Professional Documents
Culture Documents
Performance Evaluation of Modified Versions of SCS Curve Number Method For Two Watersheds of Maharashtra, India
Performance Evaluation of Modified Versions of SCS Curve Number Method For Two Watersheds of Maharashtra, India
To cite this article: R.K. Sahu , S.K. Mishra & T.I. Eldho (2012) Performance evaluation of
modified versions of SCS curve number method for two watersheds of Maharashtra, India, ISH
Journal of Hydraulic Engineering, 18:1, 27-36, DOI: 10.1080/09715010.2012.662425
Download by: [Library Services City University London] Date: 20 December 2015, At: 22:54
ISH Journal of Hydraulic Engineering
Vol. 18, No. 1, March 2012, 27–36
Performance evaluation of modified versions of SCS curve number method for two watersheds of
Maharashtra, India
R.K. Sahua, S.K. Mishrab and T.I. Eldhoc*
a
Department of Soil and Water Conservation Engineering, College of Agricultural Engineering, Rajendra Agricultural
University, Pusa, Samastipur; bDepartment of Water Resources Development and Management, Indian Institute of Technology,
Roorkee; cDepartment of Civil Engineering, Indian Institute of Technology Bombay, Powai, Mumbai, India
(Received 12 February 2011; final version received 28 July 2011)
One of the popular methods for computing the depth of surface runoff for a given rainfall event is the Soil
Downloaded by [Library Services City University London] at 22:54 20 December 2015
Conservation Service Curve Number (SCS-CN) method. Research conducted on the applicability of the SCS-CN
method suggested a need for improvement. Consequently, several modifications of the method have been
suggested and reported in literature. The important modified versions of the method include Mishra and Singh
(MS) model, Michel et al. model, Sahu et al. model and SME model. These modified models have been reported
to perform better than the original method for U.S. watersheds. In the present study, these modified models are
applied to two Indian watersheds in Maharashtra and are compared with each other for their performance. The
results indicated that the SME and the Sahu et al. model perform equally well compared to each other, and the
duo perform consistently better than the original SCS-type method and the others for both the Amba and Kalu
watersheds. Further, the performance of the MS model is found to be better than the Michel et al. model.
Keywords: rainfall-runoff modelling; curve number method; surface runoff
1. Introduction
Rainfall-runoff modelling has a great importance in hydrology and water resources management. The Soil
Conservation Service Curve Number (SCS-CN) method (SCS 1956) is one of the most popular methods for computing
the depth of surface runoff for a given rainfall event from small watersheds. The method is simple and easy to apply.
The primary reason for its wide applicability and acceptability lies in the fact that it accounts for major runoff-
generating watershed characteristics, namely, soil type, land use/treatment, surface condition and antecedent moisture
condition (Mishra and Singh 2002; Mishra and Singh 2003a; Ponce and Hawkins 1996). Though the SCS-CN method
was originally developed for computation of direct surface runoff from storm rainfall, it has since been applied to other
areas, such as long-term hydrologic simulation (Huber et al. 1976; Knisel 1980; Mishra and Singh 2004; Williams and
LaSeur 1976; Woodward and Gbuerek 1992), prediction of infiltration and rainfall-excess rates (Aron et al. 1977; Chen
Cheng-lung 1982; Gray et al. 1982; Mishra 1998; Mishra and Singh 2003b), hydrograph simulation (Mishra et al.
2002), sediment transport (Mishra and Singh 2003a), and partitioning of heavy metals (Mishra et al. 2003, 2004a).
Research conducted on the applicability of the SCS-CN method suggests a need for improvement (Ponce and
Hawkins 1996). Although several modifications of the method have been suggested and reported in the literature,
further improvement of the method is needed (Mishra and Singh 2002). The important modified versions of the
method include the Mishra and Singh (MS) model (Mishra and Singh 2002), the Michel et al. (2005) model, the Sahu
et al. (2007) model and the SME model (Sahu et al. 2010). These modified models have been reported to perform better
than the original method for U.S. watersheds (Sahu et al. 2007; Sahu et al. 2010). In the present study, therefore, these
modified models along with the original SCS-CN model are applied to two Indian watersheds in Maharashtra and are
compared with each other for their performance.
and
Ia ¼ S, ð2Þ
where Q ¼ direct runoff (mm), P ¼ total precipitation (mm), Ia ¼ initial abstraction (mm), S ¼ potential maximum
retention (mm) and ¼ initial abstraction coefficient (dimensionless). Further, the parameter S is expressed as
25400
S¼ 254, ð3Þ
CN
where S is in mm and CN is the curve number, which depends on the above-described four characteristics. The
parameter CN depends on land use, hydrologic soil group, hydrologic condition and antecedent moisture condition
(AMC) (SCS 1956). Because is taken to be equal to 0.2 (standard), the resulting equation for computing direct runoff
model is:
ðP 0:2SÞ2
Q¼ for P 4 Ia
P þ 0:8S ð4Þ
¼ 0 otherwise
Equation (4) in combination with Equation (3) forms the popular form of the SCS-CN method, and this popular
form is applied in the present study. Further, the present study calibrates the parameter CN instead of taking its value
Downloaded by [Library Services City University London] at 22:54 20 December 2015
from traditional CN table. Hence, this approach of application may be termed as a SCS-type method consisting of
Equations (2)–(4).
Following are the main advantages (Mishra and Singh 2003a; Ponce and Hawkins 1996) of the SCS-CN method:
(1) It is a simple conceptual method for predicting direct surface runoff from a storm rainfall amount, and is well
supported by empirical data and wide experience.
(2) It is easy to apply and useful for ungauged watersheds.
(3) The method relies on only one parameter CN.
(4) The parameter CN is a function of the watershed characteristics and, hence, the method exhibits responsiveness
to major runoff- producing watershed characteristics.
However, the method has the following limitations:
(1) The three AMC levels used with the SCS-CN method permit unreasonable sudden jumps in CN, and hence
corresponding sudden jumps in computed runoff is possible.
(2) There is a lack of clear guidance on how to vary antecedent moisture condition.
(3) There is no explicit dependency of initial abstraction on the antecedent moisture.
(4) The method doesn’t contain any expression for time and ignores the impact of rainfall intensity and its
temporal distribution.
(5) It is applicable to only small watersheds. Ponce and Hawkins (1996), for example, cautioned against its use to
watersheds larger than 250 sq. Km.
Here, Ia is the same as in Equation (2) and P5 denotes the amount of antecedent 5-day rainfall.
The combination of Equations (2), (5) and (6) is designated as the Mishra–Singh (MS) model. This method
advantageously obviates sudden jumps in CNs and hence in computing runoff through incorporation of the expression
of M (Equation (6)) replacing the three AMCs. However, it does not show an explicit dependency of Ia on M. Further,
in this method, S is optimised as a parameter, which is, in fact, a varying quantity depending on M for a given
watershed. Hence, it is not clear as to which moisture level/condition the optimised S would correspond to.
ISH Journal of Hydraulic Engineering 29
ðP þ V0 Sa Þ2
If Sa P 5 V0 5 Sa , then Q¼ : ð9Þ
Downloaded by [Library Services City University London] at 22:54 20 December 2015
P þ V0 Sa þ S
ðS þ Sa V0 Þ2
If Sa V0 Sa þ S, then Q¼P 1 : ð10Þ
S2 þ ðS þ Sa V0 ÞP
Equations (8)–(10) constitute the Michel et al. (Michel) model. Despite the improvement over the popular form of
SCS-CN method from the SMA view-point, the Michel model lacks an expression for the initial soil moisture store
level (V0).
0:44P5 0:004S
If P5 4 0:1S then, V0 ¼ S : ð16Þ
P5 þ 0:9S
ðP þ V0 0:1SÞ2
If 0:1S 5 V0 þ P 5 0:1S þ P, then Q ¼ : ð18Þ
P þ V0 þ 0:9S
ð1:1S V0 Þ2
If 0:1S V0 1:1S, then Q ¼ P 1 : ð19Þ
S2 þ ð1:1S V0 ÞP
¼ 0, otherwise ð20Þ
4. Applications of models
4.1. Study area
The original SCS-type method and its important modified versions, namely, the MS model, Michel model, Sahu-3p
model, Sahu-1p model and SME model, were applied to two Indian watersheds to evaluate their comparative
performance. The selected watersheds are Amba and Kalu watersheds located in Maharashtra State in India.
between the East Longitude 73 360 to 73 490 and the North Latitude 19 170 to 19 260 . The location map of the
watershed is shown in Figure 2. The catchment area of the Kalu watershed is 22,400 ha (224 km2), which is less than the
maximum limit of 250 km2 (Ponce and Hawkins 1996) for applicability of SCS-CN method. The watershed experiences
an average annual rainfall of 2450 mm. Most of the rains are received during June to October. Existing cropping
pattern of the cultivation covers 46% of paddy, 16% of nanchani vari, 3% of pulses and 35% of grass. The catchment is
covered with about 50% thickly wooded forest and about 50% cultivated area. For the present study, nine rainfall-
runoff data sets (Table 2) from years 1999 to 2001 were used. In the Kalu watershed, however, there were five
raingauges but as continuous data were available for only one raingauge, the data of only one station was used in the
present study.
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u N
u1 X
RMSE ¼ t ðQobs Qcomp Þ2i , ð25Þ
N i¼1
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
PN ffi
1 2
N i¼1 ðQ obs Q comp Þ i
NRMSE ¼ , ð26Þ
Qobs
"
PN 2 #
i¼1 Qobs Qcomp i
EF ¼ 1 P 2 , ð27Þ
N
i¼1 Qobs Qobs i
where SE and RMSE are in millimetres while NRMSE and EF are dimensionless. Qobs is the observed storm runoff
(mm), Qcomp is the computed runoff (mm), Qobs is average of observed runoff values in a watershed, N is the total
number of rainfall-runoff events, m is the number of model parameters and i is an integer varying from 1 to N.
The higher the SE or RMSE, the poorer is the performance of the model, and vice versa. SE ¼ 0 or RMSE ¼ 0
exhibits a perfect fit. SE has the advantages of having the same units (dimensions) as the variable, properly accounting
for the degrees of freedom and being valid for nonlinear as well as linear models (McCuen 2003). Obviously, the unit of
SE or RMSE here is mm. Mishra and Singh (1999) and many others used the SE criterion for comparing model
performance. The works of Madsen et al. (2002) and Mishra et al. (2004b, 2005) are but a few examples among many
others to cite the wide usage of RMSE. It may be noted that SE and RMSE are equivalent for one parameter model,
that is, for m ¼ 1. Similarly, higher the NRMSE, poorer is the performance of the model, and vice versa. It has a
minimum value of 0.0, with a perfect agreement close to it. However, it may also be represented in percentage by
multiplying it by 100. Obviously, NRMSE gives RMSE per unit mean storm runoff depth. EI-Sadek et al. (2001),
Coffey et al. (2004) and many others have used this criterion for comparison of models.
ISH Journal of Hydraulic Engineering 33
Amba watershed
An EF (efficiency) value of unity indicates perfect agreement between observed and computed values and
decreasing values indicate poorer agreement (Fentie et al. 2002). EF ¼ 0 implies that the model predicts no better than
the average of the observed data (Coffey et al. 2004). The value of model efficiency can be negative, which indicates
that the average observed value is a better estimate than the model predicted, that is, model prediction is worse than the
average observation (EI-Sadek et al. 2001; Fentie et al. 2002). Among others, Michel et al. (2005) used this criterion to
compare the model performance.
All the six models, considered for the present study, have been applied to the two selected Indian watersheds. The
values of SE, RMSE, NRMSE and EF resulting from the applications of each of the models to the Amba and Kalu
watersheds are presented in Tables 4 and 5, respectively.
5. Performance evaluation
5.1. Amba watershed
In the present study, six rainfall-runoff data sets from the years 1999–2001 are selected for application of the models to
the Amba watershed. The values of SE, RMSE, NRMSE and efficiency yielded by application of the considered
models to this watershed are presented in Table 4. It is seen from this table that the SME model and the Sahu-3p model
yield nearly equal values of SE, RMSE, NRMSE and efficiency. These values for the former model are, respectively,
34 R.K. Sahu et al.
Downloaded by [Library Services City University London] at 22:54 20 December 2015
Figure 3. Fitting of existing SCS-type method and the modified models to the Amba watershed.
Kalu watershed
0.79 mm, 0.65 mm, 0.11 and 0.98 whereas for the latter model they are 0.78 mm, 0.63 mm, 0.13 and 0.98, respectively.
Lower values of SE, RMSE, NRMSE and higher value of EF than those resulted by other models (Table 4) indicate
that these two models perform equally well and better than the SCS-type, MS, Michel and Sahu-1p models for the
Amba watershed.
Because the values of SE, accounting for the effect of number of model parameters, yielded by the SME model and
the Sahu-3p model, are considerably lower (about half) than those by the existing SCS-type method (SE ¼ 1.50) and
the Michel model (SE ¼ 1.46), the duo can be considered to have performed significantly better than the latter two
(Table 4). The performance of the MS model is better than the Michel model. Similar performance is observed based
on RMSE, NRMSE and the efficiency criteria (Table 4). On the other hand, the Sahu-1p (1-parameter) model
performs insignificantly better than the SCS-type method. The insignificant improvement of the former over the latter
asserts the general notion that the simplifications based on the particular set of climate and geomorphology are not
necessarily applicable to the other set.
Figure 3 exhibits typical fitting of the SCS-type, MS, Michel, Sahu-3p, Sahu-1p and SME models for the Amba
watershed. It is evident from this figure that the runoff values computed using the SME model and the Sahu-3p model
are very close to the corresponding observed values—and closer than those using other models for most of the events.
Hence, these two models perform equally well but better than the other models. Among the one-parameter models
(SCS-type and Sahu-1p models), there is no improvement with regard to the consistency.
Figure 4. Fitting of existing SCS-type method and the modified models to the Kalu watershed.
(Table 5). Further, the MS model shows an improvement over the Michel model. Among the one-parameter models,
the Sahu-1p model performs insignificantly better than the SCS-type method. Because the Sahu-1p model has many
hydrological advantages over the latter, it can be preferred over the other.
Figure 4 exhibits fitting of the SCS-type, MS, Michel, Sahu-3p, Sahu-1p and SME models for the Kalu watershed.
From this figure, the SME and the Sahu-3p models perform, in general, better than the others. The SCS-type method
appears to be inconsistent in performance.
6. Discussion
The results of various models indicates that the SME model and the Sahu-3p model are the best-performing models for
Kalu and Amba watersheds, and the duo perform consistently better than the SCS-type method. Among one-
parameter models, the Sahu-1p model however does not show a considerable improvement over the SCS-type method
but is more consistent in performance. It is worth mentioning that in spite of the much larger catchment area of the
Kalu watershed compared with those of the U.S. watersheds, used for developing the considered models, and the
Amba watershed, all the six SCS-CN-based models perform reasonably well. However, the performance of the models
for the Kalu watershed is poorer than that for the Amba watershed. The poorer performance when using the Kalu
watershed (area ¼ 22,400 ha) data set compared to that with the Amba watershed (area ¼ 3530 ha) is attributed to the
comparatively much larger (6.35 times) catchment area of the former. Increase in catchment area practically results in
higher infiltration and other losses before reaching the outlet, and hence results in lower direct runoff for a given
amount of rainfall. In the Kalu watershed, however, there were five raingauges but as continuous data were available
for only one raingauge, the data of only one station was used in the present study. This is also one of the reasons of
poorer performance of the models with the Kalu watershed. Further, as the SCS-CN-based models do not consider
spatial effect on runoff, they perform poorly on larger catchments.
7. Conclusions
From the present study, the following conclusions can be drawn:
(1) The SME and the Sahu-3p models perform equally well compared with each other, and the duo perform
consistently better than the SCS-type, MS, Michel and Sahu-1p models for both the Amba and Kalu
watersheds.
(2) Performance of Sahu-1p model is marginally better than the SCS-type method for the Amba watershed and is
(nearly) the same for the Kalu watershed.
(3) Performance of the MS model is better than the Michel model.
Acknowledgements
The authors are grateful to the officials of the Hydrology Project, Irrigation Department, Government of Maharastra,
Nasik (India), for providing hydrological data for the Amba and Kalu watersheds.
36 R.K. Sahu et al.
References
Aron, G., Miller Jr, A.C., and Lakatos, D.F. (1977). ‘‘Infiltration formula based on SCS curve number.’’ J. Irrig. Draing. Div.
ASCE., 103(IR4), 419–427.
Chen Cheng-lung (1982). ‘‘An evaluation of the mathematics and physical significance of the soil conservation service curve number
procedure for estimating runoff volume.’’ Rainfall-runoff relationship, V.P. Singh, ed., Water Resources Publications, Littleton,
CO, 80161.
Coffey, M.E., Workman, S.R., Taraba, J.L., and Fogle, A.W. (2004). ‘‘Statistical procedures for evaluating daily and monthly
hydrologic model predictions.’’ Trans. A.S.A.E., 47(1), 59–68.
EI-Sadek, A., Feyen, J., and Berlamont, J. (2001). ‘‘Comparison of models for computing drainage discharge.’’ J. Irrig. Draing.
Engg. ASCE., 127(6), 363–369.
Fentie, B., Yu, B., Silburn, M.D., and Ciesiolka, C.A.A. (2002). ‘‘Evaluation of eight different methods to predict hillslope runoff
rates for a grazing catchment in Australia.’’ J. Hydrol., 261(1–4), 102–114.
Gray, D.D., Katz, P.G., deMonsabert, S.M., and Cogo, N.P. (1982). ‘‘Antecedent moisture condition probabilities.’’ J. Irrig.
Draing. Div. ASCE., 108(2), 107–114.
Huber, W.C., Heaney, J.P., Bedient, B.P., and Bender, J.P. (1976). Environmental resources management studies in the Kissimmee
River basin. Rep. No. ENV-05-76-3. Department of Environmental Engineering Science, University of Florida, Gainesville,
Downloaded by [Library Services City University London] at 22:54 20 December 2015
FL.
Knisel, W.G. (1980). CREAMS, A field scale model for chemicals, runoff, and erosion from agricultural management systems.
Conservation Rep. No. 26. ARS, U.S. Department of Agriculture: Washington.
Madsen, H., Wilson, G., and Ammentorp, H.C. (2002). ‘‘Comparison of different automated strategies for calibration of rainfall-
runoff models.’’ J. Hydrol., 261(1–4), 48–59.
Marquardt, D.W. (1963). ‘‘An algorithm for least-squares estimation of nonlinear parameters.’’ J. Soc. Indust. Appl. Math., 11(2),
431–441.
McCuen, R.H. (2003). Modeling hydrologic change: statistical methods. Lewis Publishers, A CRC Press Company, Boca Raton, FL,
ISBN 1-56670-600-9.
Michel, C., Vazken, A., and Perrin, C. (2005). ‘‘Soil conservation service curve number method: how to mend a wrong soil moisture
accounting procedure.’’ Water Resource Res., 41, 1–6.
Mishra, S.K. (1998). ‘‘Operation of a multipurpose reservoir.’’ Ph.D. thesis, University of Roorkee, Roorkee, India.
Mishra, S.K., Jain, M.K., Pandey, R.P., and Singh, V.P. (2003b). ‘‘Evaluation of AMC-dependant SCS-CN-based models using
large data of small watersheds.’’ J. Water Energy Int., 60(3), 13–23.
Mishra, S.K., Jain, M.K., Pandey, R.P., and Singh, V.P. (2005). ‘‘Catchment area-based evaluation of the AMC-dependent SCS-
CN-inspired rainfall-runoff models.’’ J. Hydrol. Proc., 19(14), 2701–2718.
Mishra, S.K., Jain, M.K., and Singh, V.P. (2004b). ‘‘Evaluation of the SCS-CN-based model incorporating antecedent moisture.’’ J.
Water Resources Manag., 18, 567–589.
Mishra, S.K., Sansalone, J.J., and Singh., V.P. (2004a). ‘‘Overland metal partitioning using SCS-CN concept.’’ J. Environ. Eng.
ASCE., 130(2), 145–154.
Mishra, S.K., and Singh, V.P. (1999). ‘‘Another look at the SCS-CN method.’’ J. Hydrol. Eng. ASCE., 4(3), 257–264.
Mishra, S.K., and Singh, V.P. (2002). ‘‘SCS-CN-based hydrologic simulation package.’’ Mathematical models in small watershed
hydrology and applications, V.P. Singh and D.K. Frevert, eds., Water Resources Publications, Littleton, CO 80161, 391–464.
Mishra, S.K., and Singh, V.P. (2003a). Soil conservation service curve number (SCS-CN) methodology, Kluwer Academic
Publishers, Dordrecht.
Mishra, S.K., and Singh, V.P. (2003b). ‘‘Validity and extension of the SCSCN method for computing infiltration and rainfall-excess
rates.’’ J. Hydrol. Proc., 18(17), 3323–3345.
Mishra, S.K., and Singh, V.P. (2004). ‘‘Long-term hydrological simulation based on the soil conservation service curve number.’’ J.
Hydrol. Proc., 18(7), 1291–1313.
Mishra, S.K., Singh, V.P., Ojha, C.S.P., Aravamuthan, V., and Sansalone, J.J. (2002). ‘‘An SCS-CN-based time distributed runoff
model.’’ J. Water Energy Int. Central Board Irrig. Power, 59(2), 34–51.
Nash, J.E., and Sutcliffe, J.V. (1970). ‘‘River flow forecasting through conceptual models: Part I—a discussion of principles.’’ J.
Hydrol., 10(3), 282–290.
Ponce, V.M., and Hawkins, R.H. (1996). ‘‘Runoff curve number: has it reached maturity?.’’ J. Hydrol. Eng. ASCE., 1(1), 11–19.
Sahu, R.K., Mishra, S.K., and Eldho, T.I. (2010). ‘‘An improved AMC-coupled runoff curve number model.’’ J. Hydrol. Proc.,
24(20), 2834–2839.
Sahu, R.K., Mishra, S.K., Eldho, T.I., and Jain, M.K. (2007). ‘‘An advanced soil moisture accounting procedure for SCS curve
number method.’’ J. Hydrol. Proc., 21(21), 2872–2881.
SCS (1956). In Hydrology, national engineering of handbook, Soil Conservation Service, USDA; Washington, DC, Supplement A,
Section 4, Chapter 10.
Williams, J.R., and LaSeur, V. (1976). ‘‘Water yield model using SCS curve numbers.’’ J. Hydraulic Eng. ASCE., 102(HY9),
1241–1253.
Woodward, D.E., and Gbuerek, W.J. (1992). Progress report ARS/SCS curve number work group, Proc. ASCE Water Forum,
ASCE, New York, 378–382.