Computers and Electronics in Agriculture 115 (2015) 66–77

Contents lists available at ScienceDirect

Computers and Electronics in Agriculture

journal homepage: www.elsevier.com/locate/compag

Long-term monthly evapotranspiration modeling by several data-driven

methods without climatic data
Ozgur Kisi a,⇑, Hadi Sanikhani b, Mohammad Zounemat-Kermani c, Faegheh Niazi d
a
Canik Basari University, Civil Eng. Dept., Samsun, Turkey
b
Young Researchers and Elite Club, Saveh Branch, Islamic Azad University, Saveh, Iran
c
Shahid Bahonar University of Kerman, Department of Water Engineering, Kerman, Iran
d
University of Tabriz, Dept. of Water Eng., Tabriz, Iran

a r t i c l e i n f o a b s t r a c t

Article history: In this study, the ability of four different data-driven methods, multilayer perceptron artificial neural net-
Received 30 December 2014 works (ANN), adaptive neuro-fuzzy inference system (ANFIS) with grid partition (GP), ANFIS with sub-
Received in revised form 18 April 2015 tractive clustering (SC) and gene expression programming (GEP), was investigated in predicting
Accepted 29 April 2015
long-term monthly reference evapotranspiration (ET0) by using data from 50 stations in Iran. The period-
icity component, station latitude, longitude and altitude values were used as inputs to the applied models
to predict the long-term monthly ET0 values. The overall accuracies of the multilayer perceptron ANN,
Keywords:
ANFIS-GP and ANFIS-SC models were found to be similar to each other. The GEP model provided the
Neural networks
Adaptive neuro-fuzzy
worst estimates. The maximum determination coefficient (R2) values were found to be 0.997, 998 and
Genetic programming 0.994 for the ANN, ANFIS-GP and ANFIS-SC models in Karaj station, respectively. The highest R2 value
Geographical inputs (0.978) of GEP model was found for the Qom station. The minimum R2 values were respectively found
Reference evapotranspiration as 0.959 and 0.935 for the ANN and ANFIS-GP models in Bandar Abbas station while the ANFIS-SC and
GEP models gave the minimum R2 values of 0.937 and 0.677 in the Tabriz and Kerman stations, respec-
tively. The results indicated that the long-term monthly reference evapotranspiration of any site can be
successfully estimated by data-driven methods applied in this study without climatic measurements. The
interpolated maps of ET0 were also obtained by using the optimal ANFIS-GP model and evaluated in the
study. The ET0 maps showed that the highest amounts of reference evapotranspiration occurred in the
southern and especially southeastern parts of the Iran.
Ó 2015 Elsevier B.V. All rights reserved.

1. Introduction a difficult task because it is time-consuming and needs potential

Evapotranspiration (ET) is a process of losing water to the atmo-
sphere via the combined process of evaporation from the soil and
transpiration from the plants. In agriculture science especially in
arid and semiarid areas, proper prediction of the given crop evap-
otranspiration (ETc) is important. Therefore, the ability to obtain or
estimate evapotranspiration (ET) is of great importance for operat-
ing irrigation systems (Karimadldini et al., 2012). ETc can be calcu-
lated by multiplying ET0 (reference evapotranspiration) to the crop
coefficient (Kc) of the crop.
Several methods and techniques could be applied for calculat-
ing or estimating ET including (1) direct, (2) indirect and (3)
data-driven methods. Direct measurement method by using
lysimeter is the best options for obtaining ET crop and ET0 but is
⇑ Corresponding author. Tel.: +90 362 2801070.
E-mail address: okisi@basari.edu.tr (O. Kisi).
financial and environmental resources. As a result, a large number
of indirect methods such as empirical, semi-empirical equations
and data driven models have been developed for measuring ET0
from meteorological data (Karimadldini et al., 2012). Some of the
indirect ET0 estimation methods can be named as simple pan
evaporation-based methods and combination methods such as
(Penman, 1963), Penman–Monteith (Monteith, 1965) and etc.
One of the best known indirect method is the Food and
Agricultural Organization of the United Nations (FAO) method in
which combines the equation of Penman–Monteith modified by
Allen in 1998 (FAO-56 PM equation) as a reference equation for
ET0 estimation. The FAO-56 Penman–Monteith (PM-56) equation
is influenced by several weather parameters including measure-
ment of the maximum and minimum air temperature, maximum
and minimum relative air humidity, wind speed and solar radia-
tion. With an exception to the air temperature which is available
at most weather stations, the remaining variables are often

http://dx.doi.org/10.1016/j.compag.2015.04.015
0168-1699/Ó 2015 Elsevier B.V. All rights reserved.
 O. Kisi et al. / Computers and Electronics in Agriculture 115 (2015) 66–77 67

Fig. 1. The location of the stations in Iran.

Table 1
The geographical information of the stations used in the study.

No Station Long. (E) Lat. (N) Alt. (m) ET0 (mm/day) No Station Long. (E) Lat. (N) Alt. (m) ET0 (mm/day)
1 Tehran Merabad 51 19 35 41 1191 4.76 26 Ramsar 50 40 36 54 20 2.24
2 Sanandaj 47 0 36 20 1373 3.87 27 Babolsar 52 39 36 43 -21 2.53
3 Rasht 49 37 37 19 -8.6 2.32 28 Saqqez 46 21 36 15 1523 3.56
4 Ahwaz 48 40 31 20 22.5 6.20 29 Sabzevar 57 39 36 12 972 5.46
5 Zanjan 48 29 36 41 1663 4.04 30 Torbat-e Heydarieh 59 16 35 13 1451 4.15
6 Urmia 45 3 37 40 1328 3.10 31 Khoy 44 58 38 33 1103 2.88
7 Ardabil 48 17 38 15 1332 2.87 32 Kashan 51 27 33 59 982 3.39
8 Gorgan 54 24 36 54 0 2.84 33 Fasa 53 41 28 58 1288 4.69
9 Sari 53 0 36 33 23 2.78 34 Bandar Anzali 49 27 37 29 -23.6 2.41
10 Arak 49 55 34 6 1708 3.68 35 Bandar Lengeh 54 50 26 32 22.7 5.58
11 Kermanshah 47 9 34 21 1319 4.34 36 Jask 57 46 25 38 5.2 5.23
12 Ilam 46 26 33 38 1337 4.52 37 Bam 58 21 29 6 1067 6.47
13 Khorramabad 48 17 33 26 1148 4.11 38 Abadan 48 15 30 22 6.6 6.68
14 Shahrekord 50 51 32 17 2049 3.46 39 Dezful 48 23 32 24 143 5.00
15 Qazvin 50 3 36 15 1279 3.83 40 Zabol 61 29 31 2 489 8.42
16 Yasuj 51 33 30 41 1816 3.89 41 Mashhad 59 38 36 16 999 3.86
17 Bojnord 57 16 37 28 1112 3.59 42 Bandar Abbas 56 22 27 13 9.8 5.29
18 Birjand 59 12 32 52 1491 5.10 43 Semnan 53 25 35 35 1127 4.05
19 Zahedan 60 53 29 28 1370 5.84 44 Shiraz 52 36 29 32 1484 4.90
20 Yazd 54 17 31 54 1237 5.45 45 Kerman 56 58 30 15 1754 5.43
21 Bushehr 50 49 28 58 9 5.06 46 Tabriz 46 17 38 5 1361 4.11
22 Tabas 56 55 33 36 711 4.67 47 Hamedan 48 43 25 12 1679 3.93
23 Iranshahr 60 42 27 12 591 6.10 48 Qom 50 51 34 42 877 4.83
24 Chabahar 60 37 25 17 8 4.67 49 Isfahan 51 40 32 37 1550 4.42
25 Shahrud 54 57 36 25 1349 4.00 50 Karaj 50 54 35 55 1313 4.11

incomplete or not always available for many locations and even
not always reliable (Rahimikhoob, 2010). Regarding the above con-
texts, there are so many factors affecting ET estimation based on
indirect methods, thereafter it is extremely difficult to formulate
the ET0 equation that can produce reliable estimates.
The main problem with modeling ET process is its nonlinear
dynamic and high complexity, in this respect, using data-driven
methods (e.g. ANN, ANFIS, SVM ...) based on soft computing tech-
niques could be considered as proper methods for estimating
ET0. These methods are known for their ability in dealing with
68 O. Kisi et al. / Computers and Electronics in Agriculture 115 (2015) 66–77

Gene-expression programming (GEP) interface is a powerful

technique applied in many engineering researches with good gen-
eralization ability (Azamathulla et al., 2011; Shiri et al., 2014a,b,c;
Marti et al., 2015; Kisi and Sanikhani, 2015). It has been observed
that related studies in the literature related to the use for ET0 mod-
eling are scare. Guven et al. (2008) presented genetic programming
for the estimation of ET0 by using daily atmospheric variables
obtained from the California Irrigation Management Information
System database. The results of the GEP model were compared to
seven conventional models and were found to be in good agree-
ment. Results showed that the GEP model gave a fast and practical
method for the estimation of ET0 to obtain accurate results and
Fig. 2. General view of an ANFIS structure. encourage the use of GEP in other aspects of water engineering
studies. In another study, Traore and Guven (2012) investigated
complex problems, with only sets of available data. In the follow- the performance ability of the GEP for modeling ET0 using decadal
ing, the application of soft computing techniques to evapotranspi- climatic data in Burkina Faso. They claimed that GEP is an effectual
ration modeling is presented. modeling tool for successfully computing evapotranspiration. To
Among different soft computing methods, artificial neural net- the knowledge of the writers, there is not any published paper in
works (ANNs) due to their high ability in the modeling of nonlinear the literature that compares the ability of ANN, ANFIS with grid
processes such as ET0 are very appropriate tools. In the past dec- partition (ANFIS-GP), ANFIS with subtractive clustering
ades, ANNs were used to estimate evapotranspiration (Kumar (ANFIS-SC) and GEP in predicting long-term monthly ET0.
et al., 2002; Sudheer et al., 2003; Trajkovic, 2005; Kim and Kim, The aim of this study is to investigate the accuracy of multilayer
2008; Landeras et al., 2008; Traore et al., 2010). Rahimikhobb perceptron ANN, ANFIS-GP, ANFIS-SC and GEP in predicting
(2010) examined the potential for the use of ANNs to estimate long-term monthly ET0 by using data from 50 stations in Iran.
ET0 based on air temperature data under humid subtropical condi-
tions on the southern coast of the Caspian Sea. He claimed that the
ANN model was superior to indirect empirical equations. A compre- 2. Case study
hensive and recent review of ANN applications in ET0 modeling can
be refereed to Kumar et al. (2011). They showed that the perfor- The study area used in this study lays over the entire region of
mance of most of the ANN models depends on a few meteorological Iran’s geographical area (25°000 N and 38°390 N latitudes and
stations which have recorded weather data for a specific period. between 44°000 E and 63°250 E longitudes) of about 1650,000 km2.
In addition to ANNs, more recently adaptive neuro-fuzzy infer- According to Dinpashoh et al. (2011) generally, Iran is categorized
ence systems (ANFIS) was used to simulate/predict ET0 (Kisi and as having arid (BW) and semi-arid (BS) climates based on the
Ozturk, 2007; Aytek, 2009; Dogan, 2009; Cobaner, 2011; Kisi and Koppen climatic classification (Ahrens, 1998). Precipitation is the
Zounemat-Kermani, 2014). ANFIS provides an efficient way of main source of water in Iran, which normally amounts to
modeling the uncertainty for complicated systems lacking ade- 251 mm or 413 billion cubic meters annually. More than 70% of
quate data. Kisi and Ozturk (2007) investigated the capability of the annual precipitation is lost from evapotranspiration (Kousari
the ANFIS in the estimation of ET0 in Los Angeles. The results
showed that the ANFIS models could be considered as proper tools
in modeling ET0 process. Aytek (2009) applied a coactive
neuro-fuzzy inference system (CANFIS) technique for modeling
daily ET0, in three stations and compared the results with some
empirical formulas. They claimed that CANFIS can be employed
as an alternative ET0 model to the existing conventional methods.
Cobaner (2011) investigated the potential of grid partition-based
fuzzy inference system (G-ANFIS) and subtractive
clustering-based fuzzy inference system (S-ANFIS) and ANNs, in
modeling of ET0. They found that the S-ANFIS model achieved more
accurate results with fewer amounts of iterations as compared to
the G-ANFIS and ANN models in modeling ET0 process.
Karimaldini et al. (2012) investigated the potential of the adap-
tive neuro-fuzzy computing technique (ANFIS) for daily ET0 model-
ing under arid conditions from limited weather data. The
comparison results indicated that by using similar meteorological
inputs, the ANFIS models performed better than conventional
methods. Tabari et al. (2013) investigated the accuracy of the
ANFIS and support vector machines (SVM) for potato ETc estima-
tion. The results confirmed that the SVM models could provide
more accurate ETc estimates than the ANFIS and empirical equa-
tions. Kisi and Zounemat-Kermani (2014), applied two different
types of ANFIS (with grid partition method and ANFIS with sub-
tractive clustering method) in modeling daily ET0. Results indi-
cated that the three-and four-input ANFIS models were superior
to the corresponding empirical equations in modeling ET0 while
the calibrated two-parameter Ritchie and Valiantzas’ equations
were found to be better than the two-input ANFIS models. Fig. 3. Schematic architecture of a feed forward neural network.
 O. Kisi et al. / Computers and Electronics in Agriculture 115 (2015) 66–77 69

and Ahani, 2012). Iran is a vulnerable country in regard to climate
changes and global warming because it has low amounts of annual
precipitation; highly spatial and temporal variation in precipita-
tion in contrast to the tremendous potential of evapotranspiration
(Abbaspour et al., 2009).
The data of 50 stations were compiled from the Iran
Meteorological Organization (Tehran). The location of the stations
in Iran is represented in Fig. 1. The geographical positions of these
stations are given in Table 1. Data cover the long-term averaged
evapotranspirations between the periods of 1956–2010. The
long-term monthly ET0 ranges from as low as 2.24 mm/day in
Ramsar station to as high as 8.42 mm/day in Zabol station.
3. Methods
3.1. ANFIS
Here, ðpi ; qi ; r i Þ is the linear combination of all inputs and needs
to be adjusted as consequent parameters.
– Layer VI (summation): in this layer the summation of all
difuzzified values produce the output of the ANFIS model.
X
OutputVI ¼ W if i ð6Þ
In Takagi–Sugeno ANFIS, a hybrid-learning algorithm is applied
for training the network consisting of the combination of gradient
descent (for identifying the premise parameters) and least-squares
methods (to assign the linear consequent parameters). The learn-
ing algorithm tunes the premise (MF parameters) and consequence
parameters (the response of the network), according to the training
data. In the forward part of hybrid learning algorithm, the conse-
quent parameters are identified by the least squares method until
layer IV. In the backward part, the premise parameters are updated

Neuro-fuzzy systems which are based on the rule-based fuzzy

Create chromosomes of inial populaon
systems uses the capability of neural networks learning algorithm
for adapting their rule-base parameters (Zounemat-Kermani and
Scholz, 2013). Generally, in ANFIS a six layers multilayer neural Express chromosomes
network-based fuzzy system is constructed. In this system archi-
tecture, the input and output nodes represent the input states
(variables) and output response (target), respectively, while in Execute each program
the hidden layers membership functions (MF) and rules are placed
as node functioning (Jang et al., 1997; Zounemat-Kermani and
Evaluate fitness
Teshnehlab, 2008).
Considering two crisp input parameters (X1, X2) and member-
ship functions (A1, A2 and B1, B2) for each input, ANFIS structure
Iterate or Terminate
(see Fig. 2) can be briefly summarized in the following six layers: End
terminate?
– Layer I (Input): all input variables at this layer are transferred to
the proceeding layers.
Output ðIÞ;X1 ¼ X1; Output ðIÞ;X2 ¼ X2 ð1Þ Best of Yes
generaon
– Layer II (fuzzification): the output of each node (i) calculates the
membership grades according to the membership degrees ðlÞ
which are dependent on input values and on the used member- Selecon
ship function. The membership degrees are symbolized as
Output ðIIÞ;i ¼ lAi ðX1Þ; i ¼ 1; 2; OutputðIIÞ;i ¼ lBi ðX2Þ; i ¼ 1; 2 Replicaon
ð2Þ
In this study, several types of MFs such as triangular, Gaussian Mutaon
and trapezoidal are used.
– Layer III (rules): according to the Takagi–Sugeno fuzzy logic
Inversion
rules, by multiplying input signals from the respective fuzzifica-
tion nodes, the firing strength of each fuzzy rule is calculated
and sending the product out as summarized below IS Transposion
Reproducon

Output III;i ¼ W n ¼ lAi ðX1ÞlBi ðX2Þ; i ¼ 1; 2 ð3Þ

RIS Transposion
where Wn is the firing strength of the nth rule.
– Layer IV (normalization): all the output values from the third Gene Transposion
layer are taken as input values by normalization layer which
computes the normalized value of each rule by calculating the
One-point recombinaon
ratio of the ith rule’s firing strength to the total rule’s firing
strength.
Two-point recombinaon
Output IV;i ¼ W n ð4Þ

Layer V (deffuzification): the adjustable nodes of this layer cal- Gene recombinaon
culate the weighted output of the rth fuzzy rule by multiplica-
tion of the normalized firing strength (Output IV) and
consequent function (pi + qi + ri). New chromosomes of next generaon

Output V;i ¼ W i f i ¼ W i ðpi þ qi þ r i Þ ð5Þ Fig. 4. The flowchart of the gene expression algorithm (Ferreira, 2006).
70 O. Kisi et al. / Computers and Electronics in Agriculture 115 (2015) 66–77

by gradient descent using the back-propagation algorithm (Jang ANFIS-SC model with a radii value of 0.90. Similar to the
et al., 1997). More information about ANFIS can be found in Jang ANFIS-GP models, training of the ANFIS-SC models were stopped
(1993). after 100 epochs.
Generally, two ANFIS types can be established by using GP and
SC methods according to the input vector and output variables. In 3.2. ANN
GP method, the input and output space are partitioned into fuzzy
grids. Using these fuzzy grids the location of MFs can be optimized. Artificial neural networks are biologically inspired computa-
In SC method, each input parameter in the first layer is clustered tional model constructed of many simple interconnected elements
into several class values to build up fuzzy rules. More details about called neurons (processing elements) connected with coefficients
GP and SC can be found in (Kisi and Zounemat-Kermani, 2014). In
the present study, different membership functions, triangular
(trimf), Gaussian (gaussmf), trapezoidal (trapmf), two Gaussian
(gauss2mf), pi-shaped (pimf) and generalized bell (gbellmf) and
100 iterations were used for the ANFIS-GP models. Different num-
bers of membership functions were tested and the best one that
gave the minimum RMSE (root mean square errors) was selected
for the ANFIS-GP models. For the ANFIS-SC model, different radii
values were tried and optimal results were obtained from an

Table 2
Test results of ANN training algorithms.

Training algorithm* Model Iteration RMSE R2

structure number (mm/day)
Levenberg–Marquardt 4-2-1 1,000 0.772 0.903
backpropagation (lm)
Gradient descent back- 4-9-1 50,000 0.835 0.893
propagation (gd)
Gradient descentwith adaptive 4-4-1 50,000 0.744 0.913
learning rate
backpropagation (gda)
Gradient descent with 4-3-2 50,000 0.753 0.909
momentum backpropagation
(gdm)
Gradient descent with 4-4-1 50,000 0.739 0.913
momentum and adaptive
learning rate back-
propagation (gdx)
Scaled conjugate gradient back- 4-3-1 1,000 0.768 0.906
propagation (scg)
*
It is notable for all of training algorithms, the tangent sigmoid transfer function
was used in the hidden layers, and purelin transfer function in the output layer.

Table 3
Test results of different ANFIS-GP structures.

Type MFs Model structure RMSE (mm/day) R2

Triangular (trimf) 4-2-2-2 0.723 0.917
Gaussian (gaussmf) 3-2-3-3 0.757 0.915
Trapezoidal (trapmf) 3-2-3-2 0.697 0.928
Two Gaussian (gauss2mf) 4-2-3-2 0.845 0.888
Pi-shaped (pimf) 3-2-3-2 0.652 0.938
Generalized bell (gbellmf) 3-2-3-2 0.852 0.888

Table 4
Test results of different GEP basic and linking functions for the parse tree.

Fig. 5. Expression trees of the optimal GEP model.

 O. Kisi et al. / Computers and Electronics in Agriculture 115 (2015) 66–77 71

(weights and biases) which constitute the neural structure. ANNs
are capable of recognizing underlying relationships between input
and output procedures.
Several types of ANNs have been designed and reported
(Zounemat-Kermani et al., 2013). One of the most comprehensive
types of ANNs is multilayer perceptron models which have been
proven to be a flexible and universal approximator (Fig. 3). In mul-
tilayer perceptron ANN networks the flow of computation is from
the input layer to the output layer. The arriving variables (inputs),
multiplied by the connection weights and added by the biases are
first combined (summation). Each neuron has an adjustable weight
factor (w) and bias (b). Thereafter, the adjusted parameters passed
through an activation (transfer) function to make the output for
that neuron. The outputs produced from the first hidden layer
become the inputs to the second layer and so on to the output
layer.
In supervised learning of ANNs, there are several applicable
learning algorithms but the most often used is back-propagation
(BP) rule. In BP, optimization of the weights and biases is made
by backward propagation of the error during training phase. The
ANN compares the input and output values in the training data
set and alters the values of the processing units to reduce the dif-
ference between the predicted and target values. The difference
between predicted and target values makes the error vector, which
is used to adjust the network weights and biases.
The error performance measure is generally mean-square error
(MSE), which can be calculated as:
1 XN
E¼ e2
2N i¼1 i
where N is total number of training patterns and ei ¼ yi  t i equals
error for the ith training pattern. Y is the network predicted output
and t is the observed target.
Considering BP, there are several algorithms which can be used
for training ANNs such as gradient descent and Levenberg–
Marquadt. Gradient descent which is one of the most popular
training algorithms works by measuring the output error, calculat-
ing the gradient of this error, and then adjusting the weights and
biases in the descending gradient direction. This algorithm consists
of different versions such as: (1) simple gradient descent BP, (2)
gradient descent with adaptive learning rate, (3) gradient descent
with momentum, (4) gradient descent with momentum and adap-
tive learning rate, and (5) scaled conjugate gradient (Mukherjee
and Routroy, 2012).
The LM algorithm approximates to the Gauss–Newton method
which approximates the error of the network with a second order
expression. Hence, LM can be considered a combination of steepest
descent and the Gauss–Newton method. It is proved that LM out-
performs simple gradient descent and other conjugate gradient
methods in a wide variety of problems (Zounemat-Kermani et al.,
2013). LM algorithm uses an approximation to the Hessian matrix
in the following Gauss–Newton weight update:
wkþ1 ¼ wk  ½H þ kI1 J T e ð8Þ
where w is the weight of neural network, H is the Hessian matrix
evaluated at wk. The Jacobian matrix (J) can be computed through
a standard back-propagation technique which is much less complex
than computing the Hessian matrix. Thereafter, the Hessian matrix
can be approximated as H ¼ JT J. e implies the vector of network
errors. k is a scalar which can adjusted thorough the training proce-
dure. When the scalar k is zero, this is just Gauss–Newton’s method,
using the approximated Hessian matrix. However considering large
values for k, the algorithm becomes gradient descent with a small
step size. It should be mentioned that Newton’s method is faster
and more accurate therefore the goal is to shift toward Newton’s
ð7Þ method as fast as possible. However, the main shortcoming of LM
algorithm is its increased memory requirement to calculate the
Jacobian matrix of the error function. In the present paper, multi-
layer perceptron ANN models were used. For the ANN models, dif-
ferent training algorithms, Levenberg–Marquardt (lm), gradient
descent (gd), gradient descent with adaptive learning rate (gda),
gradient descent with momentum (gdm), gradient descent with
momentum and adaptive learning rate (gdx) and scaled conjugate
gradient (scg), were used for calculating the weights and results
were compared.

Table 5 3.3. GEP

The training and test results of the optimal ANN, ANFIS-GP, ANFIS-SC and GEP models.

Model Training period Test period

GEP was invented by Ferreira in 1999 (Ferreira, 2001), and
incorporates both the simple, linear chromosomes of fixed length
RMSE R2 RMSE R2
similar to the ones used in genetic algorithms and the ramified
ANN 0.872 0.895 0.739 0.913 structures in different sizes and shapes similar to the parse trees
ANFIS-GP 0.600 0.950 0.652 0.938
of GP. Because the ramified structures of different sizes and shapes
ANFIS-SC 0.641 0.943 0.747 0.912
GEP 1.306 0.766 0.690 0.923 are encoded in the linear chromosomes of fixed length, this is
equivalent to say that, in GEP, the genotype and phenotype are
finally separated from one another and the system can now benefit

5
Table 6
ANN ANFIS-GP ANFIS-SC GEP Summary of the test process of ANFIS and GEP models.
4
Stations ANN ANFIS-GP ANFIS-SC GEP
RMSE, (mm/day)

RMSE R2 RMSE R2 RMSE R2 RMSE R2

3
Mashhad 1.204 0.987 0.581 0.992 0.624 0.991 0.842 0.966
Bandar 0.497 0.959 0.470 0.935 1.290 0.988 0.844 0.913
2 abbas
Semnan 0.453 0.991 0.337 0.991 0.296 0.992 0.506 0.971
1 Shiraz 0.378 0.994 0.853 0.991 0.261 0.993 0.420 0.959
Kerman 0.582 0.994 0.269 0.994 0.872 0.993 0.930 0.677
Tabriz 1.474 0.984 1.090 0.969 1.270 0.937 0.930 0.924
0 Hamedan 0.723 0.990 0.382 0.988 0.632 0.978 0.643 0.939
0 5 10 15 20 25 30 35 40
Qom 0.418 0.991 1.010 0.992 0.524 0.994 0.854 0.978
Station
Isfahan 0.405 0.976 0.522 0.979 0.426 0.975 0.459 0.959
Karaj 0.238 0.997 0.365 0.998 0.289 0.994 0.427 0.967
Fig. 6. Training accuracy of the data-driven models with respect to RMSE.
72 O. Kisi et al. / Computers and Electronics in Agriculture 115 (2015) 66–77

from all the evolutionary advantages this brings about (Ferreira, In this research, different functions were utilized, including basic
2006). arithmetic operators ðþ; ; ; Þ as well as some of the other basic
GEP is an example of a full-fledged replicator/phenotype system mathematical functions as blow:
where the chromosomes/expression trees form a truly functional,
indivisible whole (Ferreira, 2001). The fundamental steps of the  pffiffiffi pffiffiffi 
ln x; e x ; x2 ; x3 ; x; 3 x; sin x; cos x; arctanðxÞ
gene expression algorithm (GEA) are schematically showed in
Fig. 4. The third major step is to choose the chromosomal architecture.
The procedure for ET0 prediction by GEP consists five major The fourth major step is to choose the kind of linking function. And
steps. The first step is to choose the fitness function. Root relative finally, the fifth major step in preparing to use GEP is to choose the
squared error (RRSE) is selected as the best fitness function in this set of genetic operators and their rates (Ferreira, 2006).
study. The second major step is to choose the set of terminals (T) Soft computing program called GeneXpro Tools 4.0 was used to
and the set of functions (F). In current work, the terminal set con- predict ET0 using GEP. The parameters used are summarized as fol-
sists of the geographical information of stations (latitude, longi- lows: Number of chromosomes: 30, head size: 8, number of genes:
tude and altitude) and the number corresponding each month. 3, mutation rate: 0.044, inversion rate: 0.1, one point recombina-
The set of functions is selected based on the nature of the problem. tion rate: 0.3, two point recombination rate: 0.3, gene

PM-FAO56 ANN
PM-FAO56 ANN
10 ANFIS-GP ANFIS-SC 10 ANFIS-GP ANFIS-SC
GEP GEP
8 8

ET0 (mm/day)
ET0 (mm/day)

6 6

4 4

2 2
(a) (b)
0 0
JAN

APR

JUN

AUG

NOV
FEB
MAR

JUL

OCT

DEC
MAY

SEP

FEB
MAR

OCT
JUL

SEP

DEC
JAN

APR
MAY
JUN

AUG

NOV
month month

Fig. 7. Penman–Monteith (FAO-56) versus predicted ET0 values in (a) Mashhad and (b) Bandar Abbas weather stations.

10 PM-FAO56 ANN PM-FAO56 ANN

ANFIS-GP ANFIS-SC 10 ANFIS-GP ANFIS-SC
GEP GEP
8 8
ET0 (mm/day)

ET0 (mm/day)

6 6

4 4

2 2
(a) (b)
0 0
DEC
FEB
MAR

JUL

OCT
MAY

SEP
JAN

APR

JUN

AUG

NOV
JAN

APR

JUN

AUG

NOV
MAR

OCT
FEB

MAY

JUL

SEP

DEC

month month

Fig. 8. Penman–Monteith (FAO-56) versus predicted ET0 values in (a) Semnan and (b) Shiraz weather stations.

PM-FAO56 ANN PM-FAO56 ANN

10 ANFIS-GP ANFIS-SC 10 ANFIS-GP ANFIS-SC
GEP GEP
8 8
ET0 (mm/day)
ET0 (mm/day)

6 6

4 4

2 2
(a) (b)
0 0
FEB
MAR

JUL

SEP
OCT

DEC
APR
JAN

JUN

AUG

NOV

JAN
FEB
MAR
APR

JUN
JUL
AUG

OCT
SEP

NOV
DEC
MAY

MAY

month month

Fig. 9. Penman–Monteith (FAO-56) versus predicted ET0 values in (a) Kerman and (b) Tabriz weather stations.
 O. Kisi et al. / Computers and Electronics in Agriculture 115 (2015) 66–77 73

10 PM-FAO56 ANN PM-FAO56 ANN

ANFIS-GP ANFIS-SC 10 ANFIS-GP ANFIS-SC
GEP GEP
8 8

ET0 (mm/day)

ET0 (mm/day)
6 6

4 4

2 2
(a) (b)
0 0

FEB
MAR

OCT

DEC
JUL

SEP
APR
JAN

JUN

AUG

NOV
APR
JAN

MAR
FEB

JUN
JUL
AUG
SEP
OCT
NOV
DEC

MAY
MAY
month month

Fig. 10. Penman–Monteith (FAO-56) versus predicted ET0 values in (a) Hamedan and (b) Qom weather stations.

PM-FAO56 ANN PM-FAO56 ANN

10 10
ANFIS-GP ANFIS-SC ANFIS-GP ANFIS-SC
GEP GEP
8 8

ET0 (mm/day)
ET0 (mm/day)

6 6

4 4

2 2
(a) (b)
0 0
FEB
MAR

JUL

OCT
SEP

DEC
JAN

APR
MAY
JUN

AUG

NOV
JAN

APR

JUN

AUG

NOV
FEB
MAR

JUL

OCT

DEC
MAY

SEP

month month

Fig. 11. Penman- Monteith (FAO-56) versus predicted ET0 values in (a) Isfahan and (b) Karaj weather stations.

Table 7 4. Application and results

Accuracy rank of the data-driven models in estimating long-term monthly ET0.

Stations ANN ANFIS-GP ANFIS-SC GEP Four different data-driven methods, ANN, ANFIS-GP, ANFIS-SC
Mashhad 4 1 2 3 and GEP, were applied for estimating long-term monthly mean
Bandar abbas 2 1 4 3 ET0. Latitude, longitude, altitude and periodicity component
Semnan 3 2 1 4 (month of the year) data were used as inputs to the each model.
Shiraz 2 4 1 3 Monthly data of 40 weather stations (40 sta-
Kerman 2 1 3 4
Tabriz 4 2 3 1
tions  12 months = 480 data) were used for training and 10 sta-
Hamedan 4 1 2 3 tions’ data (10 stations  12 months = 120 data) were used for
Qom 1 4 2 3 testing. The stations used for the testing procedure are Mashahd,
Isfahan 1 4 2 3 Bandar abbas, Semnan, Shiraz, Kerman, Tabriz, Hamedan, Qom,
Karaj 1 3 2 4
Isfahan and Karaj. The models were compared with each other
Total 24 23 22 31 with respect to RMSE (root mean square errors) and determination
coefficient (R2) statistics. The RMSE can be defined as
recombination rate: 0.1, gene transposition rate: 0.1, IS transposi- vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u N
tion rate: 0.1, RIS transposition: 0.1. Different basic functions and u1 X
linking functions were used for the GEP models and optimal one RMSE ¼ t ðET m  ET e Þ2 ð9Þ
N i¼1
was selected with respect to RMSE.

Table 8
Comparison of the applied models with respect to fitting line equations.

Stations ANN ANFIS-GP ANFIS-SC GEP

a b a b a b a b
Mashhad 1.073 0.871 1.026 0.433 1.067 0.287 0.900 1.079
Bandar abbas 1.040 0.559 0.838 0.999 1.453 1.367 1.319 1.665
Semnan 1.014 0.334 1.005 0.233 0.989 0.256 0.929 0.568
Shiraz 1.140 0.649 0.961 0.631 0.959 0.041 0.959 0.127
Kerman 1.037 0.740 1.017 0.086 0.924 0.415 0.882 0.189
Tabriz 0.617 0.590 0.723 0.454 0.675 0.597 0.770 0.693
Hamedan 0.793 0.329 0.879 0.404 0.843 0.213 0.899 0.611
Qom 0.894 0.331 0.767 0.369 0.863 0.354 0.797 0.435
Isfahan 1.048 0.351 0.970 0.277 1.011 0.273 0.991 0.104
Karaj 0.946 0.060 0.917 0.042 0.953 0.014 0.954 0.275
74 O. Kisi et al. / Computers and Electronics in Agriculture 115 (2015) 66–77
where ETo and ETe denote the measured and estimated evapotran-
spiration values, respectively and N is the number of data sets.
The comparison of different ANN training algorithms is made in
Table 2. In this table, 4-2-1 indicates an ANN model respectively
comprising 4 inputs, 2 hidden and 1 output nodes at 1,000 epochs.
The hidden nodes’ number was determined using trial and error
method. For each ANN model, hidden node number increased from
1 to 10 and the optimal ones were selected with respect to RMSE
criterion. Training of the lm, gd, gda, gdmgdx and scg algorithms
was respectively stopped after 1000, 50,000, 50,000, 50,000,
50,000 and 1000 iterations because the decrease in accuracy was
too small after these epochs. The tangent sigmoid and linear acti-
vation functions were used for the hidden and output nodes,
respectively. It is clear from Table 2 that gradient descent with
momentum and adaptive learning rate back-propagation algo-
rithm has a better accuracy than the other algorithms in the test
period.
Different ANFIS-GP models having different types of MFs is
compared in Table 3. In this table, 4-2-2-2 indicates an ANFIS-GP
model having 4, 2, 2 and 2 MFs for the month of the year, longi-
tude, latitude and altitude inputs. It is apparent from the table that
ANFIS-GP model with pi-shaped MFs has the best accuracy with
respect to RMSE and R2.
Table 4 gives the test results of the GEP models. It is clear from
the table that the best accuracy was obtained from the GEP model
with addition linking function and F5 operators. The expression
tree for the optimal GEP model is shown in Fig. 5. The equation
of the optimal GEP model is
Fig. 12. ET0 maps (mm) in January, February, March and April of Iran based on ANFIS-GP model.
ET0ðGEPÞ ¼ B þ exp ½arctg ðarctg ðarctg ð2  Longitude  ðAltitude  2:057ÞÞÞÞ
ð10Þ
6
tan1 ðLongitude Þ
where B ¼ A þ Latitude
and
monthþLongitude
 
2=3
A ¼ 3:061COS month  ðmonth Þ ð11Þ
Training and test results of the optimal data-driven models
which obtained from 50 stations are given in Table 5. It is clear
from the table that the ANFIS-GP model has the lowest RMSE
and the highest R2 in both training and test periods. The ranks of
the applied models in the test period are: ANFIS-GP, GEP, ANN
and ANFIS-SC. The RMSE accuracy of data-driven models in the
training period is compared for each station in Fig. 6. It is clear that
the ANFIS-GP model generally has lower RMSE values than the
other models. In some cases (Rasht, Yasuj, Shahrud, Babolsar,
Torbat-e Heydarieh, Bandar Anzali, Bandar Lengeh, Jask and
Zabol), ANFIS-SC model performs better than the ANFIS-GP model.
Test results of the optimal data-driven models for each station
are provided in Table 6. It is clear from the table that the RMSE
values range 0.238–1.474, 0.337–1.090, 0.261–1.290, 0.420–
0.930 mm/day for the ANN, ANFIS-GP, ANFIS-SC and GEP models,
respectively. For the ANN, ANFIS-GP and GEP models, the
maximum RMSE (1.474, 1.090, 0.930 mm/day) values were
found for the Tabriz station, respectively. For the ANFIS-SC
model, however, the maximum RMSE value was found to be
1.290 mm/day in the Bandar abbas station. The best ANNFIS-SC
 O. Kisi et al. / Computers and Electronics in Agriculture 115 (2015) 66–77
(RMSE = 0.261 mm/day) and GEP (RMSE = 0.420 mm/day) models
results were found for the Shiraz station while the ANN and
ANFIS-GP models provided best accuracy in the Karaj and
Kerman stations, respectively. It is clearly seen from Table 6 that
the accuracy of the ANFIS-GP model is generally better than the
other models in long-term monthly ET0 prediction. In four of ten
stations, the ANFIS-GP model has the best accuracy. The ANN
and ANFIS-SC models respectively performed the best in three
and two stations while the GEP was found to be the best model
in only one station. For the ANN, ANFIS-GP and ANFIS-SC models,
the maximum determination coefficient values were found to be
0.997, 998 and 0.994 in Karaj station, respectively. The GEP model
has the highest R2 value of 0.978 for the Qom station. The mini-
mum determination coefficient values were respectively found as
0.959 and 0.935 for the ANN and ANFIS-GP models in Bandar abbas
station while the ANFIS-SC and GEP models have the minimum R2
values of 0.937 and 0.677 in the Tabriz and Kerman stations,
respectively.
The test results of the data-driven models are compared in
Figs. 7–11. The monthly ET0 estimates given in the graphs gener-
ally confirm the test results given in Table 6. From the Figs. 8
and 11, it is clear that all the models provided good accuracy for
the Semnan, Isfahan and Karaj stations. ANFIS-SC and GEP models
significantly overestimated the ET0 values of Bandar abbas. All the
models gave underestimations for the summer months of Tabriz
station. The reason of this may be the fact that a sudden increase
is seen from May to June and from June to July months and sudden
decrease happens from August to September and from September
Fig. 13. ET0 maps (mm) in May, June, July and August of Iran based on ANFIS-GP model.
75
to October months in this station. The applied models cannot catch
these sudden changes. The accuracy ranks of the data-driven mod-
els are given in Table 7. It is clear from the table that the overall
rank of the methods’ accuracies is: ANFIS-SC, ANFIS-GP, ANN and
GEP. The overall accuracies of the ANFIS-SC, ANFIS-GP and ANN
models are similar to each other. The GEP model seems to be the
worst model. However, the main advantage of the GEP model is
that it has an explicit formulation can be simply used in practical
applications. Recently, Kisi and Sanikhani (2015) applied
ANFIS-GP, ANFIS-SC and ANN models for precipitation estimation
and they found the overall rank of the methods’ accuracies as
ANFIS-GP, ANN and ANFIS-SC. This indicates that the models’ accu-
racies change according to the investigated phenomenon.
Test results of the applied models are compared with respect to
fit line equations (assume that the equation is y = ax + b) in Table 8.
In this table, a values smaller than 1 indicate underestimation
while higher values show overestimation. From the table, it is clear
that the all the models give underestimations for the Tabriz,
Hamedan, Qom and Karaj stations. It is clear from the Fig. 1 that
these stations are located close to each other and thus have same
climatic conditions. It is clear from the Table 8 that the models
generally underestimates in long-term monthly ET0 prediction.
The reason of this may be the difference between the training
and validation data characteristics. Higher values used in training
period may lead data-driven models underestimations in the vali-
dation period. In Table 8, a and b values respectively close to 1 and
0 indicate that fitting line close to exact line (y = x). From Table 8, it
is apparent that the fitting lines of the ANFIS-SC models are
76 O. Kisi et al. / Computers and Electronics in Agriculture 115 (2015) 66–77
Fig. 14. ET0 maps (mm) in September, October, November and December of Iran based on ANFIS-GP model.
Fig. 15. Annual ET0 map (mm) of Iran based on average of 12 monthly ET0 maps
obtained by ANFIS-GP model.
generally closer to the exact line than those of the other models in
most of cases. In three of ten stations (Mashhad, Shiraz, Karaj), the
ANFIS-SC model has the closest fit lines. The ANN and ANFIS-GP
models has the closest fit lines in two stations (Bandar abbas,
Qom for ANN and Semnan, Kerman for ANFIS-GP) while the GEP
performed the best in only one station (Isfahan).
The interpolated maps of ET0 were obtained by using the opti-
mal ANFIS-GP model. Figs. 12–14 indicate the interpolated maps
of monthly ET0 (January to December) and Fig. 15 shows the
annual scale derived by average of monthly interpolated maps.
These figures indicate that the higher amounts of ET0 were seen
in the southern and southeastern parts of the province, while the
lower amounts could be seen in the north of the Iran province. In
the month of January, higher level of ET0 affects the southern
and southeastern parts of the province than the other regions of
this province. Then, high level of ET0 values is extended from
southern to eastern parts of the province from February to
September. From October, this extension is reduced and limited
to southern and especially southeastern parts in October,
November and December. It is clear from the monthly maps that
the amounts of ET0 increase from January to September and
decreases from October to February. According to the annual
map given in Fig. 15, the highest values of ET0 are found in the
southeastern regions, while the lowest values are seen in the
northern parts.
5. Conclusion
In this study, the accuracy of four different data-driven meth-
ods, ANN, ANFIS-GP, ANFIS-SC and GEP, were investigated to pre-
dict long-term monthly reference evapotranspiration by using
geographical input data. The periodicity component, station alti-
tude, longitude and latitude values were used as inputs to the
applied models to estimate the long-term monthly ET0 values.
Data from 50 weather stations in Iran were used for training and
testing of the models. The data-driven models were compared with
each other with respect to root mean-squared error and
 O. Kisi et al. / Computers and Electronics in Agriculture 115 (2015) 66–77
determination coefficient statistics. Different training algorithms
were used for ANN models and gradient descent with momentum
and adaptive learning rate back-propagation algorithm was found
to be better than the other algorithms. Various membership func-
tions were tried in ANFIS-GP models and pi-shaped MFs gave the
best accuracy. In the test period, the ANFIS-GP model generally
performed better than the other models in long-term monthly
ET0 prediction. Test results of the optimal data-driven models were
also compared for each station. The ANFIS-GP model had the best
accuracy in four out of ten stations. The ANN and ANFIS-SC models
respectively have the best accuracy in three and two stations while
the GEP was found to be the best model in only one station. It
should be noted that all the models gave R2 value higher than
0.900 except the one case (GEP model in Kerman station). As a con-
clusion, it can be said that the all the data-driven techniques can be
successfully used to predict the long-term reference evapotranspi-
ration of any site without climatic measurements.
Important information on various water resources issues such
as examining water use by different regions, water rights, water
allocations, water consumption and planning and management of
ground and surface water resources can be provided by ET0 maps.
The ET0 maps in different months revealed that the highest
amounts of reference evapotranspiration occurred in the southern
and especially southeastern parts of the Iran. More attention
should be focused on water resource planning of these parts.
Acknowledgement
This study was partly supported by The Turkish Academy of
Sciences (TUBA). The first author would like to thank TUBA for
their support of this study.
References
Abbaspour, K.C., Faramarzi, M., Ghasemi, S.S., Yang, H., 2009. Assessing the impact
of climate change on water resources in Iran. Water Resour. Res. 45, W10434.
http://dx.doi.org/10.1029/2008WR007615.
Ahrens, C., 1998. Essentials of Meteorology: An Introduction to the Atmosphere,
2nd ed. Publishing Company, Wadsworth.
Aytek, A., 2009. Co-active neurofuzzy inference system for evapotranspiration
Modeling. Soft Comput. 13 (7), 691–700.
Azamathulla, H.M., Ghani, A.A., Leow, C.S., Chang, C.K., Zakaria, N.A., 2011. Gene-
expression programming for the development of a stage-discharge curve of the
Pahang River. Water Resour. Manage. 25 (11), 2901–2916.
Cobaner, M., 2011. Evapotranspiration estimation by two different neuro-fuzzy
inference systems. J. Hydrol. 398 (3–4), 292–302.
Dinpashoh, Y., Jhajharia, D., Fakheri-Fard, A., Singh, V.P., Kahya, E., 2011. Trends in
reference crop evapotranspiration over Iran. J. Hydrol. 399 (3–4), 422–433.
Dogan, E., 2009. Reference evapotranspiration estimation using adaptive neuro-
fuzzy inference systems. Irrig. Drain. 58, 617–628.
Ferreira, C., 2001. Gene expression programming: a new adaptive algorithm for
solving problems. Complex Syst. 13 (2), 87–129.
Ferreira, C., 2006. Gene Expression Programming: Mathematical Modeling by an
Artificial Intelligence. Springer, Berlin: Heidelberg New York, 478.
Guven, A., Aytek, A., Ishak Yuce, M., Aksoy, H., 2008. Genetic programming-based
empirical model for daily reference evapotranspiration estimation. Clean 36
(10–11), 905–912.
Jang, J., 1993. ANFIS: adaptive-network-based fuzzy inference system. IEEE
Transact. Syst. Man. Cybern. 23 (3), 665–685.
Jang, J.S.R., Sun, C.T., Mizutani, E., 1997. Neuro-Fuzzy and Soft Computing: A
Computational Approach to Learning and Machine Intelligence. Prentice-Hall,
New Jersey.
77
Karimaldini, F., Shui, L.T., Mohamed, A.T., Abdollahi, M., Khalili, N., 2012. Daily
evapotranspiration modeling from limited weather data by using neuro-fuzzy
computing technique. J. Irrig. Drain Eng. 138 (1), 21–34.
Kim, S., Kim, H.S., 2008. Neural networks and genetic algorithm approach for
nonlinear evaporation and evapotranspiration modeling. J. Hydrol. 351, 299–
317.
Kisi, O., Ozturk, O., 2007. Adaptive neurofuzzy computing technique for
evapotranspiration estimation. J. Irrig. Drain. Eng. ASCE 133 (4), 368–379.
Kisi, O., Sanikhani, H., 2015. Prediction of long-term monthly precipitation using
several soft computing methods without climatic data. Int. J. Climatol. http://
dx.doi.org/10.1002/joc.4273.
Kisi, O., Zounemat-Kermani, M., 2014. Comparison of two different adaptive neuro-
fuzzy inference systems in modelling daily reference evapotranspiration. Water
Resour. Manage. 28 (9), 2655–2675.
Kousari, M.R., Ahani, H., 2012. An investigation on reference crop
evapotranspiration trend from 1975 to 2005 in Iran. Int. J. Climatol. 32 (15),
2387–2402.
Kumar, M., Raghuwanshi, N.S., Singh, R., Wallender, W.W., Pruitt, W.O., 2002.
Estimating evapotranspiration using artificial neural network. J. Irrig. Drain.
Eng. 128 (4), 224–233.
Kumar, M., Raghuwanshi, N.S., Singh, R., 2011. Artificial neural networks approach
in evapotranspiration modeling: a review. Irrig. Sci. 29 (1), 11–25.
Landeras, G., Ortiz-Barredo, A., Lopez, J.J., 2008. Comparison of artificial neural
network models and empirical and semi-empirical equations for daily reference
evapotranspiration estimation in the Basque Country (Northern Spain). Agric.
Water Manage. 95, 553–565.
Marti, P., González-Altozano, P., López-Urrea, R., Mancha, L.A., Shiri, J., 2015.
Modeling reference evapotranspiration with calculated targets. Assessment and
implications. Agric. Water Manage. 149 (2), 81–90.
Monteith, J.L., (1965). The state and movement of water in living organisms. In:
Proc., Evaporation and Environment, XIXth Symp., Soc. for Exp. Biol., Swansea.
Cambridge University Press, NY, pp. 205–234.
Mukherjee, I., Routroy, S., 2012. Comparing the performance of neural networks
developed by using Levenberg–Marquardt and Quasi-Newton with the gradient
descent algorithm for modelling a multiple response grinding process. Exp.
Syst. Appl. 39, 2397–2407.
Penman, H.L., 1963. Vegetation and hydrology. Tech. Communication No. 53,
Commonwealth Bureau of Soils. Harpenden, England.
Rahimikhoob, A., 2010. Estimation of evapotranspiration based on only air
temperature data using artificial neural networks for a subtropical climate in
Iran. Theor. Appl. Climatol. 101, 83–91.
Shiri, J., Nazemi, A.H., Sadraddini, A.A., Landeras, G., Kisi, O., Fard, Fakheri., Martí, P.,
2014a. Comparison of heuristic and empirical approaches for estimating
reference evapotranspiration from limited inputs in Iran. Comput. Electron.
108 (10), 230–241.
Shiri, J., Marti, P., Singh, V.P., 2014b. Evaluation of gene expression programming
approaches for estimating daily pan evaporation through spatial and temporal
data scanning. Hydrol. Processes 28 (3), 1215–1225.
Shiri, J., Sadraddini, A.A., Nazemi, A.H., Kisi, O., Landeras, G., Fakheri Fard, A., Marti,
P., 2014c. Generalizability of Gene Expression Programming-based approaches
for estimating daily reference evapotranspiration in coastal stations of Iran. J.
Hydrol. 508, 1–11.
Sudheer, K.P., Gosain, A.K., Ramasastri, K.S., 2003. Estimating actual
evapotranspiration from limited climatic data using neural computing
technique. J. Irrig. Drain. Eng. ASCE 129 (3), 214–218.
Tabari, H., Martinez, C., Ezani, A., Talaee, P.H., 2013. Applicability of support vector
machines and adaptive neurofuzzy inference system for modeling potato crop
evapotranspiration. Irrig. Sci. 31 (4), 575–588.
Trajkovic, S., 2005. Temperature-based approaches for estimating reference
evapotranspiration. J. Irrig. Drain. Eng. 131 (4), 316–323.
Traore, S., Guven, A., 2012. Regional-specific numerical models of
evapotranspiration using gene-expression programming interface in Sahel.
Water Resour. Manage. 26, 4367–4380.
Traore, S., Wang, Y.M., Kerh, T., 2010. Artificial neural network for modeling
reference evapotranspiration complex process in Sudano-Sahelian zone. Agric.
Water Manage. 97, 707–714.
Zounemat-Kermani, M., Scholz, M., 2013. Computing air demand using the Takagi-
Sugeno model for dam outlets. Water 5 (3), 1441–1456.
Zounemat-Kermani, M., Teshnehlab, M., 2008. Using adaptive neuro-fuzzy
inference system for hydrological time series prediction. Appl. Soft Comput. 8
(2), 928–936.
Zounemat-Kermani, M., Kisi, O., Rajaee, T., 2013. Performance of radial basis and
LM-feed forward artificial neural networks for predicting daily watershed
runoff. Appl. Soft Comput. 13 (12), 4633–4644.