Available online at [Link].

org

Civil Engineering Journal

Vol. 5, No. 10, October, 2019

A Comparison of Emotional Neural Network (ENN) and Artificial

Neural Network (ANN) Approach for Rainfall-Runoff Modelling

Suraj Kumar a, Thendiyath Roshni b*, Dar Himayoun a

a
Research Scholar, Department of Civil Engineering, National Institute of Technology, Bihar, Patna, 800005, India.
b
Assistant Professor, Department of Civil Engineering, National Institute of Technology, Bihar, Patna, 800005, India.
Received 14 June 2019; Accepted 06 September 2019

Abstract
Reliable method of rainfall-runoff modeling is a prerequisite for proper management and mitigation of extreme events such
as floods. The objective of this paper is to contrasts the hydrological execution of Emotional Neural Network (ENN) and
Artificial Neural Network (ANN) for modelling rainfall-runoff in the Sone Command, Bihar as this area experiences flood
due to heavy rainfall. ENN is a modified version of ANN as it includes neural parameters which enhance the network
learning process. Selection of inputs is a crucial task for rainfall-runoff model. This paper utilizes cross correlation analysis
for the selection of potential predictors. Three sets of input data: Set 1, Set 2 and Set 3 have been prepared using weather
and discharge data of 2 raingauge stations and 1 discharge station located in the command for the period 1986-2014.
Principal Component Analysis (PCA) has then been performed on the selected data sets for selection of data sets showing
principal tendencies. The data sets obtained after PCA have then been used in the model development of ENN and ANN
models. Performance indices were performed for the developed model for three data sets. The results obtained from Set 2
showed that ENN with R= 0.933, R2 = 0.870, Nash Sutcliffe = 0.8689, RMSE = 276.1359 and Relative Peak Error =
0.00879 outperforms ANN in simulating the discharge. Therefore, ENN model is suggested as a better model for rainfall-
runoff discharge in the Sone command, Bihar.
Keywords: Emotional Neural Network (ENN); Artificial Neural Network (ANN); Cross Correlation; Principal Component Analysis
(PCA); Rainfall-Runoff.

1. Introduction
The rainfall-runoff relationship is one of the most complex hydrological phenomena due to presence of complex
non-linear relationships in the transformation of rainfall into runoff. This process is quite difficult to comprehend, owing
to the presence of huge number of variables involved in the demonstration of physical process [1-4]. Therefore its precise
modelling is important for water resources management and development and the prediction of natural calamities like
droughts and floods. Based on the involvement of physical aspects, rainfall-runoff models are classified as either
physical-based models or system theoretic models [5-7]. The physical-based models also called data driven models
require the considerable information about the system mechanism as well as its parameters. However, the system
theoretic models do not concern much about the physical processes of the problem. These models are primarily based
on rainfall and runoff data and seek to characterize nonlinearity and non-stationary behaviour from those data by the use
of transfer functions [8-10]. Among the system theoretic models, Artificial Neural Network (ANN) based models for
rainfall-runoff modelling have received global attention because of their capability to capture high degree of non-

* Corresponding author: roshni@[Link]

[Link]

© 2019 by the authors. Licensee C.E.J, Tehran, Iran. This article is an open access article distributed under the terms
and conditions of the Creative Commons Attribution (CC-BY) license ([Link]

2120
Civil Engineering Journal Vol. 5, No. 10, October, 2019

linearity and complex nature of relationship between the hydrological variables without fully understanding the
processes beneath [11-16].
The ANNs have black box properties and offer a relatively quick and flexible technique of modelling. Owing to their
parallel architecture, these models can treat nonlinearity and non-stationary behavior in the data to some extent [17-19].
Several studies could be found in the technical literature that have reported that ANNs outperform the traditional
statistical rainfall-runoff models [17, 11]. The ANNs are more promising alternatives for conceptual rainfall-runoff
models.
Ghumman et al. [11] used ANN based rainfall- runoff model and compared the results with statistical conceptual
model. The results of their study reported that ANN based approach for rainfall-runoff modelling is more promising
alternative to conceptual models and this approach could be used when the dataset is of low standard and/or short range.
Demirel et al., [17] investigated the ability of both Soil and Water Assessment Tool (SWAT) and ANN model in flow
forecasting. They compared the results of SWAT model with ANN model based on the prediction accuracy and
concluded that ANNs can be more powerful tools in daily flow forecasts. Despite these promising results, several other
studies reveal inefficiency and drawbacks of ANNs over the other rainfall-runoff models [19-22]. The deficiencies like
overtraining of data and underestimation of peak values can be fixed to some extent by several data pre-processing
approaches. For instance, the ability of wavelet based data pre-processing approaches in decomposing complex
hydrological time series into sub-series can be very effective for interpreting hydrological phenomena [23-25]. This
technique extracts the useful information from data series at different scales to enhance the modelling efficiency and to
extract the seasonal features of the rainfall-runoff process across most areas of hydrology [25-28].
The flexible and data dependent structure of ANN leaves a huge room for its improvement in the context of rainfall-
runoff modelling [29]. Wu and Chau [15] employed ANN coupled with Singular spectrum Analysis (SSA) for rainfall-
runoff modelling. The purpose of coupling SSA with ANN was to reduce the lag effect in ANN and the results showed
that SSA improves the model performance and can eliminate the lag effect. Fellous [30] was the first study that
incorporated the emotions into Artificial Intelligence (AI) systems suggesting that emotions must be dynamically
interacted with together. After that other researchers investigated the role artificial emotions in AI models by the
integration of artificial emotions into the classic ANN framework as Emotional Neural Network (ENN) model [31, 32].
The neurophysiological response of animals from a biological point of view, could be affected by the emotion and mood
of animal due to the hormonal activities so that the animals at different moods may provide different actions for the
same task. Employing this concept by merging neural network with artificial emotions, there will be a feedback loop
between the hormones and neural systems that could relatively enhanced the learning ability of a neural network. A few
studies could be find in the technical literature that have successfully applied ENN in hydrological studies [31]. Sharghi
et al. [29] employed EANN and WANN approaches for modelling the rainfall runoff process. They reported the
superiority of EANN over ANN as well as the better learning capability of EANN in extraordinary and extreme
conditions of the training phase.
In this study we have applied ANN and ENN approaches to model rainfall-runoff process of Sone river command,
Bihar. We employ Cross Correlation Function (CCF) and Auto Correlation Function (ACF) for the selection of input
parameters at different lags. In order to remove the redundancy in input variable we applied Principal Component
Analysis (PCA). The results derived from different input combination were compared in both the models and the best
combinations were chosen based on the evaluation criteria. Finally the results of ENN are compared with the results of
ANN by graphical indicators as well as by the selected evaluation criteria.

2. Study Area and Research Data

The study area (Figure 1) comprises of the Sone Irrigation Project in Bihar, India. It is a river diversion scheme built
across the river Sone, which is a tributary of river Ganga. It lies at latitude 24°48’N and longitude 84°07’E. The regions
to be specific Patna, Jehanabad, Aurangabad, Gaya, Buxar, Bhojpur, Rohtas and Bhabua are secured under this task.
17,651 sq. km (25%) of the total catchment area of the river i.e. 71,259 sq. km falls in Bihar. Practically plain in
geography, it consistently inclines towards the Ganga River. These factors are ideally suited for the development of
irrigated agriculture. The major crop grown in the command in the Kharif (monsoon) season is rice and in Rabi season
is wheat. Rice is the major crop and other crops occupy less than 2% of cultivable command area. Sugarcane is the main
money crop developed in the territory. Linseed and mustard oilseeds are likewise developed in Rabi season over a little
zone. The water system framework has been employable since 1871 yet water system in a sorted out way began in 1879.

2121
Civil Engineering Journal Vol. 5, No. 10, October, 2019

Figure 1. Location of study area (Sone River Command, Bihar, India)

The data utilized for carrying out this work consists of rainfall and discharge data for the period 1986-2014 (Figure
2).The rainfall data has been obtained from Indian Meteorological Department (Pune) and the discharge data has been
obtained from CWC (Patna) (Table 1). The details of the data are shown in table
Table 1. Statistics of the monthly time series of the Sone Command

Station
Time Series Statistical Parameter
Indepuri Buxar Koelwar
Mean 78.545 78.81
Maximum 674.8 875.1
Rainfall(mm)
Minimum 0 0
Standard Deviation 118.75 132.126
Mean 462.09
Maximum 7557.577
Discharge(m3/s)
Minimum 10.7
Standard Deviation 762.87

Figure 2. Box Plot of the data used

2122
Civil Engineering Journal Vol. 5, No. 10, October, 2019

3. Research Methodology
3.1. Principal Component Analysis (PCA)
Principal Component Analysis (PCA) is a procedure mostly known for extraction of constituents for multivariate
analysis. The extracted constituents obtained using PCA are uncorrelated. Its objective is to separate the significant data
from the information table and to express this data as a lot of new symmetrical factors called principal components. In
this study PCA analysis of the datasets has been done in MATLAB.

3.2. Artificial Neural Network (ANN)

The Artificial Neural Network (ANN) is a powerful soft computational technique composed of interconnected nodes
(like neurons in the human brain) that are linked by weighted synaptic connections [2, 11, 31]. This technique inspired
by the biological cerebral activity is widely applied in the forecasting of water resource and hydrology [2, 11, 17, 19,
31, 33]. This approach being faster, flexible, highly adaptive and robust in newer and noisy environments can solve a
wide in the range of problems. Extensive research has been successfully carried out regarding the implementation of
ANN in engineering related fields like rainfall-runoff modelling, time series prediction and rule-based control. In ANN,
Back Propagation (BP) algorithm network models are common to engineers. As it has been proved that the three-layer
BP network model is satisfied for all kinds of engineering problems in terms of simulation and forecasting [15]. The
three layer Feed Forward Neural Network (FFNN) usually employed in hydrological time-series forecasting, provide a
general framework to perform an input–output mapping using an arrangement of interconnected sensorial components.
A detailed explanation regarding the properties of ANN could be found in the study of [18, 23, 31].

3.3. Emotional Neural Network (ENN)

An Emotional Neural Network (ENN) is the new generation of the conventional ANN model, which contains an
artificial emotional system where the emission of hormones take place in order to adjust the operation of neurons. The
hormonal parameters in the ENN model can be modified by input and output value of neurons of the network. It is
evident from the Figure 3, that in each neuron of the ENN model, the information repeatedly passes from inputs to
output and vice versa. In addition to that, each node (neuron) provide dynamical hormones of Ha, Hb, and Hc. These
hormones according to the input and output values are firstly initialized in the training phase of model and then are
modified through the learning process. In the training phase of the network, these hormonal coefficients can impact the
other components of the node such as activation function, weights and net function (Figure 3). The solid and dotted lines
in the Figure 3, represent the neural and hormonal paths of the information. The output of the i th neuron in the proposed
ENN model with three hormonal glands of Ha, Hb, Hc is computed as follows [6, 29]:

 
       
Yi   i    i ,h H h  f     i    i,h H h    i , j    i , j ,k H h X i,j 
  j  
h
 h  h  
1  2 3  (1)
   
   i    i ,h H h     i    i ,h H h 
h h
4 5

Whereas the overall hormone value is computed as follows [6, 29]:

H h   H i ,h ; H  (a,b,c) (2)
i

In the Equation 1, the first term represents the applied weight to the activation function (f). It consists of both the
dynamic hormonal weight of ∑ℎ 𝜎𝑖,ℎ 𝐻ℎ and the statistic neural weight of 𝜆𝑖 . The second term represents the applied
weight to the summation function, the third term represents the applied weight to the input value of 𝑋𝑖,𝑗 coming from jth
neuron of previous layer and the fourth term represents the bias of net function from both the hormonal and the neural
weights of ∑ℎ 𝑥𝑖,ℎ 𝐻ℎ and 𝛼𝑖 respectively. Finally, the fifth term contributes to the activation function, where hormonal
and neural weights contribute as ∑ℎ 𝜌𝑖,ℎ 𝐻ℎ and 𝛿𝑖 respectively.

2123
Civil Engineering Journal Vol. 5, No. 10, October, 2019

VII

VII
Ha Hb Hc
IX
XI X
XII

Input   f (x)
V
Output

II III
I
ith node IV
IV

Figure 3. ENN model architecture

The ith node of the network output (Yi) will provide hormonal feedback of Hi,h to the network as follows [6, 29]:
H i ,h  glandity i ,h  Yi (3)

Where, 𝑔𝑙𝑎𝑛𝑑𝑖𝑡𝑦𝑖,ℎ is a parameter representing the production factory of all hormones in the gland. This parameter
should be calibrated to produce desired level of hormone in each gland. By considering Equations 1 and 2 and the
network output (Yi), the hormone value is updated through the training process to get the reliable agreement between
the observed and computed time series of the target.
In this study the time series data of rainfall and runoff have been utilised to develop the best rainfall-runoff model
utilising cross correlation and PCA. The flowchart of the methodology adopted in this study is shown in Figure 4.

Figure 4. Flowchart of methodology

2124
Civil Engineering Journal Vol. 5, No. 10, October, 2019

3.3. Efficiency Parameters

In this work the author intended to use coefficient of determination (R 2), Nash Sutcliffe Efficiency (ENS), Root
mean square error as efficiency parameters for determining the accuracy of the model developed.
Coefficient of determination (R2) measures the dispersion between observed value and simulated value. It is
expressed as:-
 Coefficient of determination:
𝑛(∑ 𝑄𝑜𝑏𝑠 ∗𝑄𝑠𝑖𝑚 )−(∑ 𝑄𝑜𝑏𝑠 )∗(∑ 𝑄𝑠𝑖𝑚 )
𝑅2 = 1⁄ (4)
2
√[𝑛(∑ 𝑄𝑜𝑏𝑠 2 )−(∑ 𝑄𝑜𝑏𝑠 )2 ]𝑋[𝑛(∑ 𝑄𝑠𝑖𝑚 2 )−(∑ 𝑄𝑠𝑖𝑚 )2 ]

Where Qobs= discharge observed, Qsim= discharge simulated, and n = number of observations.
The range of R2 lies between 0 and 1 which represents no correlation and perfect correlation between observed and
simulated value.
 Nash-Sutcliffe efficiency (N):
𝑡 𝑡
∑𝑇
𝑡=1(𝑄𝑜𝑏𝑠 −𝑄𝑠𝑖𝑚 )
2
𝑁 = 1− ∑𝑇 𝑡 𝑡 2 (5)
𝑡=1(𝑄𝑜𝑏𝑠 −𝑄𝑚𝑜 )

𝑡
Where, 𝑄𝑚𝑜 = average of observed discharges

 Root mean square error (RMSE):

∑𝑁 𝑡 𝑡 2
𝑛=1(𝑄𝑜 −𝑄𝑠𝑖𝑚 )
𝑅𝑀𝑆𝐸 = √ (6)
𝑁

 Relative peak error (RPE):

To predict peak flow with accuracy in addition to the goodness of fit RPE has been utilized in this study. It signifies
the accuracy of the model to predict peak flows accurately. It is expressed as:
|𝑄𝑝 −𝑄𝑝𝑚 |
𝑅𝑃𝐸 = (7)
𝑄𝑝

Where, Qp = observed peak, and Qpm = simulated peak discharge.

4. Results and Discussions

In this research study, time series data of monthly precipitation from two stations (Inderpuri and Buxar) and discharge
(Koelwar) from one station were employed for the model development. The foremost step in the process of model
development is the selection of significant input variables for the model. In order to select the significant input variables
for the model, statistical analysis consisting of Cross Correlation Functions (CCF) and partial Auto Correlation
Functions (ACF) were used.
The CCF curves between discharge at Koelwar station and precipitation at different lags for Inderpuri and Buxar
stations are presented in Figure 5 (a and b) and the ACF curve for Koelwar station at different lags is presented in Figure
5(c). As can be seen from CCF curves Figure 5 (a and b), the correlation coefficient increases with increase in lag time
and reaches maximum at (t-1), then decreases continuously for both Inderpuri and Buxar stations. Therefore, P(t) and
P(t-1) are selected as dominant input variables from Inderpuri and Buxar precipitation stations. It is evident from the
Figure 5(c), that correlation coefficient decreases continuously with the increase in lag time. Thus, Q(t) and Q(t-1) are
selected as dominant input variables from Koelwar discharge gauging stations. Based on the results of CCF and ACF
curves three sets of input variables were chosen (Table 2).

2125
Civil Engineering Journal Vol. 5, No. 10, October, 2019

(a)
0.7 (b) 0.7
0.6 Inderpuri 0.6 Buxar
0.5 0.5
Correlation coefficient

Correlation coefficient
0.4 0.4
0.3 0.3
0.2 0.2
0.1 0.1
0 0

t-5
t-0
t-1
t-2
t-3
t-4

t-6
t-7
t-8
t-9
t-10
t-11
t-12
t-3
t-0
t-1
t-2

t-4
t-5
t-6
t-7
t-8
t-9
t-10
t-11
t-12
-0.1 -0.1
-0.2 -0.2
-0.3 -0.3
Precipitation lag time Precipitation lag time

0.6
(c) Koelwar
0.5
Correlation coefficient

0.4
0.3
0.2
0.1
0
t-1 t-2 t-3 t-4 t-5 t-6
-0.1
-0.2
Discharge lag time

Figure 5. Cross correlation curves of (a) Inderpuri; (b) Buxar; (c) Koelwar

Table 2. Inputs selected for model generation

Set 1 P obs Inder Pt-1 obs Inder P obs Bux Pt-1 obs Bux
Set 2 Qt-1 P obs Inder P obs Bux
Set 3 Qt-1 Pt-1 obs Inder Pt-1 obs Bux Pt-12 obs Inder Pt-12 obs Bux

For lowering the dimensionality of the input variables, there is a requirement of a practical and efficient method
which is capable of changing the correlated discharge-affecting factors into uncorrelated ones. PCA of the selected sets
has then been carried out. With the PCA the eigen-vector based multivariate analysis has been done. It helps in
establishing sets of variables which exhibits the observed principal tendencies. These new linear combinations are called
principal components.
Monthly observed data from 2 rainfall and 1 discharge stations for the period January 1986 to December 2014 of the
Sone command have been utilised for the model development. Two models (ANN and ENN) have been used to check
the forecasting abilities. Using the top three combinations of data 3 pairs of sets has been developed for rainfall-runoff
modelling. The foremost task in rainfall-runoff (r-r) modelling using ANN and ENN approaches is the identification of
optimal network geometry. It should be noted that, besides the selection of significant input variables for the model, the
optimal adjusting of the network parameters like training iteration epoch, the number of hidden neurons and transfer
functions of layers also plays an important role. In this study, the training iteration epoch and the number of hidden
neurons were selected based on the trial and error method for each set of input variables in both the models. Both the
models were trained on Levenberg–Marquardt Back-propagation algorithm using tangent sigmoid as activation function
in ANN model. Based on the study of Sharghi et al. [29] the models were checked for hidden neurons up to the fourfold
of the input number.
The comparisons of the efficiency of models have been done using different indices and the results are shown in
table. In the table R, R2, Nash Sutcliffe, RMSE and Relative Peak Error of the two models are shown (Table 3). The
summary of the models have been shown using table. Both the models are showing good results but ENN is showing
better results and outperforms ANN for the river runoff forecasting. The performance of the datasets are indicated in
Table 3.

2126
Civil Engineering Journal Vol. 5, No. 10, October, 2019

Table 3. Performance indices of models

ANN R R2 NSE RPE RMSE

Set 1 0.7685 0.590592 0.5885 0.431 489.322
Set 2 0.9149 0.837072 0.0197 0.01974 309.8577
Set 3 0.8072 0.651572 0.6476 0.2766 452.8103
2
ENN R R NSE RPE RMSE
Set 1 0.850884 0.724003 0.7173 0.1528 405.558
Set 2 0.933041 0.870565 0.8689 0.00879 276.1359
Set 3 0.923018 0.851962 0.8306 0.0998 313.9636

From the table and figure it can be easily deduced that the best performing sets for rainfall-runoff modelling is Set 2
for both ANN and ENN. Therefore, for further analysis the results obtained from set 2 for both the models has been
utilised.
The observed versus simulated discharge time series using ENN and ANN for Sone Command has been shown in
Figure 6. The observed runoff values are in good accord with the simulated values but it can be seen that simulation by
ANN is slightly deviated from the observed discharge. This can be due to the fact that ENN encompasses neural
parameters which enhance the network learning process.

8000
Observed
7000
ENN simulated
6000
5000 ANN simulated
Runoff

4000
3000
2000
1000
0
10/09/1985

06/06/1988

03/03/1991

27/11/1993

23/08/1996

20/05/1999

13/02/2002

09/11/2004

06/08/2007

02/05/2010

26/01/2013
Time

Figure 6. Observed versus simulated discharge time series using ENN and ANN for Sone Command
The scatter plot of the simulated discharge using ENN and ANN has been shown in Figure 7. The R2 value using ENN
is highly significant with value 0.8706 as compared to ANN which shows lesser value of 0.8371.

ENN ANN Linear (ENN) Linear (ANN)

8000

7000

6000
y = 0.8005x + 76.439
R² = 0.8371
Runoff

5000

4000

3000

2000
y = 0.8438x + 93.539
1000 R² = 0.8706
0
0 1000 2000 3000 4000 5000 6000 7000 8000
Runoff

Figure 7. Verification scatter plot of simulated runoff using ANN and ENN

2127
Civil Engineering Journal Vol. 5, No. 10, October, 2019

Figure 8 details the Taylor diagram for the two models ANN and ENN for the period 1986-2014. The Taylor diagram
shows the closeness of the observed and model output. Three statistics namely correlation, root-mean-square error and
standard deviation are used to quantify the similarity between observed and model stimulated output. RMSE values are
indicated by the brown contours. This figure clearly implies that ENN is showing minimum standard deviation,
minimum RMSE and maximum correlation.

Figure 8. Taylor diagram of the two developed models

Figure 9. Random Walk Test

Figure 9 shows the results of the random walk test of the simulated discharge using ENN and ANN at the discharge
station for three data sets for the period 1986-2014. The basis of random walk test is sign test which is free of
distributional assumptions associated with the errors related to simulation. The results associated with this site are
contrasted in each and every step. The upward movement of all the 3 sets shows that ENN model is more skilful approach
for discharge simulation. The 95% confidence interval suggests that both the approaches are equally skilful in discharge
simulation.

5. Conclusion
The impacts of flood can be considerably reduced if the relationship between rainfall and runoff can be properly
established. This study has been done to test the applicability of soft computing technique based rainfall-runoff models
namely ENN and ANN to simulate runoff in the Sone Command, Bihar. Runoff and antecedent runoff, precipitation,
antecedent precipitation over the basin, at three gauging stations in the basin were first identified as appropriate input
variables, and then CCF curves at differ time lags have been plotted to select the potential input variables. Monthly
rainfall data of two stations and discharge data of one station for the period 1986-2014 have been utilised as data sets
for the development of ENN and ANN models. Three datasets have been selected based on cross correlation as potential
inputs for rainfall-runoff modelling. PCA of the selected scaled input has then been carried out to reduce the
dimensionality.
Many checks have been done to estimate the reliability and performance of the models. Based on the statistical
indices it has been established that ENN outperformed ANN and is more accurate as compared to the traditional ANN

2128
Civil Engineering Journal Vol. 5, No. 10, October, 2019

method for rainfall-runoff modelling. Also, other graphical indicators like Taylor diagram, Random walk test and one
to one correlation signifies the outperformance of ENN over ANN in the Sone River Command, Bihar. The results of
this study will be helpful in selecting the appropriate model for the discharge simulation in the Sone command, Bihar
and thereby helping planners for effective flood mitigation.

6. Conflicts of Interest
The authors declare no conflict of interest.

7. References
[1] Chang, Tak Kwin, Amin Talei, Sina Alaghmand, and Melanie Po-Leen Ooi. “Choice of Rainfall Inputs for Event-Based Rainfall-
Runoff Modeling in a Catchment with Multiple Rainfall Stations Using Data-Driven Techniques.” Journal of Hydrology 545
(February 2017): 100–108. doi:10.1016/[Link].2016.12.024.

[2] Jain, Ashu, and Sanaga Srinivasulu. “Development of Effective and Efficient Rainfall-Runoff Models Using Integration of
Deterministic, Real-Coded Genetic Algorithms and Artificial Neural Network Techniques.” Water Resources Research 40, no. 4
(April 2004). doi:10.1029/2003wr002355.

[3] Osman, Yassin Z., and Mawada E. Abdellatif. “Improving Accuracy of Downscaling Rainfall by Combining Predictions of
Different Statistical Downscale Models.” Water Science 30, no. 2 (October 2016): 61–75. doi:10.1016/[Link].2016.10.002.

[4] Shoaib, Muhammad, Asaad Y. Shamseldin, Sher Khan, Mudasser Muneer Khan, Zahid Mahmood Khan, and Bruce W. Melville.
“A Wavelet Based Approach for Combining the Outputs of Different Rainfall–runoff Models.” Stochastic Environmental
Research and Risk Assessment 32, no. 1 (November 29, 2016): 155–168. doi:10.1007/s00477-016-1364-x.

[5] Chang, Tak Kwin, Amin Talei, Chai Quek, and Valentijn R.N. Pauwels. “Rainfall-Runoff Modelling Using a Self-Reliant Fuzzy
Inference Network with Flexible Structure.” Journal of Hydrology 564 (September 2018): 1179–1193.
doi:10.1016/[Link].2018.07.074.

[6] Nourani, Vahid. “An Emotional ANN (EANN) Approach to Modeling Rainfall-Runoff Process.” Journal of Hydrology 544
(January 2017): 267–277. doi:10.1016/[Link].2016.11.033.

[7] Rezaie-balf, Mohammad, Sujay Raghavendra Naganna, Alireza Ghaemi, and Paresh Chandra Deka. “Wavelet Coupled MARS
and M5 Model Tree Approaches for Groundwater Level Forecasting.” Journal of Hydrology 553 (October 2017): 356–373.
doi:10.1016/[Link].2017.08.006.

[8] Bartoletti, N., F. Casagli, S. Marsili-Libelli, A. Nardi, and L. Palandri. “Data-Driven Rainfall/runoff Modelling Based on a Neuro-
Fuzzy Inference System.” Environmental Modelling & Software 106 (August 2018): 35–47. doi:10.1016/[Link].2017.11.026.

[9] Chandwani, Vinay, Sunil Kumar Vyas, Vinay Agrawal, and Gunwant Sharma. “Soft Computing Approach for Rainfall-Runoff
Modelling: A Review.” Aquatic Procedia 4 (2015): 1054–1061. doi:10.1016/[Link].2015.02.133.

[10] Sharma, S.K., and K.N. Tiwari. “Bootstrap Based Artificial Neural Network (BANN) Analysis for Hierarchical Prediction of
Monthly Runoff in Upper Damodar Valley Catchment.” Journal of Hydrology 374, no. 3–4 (August 2009): 209–222.
doi:10.1016/[Link].2009.06.003.

[11] Ghumman, A.R., Yousry M. Ghazaw, A.R. Sohail, and K. Watanabe. “Runoff Forecasting by Artificial Neural Network and
Conventional Model.” Alexandria Engineering Journal 50, no. 4 (December 2011): 345–350. doi:10.1016/[Link].2012.01.005.
[12] Hlavčová, Kamila, Zuzana Štefunková, Peter Valent, Silvia Kohnová, Roman Výleta, and Ján Szolgay. “Modelling the Climate
Change Impact On Monthly Runoff in Central Slovakia.” Procedia Engineering 161 (2016): 2127–2132.
doi:10.1016/[Link].2016.08.804.

[13] Kabiri, R., V. Kanani, and C. Andrew. "Climate Change Impacts on River Runoff in Klang Watershed in West Malaysia." J.
Clim. Res 48 (2012): 57-71.

[14] Lee, Kwan Tun, Jui-Yi Ho, Hong-Ming Kao, Gwo-Fong Lin, and Tsun-Hua Yang. “Using Ensemble Precipitation Forecasts and
a Rainfall-Runoff Model for Hourly Reservoir Inflow Forecasting During Typhoon Periods.” Journal of Hydro-Environment
Research 22 (January 2019): 29–37. doi:10.1016/[Link].2018.05.002.

[15] Wu, C.L., and K.W. Chau. “Rainfall–runoff Modeling Using Artificial Neural Network Coupled with Singular Spectrum
Analysis.” Journal of Hydrology 399, no. 3–4 (March 2011): 394–409. doi:10.1016/[Link].2011.01.017.

[16] Yaduvanshi, Aradhana, Rajat K. Sharma, Sarat C. Kar, and Anand K. Sinha. “Rainfall–runoff Simulations of Extreme Monsoon
Rainfall Events in a Tropical River Basin of India.” Natural Hazards 90, no. 2 (October 31, 2017): 843–861. doi:10.1007/s11069-
017-3075-0.

2129
Civil Engineering Journal Vol. 5, No. 10, October, 2019

[17] Demirel, Mehmet C., Anabela Venancio, and Ercan Kahya. “Flow Forecast by SWAT Model and ANN in Pracana Basin,
Portugal.” Advances in Engineering Software 40, no. 7 (July 2009): 467–473. doi:10.1016/[Link].2008.08.002.

[18] Maier, Holger R., and Graeme C. Dandy. “Neural Networks for the Prediction and Forecasting of Water Resources Variables: a
Review of Modelling Issues and Applications.” Environmental Modelling & Software 15, no. 1 (January 2000): 101–124.
doi:10.1016/s1364-8152(99)00007-9.

[19] Tiwari, Mukesh K., and Chandranath Chatterjee. “A New wavelet–bootstrap–ANN Hybrid Model for Daily Discharge
Forecasting.” Journal of Hydroinformatics 13, no. 3 (July 2011): 500–519. doi:10.2166/hydro.2010.142.

[20] Adamowski, Jan, and Karen Sun. “Development of a Coupled Wavelet Transform and Neural Network Method for Flow
Forecasting of Non-Perennial Rivers in Semi-Arid Watersheds.” Journal of Hydrology 390, no. 1–2 (August 2010): 85–91.
doi:10.1016/[Link].2010.06.033.

[21] Solomatine, Dimitri P., and Khada N. Dulal. “Model Trees as an Alternative to Neural Networks in Rainfall—runoff Modelling.”
Hydrological Sciences Journal 48, no. 3 (June 2003): 399–411. doi:10.1623/hysj.48.3.399.45291.

[22] Wang, Wensheng, and Jing Ding. "Wavelet network model and its application to the prediction of hydrology." Nature and
Science 1, no. 1 (2003): 67-71.

[23] Nourani, Vahid, Mehdi Komasi, and Akira Mano. “A Multivariate ANN-Wavelet Approach for Rainfall–Runoff Modeling.”
Water Resources Management 23, no. 14 (February 3, 2009): 2877–2894. doi:10.1007/s11269-009-9414-5.

[24] Roshni, Thendiyath, Madan K. Jha, Ravinesh C. Deo, and A. Vandana. “Development and Evaluation of Hybrid Artificial Neural
Network Architectures for Modeling Spatio-Temporal Groundwater Fluctuations in a Complex Aquifer System.” Water
Resources Management 33, no. 7 (April 18, 2019): 2381–2397. doi:10.1007/s11269-019-02253-4.

[25] Seo, Youngmin, Sungwon Kim, Ozgur Kisi, and Vijay P. Singh. “Daily Water Level Forecasting Using Wavelet Decomposition
and Artificial Intelligence Techniques.” Journal of Hydrology 520 (January 2015): 224–243. doi:10.1016/[Link].2014.11.050.

[26] Du, Kongchang, Ying Zhao, and Jiaqiang Lei. “The Incorrect Usage of Singular Spectral Analysis and Discrete Wavelet
Transform in Hybrid Models to Predict Hydrological Time Series.” Journal of Hydrology 552 (September 2017): 44–51.
doi:10.1016/[Link].2017.06.019.

[27] Mallat, S.G. “A Theory for Multiresolution Signal Decomposition: The Wavelet Representation.” IEEE Transactions on Pattern
Analysis and Machine Intelligence 11, no. 7 (July 1989): 674–693. doi:10.1109/34.192463.

[28] Rezaie-Balf, Mohammad, Zahra Zahmatkesh, and Sungwon Kim. “Soft Computing Techniques for Rainfall-Runoff Simulation:
Local Non–Parametric Paradigm Vs. Model Classification Methods.” Water Resources Management 31, no. 12 (June 1, 2017):
3843–3865. doi:10.1007/s11269-017-1711-9.

[29] Sharghi, Elnaz, Vahid Nourani, Hessam Najafi, and Amir Molajou. “Emotional ANN (EANN) and Wavelet-ANN (WANN)
Approaches for Markovian and Seasonal Based Modeling of Rainfall-Runoff Process.” Water Resources Management 32, no.
10 (May 8, 2018): 3441–3456. doi:10.1007/s11269-018-2000-y.

[30] Fellous, Jean-Marc. “Neuromodulatory Basis of Emotion.” The Neuroscientist 5, no. 5 (September 1999): 283–294.
doi:10.1177/107385849900500514.

[31] Nourani, Vahid, Özgür Kisi, and Mehdi Komasi. “Two Hybrid Artificial Intelligence Approaches for Modeling Rainfall–runoff
Process.” Journal of Hydrology 402, no. 1–2 (May 2011): 41–59. doi:10.1016/[Link].2011.03.002.

[32] Sharghi, Elnaz, Vahid Nourani, Amir Molajou, and Hessam Najafi. “Conjunction of Emotional ANN (EANN) and Wavelet
Transform for Rainfall-Runoff Modeling.” Journal of Hydroinformatics 21, no. 1 (October 9, 2018): 136–152.
doi:10.2166/hydro.2018.054.

[33] Latinez Sotomayor, Karen A. "Comparison of adaptive methods using multivariate regression splines (MARS) and
artificial neural networks backpropagation (ANNB) for the forecast of rain and temperatures in the Mantaro river
basin." (March 2010).

2130