Download as pdf or txt
Download as pdf or txt
You are on page 1of 20

Muskingum river routing with dynamic bank storage

A.L. Birkhead
*
, C.S. James
Centre for Water in the Environment, University of the Witwatersrand, Private Bag 3, WITS 2050, South Africa
Received 31 August 2001; revised 19 February 2002; accepted 15 March 2002
Abstract
The Muskingum unsteady ow routing method is well established in the hydrological literature and its modest data
requirements make it attractive for practical use. Muskingum routing is based on an assumed power relationship between
channel storage and river stage, and consequently, it accounts for channel storage only and not bank storage. Routing
parameters in the models are usually derived by calibration using measured discharge hydrographs. Bank storage is, therefore,
implicitly accounted for, together with channel storage, but this may result in unrealistic parameter values. In this paper,
Muskingum procedures are modied to explicitly account for the interaction between channel ow and bank storage in rivers
with permeable river banks of varying hydraulic conductivity. The approach is veried by resynthesising a rating function used
in rigorous routing in a hypothetical channel with permeable banks. The method is also applied to determine routing parameters
for a reach along the Sabie River in South Africa by explicitly accounting for bank storage. q 2002 Elsevier Science B.V. All
rights reserved.
Keywords: Hydrological ow routing; Hydraulic data; River bank storage
1. Introduction
Flood routing in open channels may be undertaken
using a variety of modelling procedures. These range
in simplicity from Muskingum-type approximations
(McCarthy, 1938) (which have modest data require-
ments), to MuskingumCunge methods (Cunge,
1969) (where the typically calibrated Muskingum
routing parameters are related to physical and
hydraulic characteristics such as reach length, ood
wave celerity, unit width discharge and channel bed
slope (Ponce, 1979)), to the solution of the full
dynamic ow (St Venant) equations (which require
surveyed cross-sectional channel proles and ow
resistance data). The Muskingum and Muskingum
Cunge methods are well established in the hydro-
logical literature (Aldama, 1990; Gill, 1992; Koussis,
1978; Ponce and Yevjevich, 1978; Singh and
McCann, 1980), and the modest data requirements
make these procedures attractive even though more
rigorous hydraulic models are available for unsteady
ow routing.
In the nonlinear form of Muskingum routing (Gill,
1978; Yoon and Padmanabhan, 1993), storage is
related to weighted exponential functions of the
discharges owing into and out of a river reach. The
nonlinear function arises from assumed power
relationships between channel storage and river
stage, and discharge and river stage. Consequently,
Muskingum procedures account explicitly for channel
storage only, and not total storage along a river reach
0022-1694/02/$ - see front matter q 2002 Elsevier Science B.V. All rights reserved.
PII: S0022- 1694( 02) 00068- 9
Journal of Hydrology 264 (2002) 113132
www.elsevier.com/locate/jhydrol
*
Corresponding author. Present address: Streamow Solutions,
P.O. Box 752273, Gardenview 2047, South Africa.
E-mail addresses: streamow@icon.co.za (A.L. Birkhead);
csj@civil.wits.ac.za (C.S. James).
which may include lateral inows (tributary or diffuse
inputs) or outows, losses (recharge of groundwater
aquifers or evapotranspiration), and temporal changes
in bank storage. Parameters in the Muskingum models
are typically derived by calibration using measured
inow and outow discharge hydrographs which do,
however, relate to changes in total storage along a
river reach. Traditionally, total storage is, therefore,
implicitly accounted for in the calibration of the
routing parameters for the river channel.
Birkhead and James (1998) modied the tra-
ditional nonlinear Muskingum routing equations to
synthesise the rating relationship (relationship
between stage and discharge) based on a measured
short-term local stage hydrograph at the site of
interest, and a corresponding discharge hydrograph
at a remote site along the river. Application of the
procedures to a section of the Sabie River (South
Africa) showed that neglecting bank storage resulted
in poor optimisation of the storage relationships and
unrealistic estimates of the storage weighting factor.
The procedures were successfully modied to account
for bank storage by assuming instantaneous response
of seepage in the alluvial bank zone, this being
justied by the high hydraulic conductivity of the
uvial sediments. The procedures described by Birk-
head and James (1998) are further developed here to
explicitly account for the interaction between channel
Nomenclature
a
1
proportionality constant in discharge rating function (m
2
/s)
a
2
proportionality constant in storage rating function (m
2
)
C
1
; C
2
; C
3
constants in linear Muskingum ood routing equation
E objective function in optimisation procedure
h piezometric potential (m)
I
t
inow discharge at time t (m
3
/s)
k constant in linear Muskingum ood routing equation
K hydraulic conductivity (m/s)
l length of channel reach (m)
L lateral extent of the permeable river bank (m)
m initial depth of saturation in the river bank (m)
m
1
exponent in discharge rating function
m
2
exponent in storage rating function
nt number of data points on ood hydrograph
O
t
outow discharge at time t (m
3
/s)
Q
b
base ow discharge (m
3
/s)
Q
p
peak ow discharge (m
3
/s)
Qs
t
bank seepage (m
3
/s)
Q
t
discharge at time t (m
3
/s)
S storage coefcient given by the effective porosity
S
1
prism storage corresponding to initial base ow (m
3
)
S
t
storage at time t (m
3
)
t time (s)
t
b
time taken for discharge to return to base ow (s)
t
p
time corresponding to peak ow (s)
T transmissivity (m
2
/s)
x Muskingum storage weighting factor, horizontal distance from the river bank (m)
y
t
river stage at time t (m)
z distance in the vertical direction (m)
Dt time-step between data points on ood hydrograph (s)
A.L. Birkhead, C.S. James / Journal of Hydrology 264 (2002) 113132 114
ow and bank storage in rivers with permeable river
banks of any hydraulic conductivity. The effects of
accounting for bank storage explicitly and implicitly
(as in traditional Muskingum routing) on the esti-
mation of the nonlinear Muskingum routing par-
ameters from measured discharge hydrographs are
also considered.
2. Muskingum models
The net storage along a river reach can be related to
the inow and outow discharges through the
continuity equation. The storage at any time t can be
expressed in nite-difference form as
S
t

t
i2

I
i
2

O
i

_ _
Dt S
1
1
in which S
1
is the storage at time t 1; Dt is the
period between times t and t 21; and

I
i
and

O
i
are the
mean inow and outow discharges over Dt,
respectively.
For channel storage, S
t
can also be estimated using
Muskingum approximations (as originally proposed
by McCarthy (1938)), developed from the simple
exponential relationships between discharge and
stage, and prismatic storage and stage given by Eqs.
(2) and (3), respectively
Q
t
a
1
y
t
m
1
2
S
t
a
2
y
t
m
2
3
in which Q
t
is the discharge and y
t
is the stage at time
t.
Combining Eqs. (2) and (3) and estimating the
storage using a weighted average of the prismatic
upstream and downstream ow conditions leads to the
Muskingum approximation for channel storage
S
t

a
2
a
1
m
2
=m
1
xI
t
m
2
=m
1
1 2xO
t
m
2
=m
1
_ _
4
and the following approximate form of Eq. (4) (Gill,
1978; Tung, 1985; Yoon and Padmanabhan, 1993)
S
t

a
2
a
1
m
2
=m
1
xI
t
1 2xO
t

m
2
=m
1
5
where x is the storage weighting factor. Eq. (4) or (5)
can be solved conjunctively with Eq. (1) for
hydrological ood routing.
Birkhead and James (1998) modied Eqs. (1), (4)
and (5) to synthesise the rating relationship (i.e. to
provide estimates of the coefcients a
1
and m
1
in Eq.
(2)) based on a measured short-term local stage
hydrograph at the site of interest, and a corresponding
discharge hydrograph at a remote site along the river.
Additional data requirements include a cross-sec-
tional survey at the local site and an estimate of the
distance along the river between the two sites. Three
procedures were described for optimising the par-
ameters of the rating curve, with two of these based on
the nonlinear forms of the Muskingum channel
storage function and applicable to any ow sequence
(i.e. Eq. (4) (Model 2) and Eq. (5) (Model 1) with
stage y
t
rather than discharge O
t
recorded at the
downstream section). The third (Model 3) presented
by Birkhead and James (1998) does not depend on this
function; it is easier to apply, but requires a complete
event bounded by steady ows. Verication of the
approach using rigorous ood routing in a hypothe-
tical channel produced results from the rst two
models (1 and 2) that are virtually indistinguishable
and compare closely with those of the third.
Application of the procedures to a section of the
Sabie River, however, showed that neglecting bank
storage provided unrealistic estimates of the storage
weighting factor x, and poorly optimised temporal
storage responses. This shortcoming was successfully
overcome by including bank storage, which could be
assumed to respond instantaneously to channel
discharge because the course sand deposits along the
Sabie River have a high hydraulic conductivity. The
assumption of instantaneous response enabled bank
storage to be easily accounted for by increasing the
effective width of the channel. Assuming instan-
taneous bank storage response is equivalent to
accounting for it implicitly in traditional hydrological
routing. Routing coefcients are, however, usually
expressed as lumped parameters (Eqs. (4) and (5)),
and parameter values are not evaluated for use in a
rating function (Eq. (2)).
A more general approach to account for the
inuence of bank storage is described below, whereby
a temporal bank storage function is added to the
nonlinear Muskingum model for channel storage (i.e.
Eq. (4) or (5)) to estimate total reach storage. The
A.L. Birkhead, C.S. James / Journal of Hydrology 264 (2002) 113132 115
method is applicable to river banks of any hydraulic
conductivity, and requires specication of three
additional input parameters (Fig. 1): the hydraulic
conductivity of the river banks, their lateral extent,
and the depth to an impermeable lower boundary (e.g.
bedrock). This method has been applied to two cases.
(i) The Muskingum routing parameters (a
1
; m
1
and x )
are optimised for the case considered by Birkhead and
James (1998) where discharge and stage hydrographs
are provided at remote upstream and local down-
stream sites, respectively (Model 4). (ii) The effects
are examined of, both explicitly accounting for, and
neglecting bank storage in the calibration of routing
coefcients from eld data in traditional Muskingum
routing, i.e. where discharge hydrographs are
measured at both the upstream and downstream
river sites (Model 5) (The above convention for
model numbering serves to prevent confusion with the
numbers used by Birkhead and James (1998)).
Realistic addition of the bank storage contribution
to the channel storage estimate in Eq. (3) requires an
expression for bank storage in terms of ow depth.
The complexity of the function dening the bank
storage should be consistent with that provided by the
Muskingum approximation for channel storage (e.g.
also as a simple power function). This is not easily
achievable, however, since bank seepage (and hence
storage) is a function of antecedent ow conditions in
rivers. Furthermore, the bank storage expression
should ideally be differentiable with respect to the
Muskingum routing parameters for ease of inclusion
in the optimisation procedure described by Birkhead
and James (1998).
Various numerical models for computing the
temporal distribution of storage in a vertical cross-
section through a river bank are described in the
literature (e.g. Hornberger et al., 1970; Birkhead,
2000). These are not appropriate for the present
application, however, since the nite-difference
spatial approximations applied require highly com-
putational solution procedures. The analytical one-
dimensional saturated model developed by Cooper
and Rorabaugh (1963) is well suited, but limited in
practical application, since the analytical solution was
derived for a sinusoidal stage hydrograph described
by a specic mathematical function. Hall and Moench
(1972) and Moench et al. (1974) derived an
approximate solution for one-dimensional saturated
ow in a homogeneous and isotropic system in
response to arbitrary uctuations of river stage using
the convolution integral as given by Schwarz and
Friedland (1965). The relative simplicity of this
solution scheme makes it appropriate for modifying
Muskingum routing procedures to account for tem-
poral bank storage along rivers.
3. Convolution integral applied to bank seepage
The mass balance equation for one-dimensional
transient ow is given by Jacob (1950) as

2
h
x
2

S
T
h
t
6
in which h is the piezometric head, x is the horizontal
distance from the river bank not to be confused with
the Muskingum storage weighting factor, S is the
storage coefcient given by effective porosity after
gravity drainage has materially ceased, T is the
transmissivity given by the product of the hydraulic
conductivity K and the depth of saturation m; and t
is time (refer to Fig. 1).
Fig. 1. Bank characteristics for the hypothetical ood routing in rectangular channel.
A.L. Birkhead, C.S. James / Journal of Hydrology 264 (2002) 113132 116
The analytical model of Cooper and Rorabaugh
(1963) applies constant transmissivity based on initial
saturated depth, and the numerical model of Horn-
berger et al. (1970) uses a variable transmissivity
using the piezometric head (i.e. T Kh). For the case
of a nite river bank of lateral extent L and assuming
constant transmissivity, the following expression for
the bank seepage per unit length of river at time t may
be derived from the relationships given by Moench
et al. (1974) and Hall and Moench (1972)
Qs
t
2
2T
L

t
i2
_
y
i
2y
i21
Dt

1
n1
exp
_
2
_
2n 21p
2L
_
2
TDt
S
t 2i 0:5
__
7
where negative and positive seepage values represent
recharge of and discharge from bank storage,
respectively.
The bank storage per unit length of river at time t,
relative to the storage at time t 1; is, therefore,
given by
S
t

t
i2
Qs
i
Dt 8
in which the average seepage calculated from the
average ow depths between successive time-steps is
used.
Although the computational bank storage models
described by Birkhead (2000), Hornberger et al.
(1970) and Cooper and Rorabaugh (1963) are
inappropriate for computing bank storage in the
Muskingum routing procedure, they provide useful
means for assessing the performance of the approxi-
mate convolution method (Eq. (7)) for one-dimen-
sional ow and constant transmissivity. Comparisons
of the predictive performances of the numerical
solutions schemes (Birkhead, 2000; Hornberger
et al., 1970) and the analytical solution scheme of
Cooper and Rorabaugh (1963) has been described by
Birkhead (2000). Generally, the analytical model
(Cooper and Rorabaugh, 1963) tends to underestimate
the seepage into and out of the bank zone due to the
assumption of a constant transmissivity, necessary to
derive the analytical solution. Hornberger et al. (1970)
demonstrated that increased error in prediction is
concomitant with increased relative difference
between peak and initial stages (approximately 30%
for a three-fold change in river stage), with acceptable
results (error ^15%) produced for ratios of up to 1.5.
Fig. 2 is a plot of the bank seepage response to a
uctuation of river stage (inset) predicted by the
above four bank storage models for K 10 m=day;
S 0:30 and L 100 m; and nite-difference
Fig. 2. Comparison of bank seepage responses computed using the approximate convolution method, the exact analytical solution (Cooper and
Rorabaugh, 1963), and two numerical models (Birkhead, 2000; Hornberger et al., 1970).
A.L. Birkhead, C.S. James / Journal of Hydrology 264 (2002) 113132 117
temporal and spatial parameter values of Dt 0:1 days,
Dx 5 m and Dz 0:1 m (where z denotes distance in
the vertical direction). For the case analysed, the
approximate convolution method (Eq. (7)) underesti-
mates the seepage by up to 5%compared with the exact
analytical solution of Cooper and Rorabaugh (1963).
Validation of a two-dimensional numerical model by
Birkhead (2000) showed that modelled bank seepage
rates reduce when large vertical gradients arise, and also
when owin the unsaturated zone is accounted for. This
implies that under certain conditions, the approximate
convolution method (i.e. Eq. (7)) may actually give
results that are more accurate than those produced using
the more rigorous one-dimensional model of Hornber-
ger et al. (1970).
The Muskingum approximations for channel
storage (Eqs. (4) and (5)) may, therefore, be modied
to predict reach storage by adding the bank storage
contribution estimated by the convolution function
(Eqs. (7) and (8)).
4. Modications to the Muskingum models and
parameter estimation
Two models are considered in the modication of
the Muskingum procedure to account for bank
storage: Model 4 (an extension to Model 2 of
Birkhead and James (1998) using Eq. (4)) which
considers inow discharge and outow stage hydro-
graphs, and Model 5 which deals with the more
traditional Muskingum parameter estimation pro-
cedure where discharge hydrographs are specied at
both the upstream and downstream ends of a river
reach. The extension to Eq. (4) presented in this paper
(i.e. Model 4) is equally applicable to Eq. (5) which is
a simplication of the more rigorous form of the
weighted reach storage relationship (Eq. (4)). In both
the cases (Models 4 and 5), the Muskingum
parameters are determined by optimising an objective
function to minimise the difference between estimated
and measured (or data-based in the case of Model 4)
reach storage.
4.1. Model 4: measured inow discharge and outow
stage hydrographs
The objective function (E) is dened as the sum of
the squares of differences between the reach storage as
calculated from the measured data (Eq. (1)) and as
estimated from the weighted average of the channel
and bank storages calculated using the upstream and
downstream ow conditions, i.e.
E

nt
t2
_
Dt

t
i2

I
i
2
a
1
2
Dt

t
i2
_
y
i
m
1
y
i21
m
1
_
S
1
2a
2
_
x
_
I
t
a
1
_
m
2
=m
1
1 2xy
t
m
2
_
2
4TlDt
L

t
i2

i
j2
x
_

I
j
a
1
_
1=m
1
2
_

I
j21
a
1
_
1=m
1
Dt
1 2x
y
j
2 y
j21
Dt

1
n1
exp
__
2n 21p
2L
_
2
TDt
L
i 2j 0:5
__
2
9
The Muskingum storage weighting factor (x ) is
applied to both the channel and bank estimates of
storage in Eq. (9), and symmetrical cross-sectional
bank geometry is considered although this may be
adjusted for right and left banks with different
geometric and hydraulic characteristics. The
rationale for using equivalent weightings for the
bank and channel storages is discussed later (model
verication).
The unconstrained optimisation procedure of
Fletcher and Reeves (1964) for arbitrary differentiable
functions has been used to compute the calibration
parameter values (a
1
; m
1
and x ) that minimise the
objective function value (Eq. (9)). This requires the
gradients of the objective function, which are
provided in Appendix A.
4.2. Model 5: measured inow and outow discharge
hydrographs
The traditional method for determining Muskin-
gum coefcients using actual inow and outow
discharge hydrographs is also considered in the
modied approach, since it is valuable to assess the
A.L. Birkhead, C.S. James / Journal of Hydrology 264 (2002) 113132 118
effect of bank storage on the parameter values of
the calibration constants. The objective function is
given by
E

nt
t2
_

t
i2

I
i
2

O
i
Dt S
1
2
a
2
a
1
m
2
=m
1
_
xI
t
m
2
=m
1
1 2xO
t
m
2
=m
1
_
2
4TlDt
L

t
i2

i
j2
x
_

I
j
a
1
_
1=m
1
2
_

I
j21
a
1
_
1=m
1
Dt
1 2x
_

O
j
a
1
_
1=m
1
2
_

O
j21
a
1
_
1=m
1
Dt

1
n1
exp
__
2n 21p
2L
_
2
TDt
L
i 2j 0:5
__
2
10
The gradients of the objective function with respect to
the calibration parameters are given in Appendix A.
5. Model verication
The performances of the two models have been
assessed by applying them to resynthesise a stage
discharge relationship from information generated by
rigorous routing in a hypothetical situation. A single
ood event was routed along a prismatic rectangular
channel section, using an explicit nite-difference
solution scheme for the unsteady ow equations
(proposed by Koutitas (1983) and presented by
Chadwick and Morfett (1993)), with the addition
that bank seepage (computed using Eq. (7)) has also
been included in the rigorous routing. The ood event
was dened by a sinusoidal discharge hydrograph
specied at the upstream end of the reach, Mannings
equation was specied as the downstream boundary
condition, and the effect of different values for the
hydraulic conductivity of the bank material were
investigated. The channel characteristics, hydraulic
data, and nite-difference routing parameters used in
the unsteady ow computations are listed in Table 1.
The bank properties are given in Table 2 for the ve
cases analysed. The hydraulic conductivities con-
sidered range from impermeable material (Case 1) to
those characteristic of a coarse sand/ne gravel (Cases
2 and 5) (Cernica, 1982), to allow the performance of
the models to be assessed for a range of conditions.
The inow and outow discharge hydrographs for
Cases 1, 2 and 5 are shown in Fig. 3. The effects of the
hydraulic characteristics of the bank sediments on
hydrograph shape are apparent: there is a signicant
increase in peak attenuation and a marginal increase
in lag time with higher conductivities, and the rate of
rise and of recession are reduced. The more extensive
banks considered in Case 5 act to increase the bank
storage only slightly, indicating that the phreatic
Table 1
Data for unsteady ow computations in rectangular channel
Channel characteristics
Width 10 m
Reach length 1250 m
Gradient 0.001
Mannings n 0.030
Hydraulic data
Q
p
50 m
3
/s
Q
b
5 m
3
/s
t
p
1.5 h
t
b
3.0 h
Finite-difference routing parameters
Dx 250 m
Dt 20 h
Table 2
Bank characteristics for unsteady ow computations in rectangular channel
Case 1 Case 2 Case 3 Case 4 Case 5
Hydraulic conductivity (m/s) 0 0.01 0.001 0.0001 0.01
Extent of bank/side (m) 10 10 10 100
Effective porosity 0.3 0.3 0.3 0.3
A.L. Birkhead, C.S. James / Journal of Hydrology 264 (2002) 113132 119
surface response within the bank does not extend
signicantly beyond 10 m for the ow hydrograph
analysed. Rorabaugh (1963) provides the following
criterion for when a nite system can effectively be
treated as innite in extent
L .

0:2Tt
S
_
11
For Case 5, Eq. (11) implies that the bank storage is
not recharged beyond 30 m from the channel bank
over a 3-day time period, and this supports the small
difference in outow hydrographs between Cases 2
and 5 for a ten-fold increase in the lateral extent of the
bank.
Fig. 4 is a plot of the discharge hydrographs and
bank seepage for Case 2, illustrating the delayed
response of bank seepage compared with surface
channel ow: the outow hydrograph falls from t
1:7 days, whilst the net response of the river banks
Fig. 3. Inow and outow discharge hydrographs for the bank characteristics given in Table 2.
Fig. 4. Inow, outow and bank seepage for Case 2 (Table 2).
A.L. Birkhead, C.S. James / Journal of Hydrology 264 (2002) 113132 120
along the reach is to commence discharging water
from storage at t 2 days.
The parameter optimisation technique described
by Birkhead and James (1998) has been applied to
determine the parameter values of the calibration
coefcients (a
1
; m
1
and x ), with the gradients of
the objective function (E) provided by the
equations in Appendix A for Models 4 and 5.
For Model 5, inow and hydraulically routed
outow discharge hydrographs (as shown in Fig.
3) provide the input data; whereas for Model 4,
the routed outow stage hydrograph was used.
The relationship between the outow discharge
and stage hydrographs is provided by Mannings
resistance equation, which supplied the down-
stream boundary condition in the rigorous routing
procedure. The optimised rating curve parameters
obtained for each of the cases using the two
parameter estimation models are presented in
Table 3. Also presented in Table 3 are the
parameter values corresponding to Mannings
relationship, expressed in the form of Eq. (1).
For each case analysed, implicit (assuming bank
storage is included in surface channel storage) and
explicit (incorporating Eqs. (7) and (8) in the total
storage estimate) account is taken of bank storage.
Table 3
Optimised rating curve parameters for rectangular channel
Model 4 Model 5
Implicit Explicit Implicit Explicit
a
1
m
1
x a
1
m
1
x a
1
m
1
x a
1
m
1
x
Case 1 9.452 1.484 0.284 9.452 1.484 0.284 9.411 1.487 0.330 9.411 1.487 0.330
Case 2 11.067 1.366 0.597 9.461 1.488 0.312 6.543 1.406 20.107 9.263 1.490 0.297
Case 3 9.955 1.452 0.362 9.456 1.485 0.296 7.296 1.546 20.033 9.388 1.486 0.323
Case 4 9.612 1.474 0.306 9.453 1.484 0.287 8.650 1.509 0.199 9.403 1.486 0.327
Case 5 11.054 1.389 0.543 9.438 1.491 0.314 4.718 1.589 20.459 9.372 1.487 0.315
Parameters derived from Mannings equation n 0:03
a
1
9.293
m
1
1.512
Fig. 5. Model 4 optimised storage data for Case 2 (Table 2) with implicit and explicit account taken of bank storage.
A.L. Birkhead, C.S. James / Journal of Hydrology 264 (2002) 113132 121
(Note that the implicit model is effectively the
explicit model with impermeable river banks, i.e.
T 0).
5.1. Model 4
Fig. 5 is a graphical illustration of the optimised
storage estimates for Case 2 using Model 4, showing
both the implicit and explicit situations. The sums of
the weighted discharge and bank storage estimates
dened by Eq. (12) are plotted against the data-based
estimates of reach storage dened by Eq. (13)
S
t
2a
2
_
x
_
I
t
a
1
_
m
2
=m
1
1 2xy
t
m
2
_
2
4TlDt
L

t
i2

i
j2
x
_

I
j
a
1
_
1=m
1
2
_

I
j21
a
1
_
1=m
1
Dt
1 2x
y
j
2 y
j21
Dt

1
n1
exp
__
2n 21p
2L
_
2
TDt
L
i 2j 0:5
_
12
S
t

t
i2

I
i
2
a
1
2

t
i2
y
i
m
1
y
i21
m
1
_ _
_ _
Dt S
1
13
For the implicit situation, the discrepancy between the
storage estimates and the relationship indicating
equality of estimates (i.e. x y) shows that the
model is not representing the reach storage correctly.
Comparison of the optimised values for the cali-
bration parameters in Table 3 with the values derived
from Mannings equation and Case 1 (no bank
seepage) indicates that a
1
and x are increasingly
overestimated and m
1
underestimated with increasing
bank storage. This is supported by the results of the
optimisation for the Sabie River site where bank
storage was neglected (Birkhead and James, 1998).
The differences between the calibrated values for
permeable (Cases 25) and impermeable (Case 1)
bank conditions reduce as the hydraulic conductivity
reduces. The linear relationship achieved between the
two estimates of reach storage when bank storage is
explicitly accounted for (Fig. 5) conrms that the
storage description is more realistic. Comparison of
the optimised parameter values in Table 3 with those
derived from Mannings equation indicate that taking
bank storage into consideration provides more
accurate optimised parameter values.
Fig. 6 is a plot of the resynthesised rating curves for
the rectangular channel using coefcients in Table 3 for
the following cases: (i) derived from Mannings
Fig. 6. Model 4 synthesised rating curves for selected bank characteristics compared with Mannings uniform ow relationship.
A.L. Birkhead, C.S. James / Journal of Hydrology 264 (2002) 113132 122
equation; (ii) impermeable river bank (Case 1); and
(iii) derived by neglecting bank storage in Case 2
(where the deviation from Mannings relationship is
the largest out of the cases analysed). It is interesting to
note that although neglecting bank storage results in
overestimates and underestimates of a
1
and m
1
(19 and
12% for Case 2) respectively, the net effect on the
resynthesised rating curve is not as pronounced.
Rather, optimisation of the data-based and weighted
estimates of reach storage result in an increase in the
storage weighting factor x (2.1 times for Case 2), and
values higher than those expected occur (x is generally
expected to be between 0 and 0.5, e.g. Tung, 1985). This
is supported by the analysis for the Sabie River site
neglecting bank storage, where a value of 1.517 resulted
(Table 6). Model 4 optimisation with bank storage
neglected, therefore, appears to be a relatively robust
means of calibrating rating curve coefcients, even
thoughreachstorage is not well represented(as shownin
Fig. 5). Prediction of a more linear rating curve (m
1
is
underestimated in Table 3) implies that extrapolation of
the stage-discharge relationship will be increasingly
inaccuratebeyondtherangeofows usedintheanalysis.
The relative differences in ow depth predicted
using the optimised rating curve parameters and
parameter values derived from Mannings equation
are listed in Table 4. The differences are given
separately for the range of discharges associated with
the ow event (550 m
3
/s) and for lower and higher
(up to 100 m
3
/s) extrapolations. For the implicit
analysis, differences of up to 46 and 5% result for
the lower and higher extrapolations, respectively,
while up to a 16% difference in ow depth occurs
within the range of discharges used for the event.
These values reduce considerably when explicit
account is taken of bank storage (up to 9, 2 and 2%
for the three discharge ranges), similar to those
associated with the impermeable bank condition.
The storage components (channel, bank and reach)
are plotted against time in Fig. 7 for Case 2 using
rigorous unsteady ow routing (solution of the St
Venant equations) and the explicit and implicit Model
4 optimisation. Good agreement between the rigorous
routing and explicit optimisation is achieved, whilst
poor representation of reach storage occurs for the
implicit situation. For the implicit model, the
difference between the predicted reach and channel
storages are a consequence of the residual error
between the two storage estimates as plotted in Fig. 5.
The bank storage contributions are also plotted using
only the upstream x 1 and downstream x 0
ow conditions. Comparison between the optimised
bank storage estimate and the bank storage derived
from the rigorous routing supports the use of the
channel storage weighting factor for proportioning
bank storage.
5.2. Model 5
Application of Model 5 optimisation results in
remarkably different ndings. For the ve cases
analysed, neglecting bank storage (implicit situation)
results in underestimates of a
1
and x, and both
underestimates and overestimates in m
1
(Table 3). The
negative values obtained for the weighting factors are
unexpected. Taking bank storage into account,
Table 4
Relative differences (%) for discharge ranges (m
3
/s) in ow depth predicted using the optimised rating curve parameters and parameter values
derived from Mannings equation given in Table 3 (05, 550, 50100 are discharge ranges in m
3
/s)
Model 4 Model 5
Implicit Explicit Implicit Explicit
05 550 50100 05 550 50100 05 550 50100 05 550 50100
Case 1 9.4 1.9 1.9 9.4 1.9 1.9 8.2 1.5 1.9 8.2 1.5 1.9
Case 2 46.2 15.7 4.5 8.3 1.8 1.4 24.4 39.6 44.9 6.4 1.9 2.6
Case 3 21.2 6.2 1.9 9.1 1.9 1.8 29.4 18.0 14.1 8.4 1.4 2.2
Case 4 13.2 3.3 1.9 9.4 1.9 1.9 4.8 5.1 5.2 8.5 1.5 2.0
Case 5 41.3 14.8 2.6 7.3 1.6 1.2 91.5 56.2 45.2 8.0 1.3 2.2
A.L. Birkhead, C.S. James / Journal of Hydrology 264 (2002) 113132 123
however, produces realistic values for all calibration
parameters. For the implicit model, the net effect on
the resynthesised rating curves are increasingly over-
predicted ow depths concomitant with larger bank
storage contributions (Fig. 8 and Table 4). This arises
from the need to incorporate bank storage within the
channel storage estimate, and this is accommodated in
the model by providing erroneously elevated ow
depths. Plots of the storage estimates in Fig. 9,
however, show that the model provides a reasonably
accurate estimate of reach storage. This is attributed to
the fact that the data-based estimate of reach storage is
measured, and does not incorporate calibration
parameters as in Model 4. Implicit application of
Model 5 is, therefore, particularly robust when used
for routing ows (due to its accurate representation of
weighted reach storage based wholly on measured
data), even though the routing may provide inaccurate
individual calibration parameters, thereby, yielding
inaccurate rating relationships.
6. Model application: Sabie River data revisited
The rating curve for the Sabie River site was
synthesised by Birkhead and James (1998) with
account taken of bank storage by assuming instan-
taneous response. The application is revisited here
applying the Model 4 optimisation procedure, with the
bank characteristics provided in Table 5.
The hydraulic conductivity and effective porosity
values have been obtained from the calibration of a
quasi three-dimensional bank seepage model for the
Sabie River site (Birkhead, 2000), and the average
lateral extent of the banks is as determined previously
(Birkhead and James, 1998) from analysis of scale
1:10 000 aerial photographs. Delineation of ground-
water ow boundaries at the site using ground
penetrating radar and physical probing techniques
(Birkhead et al., 1996) revealed that the channel bed
lies approximately 1 m above bedrock, and this value,
together with the stage of zero discharge at the cross-
section provide an estimate of 1.56 m for the depth of
saturation (m) at the initial discharge (4.7 m
3
/s).
Fig. 10 is a plot of the weighted estimates of
channel plus bank storage against the data-based
reach estimate for the Sabie River site using the
Model 4 optimisation procedure. The relationship
between the two estimates of storage conrms that the
Fig. 7. Model 4 optimised storage relationships for Case 2 (Table 2) with implicit and explicit account taken of bank storage.
Table 5
Bank characteristics for the Sabie River analysis
Hydraulic conductivity 0.014 m/s
Extent of bank/side 62 m
Effective porosity 0.34
Saturated depth 1.56 m
A.L. Birkhead, C.S. James / Journal of Hydrology 264 (2002) 113132 124
Fig. 8. Model 5 synthesised rating curves for selected bank characteristics compared with Mannings uniform ow relationship.
Fig. 9. Model 5 optimised storage relationships for Case 2 (Table 2) with implicit and explicit account taken of bank storage.
A.L. Birkhead, C.S. James / Journal of Hydrology 264 (2002) 113132 125
Fig. 10. Optimised storage data plotted using the coefcients produced by applying Model 4 to the Sabie River ood data, with bank storage
modelled using the convolution method.
Fig. 11. Model 4 optimised storage relationships for the stretch of Sabie River, showing monitored, instantaneous and modelled bank storage
variations at the downstream site.
Table 6
Optimised rating coefcients for the Sabie River site
Parameter Model 4 Model 2, instantaneous bank storage
response (Birkhead and James, 1998)
Regression
Implicit Explicit
a
1
14.541 13.855 13.983 12.587
m
1
1.711 2.048 2.278 2.173
x 1.517 0.511 0.297
A.L. Birkhead, C.S. James / Journal of Hydrology 264 (2002) 113132 126
storage description is realistic. The optimised rating
coefcients are provided in Table 6, together with the
regression coefcients determined by tting a
relationship of the form given by Eq. (2) to the
point rating data correlated between an upstream
gauge and downstream site. The coefcients obtained
by assuming instantaneous bank storage response are
also provided.
The rigorous modelling of bank storage using
the convolution method (Eq. (7)) reduces the
storage weighting factor from 1.517 to a more
realistic value of 0.511, which is higher than that
obtained by assuming instantaneous bank storage
response (0.297). This is expected since the latter
assumption represents a lower limit for bank
storage, and model verication has shown that
neglecting bank storage results in higher esti-
mations of the weighting factor. Fig. 11 is a plot
of the optimised storage components (reach,
channel and bank) for the length of river analysed.
The bank storage estimates using (assumed)
instantaneous, monitored, and modelled (Eqs. (7)
and (8) with no optimisation required since ow
depth has been recorded) responses based on the
downstream ow conditions are also plotted, for
comparison. The results clearly indicate that
Model 4 provides an improved representation of
the bank storage than that achieved by assuming
instantaneous response. It is interesting to note
that bank storage along the 4.6 km reach of the
Sabie River accounts for up to 40% of the
temporal reach storage, indicating the signicance
of bank storage within this system.
Fig. 12 is a plot of the synthesised rating curves and
regression relationship using the (unattenuated) point
rating data. Application of Model 4 and explicitly
accounting for bank storage produces a function that
agrees closely with the regression, whilst neglecting
bank storage (Model 4, implicit) provides a good t to
the weir data at low ows (when minimal attenuation
of ow peaks is expected). This supports the earlier
nding that Model 4 provides a robust means of
calibrating rating coefcients: it provides a reasonable
characterisation of the rating relationship when bank
storage is neglected, but improved accuracy when
bank storage is explicitly accounted for by applying
the convolution method.
7. Muskingum ood routing equation with bank
storage
The extended Muskingum models described in this
paper have been developed and tested for the purpose
of calibrating routing parameters, and thereby rating
curve coefcients, for rivers with permeable banks.
To use these calibrated parameters to route hydro-
graphs through a river reach, it is necessary to derive
Fig. 12. Comparison of the Model 4 synthesised rating curves and unrouted ow data, for the Sabie River site.
A.L. Birkhead, C.S. James / Journal of Hydrology 264 (2002) 113132 127
the modied nonlinear Muskingum ood routing
equation for a river with bank storage.
The continuity equation for the change in storage
between consecutive time-steps is given by
S
t
2S
t21
Dt

I
t
I
t21
2
2
O
t
O
t21
2
14
An estimate of reach storage is provided by adding the
nonlinear Muskingum equation for channel storage
(Eq. (4)) and the bank storage equation (Eq. (8), with
Eq. (7) expressed in terms of discharge rather than
ow depth), giving
S
t

a
2
a
1
m
2
=m
1
xI
t
m
2
=m
1
1 2xO
t
m
2
=m
1
_ _

4TlDt
L

t
i2

i
j2
x

I
j
a
1
_ _
1=m
1
2

I
j21
a
1
_ _
1=m
1
Dt
1 2x

O
j
a
1
_ _
1=m
1
2

O
j21
a
1
_ _
1=m
1
Dt

1
n1
exp
2n 21p
2L
_ _
2
TDt
L
i 2j 0:5
_ _
15
Substituting for S
t
and S
t21
in Eq. (14) using Eq. (15)
leads to
O
t

2a
2
a
1
m
2
=m
1
Dt
1 2xO
t
m
2
=m
1

8Tl
a
1
1=m
1
LDt

t
i2
1 2x

O
i
1=m
1

1
n1
exp
2n 21p
2L
_ _
2
TDt
L
t 2i 0:5
_ _
I
t
I
t21
2O
t21
2
2a
2
a
1
m
2
=m
1
Dt
xI
t
m
2
=m
1
2xI
t21
m
2
=m
1
21 2xO
t21
m
2
=m
1
_ _
2
8Tl
a
1
1=m
1
LDt

t
i2
x

I
i
1=m
1
2

I
i21
1=m
1
_ _
21 2x

O
i21
1=m
1

1
n1
exp
2n 21p
2L
_ _
2
TDt
L
t 2i 0:5
_ _
16
which is the nonlinear Muskingum ood routing
equation with bank storage. Eq. (16) requires the
outow at every time-step O
t
to be determined
through trial and error due to the nonlinear charac-
teristic of the relationship. A method to compute the
outows may readily be coded using standard
iterative solution procedures. For example, Yoon
and Padmanabhan (1993) used the NONLR algorithm
(iterative optimisation of multivariate parameter
estimation of nonlinear models) for the case without
bank storage.
In the linear form of the Muskingum storage
model, the parameter values of m
1
and m
2
are assumed
to be equivalent, and k is substituted for a
2
=a
1
: If bank
storage is also neglected (by substituting T 0), Eq.
(16) leads to the well-known Muskingum ood
routing equation which may be solved directly for
O
t
; given by
O
t
C
1
I
t
C
2
I
t21
C
3
O
t21
17
in which
C
1
2
kx 20:5Dt
k 2kx 0:5Dt
;
C
2

kx 0:5Dt
k 2kx 0:5Dt
;
C
3

k 2kx 20:5Dt
k 2kx 0:5Dt
18
8. Conclusions
Muskingum routing principles can be modied
to explicitly account for bank storage by applying
the convolution method for one-dimensional tran-
sient bank seepage. This modication was applied
to one of the nonlinear forms of the Muskingum
storage function to determine routing coefcients
for the cases where measured outows from a
river reach are provided by either stage or
discharge hydrographs. The extended models
have been veried by comparison with rigorous
A.L. Birkhead, C.S. James / Journal of Hydrology 264 (2002) 113132 128
ood routing in a rectangular channel incorporating
permeable banks. The results show that accurate
representation of both reach storage and the
calibration parameters are produced by explicitly
accounting for bank storage in the two models
developed. When bank storage is neglected, the
model utilising inow discharge and outow stage
hydrographs produces reasonable predictions of
the rating curve coefcients, inates the value of
the storage weighting factor, and provides poor
representation of the reach storage. For the
traditional Muskingum routing model utilising
inow and outow discharge hydrographs, neglect-
ing bank storage results in inaccurate estimates of
the rating curve coefcients, reduces the value of
the storage weighting factor, but provides good
representation of the temporal reach storage. The
former model, therefore, provides a robust means of
assessing the rating curve coefcients, and the latter
for routing discharge through river channels.
The modied procedure is applied to a section
of the Sabie River with the hydraulic conductivity
and effective porosity of the bank alluvium
provided by calibration of a dynamic bank storage
model, and the depth of saturation taken from eld
measurements. The resynthesised rating curve at
this section agrees closely with comparable point
data collected over a number of years. The
modelled variation in bank storage for the section
also compares well with monitored changes over
the duration of the ow event. This application
shows that bank storage accounts for up to 40% of
the temporal reach storage, indicating the signi-
cance of bank storage along the length of the Sabie
River.
Acknowledgments
This study formed a part of a research project
funded by the Water Research Commission (South
Africa). The project was supported logistically by
the National Parks Board. We thank the reviewers
Professors Denis Hughes and Jozsef Szilagyi for
detailed scrutiny of the paper and valuable
suggestions.
Appendix A.
A.1. Model 4
The gradients of the objective function (Eq. (9))
with respect to the calibration parameters a
1
; m
1
and x
are given by
E
a
1

nt
t2
_
22E
t
_
Dt
2

t
i2
_
y
i
m
1
y
i21
m
1
_
2
m
2
a
2
x
m
1
a
1
m
2
m
1
=m
1
I
t
m
2
=m
1
2
4TlDt
L
x
m
1
a
1
1m
1
=m
1

t
i2

i
j2

I
j
1=m
1
2

I
j21
1=m
1
Dt

1
n1
exp
__
2n 21p
2L
_
2
TDt
L
i 2j 0:5
___
E
m
1

nt
t2
_
22E
t
_
a
1
Dt
2

t
i2
y
i21
m
1
ln y
i21
y
i
m
1
ln y
i
2
a
2
m
2
x
m
1
2
_
I
t
a
1
_
m
2
=m
1
ln
_
I
t
a
1
_
2
4TlDt
L
x
m
1
2

t
i2

i
j2

_

I
j
a
1
_
1=m
1
ln
_

I
j
a
1
_
1=m
1
2
_

I
j21
a
1
_
1=m
1
ln
_

I
j21
a
1
_
1=m
1
Dt

1
n1
exp
__
2n 21p
2L
_
2
TDt
L
i 2j 0:5
___
A.L. Birkhead, C.S. James / Journal of Hydrology 264 (2002) 113132 129
E
x

nt
t2
_
22E
t
_
a
2
_
I
t
a
1
_
m
2
=m
1
2a
2
y
t
m
2

4TlDt
L

t
i2

i
j2
_

I
j
a
1
_
1=m
1
2
_

I
j21
a
1
_
1=m
1
Dt

1
n1
exp
__
2n 21p
2L
_
2
TDt
L
i 2j 0:5
___
with
E
t

_
Dt

t
i2

I
i
2
a
1
2
Dt

t
i2
_
y
i
m
1
y
i21
m
1
_
S
1
2a
2
_
x
_
I
t
a
1
_
m
2
=m
1
1 2xy
t
m
2
_
2
4TlDt
L

t
i2

i
j2
x
_

I
j
a
1
_
1=m
1
2
_

I
j21
a
1
_
1=m
1
Dt
1 2x
y
j
2 y
j21
Dt

1
n1
exp
__
2n 21p
2L
_
2
TDt
L
i 2j 0:5
__
2
A.2. Model 5
The gradients of the objective function (Eq. (10)) with respect to the calibration parameters a
1
; m
1
and x are
given by
E
a
1

nt
t2
_
22E
t
_
2
a
2
m
2
m
1
a
1
m
1
m
2
=m
1
_
xI
t
m
2
=m
1
1 2xO
t
m
2
=m
1
_
2
4TlDt
L
1
m
1
a
1
1m
1
=m
1

t
i2

i
j2
x

I
j
1=m
1
2

I
j21
1=m
1
Dt
1 2x

O
j
1=m
1
2

O
j21
1=m
1
Dt

1
n1
exp
__
2n 21p
2L
_
2
TDt
L
i 2j 0:5
___
A.L. Birkhead, C.S. James / Journal of Hydrology 264 (2002) 113132 130
E
m
1

nt
t2
_
22E
t
_
2
a
2
m
2
m
1
2
_
x
_
I
t
a
1
_
m
2
=m
1
ln
_
I
t
a
1
_
1 2x
_
O
t
a
1
_
m
2
=m
1
ln
_
O
t
a
1
__
2
4TlDt
L
1
m
1
2

t
i2

i
j2
x
_

I
j
a
1
_
1=m
1
ln
_

I
j
a
1
_
2
_

I
j21
a
1
_
1=m
1
ln
_

I
j21
a
1
_
Dt
1 2x
_

O
j
a
1
_
1=m
1
ln
_

O
j
a
1
_
2
_

O
j21
a
1
_
1=m
1
ln
_

O
j21
a
1
_
Dt

1
n1
exp
__
2n 21p
2L
_
2
TDt
L
i 2j 0:5
___
E
x

nt
t2
_
22E
t
_
a
2
a
1
m
2
=m
1
_
I
t
m
2
=m
1
2O
t
m
2
=m
1
_

4TlDt
L

t
i2

i
j2
_

I
j
a
1
_
1=m
1
2
_

I
j21
a
1
_
1=m
1
Dt
2
_

O
j
a
1
_
1=m
1
2
_

O
j21
a
1
_
1=m
1
Dt

1
n1
exp
__
2n 21p
2L
_
2
TDt
L
i 2j 0:5
___
with
E
t

_

t
i2

I
i
2

O
i
Dt S
1
2
a
2
a
1
m
2
=m
1
_
xI
t
m
2
=m
1
1 2xO
t
m
2
=m
1
_
2
4TlDt
L

t
i2

i
j2
x
_

I
j
a
1
_
1=m
1
2
_

I
j21
a
1
_
1=m
1
Dt
1 2x
_

O
j
a
1
_
1=m
1
2
_

O
j21
a
1
_
1=m
1
Dt

1
n1
exp
__
2n 21p
2L
_
2
TDt
L
i 2j 0:5
__
2
A.L. Birkhead, C.S. James / Journal of Hydrology 264 (2002) 113132 131
References
Aldama, A.A., 1990. Least-squares parameter estimation for
Muskingum routing. J. Hydraul. Engng 116 (4), 580586.
Birkhead, A.L., 2000. Interaction between channel ow and bank
storage in rivers. Unpublished PhD Thesis. Faculty of
Engineering, University of the Witwatersrand, Johannesburg,
South Africa.
Birkhead, A.L., James, C.S., 1998. Synthesis of rating curves from
local stage and remote discharge monitoring using nonlinear
Muskingum routing. J. Hydrol. 205, 5265.
Birkhead, A.L., Heritage, G.L., White, H., van Niekerk, A.W.,
1996. Ground-penetrating radar as a tool for mapping the
phreatic surface, bedrock prole, and alluvial stratigraphy in the
Sabie River, Kruger National Park. J. Soil Water Cons. 51 (3),
234241.
Cernica, J.N., 1982. Geotechnical Engineering, CBS College
Publishing, New York, NY, 488 pp.
Chadwick, A., Morfett, M., 1993. Hydraulics in Civil and
Environmental Engineering, second ed, Chapman & Hall,
London, 557 pp.
Cooper, H.H. Jr., Rorabaugh, M.I., 1963. Changes in ground-water
movement and bank storage caused by ood waves in surface
streams. US Geol. Surv. Prof. Pap. 475-B, B192B195.
Cunge, J.A., 1969. On the subject of ood propagation compu-
tational method (Muskingum method). J. Hydrol. Res. 7 (2),
205230.
Fletcher, R., Reeves, C.M., 1964. Function minimisation by
conjugate gradients. Comput. J. 7, 149154.
Gill, M.A., 1978. Flood routing by the Muskingum method.
J. Hydrol. 36, 353363.
Gill, M.A., 1992. Numerical solution of Muskingum equation.
J. Hydraul. Engng 118 (5), 804809.
Hall, F.R., Moench, A.F., 1972. Application of the convolution
equation to streamaquifer relationships. Water Resour. Res. 8
(2), 487493.
Hornberger, G.M., Erbert, J., Remson, I., 1970. Numerical solution
of the Boussinesq equation for aquiferstream interaction.
Water Resour. Res. 6 (2), 601608.
Jacob, C., 1950. Flow of groundwater. In: Rouse, H., (Ed.),
Engineering Hydraulics, Wiley, New York.
Koussis, A.D., 1978. Theoretical estimations of ood routing
parameters. J. Hydraul. Div., Proc. ASCE 104 (HY10),
109115.
Koutitas, C.G., 1983. Elements of Computational Hydraulics,
Pentech Press, Plymouth.
McCarthy, G.T., 1938. The unit hydrograph and ood routing.
Unpublished Paper. Conference of North Atlantic Division, US
Army Corps of Engineers, New London, CT. US Engineering
Ofce, Providence RI.
Moench, A.F., Sauer, V.B., Jennings, M.E., 1974. Modication of
routed streamow by channel loss and base ow. Water Resour.
Res. 10 (5), 963968.
Ponce, V.M., 1979. Simplied Muskingum routing equation.
J. Hydraul. Div., ASCE 105 (HY1), 8591.
Ponce, V.M., Yevjevich, V., 1978. MuskingumCunge method
with variable parameters. J. Hydraul. Div., Proc. ASCE 104
(HY21), 16631667.
Rorabaugh, J.I., 1963. Estimating changes in bankstorage and
groundwater contribution to streamow. Int. Assoc. Sci. Hydrol.
Publ. 63, 432441.
Schwarz, R.J., Friedland, B., 1965. Linear Systems, McGraw-Hill,
New York.
Singh, V.P., McCann, R.C., 1980. Some notes of Muskingum
method of ood routing. J. Hydrol. 48, 343361.
Tung, Y.K., 1985. River ood routing by nonlinear Muskingum
method. J. Hydraul. Engng. 111 (12), 14471460.
Yoon, J., Padmanabhan, G., 1993. Parameter estimation of linear
and nonlinear Muskingum models. J. Water Resour. Planning
Mgmt 119 (5), 600610.
A.L. Birkhead, C.S. James / Journal of Hydrology 264 (2002) 113132 132

You might also like