Вы находитесь на странице: 1из 19

Water Resour Manage

DOI 10.1007/s11269-008-9388-8

Estimation of Clark’s Instantaneous Unit Hydrograph


Parameters and Development of Direct Surface
Runoff Hydrograph

Muhammad Masood Ahmad ·


Abdul Razzaq Ghumman · Sajjad Ahmad

Received: 1 March 2008 / Accepted: 12 December 2008


© Springer Science + Business Media B.V. 2009

Abstract We present a method to estimate Time of Concentration (Tc ) and Storage


Coefficient (R) to develop Clark’s Instantaneous Unit Hydrograph (CIUH). Tc is
estimated from Time Area Diagram of the catchment and R is determined using
optimization approach based on Downhill Simplex technique (code written in FOR-
TRAN). Four different objective functions are used in optimization to determine
R. The sum of least squares objective function is used in a novel way by relating it
to slope of a linear regression best fit line drawn between observed and simulated
peak discharge values to find R. Physical parameters (delineation, land slope, stream
lengths and associated drainage areas) of the catchment are derived from SPOT
satellite imageries of the basin using ERDAS: Arc GIS is used for geographic data
processing. Ten randomly selected rainfall–runoff events are used for calibration
and five for validation. Using CIUH, a Direct surface runoff hydrograph (DSRH) is
developed. Kaha catchment (5,598 km2 ), part of Indus river system, located in semi-
arid region of Pakistan and dominated by hill torrent flows is used to demonstrate the
applicability of proposed approach. Model results during validation are very good
with model efficiency of more than 95% and root mean square error of less than
6%. Impact of variation in model parameters Tc and R on DSRH is investigated.
It is identified that DSRH is more sensitive to R compared to Tc . Relatively equal
values of R and Tc reveal that shape of DSRH for a large catchment depends on both
runoff diffusion and translation flow effects. The runoff diffusion effect is found to
be dominant.

M. M. Ahmad · A. R. Ghumman
Faculty of Civil Engineering, University of Engineering and Technology, Taxila, Pakistan
e-mail: mmasood567@yahoo.com
A. R. Ghumman
e-mail: abdulrazzaq@uettaxila.edu.pk

S. Ahmad (B)
Department of Civil and Environmental Engineering, University of Nevada,
Las Vegas, NV, USA
e-mail: sajjad.ahmad@unlv.edu
M.M. Ahmad et al.

Keywords Time of concentration · Storage coefficient ·


Clark’s instantaneous unit hydrograph · Direct surface runoff hydrograph ·
Hydrograph parameter estimation

Nomenclature

Qi+ 1 (i + 1)th ordinate of the CIUH


i index varying from 1 to N, N is number of ordinates of the time area
diagram
RE uniformly distributed rainfall excess
C0 and C1 are weighting coefficients
R storage coefficient
t computational time interval
Tc time of concentration
Lj length of stream of sub basin in km
Sj representative land slope of sub basin j
j index varying from 1 to M and M is number of identified sub basins
Qpo observed peak discharge
Qps computed/simulated peak discharge
Qoj jth ordinate of the observed hydrograph
Qcj jth ordinate of the computed hydrograph
j index varying from 1 to n, n being no. of ordinates of observed
hydrograph
Qpo(i) peak discharge of observed hydrograph for ith event
Qps(i) peak discharge of simulated hydrograph for ith event
Tpo(i) time to peak of observed hydrograph for ith event
Tps(i) time to peak of simulated hydrograph for ith event
NE number of calibration events
η model efficiency
NQ number of ordinates of the simulated hydrograph
i index varying from 1 to NQ
Qoi ith ordinate of the observed hydrograph
Qci ith ordinate of the computed/simulated hydrograph
Q̄o mean of the ordinates of the observed hydrograph
Z peak weighted root mean square (RMS) error objective function.

1 Introduction

The idea of Instantaneous Unit Hydrograph (IUH) was first introduced by Clark
(Clark 1945). In IUH unit rainfall excess is applied uniformly to the entire catchment
momentarily. Being independent of time, a unique hydrograph is obtained for
the catchment (Noorbaksh 2005). To derive IUH, the Clark model requires two
parameters: time of concentration Tc and coefficient of storage R. The definition of
time of concentration in Clark model is somewhat different than that used elsewhere
in hydrology. Clark’s Tc is time of travel required by last drop of rainfall excess at the
hydraulically most remote point in the catchment to reach the channel network, in
Estimation of Clark’s instantaneous unit hydrograph parameters

contrast to the outlet that is used elsewhere (Straub et al. 2000). For linear system and
purely translation flow both definitions are equivalent. The Clark model interprets
hydrologic response into two separate functions, the translation and attenuation.
The translation flow is taken care of by time area diagram where as attenuation is
produced due to channel storage (Schulz 1976).
The Clark Unit Hydrograph (CUH) is essentially a synthetic unit hydrograph
(Ponce 1989). It is very effective tool for simulating rainfall–runoff, especially in
the catchments having unusual shapes with large length to width ratios and complex
geomorphology (Sabol 1988). However, CUH could not gain much popularity due
to difficulty in estimating Clark model parameters. Various researchers have tried to
correlate Clark’s model parameters with physical characteristics of catchment and
channel system. Jawed (1973) described that R is equal to discharge at point of
inflection on observed hydrograph divided by the slope at that point. Although this
procedure is simple, it requires an observed hydrograph with well defined point of
inflection which is usually not the case. The identification of inflection point on the
recession limb of the observed hydrographs is an approximate procedure. Further
more, each hydrograph varies in shape, therefore different R values will be obtained
from different hydrographs and often the simulated hydrographs, based on average
value of storage coefficient R, do not fit well with the observed hydrographs. Sabol
(1988) presented a procedure for finding R from observed hydrograph recession
limb analysis. It, however, requires averaging of different values of R for different
rainfall–runoff events. Wilson (1990) proposed a similar method of finding R but
it again required representative flood hydrograph, whereas shape of hydrograph
almost always varies in every flood event and consequently a unique value of R
representing unique instantaneous unit hydrograph is difficult to achieve. Melching
and Marquard (1996) presented empirical equations for estimating Tc and R. These
equations require estimation of representative channel and overland slopes and are
region specific. Straub et al. (2000) presented a procedure for developing empirical
relationship for estimating R based on regional analysis for small catchments in
Illinois USA and identified limitations of using these equations to mid-sized or
large catchments with different physiographic characteristics. Kumar et al. (2004)
demonstrated that peak of geomorphologic instantaneous unit hydrograph (GIUH)
is very sensitive to dynamic velocity and length of highest order stream, which in turn
affects the estimated Clark model parameters. The method requires velocity data
that may not be available for the catchment of interest. Sahoo et al. (2006) estimated
Tc and R by equating GIUH peak to that of Clark model. This procedure results in
large diversity in Tc and R values. Also, peak of GIUH depends on representative
dynamic velocity, the estimation of which requires a parallel relationship between
observed peak discharge or intensity of rainfall excess and flow velocity.
We present a method for estimating CIUH parameters Tc and R from rainfall
and peak discharge observations. The proposed method overcomes the catchment
size limitations. A single parameter i.e., lumped storage coefficient R takes care
of the dynamics of the rainfall–runoff process and the need for velocity estimation
is eliminated. The velocity is accounted for by the time of concentration in terms
of slope variation at sub-catchment level. Time of concentration is derived from
geomorphology of the catchment, whereas storage coefficient is determined using
optimization approach. The criterion adopted in proposed method is to find ‘R’ and
‘Tc ’ which results in the best fit between peak discharges of observed and simulated
M.M. Ahmad et al.

DSRH. Design of hydraulic structures and flood protection structures need mainly
the peak flood values but the complete hydrograph is required in many cases such as
for water resources planning and reservoir operations. The method we present can
also handle cases where complete hydrograph is required.
We present four different objective functions for estimation of CIUH parameters.
The objective function based on the least sum of the square of difference between
observed and simulated hydrograph ordinates has long been applied for hydrologic
modeling, however the current study applies the least sum of squares objective
function in a novel way by relating it to slope of a linear regression best fit line
drawn between observed and simulated peak discharge values. A classical approach
is applied in a novel and simplified way to find pair of Clark’s model parameters.
Remaining paper is organized as follows. First, theoretical background on Clark’s
instantaneous unit hydrograph model, time area diagram, optimization of para-
meters, and evaluation of model performance is provided. This is followed by
description of study area. Then method is explained, and results are presented.
Finally, results are discussed and conclusions are drawn.

2 Theoretical Background

2.1 Clark’s Instantaneous Unit Hydrograph Model

The mathematical form of Clark’s Instantaneous Unit Hydrograph (CIUH) after


Clark (1945) is represented as:

Qi+1 = 2Co RE (i) + C1 Qi (1)

Where i index varying from 1 to N, and N is number of ordinates of the time area
diagram
RE uniformly distributed rainfall excess (ordinates of time–area diagram in units
of km2 multiplied by a constant for unit conversion)

Qi+1 = (i+1)th ordinate of the Clark’s Instantaneous Unit Hydrograph


Co and C1 are weighting coefficients proposed by Muskingham defined as Co =
0.5t/(R + 0.5t) and C1 = (R − 0.5t)/(R + 0.5t)
t computational time interval

A unit hydrograph for finite time interval T can be found by lagging IUH equal to
time T and averaging the IUH ordinates for time period T.

2.2 Time Area Diagram

To apply CIUH concept it is necessary to develop time area diagram for the
catchment. The time area diagram shows the relationship between travel time of
runoff to the catchment outlet and area of the catchment contributing towards
surface runoff generation. Therefore the only independent variable in time area
diagram is the travel time, which is time of concentration of the runoff to the
Estimation of Clark’s instantaneous unit hydrograph parameters

catchment outlet. It depends on the velocity of flow of runoff over the land surface.
The time area diagram only does not fully define the phenomena of rainfall–runoff
as runoff generated is diffused by the undulating land topology and similar other
factors. The storage coefficients R of the catchment takes care of such effects.
In order to find time of concentration and thus developing time area diagram,
travel time from the most remote point on the hydrologic boundary of each sub-basin
is required which can be found using classical Kirpich formula given as (Kirpich 1940)


(Tc ) j = 0.06628 L0.77
j S0.385
j (2)

Where (Tc ) j is time of concentration in hours for jth sub-basin, j is index varying
from 1 to M and M is number of identified sub-basins. L j is length of stream of jth
sub basin in km and S j is representative land slope of sub-basin j.

2.3 Optimization of Parameters

Clark model is used to simulate runoff when its various parameters are known. The
parameter optimization is the inverse of this. It consists of determining the best
values of parameters that produce results close to the observed values of discharge
hydrograph. The goal of optimization is to minimize a scalar quantity known as
objective function or error. The objective function/error may be defined in several
ways. The following four objective functions are adopted in this study.

1. Based on minimizing difference between observed and simulated peak dis-


charges for all events. A linear regression line is drawn between scatter plot of
observed and computed peak discharge of all the calibration events and slope of
the line is determined by the following relationship (Blueman 2003)

     
NE Qpo Qps − Qpo Qps
Objective Function (F1 ) =  2  2 (3)
NE Qpo − Qpo

Where NE is number of calibration events, Qpo is observed peak discharge and Qps
is computed peak discharge. The desired value of objective function being the slope
of linear regression line is unity which shows all discharges equi-positioned around
the line with least distance between straight line and observed and computed values
of peak discharge.

2. Based on minimizing difference between observed and simulated peak dis-


charges for single event:

 2
Qpo − Qps
Objective Function (F2 ) =  2 (4)
Qpo
M.M. Ahmad et al.

3. Based on the method of least squares i.e. minimizing the sum of squares
of deviations between observed and computed values of runoff hydrograph.
Mathematically, this objective function is expressed as:
j=n  2
 Qoj − Qcj
Objective Function (F3 ) =  2 (5)
j=1 Qoj

Where Qoj is observed value of jth ordinate of direct runoff hydrograph, Qcj is
simulated value of jth ordinate of direct runoff hydrograph. j is an index varying
from 1 to n, n being number of hydrograph ordinates of observed hydrograph.
4. This objective function defined by Lee et al. (1972) and adopted by Al-Wagdany
and Rao (1997) takes consideration of both peak discharge Q p and time to peak
T p and is defined as follows:
 2 
12
E
i=N
Qpo(i) − Qps(i) Tpo(i) −Tps(i) 2
Objective function (F4 ) = + (6)
i=1
Qps(i) Tps(i)

Where NE is number of calibration events, Qpo(i) is peak discharge of observed ith


event hydrograph, Tpo(i) is time to peak of observed ith event hydrograph, Qps(i) is
peak discharge of simulated ith event hydrograph and Tps(i) is the time to peak of the
simulated ith event hydrograph, i vary from 1 to NE.
These objective functions can be minimized using a suitable optimization routine.

2.4 Model Performance

The model efficiency and peak weighted root mean square error are selected to test
the performance of the model as proposed by Nash and Sutcliffe (1970) and USACE
Hydrologic Modelling System HEC-HMS (1998).
⎡ ⎤
Q
i=N
⎢ (Qoi − Qci ) ⎥
2
⎢ i=1 ⎥
η = ⎢1 − ⎥ × 100 (7)
⎣ Q 
i=N 2 ⎦
Qoi − Q̄o
i=1
  NQ  
 12
1  Qoi + Q̄o
Z= (Qoi − Qci )2 (8)
NQ 1 2 Q̄o

η is Model Efficiency, NQ is number of ordinates of the hydrograph, i is index varying


from 1 to NQ, Qoi is ith ordinate of the observed hydrograph, Qci is ith ordinate of
the computed/simulated hydrograph, Q̄o is mean of the ordinates of the observed
hydrograph and Z is Peak weighted Root Mean Square Error (RMSE).

3 Study Area

Pakistan, like other developing countries, depends heavily on its water resources
both for agriculture and power. Arid/semi arid regions in the country come across
with storms only for a very short time during monsoon and winter periods. We
Estimation of Clark’s instantaneous unit hydrograph parameters

investigate the applicability of conceptual models like Clark’s Instantaneous Unit


Hydrograph for a semi arid catchment in Pakistan. Currently, NRCS curve number
and SCS synthetic unit hydrograph methods are widely used in Pakistan. The study
area “Kaha Catchment” is selected in semi-arid subtropical continental monsoon
region of Pakistan. It is characterized by two distinct seasons in summer and winter.
Mean annual precipitation in the area is around 350 mm. Most of the rainfall takes
place during the months of June to September under the influence of monsoon rains.
Mean annual summer day temperature in the area is 34◦ C. The hottest month is
June with the mean maximum day time temperature of 41◦ C, whereas the mean
minimum daytime temperature is 4◦ C observed in January. Geographically, the area
is located in Suleman Range in south of the country between latitude and longitude
of 29◦ –31◦ N and 69◦ –71◦ E, respectively (Fig. 1a). The mean elevation is around
1,100 m. Nearly 55% of the area lies between elevation of 900 m and 1,300 m. The
minimum elevation is less than 300 m and maximum elevation exceeds 2,000 m. The
65% of total catchment area (mountain ranges) comprises lime stone, shale and
sand stone, 30% gravelly loamy soils and 5% limy silty soil cover. The mountain
ranges are generally running in the north east-south west direction. There are a
few perennial irrigation schemes originating from aquifers having small quantities
of flow from 0.03 to 0.04 m3 /s. The stream flow velocity is high and can carry gravels
and boulder sized stones. The high flow velocities are due to high intensity flashy
rains resulting in torrential flows (fast flowing water). These fast flowing hill streams
are called hill torrents. Very few aquifers in the area are charged by infiltration
and subsurface component of hill torrent flows. To divert flood waters to cultivate
crops local farmers build various temporary embankments. The geomorphology of
the area is characterized by rugged terrain, large variations in land slope and sharp
bends in natural streams. Around sixty five percent of catchment areas consist of
barren mountains without vegetative cover. Almost all the streams are ephemeral
in nature flowing during monsoon or winter rains only. Kaha Catchment drains into
Indus River.

4 Method

To develop direct surface runoff hydrograph (DSRH) from Clark’s instantaneous


unit hydrograph (CIUH), following tasks are carried out.

4.1 Digitization of Catchment Maps and Data Processing

Detailed topographic and soil maps are not available for Kaha catchment as the area
is scarcely populated and difficult to access. However, presence of satellite imageries
of the area has made possible to map the soil cover, aspect, and topographical details.
The satellite imageries were obtained by Hanson et al. (1995) from the French-
Swedish SPOT satellite. The slope map developed by Hanson et al. (1995) at 1:
125,000 scale from the SPOT stereo 15 imageries using ORTHOMAX software
are used to estimate the average land slope. Data processing of the catchment
is done using ArcGIS. Using the Digital Elevation Model of the catchment each
stream length and area draining in respective stream is measured. A total of 326
sub-catchments are identified by delineation of the catchment as shown in Fig. 1b.
M.M. Ahmad et al.

(a)

(b)
Estimation of Clark’s instantaneous unit hydrograph parameters

 Fig. 1 a Geographic location of Kaha catchment, Pakistan. b Drainage pattern in Kaha catchment,
Pakistan

The catchment area is determined as 5598 km2 . The total length of streams is
1711 km. The drainage density (stream lengths per unit area of catchment) of the
catchment is 300 m/km2 . The length of longest stream is 171 km.

4.2 Development of Time Area Diagram

Two model equations for computation of time of concentration were applied prior
to making decision that which model will be best applicable to study catchment
and subsequent development of time area diagram. First, the flow regimes were
broken into overland, shallow concentrated, and channel flows. The kinematic wave
model was applied to overland flow. This resulted in flow velocities at the catchment
outlet not high enough to justify carrying of large sized boulder stones present at
the catchment outlet. On the other hand, Kirpich equation yielded high velocities at
catchment outlet ranging from 1.9 m/s to 2.75 m/s. This was found applicable due to
the fact that big boulders transported by the streams are present in large quantities
at the downstream of the catchment outlet. The kirpich equation has been used by
various researchers for development of time area diagram (Sahoo et al. 2006; Sarangi
et al. 2006; Kumar et al. 2002; Jain et al. 2000). We also found that Kirpich equation
was able to adequately describe the flow process in the Kaha catchment.
Time of concentration (Tc ) for each sub-catchment was computed using Eq. 2. The
Tc ’s for each sub-catchment were summed up to get total time of travel from each
sub-catchmnet to the catchment outlet. The sub-catchments were grouped for each
hour of travel times and respective sub-basin areas were summed to get isochronal
areas. Isochrones are lines delineating catchment area into sub areas so that all rain
falling instantaneously on particular sub-area has the same time of travel to the
catchment outlet. This procedure eliminated the necessity of plotting isochrones on
the plan of catchment. The developed time area diagram is shown in Fig. 2. The base
length of the time area diagram gives the time of concentration of the catchment.
A polynomial function was determined by regression that best described the time
area relationship. This function has dependent variable as isochrone area and
independent variable as time of concentration. The function was plotted for various
values of time of concentration as shown in Fig. 3. This shows how the distribution
of runoff contributing areas changes with change in time of concentration of the
catchment. It indirectly defines the translation hydrograph with choice of time of
concentration of the model. The base length of time area diagram is a single value of
time of concentration, which is subsequently unchanged in rest of the analysis.
The ordinates of the time area diagram were normalized. The area ordinates
were divided by area of the catchment and time ordinates by the maximum time
of concentration estimated. The cumulative normalized ordinates were plotted as
shown in Fig. 4. A synthetic function is defined in USACE HEC-HMS (1998) for de-
   1.5 
velopment of time area diagram as A Ac = 1.414 t Tc for t ≤ Tc /2 and A Ac =
  1.5
1 − 1.414 1 − t Tc for t ≥ Tc /2. For comparison synthetic function derived by
USACE HEC-HMS (1998) is also shown in Fig. 4. The time area relationship of the
catchment under study lies below the synthetic HEC-HMS relation showing slower
M.M. Ahmad et al.

Fig. 2 Time area diagram for


Kaha catchment
500

400

300

200

100

-
- 5 10 15 20 25
Time of Concentration (Tc ) in hours

hydrologic response by the catchment under study as compared to the catchments


whose database was used to develop average synthetic time area function by HEC-
HMS.
Development of time area diagram is required to define static model inputs.
The time area diagram ordinates are lagged appropriately and summed to get unit
hydrographs of various durations.

4.3 Observed Rainfall–Runoff Data and Abstractions

Irrigation and Power Department, a provincial water management agency, is re-


sponsible for the management of the hill torrents in the study region. The Rainfall–
Runoff data was obtained from this agency. There are 12 rain gauge stations in the
Suleman Range. The record length for rainfall and runoff varied from 8 to 26 years.
The study area i.e., Kaha catchment has three rain gauge stations and has 23 years

Fig. 3 Variation in shape of 600


time area diagram with
Isochrone Area, A (km )
2

variation in Tc 500

400

300

200

100

-
0 5 10 15 20 25
Tc (Hours)

15 16 18 19 20 23
21 17 22 24 25
Estimation of Clark’s instantaneous unit hydrograph parameters

Fig. 4 Normalized time area 1.0


diagram for Kaha catchment 0.9
(Ac = area of catchment) 0.8
0.7
0.6

A/Ac
0.5
0.4
0.3
0.2
0.1
-
- 0.2 0.4 0.6 0.8 1.0
t/Tc
Poly. (Time Area Relation for Kaha Watershed)
Poly. (HEC-HMS Synthetic time-area relation)

of record. This data was analyzed to obtain maximum daily rainfall corresponding
to annual peak discharge. Fifteen largest flood events were selected for this study.
The total rainfall depth, obtained by isohyetal method, was converted to excess
rainfall by subtracting the losses. Three approaches of loss separation i.e. Curve
Number Approach, Percentage Runoff Approach and Constant Loss Approach were
considered (for details on these approaches see Linsley et al. 1982). It was noted that
1% error in selection of curve number could vary flood peak discharge by 10%. The
constant loss approach is less realistic as abstractions do not remain constant during
storm duration. The percentage runoff approach has certain advantage over other
methods and proved to be relatively reliable and simple to use. In this approach, the
ratio of the net rainfall to the gross rainfall is expressed as a constant and is used for
estimation of excess rainfall. Determination of excess rainfall is required to define
dynamic model inputs. The dynamic input is time-distributed excess rainfall whereas
output is direct runoff hydrograph.

4.4 Optimization to Identify Clark Model Parameters

Considering the four objective functions used in this study and decision tree given
by NAG (NAG 1985) a combination of second and first order gradient techniques
(Gauss Newton and modified Newton method Gill et al. 1981) were found suitable to
be used for the present study. Subroutine E04FCF from the NAG Library, based on
Downhill Simplex method (Press et al. 1986), for automatic optimization of the para-
meters of the Clark model was selected. This method is suitable for multidimensional
minimization and requires only function evaluation not derivatives. In this method
search is terminated when the error between the observed and simulated values of
runoff is minimized. Different steps involved in estimating R are shown in a flow
chart in Fig. 5.

4.5 Model Calibration and Validation

Model was calibrated using 10 randomly selected flood events (E1–E10) and vali-
dated on remaining five independent flood events (E11–E15). A sensitivity analysis
was performed to evaluate the impact of selection of objective functions on value
M.M. Ahmad et al.

Fig. 5 Flow chart for computer program to estimate R

of storage coefficient R. Model performance was evaluated using different error


measures including model efficiency, and RMSE.

5 Results and Discussions

5.1 Calibration of Clark Model and Sensitivity Analysis of Parameters

The objective function F1 plotted against Tc and R is shown in Fig. 6. The X-axis
intersects the Y-axis (objective function) at value of unity, the desired value of
objective function. The objective function lines intersect X-axis at Tc = 25 h and
R = 24 h, respectively. The slope of F1 line shows that it converges rapidly for
same range of R values as compared to Tc values, indicating that F1 and therefore
DSRH is more sensitive to R than Tc . The objective functions F1, F3, and F4 for
Tc of 25 h are plotted against R in Fig. 7. A unique value of R corresponding to
minimum value of each objective function can be determined from Fig. 7. The values
of R corresponding to four different objective functions are shown in Table 1. The
suitability of a particular objective function depends upon type of application and
data availability. If estimates are required for design of hydraulic structures and flood
Estimation of Clark’s instantaneous unit hydrograph parameters

Fig. 6 Unique pair of (Tc , R) 1.10


using objective function F1

1.05

Objective Function
1.00
20 21 22 23 24 25 26 27 28 29 30

0.95

0.90
Tc & R (Hours)
F1 vs Tc F1 vs R

protection schemes then objective function based on peak flood values (F1, F2, and
F4) will be suitable. In case of water resources planning and reservoir operations the
information on entire hydrograph is required so the objective function based on the
entire hydrograph (F3) should be selected.
The objective function F1 gives R value of 24 h (column 2 in Table 1), whereas the
event based objective function F2 gives different values of R as shown in column
3 of Table 1. These values of R vary by −11% to +10% from those computed
using objective function F1. The objective function F3 gives average R value of
19.81 h (column 4 of Table 1). This value differs by 17% from R value obtained
from objective function F1. The objective function F4 yielded R of 23.24 h being 3%
less than the value of R obtained using objective function F1.

Fig. 7 Optimum R yielded by 1.0


various objective functions
(Tc = 25 h)
0.8

0.6

0.4

0.2

-
15 20 25 30
R (Hours)

F4 F3 F1
M.M. Ahmad et al.

Table 1 Value of R (storage coefficient) obtained from four different objective functions
Event Storage coefficient
number Based on minimizing Based on minimizing Based on
difference between sum of square errors minimizing ERR
observed and simulated between observed (Lee et al. 1972)
peak discharges and simulated
DSRH ordinates
All events FI Single event based F2 Single event based F3 All events F4
[1] [2] [3] [4] [5]
1 24.00 21.68 19.53 23.24
2 22.14 18.57
3 21.74 18.63
4 25.04 20.70
5 23.23 19.98
6 26.55 20.29
7 23.87 21.64
8 20.83 18.78
9 24.30 21.03
10 21.28 18.91
Mean value 23.07 19.81

Corresponding to objective function F1, the computed value of R is 24 h. It


is observed that Clark IUH based DSRH model gives fairly accurate results as
demonstrated by the model performance measures computed and shown in Table 2.
The model efficiency varies from 92% to 98%, which is acceptable as unit hydrograph
technique inherent itself with assumptions and approximations (Linsley et al. 1982).
Most hydrologists while reporting flood hydrograph peak increase it by 5–20%
before using unit hydrograph for the estimation of extreme floods (Linsley et al.
1982). The maximum percent error in runoff volume is around 12%, however, its
value is generally less than 7% for other events. The error in runoff volume is
believed to be due to the reason that recession limb of the DSRH is underestimated
by Clark’s model, whereas the recession limb after the inflection point do carry
indirect surface runoff resulting from recharge from stream banks. The hydrograph
peak was observed to be over estimated by 9% in one case and under estimated
by 6% in the other, which is expected to be due to rounding-off of parameter ‘R’
to nearest hour to get a practical computational time interval. The RMSE (Z) has
values less than 8%. The value of R is about 0.95 times that of Tc showing that flow
diffusion and translation flow effects play equal role in mechanism of development
of DSRH by CIUH technique. This corresponds to a value of 0.5 for [R/(Tc + R)],
which is one of characteristics of large catchments.
Corresponding to objective function F2 the average value of R computed for 10
calibration events is 23.07 h, as shown in column [3] of Table 1 When this average of
10 events was used for all the calibration events, the model efficieny varied between
94% and 98% and maximum error in volume, peak flow, time of peak and RMSE was
12%, −15%, 0% and 8%, respectively (Table 2). The peak of the DSRH deviated
from observed peaks by 15% vs 9% in deviation observed in case when objective
function F1 is used.
Table 2 CIUH model performance using different error measures during calibration and validation
Event Based on minimizing Based on minimizing Based on minimizing Based on minimizing ERR
number difference between difference between sum of square errors (Lee et al. 1972) F4
observed and simulated observed and simulated between observed
peak discharges peak discharges and simulated DSRH
(all events) F1 (single event) F2 ordinates F3
R= 24 h R= 23.07 R= 19.80 R= 23.24
H % V % Qp % Tp % Z η % V % Qp % Tp % Z η % V % Qp % Tp % Z η % V % Qp % Tp % Z
1 Calibration 92 −2 6 0 8 94 −2 4 0 8 96 −2 −5 0 7 94 −2 4 0 8
2 93 −3 5 −3 7 94 3 2 0 6 96 3 −6 0 6 94 3 3 0 6
3 96 4 6 0 6 94 4 3 0 5 98 4 −6 0 5 97 4 4 0 6
4 98 12 −2 0 5 97 12 −15 0 5 98 12 −15 0 5 98 12 −5 0 5
5 96 2 3 0 6 97 2 0 0 5 97 2 −10 0 5 97 2 0 0 5
6 97 11 −6 0 5 97 11 −9 0 5 97 11 −19 0 7 97 11 −9 0 5
7 97 4 1 0 5 97 4 −2 0 5 98 4 −11 0 5 97 4 −2 0 5
8 96 −2 9 0 6 97 −2 6 0 6 98 −2 −3 0 5 97 −2 6 0 6
Estimation of Clark’s instantaneous unit hydrograph parameters

9 97 7 0 0 5 98 7 −3 0 5 98 7 −13 0 5 98 7 −3 0 5
10 97 −3 7 0 6 97 −3 5 0 6 98 −3 −4 0 5 97 −3 5 0 6
11 Validation 96 5 2 0 5 97 5 −1 0 5 98 5 −10 0 5 97 5 0 0 5
12 97 5 2 0 5 97 5 −1 0 5 98 5 −11 0 5 97 5 −1 0 5
13 96 8 −1 0 5 97 8 −4 0 5 97 8 −14 0 6 97 8 −4 0 5
14 97 13 −6 0 5 97 13 −9 0 5 97 13 −19 0 6 97 13 −9 0 5
15 97 8 1 0 5 98 8 −2 0 5 98 8 −12 0 5 98 8 −2 0 5
M.M. Ahmad et al.

The objective function F3 yielded average R value of 19.8 h compared to 24 h


determined by the objective function F1. This relatively low value resulted in slight
increase in model efficiency and more deviation in peak discharge of the order of
19%. Both of these objective functions are important. F1 is based on peak flow
that is relevent for design and operation of hydraulic structures, whereas F3 is
based on complete hydrograph that is important for water resources planning and
management.
Corresponding to objective function F4 the unique value of R for ten calibration
events was found to be 23.24 h, which is close to the value of 24 h determined
for objective function F1. However, this slightly lower value of R resulted in 13%
maximum error in peak discharge as compared to the case of the objective function
F1, which yielded maximum error of 9%. The runoff volumes and RMSE also varied
similarly. The performance comparison of different objective functions reveals the
superior performance of objective function F1 with Tc and R parameters of (25,24).
The calibration events E1 to E10 are plotted in Fig. 8a to d using best pair of
(Tc , R) as (25,24). These events were grouped keeping in view the range of peak
discharges. The mildest and extreme events show relatively more deviations (Fig. 8a,
d), whereas the moderate events show best fit (Fig. 8b, c). The peak discharge
matches the observed discharge well, whereas the rising and falling limbs of the
hydrograph are under estimated. This error is due to the simplified storage effects

1000
2000
800
Flow (m3/s)

Flow (m3/s)

1500
600
400 1000
200 500
0 0
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
Time (hour) Time (hour)
Event 1 (Observed) Event 1 (Computed) Event 3 (Observed) Event 3 (Computed)
Event 4 (Observed) Event 4 (Computed) Event 8 (Observed) Event 8 (Computed)
Event 9 (Observed) Event 9 (Computed)
(a) (b)
3500
3000 3000
2500 2500
Flow (m3/s)
Flow (m3/s)

2000 2000
1500 1500
1000 1000
500 500
0 0
0 10 20 30 40 50 60 70 0 10 20 30 40 50 60 70
Time (hour) Time (hour)
Event 5 (Observed) Event 5 (Computed) Event 2 (Observed) Event 2 (Computed)
Event 7 (Observed) Event 7 (Computed) Event 6 (Observed) Event 6 (Computed)
Event 10 (Observed) Event 10 (Computed)
(c) (d)
Fig. 8 a to d Observed vs computed DSRH for calibration (events E1 to E10). a Computed vs
observed DSRH for event 1 and 4. b Computed vs observed DSRH for events 3, 8 and 9. c Computed
vs observed DSRH for events 5, 7 and 10. d Computed vs observed DSRH for events 2 and 6
Estimation of Clark’s instantaneous unit hydrograph parameters

of the catchment inherited in Clark’s model. The recession limb has more time base
as actually observed in hill torrents. The recession limb might have better fit for
catchments having perennial flows at the outlet. Since Clark’s model is event based
and base flow contribution is not accounted for in R, the recession limb is under
estimated.

5.2 Validation of Model

The optimum pair of Clark IUH Model parameters (Tc , R) determined as (25, 24)
in units of hours were used to compute, DSRH for validation (events 11 to 15).
Instantaneous unit hydrograph was derived by Clark’s method and convolution of
the same was done with excess rain hyetograph to get the direct surface runoff
hydrograph (DSRH). The resultant DSRH is plotted (events E11 to E15) as shown
in Fig. 9a–c. The events with lowest peak were plotted separately (Fig. 9a, c) whereas
moderate events were grouped together as shown by Fig. 9b. The performance of
model during validation was comparable to that during calibration. Figure 9a–c show
that the peak discharge, time to peak and time base of the observed and computed
hydrographs match well, however the falling limb of hydrograph is under estimated.
There is 13% error in runoff volume computed from DSRH developed from Clark
IUH model (Table 2). As rainfall–runoff is a complex process and single pair of

2000

1500 2000
Flow (m3/s)

Flow (m3/s)

1000 1500
1000
500
500
0 0
0 10 2030 40 50 60 70 0 10 20 30 40 50 60 70
Time (hour) Time (hour)
Event 12 (Observed) Event 12 (Computed) Event 11 (Observed) Event 11 (Computed)
Event 13 (Observed) Event 13 (Computed)
Event 14 (Observed) Event 14 (Computed)
(a) (b)
1400
1200
1000
Flow (m3/s)

800
600
400
200
0
0 10 20 30 40 50 60 70
Time (hour)
Event 15 (Observed) Event 15(Computed)
(c)
Fig. 9 a to c Observed vs computed DSRH for validation (events E11 to E15). a Computed vs
observed DSRH for event 12. b Computed vs observed DSRH for event 11, 13 and 14. c Computed
vs observed DSRH for event 15
M.M. Ahmad et al.

(Tc , R) does not fully define complex geophysical characteristics of the catchment,
certain deviations are always there. For solution to practical problems the model has
produced reasonable results.

6 Conclusions

A method to estimate Time of Concentration (Tc ) and Storage Coefficient (R)


to develop Clark’s Instantaneous Unit Hydrograph (CIUH) is presented. Tc is
estimated from Time Area Diagram of the catchment and R is determined using
optimization approach based on Downhill Simplex technique. Using CIUH a Direct
surface runoff hydrograph (DSRH) is developed for a large catchment with area
5,598 km2 in a semi-arid region of Pakistan that experiences hill torrent flows. Four
different objective functions are used in optimization to determine the sensitivity
of storage coefficient R. The results during validation are very good with model
efficiency of more than 95% and root mean square error of less than 6%.
The following main conclusions are derived from this study:

• It is not always necessary to develop regional equations for estimating Clark IUH
model parameters. These can be determined from rainfall–runoff observations
for a particular catchment and depend on the length of data record used for
determination of these parameters. To get stable values of R, longer rainfall–
runoff records are preferable. Value of Tc depends on the shape of the Time
Area Diagram which is developed by calculating travel time to the catchment
outlet using empirical model.
• For limited data the objective function based on peak values of discharge gives
better results as compared to the objective function based on the complete
hydrograph.
• It is found that hydrologic response of the catchment is closely related to the ratio
[R/(R + Tc )]. For large catchments, [R/(R + Tc )] ratio is around 0.5 showing
runoff diffusion and translation flow effects play equal role in slow hydrologic
response i.e., long time base of the DSRH.
• DSRH shape is more sensitive to R value than that of Tc showing that runoff
diffusion phenomenon is dominant as compared to translation flow effects when
evaluating hydrologic response of catchments of large size.
• Clark’s unit hydrograph generally under estimates recession side of hydrograph
resulting in errors in runoff volume.
• DSRH derived from Clark IUH model gives acceptable accuracy and model
parameters can be easily updated as additional rainfall–runoff data becomes
available. However, updating of the parameters is possible only for gauged
catchments.

Although the proposed method is applied to a catchment having observed rainfall


and peak flow data, it can be applied to un-gauged catchments by simulating
hypothetical storms and survey of highest flood marks at the outlet. A value of R
determined for single flood event (corresponding to highest flood marks) can be used
to compute different runoff hydrographs for different design storms.
Estimation of Clark’s instantaneous unit hydrograph parameters

References

Al-Wagdany AS, Rao AR (1997) Estimation of the velocity parameter of the geomorphologic in-
stantaneous unit hydrograph. J Water Resour Manage 11(1):1–16. doi:10.1023/A:1007923906214
Blueman AG (2003) Elementary statistics. Mc Graw Hill, New York
Clark CO (1945) Storage and the unit hydrograph. Trans ASCE 110:1419–1446
Gill PE, Murray W, Wright MH (1981) Practical optimization. Academic, London
Hanson G, Hashmi MA VBB VIAK, Sweeden (1995) Conservation and development of land and
water resources of upper Kaha hill torrent catchment. Report No. NES/4804
Jain SK, Singh RD, Sethi SM (2000) Design flood estimation using GIS supported GIUH approach.
Water Resour Manage 14:369–376. doi:10.1023/A:1011147623014
Jawed K (1973) Comparison of methods of deriving unit hydrographs. Colorado State University MS
Thesis
Kirpich ZP (1940) Time of concentration of small agricultural catchments. Civ Eng 10(6):362–365
Kumar R, Chaterjee C, Lohani AK, Kumar S, Singh RD (2002) Sensitivity analysis of GIUH based
Clark model for a catchment. Water Resour Manage 16:263–278. doi:10.1023/A:1021920717410
Kumar R, Chaterrjee C, Singh RD, Lohani AK, Kumar S (2004) GIUH based Clark and Nash models
for runoff estimation for an ungauged basin and their uncertainty analysis. Int J River Manage
2(4):281–290
Lee MT, Blank D, Delleur JW (1972) A program for estimating runoff from Indiana watershed, Part
II. Assembly of hydrologic and geomorphologic data for small watersheds in Indiana, Tech. Rep.
N0. 23. Purdue University Water Resources Research Center, Lafayette
Linsley RK, Kohler MA, Paulhus JLH (1982) Hydrology for engineers, 3rd edn. McGraw-Hill, New
York
Melching CS, Marquard JS (1996) Equations for estimating synthetic unit hydrograph parameter
values for small catchments in Lake County, Illinois, USGS, Rep. 96–474, 49p
NAG (1985) NAG FORTRAN Library. Numerical Analysis Group, London
Nash JE, Sutcliffe JV (1970) River flow forecasting through conceptual models, Part-I: a discussion
of principles. J Hydrol (Amst) 10(3):282–290. doi:10.1016/0022-1694(70)90255-6
Noorbaksh ME (2005) Estimation of instantaneous unit hydrograph with Clark’s method using GIS
technique. J Appl Sci 5(3):445–458
Ponce VM (1989) Engineering hydrology principles and practices. Prin. Hall, NJ
Press WH, Flannery BP, Teukolsky SA, Vetterling WT (1986) Minimization or maximization of
functions. Numerical Recipes, the art of scientific computing, pp 274–334
Sabol GV (1988) Clark unit hydrograph and R-parameter estimation. J Hydraulic Engg, ASCE,
114(1):103–111
Sahoo B, Chatterjee C, Raghuwanshi NS, Singh R, Kumar R (2006) Flood estimation by GIUH-
based Clark and nash models. J Hydrol Eng 11(6):515–525. doi:10.1061/(ASCE)1084-0699
(2006)11:6(515)
Sarangi A, Madramootoo CA, Enright P (2006) Evaluation of three unit hydrograph models to
predict the surface runoff from a Canadian watershed. Water Resour Manage 21:1127–1143.
doi:10.1007/s11269-006-9072-9
Schulz EF (1976) Problems in applied hydrology. Water Resource Publication, Fort Collins
Straub TD, Melching CS, Kocher KE (2000) Equations for estimating Clark unit hydrograph pa-
rameters for small rural catchments in Illinois, Water-Resources Investigations Report 00-4184
USGS
USACE Hydrologic Modelling System HEC-HMS (1998) Technical Reference Manual. (www.hec.
usace.army.mil/software/hec-hms)
Wilson EM (1990) Engineering Hydrol. Macmillan, London, pp 149–152

All in-text references underlined in blue are linked to publications on ResearchGate, letting you access and rea

Вам также может понравиться