Estimation of Annual Runoff in Indravati Sub Basin of Godavari River using Statistical Approach

International Journal of Innovative Research in Advanced Engineering (IJIRAE) ISSN: 2349-2763
Issue 07, Volume 3 (July 2016) www.ijirae.com
_________________________________________________________________________________________________
IJIRAE: Impact Factor Value – SJIF: Innospace, Morocco (2015): 3.361 | PIF: 2.469 | Jour Info: 4.085 |
Index Copernicus 2014 = 6.57
© 2014- 16, IJIRAE- All Rights Reserved Page -1
Estimation of Annual Runoff in Indravati Sub Basin of
Godavari River using Statistical Approach
Dr. P. D. Dahe *
Miss B. B. Deshmukh
Department of Civil Engineering Department of Civil Engineering
SGGSIE&T, Vishnupuri, Nanded (Maharashtra) SGGSIE&T, Vishnupuri, Nanded (Maharashtra)
Abstract— Prediction of runoff from known rainfall is one of the major problems confronted by hydrologists. There is
lack of availability of long period runoff records in large number of catchments in India. Investigators have proposed
many empirical relationships for runoff estimation in different catchments based on limited data of parameters
affecting runoff. These regional relationships are useful in planning of water resource projects. This study was
carried out to obtain simple yet effective relationship for estimation of annual runoff in Indravati sub basin of
Godavari river. Regression analysis was carried out using annual rainfall, annual runoff and average annual
temperature data to develop empirical models for annual runoff estimation. GIS software was used for preparing
maps for the study area and to extract the precipitation and temperature data available in grid format from IMD. The
best suited empirical model is then selected as per statistical criteria with lower values of standard error, standard
deviation, mean absolute deviation (MAD), root mean square error (RMSE) and higher values of R square and
correlation coefficient. Statistical significance of selected empirical model was evaluated by paired t test, F test and P
value at 95 % confidence level. The developed relationship is then compared with the existing Khosla and Inglis and
DeSouza relationships. Outcome of this comparison produces encouraging inferences to suggest an effective regional
relationship for annual runoff estimation in the Indravati sub basin of Godavari river in India.
Keywords— Godavari river, runoff estimation, regression analysis, empirical relationship, ungauged catchments.
I. INTRODUCTION
Estimates of annual runoff volumes are needed for efficient planning of water resources projects. There is lack of
availability of long period runoff records in large number of catchments in India. Use of regional empirical relationships
for annual runoff estimation is essential for planning of water resources projects in the absence of historical runoff data.
These empirical relationships for estimation of annual runoff are based on factors affecting runoff like precipitation,
temperature, vegetation cover and drainage density, for which data is readily available. The influence of these factors
affecting runoff is studied using statistical approach and the relationships are proposed. The efficiency and applicability
of these relationships is dependent on the number of independent variables considered along with the length and
reliability of data used in developing such relationships. Further, these relationships cannot be easily generalized for a
large region as the influence of various factors affecting runoff may vary from region to region due to climatic and
physiographic changes.
Based on limited data, earlier investigators in India have developed empirical relationships for annual runoff
relating to factors such as annual rainfall, annual temperature and catchment characteristics. Khosla (1949) [6] developed
relationships for monthly and annual runoff based on a proposition that runoff can be expressed as a difference between
rainfall and losses. He proposed that temperature can be taken as a major factor representing climatic parameters like
sunshine, clouds and wind velocity which are responsible for loss due evaporation and transpiration. Kothyari and Garde
(1991) [3] analyzed data of 55 non snow fed catchments in India and proposed relationships for estimation of mean
annual runoff in five zones of India. They considered mean annual rainfall, mean annual temperature and vegetal cover
as the factors influencing annual runoff.
Subsequently, several attempts have been made to modify the original formula proposed by Khosla (1949) [6]
with varying degree of success in obtaining better results {Panchang (1954) [12], Seghal and Ghulati (1969) [14]}.
However Raja Rao and Pentaiah (1971) [13] found that neither original Khosla formula nor the modified version yielded
results that are comparable with observed runoff in some river basins in Andhra Pradesh. In the present study an attempt
is made to obtain simple yet effective relationships for estimation of annual runoff for Indravati sub basin using statistical
approach.

_________________________________________________________________________________________________
II. STATISTICAL APPROACH FOR RUNOFF ESTIMATION
The importance of estimating the water availability from the available data of hydrologic parameters for
purposes of planning water-resource projects was recognized by engineers even in the last century. With a keen sense of
observation in the region of their activity many engineers of the past have developed empirical runoff estimation
formulae. However, these formulae are applicable only to the region for which they were developed. Moreover these
relationships were developed based on limited data available during these studies. These formulae are essentially rainfall-
runoff relations with additional third or fourth parameters to account for climatic and/or catchment characteristics. There
is a scope to examine the existing relationships and a potential to develop new relationships based on long period data of
hydrologic parameters like precipitation and temperature. Statistical analysis helps in the study of effect of various
hydrologic parameters on annual runoff and development of such empirical relationships.
Regression analysis is a branch of statistical theory that is widely used in almost all scientific disciplines. It is
the basic technique for measuring or estimating the relationship between a dependent variable and one or more
independent variables. The form of simple linear regression is Y = m X + c for one independent variable X. In multiple
regression analysis two or more independent variables are used to estimate the values of the dependent variable. The
general form of equation for multiple linear regression is, Y = β1 X1 + β2 X2 + β3 X3 +……+ βn Xn. The nonlinear form of
equation involving one or more independent variables can be transformed into linear form by taking logarithm of both
the sides of equation and applying least square method for sum of residuals. Statistical significance of obtained
relationship is evaluated by conducting F test, t test, P value and some other measures like root mean square error
(RMSE), standard error, standard deviation and mean absolute deviation (MAD).
III. STUDY AREA
The Indravati river originates in eastern ghats of Dandakaranya range in Kalahandi district of Odisha at an
elevation of 914 m. It follows a westerly path and enters Jagadalpur in the state of Chhattisgarh. The river then moves
towards south before uniting with the Godavari river at the borders of three states namely Maharashtra, Chhattisgarh and
Andhra Pradesh at an elevation of 82.3 m. The Indravati river basin lies between longitude of 800
16’
19’’
E to 830
07’ 10”
E and latitude of 180
43’
25’’
N to 190
26’ 46” N. Average annual rainfall over the basin is 1.588 m. It covers a catchment
area of about 38306 km2
and the length of Indravati river basin is 535 km. The important right bank tributaries of river
Indravati are Bhaskal River, Narangi River, Boarding River, Nibra River, Kotri River and Bandia River, and the left bank
tributary is Nandiraj River. The Upper Indravati hydropower project is one of the biggest dams in India (Fig. 1).
Godavari
River
Basin
Indravati
Sub Basin
Fig. 1. Map of Indravati Sub Basin

_________________________________________________________________________________________________
IV.DEVELOPMENT OF EMPIRICAL RELATIONSHIPS
A. Data and Methodology
Daily data for runoff, precipitation and temperature for the period 1971 to 2005 is used in this study. The runoff
data is collected from WRIS and the precipitation and temperature data from IMD. The daily gauge discharge data is
converted to annual runoff data assuming a water year from 1st
June to 31st
May next year. The catchment area up to the
gauge discharge site was used for this conversion. Precipitation and temperature data is available with IMD in grid
format. A 10
grid data is used in this study. The geo referenced boundary of Indravati basin is clipped and superimposed
on the precipitation and temperature grid provided by IMD to extract the data. The variation of rainfall, runoff and
temperature is presented in Fig. 2 and Fig. 3.
Fig. 2.
Fig. 3.
B. Development of Relationships
Relationships are proposed for annual runoff estimation in Indravati basin for the following two cases:
Case 1 – Relationship considering runoff as a function of rainfall. R = f (P)
Case 2 – Relationship considering runoff as a function of rainfall and temperature. R = f (P&T)
Where, R is annual runoff in centimetres, P annual rainfall in centimetres and T is average annual temperature in 0
C.
Simple regression analysis is employed for Case 1 and Case 2 to obtain empirical relationships. Three different
form of relationships were attempted for the Case 1. In the Case 2, there are two independent variables P and T affecting
the dependent variable R. Runoff is a function of the difference between precipitation and the losses. The evaporation
and transpiration losses are assumed be a function of temperature [R = f (P – T)]. With this assumption (P - T) is
considered as one independent variable and a simple regression is carried out for the Case 2. The best suited empirical
model is then selected as per statistical criteria {Khabat et.al. (2013) [1], Horn (1987) [7]} with lower values of standard
error, standard deviation, mean absolute deviation (MAD), root mean square error (RMSE) and higher values of R square
and correlation coefficient. Statistical significance of selected empirical model was evaluated by paired t test, F test and
P value at 95 % level of confidence.

_________________________________________________________________________________________________
V. RESULTS AND DISCUSSION
A. Case 1: Relationship considering Runoff as a function of Rainfall
In this case three models; (i) bivariate linear (R = a P) shown in Fig. 4, (ii) bivariate nonlinear (R = a Pb
) shown
in Fig. 5 and (iii) polynomial model shown in Fig. 6 were developed. The statistically significant model amongst these
three models is selected with lower values of standard error, standard deviation, mean absolute deviation (MAD), root
mean square error (RMSE) and higher values of R square and correlation coefficient. Statistical study is carried out in
Microsoft Excel. The statistical measures for these three models are computed and presented in Table 1.
Fig. 4. Case 1 (i) Linear Variation of R and P
Fig. 5. Case 1 (ii) Nonlinear Variation of R and P
Fig. 6. Case 1 (iii) Polynomial Variation of R and P

_________________________________________________________________________________________________
TABLE 1. STATISTICAL MEASURES FOR SELECTION OF EMPIRICAL MODEL
Empirical Model Mean
Absolute
Deviation
Standard
Error
Standard
deviation
R2
Correlation
Coefficient
Root Mean Square
Error (RMSE)
R= 0.7779 P - 49.95 11.23 3.46 19.10 0.77 0.88 13.15
R = 0.0078 P1.7998
10.44 2.93 16.86 0.79 0.88 12.87
R = 0.0039 P - 0.34 P + 27.59 18.32 2.36 13.34 0.78 0.88 22
The selected model is of the form bivariate nonlinear (R = a Pb
) and the recommended relationship is R = 0.007
P1.799
. Some statistical tests are performed in Microsoft-Excel to check statistical significance of selected empirical model
and the summary output is shown in Table 2.
TABLE 2. SUMMARY OUTPUT FOR CASE 1 SELECTED MODEL ( R = 0.0078 P1.7998
)
1 Regression Statistics
Multiple R 0.887
R Square 0.787
Adjusted R Square 0.780
Standard Error 0.173
Observations 33
2 Analysis of Variance
ANOVA df SS MS F F critical Significance F
Regression 1 3.462 3.462 114.63 4.16 6.12*
10-12
Residual 31 0.936 0.0302
Total 32 4.398
3 Significance Tests
Coefficients Standard
Error
t Stat tcritical P Value Lower 95% Upper 95%
Intercept -4.855 0.83 -5.85 2.04 1.83*
10-6
-6.554 -3.164
Variable (P) 1.799 0.17 10.71 2.04 6.12*
10-12
1.456 2.142
The summary output Table 2 for Case 1 indicates that the correlation between R and P is positive because the
correlation coefficient (multiple R) between R and P is 0.887. The coefficient of determination R2
is 0.787 which
indicates that 78% of variation in R is explained by the independent variable P. The standard error of regression is 0.173
which measures the variation of R about regression line. The reliability of individual coefficients is tested using paired t
test. The coefficient of P is 1.799 which has a standard error of only 17% and t stat value of 10.71 which is greater than t
critical having a value of 2.04 [tcritical = t (P,n-k-1)]. Hence, the coefficient of P is statistically significant at 95% level of
confidence. The overall significance of obtained relationship is tested by F test and P value. The F value is 114.63 which
is greater than Fcritical [Fcritical = F(P, k, n-k-1)], and the P value is less than 0.05 which indicates that this relationship is
statistically significant at 95% level of confidence.
Considering the test values obtained for F, Significance F, t Stat and P value; the null hypothesis is rejected and
hence the proposed relationship is significant at 95% level of confidence.
B. Case 2: Relationship considering Runoff as a function of Rainfall and Temperature
Several meteorological factors which are responsible for loss from rainfall due to evaporation and transpiration
are directly or indirectly related to temperature. Average annual temperature can be taken as an independent variable to
represent the losses from catchments during the year. Correlation between temperature (T) and observed runoff (R) was
studied and plotted as shown in Fig.7.

_________________________________________________________________________________________________
Fig. 7. Variation of R with T
The plot in Fig. 7. indicates that the runoff decreases with rise in temperature and has a negative correlation
coefficient of -0.35. Khosla (1949) [6] analyzed the rainfall, runoff and temperature data for various catchments in India
and USA to arrive at an empirical relationship. The relationship proposed by him is of the form R = P – xT. Kothyari
and Garde (1991) [3] used the data from 55 non-snow-fed catchments in the Indian subcontinent to verify with existing
relationship given by Kothyari et. al. (1985) [5] and obtained a better relationship for estimation of mean annual runoff
using multiple regression analysis for five zones (Eastern, Western, Northern, Southern and Central) of India. This
relationship is in the form of R = P – a Tb
, where a and b are constants which vary for different catchments. With
reference to previous studies {Khosla (1949) [6], Kothyari and Garde (1991) [3]}, the relationship for Indravati basin
was assumed in the form of R = P – a Tb
and after making some trials the best values for correlation coefficient and R2
were obtained for a = 1.8 and b = 1. Two additional trials were made assuming linear and exponential relationships for R
versus (P - 1.8 T). However, the R2
values for these trials were found to be inferior. Hence the power relationship was
adopted. The variation of R with P – 1.8T as a power relationship is shown in Fig. 8.
Fig. 8. Variation of R with P – 1.8T
Some statistical tests are performed in Microsoft-Excel to check statistical significance of selected empirical
model and the summary output for the Case 2 is shown in Table 3.
TABLE 3. SUMMARY OUTPUT FOR CASE 2
1 Regression Statistics
Multiple R 0.883
R Square 0.78
Adjusted R Square 0.77
Standard Error 0.176
Observations 33

_________________________________________________________________________________________________
2 Analysis of Variance
ANOVA df SS MS F Fcritical Significance F
Regression 1 3.431 3.431 109.94 4.16 1.02*
10-11
Residual 31 0.967 0.031
Total 32 4.398
3 Significance Tests
Coefficients Standard
Error
t Stat tcritical P Value Lower 95% Upper 95%
Intercept -1.031 0.482 -2.14 2.04 0.040 -2.014 -0.0487
Variable ln (P-1.8T) 1.124 0.107 10.48 2.04 1.02*
10-11
0.905 1.343
Considering the test values obtained for F, Significance F, t Stat and P value; the null hypothesis is rejected and
hence the proposed relationship is significant at 95% level of confidence. Results of the obtained rainfall runoff
relationships for Case 1 and Case 2 are then compared with Inglis and DeSouza equation and the Khosla’s formula. This
comparison of statistical test results is presented in Table 4.
TABLE 4. COMPARISON OF CASE1 AND CASE 2 WITH EXISTING RELATIONSHIPS FOR STATISTICAL TESTS RESULTS
Equation Mean Absolute
Deviation
Root Mean Square
Error (RMSE)
Standard
Deviation
Standard
Error
Case 1: R = 0.0078 P1.7998
10.44 12.87 16.86 2.93
Case 2: R= 0.356 (P-1.8T)1.124
8.07 10.79 17.93 3.22
Inglis and DeSouza: P / 254 P ( P – 17.8 ) 11.83 16.52 26.94 4.69
Khosla’s formula: R = P – T / 3.74 74.72 76.91 24.29 4.22
It is evident from the comparison of statistical test results that both the developed relationships for Case 1 and
Case 2 are superior to the existing relationships. The relationship in Case 2 based on temperature as a parameter for
representation of losses is better that the relationship developed for Case 1. The graphical representation of computed
versus observed runoff is presented in Fig. 9.
Fig. 9. Comparison of Case1 and Case 2 with existing relationships for computed versus observed runoff
The observation on superiority of the developed relationships for Case 1 and Case 2 is reinforced by the
graphical representation of computed versus observed runoff presented in Fig. 9. The computed runoff by the proposed
relationships is in close agreement with the observed runoff as compared to the computed runoff by the existing
relationships.

_________________________________________________________________________________________________
VI.CONCLUSIONS
Two empirical relationships are developed for estimation of annual runoff in Indravati sub basin of Godavari
river, using a statistical approach and a long period data record of thirty five years. Both the relationships developed are
superior to the existing relationships. The relationship R = 0.356 (P - 1.8T)1.124
is recommended for estimation of annual
runoff in Indravati sub basin. The relationship R = 0.0078 P1.7998
is recommended if the runoff estimation is to be done on
the basis of only one parameter of annual rainfall. Such relationships are very useful for planning of water resource
projects in ungauged catchments wherein measured runoff data is not available.
REFERENCES
[1]. Khabat, Khosravi, Haidar Mirzai and Iman Saleh. (2013), “Assessment of empirical methods of runoff estimation
by statistical test”, International Journal of Advanced Biological and Biomedical Research, Vol.1, pp. 285-309.
[2]. Regional Model For Annual Runoff Estimation (2000), The Report of National Institute of Hydrology, Jal Vigyan
Bhavan, Vol. No.1, Roorkee.
[3]. U.C. Kothyari and R.J. Garde (1991), “Annual runoff estimation for catchments in India”, Journal of Water
Resource Planning and Management, ASCE, Vol.117, No.1, pp.1-10.
[4]. Kothyari, U.C. (1984), “Estimation of annual runoff from catchments”, M.E. Thesis, University of Roorkee,
Roorkee (India).
[5]. Kothyari, U. C , Garde, R. J., and Seth, S. M. (1985), “Estimation of Annual runoff from catchments”, Central
Board of Irrigation and Power, (India), Vol. 42, No. 4, pp. 319-327.
[6]. Khosla, A.N. (1949), “Analysis and utilization of data for the appraisal of water resources”, Journal of Central
Board of Irrigation and Power, (India), pp. 319-327.
[7]. Dennis R. Horn (1987), “Annual flow statistics for ungauged streams in IDAHO”, Journal of Irrigation and
Drainage Engineering, ASCE, Vol.114, No.3, pp. 463-475.
[8]. Umesh C. Kothyari (1995), “Estimation of monthly runoff from small catchments in India”, Hydrological Sciences
Journal, Vol. 40, No. 4, August 1995, pp. 533-541.
[9]. Linsley, R.K., Kohler, M.A. & Paulhus, J. L. A. (1949), Applied Hydrology, McGraw Hill, New York, USA, p.p.
412-434.
[10]. Ven Te Chow (1964), Hand Book of Applied Hydrology, The McGraw-Hill Companies, New York.
[11]. A.W. Salami, A.A. Mohammad, J.A. Adeyemo and O.K.Olanloku “Assessment of impact of climate change on
runoff in the Kainji lake basin using statistical methods”, Journal of Water Resources and Environmental
Engineering, Vol. 7(2), pp.7-16.
[12]. Panchang, G. M. (1954), Improving goodness of fit of Khosla’s rainfall-runoff formula. Annual Report, Vol. II,
CBIP No. 62.
[13]. Raja Rao, K. N. and Pentiah, A. R. (1971), “A statistical analysis of Khosla’s original and modified formula for
runoff ”, Proc. Symp. Water Resources IISc, Bangalore, A-15.
[14]. Sehagal, S. R. and Ghulati, A. D. (1969), “Estimation of yield from catchments”, Proc. 39th
Annual Res. Session,
CBIP, Ahmedabad, pp. 261-290.

Estimation of Annual Runoff in Indravati Sub Basin of Godavari River using Statistical Approach

More Related Content

Similar to Estimation of Annual Runoff in Indravati Sub Basin of Godavari River using Statistical Approach (20)

More from AM Publications (20)

Recently uploaded (20)

Estimation of Annual Runoff in Indravati Sub Basin of Godavari River using Statistical Approach