . . . 2
2. The long wave equations . . . . . . . . . . . . . . . . . . 2
3. Non-dimensionalisation . . . . . . . . . . . . . . . . . . 4
4. Linearisation of the equations . . . . . . . . . . . . . . . . 5
4.1 Mass and momentum conservation equations . . . . . . . . . . 6
4.2 The Telegraphers equation . . . . . . . . . . . . . . . 7
5. Propagation behaviour of waves . . . . . . . . . . . . . . . . 9
5.1 Dimensionless Telegraphers equation . . . . . . . . . . . . 9
5.2 Solutions for waves periodic in time . . . . . . . . . . . . 9
5.3 Approximate solutions in limits of not-so-long and very-long waves . . 10
5.4 Presentation of results . . . . . . . . . . . . . . . . . 12
5.5 Stability of ow roll waves . . . . . . . . . . . . . . . 14
6. The momentum equation relative importance of terms and some
approximations . . . . . . . . . . . . . . . . . . . . . 14
7. The slow-change/slow-ow routing equation . . . . . . . . . . . . 18
8. Numerical solution of the long wave equations an explicit nite difference
scheme . . . . . . . . . . . . . . . . . . . . . . . . 20
8.1 A Forward-Time-Central-Space scheme . . . . . . . . . . . 20
Long waves in open channels their nature, equations, approximations, and numerical simulation John D. Fenton
8.2 Linear stability . . . . . . . . . . . . . . . . . . . 20
8.3 Approximate stability criteria . . . . . . . . . . . . . . 22
8.4 Test of linear stability . . . . . . . . . . . . . . . . . 22
8.5 Determining stability limits for unsteady ow problems . . . . . . 23
9. Characteristics . . . . . . . . . . . . . . . . . . . . . 24
References . . . . . . . . . . . . . . . . . . . . . . . 25
10.Introduction . . . . . . . . . . . . . . . . . . . . . . 27
10.1 Approximate stability criteria . . . . . . . . . . . . . . 28
10.2 Test of linear stability . . . . . . . . . . . . . . . . . 29
1. Introduction
This is not ready
2. The long wave equations
Consider the long wave equations in the form obtained by Fenton (2010b, equations 19):
In terms of cross-sectional area A and discharge Q: For some theoretical studies, this is the
simplest formulation:
A
t
+
Q
x
i = 0, (1a)
Q
t
+ 2
Q
A
Q
x
+

gA
B

Q
2
A
2

A
x
+P
Q|Q|
A
2
gA

S = 0, (1b)
where t is time, x is a horizontal streamwise co-ordinate, i is inow per unit length of stream, g is
gravitational acceleration, B is surface width, is the Boussinesq momentum coefcient, is a dimen-
sionless resistance coefcient, which is dened below, and where

S is the local downstream bed slope in
a mean sense, evaluated around the perimeter. In equation (1b) some relatively unimportant terms have
not been shown: the inow contribution to momentum has been neglected; there are no converging or
diverging vertical walls (otherwise an extra contribution to the slope term would have to be made); and
where the Boussinesq momentum coefcient has been assumed to be constant.
In terms of surface elevation and Q: In practice it is more convenient to use surface elevation
instead of area as the dependent variable, where the derivatives are related by
A
t
= B

t
and
A
x
= B

x
+

S

, (2)
provided there are no vertical walls that move in time t or converge or diverge in x. Equations (1)
become
B

t
+
Q
x
i = 0, (3a)
Q
t
+ 2
Q
A
Q
x
+

gA
Q
2
B
A
2


x
+P
Q|Q|
A
2

Q
2
B
A
2

S = 0. (3b)
The resistance term What to do: In most situations, however, the geometry is poorly known and only
an approximate

S is used, and similarly is usually poorly known. Certainly the slope correction

S
2
in
the expression for is almost always unnecessary.
2
Long waves in open channels their nature, equations, approximations, and numerical simulation John D. Fenton
The resistance term in both momentum equations is P Q
2
/A
2
, in which is the dimensionless resis-
tance coefcient
=

8

1 +

S
2

, (4)
where is the dimensionless Weisbach resistance coefcient.This denition allows for nite slopes to
be used, however in almost all rivers and canals,

S is so small that = /8 to good approximation, and
is simply related to Chzys C by = g/C
2
. We will nd that using the dimensionless makes a
number of our expressions simpler.
In physical terms connects the mean stress on the wetted boundary to the mean velocity in the stream
Q/A by (e.g. eqn (6.11) of White 2003)
=

8

Q|Q|
A
2
=
Q|Q|
A
2
,
where is uid density, and so the manner in which the resistance appears in the momentum equations
(1b and 3b), as PQ|Q| /A
2
, has a simple physical signicance as a dimensionless coefcient (in which
the necessary horizontal resolution of force has been included) multiplied by the wetted perimeter and
the square of the mean velocity with sign attached.
Other, more familiar, ways of writing the resistance term are
Manning!
gAQ|Q| /K
2
: where K is the conveyance of the stream
K =
r
gA
3
P
, (5)
which is a function of the resistance coefcient and the stream geometry. Although this form loses
some of the physical signicance of PQ|Q| /A
2
, occasionally it is more convenient to use it as a
shorthand.
gAS
f
: where S
f
is called the resistance slope. This terminology has its pitfalls, for example in some
works (e.g. Lyn 1987, Lyn & Altinakar 2002) it has been assumed to be a constant even where Q
and A vary.
The introduction of makes notation rather simpler further below. If we consider the case of steady uni-
form ow in a prismatic canal of constant slope S
0
and resistance
0
, when all derivatives in equations
(1) are zero and Q = Q
0
, A = A
0
, and P = P
0
are all constant, equation (1b) becomes

0
P
0
Q
2
0
A
2
0
gA
0
S
0
= 0, (6)
In general, the resistance coefcient as given by equation (4) for is a function of the relative rough-
ness of the channel boundary = k
s
/(A/P) where k
s
is equivalent sand grain roughness and the
channel Reynolds number R = (Q/P)/ where is kinematic viscosity, as given by Yen (1991, eqn
30), as explained in Fenton (2010a). For a particular channel shape, P is a known function of A so that
equation (6) actually involves A
0
and Q
0
in quite a complicated manner. If the resistance coefcient
0
is taken as independent of Reynolds number and so independent of Q
0
, the equation can be written as an
explicit expression for mean velocity U
0
= Q
0
/A
0
, the familiar Chzy-Weisbach form of the uniform
ow equation:
U
0
=
r
g

0
A
0
P
0
S
0
. (7)
This has an interesting signicance if we re-write it to give an expression for the Froude number F
0
of
the ow
F
2
0
=
U
2
0
gA
0
/B
0
=
S
0

0
B
0
P
0
. (8)
3
Long waves in open channels their nature, equations, approximations, and numerical simulation John D. Fenton
As the wetted perimeter P is approximately equal to the top width B for wide streams, we have the
relationship F
2
0
S
0
/
0
, so that the square of the Froude number is approximately the ratio of the
slope S
0
to the dimensionless resistance coefcient
0
.
A single equation in terms of upstream volume V : For theoretical studies it is easier to consider
a single equation in terms of the volume of uid V upstream of a point. That rate of change of volume
with time is equal to the rate of inow Q(x
0
, t) at the upstream end of the stream x
0
plus the increase
due to distributed inow i along the stream, less Q, the discharge at the general point x, as this is the
rate at which uid is becoming no longer upstream. The result is
V
t
= Q(x
0
, t) +
Z
x
x0
i(x
0
) dx
0
Q. (9)
The rate of change of volume with x is simply the cross-sectional area of the ow
V
x
= A. (10)
Solving these two equations for A and Q:
A =
V
x
and Q = Q(x
0
, t) +
Z
x
x
0
i(x
0
) dx
0

V
t
, (11)
(3 of Fenton 2010b). The inclusion of Q(x
0
, t) has been suggested by Fatemeh Soroush (2011, Personal
Communication), improving on an earlier denition by the author (Fenton, Oakes & Aughton 1999,
Fenton & Keller 2001, Barlow, Fenton, Nash & Grayson 2006). With this addition V can now more
consistently be understood as the total volume in the channel between the upstream point x
0
and a
general point x. Equations (11) satisfy the mass conservation equation (1a) identically. The momentum
conservation equation (1b) becomes

2
V
t
2
+ 2
Q
A

2
V
xt
+

Q
2
A
2

gA
B(A)


2
V
x
2
=
gA

S +P(A)

Q(x
0
, t) +
R
x
i(x
0
) dx
0
V/t
V/x

2
+
dQ(x
0
, t)
dt
. (12)
where for notational simplicity, single symbols Q and A have been retained in coefcients, and for
conciseness the resistance term has been shown in terms of the usual unidirectional ow result Q
2
rather
than Q|Q|. The momentum equation has become a second-order partial differential equation in terms
of the single variable V . It is a single equation in a single unknown, and could be used for ood routing
and wave propagation studies, but as it is a second-order equation, in computations one would probably
proceed by introducing an intermediate variable V/t, which is the reverse of the direction from which
we have come. The equation is probably more useful in theoretical works such as this. However, we
will see that there is an interesting and useful approximation to it.
3. Non-dimensionalisation
The aim of non-dimensionalisation is to simplify the problem by grouping physical quantities, reducing
the number of parameters, and determining the relative magnitudes of quantities in the problem. We now
attempt to do this for the long wave equations with boundary conditions, and will show that previous
approaches have been misleading because scales for the independent variables have not been made clear,
and neither have the magnitudes of terms in the differential equations.
Consider equations (1) with no inow (i = 0) for a channel of general cross-section. The variation with
time in the channel has a scale T, which is determined by the time variation of the input to the system,
while the scale of the length of disturbances L is not known at this stage. We introduce dimensionless
quantities denoted by asterisks, such that t = t

T, and x = x

L. For channel width we use the width


scale B
0
such that B = B

B
0
, and for the perimeter we write P = P

P
0
, in terms of a perimeter scale
4
Long waves in open channels their nature, equations, approximations, and numerical simulation John D. Fenton
P
0
. We could have used B
0
for this, but the end result is slightly more concise if we use a separate
scale. Now introducing the area scale A
0
, the dependent variables Aand Qare scaled as A = A

A
0
and
Q = Q

U
0
A
0
. In addition we write the channel slope and friction factors in terms of reference values
with a 0 subscript as

S = S
0
S

, and =
0

. The equations we obtain are


L
U
0
T
A

+
Q

= 0, (13a)
U
0
gS
0
T
Q

+
U
2
0
gS
0
L

x

Q
2

+
A
0
/B
0
S
0
L
A

Q
2

A
2

= 0, (13b)
where we have used the uniform ow equation (7) to write
0
in terms of U
0
, and where it has been
assumed that ow does not reverse so that we can use Q
2
in the resistance term.
In the traditional approach (e.g. Woolhiser & Liggett 1967, and subsequent researchers), the length
and time scales of disturbances, and hence their velocity of propagation, are assumed related by the
mean uid velocity, such that it is assumed that T = L/U
0
, the equations become, after dividing the
momentum equation through by U
2
0
/gLS
0
:
A

+
Q

= 0,
Q

1
F
2
0

Q
2

A
2

+ 2
Q

+
LS
0
F
2
0
A
0
/B
0

2
A

!
= 0.
There are two dimensionless ow parameters: the Froude number appearing as 1/F
2
0
, and the quantity
LS
0
/

F
2
0
A
0
/B
0

, which Liggett (1968) called the Kinematic Flow Number. In fact it can be written as
1/F
2
0
times LS
0
/ (A
0
/B
0
), the Drop/depth ratio, which is the ratio of the drop in channel bed LS
0
over
a disturbance length scale of L to the mean depth A
0
/B
0
, so the Drop/depth ratio could be thought of as
the second parameter.
Strelkoff & Clemmens (1998, eqn 12) made the additional assumption that one can attach a magnitude
to L, such that in present terms L = (A
0
/B
0
) /S
0
, so that the drop/depth ratio is 1, and the Kinematic
Flow Number becomes 1/F
2
0
. The momentum equation can be written
F
2
0


Q
2

A
2

+ 2
Q

+
A

2
A

= 0,
in terms of a single parameter F
2
0
, and which suggests that the leading terms can be ignored for F
2
0
1,
leading to the so-called Zero-inertia model.
The problem with these formulations is that in the rst a relationship between Land T has been assumed
and additionally in the second a value of L has been assumed in terms of channel geometry. This means
that we cannot be sure what the magnitudes of the derivatives are for example in the dynamic wave
limit, when much experience suggests that waves travel at a velocity related to
p
gA
0
/B
0
, then L/T/U
0
,
assumed to be 1 is actually of a magnitude
p
gA
0
/B
0
/U
0
= 1/F
0
, and the non-dimensionalisation as
given does not tell us the real relative importance of terms.
In this work, we prefer not to use the above approach. In fact, we believe that such scalings cannot be
done at this stage, as we assert that while the input time scale T is an independently-speciable quantity
from the boundary conditions imposed on the problem, the length scale L is determined by the actual
solutions to the equations. To obtain the relative importance of various terms in the equations without
solving them seems not to be possible. We now proceed to a solution of the linearised equations and
then use that to examine the relative importance of terms in the momentum equation.
4. Linearisation of the equations
5
Long waves in open channels their nature, equations, approximations, and numerical simulation John D. Fenton
4.1 Mass and momentum conservation equations
To linearise the equations, it is actually more insightful to go back to their dimensional form. We
consider relatively small disturbances about a uniformow, where there is no inowi = 0, and we write
A = A
0
+A
1
and Q = Q
0
+Q
1
,
where is a small quantity. As A
0
and Q
0
are constant, all derivatives of A and Q in equations (1) are
of order so that the coefcients need only be written to zeroeth order, and the linearisation is simple.
The only non-trivial operations are in the remaining resistance and slope terms of equation (1b). We
introduce the function for those terms
= P
Q
2
A
2
gA

S, (14)
where we have written Q|Q| = Q
2
, as we consider small perturbations about a uniform ow, which is
unidirectional. For small perturbations in area and discharge of A
1
and Q
1
the Taylor expansion of
is
=
0
+Q
1

Q

0
+A
1

A

0
+O

, (15)
where in the case of steady uniform ow

0
=
0
P
0
Q
2
0
A
2
0
gA
0
S
0
= 0.
For the rst derivative in (15) we obtain from (14)

0
= 2
0
Q
0
P
0
A
2
0

1 +
1
2

Q

0
Q
0

. (16)
As
0
is in general a function of discharge Q
0
(because of its variation with Reynolds number) this
expression is a somewhat complicated function of Q
0
. Below we will see that it, with units of T
1
is
actually an important channel parameter, determining the nature of wave behaviour, so it helps us here
to obtain a more physical feel for it. We obtain an approximation by noting that the variation of with
Qis relatively unimportant so that we neglect the derivative /Qand as is independent of Qin this
approximation we can use the Chzy-Weisbach resistance equation (7) to eliminate Q
0
= U
0
A
0
, giving
the expression in terms of the geometrical and roughness properties of the channel

0
2
s
gS
0

0
A
0
/P
0
, (17)
which would be quite accurate in most applications.
For the other derivative we obtain

0
= gS
0
+

0
P
0
Q
2
0
A
3
0

2 +
A
0
P
0
P
A

0
+
A
0

and we can make use of the uniform ow relationship (7) to give

0
= 3gS
0

1
1
3

A
0
P
0
P
A

0
+
A
0

(18)
In this case, both derivative terms in P/A and /A in general may have nite contributions, the
rst expressing the effect of nite channel width, and the second because the resistance coefcient has a
strong variation with relative roughness, as shown and quantied in Fenton (2010a, p4).
Now to combine the results into linear equations. The mass conservation equation (1a) is already linear.
For the momentum conservation equation (1b), we take the derivative terms with coefcients from the
6
Long waves in open channels their nature, equations, approximations, and numerical simulation John D. Fenton
zeroeth order solution and combine with the linearised resistance-slope contribution (15) to give the
equations
A
1
t
+
Q
1
x
= 0, (19a)
Q
1
t
+

C
2
0

2
U
2
0

A
1
x
+ 2U
0
Q
1
x
+

Q

0
Q
1
+

A

0
A
1
= 0, (19b)
in which the mean uid velocity U
0
= Q
0
/A
0
has been used for simplicity, as well as the speed C
0
which is given by
C
0
=
q
gA
0
/B
0
+

U
2
0
, (20)
as obtained by Keulegan & Patterson (1943, eqn 87) for the wave speed of small disturbances when
there is no resistance, and who noted that a similar expression had been obtained by Boussinesq (1877,
p285, eqn 265). This result is a generalisation that puts into context the fact that in most textbooks this
is written with = 1 such that C
0
=
p
gA
0
/B
0
, an approximation, and which is usually said to be the
"celerity" or "long wave speed" or "dynamic wave speed". Remarkably, the same expression, but with
no limitations of linearisation, is valid for the speed of characteristics, shown in 9. However, it is not,
in general, the speed at which disturbances travel. Below it will be shown that it is actually the speed
of waves only in the limit of shorter waves, but where they are still long enough that the hydrostatic
approximation holds. That limit is achieved relatively rarely, for example, when waves are due to rapid
gate movements. The dynamic wave speed seems to be a much less-important quantity in hydraulics
than is generally believed.
It is occasionally useful when we consider velocities of propagation to use the coefcient of the sec-
ond term in the form C
2
0

2
U
2
0
, as written in equation (19b), while at other times when we con-
sider the relative importance of terms it will be useful to use equation (20) to write it in the form
(gA
0
/B
0
)

1 F
2
0

.
4.2 The Telegraphers equation
A pair of equations such as (19) has been obtained and analysed by Ponce & Simons (1977) for wide
channels, requiring a number of elementary matrix operations for subsequent manipulations. It is math-
ematically and physically simpler if, instead, the equations are combined into a single equation similar
to equations for wide channels obtained by Deymie (1938, eqn 6), Iwasa (1954, eqn 8), Lighthill &
Whitham (1955, eqn 28) and Ponce (1990, eqn 4). Continuing our analysis of prismatic channels of
arbitrary section we introduce a perturbation upstream volume v (x, t) (see equation 11 above) such that
A
1
= v/x and Q
1
= v/t, so that it satises the mass conservation equation (19a) identically.
The momentum equation (19b) becomes the Telegraphers equation:

v
t
+c
0
v
x

+

2
v
t
2
+ 2U
0

2
v
xt
(C
2
0

2
U
2
0
)

2
v
x
2
= 0. (21)
We have introduced the symbols and c
0
, where
1. The quantity is simply /Q|
0
as given by equations (16) and (17):
=

Q

0
=
2gS
0
U
0

1 +
1
2
Q
0

2
s
gS
0

0
A
0
/P
0
, (22)
in which we have shown two forms, one in terms of U
0
and the other approximation in terms of
the fundamental geometrical channel quantities and resistance where it has been assumed that
does not depend on Q. It has dimensions of T
1
; we see here that it increases with both slope and
resistance and decreases with depth.
7
Long waves in open channels their nature, equations, approximations, and numerical simulation John D. Fenton
2. The term c
0
is the ratio of the two derivatives
c
0
=
/A|
0
/Q|
0
, (23)
and using equations (16) and (18):
c
0
=
3
2
U
0
1
1
3

A
0
P
0
P
A

0
+
A
0

1 +
1
2
Q
0

0
(24)
= U
0
, (25)
where for convenience, the symbol has been introduced, which from equation (24) is roughly
3/2. In most practical situations we know the resistance coefcient only approximately, and
including the derivatives /Q and /A might be thought excessively enthusiastic. However,
for possible theoretical studies, they will be retained here. The actual derivatives of with respect
to Aand Qcould be obtained from the formulae (7) of (Fenton 2010a), which give the derivatives of
( 8) with respect to relative roughness and Reynolds number. In many situations the Reynolds
number of the ow is so large that there is no longer any variation with it, such that /Q = 0. A
more important situation is where bed forms can develop, when would show a stronger variation
with discharge Q.
The quantity c
0
is a wave speed, as will be shown below. It has previously been called the "kinematic
wave speed", but as we shall show, that is actually a misleading term: it is actually the speed of
propagation of very long waves.
The denition of c
0
in equation (23) is a generalisation of the usual Kleitz-Seddon equation
c
0
=
dQ
0
(A
0
)
dA
0
,
where Q
0
(A
0
) is the discharge given by any of the usual resistance formulae. The generalisation is
necessary for the model we have set up here, where the resistance coefcient may be a function
of Q , such as when it depends on channel Reynolds number.
Nature of wave propagation in limiting cases: At this stage it is already possible to anticipate
results that will be established more fully below. The terms in the Telegraphers equation (21) can be
grouped: the rst two terms can be characterised as rst derivatives, and the last three terms are all
second derivatives. We now examine the governing equation in two limits:
Very-long waves: For disturbances that have a long period, such that
2
/t
2
/t, which we will
call "very long waves", the last three terms in equation (21) can be neglected, and the equation becomes
v
t
+c
0
v
x
= 0,
with a general solution v = f
1
(x c
0
t), where f
1
(.) is an arbitrary function given by the upstream
conditions. This solution is a wave propagating downstream at speed c
0
= U
0
.
Not-so-long waves: In the other limit, for disturbances which are shorter, such that
2
/t
2
/t,
for which we use the term "not-so-long" waves, equation (21) becomes

2
v
t
2
+ 2U
0

2
v
xt
(C
2
0

2
U
2
0
)

2
v
x
2
= 0,
which is a second-order wave equation with solutions
v = f
21
(x (U
0
+C
0
) t) +f
22
(x (U
0
C
0
) t)
where f
21
(.) and f
22
(.) are arbitrary functions determined by boundary conditions. In this case the
8
Long waves in open channels their nature, equations, approximations, and numerical simulation John D. Fenton
solutions are waves propagating upstream and downstream at velocities of U
0
C
0
. It is noteworthy
that the water carries disturbances at a theoretical transport velocity of U
0
, rather than just U
0
, its mean
velocity. The waves have a speed of C
0
relative to the transport velocity. If = 1, with no velocity
prole or turbulence, equation (20) gives C
0
=
p
gA
0
/B
0
, which in many places is referred to as "the"
long wave speed. What we have shown here is that it is the speed of waves just in the shorter limit and
then only for = 1. Accordingly we will not call C
0
the long wave speed. In fact, it will be shown that
there is no such thing as a long wave speed.
We shall now spend some effort on the intermediate general case of long waves intermediate between
the limits of "not-so-long" and "very long".
5. Propagation behaviour of waves
5.1 Dimensionless Telegraphers equation
Now it is simpler to make the Telegraphers equation (21) dimensionless. We introduce dimensionless
varibles t

= t and x

= x/U
0
, and simply consider v to have been made dimensionless in some
manner that does not matter. The equation becomes
v
t

+
v
x

+

2
v
t
2

+ 2

2
v
x

(F
2
0
)

2
v
x
2

= 0. (26)
We note that the physical parameters and are constant for a particular channel, which leaves the
equation containing a single parameter determining the nature of solutions, the Froude number in the
form F
2
0
. It might be thought awkward to have a quantity F
2
0
which is often large, but it is simpler to
have it appearing in this way, in only one place and where it actually expresses the relative importance
of this term, rather than multiplying the whole equation through by F
2
0
and having that appear as an
artefact in every term but one.
It might be thought that there is a large problemwith this in 3 on conventional non-dimensionalisations
we were critical of exactly the scaling for x

and t

we have used here. However, we are about to obtain


an analytical solution of equation (26) for the wavelength in terms of the period, and with that solution,
in 6 below we calculate the magnitudes of the various terms in the equation for different periods and
Froude numbers, the results being independent of our scaling.
5.2 Solutions for waves periodic in time
We start by obtaining solutions of the form
v = exp(i (x

)) , (27)
where i =

1, and where in general and are complex quantities, whose nature determines the
behaviour of the solutions. Substituting this solution into the dimensionless Telegraphers equation (26)
and dividing through by common terms gives the polynomial equation
i
2
+ 2i + i
2

F
2
0

= 0, (28)
where the order of the terms corresponds to those in equation (21). Ponce & Simons (1977) obtained
a similar equation for a wide channel, however different terms had different coefcients, each of which
could take a magnitude of 0 or 1, so that they could be switched in or out according to different approx-
imations. We will not follow that path, preferring to let the equation make its own approximations.
Another departure here is that we are not so interested in the response of waves that are periodic in space,
as in rivers and canals usually the lengths of disturbances are not xed. Instead we will consider waves
that are periodic in time, as any input disturbance can be written as a Fourier series in time. Then from
the governing equation we will determine the behaviour in space. In this case, is real, and equation
9
Long waves in open channels their nature, equations, approximations, and numerical simulation John D. Fenton
(28) is a quadratic equation for with solutions
=
(i/2 +) +
q
(i/2 +)
2
+

F
2
0

(
2
+i)
F
2
0

, (29)
where = 1 and where we have neither expanded the products nor have we written it in standard
complex form, as either gives a much longer expression.
For any frequency of input , this equation (29) gives the complex value of determining the wave
propagation characteristics. If we write it as =
r
+i
i
, the solution (27) can be written
v = exp(
i
x

) exp

i
r

r
t

, (30)
so that in (x

, t

) space, the decay coefcient in x

is
i
, the length of the waves is 2/
r
, and the
propagation velocity is /
r
. In terms of the physical independent variables, as t

= t and x

=
x/U
0
, the solution is
v = exp

i
U
0
x

exp

r
U
0

x
U
0

r
t

, (31)
so that the physical propagation velocity is U
0
/
r
, for which we will use the symbol c. To plot the
results for dimensionless propagation velocity, the widely-used dimensionless velocity c/
p
gA
0
/B
0
is
suggested, which is related to the
r
calculated from equation (29) by
c
p
gA
0
/B
0
=
U
0

r
p
gA
0
/B
0
=
F
0

r
, (32)
and as
r
calculated from equation (29) is a function of Froude number F
0
already, as well as a function
of frequency , this does not complicate the representation. For plotting the spatial decay rate, it seems to
be too complicating to use anything other than the dimensionless
i
, the decay rate in the dimensionless
space variable x

.
5.3 Approximate solutions in limits of not-so-long and very-long waves
Unfortunately, before presenting the results, some more mathematical manipulations are necessary, to
relate the present work to known limiting behaviour and to explain some of the behaviour to be shown
graphically.
Above we have obtained solutions in terms of dimensionless frequency , whereas dimensionless period
seems to have more signicance. Consider the time variation in equation (27): t

= t, which varies
by 2 while physical time varies by a period T, so that
=
2
T
, (33)
and T is the dimensionless period. We will consider two limits for T.
5.3.1 Not-so-long waves, T small
The rst limit is where the dimensionless wave period is actually relatively small. The theory we have
been using, based on an approximation that the waves are much longer than the depth, is correctly
called a long wave theory, such that the pressure distribution is hydrostatic, however in the case we are
considering the waves are short enough that T is small, and we call this the "not-so-long" wave limit.
In this case where T is relatively small so that is large, the solution (29) can be written as an asymp-
10
Long waves in open channels their nature, equations, approximations, and numerical simulation John D. Fenton
totic series, valid for large . The solution for becomes
=
F
0
(F
0
+C

0
)
1 F
2
0
+i
F
0

F
0
C

0
+

1 +F
2
0
( 1)

1 F
2
0

0
+O

, (34)
where C

0
= C
0
/
p
gA
0
/B
0
=
q
1 +

F
2
0
. The solution shows the rst two terms of the series
the rst is of order
1
, which is real, and the next
0
which is purely imaginary. The next neglected
term is O

which is also real. From equation (32) these real terms give the physical velocity of
propagation, which after several steps of simple mathematics, including rationalising surds,
c = U
0
C
0
+O

, (35)
where C
0
is the dynamic wave speed, given by equation (20). This is the result, of course, widely-
believed to be the case for all long waves, where the waves are carried downstream at an advective
velocity (curiously, U
0
and not just U
0
), but propagate up- and down-stream relative to this at velocities
U
0
. The error term in equation (35) which we can write as O

(T)
2

shows how the result applies


only to not-so-long waves, and not to all long waves. We will see below how limited is the range of
wave periods for which this is a good approximation.
Having established that slightly unexpected result, other results (for example the next term in equation
34) are more complicated, and it is just as insightful here to take the common case = 1, for uid
velocity constant over the depth, and = 3/2, for a wide channel, to give the result for the leading
imaginary term in equation (34), which as it is independent of and hence period, is a decay rate. It can
be shown to be

i
=
F
0
2

1
2
F
0
1 +F
0
+O

, (36)
and so for F
0
small it is approximately
i
= F
0
/2, so that it is positive for downstream propagation so
that the wave decays downstream, and decays upstream for propagation in that direction.
5.3.2 Very-long waves, T large
The other limit is where the dimensionless wave period is relatively large, which we will call the "very-
long" wave limit, where T is large, small. The series expansion of the solution (29) in terms of
:
=
iF
2
0
/2
1 F
2
0
( 1)+

+
F
2
0
( 1)
1 F
2
0

+i

2
F
2
0

1 F
2
0

2
(2 1)

+O

. (37)
Taking the real part, equation (32) gives for the velocity of propagation
c = U
0
+O

for downstream propagation, = 1 (38a)


c =
U
0
1 + 2F
2
0
/

1 F
2
0
+O

for upstream propagation, = 1 (38b)


thus in the very long wave limit waves that travel downstream do so at a velocity c = U
0
= c
0
, where c
0
is the kinematic wave speed, while those that travel upstream do so at a similar, slightly smaller, speed.
For upstream propagation c has a value of approximately c
0
, where the value is actually modied by
a function of F
2
0
. There is no apparent advection component from the ow in either direction, unlike
"not-so-long" waves. It is interesting that this theory has shown that the condition that waves travel at
the kinematic wave speed is that the waves be very long, whereas it has traditionally been believed that
the requirement is that the Froude number be small.
Considering the decay rate of the waves, for downstream propagation = +1, the imaginary part of
11
Long waves in open channels their nature, equations, approximations, and numerical simulation John D. Fenton
equation (37) has a zero rst part, leaving

+
i
=

2
2
3
F
2
0

1 F
2
0

2
(2 1)

+O

, (39)
which is of a diffusion-like nature, that the decay rate in space varies like the square of the frequency in
time, so that shorter waves decay faster, much faster, while in the limit of sufciently long waves there
is no diminution.
For upstream propagation

i
=
F
2
0
1 F
2
0

+
i
, (40)
and we see that the decay rate consists of a frequency-independent term plus the negative of the diffusion
termfor downstreamwaves. These waves decay as x becomes increasingly negative, as we might expect.
5.4 Presentation of results
To estimate over what range we should obtain and present solutions we consider equation (22) for ,
which gives
T
2gS
0
U
0
T (41)
and take g 10 ms
2
, and a representative value of U
0
= 1 ms
1
. To estimate the larger magnitudes
of T, we consider a slow-rising ood such that the period might be T = 4 days on a stream of steeper
gradient S
0
= 10
3
, which gives
T 2 4 24 3600
10 10
3
1
6900.
For the smallest likely value, consider the fast movement of a gate T = 20 min (time to open 10 min)
on an irrigation canal of gentle slope S
0
= 10
4
, giving
T 2 20 60
10 10
4
1
= 2.4.
In view of these values we will plot results over a range of dimensionless wave periods 1 6 T 6 10
4
.
Figure 1 shows the results for wide channels with = 3/2, and for = 1 as a function of dimensionless
period T with Froude number as parameter. For the moment we are concerned only with the solid
black lines; the coloured dashed lines are an approximation that will be considered below.
The upper part (a) of the gure shows the dimensionless propagation velocity c/
p
gA
0
/B
0
as computed
from equations (29) and (32). The set of curves for positive/negative c are for downstream/upstream
propagation. It has been obvious since the work of Lighthill & Whitham (1955) and more recently
Ponce & Simons (1977) that propagation velocity is in general a function of wave length. Here, we have
put it in the context of wave period rather than length; the gure shows that waves of different periods
travel at different velocities. In mathematical terms this means that the physical system of long waves
with resistance actually shows the phenomenon of dispersion. It is clear and noteworthy that shorter
waves travel faster than longer waves, whether going upstream or downstream. There is no such thing
as a unique propagation speed. The results are more complicated than is widely believed.
For not-so-long waves, T < 2, say, at the left of the gure, corresponding to rapid changes such as
due to gate movements, the curves are almost horizontal and as shown by equation (35), c U
0
C
0
is a good approximation to the wave velocity, for both downstream and upstream propagation. For
T > 2, say, and certainly for T > 10, U
0
C
0
is no longer a good approximation to the speed of
propagation, contradicting the widely-held belief in hydraulic engineering. For 2 < T < 100 or 200
the propagation velocity continuously depends on wave period. For T > 100 or 200, the curves are
12
Long waves in open channels their nature, equations, approximations, and numerical simulation John D. Fenton
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
Velocity
c
p
gA
0
/B
0
F
0
= 0.1
0.2
0.4
0.6
0.8
F
0
= 0.1
F
0
= 0.3
c U
0
+C
0
c U
0
C
0
c U
0
c eqn (38b)
Downstream propagation
Upstream propagation
(a) Propagation velocity
-0.3
-0.2
-0.1
0.0
0.1
1 10 20 50 100 1000 10000
Decay
rate

i
Dimensionless period T
(b) Spatial decay rate
F
0
= 0.1
F
0
= 0.2
F
0
= 0.3
F
0
= 0.1
F
0
= 0.8
Downstream propagation
Upstream propagation

i
eqn (36), = +1

i
eqn (36), = 1

i
eqn (39)

i
eqn (40)
Flood
waves
Fast gate
movements
Linear theory, eqn (29)
Figure 1. Dependence of propagation properties on wave period and Froude number
again almost horizontal, but this time with a propagation velocity corresponding to equation (??) with
c c
0
= U
0
for downstreampropagation, more honoured in the hydrology literature, and with a similar
magnitude for upstream propagation. The conventional view is that this is a low-inertia or small Froude
number approximation. Figure 1 shows graphically that it is actually a slow variation approximation for
sufciently long period waves, as shown by the neglected terms O

=O

(T)
2

in equation (??)
and apparently valid for any (sub-critical) Froude number as predicted by the theory.
Figure 1(b) shows the spatial decay rate of disturbances. Again it can be seen that there is a dependence
on wave period, in this case rather more marked than for wave speed. In mathematical terms, this
dependence of decay rate on period is known as diffusion. Classically that refers to decay in time of
disturbances in space, here we use the same term, to describe the damping in space of disturbances in
time. The gure shows that in all cases, just as in classical diffusion, shorter disturbances decay more
quickly. For downstream propagation, shown by positive decay rates, for sufciently long waves the
decay rate goes to zero which means that very long waves suffer little decay. The behaviour is like
2
so that if we write equation (39) as
i
=
2
where is a diffusion-like coefcient, giving the spatial
13
Long waves in open channels their nature, equations, approximations, and numerical simulation John D. Fenton
decay rate in terms of the square of the frequency in time, for downstream propagation
=
1

F
2
0
+

(2 1)
2

+O

,
and in the common approximations = 1 and for wide channels = 3/2, =

8/F
2
0
2

/27.
For upstream propagation the behaviour is very different, and we nd that
lim
0

i
=

F
2
0


2

such that there is a contribution independent of frequency plus a negative diffusion term, the same as the
downstream one in magnitude, but meaning that waves decay as they propagate in the x direction. The
most noteworthy feature of this is the leading term independent of frequency, very much greater than for
downstream-travelling waves, so that the upstream-waves decay much more quickly.
5.5 Stability of ow roll waves
The stability limit of ow in the channel can be determined by writing the solution (29) in standard
complex form and satisfying the condition for stability, that
i
= =() > 0. After a number of long
algebraic manipulations, the limiting condition leads to the limit in terms of Froude number
F
0
6
1
q

2
(2 1)
(42)
for stability. For F
0
above this limit the ow becomes unstable and roll waves develop. This result can
be obtained quickly from equation (39) for very long waves, for the decay coefcient
+
i
to be positive.
It can also be obtained from the imaginary term of equation (34) for not-so-long waves, after some
manipulations, or, for the case of a wide channel = 3/2 and with the traditional = 1,the condition is
obtained from equation (42) (or immediately from approximation 36 for not-so-long waves) that F
0
6 2
for stability, the result obtained by Jeffreys (1925). If we were to use Gauckler-Manning resistance we
would nd = 5/3 for a wide channel, and using this with the more realistic value of = 1.05 gives
F
0
6 1.75 for stability. For = 3/2 and = 1.05, the limiting value is F
0
6 2.58. It can be seen that
this stability limit is quite sensitive to the values of and , unlike most of the results in this work.
6. The momentum equation relative importance of terms and
some approximations
Having established the general nature of wave propagation now we examine the relative importance
of terms in the momentum equation, using the dimensionless Telegraphers equation (26), so that we
know when approximations can be made. Consider the various cells of the second row of table 1 with
individual terms taken from the full momentum equation (1b), where the resistance and slope terms are
best considered as a single term shown bracketted. In the third row are shown the corresponding terms in
the dimensionless Telegraphers equation, where the resistance and slope terms are now separated into a
time and a space derivative. In the fourth row are shown the terms in polynomial (28) after substitution
of the periodic solution. We could substitute solutions from equation (29), but it is more insightful to
consider the approximations of small and large .
Not-so-long waves, T small, large: we have the asymptotic solution (34). Results for the relative
importance of terms are shown in the fth row of table 1. The rst term corresponding to Q/t in the
original equation is the same for all linear solutions here, as it contains a contribution i
2
and none
from a term. Examining the approximations in the subsequent terms along the row it can be seen that
this and the next two terms in the momentum equation are of order
2
, whereas the resistance and slope
terms are of order . In this limit of large ,this would suggest an approximate equation, in which the
14
Long waves in open channels their nature, equations, approximations, and numerical simulation John D. Fenton
Term number 1 2 3 4 5
Momentum eqn (1b)
Q
t
2
Q
A
Q
x

gA
B

Q
2
A
2

A
x

P
Q
2
A
2
gA

Telegraphers eqn (26)



2
v
t
2

2

2
v
x

(1/F
2
0
)

2
v
x
2

v
t


v
x

Polynomial (28) i
2
i2 i
2

F
2
0


Substituting the solution () from equation (29) results shown just for limiting cases from (34) & (37):
Not-so-long waves, large i
2
O

O()
Very-long waves, small
Downstream i
2
O

1 F
2
0

2
F
2
0

2
+O

+O

Upstream i
2
2F
2
0
1 F
2
0
+O

i

2
F
2
0
1 F
2
0
+O() i

2
F
2
0
1 F
2
0
+O()
Table 1. Terms in momentum equation and their relative magnitudes in the limits of not-so-long waves and very
long waves
latter terms could be omitted. However, this dropping of the relatively simple resistance and slope terms,
and retaining the terms including the time derivative and two space derivatives does not give much of a
simplication, and we will not consider this limit further.
Very long waves, T large, small: substituting the solution (37) the values are given in the last
two rows of the table, for downstream and upstream propagation respectively, as they are different:
Downstream propagation: in this case the rst two terms are of magnitude
2
, which we shall sub-
sequently ignore in this limit of small . The third term is also of this order, but in the limit of small
Froude number the term varies like i
2
/

2
F
2
0

, which is not necessarily small relative to the remain-


ing two terms of magnitude in columns four and ve. This means that if we retain all of term three as
we are not making a low-inertia (small Froude number) approximation here, plus terms four and ve we
would approximate the momentum equation (1b) by

gA
B

Q
2
A
2

A
x
+P
Q
2
A
2
gA

S = 0. (43)
To examine the wave propagation properties of the approximation compared with the full momentum
equation, we consider the corresponding terms in the second row of table 1, excluding terms 1 and 2, to
give the quadratic equation in
i

F
2
0

2
+ = 0, (44)
with solution
=
i/2 +i
q

2
/4 i

F
2
0

F
2
0

. (45)
Figure 2 is a repeat of part of gure ??, but for slightly longer waves, T > 10 and with smaller vertical
range, plus results from the approximation (43) in the form of the linear solution (45) shown by blue
dashed lines. Here, for the moment, we conne our attention to downstream-propagating waves, the
upper part of each of the gures 2(a) and (b). The result is immediately clear that equation (43) is
indeed a slow-change approximation, and not a low-inertia approximation, because agreement of both
propagation velocity and decay rate with the full linear theory is very close for all waves longer than
T 50, and shows very little dependence on Froude number at all.
Now we introduce a further approximation, where we neglect the inertial term in the brackets in
15
Long waves in open channels their nature, equations, approximations, and numerical simulation John D. Fenton
-0.5
0.0
0.5
1.0
1.5
Velocity
c
p
gA
0
/B
0 F
0
= 0.1
0.2
0.4
0.6
0.8
F
0
= 0.1
F
0
= 0.3
c U
0
c eqn (38b)
Downstream propagation
Upstream propagation
(a) Propagation velocity
-0.1
-0.1
-0.0
0.0
0.1
10 20 50 100 1000 10000
Decay
rate

i
Dimensionless period T
(b) Spatial decay rate
F
0
= 0.1
F
0
= 0.8
F
0
= 0.1
F
0
= 0.2
F
0
= 0.3
Downstream propagation
Upstream propagation
Linear theory, eqn (29)
Approxmn: eqn (45) based on (43)
Approxmn: eqn (48) based on (46)
Figure 2. Dependence of propagation properties on wave period and Froude number
equation (43), of relative magnitude F
2
, believing that now for the rst time we are making a low-inertia
approximation, to give
gA
B
A
x
+P
Q
2
A
2
gA

S = 0. (46)
This is a well-known approximation (*************), however, it is widely believed that the approxi-
mation just consists of ignoring terms of order F
2
, as suggested by the traditional scaling of the dimen-
sionless equation (16), as also done by the author previously (Fenton & Keller 2001, Appendix B.2).
To examine the wave propagation properties of this approximation we ignore the terms in equation
(44) to give the polynomial
i
2
F
2
0
+ = 0, (47)
with solution, similarly simply obtained from equation (45)
= iF
2
0

/2 +
q

2
/4 i/F
2
0

. (48)
16
Long waves in open channels their nature, equations, approximations, and numerical simulation John D. Fenton
This result is plotted as red dashed lines on gure 2. Examination of the top halves of gures (a) and
(b) shows the very surprising result that, if anything, agreement with the full linear solution is even
better, especially for velocity of propagation, and results are even better for large Froude number. For
these downstream travelling waves (propagation velocity and decay rate both positive) the accuracy of
the approximation does depend only on the period, such that it is accurate for T & 20 and does not
deteriorate with increasing Froude number. In this case, it really does seem to be a "slow-change" ap-
proximation. For decay rate, there is some loss of accuracy with increasing Froude number, nevertheless
it is clear that the approximation is still very much a very-long-wave approximation, that for T & 50
is an excellent approximation, for all (sub-critical) Froude numbers considered. Even though formally
equation (46) was obtained from the the very-long-wave approximation (43) by dropping a term of order
F
2
, it shows almost no loss of accuracy with increasing Froude number.
Hence, for downstream-propagating waves the simplied momentum equation (46) in effect is just a
long period or slow-change approximation.
Upstream propagation: A different situation arises when we consider upstream propagating waves,
possibly caused by control changes such as gate operations or tides. . Considering the last row of table 1
it can be seen that the rst, unsteady, term Q/t still has a relatively small contribution of i
2
in this
long wave limit; the next term involving Q/x has a contribution O

F
2
0

; and in the last three cells


that each has a contribution of order O() and in the third and fth there are leading order terms in F
2
0
that cancel. The result is that the second term, neglected in the above approximations, has a magnitude
relative to the last three terms of O

F
2
0

, which is as was previously believed to be the order of accuracy


of the approximation (46). In this case the approximate equations (43) and (46) are both a slow change
and a slow-ow approximation.
Considering the results of the linear solutions (45) and (48) for propagation velocity plotted in the
lower half of gure 2(a), it can be seen the results are now dependent on Froude number, even for the
longest waves, and already for F
0
= 0.3 are not accurate, while for all results there is little difference
between the two simplied equations (43) and (46). For decay rate, shown in the lower half of gure
2(b), the nominally more accurate equation (43) agrees closely with the full linear solution for all Froude
number, while the nominally low-inertia approximation (46) really is a low-inertia approximation for
upstream propagating waves. It is accurate for F
0
= 0.2 but again for F
0
= 0.3 it is no longer. In
any case, it can be seen that the decay rates for upstream travelling waves are rather greater than for
downstream propagation.
Combining the results from downstream and upstream propagation, we can conclude that the use of
the nominally low-inertia approximation (46) is preferable to the in-principal more accurate equation
(43). It seems generally to mimic the effects of time and space derivatives in the full equation that
both approximations have ignored. For downstream propagating waves the approximate equation (46) is
only that the change of boundary conditions (ood inow etc.) be slow, while for upstream propagating
waves both change should be slow and ow should be slow such that F
2
is small. In view of these
two slightly-different results for downstream and upstream propagation we will refer to equation (46) as
the slow-change/slow-ow momentum equation, where the terminology is an abbreviation for the rather
longer term "slow-change and possibly slow-ow momentum equation".
We can estimate typical values of wave period for which the simplied momentum equation should be
accurate. For example, in a channel of slope 10
4
and velocity 0.5 ms
1
, from equation (41) T = 20
corresponds, to a physical period T = 20 U
0
/2gS
0
20 0.5/(2 1010
4
) = 5000 s, or the time-
to-rise of a ood half of that, roughly 40 min. If the slope were 10
3
, the corresponding limitation (for
the same U
0
) would be about 4 min. For any changes slower than those the simplied theory should be
accurate. Of course, that limit can easily be calculated in any application.
As we obtained equation (47) with an approximation neglecting
2
and higher order terms, it is ap-
propriate to write the solution (48) as a power series, nominally in , but equation (48) shows that the
expression is a function of /F
2
0
so that this is the effective expansion parameter. The leading terms
17
Long waves in open channels their nature, equations, approximations, and numerical simulation John D. Fenton
are
=

+O

3
/F
4
0

+ i

F
2
0
2
( 1) +

3
F
2
0

2
+O

4
/F
6
0

. (49)
From the real part and equation (32), we have to lowest order, c = U
0
, as in equation (38a) from
the full linearised solution for downstream propagating waves, and also as an approximation to equa-
tion (38b) for F
2
0
small. The notable and surprising result here, although obvious on gure 2, is that
using the slow-change/slow-ow momentum equation (46), and the algebraic equivalent equation (47)
+i
2
F
2
0
= 0, gives the expected result to lowest order of = for wave propagation
downstream at physical velocity +U
0
but also, not at all obvious from the equation, for upstream prop-
agation, a velocity of U
0
as an approximation, so that very-long waves travel at the same speed in
both directions.
From the imaginary part of equation (49) similar results for the decay coefcient are obtained, which
are small F
2
0
approximations to equations (39) and (40).
7. The slow-change/slow-ow routing equation
We now examine the use of the slow-change/slow-ow momentum equation (46) and develop a simple
single equation which can be used for simulation purposes for longer waves, say for T & 20.
If the resistance is not a function of Q then equation (46) can be solved for Q
Q =
r
gA
3
P
r

S
1
B
A
x
= K
r

S
1
B
A
x
, (50)
where the conveyance K has been introduced for convenience, from equation (5). From here we will
not necessarily limit ourselves to Chzy-Weisbach friction, so that for K one could use the formulation
used above, where 8, or the Chzy form or the Gauckler-Manning formula:
K =
r
g

A
3
P
= C
r
A
3
P
=
1
n
A
5/3
P
2/3
.
To use equation (50) to give a single equation in a single unknown, there are two ways that we can
proceed.
Routing equations in terms of A(x, t) or (x, t): The rst is to substitute Q from equation (50)
into the mass conservation equation (1a) to give the equation
A
t
+

x

K
r

S
1
B
A
x
!
= i. (51)
Routing problems could be solved by solving for A down the channel and if necessary using equation
(50) at any stage to calculate the discharge Q. However the free surface elevation has more direct
signicance. Equations (14) and (16) of Fenton (2010b) are
A
t
= B

t
and
1
B
A
x
=

x
+

S,
provided the banks of the stream are not both vertical and converging/diverging. Equation (51) becomes
the even simpler equation in terms of surface elevation :

t
+
1
B

x

K
r

x
!
=
i
B
. (52)
In many problems, however, the discharge is specied as a boundary condition, which does not naturally
lead to a boundary condition on A or as we have considered here.
18
Long waves in open channels their nature, equations, approximations, and numerical simulation John D. Fenton
The slow-change/slow-ow routing equation in terms of upstream volume V (x, t): A more
general formulation is obtained if we use the upstream volume identity (11) for Q: Q = Q(x
0
, t) +
R
x
x
0
i(x
0
) dx
0
V/t used directly in equation (50) to give the partial differential equation in terms of
V :
V
t
+K (V
x
)
s

S
1
B(V
x
)

2
V
x
2
= Q(x
0
, t) +
Z
x
x0
i(x
0
) dx
0
, (53)
where we have shown that both K and B are functions of V
x
= V/x, as they are functions of ow
area. In practice, this equation would be solved by nite differences, stepping forward in time.
The author has used such an equation in several studies (, Fenton et al. 1999, Fenton & Keller 2001, Bar-
low et al. 2006) where its simplicity and/or its relative stability for zero or low ow was an advantage.
Here, the inclusion of the inow at the upstream boundary Q(x
0
, t) in the denition of V/t (equation
11) is more consistent because it now includes inow both at the upper end and distributed inow in the
same term, such that V really is the volume in the channel between the upstream boundary and the point
under consideration. This modication has been suggested by Fatemeh Soroush (2011, Personal Com-
munication). Previously the present author called the resulting equation the Volume Routing Equation.
Fatemeh Soroush also suggested that that is an ambiguous term in irrigation engineering. In view of that
and the discovery above of the real nature of the approximation to the momentum equation (46) here we
propose the name slow-change/slow-ow routing equation for equation (53).
Almost any common boundary condition can be used with the equation, as (equation 11) Q can be
expressed as a function of V/t and the surface elevation , via cross-sectional area Acan be expressed
as a function of V/x, which variables are often connected at control structures:
Upstream boundary: It is usually the upstream boundary condition that drives the whole model,
where a ood or wave enters, via the specication of the time variation of Q = Q(x
0
, t) at the bound-
ary. However, the boundary condition on V is exceptionally simple as there is no ow upstream of
the most upstream point, we have simply the boundary condition V (x
0
, t) = 0. The actual inow hy-
drograph appears instead in the partial differential equation (53). The surface elevation at the upstream
boundary is obtained as part of the computations, as if we know the variation of V with x (probably in-
terpolated from point values) at any time level, then from equation (11) we have A = V/x, and the
surface height comes from knowledge of the cross-section.
Downstream boundary: If a control exists: if there is a structure downstreamthat restricts or con-
trols the ow with a ow formula such as Q
out
= f(
out
), then as
out
is related to area A
out
=
V/x|
out
from the cross-section, and Q
out
is related to V/t|
out
, from equation (11), the discharge
equation for most boundary structures gives an equation for V/t|
out
in terms of V/x|
out
, which
can be evaluated as part of a numerical solution and the value of V updated there, in a similar sense
to equation (53) giving a value of V/t at interior points which would be similarly used to give V
at the next time level.
Open boundary condition: often the computational domain might be truncated without the presence
of a control structure on the stream. One solution is to use a uniform ow boundary condition there,
just as if it were a control. The author prefers a different approach, and this is simply to treat the
boundary as if it were just any other part of the river (which it is!) and to use the slow-change/slow-
ow routing equation to update V there, calculating the necessary x-derivatives from upstream nite
difference formulae. I have found that it works very well in practice, but a senior hydraulic engineer
was horried when he heard this, believing it to violate the physics of the problem. To me it is a
sensible step. One still has to truncate somewhere. If one has truncated the computational domain,
one has already abandoned any idea of information coming from downstream anyway. So, if all
information is coming from upstream, we just use that and the solution at the exit point from the
equations, using approximations to derivatives from the conditions immediately upstream. To me,
it is more sensible than applying the wrong boundary condition such as a uniform ow boundary
condition at the downstream end.
19
Long waves in open channels their nature, equations, approximations, and numerical simulation John D. Fenton
8. Numerical solution of the long wave equations an explicit
nite difference scheme
A Forward-Time-Centred-Space scheme is probably the simplest possible nite difference scheme for
numerical solution of the equations. Liggett & Cunge (1975) suggested that such a scheme was uncon-
ditionally unstable and named it as such. The author believes that the deductions from their analysis are
wrong. The scheme has a quite acceptable stability limitation, and it opens up the possibility of a much
simpler method for computations of oods and ows in open channels than the implicit methods often
used.
8.1 A Forward-Time-Central-Space scheme
I think I should put Manning in here
Consider the forward-time nite difference approximation to equations (3):
(x, t +) = (x, t) +

i
B

1
B
Q
x

, (54a)
Q(x, t +) = Q(x, t) +

Q
2
B
A
2

S P
Q|Q|
A
2
2
Q
A
Q
x

gA
Q
2
B
A
2

, (54b)
where all quantities on the right are evaluated at (x, t), and we use the central difference expressions for
the derivatives:

(x,t)
=
(x +, t) (x , t)
2
(55a)

Q
x

(x,t)
=
Q(x +, t) Q(x , t)
2
. (55b)
This combination of equations (54) for stepping forward in time using the central difference expressions
(55) for space derivatives, gives the scheme and its name.
8.2 Linear stability
Now we examine the stability properties of the scheme. We consider, in the same way as in 4, the case
of no inow i = 0, and small perturbations about a uniform ow such that we write
= h
0
S
0
x +h
1
and Q = Q
0
+Q
1
,
In fact, the linearisation would be the same as that done previously so that the linear equations in terms
of h
1
and Q
1
are obtained more simply by substituting A
1
= B
0
h
1
into the linearised equations (19)
plus from equation (22) and c
0
from equation (23), and we replace derivatives by the nite difference
approximations to give the linearised nite difference equations
B
0
h
1
(x, t +) h
1
(x, t)

+
Q
1
(x +, t) Q
1
(x , t)
2
= 0, (56a)
Q
1
(x, t +) Q
1
(x, t)

C
2
0

2
U
2
0

B
0
h
1
(x +, t) h
1
(x , t)
2
+2U
0
Q
1
(x +, t) Q
1
(x , t)
2
+ (Q
1
c
0
B
0
h
1
) = 0. (56b)
We look for solutions of the form

h
1
(x, t)
Q
1
(x, t)

=

h
10
Q
10

expikx expt,
20
Long waves in open channels their nature, equations, approximations, and numerical simulation John D. Fenton
where h
10
and Q
10
are coefcients and i=

1. This has assumed a periodic solution in x, expikx,
possibly as part of a Fourier series, as computations are performed on a nite length of channel. The as-
yet unknown behaviour in t is written as expt, where the nature of determines the solution behaviour
in time. Substituting into the equations (56a) and (56b), introducing the notation
K =
sin(k)

and M =
exp() 1

, (57)
and dividing through by common factors gives the matrix equation

M iK
i

C
2
0

2
U
2
0

K c
0
M + + 2iU
0
K

B
0
h
10
Q
10

=

0
0

.
For non-trivial solutions to exist, the determinant of the 2 2 matrix must be zero, giving the quadratic
in M:
M
2
+M (i2U
0
K +) + iKc
0
+K
2

C
2
0

2
U
2
0

= 0,
and solving for M and using equation (57) we have the solution for the multiplication factor at each time
step:
exp() = 1 /2 iKU
0

q
(/2)
2
K
2
C
2
0
iK (c
0
U
0
), (58)
where = 1 for downstream/upstream propagation. It is convenient to introduce the modied Froude
number F
0
0
= U
0
/C
0
, where from equation (20), C
0
=
q
gA
0
/B
0
+

U
2
0
; to use the relation-
ship (25) c
0
= U
0
; and to express K using the dimensionless-frequency-like term
=
2KC
0

=
2C
0

sin(k)

, (59)
so that the solution (58) becomes
exp() = 1

1 + iF
0
0
+
q
1
2
i2F
0
0
( )

. (60)
We write the square root term in standard complex form as
q
1
2
2iF
0
0
( ) = + i,
where =
1

2
p
+ 1
2
and =
1

2
p
1 +
2
,
in which =
q
(1
2
)
2
+ 4
2
F
02
0
( )
2
,
so that the solution is written
exp() = 1

1 + iF
0
0
+ ( + i)

.
The condition for stability of the numerical solutions is that the magnitude of the factor exp() by
which the solution changes at each time step must be less than or equal to 1, so that |exp()|
2
6 1.
Multiplying the solution for exp() by its complex conjugate then gives the condition for stability of
the FTCS scheme:
6 4
1 +
(1 +)
2
+ (F
0
0
+)
2
, (61)
giving a limiting value of . As and are functions of , the dimensionless frequency parameter
dened in equation (59), and F
0
0
( ) this condition (61) means that the stability criterion is a function
of , , F
0
0
and F
0
0
. As and are constant for practical purposes it is a function of , and F
0
0
. In
any particular problem we have to examine both values of = 1 and all the values of from equation
(59) over the range of values of k = 2m/L, where m can take any value between 1 and N/2 for N
computational points.
21
Long waves in open channels their nature, equations, approximations, and numerical simulation John D. Fenton
8.3 Approximate stability criteria
We obtain an approximation to the stability criterion (61) by writing it as a series in terms of Froude
number. It is necessary to consider two alternatives, depending on whether 1. The results are:
6
4
1 +

1
2
+O

F
02
0

for 6 1, (62a)
6
4

1 +O

F
0
0

for > 1. (62b)


It can be seen that neglected terms in the rst are proportional to the square of the Froude number, and
in the second, the Froude number itself. Usually we do not need a particularly accurate result for the
limiting step size, so these should be adequate for practical purposes. It is interesting that the inverse
time scale has again appeared, and that the criteria have been able to be expressed simply in terms of
it.
The most limiting case for the rst condition (69a) is when = +1. For we make the further
approximations that = 1 plus the wide-channel approximation such that A
0
/B
0
= A
0
/P
0
= d
0
,
where d
0
is the depth of ow. In equation (59) in this case C
0
=

gd
0
and from equation (22) =
2
p
gS
0

0
/d
0
, and we use, for k small, sin(k) / k = 2m/L, which in this case, the smallest
value possible is when m = 1, giving

min
=
2C
0

sin(k)

= 2
d
0
/L

S
0

0
From equation (8) we have for a wide channel

S
0
F
2
0
which gives

min
= 2F
0
d
0
/L
S
0
and is a minimum,
min
, for which m = 1
For
max
we take sin(k) = 1. This gives

max
=
d
0
/

S
0

0
= N
d
0
/L

S
0

0
= NF
0
d
0
/L
S
0

min
= 2
h
0
/L

S
0

0
, and (63a)

max
=
h
0
/

S
0

0
= N
h
0
/L

S
0

0
, (63b)
where we assume that N > 2 6.
In the limiting case of zero slope, S
0
0 the scheme is unstable, as
max
and equation (73)
shows that the time step for stability 0.
8.4 Test of linear stability
Now we test the stability theory numerically. To do this we considered a wide model stream of rectan-
gular section. Resistance coefcients and slopes were taken from the data given in section ??, generally
limited by the region shown dashed in gure **** ?? ****. Four values of resistance were used:
[0.003, 0.01, 0.03, 0.1], and six Froude numbers, [0.05, 0.1, 0.2, 0.4, 0.6, 0.8]. In each case the slope
22
Long waves in open channels their nature, equations, approximations, and numerical simulation John D. Fenton
0.003
0.01
0.03
0.1
10
5
10
4
10
3
10
2
1
10
100
300

(sec.)
From computations
Linear theory, eqn (61)
Approximations, eqns (72 & 73)

Slope

(sec.)
Figure 3. Limiting time step for stability, as determined theoretically and by performing computations
was determined from S = F
2
. It was thought not unreasonable to take depth of 1 m for a depth scale,
a large width of 100 m, a computational length L = 10 km, and N = 20 such that = 500 m.
To test the time step in actual calculations, the initial normal ow was calculated, and then a ood with a
peak at 24 h was introduced, but with a peak increase of only 1% of the normal ow to test the stability.
A series of calculations using the FTCS scheme were performed, varying the time step until the scheme
was just stable. Results are shown in gure 5. It can be seen that over the region of most streams as
given in gure ??, the computations were stable with time steps between 10 s and 180 s. In the limit of
very rough and steep streams with Froude numbers of up to 0.8, the limiting step went down to 4 s. Of
course, the implicit Preissmann Box scheme would allow much larger time steps.
The results from the linear theory, equation (61) with a scan of both values and all values of mbetween
1 and N/4 were found to be highly accurate. Also, the approximate theory presented here, equations
(72 & 73), gave results good enough for practical purposes. For large Froude numbers the results could
be in error by a factor of 2, but as an estimate only is required, that seems not to be a problem.
8.5 Determining stability limits for unsteady ow problems
In all the above we have considered only small variations about a uniform ow. Now we proceed to a
method by which we might estimate what is the time limitation for an unsteady ow computation. We
will base this on the assumption that the above theory holds even for the unsteady case.
If we consider the stability criterion (69a) for 6 1 and take the most stringent limitation, . 2.
Substituting equation (22) for and using the wide-channel approximation
.
s
h
0
gS
0

0
, (64)
and the minimum time step is dictated by the minimum depth. To express this in terms of discharge
we use the wide-channel approximation, from the Chzy-Weisbach expression (7), and introduce q
0
=
23
Long waves in open channels their nature, equations, approximations, and numerical simulation John D. Fenton
U
0
h
0
, the discharge per unit width, to give
h
0
=

q
2
0

0
gS
0

1/3
(65)
from which we deduce
6

q
0
(gS
0
)
2

1/3
, (66)
determined by the minimum ow.
Now we consider the other condition (73) for > 1. As we are considering approximations only, we
neglect the term in Froude number to give the condition . 4/
2
. Substituting equation (22) for ,
using =
max
from equation (71b) with the wide-channel approximation gives
. 2
s
h
0
S
0

0
g


h
0

2
. (67)
In this case the most demanding criterion is when h
0
is a maximum. Using equation (66) to express it in
terms of q
0
the criterion becomes
. 2
S
0

2
q
0
. (68)
It is interesting that resistance has disappeared completely.
The procedure we suggest tentatively to determine the time step such that computations are stable is
then:
1. With q
0
the minimum ow per unit width expected, use equation (69) to estimate h
0
then calculate

min
from equation (71a). If
min
6 1 then calculate the limiting value of from equations (64)
or (66).
2. With q
0
the maximum ow per unit width expected, use equation (69) to estimate h
0
then calculate

max
using equation (71b). If
max
6 1 the previous criterion applies, else if
max
> 1 then
calculate the limiting value of from equations (67) or (68).
3. The minimum of the two values determines the real limiting value of .
As an example, we consider the 100 m wide channel used above, but now with conditions more likely in
a practical problem: a longer stream, with L = 100 km, N = 100, and a ow that rises from 100 m
3
s
1
to a peak of 1000 m
3
s
1
after 24 hours and then diminishes again. We computed for 96 h. A number of
different cases of resistance and slope were considered. In each case the approximate limit to stability
was found by trial and error, and compared with the results from the procedure suggested immediately
above. Results are shown in gure 4. It can be seen that over the range of most natural rivers and
canals, the estimates of the approximate formulae here are quite sharp. For streams on a small slope,
they underestimated the stability limits, but the order of magnitude was correct. Computational times
for each simulation varied between less than a second to several seconds.
That is not a conclusive proof of the applicability of the limits in real streams and oods, of course,
but provides a guide to the estimation procedure that might be followed and to the sorts of time steps
expected: very small in the case of streams on small slopes, but nite for the majority of streams.
While the time steps obtained can sometimes be small, a typical run time on a personal computer was
only second(s). Such small time steps do not seem to be a problem in practice, and are a small price to
pay for being able to use such a simple scheme.
9. Characteristics
24
Long waves in open channels their nature, equations, approximations, and numerical simulation John D. Fenton
0.003
0.01
0.03
0.1
10
5
10
4
10
3
10
2
1
10
100

(sec.)
From computations
Approximations, eqns ()

Slope

(sec.)
Figure 4.
References
Barlow, K., Fenton, J. D., Nash, D. & Grayson, R. (2006), Modelling phosphorus trans-
port in a surface irrigation drain, Advances in Water Resources 29(9), 13831398.
http://johndfenton.com/Papers/Barlow06-Modelling-phosphorus-transport-in-a-surface-
irrigation-drain.pdf
Boussinesq, V. J. (1877), Essai sur la thorie des eaux courantes, Mmoires
prsents par divers savants lAcadmie des Sciences, Paris 23, 1680.
http://gallica.bnf.fr/ark:/12148/bpt6k56673076.r=Boussinesq.langEN.swf
Deymie, P. (1938), Propagation dune intumescence allonge (problme aval), in Proc. Fifth Int. Congr.
Appl. Mech., Cambridge, Mass., pp. 537544.
Fenton, J. D. (2010a), Calculating resistance to ow in open channels, Technical report, Alter-
native Hydraulics Paper 2. http://johndfenton.com/Papers/02-Calculating-resistance-to-ow-in-
open-channels.pdf
Fenton, J. D. (2010b), The long wave equations, Technical report, Alternative Hydraulics Paper 1.
http://johndfenton.com/Papers/01-The-long-wave-equations.pdf
Fenton, J. D. & Keller, R. J. (2001), The calculation of streamow from measurements of stage,
Technical Report 01/6, Cooperative Research Centre for Catchment Hydrology, Melbourne.
http://www.catchment.crc.org.au/pdfs/technical200106.pdf
Fenton, J. D., Oakes, A. M. & Aughton, D. J. (1999), On the nature of waves in
canals and gate stroking and control, in Proc. Workshop on Modernization of Irriga-
tion Water Delivery Systems, US Committee on Irrigation and Drainage, October 17-21,
Phoenix, pp. 343357. http://johndfenton.com/Papers/Fenton99Phoenix-On-the-nature-of-waves-
in-canals-and-gate-stroking-and-control.pdf
Iwasa, Y. (1954), The criterion for instability of steady uniform ows in open channels, Memoirs of the
Faculty of Engineering, Kyoto University 16(4), 264275.
Jeffreys, H. (1925), The ow of water in an inclined channel of rectangular section, Phil. Mag. (6)
49, 793807.
25
Long waves in open channels their nature, equations, approximations, and numerical simulation John D. Fenton
Keulegan, G. H. & Patterson, G. W. (1943), Effect of turbulence and channel slope on translation waves,
J. Res. Nat. Bureau Standards 30, 461512.
Liggett, J. A. (1968), Mathematical ow determination in open channels, J. Engng Mech. Div. ASCE
94(EM4), 947963.
Liggett, J. A. & Cunge, J. A. (1975), Numerical methods of solution of the unsteady ow equations, in
K. Mahmood & V. Yevjevich, eds, Unsteady Flow in Open Channels, Vol. 1, Water Resources
Publications, Fort Collins, chapter 4.
Lighthill, M. J. & Whitham, G. B. (1955), On kinematic waves. I: Flood movement in long rivers, Proc.
Roy. Soc. London Ser. A 229, 281316.
Lyn, D. A. (1987), Unsteady sediment-transport modeling, J. Hydraulic Engineering 113(1), 115.
http://link.aip.org/link/?QHY/113/1/1
Lyn, D. A. & Altinakar, M. (2002), St. VenantExner equations for near-critical and transcritical Flows,
J. Hydraulic Engineering 128(6), 579587. http://link.aip.org/link/?QHY/128/579/1
Ponce, V. M. (1990), Generalized diffusion wave equation with inertial effects, Water Resources Re-
search 26(5), 10991101.
Ponce, V. M. & Simons, D. B. (1977), Shallow wave propagation in open channel ow, J. Hydraulics
Div. ASCE 103(12), 14611476.
Strelkoff, T. S. & Clemmens, A. J. (1998), Nondimensional expression of unsteady canal ow, J. Irriga-
tion and Drainage Engng 124(1), 5962.
White, F. M. (2003), Fluid Mechanics, fth edn, McGraw-Hill, New York.
Woolhiser, D. A. &Liggett, J. A. (1967), Unsteady, one-dimensional owover a plane the rising hydro-
graph, Water Resources Research 3(3), 753771. http://dx.doi.org/10.1029/WR003i003p00753
Yen, B. C. (1991), Hydraulic resistance in open channels, in B. C. Yen, ed., Channel Flow Resistance:
Centennial of Mannings Formula, Water Resource Publications, Highlands Ranch, USA, pp. 1
135.
26

You might also like