Springer Nature is making SARS-CoV-2 and COVID-19 research free. View research | View latest news | Sign up for updates

Application of Soft Computing Techniques for River Flow Prediction in the Downstream Catchment of Mahanadi River Basin Using Partial Duration Series, India


The delta region of the Mahanadi river basin is faced with high floods having arisen many questions like (1) due to the release of Hirakud dam, (2) due to the middle region catchment contribution (Tel tributary at Kantamal station and Ong tributary at Salebhata station) and (3) due to an increase in heavy rainfalls over the basin. To answer this, we choose the partial duration series (PDS) having an average number of flood peaks per year (λ = 3) at Naraj based on the capacity of the existing embankment which mitigates floods up to 28,400 m3/s and analyze the trends of peak discharge of PDS having λ = 3 at the Naraj, Hirakud dam and middle region to know the flow behavior which is responsible for flooding problem in the delta of Mahanadi basin. Subsequently, the trend of extreme rainfall over the entire basin and at the middle region of the basin is analyzed. Analysis of trends of extreme floods in relation to the trends of extreme rainfall to have a proper understanding of the causes of changing floods over the basin is important. Then, runoff at Naraj is estimated based on two techniques: artificial neural networks with different algorithms and an adaptive neuro-fuzzy inference system (ANFIS) using five different input combinations of peak runoff data and rainfall data of upstream locations and compare their performance based on different indices, namely correlation coefficient (R), index of agreement (d), root-mean-square error and Nash–Sutcliffe efficiency (E) to obtain the best input–output model. From the analysis, it is shown that ANFIS techniques as well as the neural network with Levenberg–Marquardt algorithm performs better for estimating runoff at Naraj and that gives the clear idea about the delta flooding. Also, the analysis revealed the answer arisen by many questions regarding the delta flood problem and which are due to the middle region catchment contribution and an increase in extreme rainfall.

This is a preview of subscription content, log in to check access.

We’re sorry, something doesn't seem to be working properly.

Please try refreshing the page. If that doesn't work, please contact support so we can address the problem.

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5
Fig. 6
Fig. 7
Fig. 8
Fig. 9
Fig. 10
Fig. 11
Fig. 12
Fig. 13
Fig. 14


  1. Asokan SM, Dutta D (2008) Analysis of water resources in the Mahanadi River Basin, India under projected climate conditions. Hydrol Process 22:3589–3603

  2. Chau KW et al (2010) A hybrid model coupled with singular spectrum analysis for daily rainfall prediction. J Hydroinform 12(4):458–473

  3. Chen XY et al (2015) A comparative study of population-based optimization algorithms for downstream river flow forecasting by a hybrid neural network model. Eng Appl Artif Intell 46(A):258–268

  4. Dawson CW, Wilby RL (1998) An artificial neural network approach to rainfall runoff modeling. Hydrol Sci J 43(1):47–67

  5. Demuth H, Beale M (1998) Neural network toolbox for use with MATLAB, user’s guide, version 3. The MathWorks Inc., Natick

  6. Fausett L (1994) Fundamentals of neural networks architectures, algorithms, and applications. Prentice-Hall, Upper Saddle River

  7. Gholami V et al (2015) Modeling of groundwater level fluctuations using dendrochronology in alluvial aquifers. J Hydrol 529(3):1060–1069

  8. Ghosh S, Raje D, Mujumdar PP (2010) Mahanadi stream flow: climate change impact assessment and adaptive strategies. Curr Sci 98:1084–1091

  9. Goldberg DE (1989) Flood routing with neural networks using genetic algorithm. In: Engineering Hydrology ASCE (ed) Proceedings. Reston, Va., pp 754–759

  10. Gosain AK, Rao S, Basuray D (2006) Climate change impact assessment on hydrology of Indian River basins. Curr Sci 90:346–353

  11. Haykin S (1999) Neural networks: a comprehensive foundation. Macmillan, New York

  12. Kendall MG (1975) Rank correlation methods. Griffin, London

  13. Mann HB (1945) Nonparametric tests against trend. Econometrica 13:245–259

  14. Mishra SK, Behera K (2009) Development and management of water and energy resource. In: 7th International R&D Conference, 4–6 February, Bhubaneswar, Odisha

  15. Parhi PK, Mishra SK, Singh R, Tripathy VK (2012) Floods in Mahanadi river basin, Orissa (India): a critical review. India Water Week. Water, Energy and Food Security: Call for Solutions, 10–14 April, New Delhi

  16. Patterson D (1996) Artificial neural networks. Prentice-Hall, Singapore

  17. Raje D, Mujumdar PP (2009) A conditional random field-based downscaling method for assessment of climate change impact on multisite daily precipitation in the Mahanadi basin. Water Resour Res 45:W10404

  18. Rao PG (1993) Climatic changes and trends over a major river basin in India. Climate Res 2:215–223

  19. Rao PG (1995) Effect of climate change on stream flows in the Mahanadi river basin, India. Water Int 20:205–212

  20. Sen PK (1968) Estimates of the regression coefficient based on Kendall’s tau. J Am Stat Assoc 63:1379–1389

  21. Seo Y, Kim S, Kisi O, Singh VP (2015) Daily water level forecasting using wavelet decomposition and artificial intelligence techniques. J Hydrol 520:224–243.

  22. Takagi T, Sugeno M (1985) Fuzzy identification of systems and its application to modeling and control. IEEE Trans Syst Man Cybern 15(1):116–132

  23. Taormina R et al (2015) Data-driven input variable selection for rainfall-runoff modeling using binary-coded particle swarm optimization and Extreme Learning Machines. J Hydrol 529(3):1617–1632

  24. Wang W, Xu D, Chau K, Chen S (2013) Improved annual rainfall-runoff forecasting using PSO-SVM model based on EEMD. J Hydroinform 15(4):1377–1390.

  25. Wang WC et al (2015) Improving forecasting accuracy of annual runoff time series using ARIMA based on EEMD decomposition. Water Resour Manag 29(8):2655–2675

  26. Wu CL et al (2010) Prediction of rainfall time series using modular artificial neural networks coupled with data preprocessing techniques. J Hydrol 389(1–2):146–167

  27. Xu J, Chen Y, Bai L, Xu Y (2016) A hybrid model to simulate the annual runoff of the Kaidu River in northwest China. Hydrol Earth Syst Sci 20(4):1447–1457.

Download references


The authors would like express their thanks to the National Institute of Technology, Rourkela, for providing the research environment.

Author information

Correspondence to Nibedita Guru.



Mann–Kendall Test

The Mann–Kendall test is a rank-based test first used by Mann (1945), and then the test statistic was derived by Kendall (1975). The Mann–Kendall S Statistic is computed as follows:

$$S = \sum\limits_{i = 1}^{n - 1} {\sum\limits_{j = i + 1}^{n} {{\text{sign}}\left( {T_{j} - T_{i} } \right)} }$$


$${\text{sign}}\left( {T_{j} - T_{i} } \right) = \left\{ \begin{aligned} 1\quad if\; T_{j} - T_{i} > 0 \hfill \\ 0 \quad if\; T_{j} - T_{i} = 0 \hfill \\ - 1 \quad if\;T_{j} - T_{i} < 0 \hfill \\ \end{aligned} \right.$$

where Tj and Ti are the annual values in years j and i, j > i, respectively.

If n < 10, the value of |S| is compared directly to the theoretical distribution of S derived by Mann and Kendall. The two-tailed test is used. At certain probability level, H0 is rejected in favor of H1 if the absolute value of S equals or exceeds a specified value Sα/2, where Sα/2 is the smallest S that has probability less than α/2 to appear in case of no trend. A positive (negative) value of S indicates an upward (downward) trend.

For n ≥ 10, the statistic S is approximately normally distributed with the mean and variance as follows:

$$E\left( S \right) = 0$$

The variance (σ2) for the S statistic is defined by:

$$\sigma^{2} = \frac{{n\left( {n - 1} \right)\left( {2n + 5} \right) - \sum {t_{i} \left( i \right)\left( {i - 1} \right)\left( {2i + 5} \right)} }}{18}$$

in which ti denotes the number of ties to extent i. The summation term in the numerator is used only if the data series contains related values. The standard test statistic Zs is calculated as follows:

$$Z_{s} = \left\{ \begin{aligned} \frac{s - 1}{\sigma }\quad {\text{for}}\,s > 0 \hfill \\ 0\quad {\text{for}}\,s = 0 \hfill \\ \frac{s + 1}{\sigma }\quad {\text{for}}\,s < 0 \hfill \\ \end{aligned} \right.$$

The test statistic Zs is used a measure of the significance of the trend. In fact, this test statistic is used to test the null hypothesis, H0. If | Zs| is greater than Zα/2, where α represents the chosen significance level (e.g., 5% with Z 0.025 = 1.96); then, the null hypothesis is invalid, implying that the trend is significant.

Liner Regression

Liner regression is a simple technique which stated the positive or negative slope values represent the direction, and the magnitude of the trend was also analyzed. A regression line of the order y = a + bx is fitted to the sample data. If b is found significantly different from zero, then we assume trend to be present in the data. The linear regression-based trends are analyzed using the R-square values.

Levenberg–Marquardt (LM)

The algorithm uses the Hessian matrix which is a second-order training (Demuth and Beale 1998). The LM method is the standard method for minimization of the MSE criterion, due to its rapid convergence properties and robustness.

Gradient Descent with Momentum and Adaptive Learning Rate Algorithm (GDX)

The algorithm calculates derivatives of performance (p) with respect to the weight and bias variables (X) using various epochs. Hagan et al. (1996) provided the detailed description of this training method.

Conjugate Gradient Backpropagation Algorithm (CGF)

The algorithm can use the expression X = X + adX, where dX is the search direction. The parameter a is a constant that minimizes the error functions along the search direction. Fletcher (1987) provided the details of this algorithm.

Root-Mean-Square Error (RMSE) Root-mean-squared error is a measure of the differences between values predicted by model or an estimator and the actually observed values.

$${\text{RMSE = }}\sqrt {\frac{{\sum\limits_{i = 1}^{n} {\left( {O_{i} - P_{i} } \right)^{2} } }}{n}}$$

where Oi is the observed value, Pi is the model-simulated value and n is the total number of observations of the dataset.The Nash–Sutcliffe Efficiency (E) Defined as

$$E = 1 - \frac{{\sum\limits_{i = 1}^{n} {\left( {O_{i} - P_{i} } \right)^{2} } }}{{\sum\limits_{i = 1}^{n} {\left( {O_{i} - \overline{O} } \right)^{2} } }}$$

where Oi is the observed value, Pi is the model-simulated value and \(\overline{O}\) is the mean observed value. Essentially, the closer the model efficiency is to 1, the more accurate the model is.

Index of agreement (d) The index of agreement represents the ratio of the mean square error and the potential error and is defined as:

$$d = 1 - \frac{{\sum\limits_{i = 1}^{n} {\left( {O_{i} - P_{i} } \right)^{2} } }}{{\sum\limits_{i = 1}^{n} {\left( {\left| {P_{i} - \overline{O} } \right| + \left| {O_{i} - \overline{O} } \right|} \right)}^{2} }}$$

The range of d is similar to that of r2 and lies between 0 (no correlation) and 1 (perfect fit).

Correlation coefficient (R) The correlation coefficient (R) is given below:

$$R = \frac{{\sum\limits_{i = 1}^{n} {\left( {Q_{o} - M_{0} } \right)\left( {Q_{p} - M_{p} } \right)} }}{{\sqrt {\sum\limits_{i = 1}^{n} {\left( {Q_{o} - M_{o} } \right)^{2} \sum\limits_{i = 1}^{n} {\left( {Q_{p} - M_{p} } \right)^{2} } } } }}$$

where Q0 and QP are the observed and estimated value, M0 and MP are the means of the observed and estimated value, respectively, and n is the total number of observations of the dataset.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Guru, N., Jha, R. Application of Soft Computing Techniques for River Flow Prediction in the Downstream Catchment of Mahanadi River Basin Using Partial Duration Series, India. Iran J Sci Technol Trans Civ Eng 44, 279–297 (2020).

Download citation


  • Trend analysis
  • Flow prediction
  • ANN
  • Mahanadi river