You are on page 1of 12

Water Resources Management (2006) 20: 979–990

DOI: 10.1007/s11269-006-9018-2 
C Springer 2006

Approximate Methods for the Estimation of


Muskingum Flood Routing Parameters

JASEM M. AL-HUMOUD and ISMAIL I. ESEN


Civil Engineering Department, Kuwait University, P.O. Box 5969, Safat 13060, Kuwait
(e-mails: jasem@kuc01.kuniv.edu.kw, jasem@civil.kuniv.edu.kw)

(Received: 20 May 2005; in final form: 2 February 2006)

Abstract. There are a variety of techniques for estimating the parameters x and K of the Muskingum
method of flood routing. One common difficulty in all the approaches is that different storm sequences
along the same river reach would typically yield different parameter estimates. The a statistical analysis
of these parameters also shows that they are highly variable. As a result achieving of a high level
accuracy may not be the principle issue in describing x and K. This paper presents two approximate
methods for estimating these parameters rather easily. The first method requires the computation of
the slopes of the inflow and outflow hydrographs at their point of intersection, and the computation
of the maximum storage within the reach. The second method requires the computation of the inflow
and outflow hydrographs at two specific points. Three case studies investigated show that the first
method gives estimates for the Muskingum parameters comparable to those derived by traditional
estimation procedures for hydrographs showing linear characteristics.

Key words: K, x, flood routing, Muskingum method, parameter estimation

Introduction
The Muskingum method is a hydrologic river routing technique based on the
equation of continuity. Given the inflow at the upstream end of a river reach, the
outflow at the downstream end is expressed as
I1 + I2 Q1 + Q2 S2 − S1
− = (1)
2 2 t
where I is the inflow rate to the reach, Q the outflow; S the volume of water stored,
and t the time increment. The subscripts 1 and 2 denote the values of the respective
terms at the beginning and end of the time interval considered. The storage within
the reach is modeled by

S = S0 + K [x I n + (1 − x)Q n ] (2)

where S0 represents the initial storages, K the storage-time constant for the reach,
x a proportionality constant, and n an exponent used for considering the effects of
980 J. M. AL-HUMOUD AND I. I. ESEN

nonlinearity. In linear Muskingum models, the exponent n is taken as unity, and

S = S0 + K [x I + (1 − x)Q] (3)

It is convenient to set S0 equal to zero for the linear model.


The substitution of Equation (3) into Equation (1) gives the well known routing
equation:

Q 2 = C 0 I2 + C 1 I1 + C 2 Q 1 (4)

where

C0 = (−2K x + t)/(2K − 2K x + t) (5)


C1 = (2K x + t)/(2K − 2K x + t) (6)
C2 = (2K − 2K x − t)/(2K − 2K x + t) (7)

If the data on the inflow and outflow are sparse, then the value of K is taken as the
travel time in the reach, and x is assumed to have an average value of 0.2 (Viessman
and lewis, 2003). If both the inflow and outflow hydrographs are available, then the
Muskingum routing parameters can be determined by one of the several estimation
techniques reported in literature (HEC-HMS, 2001). In the traditional trial-and-
error graphical approach developed by McCarthy (1938) for the linear Muskingum
model, the so-called weighted discharge term [x I + (1−x)Q] is plotted against the
accumulated storage for different values of x assumed. Such a plot is referred to as
the “discharge-storage curve”. The particular value which generates the narrowest
loop is accepted as the best estimate of x. The loop is then approximated by a straight
line and, as Equation (3) stipulates the value for K is estimated as the reciprocal of
the slope of the line. Figure 1 illustrates the application of this procedure using the
inflow and outflow hydrographs (Wanielista, 1990; Bedient and Huber, 1992). For
this case, it is observed that the value of x which results in the narrowest loop is
about 0.3, and the value of K can be estimated as 1.1 days, which is the inverse of
the slope of the narrowest loop as shown in the figure.
Clearly the trial-and-error graphical procedure briefly described here is some-
what subjective and time consuming. For the most part, it has become obsolete.
To avoid subjective interpretations of data in estimating K and x, several alternate
numerical techniques have been developed. These include linear and nonlinear
regression schemes based on the least-squares method, orthogonal least-squares
methods, piecewise linear iterative regression methods, iterative optimization tech-
niques, methods of moments and cumulants, and others. The early linear models are
reviewed by Singh and McCann (1979), and the more recent linear and nonlinear
models were compared by Yoon and Padmanabhan (1993). Various other proce-
dures have also been developed for estimating the coefficients C0 , C1 and C2 without
APPROXIMATE METHODS FOR THE ESTIMATION OF MUSKINGUM flOOD 981

Figure 1. Traditional graphical procedure for the estimation of K and x (Wanielista, 1990;
Bedient and Huber, 1992).

prior knowledge of K and x (Stephenson, 1979). Similarly genetic algorithms have


also developed to estimate the parameters in nonlinear models (Mohan, 1997).
This paper develops procedures for estimating the parameters K and x rather
easily, and thus avoiding iterative estimation techniques. These procedures are
applied to three cases exhibiting different characteristics, and the results obtained
are compared with those obtained from a least-squares scheme. The confidence
regions for the parameters are also discussed.

Least-Squares Procedure
Gill (1978) developed a least-squares scheme in which the error sum of squares
was defined as the sum of the squares of the differences between the observed
and computed values of storage. Here, we shall modify Gill’s procedure slightly to
emulate McCarthy’s graphical procedure. First, Equation (3) is written as

y =a+b S (8)

where, y is the weighted discharge defined by

y = x I + (1 − x)Q (9)
982 J. M. AL-HUMOUD AND I. I. ESEN

and

1
b= (10)
K

The constant a in Equation (8) is treated as a statistical parameter. The cumulative


storage S can then be evaluated from computing Equation (1).
The function which has to be minimized by the least-squares scheme is


n
E y (a, b) = (yi − a − bSi )2 (11)
i=1

Where E y (a,b) is the sum of the squares of the difference between the observed and
computed values of the weighted discharge, yi is the observed weighted discharge
for a given value of x, Si is the observed value of storage and, n is the total number
of observations. The function E y defined by Equation (11) is usually referred to
as the error sum of squares. Again, for a given value of x, the values of a and
b for which E y becomes a minimum can be determined from ∂ E y /∂a = 0 and
∂ E y /∂b = 0. The solution of the resulting equations will lead to
         
yi Si − yi Si Si
a=   2    2 (12)
n Si − Si
      
n yi Si − yi Si
b=   2    2 (13)
n Si − Si

where the summations run from i = 1 to n.


As in the graphical procedure, a value of x is selected, the parameters a and b
are determined from Equations. (12) and (13), and then the error sum of squares is
computed from Equation (11). The procedure is repeated with different values of x,
and the value of x giving the minimum error sum of squares is accepted as the least-
squares estimate of x. The procedure of increasing x in increments until an increase
in E y is observed, and then halving the increment in x, and continuing backwards
in a similar fashion until the minimum value of E y is reached has been observed
to give extremely short computation times. The reciprocal of the corresponding b
value is then taken as the least-squares estimate of K.

Confidence Regions
Confidence contours are defined by considering E y as a constant. Although an exact
confidence level associated with a contour can be defined for linear least-squares,
for convenience we shall use the approximate 100 (1 − q)% confidence contours
APPROXIMATE METHODS FOR THE ESTIMATION OF MUSKINGUM flOOD 983

given by Draper and Smith (1981) as


 
p
Sq = E y 1 + F( p, n − p, 1 − q) (14)
n−p
where E y is the minimal sum of squares, p the number of model parameters, n
the number of observations, and F( p, n − p, 1 − q) represents the inverse of an F
distribution. In the case studies investigated here, it was observed that the parameter
a only changed very little with the value of x selected. This is expected so since
Equation (8) implies that a corresponds to the value of the weighted discharge
when S is zero, or simply the initial value of y. In the beginning of a storm, inflow
to the river reach equals the outflow for most cases, and the weighted discharge,
or the parameter a, simply equals the initial inflow. Thus, in order to avoid three-
dimensional representation for the confidence contours, we only consider the model
parameters x and K = 1/b. Consequently, the number of model parameters p = 2,
and so the inverse of the F distribution is given by
n − 2  −2/(n−2)
F(2, n − 2, 1 − q) = q −1 (15)
2
The substitution of Equation (15) into Equation (14) gives the approximate 100
(1 − q)% confidence contour as

Sq = E y q −2/(n−2) (16)

Using the definition of the weighted discharge by Equation (8), the confidence
contour is then given by

n
Sq = {(y − a)−bS}2 (17)
i=1

Solving this equation for b yields



  
(y − a)S ± [ (y − a)S]2 − [ S 2 ][ (y − a)2 − Sq ]
b=  2 (18)
S
Given value of Sq , the preceding equation yields true two values of K = 1/b for
delineating true corresponding confidence contour.

Approximate Methods
For the approximate procedures, our models are based on the limiting form of the
continuity equation given by Equation (1), i.e.
∂S
I−Q= (19)
∂t
984 J. M. AL-HUMOUD AND I. I. ESEN

The partial derivative of S with respect to time follows from Equation (3) as

∂S ∂I ∂Q
=K x + (1 − x) (20)
∂t ∂t ∂t

Now consider the inflow and outflow hydrographs for a river reach as shown in
Figure 1. At time t B when the inflow and outflow hydrographs intersect, I B = Q B ,
and Equation (19) gives (∂ S/∂t) B = 0. Thus, from Equation (20)

(∂ Q/∂t) B
x= (21)
(∂ Q/∂t) B − (∂ I /∂t) B

It is noted that Equation (21) can be regarded as a preliminary estimate of the


parameter x (Linsley and Franzini, 1972; Quimpo, 1992).
Next, noting that at t = t B , I B = Q B and storage is maximum. Thus, Equation
(3) leads to

Sm
K = (22)
IB

where Sm represents the maximum storage, shown by the shaded area in Figure 2.
At t = t A , (∂ I /∂t) A = 0, and Equations (19) and (20) give

∂Q
I A − Q A = K (1 − x) (23)
∂t A

Figure 2. Inflow and outflow hydrographs for a river reach.


APPROXIMATE METHODS FOR THE ESTIMATION OF MUSKINGUM flOOD 985

Similarly, at t = tC , (∂ Q/∂t)C = 0, and the same two equations lead to

IC − Q C = K x(∂ I /∂t)C (24)

Consequently, Equations (21) through (24) allow x and K to be estimated in various


ways.
But, because Equations (21) and (22) express these parameters in an uncoupled
manner, they will be used here in the forms given above. This approach will hereafter
be referred to as “method 1”. The simultaneous solution of Equations (23) and (24)
yields
1
x = (25)
(∂ I /∂t)C (IA − Q A )
1+ •
(∂ Q/∂t)A (IC − Q C )
(IA − Q A ) (IC − Q C )
K = + (26)
(∂ Q/∂t)A (∂ I /∂t)C

Estimating x and K based on the preceding equations will be referred to as method


2. Note that in all these equations, the time derivatives must be determined from the
inflow and outflow hydrographs shown in Figure 2 by either graphical or numerical
procedures.

Case Studies
Three case studies will be considered here. Case study 1 is based on the inflow and
outflow hydrographs exhibiting linear characteristics (Viessman and Lewis, 2003).
Case study 2 uses the data set from Wilson (1974). Data reported by Wilson are
known to present a nonlinear relationship between weighted discharge and storage.
This data set has also been extensively studied by others (Gill, 1978; Tung, 1984;
Yoon and Padmanabhan, 1993; and Mohan, 1997). The third case is based on the
data reported by Wu et al. (1985). It was shown by Wang and Sing (1992) that
the outflow hydrograph for this case could be best represented by Muskingum
parameters varying along the river reach.
Table I summarizes the values of x and K computed by the least-squares method,
and by the approximate methods 1 and 2 for these three cases. Further, Tables II,
III, and IV list the observed inflow and outflow, and the outflow predicted based on
these methods. Finally, Figure 3 delineates the contours for 95% confidence for x
and K determined by the least-squares method.

Discussion and Conclusion


It is mostly accepted that the effectiveness of the Muskingum method depends on
the accuracy with which its parameters are estimated (Singh and McCann, 1980).
However, different storm sequences for the same river reach would obviously yield
986 J. M. AL-HUMOUD AND I. I. ESEN

Table I. Muskingum parameters estimated by various methods

Least squares Method 1 Method 2

Data set x K x K x K
Viessman and 0.23 1.89 0.37 1.77 0.16 3.38
Lewis (2003)
(K in days)
Wilson (1974) 0.25 29.1 0.32 22.4 0.26 51.3
(K in hours)
Wu et al. (1985) 0.053 20.5 0.269 16.3 0.202 42.2
(K in hours)

Table II. Observed and computed values of outflow for data set given by Viessman and
Lewis (2003)

Computed Q (m3 /s)


t I Observed Q
(d) (m3 /s) (m3 /s) Least-squares Method 1 Method 2
0 166.2 118.4 118.4 118.4 118.4
1 263.6 197.4 146.1 138.7 131.6
2 365.3 214.1 209.6 206.3 170.0
3 580.5 402.1 296.4 284.1 226.0
4 594.7 518.2 442.2 466.3 331.9
5 662.6 523.9 522.4 539.3 409.8
6 920.3 603.1 602.7 590.9 482.5
7 1568.8 829.7 786.8 732.7 606.2
8 1775.5 1124.2 1193.6 1230.5 891.9
9 1489.5 1379.0 1481.7 1595.4 1159.4
10 1223.3 1509.3 1476.8 1555.4 1261.2
11 713.6 1379.0 1330.1 1398.6 1255.5
12 645.6 1050.6 1012.5 981.0 1094.2
13 1166.7 1013.7 842.3 723.4 954.1
14 1427.2 1013.7 1016.9 972.9 1014.8
15 1282.8 1013.7 1221.9 1268.0 1139.8
16 1098.7 1209.1 1246.9 1294.8 1184.6
17 764.6 1248.8 1160.0 1205.4 1162.7
18 458.7 1002.4 947.5 961.8 1047.0
19 351.1 713.6 693.9 660.6 872.2
20 288.8 464.4 516.5 475.0 717.0
21 228.8 325.6 398.1 365.5 589.6
22 170.2 265.6 309.5 286.5 482.3
23 143.0 222.6 237.4 217.1 389.2
APPROXIMATE METHODS FOR THE ESTIMATION OF MUSKINGUM flOOD 987

Table III. Observed and computed values of outflow for data set given by Wilson (1974)

Computed Q (m3 /s)


t I Observed Q
(h) (m3 /s) (m3 /s) Least-squares Method 1 Method 2
0 22 22 22.0 22.0 22.0
6 23 21 21.8 21.8 21.7
12 35 21 20.0 19.5 18.9
18 71 26 17.5 16.4 12.0
24 103 34 24.9 27.1 12.5
30 111 44 42.4 50.2 23.8
36 109 55 59.3 70.6 37.1
42 100 66 72.9 85.3 49.9
48 86 75 81.8 93.3 60.9
54 71 82 85.4 94.3 68.4
60 59 85 84.0 89.3 71.8
66 47 84 80.0 82.1 73.0
72 39 80 73.4 72.4 71.2
78 32 73 66.3 63.0 68.3
84 28 64 58.7 53.7 64.0
90 24 54 52.0 46.2 59.7
96 22 44 45.6 39.3 55.0
102 21 36 40.0 33.9 50.4
108 20 30 35.6 29.9 46.3
114 19 25 33.0 26.9 42.7
120 19 22 28.9 24.3 39.2
126 18 19 26.6 22.8 36.5

different Muskingum parameters. Also, the confidence contours shown in Figure 3


imply that it is possible to choose a wide range of values for x and K.
More specifically, the confidence contours shown in Figure 3 show that a multi-
tude pairs of x and K values would result in almost the same value of the error sum
of squares. This possibly indicates that the parameters of the Muskingum method
depend on each other. The shape of the confidence contours also suggests that x is
determined less precisely than K. It should be noted that both x and K values esti-
mated by the approximate method 1 lie almost within the 95% confidence contour
for the first two case studies. The value of x estimated by the approximate method
2 lies within this confidence contour for the first two case studies, but the value of
K does not. Both approximate methods do not work very well for the third case.
The comparisons of the computed values of outflow with the observed values
show that method 1 performs almost as well as the least-squares method. This
is particularly noticeable for the data set of Viessman et al. (1989) with linear
988 J. M. AL-HUMOUD AND I. I. ESEN

Table IV. Observed and computed values of outflow for data set given by Wu et al. (1985)

Computed Q (m3 /s)


t I Observed Q
(d) (m3 /s) (m3 /s) Least-squares Method 1 Method 2
0 1.87 1.87 1.87 1.87 1.87
6 7.08 2.89 2.31 1.39 1.09
12 15.57 5.24 4.32 2.89 0.79
18 18.85 7.50 7.61 7.69 2.71
24 11.89 9.49 10.02 12.82 6.40
30 8.35 10.48 10.22 12.78 7.83
36 5.95 10.42 9.52 11.22 8.28
42 4.16 8.78 8.41 9.27 8.17
48 2.83 6.94 7.16 7.34 7.71
54 2.10 5.66 5.94 5.59 7.02
60 1.70 4.67 4.88 4.22 6.28
66 1.44 3.74 4.00 3.23 5.57
72 1.30 2.83 3.31 2.52 4.91

characteristics. The computed values of outflow in Table I compare reasonably


well with the observed values even though there was more than one peak for the
inflow and outflow. On the other hand, none of the methods, including the least-
squares technique, perform well for the third case study, representing a routing
procedure with variable x and K values.
In the approximate method 1, the slopes of the inflow and outflow hydrographs
need be determined only in estimating x. As was noted earlier, the outflow hy-
drographs computed are not very sensitive to the estimated value of x. However,
true estimation of K requires the computation of the maximum storage. The latter
follows easily from the hydrographs given.
In view of the preceding results, and based on the case studies investigated, the
approximate method 1 based on Equations (21) and (22) can effectively be used
for the estimation of Muskingum routing parameters x and K.
Method 1 can also be used for the estimation of the exponent n characterizing a
nonlinear Muskingum model in the form of Equation (2), provided that two sets of
inflow and outflow hydrographs are given. Taking initial storage S0 = 0, and since
at maximum storage I B = Q B as shown in Figure 2, then

Sm = K I Bn (27)

Using the subscripts 1 and 2 to indicate the values related to the two different storms,
can now be determined in the form
n(Sm 1 /Sm 2 )
n= (28)
n(I B1 /I B2 )
APPROXIMATE METHODS FOR THE ESTIMATION OF MUSKINGUM flOOD 989

Figure 3. The 95% confidence contours for K and x. Data sets are from: (a) Viessman and
Lewis, 2003; (b) Wilson, 1974; (c) Wu et al. (1985).

The principal advantage of the approximate methods are that the Muskingum pa-
rameters can be directly estimated from the properties of the inflow and outflow
hydrographs. Thus eliminating the need for trial-and-error and other types of nu-
merical solutions. On the other hand, their accuracy depends on the reliability of
the plotted hydrographs. Computations needed for the slopes of the hydrographs
990 J. M. AL-HUMOUD AND I. I. ESEN

are also be prone to subjective interpretation. The two approximate methods do


not rely on all the data provided in determining the two parameters, x and K . In
particular, method 1 uses mostly the data segment before the intersection point
of the inflow and outflow hydrographs. Method 2 uses two relatively small data
segments. However, if points B and C defined in Figure 2 coincide, then the latter
method would not work.

Acknowledgements
The two approximate methods discussed in this paper were suggested by the late
Professor Mahdi S. Hantush (Abdulla, 1983). In addition, the authors would like
to thank the distinguished reviewers of the journal for their useful comments and
suggestions.

References
Abdulla, S. A. S., 1983, Flood Routing Using Muskingum Method. Unpublished Project Report
Supervised by M.S. Hantush, Department of Civil Engineering, Kuwait University, Kuwait.
Bedient, P. B. and Huber, W. C., 1992, Hydrology and Flood Plain Analysis, 2nd edition, Addison-
Wesley, New York.
Draper, N. R. and Smith, H., 1981, Applied Regression Analysis, 2nd ed. John Wiley, New York.
Gill, M. A., 1978, ‘Flood routing by the muskingum method’, Journal of Hydrology 30, 47–61.
HEC-HMS, 2001, U.S. Army Corps of Engineers, Hydrologic Engineering Center, Davis, California,
USA.
Linsley, R. K. and Franzini, J. B., 1972, Water-Resources Engineering, 2nd ed. McGraw-Hill, New
York.
McCarthy, G. T., 1938, The Unit Hydrograph and Flood Routing, Presented at Conf. North Atlantic
Div., U.S. Army Corps Eng. (unpublished).
Mohan, S., 1997, ‘Parameter estimation of nonlinear Muskingum models using genetic algorithm’,
J. Hydraul. Eng. ASCE 123(2), 137–142.
Quimpo, R. G., 1992, Hydrology and hydraulics, Department of Civil Engineering, University of
Pittsburgh, Pittsburgh.
Singh, V. P. and McCann, R. C., 1980, ‘Some notes on Muskingum method of flood routing’, Journal
of Hydrology 48, 343–361.
Stephenson, D., 1979, ‘Direct optimization of Muskingum routing coefficients’, J. Hydrol. 36,
353–363.
Tung, Y. K., 1984, ‘River flood routing by nonlinear Muskingum method’, ASCE J. Hydraul. Div.
111(12), 1447–1460.
Viessman, Jr., W. and Lewis, G. L., 2003, Introduction to Hydrology, Pearson Education, Inc., Upper
Saddle River, New Jersey, USA.
Wanielista, M., 1990, Hydrology and Water Quantity Control, John Wiley and Sons, New York.
Wang, G. T. and Singh, V.P., 1992, ‘Muskingum method with variable parameters for flood routing
in channels’, J. Hydrol. 134, 57–76.
Wilson, E.M., 1974, Engineering Hydrology, MacMillan Education Ltd., Hampshire, U.K.
Wu, Jy. S., King, E. L. and Wang, M., 1985, ‘Optimal identification of Muskingum routing coeffi-
cients’, Water Resour. Bull. 21(3), 417–421.
Yoon, J. and Padmanabhan, G., 1993, ‘Parameter estimation of linear and nonlinear Muskingum
models’, Journal of Water Resources Planning and Management, ASCE 119(5), 600–610.

You might also like