Professional Documents
Culture Documents
Sushil K. Singh
To cite this article: Sushil K. Singh (2005) Clark's and Espey's unit hydrographs vs the gamma
unit hydrograph / Les hydrogrammes unitaires de Clark et de Espey vs l'hydrogramme unitaire de
forme loi gamma, Hydrological Sciences Journal, 50:6, -1067, DOI: 10.1623/hysj.2005.50.6.1053
SUSHIL K. SINGH
National Institute of Hydrology Roorkee, Roorkee 247 667, Uttar Anchal, India
sukusi1@yahoo.com
Open for discussion until 1 June 2006 Copyright 2005 IAHS Press
1054 Sushil K. Singh
bénéficier ainsi d’une interprétation conceptuelle. Cela peut donc également être
utilisé pour élaborer un HUS.
Mots clefs hydrogramme unitaire de Clark; hydrogramme unitaire de Espey; loi gamma;
modélisation pluie–débit; hydrogramme unitaire synthétique; hydrogramme unitaire
INTRODUCTION
BACKGROUND
Clark’s method uses a time–area curve and its routing through a linear reservoir that
has a constant storage factor for calculating a runoff hydrograph resulting from a given
rainfall excess. Hence, the intermediate steps and the use of convolution in calculating
the UH can be avoided. However, a UH, being a basic property of a catchment, needs
to be calculated and stored for further application to a different rainfall excess, using
convolution. The routing of a time–area curve for an instantaneous unit effective
rainfall, through a linear reservoir, gives an instantaneous UH (IUH). This IUH is then
used to obtain a UH of specified duration.
Espey et al. (1977) and Espey & Altman (1978) proposed an empirical method to
obtain a 10-min SUH for urban watersheds; the method is based upon the concept of
Snyder’s UH. They proposed several empirical equations; the two that are used in this
paper are:
31620 A0.96
Qp = (1)
t1p.07
3.1L0.23 φ1.57
tp = (2)
S a0.25 I a0.18
where Qp is the peak of UH (cfs—cubic feet per second), tp is the time to peak of UH
(min), L is the mean length of the stream (ft), Sa is the average slope of the main
channel, A is the basin area (mi2), Ia is the percentage of impervious area in the
catchment, and φ is a dimensionless conveyance factor that depends upon the weighted
main channel Manning’s roughness coefficient. The other empirical equations pro-
posed by Espey et al. (1977) and Espey & Altman (1978) are for the time base of the
UH and hydrograph widths at 50% and 75% of the UH peak, to define the UH. The
time base of the UH is defined as the hydrograph width at 0% of the UH peak. They
proposed manual sketching through the points defined by the empirical equations with
few trials so that the area under the UH is unity.
The two-parameter gamma distribution has long been used for modelling event-based
rainfall–runoff processes (e.g. Edson, 1951; Nash, 1958). In the present context, a two-
parameter gamma distribution is expressed as:
( n −1) t
1 æ t ö −
q (t ) = ç ÷ e K
(3)
KΓ ( n ) è K ø
where n and K are parameters that define the shape and scale of the gamma
distribution; and q is the IUH (runoff depth resulting from effective rainfall in the form
of Dirac delta-function, cf. Dooge, 1959). In conceptual terms, n is the number of
linear reservoirs with equal storage coefficient K (Nash, 1958, 1959; Dooge, 1959).
These parameters have been referred to as Nash-model parameters in the subsequent
literature. The time to peak can be obtained by equating the first derivative of q with
respect to t, to zero (e.g. see Nash, 1960; Wu, 1963):
t p = (n − 1) K (4)
The other works on the use of the gamma IUH are by Nash (1960), Wu (1963), Valdes
et al. (1979), Croley (1980), Aron & White (1982), Collins (1983), Rosso (1984),
Hann et al. (1994), and Singh (1998, 2000, 2004). The procedure outlined below is for
obtaining a gamma UH from a SUH or a UH derived using other methods. While
proposing the transmutability of different synthetic unit hydrographs into a gamma
distribution, Singh (2000) showed that the gamma distribution is defined by the
dimensionless parameter β = qptp (where qp and tp are the peak and time to peak of the
UH) and tp. He proposed the following equations for computing the ordinates of a
gamma UH:
7
n = 2πβ 2 + (5)
6
Q q
= = e ( n −1) α (6)
Qp q p
t æ t ö
α = 1− + lnç ÷ (7)
tp çt ÷
è pø
where Q and Qp are discharge and peak discharge in volumetric units [L3 T-1]. The
parameters β or n depend upon catchment characteristics. The constants 2π and 7/6
(equation (5)) are derived from analytical approximation and remain the same for all
catchments.
RECESSION CHARACTERISTICS
Clark’s method assumes the storage, S, to be proportional to discharge Q, for routing
the time–area curve through a linear reservoir:
S = RQ (8)
Equation (8) gives:
∆t
Qm −
=e R (9)
Qm−1
where S is catchment storage [L3]; R is a recession or storage constant of the linear
reservoir [T]; Q is outflow or discharge [L3 T-1]; ∆t is the computational time interval
or time-step size [T]; and Qm is the discharge at the end of the mth time step [L3 T-1],
where m is an index for denoting time steps [-]. Let Qm/Qm-1 be defined as recession
constant. From equation (3), the recession characteristics of the two-parameter gamma
unit hydrograph can be written as:
( n −1) ∆t
Qm æ m ö −
=ç ÷ e K
(10)
Qm−1 è m − 1 ø
The hydrograph recession is observed due to the depletion of (a) channel storage,
(b) interflow storage and (c) groundwater storage. The UH recession consists of only
(a) and (b), as the contribution from groundwater is separated from the hydrograph to
obtain the UH. The response times for the recession of channel storage and interflow
storage are different; hence, the recession constants representing these are also
different. Equation (9) shows that the recession constant for the gamma UH changes as
the time or m increases. A gamma UH can reasonably represent the varying recession
constants due to the contributions from the channel and interflow storages.
From equation (9), it is observed that Clark’s method uses a single recession
constant that does not vary with time to represent the entire hydrograph recession.
Therefore, the depletions of, or the contributions from, interflow and channel storages
are not distinguished in Clark’s method. An unvarying recession constant may not
satisfactorily reproduce the hydrograph recession. Due to this, an unnecessary long tail
of the unit hydrograph is obtained. Viessman et al. (1989, pp. 222–223) also observed
that Clark’s method often produces prolonged runoff. The author’s scrutiny of a
number of flood events from different catchments showed that the hydrograph
recession may not follow equation (9) in many cases. When the hydrograph recession
(equation (9)) is plotted on a semi-logarithmic graph with Q on the log-axis, two or
more distinct linear segments are obtained. However, if a single straight line can be
fitted on such a graph, its slope gives the value of R (equation (9)). Straight lines with
different slopes give different values of R. With different straight lines, the estimation
of a representative R for use in Clark’s method is not justified and may cause errors.
used in the Clark’s method. Singh (2000) showed that n is a function of β that depends
on catchment characteristics. Also, the time–area curve depends on the catchment
characteristics. The runoff at the outlet at any time is proportional to the contributing
area. The area contributing to the runoff at any time, i.e. A(t), for a gamma distribution
may be written from equation (3) as:
A(t ) = Ct ( n −1) (13)
The constant of proportionality C in equation (13) depends on n and K. Therefore, the
time–area curve is implicitly incorporated into the gamma distribution, while in
Clark’s method it has to be externally supplied. The parameters of the time–area curve
as considered in HEC-1, i.e. 1.414 and 1.5, are assumed invariant irrespective of the
shape, size and location of the catchments. The shape and scale of the time–area curve
may vary from one catchment to another, because it depends upon the shape, size, and
hydraulic factors of the catchment, which govern the runoff at the outlet. Also, the
time–area curve defined by equations (11) and (12) is symmetrical with respect to t =
0.5Tc, which may not be appropriate for many catchments. These problems may cause
severe errors and inconsistency in the predicted runoff for ungauged basins. Use of
equations (5)–(7) is theoretically more justified than that of equations (11)–(12). If a
parametric form of time–area curve is assumed and its parameters are considered to
vary from one catchment to another, Clark’s method will have three to four parameters
to be optimized (including one or two parameters for the time–area curve). Since no
parametric representation of the time–area curve is currently available, a proper
optimization in this case is not feasible. However, in the case of a gamma UH, if qp or
β and tp are known, the UH can be determined even without optimization (equations
(5)–(7)). However, optimization is feasible only in cases of gauged catchments. The
gamma UH uses qp and tp (β = qptp) to completely define the shape of a UH. Both these
parameters have physical significance, and empirical relations that incorporate catch-
ment characteristics are available for their estimation (e.g. that of Snyder, 1938, which
incorporates slope and area). On the other hand, the parameters defining the time–area
curve (to be used in Clark’s method) are unavailable and empirical equations to relate
them to catchment characteristics are the subject of future studies (these can only be
available if a parametric form of time–area curve becomes available). Therefore, a
gamma UH offers more flexibility with fewer parameters than does Clark’s UH.
compared in Fig. 1 with that obtained (Akan, 1993) using Espey & Altman’s (1978)
method. A statistical measure for the difference between the two UHs, RMSE =
1.17 m3 s-1, is obtained. It is observed that, using the relation for only Qp and tp, the
gamma UH can be obtained without any trial and error. Therefore, other empirical
equations proposed by Espey et al. (1977) are found redundant for developing a
gamma UH.
For these values of n, tp and Qp, the ordinates of the gamma UH at computational
interval of 0.25 h were obtained using equations (6) and (7). The sample calculation
for the selected ordinates of the UH are given in Table 1. The gamma UH thus
obtained is compared in Fig. 2 to the Clark’s UH obtained using optimized parameters.
Although the Clark’s UH uses Tc as its parameter, the inflection point is observed just
close to the peak and the entire recession part of the Clark’s UH has unidirectional
curvature (see Fig. 2). The inflection point needs to be at Tc. This shows Tc loses its
significance in Clark’s method and serves as a fitting parameter. Mention is made here
that the gamma UH is obtained without optimization. The computed hydrographs for
the example using the two methods are shown in Fig. 3 (the baseflow and infiltration
losses were assumed the same in both UHs). The hydrograph obtained using the
gamma UH (gamma UH is convoluted with rainfall excess to get the computed
hydrograph) is more close to the observed hydrograph than that using Clark’s UH. The
integral squared error in the case of the gamma UH is 2.83 × 105 (m3 s-1)2
(3.046 × 106 (cfs)2), while with Clark’s UH it is 4.125 × 105 (m3 s-1)2 (4.441 × 106
(cfs)2) with 61 ordinates. In this example, for Clark’s UH, the optimized Tc is 3.16 h
while tg is 4.38 h, which is more than Tc. This shows inconsistency because tg should
be less than Tc (this condition can be derived, see, equation (16)) The lag obtained
using the parameters of the gamma UH is 4.38 h which is less than the calculated Tc =
5.0 h (taking tp/Tc = 0.58 using equation (17)). It may be noted here that the lag in the
case of the gamma UH comes out to be the same as for Clark’s UH. The different
values of Tc in the two methods may be due to inappropriate parameterization of
Clark’s method as used by HEC-1.
The example (Viessman et al., 1989, Example 25.1) taken next is for Oak Creek
catchment, in which the optimized UH for the sub-areas A and B were obtained using
Clark’s method through the HEC-1 program. For Sub-area A (86.5 km2 (33.4 mi2)),
Cp = 0.8 and tp = 2.93 h, the optimized Clark’s parameters are Tc = 3.74 h and R =
1.45 h with tg (lag) calculated as 2.92 h. The tg and tp in Clark’s UH come out to be
equal. This shows an inconsistency in the derived Clark’s UH because tg should be
more than tp for a UH with recession. Taking β ≈ Cp = 0.8 and tp = 2.93 h, Qp =
166.6 m3 s-1 (5885 cfs) is obtained. Thus, the values of n and K are obtained as 5.188
and 0.7 h, respectively. With these values, tg = 3.63 h, which is more than tp; hence, the
gamma UH has no such inconsistency. The Sub-area B (69.6 km2 (26.9 mi2)) has Cp =
0.8. The Clark’s UH parameters obtained through HEC-1 application are Tc = 3.4 h,
R = 1.065 h. The calculated value of tp = 2.51 h and lag = 2.49 h. Here, tg < tp, which
shows inconsistency in the derived Clark’s UH. For Cp = 0.8 and tp = 2.51 h, the
calculated parameters of gamma UH are n = 5.188 and K = 0.6 h, which give tg =
3.11 h, which is consistent. The Clark’s and gamma UHs are compared in Figs 4 and 5
for the sub-areas A and B, respectively.
Two other examples were taken from HEC-1 (1990). One is for Minnesota River
basin (Example 3 of HEC-1), where optimized Clark’s parameters are Tc = 46.00 h,
R = 183 h with Cp = 0.23 and tp = 49.11 h. Here, Tc < tp, which is an inconsistency of
the derived UH. The values of n and K for a gamma UH are obtained as 1.5 and
98.22 h, respectively. The gamma and Clark’s UHs for this example are shown in Fig.
6. The second is Example 6 of HEC-1 (1990). It has A = 9072 km2 (3503 mi2), Tc =
12.30 h, R = 8.60 h with Cp = 0.65 and tp = 10.29 h. Here, the difference between Tc
and tp is less than one sixth of Tc, which is very low and the UH may be termed
inconsistent. The gamma UH gives a larger difference. The gamma UH for this
example has n = 3.821 and K = 3.647 h.
where t1 and q1 correspond to the first point of the triplet and q1, q2, q3 are the
consecutive ordinates of the UH for the selected triplet within which the peak occurs.
The times t1, t2 and t3 correspond to the ordinates q1, q2 and q3, respectively.
For the examples from Singh (1976, Table 2), the values of tp and qp are obtained
using equations (15) and (16) as 13.38 h, 0.0582 h-1 and 18.29 h, 0.0454 h-1 for
Storm 1 and Storm 2, respectively. The values of β calculated from these values are
0.7787 and 0.8304 for Storm 1 and Storm 2, respectively. The UHs obtained by Singh
(1976) using LP and the corresponding gamma UHs are compared in Fig. 7. Another
example for a numerical UH was taken from Diskin & Boneh (1975); they obtained an
optimal UH utilizing the rainfall–runoff data of eight storms on a catchment of the
Cache River at Forman in southern Illinois, USA. The area of the catchment is
627 km2 (242 mi2). The catchment has a fairly flat (mild slopes) topography with a
well developed drainage network. The rainfall data were taken from daily
measurements at three stations. The storms were observed during February–May from
the years 1935–1951. Their optimal UH (Table III, last line) gave a triplet containing
the peak as (t1, q1) = (2 h, 0.2328 h-1); (t2, q2) = (3 h, 0.2817 h-1); (t3, q3) = (4 h,
0.1790 h-1). Using equations (15) and (16), tp = 2.83 h and qp = 0.2843 h-1 are obtained,
which give β = 0.8016 and n = 5.204. The gamma UH obtained (without optimization)
from equation (5) using these values is compared to the optimal UH obtained by
Diskin & Boneh (1975) using linear programming (Fig. 8). Both the UHs are in close
agreement. If the few points near the peak are corrupted by noise, a smooth curve
needs to be sketched through these points and then the triplet is selected.
especially the parameters of Clark’s UH, are not completely defined due to the lack of
a parametric form of the time–area curve, its direct use as a SUH is not feasible.
Hence, the empirical equations developed on the basis of the results obtained from
HEC-1 should be used with a caution. The parameters of a gamma UH can be obtained
using the existing empirical equations and guidelines for obtaining tp and qp or Cp. Few
empirical equations are also available for calculating the parameters n and K of a
gamma UH (e.g. Nash, 1960). Sometimes, the empirical equation for Tc is available.
For a gamma UH, Tc can be obtained using:
tp
Tc = t p + (16)
n −1
Equation (16) is re-arranged to obtain the value of n from known values of Tc and tp:
−2
æT ö
n = 1 + ç c − 1÷ (17)
çt ÷
è p ø
Once Tc and tp are known, n can be obtained using equation (17) and then the gamma
UH can be obtained using equations (6) and (7). Knowing the peak rate factor Df
(NRCS, 1972) and tp, the gamma UH can be obtained (see, Singh 2000). Using the tg
(lag of UH) and tp, a gamma UH can easily be calculated:
−1
æ tp ö
n = ç1 − ÷ (18)
ç t ÷
è g ø
tg
K= (19)
n
For the urban UH (e.g. Espey’s), the empirical equations for Qp and tp (equations
(1) and (2)) can be used to obtain the gamma UH. These equations in SI units take the
forms:
β = 1.4738t p−0.07 A −0.04 (20)
4.07 L0.23 φ1.57
tp = (21)
S a0.25 I a0.18
where L is in m, A is in m2, and t in min. Thus, a gamma UH can be obtained using the
available empirical equations for the parameters of other SUHs.
CONCLUSIONS
Clark’s method and a two-parameter gamma distribution for obtaining unit hydro-
graphs (UHs) for a gauged or an ungauged catchment have been critically compared in
terms of recession characteristics and time–area curve for modelling event-based
runoff. Several examples of Clark’s UH were taken (especially from HEC-1
applications) to show the inconsistencies in the derived Clark’s UHs; the
inconsistencies have been judged upon the relative values of tp, tg and Tc. The use of
the gamma UH does not show such inconsistencies. The urban unit hydrograph using
Espey’s (Espey et al., 1997) method has been transmuted to a two-parameter gamma
UH. The conclusions from the study on several examples are:
– As optimized in the HEC-1 program, Clark’s UH parameters, R and Tc, lose their
physical significance and serve merely as fitting parameters. The main problem in
using Clark’s method for developing a synthetic unit hydrograph is the unavail-
ability of a parametric form of the time–area curve. The equations that represent
the time–area curve in HEC-1 are found inadequate and unjustified, which results
in an inconsistent estimate of Tc (time of concentration).
– In Clark’s method, the entire hydrograph recession is represented by a single
recession constant, while a recession constant that varies with time is implicitly
incorporated into the gamma UH. The hydrograph recession is better represented
in a gamma UH than in Clark’s UH.
– HEC-1 uses Snyder’s Cp and tp to optimize the parameters (Tc and R) of Clark’s
UH. In addition to this, it requires that the time–area curve should be supplied
(empirical equations relating the time–area curve to catchment characteristics are
not available). This is a limitation of Clark’s UH to be used as a SUH (synthetic
UH). Once Cp and tp are known, the optimization of parameters of Clark’s UH is
not needed for developing a UH (i.e. gamma UH); the gamma UH is completely
defined without the need for optimization.
– Inconsistencies in the optimized Clark’s UH were observed in several examples of
HEC-1 applications. These are due to the unavailability of a general parametric form
of the time–area curve and lack of proper distinction between recession of interflow
and channel storages. The gamma UH takes into account interflow and channel stor-
ages by a variable recession constant. Better-fitted hydrographs and consistent UHs
are obtained for these examples using a gamma UH (even without optimization).
– Empirical relations for the parameters are not feasible in the case of Clark’s UH
due to lack of a parametric representation of the time–area curve. If such a
representation is assumed, the number of parameters of Clark’s UH is likely to be
increased to three or four; a gamma UH has only two parameters.
– Numerical UHs obtained for gauged catchments using linear programming or least
square approach have been shown easily transmutable to a gamma UH. Hence,
they can be imparted conceptual interpretation in terms of the parameters of a
gamma UH. This enhances the utility of numerical UHs for developing a SUH.
– The urban UH proposed by Espey et al. (1977) is shown to be easily transmutable
to a two-parameter gamma UH. It does not involve any trial and error and uses
empirical equations for only peak and time to peak of the UH, to define completely
the gamma UH. Other empirical equations proposed by them are found redundant
for developing the UH.
REFERENCES
Akan, A. O. (1993) Urban Stormwater Hydrology, 82–87. Technomic Publ. Co. Inc., Lancaster, USA.
Aron, G. & White, E. L. (1982) Fitting a gamma distribution over a synthetic unit hydrograph. Water Resour. Bull. 18(1),
95–98.
Clark, C. O. (1945) Storage and unit hydrograph. Trans. Am. Soc. Civil Engrs 110, 1419–1446.
Collins, M. A. (1983). Discussion of “Fitting a gamma distribution over a synthetic unit hydrograph” by G. Aron &
E. L. White. Water Resour. Bull. 19(2), 303–304.
Croley, T. E., II (1980) Gamma synthetic hydrographs. J. Hydrol. 47, 41–52.
Diskin, M. H. & Boneh, A. (1975) Determination of an optimal IUH for linear, time invariant systems from multi-storm
records. J. Hydrol. 22, 57–76.
Dooge, J. C. I. (1959) A general theory of the unit hydrograph. J. Geophys. Res. 64(2), 241–256.
Edson, C. G. (1951) Parameters for relating unit hydrograph to watershed characteristics. Trans. Am. Geophys. Union
32(4), 591–596.
Espey, W. H. & Altman, D. G. (1978). Nomograph for 10-minute unit hydrographs for small watersheds. Addendum 3 of
Urban Runoff Control Planning, Report EPA-600/9-78-035, Environmental Protection Agency, Washington DC, USA.
Espey, W. H., Jr, Altman, D. G. & Graves, C. B., Jr. (1977) Nomograph for 10 minutes unit hydrographs for urban
wartersheds. Tech. Memo. 32, Am. Soc. Civil Engrs, New York, USA.
Gray, D. M. (1961) Synthetic unit hydrograph for small drainage areas. J. Hydraul. Div. ASCE 87(4), 33–54.
HEC-1 (1990) Flood Hydrograph package, User’s Manual, Version 4.0. Hydrologic Engineering Center, US Army Corps
of Engineers, Davis, California, USA.
Hann, C. T., Barfield, B. J. & Hays, J. C. (1994) Design Hydrology and Sedimentology for Small Catchments. Academic
Press, New York, USA.
Kirpich (1940) Time of concentration of small agricultural watersheds. Civil Engng ASCE 10, 362.
Linsley, R. K., Kohler, M. A. & Paulhus, J. L. H. (1982) Hydrology for Engineers (third edn), 278–282. McGraw-Hill,
New York, USA.
Littlewood, I. G. (2003) Improved unit hydrograph identification for seven Welsh rivers: implications for estimating
continuous streamflow at ungauged sites. Hydrol. Sci. J. 48(5), 743–762.
Nash, J. E. (1958) Determining runoff from rainfall. Proc. Instn Civil Engrs London 10, 163–184.
Nash, J. E. (1959) Systematic determination of unit hydrograph parameters. J. Geophys. Res. 64(1), 111–115.
Nash, J. E. (1960) A unit hydrograph study with particular reference to British catchments. Proc. Instn Civil Engrs London
17, 249–282.
NRCS (National Resources Conservation Service) (1972) Hydrology, Sect. 4, National Engineering Handbook. NRCS
[formerly Soil Conservation Service (SCS)], USDA, Washington DC, USA.
Rosso, R. (1984) Nash model relation to Horton order ratios. Water Resour. Res. 20(7), 914–920.
Singh, K. P. (1976) Unit hydrograph—a comparative study. Water Resour. Bull. 12(2), 381–392.
Singh, S. K. (1998) Reconstructing a synthetic unit hydrograph into a gamma distribution. In: Proc. Int. Conf. on Integrated
Water Resources Management (Univ. of Alexandria, Egypt, 11–14 October 1998), 104–110. Alexandria Univ., Egypt.
Singh, S. K. (2000) Transmuting synthetic unit hydrograph into a gamma distribution. J. Hydrol. Engng ASCE 5(4), 380–385.
Singh, S. K. (2004) Simplified use of gamma-distribution/Nash model for runoff modelling. J. Hydrol. Engng ASCE 9(3),
240–243.
Snyder, F. F. (1938) Synthetic unit-graphs. Trans. Am. Geophys. Union 19, 447–454.
Valdes, J. B., Fiallo, Y. & Rodriguez-Iturbe, I. (1979) A rainfall–runoff analysis of the geomorphologic IUH. Water
Resour. Res. 15(6), 1421–1434.
Viessman, W., Lewis, G. L., Jr & Knapp, J. W. (1989) Introduction to Hydrology, 222–223, 596–608. Harper & Row,
New York, USA.
Wu, I. P. (1963) Design hydrographs for small watersheds in Indiana. J. Hydraul. Div. ASCE 89(6), 35–66.