Journal of Environmental Management 184 (2016) 255e270

Contents lists available at ScienceDirect

Journal of Environmental Management

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

Research article

A fuzzy-logic based decision-making approach for identification of

groundwater quality based on groundwater quality indices
M. Vadiati a, b, A. Asghari-Moghaddam a, M. Nakhaei c, J. Adamowski b, *,
A.H. Akbarzadeh b
a
Department of Earth Sciences, University of Tabriz, Tabriz, Iran
b
Department of Bioresource Engineering, McGill University, Sainte-Anne-de-Bellevue, Quebec, H9X 3V9, Canada
c
Department of Applied Geology, Kharazmi University, Tehran, Iran

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

Article history: Due to inherent uncertainties in measurement and analysis, groundwater quality assessment is a difficult
Received 4 November 2015 task. Artificial intelligence techniques, specifically fuzzy inference systems, have proven useful in eval-
Received in revised form uating groundwater quality in uncertain and complex hydrogeological systems. In the present study, a
26 September 2016
Mamdani fuzzy-logic-based decision-making approach was developed to assess groundwater quality
Accepted 28 September 2016
Available online 6 October 2016
based on relevant indices. In an effort to develop a set of new hybrid fuzzy indices for groundwater
quality assessment, a Mamdani fuzzy inference model was developed with widely-accepted ground-
water quality indices: the Groundwater Quality Index (GQI), the Water Quality Index (WQI), and the
Keywords:
Groundwater quality
Ground Water Quality Index (GWQI). In an effort to present generalized hybrid fuzzy indices a significant
Mamdani fuzzy system effort was made to employ well-known groundwater quality index acceptability ranges as fuzzy model
Drinking water quality output ranges rather than employing expert knowledge in the fuzzification of output parameters. The
Fuzzy logic proposed approach was evaluated for its ability to assess the drinking water quality of 49 samples
Physicochemical analysis collected seasonally from groundwater resources in Iran's Sarab Plain during 2013e2014. Input mem-
bership functions were defined as “desirable”, “acceptable” and “unacceptable” based on expert
knowledge and the standard and permissible limits prescribed by the World Health Organization. Output
data were categorized into multiple categories based on the GQI (5 categories), WQI (5 categories), and
GWQI (3 categories). Given the potential of fuzzy models to minimize uncertainties, hybrid fuzzy-based
indices produce significantly more accurate assessments of groundwater quality than traditional indices.
The developed models' accuracy was assessed and a comparison of the performance indices demon-
strated the Fuzzy Groundwater Quality Index model to be more accurate than both the Fuzzy Water
Quality Index and Fuzzy Ground Water Quality Index models. This suggests that the new hybrid fuzzy
indices developed in this research are reliable and flexible when used in groundwater quality assessment
for drinking purposes.
© 2016 Elsevier Ltd. All rights reserved.

1. Introduction
Groundwater is a major water resource in semiarid areas; its
quality and availability are key issues for environmental managers
and hydrogeologists (Jiang et al., 2009; Bain et al., 2014).
Groundwater is often one of the few reliable resources of water
available to rural communities in semiarid areas such as Iran
* Corresponding author.
E-mail addresses: mey_vadiati@tabrizu.ac.ir (M. Vadiati), moghaddam@tabrizu.
ac.ir (A. Asghari-Moghaddam), nakhaei@khu.ac.ir (M. Nakhaei), jan.adamowski@
mcgill.ca (J. Adamowski), hamid.akbarzadeh@mcgill.ca (A.H. Akbarzadeh).
(Girman et al., 2007) where population growth and the resulting
intensification of water demand is causing increased scarcity of
drinking water resources. Temporal changes in the origin of the
recharged water, hydrologic events, and anthropogenic activities
may cause periodic changes in groundwater quality
(Vasanthavigar et al., 2010). Generally, water quality declines
during passage downstream and through movement from a
recharge zone to a discharge. Numerous methods to evaluate
drinking water quality criteria are available in the published
literature (Lumb et al., 2011).
A pioneer in the field, Horton (1965) defined the Water Quality
Index (WQI), ranging from 0 (poor) to 100 (ideal), based on eight

http://dx.doi.org/10.1016/j.jenvman.2016.09.082
0301-4797/© 2016 Elsevier Ltd. All rights reserved.
256 M. Vadiati et al. / Journal of Environmental Management 184 (2016) 255e270
Nomenclature
AI Artificial Intelligence
AMSL Above Mean Sea Level
BOD5 Biological Oxygen Demand (over 5 days of
incubation at 20
C)
CAM Cosine Amplitude Method
CCME-WQI Canadian Council of Ministers of the
Environment Water Quality Index
DO Dissolved Oxygen
FGQI Fuzzy Groundwater Quality Index
FGWQI Fuzzy Ground Water Quality Index
FIS Fuzzy Inference Systems
FWQI Fuzzy Water Quality Index
GIS Geographic Information System
GQI Groundwater Quality Index
GWQI Ground Water Quality Index
TDS Total Dissolved Solids
WHO World Health Organization
WQI Water Quality Index
water quality parameters weighted according to their relative
importance. Horton's efforts were followed by research that
strived to develop less subjective and more sensitive WQIs (e.g.
Bolton et al., 1978; Liou et al., 2004; Said et al., 2004; Nasiri et al.,
2007; Dos Santos Simoes et al., 2008; Cordoba et al., 2010;
Vicente et al., 2011; Zhao et al., 2013; Nazeer et al., 2014; Al-
Mutairi et al., 2014; Aminu et al., 2014). An improved version of
the WQI (Brown et al., 1970), in which relevant parameter se-
lection followed the Delphi method, was adopted in the United
States by the National Sanitation Foundation (NSF-WQI). The
Canadian Council of Ministers of the Environment Water Quality
Index (CCME-WQI) developed an updated approach (CCME,
2001). The main changes in recently developed WQIs are found
in the interpretation and calculation procedures (Lumb et al.,
2011). The WQI methods are not perfect and weaknesses
include assigning input parameter weights, their interpretation,
index calculation methods, and appropriate WQI ranges denoting
acceptable water quality (Lumb et al., 2011; Abbasi and Abbasi,
2012; Mohebbi et al., 2013).
Most WQI studies have used surface water quality indices for
groundwater quality assessment even though the important pa-
rameters in surface and groundwater quality are different. Biolog-
ical and physicochemical parameters are important for surface
water, whereas hydrochemical parameters like major anions and
cations are important in groundwater quality assessment. In a first
attempt to develop a WQI for groundwater, Tiwari and Mishra
(1985) employed the surface WQI to rank groundwater for drink-
ing purposes. While many groundwater quality assessment studies
mimicked this procedure over the years (Ramachandramoorthy
et al., 2010; Chaturvedi and Bassin, 2010; Ketata-Rokbani et al.,
2011; Sadat-Noori et al., 2014), others used different approaches
and modes of interpretation to develop groundwater specific WQIs
(Banoeng-Yakubo et al., 2009; Giri et al., 2010; Banerjee and
Srivastava, 2011; Shi et al., 2013; Mohebbi et al., 2013; El-Fadel
et al., 2014; Abtahi et al., 2015).
A simple WQI based on nine hydrochemical parameters
including heavy metals was proposed by Soltan (1999) to assess
the water quality of 10 artesian wells located near the Dakhla
oasis in Egypt. On the basis of drinking water guidelines, Stigter
et al. (2006) developed a groundwater quality index and a
composition index using a correspondence factor analysis on
several important groundwater hydrochemical parameters.
Vasanthavigar et al. (2010), working in Tamilnadu, India, used 12
hydrochemical parameters to evaluate groundwater quality for
drinking purposes.
In some cases, researchers have attempted to develop ground-
water quality indices that specifically considered important phys-
icochemical parameters. Based on the NSF-WQI and using
normalized parameter values rather than sub index values, Saeedi
et al. (2010) developed a robust indexing methodology for
groundwater - the Ground Water Quality Index (GWQI). When
faced with the difficulty of water quality studies in inaccessible
regions, hydrogeologists began to use Geographic Information
Systems (GIS) in an effort to better understand the spatial distri-
bution of water quality parameters (Selvam et al., 2014). Babiker
et al. (2007) proposed a GIS-based Groundwater Quality Index
(GQI) that used a statistical methodology to translate water quality
parameters into a new index consistent with WHO standards.
Notwithstanding the wide scope of these studies, the deter-
ministic indices that have been proposed, and the WQI approaches
that have been implemented in water quality assessment have not
considered environmental and experimental uncertainties which
occur throughout their assessment. Therefore, an advanced
decision-making approach is required. Introducing a threshold of
safety instead of a single value to WQI standards is one approach to
overcome the difficulties in handling uncertainty in water quality
evaluation (Dahiya et al., 2007).
To address problems in the WQI approach such as its failure to
incorporate significant parameters of water quality and their
inherent uncertainty, some new water quality assessment ap-
proaches based on Artificial Intelligence (AI) have been proposed
(Nikoo et al., 2011; Gazzaz et al., 2012; Maiti et al., 2013; Patki et al.,
2015). Fuzzy logic is one of the AI methods which has been
extensively applied in the development of WQIs. Fuzzy logic is a
promising tool for the development of environmental indices since
it has the ability to reflect human understanding and expert
knowledge. Also, it can deal with non-linear, uncertain and
ambiguous datasets. The linguistic format of the model makes it
more understandable for the public, managers, and non-experts
(Gharibi et al., 2012a).
Introduced by Zadeh (1965), fuzzy logic is a mathematical
discipline based on fuzzy set theory instead of classical mathe-
matics. A recent increase in fuzzy model applications in the field of
water quality assessment indicates that fuzzy-rule-based models
are useful in solving water resources problems (Liou et al., 2003;
Lermontov et al., 2009; Ocampo-Duque et al., 2013; Wang et al.,
2014). Employing fuzzy sets instead of crisp values, Sii et al. (1993)
addressed the uncertainties associated with water quality to create
an applicable fuzzy set rather than the conventional scale of 0e100
generally used in WQI methodology. Based on fuzzy inference
systems (FIS), Ocampo-Duque et al. (2006) proposed a fuzzy WQI
wherein dissolved oxygen (DO) and organic matter (BOD5) were
represented by trapezoidal membership functions, and input levels
were assumed to be adequate to evaluate water quality. Lermontov
et al. (2009) developed a fuzzy WQI and compared its applicability
to three conventional WQIs in Brazilian river watersheds, specif-
ically testing the Fuzzy Water Quality Index on the Ribeira do
Iguape River.
The novel fuzzy-logic-based surface water WQI for drinking
water purposes devised by Gharibi et al. (2012a) provided more
precise results than traditional methods. Another fuzzy-logic-based
surface water WQI used by Ocampo-Duque et al. (2013) showed
that the flexible boundaries among the linguistic terms in the
fuzzy-logic-based WQI allowed for more accurate classification of
 M. Vadiati et al. / Journal of Environmental Management 184 (2016) 255e270
water quality than traditional methods. Notwithstanding the
notable increase in studies employing hybrid fuzzy-logic-based
WQIs in recent years, there is still no comprehensive exhibition
of fuzzy logic based WQIs for groundwater. A special index for
groundwater quality evaluation should be considered given the
flexibility of fuzzy set theory in decision-making in imprecise en-
vironments, and the importance of the inherent uncertainty in
drinking water quality assessments.
Some recent studies have attempted to assess groundwater
quality using FIS. Dahiya et al. (2007) used the Mamdani Fuzzy
Inference System evaluation (Mamdani and Assilian, 1975) to
analyze groundwater quality. The optimization model was tested
with 42 groundwater samples on the basis of 10 drinking water
quality parameters. Kumar et al. (2009) employed fuzzy set theory
for decision making in groundwater quality assessment for 30
groundwater samples. The groundwater quality was classified into
desirable, acceptable and unacceptable categories. The authors
concluded that the fuzzy groundwater quality model was able to
accurately assess groundwater quality for drinking purposes.
Jinturkar et al. (2010) proposed a fuzzy logic model that dealt with
the physicochemical quality of groundwater in Chikhli, India.
Rather than being specifically designed for groundwater, this
method was developed in a manner similar to that employed for
surface WQIs. Triangular membership functions were developed
for the six physicochemical parameters: pH, Total Dissolved Solids
(TDS), Ca2þ, Mg2þ, Hardness, and Cl.
Based on previous studies, the selection of an appropriate index
as well as an acceptable methodology for groundwater quality
assessment remains a challenge. The main problems with the WQI
and FIS developed for groundwater quality assessment can be
summarized as follows:
(1) There are many WQIs but an appropriate index specific to
groundwater quality assessment remains a challenge.
(2) Regarding data assessment, WQIs focus on individual pa-
rameters and are not able to provide a comprehensive
framework of water quality.
(3) There are some weaknesses in selecting and weighting pa-
rameters to be included in WQIs, in addition to weaknesses
in interpretation, the methodology of calculations and
appropriate water quality acceptability ranges.
(4) A significant amount of research has dealt with surface water
quality, whereas few studies have been conducted on
groundwater quality.
(5) The WQIs are used without considering inherent un-
certainties which appear in various steps of the approach.
(6) In previous studies using fuzzy-logic-based groundwater
quality assessment, output membership functions were
defined based only on expert knowledge, thereby generating
models which differed significantly from modeler/expert to
modeler/expert.
The novelty of this study is to develop a generalized rule-based
fuzzy model based on widely accepted WQIs (GQI, WQI and GWQI).
The developed model uses acceptability ranges of well-known
indices for output ranges of fuzzy models to present a new set of
hybrid fuzzy indices: Fuzzy Groundwater Quality Index (FGQI),
Fuzzy Water Quality Index (FWQI) and Fuzzy Ground Water Quality
Index (FGWQI) for groundwater quality assessment for drinking
purposes following the limits prescribed by the WHO. Conditions
such as spatial and temporal variations, the impacts of physico-
chemical parameters on human health, and expert knowledge are
considered to improve groundwater quality assessment over
traditional indices. The overall process of this study is illustrated in
Fig. 1.
257
2. Methods and materials
2.1. Site characteristics
The applicability of fuzzy-logic-based models is best revealed
when models are tested with large datasets instead of specific data.
Given the high variability in groundwater quality across the region,
Iran's Sarab Plain was selected to test the performance of the
groundwater quality assessment models developed in the present
study. This variability is the result of geological uncertainty stem-
ming from the dissolution of evaporative deposits, hydrothermal
activities of Mount Sabalan in the north, high seasonal ground-
water quality changes, and the strong impact of agricultural and
anthropogenic activities.
Located in northwestern Iran, the Sarab Plain is a sub-basin of
Lake Urmia watershed which ranges in elevation from 1660 to
4850 m AMSL (Mount Sabalan). Mean annual rainfall in the city of
Sarab is almost 343 mm. The mean annual temperature across the
study area is 7
C, and ranges from 7.9
C on the plain to 6.2
C at the
highest elevation. Sarab Plain sediments are composed of quater-
nary deposits originating from the alteration of the Sabalan and
Bozqoush mountains in the north and south, respectively. The main
rocks in the north and south mountains are Eocene andesite, latite,
ignimbrite, dacite, and basalt. The most important outcrops in the
study area are Oligocene nepheline-syenite, Miocene evaporate
sediments with gypsum and salt minerals, Miocene siltstone and
marl, Miocene red conglomerate, Miocene sandstone, Pliocene
andesite, olivine-basalt, trachyandesite, and ignimbrite. A geologic
map of the study area is presented in Fig. 2.
Rivers, springs, qanats and shallow wells formerly supplied
water demands, but recently increased irrigation activities and
population growth have led to the drilling of a number of deep
wells to obtain supplemental water. This has caused deterioration
in groundwater quality.
A total of 49 representative groundwater samples were selected
and sampled in the dry and wet seasons during 2013e2014. The
fluctuation of groundwater level and changes in groundwater
quality in the study area during 10 years are shown in Fig. S1. This
graph allows for the comparison of the quality and quantity
changes of the groundwater from year to year. As we can see in
Fig. S1, there is a considerable decline in groundwater level in the
study area in recent years. A water resource map showing the
location of groundwater resource sampling sites, including qanats,
spring and wells, is presented in Fig. S2.
Major cations and anions including bicarbonate (HCO 3 ), chlo-
ride (Cl), sulfate (SO2  
4 ), nitrate (NO3 ), fluoride (F ), lithium (Li )
þ
calcium (Ca ), magnesium (Mg ), sodium (Na ), potassium (Kþ)
2þ 2þ þ
and silicon oxide (SiO2) and TDS were analyzed using standard
procedures (APHA et al., 1998) in the hydrogeological laboratory of
the University of Tabriz.
2.2. Model development
2.2.1. Theoretical background
Fuzzy logic, a methodology proposed by Zadeh (1965, 1973) to
address issues of vagueness and uncertainty, has been used in
different branches of science and technology. FISs are able to
transfer the qualitative aspects of expert knowledge and
reasoning processes through fuzzy if-then rules (Zadeh, 1973).
Therefore, fuzzy sets and fuzzy logic can collect expert knowledge
and deal with uncertainties in the decision-making process
(Mamdani and Assilian, 1975). Capable of addressing different
types of uncertainty (e.g., vagueness, variability, and imprecision
intrinsic to environmental problems), fuzzy inference rule-based
systems generally consist of four basic elements: fuzzification,
258 M. Vadiati et al. / Journal of Environmental Management 184 (2016) 255e270

Fig. 1. Overall procedure to develop the hybrid Fuzzy Indices models.

fuzzy rule base, fuzzy inference system and defuzzification
(Fig. S3).
Given the fact that intrinsic uncertainty can influence results,
fuzzy inference rule-based systems are useful in decision-making,
especially when dealing with water quality problems (Chau,
2006; Lermontov et al., 2009). The Mamdani FIS has been chosen
by a number of investigators in the field of water quality assess-
ment due to its flexibility (Dahiya et al., 2007; Kumar et al., 2009;
Jinturkar et al., 2010; Gharibi et al., 2012a; Ocampo-Duque et al.,
2013). The FIS approach employs uncertain data (due to errors in
measurements and inherent uncertainties in the complex aquifer
system) and provides better water quality evaluation than other AI
methods in cases where water quality data is marginal or varies
little (Dahiya et al., 2007).
The selection and weighting parameters in crisp water quality
indices are based on simple mathematical equations while the
present study uses fuzzy inference models to develop hybrid fuzzy
water quality indices based on expert knowledge. More specifically,
 M. Vadiati et al. / Journal of Environmental Management 184 (2016) 255e270
Fig. 2. Geologic map of the study area (Sarab Plain, Iran).
the ranges of the input parameters are based on prescribed stan-
dard limits of the WHO and the ranges of the output parameters are
based on widely accepted water quality indices for groundwater
(GQI, WQI, and GWQI). As a white box model, one of the main
advantages of fuzzy models is their ability to deal simultaneously
with both linguistic information and numerical accuracy. Further-
more, the models are transparent regarding the fuzzy system in the
form of if-then rules and membership functions, especially when
the input-output measurements are difficult to obtain or have
inherent uncertainties that are difficult to address (Yen and Langari,
1998). The Mamdani FIS was chosen in the present study because it
is readable, flexible, tolerant to noise and imprecise data, inter-
pretable, resistant to uncertainty, and a universal approximate
model (Sivanandam et al., 2007; Wang et al., 2009).
In our study an attempt is made to minimize model sensitivity
by employing WHO prescribed standard limits for input member-
ship function ranges and similar acceptability ranges for the output
ranges of well-known indices. In doing so, expert knowledge,
previous studies and trial and error directed the development of
the generalized model. Instead of using expert knowledge for input
and output ranges of the fuzzy model, the development of the
present fuzzy models are based on WHO limits as input categories,
and an acceptability range for the selected indices as output
categories.
2.2.1.1. Fuzzy systems. The fuzzy sets are defined as a membership
function which assigns a domain of interest for the interval [0, 1].
The fuzzy set A 3 X is given as:
A ¼ fðx1 ; mA ðxÞÞjx2Xg; 0  mA ðxÞ ¼ 1
where x1 are the elements of the universe of discourse (X), mA(x) is
the membership function of x in A.
259
2.2.1.2. Fuzzification. Fuzzification is constructed with different
shapes of membership functions, either linear (trapezoidal or
triangular) or nonlinear. Trapezoidal and triangular shapes are the
most commonly used membership functions (Kusko, 1993). The
development of the fuzzy models in the present study proceeded
through various steps, including the fuzzification of the input range
of physicochemical parameters and the output range of models, as
well as the application of “min” and “max” for “and” and “or”
methods, respectively, in the antecedent that the minimum and
maximum of membership functions were drawn from. The “min” is
then applied in the implication from the antecedent to the conse-
quent and the “max” is used for the aggregation of the consequent
through the rules.
2.2.1.3. Fuzzy rule base. A fuzzy rule base is comprised of mathe-
matical, fuzzy ifethen rules that are provided by experts and
extracted from the datasets. The fuzzy rules in the inference system
translate expert judgments expressed in linguistic terms. The forms
of a standard mathematical ifethen rule is noted as “If x1 is A, and x2
is A, then Y is B”, where A and B are linguistic terms defined on the
universal sets of x1, x2 and Y. The ‘if’ portion is called the antecedent
while the ‘then’ portion is called the consequent (Ross, 2012; Wang
et al., 2009).
2.2.1.4. Fuzzy inference system. Knowledge-based or expert sys-
tems are examples of AI that are represented by a set of linguistic
terms employed in acquiring heuristic knowledge. These systems
are capable of increasing the initial knowledge base through a
procedure of learning or inference. There is increasing interest in
(1) developing fuzzy-logic-based decision-making models based on
measured data instead of the traditional linguistic rules usually
employed. A fuzzy inference system can help to make decisions or
to recognize patterns by a reasoning mechanism or an aggregation
260 M. Vadiati et al. / Journal of Environmental Management 184 (2016) 255e270
of individual rules (Ross, 2012). The procedure of fuzzy inference
involves three significant concepts: membership functions, logical
operations, and if-then rules (Zadeh, 1965).
2.2.1.5. Defuzzification. The defuzzification procedure is the last
step in the processing of FIS. Through the defuzzification process,
the results obtained in the form of fuzzy sets are converted into
numerical values. The centroid of area (COA), the center of gravity,
the mean of the maximums, and the smallest of the maximums are
frequently used defuzzification methods. More detailed de-
scriptions of fuzzy logic and FIS models can be found in Ross (2012).
In the present work the commonly employed COA defuzzifier
method (Hellendoorn and Thomas, 1993) is employed in obtaining
crisp output values. These processes are applied using the FIS editor
of ‘‘fuzzy logic toolbox’’ in MATLAB (Mathworks, 2014). The struc-
tures of the proposed fuzzy models are shown in Fig. 3.
2.3. Fuzzy water quality index development
In this study, several well-known crisp indices for groundwater
quality assessment: (i) the GQI (Babiker et al., 2007), the well-
established GIS-based groundwater quality assessment method
(ii) the WQI (Tiwari and Mishra, 1985) which is the basis of many
newly developed indices, and (iii) the GWQI (Saeedi et al., 2010),
the most recent groundwater quality assessment index (Lumb et al.,
2011), were selected to develop new fuzzy logic based indices:
FGQI, FWQI, and FGWQI. Output membership functions are defined
based on the prescribed index values range, water quality classes
and expert knowledge, while input membership functions are the
same for all the models (FGQI, FWQI, and FGWQI).
The final model was developed based on the results of an expert
panel discussion in which different experts, with various scientific
backgrounds related to groundwater quality assessment, worked
together at the Department of Earth Sciences, University of Tabriz,
Tabriz, Iran in November 2014. The expert team, consisting of 10
people, came from different scientific backgrounds (three from
hydrochemistry analysis, one form laboratory analysis, four from
hydrogeology, and two from fuzzy inference system modeling).
These experts worked on different steps of the FIS model devel-
opment, including the selection of groundwater quality parame-
ters, fuzzification of input and output parameters, and the
development of fuzzy if-then rules.
2.3.1. Fuzzification of input parameters
The parameter selection for any groundwater quality assess-
ment depends upon the purpose of the assessment and the ability
of an organization to carry out accurate parameter evaluation. Se-
lection should include those parameters most suitable to local
environments and related to background groundwater quality.
Moreover, water availability is an important influence on the
acceptability of groundwater quality; a lower quality might be
acceptable when water resources are scarce (Chapman, 1996).
In this research, seven key parameters including TDS, Cl, Ca2þ,
Mg2þ, SO2 4 , Na
þ
and NO 3 were considered in drinking water
quality assessment based on previous studies, expert knowledge
and their impacts on human health. These were selected for the
development of a FIS model to assess groundwater quality. Major
ions like Naþ, SO2
4 and Ca
2þ
exist naturally in groundwater, having
leached from sedimentary rocks and are commonly considered in
groundwater quality assessment for drinking purposes. Along with
Ca2þ, Mg2þ is one of the main contributors to water hardness. High
concentrations of Cl and NO 3 in groundwater can increase the risk
of impacts on human health and render water unpalatable.
Although more parameters improve the suitability of a model,
they also increase the uncertainty. Moreover, a good index should
Calcium
Magnesium
FGQI
Sodium
FIS
Chloride FWQI
(Mamdani)
Nitrate
FGWQI
Sulfate
Total Dissolved
Solids
Fig. 3. Schematic illustration of the FISs.
be able to assess groundwater quality with easily available, inex-
pensive data. A summary of the selected parameters for the dry and
wet seasons is shown in Table 1.
The parameters of the if-then rules define a fuzzy region of the
input space, and the output parameters specify the corresponding
output. Hence, the efficiency of the FIS depends on the relevant
inputs, number of membership functions for each input and the
corresponding numerical data. From a purely mathematical view-
point, there are many different shapes of membership functions. In
most practical applications, however, simple “trapezoid” member-
ship functions work well and linear interpolation is used to obtain
both endpoints of the interval (Barua et al., 2014). The shape of
membership functions depends on the subject to be solved (Klir
and Yuan, 1995); trapezoidal and triangular membership func-
tions are chosen based on the nature of physicochemical data and
past research (Ocampo-Duque et al., 2006; Dahiya et al., 2007;
Kumar et al., 2009; Gharibi et al., 2012a) to represent “desirable”,
“acceptable” and “unacceptable” fuzzy sets using the crisp input
categories of standard limits prescribed by the WHO (2004).
Furthermore, the parameters of the fuzzy rules and the shape of
membership functions have been adjusted in order to obtain an
optimal fuzzy system. Membership functions of input parameters
are shown in Fig. 4.
2.3.2. Fuzzification of output parameters
Previous studies predominantly dealt with the assessment of
data and focused on individual parameters based on observed data,
thereby failing to provide a comprehensive and interpretive
framework of water quality. Many methodologies that were
developed were not specific to groundwater because most WQI
parameters did not include common minerals found in ground-
water. In the present study, the GQI method (Babiker et al., 2007),
the WQI (Tiwari and Mishra, 1985), and the GWQI (Saeedi et al.,
2010), were selected to develop new fuzzy logic based indices,
FGQI, FWQI, and FGWQI.
The dry and wet season results of GQI are presented in Tables S1
and S2, respectively. The GQI values ranged from 0 to 100 and are
categorized into five classes. The output membership functions of
the FGQI model are categorized based on the GQI classes:
“completely unacceptable” (0e25); “unacceptable” (25e50);
“moderate” (50e70); “acceptable” (70e90); and “good” (90e100).
The membership functions defined for the output of FGQI are
presented in Fig. 5.
 M. Vadiati et al. / Journal of Environmental Management 184 (2016) 255e270 261

Table 1
Seasonal variations and thresholds of drinking water parameters (Sarab Plain, Iran).

Season Ca2þ (mg/l) Mg2þ (mg/l) Naþ (mg/l) Cl (mg/l) NO
3 (mg/l) SO2-
4 (mg/l) TDS (mg/l)

Minimum Dry 28 7.6 11 6.9 0.1 4.9 99.6

Wet 22 3.6 2.3 0.1 0.0 4.8 14.6
Maximum Dry 400 89 3350 4900 46 826 9050
Wet 385 83 3334 4807 174 758 8649
Mean Dry 104.7 24.1 132 180 8.3 129 666
Wet 107 25.3 143 181 12.6 148 542
SD Dry 67 183 477 697 10 139 128
Wet 74 18 478 685 26 165 1264
Threshold standard (WHO, 2004) 300 300 200 300 50 250 600

In the present study, the modified WQI method for groundwater (2) If Ca2þ is acceptable; Mg2þ is acceptable; Naþ is desirable;
quality assessment (Tiwari and Mishra, 1985) was used to develop Cl is acceptable; NO 2
3 is desirable; SO4 is desirable and
the FWQI model. The WQI for all dry and wet season samples are TDS is acceptable; then: FGQI is acceptable.
presented in Tables S1 and S2, respectively. The WQI classes are (3) If Ca2þ is desirable; Mg2þ is desirable; Naþ is desirable; Cl is
categorized as “excellent” (0e25); “good” (26e50); “poor” (51e75); desirable; NO 3 is desirable; SO4
2
is desirable and TDS is
“very poor” (76e100); and “unfit for drinking” (WQI>100) desirable; then: FWQI is excellent;
(Srivastava et al., 2011). The output membership functions of the (4) If Ca2þ is unacceptable; Mg2þ is acceptable; Naþ is accept-
FWQI model are defined based on the five WQI classes and are able; Cl is acceptable; NO 3 is acceptable; SO4
2
is unac-
presented in Fig. 5. ceptable and TDS is unacceptable; then: FWQI is poor.
An innovative method in the development of a WQI for (5) If Ca2þ is acceptable; Mg2þ is desirable; Naþ is desirable; Cl
groundwater, the GWQI, was proposed by Saeedi et al. (2010). The is desirable; NO 2
3 is acceptable; SO4 is desirable and TDS is
GWQI is a NSF-WQI-based index that utilizes the standardized acceptable; then: FGWQI is high;
concentrations of each parameter instead of the subindex values (6) If Ca2þ is unacceptable; Mg2þ is desirable; Naþ is desirable;
(Lumb et al., 2011). Using this method, the studied parameters are Cl is desirable; NO 2
3 is desirable; SO4 and TDS is unac-
weighted according to their perceived importance to aggregate all ceptable; then: FGWQI is low.
the normalized values. The calculated values of the GWQI for the
dry and wet seasons are presented in Tables S1 and S2, respectively. 2.4. Modeling performance criteria
The output membership functions of the FGWQI model are cate-
gorized based on the GWQI classes: “high” (GWQI > 0.15); “suit- Common performance indices, namely the Variance Accounted
able” (0.04 < GWQI<0.15); and “low” (GWQI < 0.04) (Fig. 5). For (VAF), Root Mean Square Error (RMSE) and Coefficient of
Determination (R2) (Yilmaz et al., 2012; Rezaei et al., 2015; Shams
et al., 2015), were selected to assess the performance of the
2.3.3. Development of fuzzy rules
hybrid fuzzy indices developed in the present study. The VAF shows
Based on available datasets and expert knowledge on the im-
the degree of difference between the variances of measured and
pacts of physicochemical input parameters on human health, rules
predicted outputs. The RMSE is selected as the common perfor-
were developed for the construction of the model. The aim was to
mance index with the ability to show the bias between measured
propose a comprehensive model that could be applied to new
and predicted data. The R2 between measured and predicted values
groundwater quality assessment datasets.
is a good indicator of the predictive performance of the models. A
The number of rules depends on the number of input parame-
high level of R2 signifies conformity between the output of the
ters and membership functions. When the number of fuzzy rules is
model and the real measured values. These indices are calculated
large, the complexity of the model increases. Therefore, removal of
by Eqs. (2) and (3):
redundant or less important fuzzy rules from the rule base can
 0
!
result in a compact fuzzy model with better generalizing ability and var y  y
an overall simplification of the system architecture (Yen and Wang, VAF ¼ 100 1  (2)
varðyÞ
1999). To decrease the number of rules, Dahiya et al. (2007) cate-
gorized input parameters into three groups: pH, TDS, Cl and SO2 4 vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
in the first group, Ca2þ, Mg2þ, Total Alkalinity and Total Hardness in u i¼N
u1 X
the second group, and NO RMSE ¼ t

ðy  y Þ2
0
3 and F in the third group. This cate- (3)
gorization of input parameters into groups may have resulted in N
i¼1
important data losses and to overcome this deficiency, all input 0
Where y, and y are the measured and predicted values,
parameters in the present study were assessed individually in one
respectively, and N is the number of values in the dataset. If the VAF
FIS model. The models developed here consisted of seven input
is 100, the RMSE is zero, and the R2 is 1.0, then the model is
parameters, with each parameter consisting of three membership
perfectly accurate.
functions. Such a model would require many rules; to avoid model
complexity, a total of 204 of the most important rules - of which six
examples are provided below d were selected for constructing the 3. Results and discussion
fuzzy rule base on the basis of available datasets and the experience
of experts. Fig. S4 shows an example of the application of the FGQI model
where columns represent the inputs/output fuzzy sets, and rows
(1) If Ca2þ is desirable; Mg2þ is desirable; Naþ is desirable; Cl- the fuzzy rules. It can be seen that when the input parameters are:
is desirable; NO 2
3 is desirable; SO4 is desirable and TDS is [Ca2þ] ¼ 76.2 mg L1, [Mg2þ] ¼ 15.8 mg L1, [Naþ] ¼ 18.4 mg L1,
acceptable; then: FGQI is good. [Cl] ¼ 7.1 mg L1, ½NO 1 2
3  ¼ 6.39 mg L , ½SO4  ¼ 52.8 mg L
1
and
262 M. Vadiati et al. / Journal of Environmental Management 184 (2016) 255e270

Desirable Acceptable Unacceptable Desirable Acceptable Unacceptable

Membership Fuction (μ)

1 1
Membership Fuction (μ)

0.75 0.75
0.5 0.5
0.25 0.25
0 0
0 50 100 150 200 250 300 350 400 0 500 1000 1500 2000 2500 3000 3500
Ca (mg/L) Cl (mg/L)

Desirable Acceptable Unacceptable Desirable Acceptable Unacceptable

1

Membership Fuction(μ)
1
Membership Fuction (μ)

0.75 0.75
0.5 0.5
0.25 0.25
0 0
0 50 100 150 200 250 300 350 400 0 20 40 60 80 100 120 140 160 180
Mg (mg/L) NO3 (mg/L)

Desirable Acceptable Unacceptable Desirable Acceptable Unacceptable

1
Membership Fuction(μ)
Membership Fuction(μ)

1
0.75 0.75

0.5 0.5

0.25 0.25

0 0
0 500 1000 1500 2000 2500 3000 3500 0 1000 2000 3000 4000 5000 6000 7000 8000
Na (mg/L) TDS (mg/L)

Desirable Acceptable Unacceptable

1
Membership Fuction(μ)

0.75
0.5
0.25
0
0 100 200 300 400 500 600 700 800
SO4 (mg/L)

Fig. 4. Membership functions defined for the input of Fuzzy models.

TDS ¼ 67.9 mg L1, then the FGQI (output) is 96.1%. Subsequently, season sample number 19, whose [Ca2þ], [Naþ] and TDS are
the results of all 49 samples were evaluated using the FGQI, FWQI ‘acceptable,’ is considered ‘acceptable’ by the FGQI. However, with
and FGWQI along with the crisp indices they originated from wet season sample number 33 differences in decision-making be-
(Tables S1 and S2 for dry and wet seasons, respectively). tween the two models are evident: all the inputs of sample 33 are
classified as ‘desirable’ but the values of GQI and FGQI are 89.7 and
96.2, respectively. The calculated value of the GQI is a border value
3.1. Results of the FGQI model
and as such, classified as ‘acceptable’. For marginal data and near
identical qualities, decision-making is a difficult task due to
The values of GQI and FGQI for the dry and wet seasons are
inherent uncertainties involved in the measurement and analysis
shown in Fig. 6a. As uncertainty in water quality parameters in the
steps. Therefore, both expert knowledge and the standard and
border area between linguistic parameters can play a key role in
permissible limits are used to provide a better decision on
misdiagnosing groundwater quality, the advantages of fuzzy
groundwater quality where the input parameter values fall in the
models is best revealed in samples with near identical qualities.
margin of the standard and permissible limits by using fuzzy
Due to similarly high concentrations of target quality parameters,
membership functions.
wet season samples 7, 13 and 30 are classified as ‘moderate’ in
With wet season sample 33 differences in decision-making
quality by both GQI and FGQI models. Sample 9 (wet season), all of
between the two models are evident: all inputs of sample 33 are
whose parameters are in the ‘desirable’ category, is classified as
classified as ‘desirable’ but values of GQI and FGQI are 89.7 (a
‘good’ according to both the GQI and FGQI. The water quality of wet
 M. Vadiati et al. / Journal of Environmental Management 184 (2016) 255e270 263

Completely Unacceptable Unacceptable season) is classified in the ‘good’ and ‘excellent’ categories by the
Moderate Acceptable WQI and FWQI models, respectively. In sample 18, all the input
Good parameters fall in the ‘desirable’ category except for [Ca2þ], [Mg2þ]
Membership Fuction (μ)

1 and TDS, which fall in the ‘acceptable’ category. The results of the
0.8
groundwater quality assessment in samples 44 and 45 (wet and
dry seasons, respectively) using the WQI and FWQI are identical:
0.6
groundwater quality is classified as ‘excellent,’ with all input pa-
0.4 rameters being placed in the ‘desirable’ category in the crisp
0.2 method.
0
0 10 20 30 40 50 60 70 80 90 100 3.3. Results of FGWQI model
FGQI
The values of GWQI and FGWQI for the dry and wet seasons are
Excelent Good shown in Fig. 6c. In sample 3 (wet season), all inputs of the FGWQI
Poor Very Poor model are classified in the ‘desirable’ category except for [Ca2þ]
Membership Fuction (μ)

Unfit for Drinking which is ‘acceptable.’ Regarding the calculated GWQI and devel-
1 oped FGWQI, sample 3 is classified in the ‘suitable’ and ‘high’ cat-
0.8 egories, respectively, while water quality based on the GQI and
FGQI, and the WQI and FWQI are in the ‘good’ and ‘excellent’ cat-
0.6
egories, respectively. These differences occurred in many samples,
0.4 demonstrating that the ranges of output categories for GWQI are
0.2 not as well defined as those of the crisp methods used in the study.
0 Although the FGWQI, which is capable of dealing with inherent
0 20 40 60 80 100 120 140 160 180 200 uncertainties in various steps of the approach, can be useful in
FWQI overcoming this deficiency, defining a specific range of GWQI re-
mains a challenge.
Low Acceptable High
Membership Fuction (μ)

3.4. Comparison of results from crisp and hybrid fuzzy indices

1 models
0.75
0.5
0.25
0 GQI and FGQI, in the ‘excellent’ category by the WQI and FWQI,
0 0.1 0.2 0.3
Fig. 5. Membership functions defined for the output of Fuzzy models.
border value and classified as ‘acceptable’) and 96.2 (‘good’),
respectively. When marginal data or near identical qualities are
used, decision-making is difficult due to inherent uncertainties
involved in the measurement and analysis steps. Therefore, where
input parameter values fall in the margin of standard and
permissible limits both expert knowledge and the prescribed
standard limits are used to provide a better decision on ground-
water quality by using fuzzy membership functions and if-then
rules prior to using a single value for standards limits. The pre-
dicted FGQI overcame the above deficiency by placing the sample
in the appropriate category by using the present data. Based on
expert judgment and the sensitivity of the model to border values,
the FGQI model clearly exhibited a better performance in decision-
making than the GQI, which supports the superiority of the fuzzy
model compared to the crisp model in groundwater quality
assessment.
3.2. Results of FWQI model
Fig. 6b shows the calculated values of the WQI and the pre-
dicted values of the FWQI for the dry and wet seasons. Samples 6
and 29 are classified in the ‘excellent’ category for the dry season
while with the crisp method, TDS and [Ca2þ] fall in the ‘unac-
ceptable’ and ‘acceptable’ category, respectively. Sample 18 (dry
The main aim of this research is to develop new hybrid fuzzy
indices (FGQI, FWQI and FGWQI) for groundwater quality assess-
ment based on widely accepted indices. For both dry and wet
seasons, most samples were classified in the ‘good’ category by the
0.4 0.5 0.6 0.7 0.8 0.9 1
and in the ‘high’ category by the GWQI and FGWQI. Numerical
FGWQI values of determinist indices and hybrid fuzzy indices (GQI and
FGQI, WQI and FWQI, and GWQI and FGWQI) involved in decision-
making in both dry and wet seasons are presented in Tables 2e4,
respectively.
The results of decision-making based on the FGQI model
showed the number of samples classified as ‘good,’ ‘acceptable’ and
‘moderate’ to be 26, 21 and 2 for the dry season. For the GQI model
the respective distribution of samples was 33, 14 and 2 for the dry
season while for the wet season, 31, 15, and 3 samples were clas-
sified as ‘good,’ ‘acceptable’ and ‘moderate’ quality, respectively for
the GQI. Based on the FWQI model (Table 3), for the dry season, 37,
10, 1, and 1 groundwater samples were classified as ‘excellent,’
‘good,’ ‘poor,’ and ‘very poor’ while in the wet season the number of
samples in each category was 31, 15, 2, and 1, respectively.
Comparatively, the WQI showed that in the dry season 37, 8, 3, and
1 groundwater samples were ‘excellent,’ ‘good,’ ‘poor,’ and ‘very
poor’ quality, respectively; the corresponding numbers in the wet
season were 32, 12, 4 and 1. Based on the FGWQI model, 46, 1, and 2
groundwater samples (dry season) and, 39, 6, and 4 samples (wet
season) were classified as ‘high,’ ‘suitable,’ and ‘low’, respectively
(Table 4). For the GWQI, 42, 5, and 5 samples (dry season), and 25,
20, and 4 (wet season) were classified as ‘high,’ ‘suitable,’ and ‘low,’
respectively.
Groundwater quality changes during the dry and wet seasons,
due to different meteorological, well exploration rates, and
hydrogeological factors. The physicochemical characteristics of the
groundwater reflect the geological formation of the study area and
the variation in groundwater quality provides key information on
the origin of the physicochemical parameters. Many of the major
264 M. Vadiati et al. / Journal of Environmental Management 184 (2016) 255e270

Fig. 6. The values of the deterministic and fuzzy indices for the dry and wet seasons of the Sarab Plain, Iran. GQI and FGQI (a, top panel), WQI and FWQI (b, middle panel), GWQI and
FGWQI (c, bottom panel).

ions in groundwater originate from water-rock interactions,
although NO 3 is a result of anthropogenic sources such as agri-
cultural and industrial activities (Yammani et al., 2008; Koh et al.,
2009; Güler et al., 2012).
Groundwater quality in the dry and wet seasons in the study
area is affected by irrigation return flows to groundwater with a
high concentration of solutes. Spatial distribution of the ground-
water quality assessment based on the FGQI, FWQI, and FGWQI
indices and the difference between dry and wet seasons are
shown in Fig. 7a, b and 7c, respectively. These figures indicate that
groundwater quality generally decreases toward the west. In
addition, Miocene formations containing gypsum and salts in the

Table 2
Results of groundwater quality assessment using GQI and FGQI models in the dry and wet seasons in the Sarab Plain, Iran.

Rank Dry season Wet season

GQI FGQI GQI FGQI

No. of samples Percent No. samples Percent No. samples Percent No. samples Percent

Good 33 67 26 53 31 63 28 57
Acceptable 14 29 21 43 15 31 18 37
Moderate 2 4 2 4 3 6 3 6
 M. Vadiati et al. / Journal of Environmental Management 184 (2016) 255e270 265

Table 3
Results of groundwater quality assessment using WQI and FWQI models in the dry and wet seasons in the Sarab Plain, Iran.

Rank Dry season Wet season

WQI FWQI WQI FWQI

No. of samples Percent No. samples Percent No. samples Percent No. samples Percent

Excellent 37 76 37 76 32 65 31 63
Good 8 16 10 20 12 24 15 31
Poor 3 6 1 2 4 8 2 4
Very Poor 1 2 1 2 1 2 1 2

Table 4
Results of groundwater quality assessment using GWQI and FGWQI models in the dry and wet seasons in the Sarab Plain, Iran.

Rank Dry season Wet season

GWQI FGWQI GWQI FGWQI

No. of samples Percent No. samples Percent No. samples Percent No. samples Percent

High 42 86 46 94 25 51 39 80
Suitable 5 10 1 2 20 41 6 12
Low 5 10 2 4 4 8 4 8

central part of the study area affect the deterioration of ground-
water quality. The dominant seasonal effects on groundwater
quality on the Sarab Plain are from agricultural activities and
meteorological factors (evaporation, precipitation, changes in the
dissolution rate and the dilution of contaminants). Samples 7, 13,
14, 29 and 30 revealed the greatest changes among the different
seasons and indices.
3.5. Evaluation of model performance
In order to assess the seasonal groundwater quality of the 49
samples, the GQI, WQI and GWQI indices were predicted using the
Mamdani fuzzy approach. The performance indices (VAF, RMSE and
R2) were calculated to evaluate the applicability of the proposed
fuzzy models. The comparison between calculated crisp indices and
predicted hybrid fuzzy indices are given in Table 5.
The respective VAF values for the FGQI, FWQI and FGWQI are
71, 76.4 and 75.4 percent in the dry season and 84.6, 79, and 80
percent in the wet season. Also, the respective RMSE values for the
FGQI, FWQI and FGWQI are 4.55, 9.22 and 0.08 in the dry season
and as 3.87, 9.47 and 0.09 in the wet season. R2 between the GQI
and FGQI in the dry and wet seasons are 0.73 and 0.85, respec-
tively. The values of R2 between WQI and FWQI and between
GWQI and FWGQI are 0.76 and 0.76 in the dry season and 0.79 and
0.80 in the wet season, respectively. The results showed good
accordance between the values of calculated and predicted indices
using fuzzy inference models. All the developed hybrid fuzzy in-
dex models were reliable in assessing groundwater quality. The
difference between indices criteria of the models in the dry and
wet seasons is a result of variations in groundwater quality and
values of input parameters. The input parameters are the same for
the FWQI, FGWQI, and FGQI models, while outputs were defined
based on the output ranges of the indices. Therefore, the differ-
ence between values of crisp and predicted GQI, WQI and GWQI
using the fuzzy model reflect the difference between output
ranges of the models. Previous studies of groundwater assessment
based on fuzzy logic did not use performance indices to evaluate
the results of models (e.g., Dahiya et al., 2007; Gharibi et al., 2012a;
Ocampo-Duque et al., 2013). Based on these results, it can be
concluded that the proposed Mamdani fuzzy models are good
alternative tools for minimizing the inherent uncertainties in
water quality assessment in comparison to crisp groundwater
quality indices.
3.6. Sensitivity analysis
Validation of the developed indeces, parameters, and rule sets is
achievable by addressing the robustness of the indices using a
sensitivity analysis since the FIS models are sensitive in developing
membership functions, fuzzy rules and output results. The cosine
amplitude method (CAM) is a well-known and widely used
approach, based on the fuzzy relation concept, and is used to obtain
similarity relation values between the relevant parameters (Ross,
2012; Monjezi and Rezaei, 2011; Ghasemi et al., 2014). The CAM
is a useful method for sensitivity analysis to determine the strength
of relation values between the input and output parameters and is
used in the present study to find the most sensitive input param-
eters for the FGQI, FGWQI and FWQI indices.
In the CAM method, all relevant data are expressed in an X-
space. The data pairs are collected in order to construct a data array,
X (Ross, 2012):
X ¼ fX1 ; X2 ; X3 ; …; Xm g (4)
Each of the elements, Xi, in the data array X is a vector of length:
Xi ¼ fXi1 ; Xi2 ; Xi3 ; …; Xim g1 (5)
The similarity relations (rij) between output and input param-
eters are computed as:
Pk¼m
k¼1 Xik Xjk
rij ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pk¼m 2 Pk¼m 2 (6)
k¼1 X ik k¼1 X jk
According to this equation, values of rij close to 0 indicate the
dissimilarity, while values of rij close to 1 indicate the similarity of
two data sets (Ross, 2012). The similarity relations (rij) between the
input parameters and FGQI, FGWQI and FWQI (Fig. 8) indicate that
[Ca2þ] and [Mg2þ] are the most significant parameters controlling
the outputs, whereas [Naþ] is the least sensitive parameter in the
hybrid fuzzy indices.
By changing the rule set numbers, rule data-based validation is
achievable for addressing the sensitivity of the FGQI, FWQI, and
266 M. Vadiati et al. / Journal of Environmental Management 184 (2016) 255e270

Fig. 7. Groundwater quality assessment based on the developed fuzzy indices in the dry and wet seasons (Sarab Plain, Iran). FGQI (a, top panel), FWQI (b, middle panel), FGWQI (c,
bottom panel).
 M. Vadiati et al. / Journal of Environmental Management 184 (2016) 255e270 267

Table 5 than traditional indices. The developed models' accuracy was

Performance indices between calculated crisp indices and predicated hybrid fuzzy
indices.
Model Dry season Wet season
VAF (percent) RMSE R2 VAF (percent) RMSE
FGQI 71 4.55 0.73 84.6 3.87
FWQI 76.4 9.22 0.76 79 9.47
FGWQI 75.4 0.08 0.76 80 0.09
FGWQI. Gharibi et al. (2012a, b) evaluated the effect of changes in
the rule set on the output of indices. Following this approach, the
developed rule set for FGQI, FWQI, and FGWQI were chosen for the
sensitivity analysis in this study.
After the approval of numerous experts (the selected experts
were discussed earlier in the paper), the final agreed rule set of
every index (68 rules for every FGQI, FWQI and FGWQI) was
considered as a baseline. Following this, deviations in the number
of rules were intentionally made from the baseline to evaluate the
changes in the correlation coefficients for the FGQI, FGWQI and
FWQI indices. The results of the sensitivity analysis of the different
rule numbers in the dry and wet season are shown in Fig. 9. This
figure suggests that the FGQI, FGWQI and FWQ outputs were not
significantly affected by changes in the number of rules.
4. Conclusion
When faced with marginal data and near identical qualities, due
to inevitable errors in measurements and inherent uncertainties,
decision-making is a difficult task. A fuzzy-logic-based decision-
making approach can collect expert knowledge and deal with un-
certainties by introducing a threshold domain instead of a single
value for WQI standards. Therefore, the FIS can play a key role in
correctly diagnosing groundwater quality in water quality param-
eters in the border area between the developed linguistic terms (for
example: the difference in decision-making in sample 33 for the
wet season based on GQI and FGQI is 89.7, ‘acceptable’, and 96.2,
‘good’, respectively). Consequently, both expert knowledge and
permissible limits are involved to provide a better decision
regarding groundwater quality assessment by using fuzzy mem-
bership functions. Given the potential of fuzzy models to minimize
uncertainties, it was found that hybrid fuzzy-based indices produce
significantly more accurate assessments of groundwater quality
assessed and a comparison of the performance indices was pro-
vided to confirm that the FGQI model is more accurate than both
FWQI and FGWQI models. It was found that the new hybrid fuzzy
R2 indices developed in this study are reliable for groundwater quality
assessment.
0.85
0.79 The Mamdani method, a common fuzzy FIS model, was devel-
0.80 oped to predict groundwater quality based on widely accepted
indices. Membership functions and rules were employed based on
the judgment of various groundwater quality experts to introduce a
generalized model that can be used for varied data. Since the FIS
models are sensitive to membership functions and developed rules,
the application of WHO standards and WQIs helped decrease the
uncertainty in the models compared to those based solely on expert
judgement. The robustness of the developed fuzzy model was
shown in samples with identical qualities, especially around border
values between linguistic terms which are affected by uncertainties
and consequently impact water quality evaluation. For an assess-
ment of the flexibility of the developed hybrid fuzzy indices, the
results of the models were compared in the dry and wet seasons.
The VAF, RMSE and R2 performance indices for the FGQI in the wet
season showed close accordance between calculated and predicted
indices using fuzzy inference models vs. the other developed
models. The most important parameters for the FGQI, FGWQI and
FWQI models were [Ca2þ] and [Mg2þ], while [Naþ] was the least
sensitive in the hybrid fuzzy indices. Results of the sensitivity
analysis showed that the developed indices are robust against
changes in the number of rules. The Mamdani fuzzy-logic-based
decision-making model is a reliable approach for groundwater
quality assessment, and the combination of the Mamdani fuzzy
inference model with suitable WQIs for groundwater, as imple-
mented in the present study, provide more reliable results
compared to conventional WQIs and the simple fuzzy inference
methods.
The development of new hybrid fuzzy indices using accept-
ability ranges of well-known indices for output ranges of fuzzy
models is an alternative approach for groundwater quality
assessment. Although results of the FIS model are promising,
further research is necessary to investigate the role of the various
indices, and the different standards and permissible limits pre-
scribed by different organizations. Further studies can compare
fuzzy inference systems with other AI methods and their appli-
cation in industrial and agricultural contexts. The parameter se-
lection for groundwater quality assessment depends upon the

0.7
FGWQI FWQI FGQI
0.6

0.5

0.4

0.3

0.2

0.1

0
Calcium Magnesium Sodium Chloride Nitrate Sulfate TDS

Fig. 8. Strength of relation (rij) between the input parameters and FGQI, FGWQI and FWQI indices.
268 M. Vadiati et al. / Journal of Environmental Management 184 (2016) 255e270
0.9
0.8
A FGQI FWQI FGWQI
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0 0
34 51 68 85 102
Fig. 9. Results of the changes in the correlation coefficients for FGQI, FGWQI and FWQI indices in the dry (A) and wet (B) seasons in the different rule numbers.
characteristics of the study area. The appropriate parameters in
this study were selected based on the most suitable parameters
that can address the local features of the environment in the study
area (i.e. agricultural development and not industrial develop-
ment). However, the integration and comparison of other pa-
rameters in a FIS model could also be explored, but this was
outside the scope of the present research. The main objective of
this study was to develop a generalized rule-based fuzzy model.
Moreover, the present study focused on well-accepted parameters
to compare the applicability of the developed fuzzy WQIs with
deterministic WQIs. Therefore, future research is necessary to
investigate the applicability of the developed fuzzy WQIs and the
integration of various additional parameters such as physi-
ochemical, organic matter, microbiological, major anions, major
cations and heavy metals.
Acknowledgements
The Iranian Ministry of Science, Research, and Technology pro-
vided a scholarship to M. Vadiati to conduct this research at McGill
University under the supervision of Professor J. Adamowski. M.
Vadiati, A. Asghari-Moghaddam, and M. Nakhaei acknowledge the
East Azerbaijan Regional Water Authority for providing part of the
data. Partial funding was provided by an NSERC Discovery Grant
held by J. Adamowski.
Appendix A. Supplementary data
Supplementary data related to this article can be found at http://
dx.doi.org/10.1016/j.jenvman.2016.09.082.
References
Abbasi, T., Abbasi, S.A., 2012. Water Quality Indices. Elsevier, Amsterdam.
Abtahi, M., Golchinpour, N., Yaghmaeian, K., Rafiee, M., Jahangiri-rad, M., Keyani, A.,
Saeedi, R., 2015. A modified drinking water quality index (DWQI) for assessing
drinking source water quality in rural communities of Khuzestan Province. Iran.
Ecol. Indic. 53, 283e291. http://dx.doi.org/10.1016/j.ecolind.2015.02.009.
Al-Mutairi, N., Abahussain, A., El-Battay, A., 2014. Spatial and temporal character-
izations of water quality in Kuwait Bay. Mar. Pollut. Bull. 83 (1), 127e131. http://
dx.doi.org/10.1016/j.marpolbul.2014.04.009.
American Public Health Association, American Water Works Association, Water
Environment Federation, APHA, AWWA, WEF, 1998. In: Clesceri, L.S., Eaton, A.D.,
Greenberg, A.E. (Eds.), Standard Methods for the Examination of Water and
Wastewater, twentieth ed. APHA-AWWA-WEF, Washington, DC.
Aminu, M., Matori, A.N., Yusof, K.W., Malakahmad, A., Zainol, R.B., 2014. A GIS-based
water quality model for sustainable tourism planning of Bertam River in
Cameron Highlands, Malaysia. Environ. Earth Sci. 73 (10), 6525e6537. http://
dx.doi.org/10.1007/s12665-014-3873-6.
Babiker, I.S., Hohamed, M.A.A., Hiyama, T., 2007. Assessing groundwater quality
using GIS. Water Resour. Manage 21 (4), 699e715. http://dx.doi.org/10.1007/
s11269-006-9059-6.
0.9
0.8
B FGQI FWQI FGWQI
0.7
0.6
0.5
0.4
0.3
0.2
0.1
34 51 68 85 102
Bain, R.E.S., Wright, J.A., Christenson, E., Bartram, J.K., 2014. Rural: urban in-
equalities in post 2015 targets and indicators for drinking-water. Sci. Total
Environ. 490, 509e513. http://dx.doi.org/10.1016/j.scitotenv.2014.05.007.
Banerjee, T., Srivastava, R.K., 2011. Evaluation of environmental impacts of Inte-
grated Industrial EstatedPantnagar through application of air and water
quality indices. Environ. Monit. Assess. 172 (1e4), 547e560. http://dx.doi.org/
10.1007/s10661-010-1353-3.
Banoeng-Yakubo, B., Yidana, S.M., Emmanuel, N., Akabzaa, T., Asiedu, D., 2009.
Analysis of groundwater quality using water quality index and conventional
graphical methods: the Volta region, Ghana. Environ. Earth Sci. 59, 867e879.
http://dx.doi.org/10.1007/s12665-009-0082-9.
Barua, A., Mudunuri, L.S., Kosheleva, O., 2014. Why Trapezoidal and Triangular
Membership Functions Work So Well: Towards a Theoretical Explanation.
Departmental Technical Reports (CS). Paper 783. http://digitalcommons.utep.
edu/cs_techrep/783.
Bolton, P.W., Currie, J.C., Tervet, D.J., Welsh, W.T., 1978. Index to improve water
quality classification. Water Pollut. control 77, 271e284.
Brown, R.M., McClelland, N.I., Deininger, R.A., Tozer, R.G., 1970. Water quality index
d do we dare? Water Sew. Works 117 (10), 339e343.
Canadian Council of Ministers of the Environment (CCME), 2001. Canadian Water
Quality Index 1.0. Technical Report and User's Manual. Canadian Environmental
Quality Guidelines Water Quality Index Technical Subcommittee, Gatineau, QC,
Canada. Seen 23 March 2016 at. http://www.ccme.ca/files/Resources/
calculators/WQI%20User's%20Manual%20(en).pdf.
Chapman, D. (Ed.), 1996. Water Quality Assessments - a Guide to Use of Biota,
Sediments and Water in Environmental Monitoring, second ed. E&FN Spon,
London. Seen 23 March 2016 at. http://www.who.int/water_sanitation_health/
resourcesquality/wqa/en/.
Chaturvedi, M.K., Bassin, J.K., 2010. Assessing the water quality index of water
treatment plant and bore wells, in Delhi, India. Environ. Monit. Assess. 163
(1e4), 449e453. http://dx.doi.org/10.1007/s10661-009-0848-2.
Chau, K.-W., 2006. A review on integration of artificial intelligence into water
quality modelling. Mar. Pollut. Bull. 52 (7), 726e733. http://dx.doi.org/10.1016/
j.marpolbul.2006.04.003.
Cordoba, E.B., Martinez, A.C., Ferrer, E.V., 2010. Water quality indicators: compari-
son of a probabilistic index and a general quality index. The case of the Con-
federacion Hidrografica del Júcar (Spain). Ecol. Indic. 10 (5), 1049e1054. http://
dx.doi.org/10.1016/j.ecolind.2010.01.013.
Dahiya, S., Singh, B., Gaur, S., Garg, V.K., Kushwaha, H.S., 2007. Analysis of
groundwater quality using fuzzy synthetic evaluation. J. Hazard. Mater 147 (3),
938e946. http://dx.doi.org/10.1016/j.jhazmat.2007.01.119.
Dos Santos Simoes, F., Moreira, A.B., Bisinoti, M.C., Gimenez, S.M.N., Yabe, M.J.S.,
2008. Water quality index as a simple indicator of aquaculture effects on
aquatic bodies. Ecol. Indic. 8 (5), 476e484. http://dx.doi.org/10.1016/
j.ecolind.2007.05.002.
El-Fadel, M., Tomaszkiewicz, M., Adra, Y., Sadek, S., Najm, M.A., 2014. GIS-based
assessment for the development of a groundwater quality index towards sus-
tainable aquifer management. Water Resour. Manage 28, 3471e3487. http://
dx.doi.org/10.1007/s11269-014-0683-2.
Gazzaz, N.M., Yusoff, M.K., Aris, A.Z., Juahir, H., Ramli, M.F., 2012. Artificial neural
network modeling of the water quality index for Kinta River (Malaysia) using
water quality variables as predictors. Mar. Pollut. Bull. 64 (11), 2409e2420.
http://dx.doi.org/10.1016/j.marpolbul.2012.08.005.
Gharibi, H., Mahvi, A.H., Nabizadeh, R., Arabalibeik, H., Yunesian, M., Sowlat, M.H.,
2012a. A novel approach in water quality assessment based on fuzzy logic.
J. Environ. Manage 112, 87e95. http://dx.doi.org/10.1016/
j.jenvman.2012.07.007.
Gharibi, H., Sowlat, M.H., Mahvi, A.H., Mahmoudzadeh, H., Arabalibeik, H.,
Keshavarz, M., Karimzadeh, N., Hassani, G., 2012b. Development of a dairy cattle
drinking water quality index (DCWQI) based on fuzzy inference systems. Ecol.
Indic. 20, 228e237. http://dx.doi.org/10.1016/j.ecolind.2012.02.015.
Ghasemi, E., Ataei, M., Shahriar, K., 2014. An intelligent approach to predict pillar
 M. Vadiati et al. / Journal of Environmental Management 184 (2016) 255e270
sizing in designing room and pillar coal mines. Int. J. Rock Mech. Min. Sci. 65,
86e95. http://dx.doi.org/10.1016/j.ijrmms.2013.11.009.
Giri, S., Singh, G., Gupta, S.K., Jha, V.N., Tripathi, R.M., 2010. An evaluation of metal
contamination in surface and groundwater around a proposed Uranium mining
site, Jharkhand, India. Mine Water Environ. 29 (3), 225e234. http://dx.doi.org/
10.1007/s10230-010-0107-3.
Girman, J., Van der Gun, J., Haie, N., Hirata, R., Lipponen, A., Lopez-Gunn, E.,
Neupane, B., Shah, T., Vrba, J., Wallin, B., 2007. Groundwater Resources Sus-
tainability Indicators, IHP-VI. Series on Groundwater No. 14. UNESCO, Paris.
Seen 23 March 2016 at. http://unesdoc.unesco.org/images/0014/001497/
149754e.pdf.
Güler, C., Kurt, M.A., Alpaslan, M., Akbulut, C., 2012. Assessment of the impact of
anthropogenic activities on the groundwater hydrology and chemistry in Tarsus
coastal plain (Mersin, SE Turkey) using fuzzy clustering, multivariate statistics
and GIS techniques. J. Hydrol. 414e415, 435e451. http://dx.doi.org/10.1016/
j.jhydrol.2011.11.021.
Hellendoorn, H., Thomas, C., 1993. Defuzzification in fuzzy controllers. J. Intell.
Fuzzy Syst. 1 (2), 109e123. http://dx.doi.org/10.3233/IFS-1993-1202.
Horton, R.K., 1965. An index number system for rating water quality. J. Water Pollut.
Control Fed. 37 (3), 300e306.
Jiang, Y., Wu, Y., Groves, C., Yuan, D., Kambesis, P., 2009. Natural and anthropogenic
factors affecting the groundwater quality in the Nandong karst underground
river system in Yunan, China. J. Contam. Hydrol. 109 (1e4), 49e61. http://
dx.doi.org/10.1016/j.jconhyd.2009.08.001.
Jinturkar, A., Deshmukh, S., Agarkar, S., Chavhan, G., 2010. Determination of water
quality index by fuzzy logic approach: a case of ground water in an Indian town.
Water Sci. Technol. 61 (8), 1987e1994. http://dx.doi.org/10.2166/wst.2010.095.
Ketata-Rokbani, M., Gueddari, M., Bouhlila, R., 2011. Use of geographical informa-
tion system and water quality index to assess groundwater quality in El Khairat
deep aquifer (enfidha, tunisian sahel). Iran. J. Energy Environ. 2 (2), 133e144.
Seen 23 March 2016 at. http://www.ijee.net/Journal/ijee/vol2/no2/5.pdf.
Klir, G., Yuan, B., 1995. Fuzzy Sets and Fuzzy Logic, fourth ed. Prentice Hall, Upper
saddle River, NJ.
Koh, D.-C., Chae, G.-T., Yoon, Y.-Y., Kang, B.-R., Koh, G.-W., Park, K.-H., 2009. Baseline
geochemical characteristics of groundwater in the mountainous area of Jeju
Island, South Korea: implications for degree of mineralization and nitrate
contamination. J. Hydrol. 376 (1e2), 81e93. http://dx.doi.org/10.1016/
j.jhydrol.2009.07.016.
Kumar, N.V., Mathew, S., Swaminathan, G., 2009. Fuzzy information processing for
assessment of groundwater quality. Int. J. Soft Comput. 4 (1), 1e9.
Kusko, B., 1993. Fuzzy Thinking: the New Science of Fuzzy Logic. Hyperion, New
York.
Lermontov, A., Yokoyama, L., Lermontov, M., Machado, M.A.S., 2009. River quality
analysis using fuzzy water quality index: Ribeira do Iguape river watershed,
Brazil. Ecol. Indic. 9 (6), 1188e1197. http://dx.doi.org/10.1016/
j.ecolind.2009.02.006.
Liou, S.-M., Lo, S.-L., Hu, C.-Y., 2003. Application of two-stage fuzzy set theory to
river quality evaluation in Taiwan. Water Res. 37 (6), 1406e1416. http://
dx.doi.org/10.1016/S0043-1354(02)00479-7.
Liou, S.-M., Lo, S.-L., Wang, S.-H., 2004. A generalized water quality index for
Taiwan. Environ. Monit. Assess. 96 (1e3), 35e52. http://dx.doi.org/10.1023/B:
EMAS.0000031715.83752.a1.
Lumb, A., Sharma, T.C., Bibeault, J.F., 2011. A review of genesis and evolution of
water quality index (WQI) and some future directions. Water Qual. Expo. Health
3 (1), 11e24. http://dx.doi.org/10.1007/s12403-011-0040-0.
Maiti, S., Erram, V.C., Gupta, G., Tiwari, R.K., Kulkarni, U.D., Sangpal, R.R., 2013.
Assessment of groundwater quality: a fusion of geochemical and geophysical
information via Bayesian neural networks. Environ. Monit. Assess. 185 (4),
3445e3465. http://dx.doi.org/10.1007/s10661-012-2802-y.
Mamdani, E.H., Assilian, S., 1975. An experiment in linguistic synthesis with a fuzzy
logic controller. Int. J. Man. Mach. Stud. 7 (1), 1e13. http://dx.doi.org/10.1016/
S0020-7373(75)80002-2.
Mathworks, 2014. MATLAB and Fuzzy Logic Toolbox. Release 2014a. MathWorks,
Natick, Massachusetts.
Mohebbi, M.R., Saeedi, R., Montazeri, A., Vaghefi, K.A., Labbafi, S., Oktaie, S.,
Abtahi, M., Mohagheghian, A., 2013. Assessment of water quality in ground-
water resources of Iran using a modified drinking water quality index (DWQI).
Ecol. Indic. 30, 28e34. http://dx.doi.org/10.1016/j.ecolind.2013.02.008.
Monjezi, M., Rezaei, M., 2011. Developing a new fuzzy model to predict burden from
rock geomechanical properties. Expert Syst. Appl. 38 (8), 9266e9273. http://
dx.doi.org/10.1016/j.eswa.2011.01.029.
Nasiri, F., Maqsood, I., Huang, G., Fuller, N., 2007. Water quality index: a fuzzy river-
pollution decision support expert system. J. Water Resour. Plann. Manag. 133
(2), 95e105. http://dx.doi.org/10.1061/(ASCE)0733-9496 (2007)133:2(95).
Nazeer, S., Hashmi, M.Z., Malik, R.N., 2014. Heavy metals distribution, risk assess-
ment and water quality characterization by water quality index of the River
Soan. Pak. Ecol. Indic. 43, 262e270. http://dx.doi.org/10.1016/
j.ecolind.2014.03.010.
Nikoo, M.R., Kerachian, R., Malakpour-Estalaki, S., Bashi-Azghadi, S.N., Azimi-
Ghadikolaee, M.M., 2011. A probabilistic water quality index for river water
quality assessment: a case study. Environ. Monit. Assess. 181 (1e4), 465e478.
http://dx.doi.org/10.1007/s10661-010-1842-4.
Ocampo-Duque, W., Ferre -Huguet, N., Domingo, J.L., Schuhmacher, M., 2006.
Assessing water quality in rivers with fuzzy inference systems: a case study.
Environ. Int. 32 (6), 733e742. http://dx.doi.org/10.1016/j.envint.2006.03.009.
269
Ocampo-Duque, W., Osorio, C., Piamba, C., Schuhmacher, M., Domingo, J.L., 2013.
Water quality analysis in rivers with non-parametric probability distributions
and fuzzy inference systems: application to the Cauca River, Colombia. Environ.
Int. 52, 17e28. http://dx.doi.org/10.1016/j.envint.2012.11.007.
Patki, V.K., Shrihari, S., Manu, B., Deka, P.C., 2015. Fuzzy system modeling for
forecasting water quality index in municipal distribution system. Urban Water J.
12 (2), 89e110. http://dx.doi.org/10.1080/1573062X.2013.820333.
Ramachandramoorthy, T., Sivasankar, V., Subramanian, V., 2010. The seasonal status
of chemical parameters in shallow coastal aquifers of Rameswaram Island, In-
dia. Environ. Monit. Assess. 160 (1e4), 127e139. http://dx.doi.org/10.1007/
s10661-008-0663-1.
Rezaei, M., Asadizadeh, M., Majdi, A., Hossaini, M.F., 2015. Prediction of represen-
tative deformation modulus of longwall panel roof rock strata using Mamdani
fuzzy system. Int. J. Min. Sci. Technol. 25 (1), 23e30. http://dx.doi.org/10.1016/
j.ijmst.2014.11.007.
Ross, T.J., 2012. Fuzzy Logic with Engineering Applications, third ed. Wiley, New
York.
Sadat-Noori, S.M., Ebrahimi, K., Liaghat, A.M., 2014. Groundwater quality assess-
ment using the water quality index and GIS in saveh-nobaran aquifer. Iran.
Environ. Earth Sci. 71 (9), 3827e3843. http://dx.doi.org/10.1007/s12665-013-
2770-8.
Saeedi, M., Abessi, O., Sharifi, F., Meraji, H., 2010. Development of groundwater
quality index. Environ. Monit. Assess. 163 (1e4), 327e335. http://dx.doi.org/
10.1007/s10661-009-0837-5.
Said, A., Stevens, D.K., Sehlke, G., 2004. An innovative index for evaluating water
quality in streams. J. Environ. Manage 34 (3), 406e414. http://dx.doi.org/
10.1007/s00267-004-0210-y.
Selvam, S., Manimaran, G., Sivasubramanian, P., Balasubramanian, N.,
Seshunarayana, T., 2014. GIS-based Evaluation of Water Quality Index of
groundwater resources around Tuticorin coastal city, south India. Environ. Earth
Sci. 71 (6), 2847e2867. http://dx.doi.org/10.1007/s12665-013-2662-y.
Shams, S., Monjezi, M., Majd, V.J., Armaghani, D.J., 2015. Application of fuzzy
inference system for prediction of rock fragmentation induced by blasting.
Arab. J. Geosci. 8 (12), 10819e10832. http://dx.doi.org/10.1007/s12517-015-
1952-y.
Shi, J., Ma, R., Liu, J., Zhang, Y., 2013. Suitability assessment of deep groundwater for
drinking, irrigation and industrial purposes in Jiaozuo City, Henan Province,
north China. Chin. Sci. Bull. 58 (25), 3098e3110. http://dx.doi.org/10.1007/
s11434-013-5952-6.
Sii, H.I., Sherrard, J.H., Wilson, T.E., 1993. A water quality index based on fuzzy sets
theory. In: Proceedings of the 1993 Joint ASCEeCSCE National Conference on
Environmental Engineering, July 12e14, Montreal, Quebec, Canada,
pp. 253e259.
Sivanandam, S.N., Sumathi, S., Deepa, S.N., 2007. Introduction to Fuzzy Logic Using
MATLAB, first ed. Springer, Berlin.
Soltan, M.E., 1999. Evaluation of ground water quality in Dakhla oasis (egyptian
Western desert). Environ. Monit. Assess. 57 (2), 157e168. http://dx.doi.org/
10.1023/A: 1005948930316.
Srivastava, P.K., Mukherjee, S., Gupta, M., Singh, S.K., 2011. Characterizing
monsoonal variation on water quality index of River Mahi in India using
geographical information system. Water Qual. Expo. Health 2 (3e4), 193e203.
http://dx.doi.org/10.1007/s12403-011-0038-7.
Stigter, T.Y., Ribeiro, L., Dill, A.C., 2006. Evaluation of an intrinsic and a specific
vulnerability assessment method in comparison with groundwater salinisation
and nitrate contamination levels in two agricultural regions in the south of
Portugal. Hydrogeol. J. 14 (1), 79e99. http://dx.doi.org/10.1007/s10040-004-
0396-3.
Tiwari, T.N., Mishra, M., 1985. A preliminary assignment of water quality index of
major Indian rivers. Indian J. Environ. Prot. 5 (4), 276e279.
Vasanthavigar, M., Srinivasamoorthy, K., Vijayaragavan, K., Ganthi, R.R.,
Chidambaram, S., Anandhan, P., Manivannan, R., Vasudevan, S., 2010. Applica-
tion of water quality index for groundwater quality assessment: thir-
umanimuttar sub-basin, Tamilnadu, India. Environ. Monit. Assess. 171 (1e4),
595e609. http://dx.doi.org/10.1007/s10661-009-1302-1.
Vicente, J., Colmenarejo, M.F., Sa nchez, E., Rubio, A., García, M.G., Borja, R.,
Jimenez, A.M., 2011. Evaluation of the water quality in the Guadarrama river at
the section of Las Rozas-Madrid. Spain. Water Environ. J. 25 (1), 55e66. http://
dx.doi.org/10.1111/j.1747-6593.2009.00188.x.
Wang, W.C., Xu, D.M., Chau, K.W., Lei, G.J., 2014. Assessment of river water quality
based on theory of variable fuzzy sets and fuzzy binary comparison method.
Water Resour. Manag. 28 (12), 4183e4200. http://dx.doi.org/10.1007/s11269-
014-0738-4.
Wang, X., Ruan, D., Kerre, E.E., 2009. Mathematics of Fuzzinessdbasic Issues.
Studies in Fuzziness and Soft Computing, vol. 245. Springer, Berlin/Heidelberg.
http://dx.doi.org/10.1007/978-3-540-78311-4.
World Health Organization (WHO), 2004. Guidelines for Drinking-water Quality,
second ed., vol. 2. Health Criteria and Other Supporting Information. WHO,
Geneva, pp. 231e233. Seen 23 March 2016 at. http://www.who.int/water_
sanitation_health/dwq/gdwq2v1/en/index1.html.
Yammani, S.R., Reddy, T.V.K., Reddy, M.R.K., 2008. Identification of influencing
factors for groundwater quality variation using multivariate analysis. Environ.
Geol. 55 (1), 9e16. http://dx.doi.org/10.1007/s00254-007-0958-5.
Yen, J., Langari, R., 1998. Fuzzy Logic: Intelligence, Control, and Information, first ed.
Prentice Hall, Upper saddle River, NJ.
Yen, J., Wang, L., 1999. Simplifying fuzzy rule-based models using orthogonal
270 M. Vadiati et al. / Journal of Environmental Management 184 (2016) 255e270

transformation methods. IEEE Trans. Syst. Man. Cybern. 29 (1), 13e24.
Yilmaz, I., Marschalko, M., Bednarik, M., Kaynar, O., Fojtova, L., 2012. Neural
computing models for prediction of permeability coefficient of coarse-grained
soils. Neural Comput. Appl. 21 (5), 957e968. http://dx.doi.org/10.1007/
s00521-011-0535-4.
Zadeh, L.A., 1965. Fuzzy sets. Inf. control 8 (3), 338e353. http://dx.doi.org/10.1016/
S0019-9958(65)90241-X.
Zadeh, L.A., 1973. Outline of a new approach to the analysis of complex systems and
decision processes. IEEE Trans. Syst. Man. Cybern. 1, 28e44. http://dx.doi.org/
10.1109/TSMC.1973.5408575.
Zhao, P., Tang, X., Tang, J., Wang, C., 2013. Assessing water quality of three gorges
reservoir, China, over a five-year period from 2006 to 2011. Water Resour.
Manag. 27 (13), 4545e4558. http://dx.doi.org/10.1007/s11269-013-0425-x.