Groundwater for Sustainable Development 10 (2020) 100350

Contents lists available at ScienceDirect

Groundwater for Sustainable Development

journal homepage: http://www.elsevier.com/locate/gsd

Research paper

Development of water quality index for Godavari River (India) based on

fuzzy inference system
Jyotiprakash G. Nayak a, *, L.G. Patil b, Vinayak K. Patki c
a
Department of Civil Engineering, Sandip Institute of Technology and Research Centre, Nashik, Maharashtra, India
b
Department of Civil Engineering, Shri Guru Gobind Singhji Institute of Engineering and Technology, Nanded, Maharashtra, India
c
Department of Civil Engineering, Nagesh Karajgi Orchid College of Engineering and Technology, Solapur, Maharashtra, India

A R T I C L E I N F O A B S T R A C T

Keywords: In view of higher pollution strength of Indian rivers, prevalent water quality indices of the western countries like
Fuzzy inference system the National sanitation foundation water quality index (NSFWQI) and indigenous Vedprakash water quality
Fuzzy water quality index index (VWQI) cannot truly represent the water quality status of Indian rivers. To overcome this limitation, fuzzy
Membership function
modeling has been used in this study for the prediction of water quality of Indian rivers. The fuzzy models have
National sanitation foundation water quality
been developed using triangular and trapezoidal membership functions with centroid, bisector and mean of
index
Water quality index maxima (MOM) methods for defuzzification. It is observed that the fuzzy model with triangular membership
function utilizing the bisector method of defuzzification performs better, compared to triangular and trapezoidal
membership function utilizing the centroid and MOM method of defuzzification. The values of water quality
index based on fuzzy logic have been compared with the NSFWQI and VWQI. It is observed that the values of
fuzzy based water quality index are more representative to actual river water quality status of Indian rivers as
compared to NSFWQI and VWQI. This is due to the fact that the adopted fuzzy logic approach is equally sensitive
to all parameters and can truly represent the minor change in the value of any parameter, especially in case of
river stretches having higher pollution.

be same.
1. Introduction The subsequent step is establishing weights to the parameters. Equal
or unequal weights have been assigned to the parameters by the index
To maintain human health and to secure the environment, evalua­ developers, as per their judgment. Some researchers claimed that all
tion of water quality is indispensable. Water quality index (i.e WQI) is a parameters are equally important and assigned equal weights to all the
single dimensionless term, by which the water quality of a stream can be parameters, during the development of WQIs (e.g. Cude, 2001; CCME,
expressed as good, medium or bad. The water quality of different regions 2001). The index may suffer from ‘Sensitivity’ issue, if the different
can easily be compared by comparing the values of WQI, instead of weights are assigned to the parameters. Equation of the final index is
comparing the numerical values of several water quality parameters. To decided on aggregation of sub-indices and considering the weights
evolve a WQI, four steps have been utilised in the past (Abbasi and assigned to the different parameters. Additive and multiplicative
Abbasi, 2011); i) Choice of parameters ii) Determination of values of methods have been used for aggregation of sub-indices. Most of the
sub-index iii) Ascertaining appropriate weights iv) Establishing the final existing WQIs have used additive method (e.g. Brown et al., 1970;
index after compounding the sub-indices. While establishing a WQI, the Sargaonkar and Deshpande, 2003) but the additive method suffers from
selection of water quality parameters is done based on, a) Referring the ‘eclipsing’ issue (Swamee and Tyagi, 2007). While, some other re­
work of previous researchers (Said et al., 2004) b) Availability of the searchers used multiplicative method for the development of the WQIs
data sets (Cude, 2001) c) True representation of the quality scenario (e.g. Bhargava, 1985). It was reported that the multiplicative approach
(Hanh et al., 2011) d) The proposed use of stream for a specific purpose can not completely resolve the ‘eclipsing’ problem and also leads to
(Hurley et al., 2012). The subsequent step is the development of ‘ambiguity’ issue sometimes (Swamee and Tyagi, 2007). NSFWQI was
sub-indices, which is done to convert the selected parameters onto a developed by Brown et al. (1970) and is used worldwide, including the
common platform, since the units of all the selected parameters cannot Indian subcontinent to determine the water quality of the rivers

* Corresponding author.
E-mail addresses: [email protected] (J.G. Nayak), [email protected] (L.G. Patil), [email protected] (V.K. Patki).

https://doi.org/10.1016/j.gsd.2020.100350
Received 21 December 2019; Received in revised form 8 February 2020; Accepted 11 February 2020
Available online 15 February 2020
2352-801X/© 2020 Elsevier B.V. All rights reserved.
J.G. Nayak et al. Groundwater for Sustainable Development 10 (2020) 100350

(Sutadian and Yilmaz, 2016). Considering the present scenario of Indian

Nomenclature rivers, this index can no longer be used in the Indian subcontinent
because most of the Indian rivers carry high pollution potential in the
DO Dissolved Oxygen form of high BOD5 and total solids content. The use of NSFWQI becomes
BOD Biological Oxygen Demand (over 5 days of incubation at deficient in such cases, since its Q-value curves do not take into
20 � C) consideration high BOD5 and total solids content of natural streams
WQI Water Quality Index (Nayak et al., 2017).
NSFWQI National Sanitation Foundation Water Quality Index Godavari River is one of the sacred river of India and Kumbhmela is
VWQI Vedprakash Water Quality Index celebrated on the banks of the Godavari in and around Nashik every
FWQI Fuzzy Water Quality Index twelve years. Thousands of people take a holy dip in Godavari during
FIS Fuzzy Inference System Kumbh and other festivals, hence it is the need of the hour to determine
MF Membership Function water quality of this river in and around Nashik. In view of the limita­
MOM Mean of Maxima tions of the prevalent WQIs viz. ‘eclipsing’, ‘ambiguity’ and inability to
be representative on downstream side of a wastewater treatment plant,
it is observed that these existing WQIs cannot represent the actual

Fig. 1. Location map of study area.

Fig. 2. Location map of water quality monitoring stations and wastewater addition points in Godavari river at Nashik.

2
J.G. Nayak et al. Groundwater for Sustainable Development 10 (2020) 100350

Table 1 Table 2
NSFWQI water quality parameters. NSF water quality index.
Sr. No. Parameter Weightage Sr. No. NSFWQI Range Water Quality

1 Dissolved Oxygen 0.17 1 90–100 Excellent

2 Faecal Coliform 0.16 2 70–90 Good
3 pH 0.11 3 50–70 Medium
4 Biochemical Oxygen demand 0.11 4 25–50 Bad
5 Temperature change 0.10 5 0–25 Very Bad
6 Total Phosphates 0.10
7 Nitrates 0.10
8 Turbidity 0.08
9 Total Solids 0.07 Table 3
Subindex equations of water quality index by Vedprakash.
Parameter Range Applicable Equation
picture of water quality status of Indian rivers like Godavari. In the
present study, an attempt has been made to utilise fuzzy logic approach DO (% saturation) 0–40% saturation IDO ¼ 0.18 þ 0.66 x(% sat)
40–100% saturation IDO ¼ 13.5 þ 1.17 x (% sat)
to develop fuzzy inference system; to overcome the limitations of
BOD (mg/l) 0–10 IBOD ¼ 96.67-7 x
existing WQI approaches and for the identification of actual river water 10–30 IBOD ¼ 38.9-1 x
quality status (especially in such Indian scenario). Fuzzy logic approach >30 IBOD ¼ 2
has its inherent merits of ‘flexibility’ and capability to deal with pH 2–5 IpH ¼ 16.1 þ 7.35 x
5–7.3 IpH ¼ 142.67 þ 33.5 x
‘vagueness’ and ‘uncertainty’ more effectively. This approach also
7.3–10 IpH ¼ 316.96-29.85 x
effectively addresses to ‘sensitivity’ and ‘ambiguity’ issues encountered 10–12 IpH ¼ 96.17-8.0 x
in the determination of WQIs. In the present study suitability of God­ <2, >12 IpH ¼ 0
avari river water have been investigated for drinking purpose and out­
door bathing purpose.
that the same are also applicable for the present study.
2. Materials and methods
2.3. Vedprakash water quality index (VWQI)
2.1. Study area
Vedprakash WQI (VWQI) is the other most commonly used WQI for
determination of water quality of surface water bodies in Indian sub­
The Godavari, one of the biggest river in India, emerges from Brah­
continent (Bhutiani et al., 2014; Water quality report, MPCB, 2015).
magiri Mountain (at 19.560N, 73.200E) is situated at Triambakeswar in
Vedprakash (1990) developed the index on observing the fact that in
Nashik District of Maharashtra state (Fig. 1) (Chavan et al., 2009). The
routine water quality analysis work, only a few numbers of important
present study was carried out to assess the water quality of the Godavari
parameters viz. DO, BOD & pH should be considered, to ascertain the
river in about 24 Km stretch in and around Nashik city. In the present
real water quality status of the streams in the Indian subcontinent. The
study, fourteen river water quality sampling stations from the Gangapur
list of parameters was selected using Delphi. This index was utilised to
dam to Dasak village were selected. The sampling stations are marked as
identify the stretches of river Ganga where the difference between the
S1 to S14 in Fig. 2. The water samples for analysis were collected once in
required and the actual water quality is considerably high, to indicate
a month for 4 years at a point 30 cm below the water surface. Analysis of
the need of pollution abatement initiatives. The index is expressed in
all the parameters was performed using standard methods (APHA,
equation (2) here:
2012). Intense reconnaissance survey work was done to identify the
point and nonpoint sources of wastewater addition in a selected stretch p
X
of the Godavari river and on that basis Fig. 2 has been developed. During VWQI ¼ W i Ii (2)
the survey work, it was observed that partially treated or untreated
i¼1

sewage is getting added in the selected stretch of the Godavari river. where Wi is weightage of ith water quality parameter and Ii is subindex
These are the main point sources of pollution and are marked as P1 to for ith water quality parameter and p is the number of water quality
P12 presented in Fig. 2. parameters.
The index is inspired from the NSFWQI, with little alteration in the
2.2. National Sanitation Foundation water quality index (NSFWQI) weightages to affirm the water quality criteria for the uses of different
categories. Subindex equations were used to determine the subindex
In the present study, one of the most extensively used index in the values, as presented in Table 3. In the present study Vedprakash water
entire world, National sanitation foundation water quality index quality index (VWQI) has also been determined.
(NSFWQI) developed by Brown et al. (1970) has been used to find WQI Weightages (Wi ) of Parameters: DO- 0.43, BOD - 0.26, pH - 0.31
for surface water (Sutadian and Yilmaz, 2016; Bhutiani et al., 2014:
Basin, 2001). 2.4. Fuzzy logic system
X
NSFWQI ¼ Wi Qi (1) 2.4.1. Fuzzy inference system
Fuzzy inference system is resulted on combining the fuzzy logic with
where Wi ¼ Weightage of the concerned parameter & Qi ¼ Q Value of expert system (Zadeh, 1965). The real conditions are often uncertain
the concerned parameter based on observed value, determined from and Zadeh (1965) defined this vagueness using the term “Fuzziness”.
concerned graph Chau (2006) had used concept of fuzzy sets to carry out water quality
For calculation of NSFWQI five parameters (excluding Faecal Coli­ modeling, since it put forth a appropriate approach to deal with the is­
form, temperature change, phosphates and Nitrates) have been consid­ sues, where the goals and boundaries are not properly defined or
ered in the present study. The selection of parameters was done by using imprecise. Nasiri et al. (2007) developed a fuzzy river-pollution decision
Delphi method. support system to make expert knowledge available for nonexpert users.
The weightages of the parameters and water quality class for Bai and Mohan (2009) studied water quality in the Semenyih river
NSFWQI are presented in Tables 1 and 2 respectively and considered (Malaysia) with physico chemical parameters and commented that fuzzy

3
J.G. Nayak et al. Groundwater for Sustainable Development 10 (2020) 100350

Fig. 3. Process flow diagram for fuzzy inference system (F.I.S.).

water quality index will assist the decision makers in reporting the 0 1 8 9
0 x < a or d < x
condition of water quality. Semiromi and Hassani (2011) et al. devel­ >
> >
>
> >
oped fuzzy water quality index for Karoon river (Iran) and identified B
B
C
C
>
>
>
> ða xÞ
>
>
>
>
B C > >
that fuzzy logic approach have superior capabilities to deal with B C <
ða bÞ
a�x�b =
non-linear, complex and uncertain systems. Gharibi (2012) commented Trapezoidal: ​ f ​ B
Bx; ​ a; b; c ​C
C​ ¼ >
B C > 1 b �x�c >
>
> >
that fuzzy logic approach is a comprehensive tool for water quality B
@
C
A
>
>
>
>
>
>
> xÞ >
assessment, especially for the analysis of human drinking water. Patki >
: ðd
c�x �d
>
;
(2015) used fuzzy inference system to assess the physicochemical pa­ ðd cÞ
rameters of municipal water, under the distribution system to determine
its portability. Li and Zhihong (2016) developed fuzzy water pollution where a, b, c and d are membership function parameters and x rep­
index method for water quality determination of Qu river (China) and resents each single point on the x axis.
found that the method gives comprehensive water quality rank. b) Evaluation of the rules: Here compounding of the outcome of a fuzzy
There are some predominant reasons, which signify the merits of the if-then rule is done. In this process, the lowermost class of any of the
models working on the concept of fuzzy logic. The reasons are: first, selected parameter becomes the class of fuzzy index for that partic­
Issues involving nonlinear relationships among the variables can be ular rule. The lowermost class and corresponding value of mem­
addressed; second, local simple models have been utilised, rendering bership function of any of the selected parameter becomes the class
them simple to work with; third, they can be communicated and un­ of the fuzzy index for the rule. In this step, applicable fuzzy rules for
derstood verbally; fourth, individual expertise and experience is used, the given data are identified.
which is not utilised by other methods (Lermontov et al.,2009; Gharibi, c) Aggregation of rule outputs: In this step, applicable fuzzy if-then
2009). rules are considered. The maximum value of the fuzzy index hav­
ing the same class, among the available applicable rules is consid­
2.4.2. Fuzzy inference rules ered. Outputs of all applicable rules are then clubbed into a single
In fuzzy logic, if-then rules and fuzzy set operators are utilised to fuzzy distribution by carrying out the fuzzy union of applicable rules.
express the relationship between input and output variables of a system. d) Defuzzification: Defuzzification is the conversion of the fuzzified
Concepts of the formation of the fuzzy rules have been presented in output to the crisp value by using the methods of defuzzification.
Table 6.
2.5. Modeling performance criterions
2.4.3. Basic structure of a fuzzy inference system (FIS)
A fuzzy inference system (FIS) works on fuzzy set theory, and in­ To ascertain the prediction accuracy of the fuzzy models developed
corporates the following four main steps (Ross, 1977): in the present study, two criterions have been utilised to assess the
performance of the models. The criterions utilised are mean square error
a) Fuzzification: In this step, Crisp values of the input parameters are (MSE) and coefficient of correlation (CC).
converted into corresponding linguistic terms by making use of
membership functions. The adopted ranges of the different class for 2.5.1. Mean square Error(MSE)
concerned parameters are used to find the value of membership The mean squared error (MSE) of a particular case measures the
function for different parameters. For determination of the value of average of the squared difference between the observed values and the
triangular and trapezoidal membership functions, undermentioned predicted values. The MSE gives an indication of the error or residual
equations are used variance.
0 1 8 0 x < a or c < x 9 1X n
>
> >
> MSE ​ ¼ ½observed predicted�2 (3)
B C >
> >
> n i¼1
B C >
< ða xÞ >
=
B C a�x�b
Triangular ​ : ​ f ​ Bx; ​ a; b; c ​C​ ¼ ða bÞ
B
@
C
A
>
>
>
>
>
>
>
> 2.5.2. Coefficient of correlation
> ðc xÞ >
: b �x�c ; The Coefficient of correlation is an indication of the linear relation­
ðc bÞ ship between the two considered variables as presented here,

4
J.G. Nayak et al. Groundwater for Sustainable Development 10 (2020) 100350

Table 4
Values of input and output parameters for trapezoidal membership function.
Class DO BOD Turbidity
a b c d a b c d a b c d

Excellent 6.5 8.0 9.0 9.0 0 0 2.0 3.0 0.0 0.0 7.5 15.0
Good 4.5 5.5 6.5 8.0 2.0 3.0 5.0 6.0 7.25 15.0 22.5 30.0
Medium 2.75 3.75 4.5 5.5 5.0 6.0 8.0 10.0 22.5 30.0 45.0 60.0
Bad 1.0 2.0 2.75 3.75 8.0 10.0 35.0 45.0 45.0 60.0 75.0 90.0
Very Bad 0.0 0.0 1.0 2.0 35.0 45.0 75.0 75.0 75.0 90.0 125.0 125.0

Class Total Solids pH FWQI

a b c d a b c d a b c d

Excellent 0 0 150.0 244.0 6.5 6.5 8.25 8.50 80.0 90.0 100.0 100.0
Good 150.0 244.0 338.0 433.0 8.25 8.50 8.75 9.0 60.0 70.0 80.0 90.0
Medium 338.0 433.0 464.0 537.0 8.75 9.0 9.25 9.50 40.0 50.0 60.0 70.0
Bad 464.0 537.0 800.0 900.0 9.25 9.50 10.00 10.50 15.0 25.0 40.0 50.0
Very Bad 800.0 900.0 1000 1000 10.00 10.50 12.00 12.00 0.0 0.0 15.0 25.0

Table 5
Values of input and output parameters for triangular membership function.
Class DO BOD Turbidity
a b c a b c a b c

Excellent 6.5 8 9.0 0 2.0 4.0 0 7.5 19.0

Good 4.5 6.5 8.0 2.0 4.0 8.0 7.5 19.0 40.0
Medium 2.5 4.5 6.5 4.0 8.0 28.0 19.0 40.0 65.0
Bad 1.0 2.5 4.5 8.0 28.0 45.0 40.0 65.0 90.0
Very Bad 0 1.0 2.5 28.0 45.0 75.0 65.0 90.0 125.0

Class Total Solids pH FWQI

a b c a b c a b c

Excellent 0 150.0 290.0 6.50 8.25 8.50 80.0 90.0 100.0

Good 150.0 290.0 433.0 8.25 8.50 9.00 60.0 80.0 90.0
Medium 290.0 433.0 668.0 8.50 9.00 9.50 38.0 60.0 80.0
Bad 433.0 668.0 900.0 9.00 9.50 10.50 15.0 38.0 60.0
Very Bad 668.0 900.0 1000.0 9.50 10.50 12.00 0 15.0 38.0

P
n comparison of river water quality using NSFWQI, VQWI and Fuzzy
ðxi x’Þðyi y’Þ
water quality index (i.e.FQWI) based on different membership func­
Cc ¼ rffii¼1
ffinffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi�ffiffiffi2ffi (4)
P tions; involving different methods of defuzzification have been done. To
ðxi x’Þ2 ðyi y’
i¼1 keep a common basis for ascertaining the value of the indices, five
representative and important water quality parameters named DO, BOD,
where, n ¼ number of data in the dependent data set, xi ¼ the observed Turbidity, Total solids and pH have been considered. It was also
values, yi¼ the predicted values, x’ ¼ mean of the observed values and y’ considered that the number of fuzzy rules increases exponentially, with
¼ mean of the predicted values an increase in the number of parameters. So, considering only important
Patki (2015) and Tiri and Belkhiri (2018) used mean square error parameters helps in reducing the number of fuzzy rules.
and coefficient of correlation as modeling performance criteria to assess Five types of class of water quality as Excellent, Good, Medium, Bad
the performance of the models. and Very Bad have been selected for both the indices viz. NSFWQI and
FWQI. In this study, the ‘Mamdani’ approach has been utilised for
development of the fuzzy inference system, to determine the FWQI. The
2.6. Development of a water quality index based on fuzzy logic approach process flow diagram for fuzzy inference system is as shown in Fig. 3.
The limits of the Classes for the considered water quality parameters,
In the present study, to investigate the water quality of the Godavari for the development of FWQI have been presented as per Tables 4 and 5
river, fuzzy logic approach with five main parameters namely DO,BOD, respectively.
Turbidity, Total solids and pH have been used. In the present study,

Table 6
Development of fuzzy rules.
Sr. No. DO BOD Turbidity Total Solids pH FWQI

1 Excellent Excellent Excellent Excellent Excellent Excellent

2 Excellent Excellent Excellent Good Excellent Good
3 Excellent Excellent Excellent Good Medium Medium
4 Medium Bad Medium Bad Good Bad
5 Bad Very Bad Medium Bad Good Very Bad
6 Very Bad Bad Bad Bad Medium Very Bad
7 Medium Bad Very Bad Very Bad Bad Very Bad
– – – – – – –
– – – – – – –
– – – – – – –
3125 Very Bad Very Bad Very Bad Very Bad Very Bad Very Bad

5
J.G. Nayak et al. Groundwater for Sustainable Development 10 (2020) 100350

Fig. 4. Triangular MF for DO, BOD, Turbidity, Total solids, pH and FWQI.

For development of FWQI five input parameters have been utilised. quality parameters with five classes, the number of fuzzy rules would be
Each input parameter has been classified into five water quality classes 3125 (i.e. 55). Total 3125 rules were used to develop fuzzy inference
as Very bad, Bad, Medium, Good and Excellent. As there are five water system. Some of the sample fuzzy rules are presented in Table 6.

6
J.G. Nayak et al. Groundwater for Sustainable Development 10 (2020) 100350

Fig. 4. (continued).

The ‘Mamdani’ approach has been used due to its simple structure triangular membership functions have been used for the development of
and Max-Min inference. The implication method used was ‘Min’ and the fuzzy inference system. Water quality data of all fourteen stations for
aggregation method used was ‘Max’. The MATLAB ® 2014 toolbox was three years from 2013 to 2015 have been used for training, while data of
used for modeling the fuzzy inference system. Trapezoidal and one year viz. 2016 has been utilised for testing of the FIS. The study was

7
J.G. Nayak et al. Groundwater for Sustainable Development 10 (2020) 100350

Fig. 5. Variation of WQIs for sampling stations with ‘good’ water quality.

carried out by considering triangular and trapezoidal Membership membership functions has been referred as FQWI-Triangular& FQWI-
function (i.e.) for each of the input and output parameters. The defuz­ Trapezoidal respectively.
zification was carried out by using Centroid, Mean of Maxima (MOM) In the present study, for the development of FIS, both triangular and
and bisector methods. The triangular MF for each input and output trapezoidal membership functions have been adopted using centroid,
parameter is shown in Fig. 4. Bisector and Mean of Maxima (i.e. MOM) as methods of defuzzification.
The triangular membership functions used, have been presented in
3. Results and discussions Fig. 4. The ranges for input and output parameters for triangular and
trapezoidal MF are mentioned in Table 4 and Table 5. These ranges have
In the present study, water quality analysis has been done for four been used to ascertain the water quality at different sampling stations in
years from 2013 to 2016 and for illustration purpose water quality data a selected stretch of Godavari river as Fuzzy water quality index (FWQI).
at station two, three, four, five, thirteen and fourteen have been pre­ FWQIs using Triangular and trapezoidal MF have been used as FWQI-
sented in Figs. 5–7. In Figs. 5–7 four water quality indices have been Triangular and FWQI-Trapezoidal in the subsequent discussion. The
presented viz. VWQI, NSFWQI & two no of FWQI based on triangular & NSFWQI and FWQI values have been taken as observed and predicted
trapezoidal membership functions respectively. Vedprakash WQI (i.e. values respectively. Modeling performance criteria in the form of Mean
VWQI) is developed in India and mainly inspired by NSFWQI. Therefore, square error (i.e. MSE) and coefficient of correlation (i.e. Cc) have been
there is a close correlation between the values of VWQI & NSFWQI. In used for validation of predicted FWQIs. In Table 9, error analysis during
India NSFWQI has been used both by central and various state pollution training and testing of fuzzy models for selected sampling stations has
control boards, so the values of FWQI have been compared with been mentioned. It is evident from Table 10 that a change in the model’s
NSFWQI, by identifying Cc at various sampling stations. In the subse­ performance occurs on changing the membership function and defuz­
quent deliberation, FQWI based on triangular & trapezoidal zification method. The best fitting model for each sampling station has

8
J.G. Nayak et al. Groundwater for Sustainable Development 10 (2020) 100350

Fig. 6. Variation of WQIs for sampling stations with ‘bad’ water quality.

been selected, based on the value of Cc between the observed and pre­ that, they cannot truly represent water quality, when BOD5 and total
dicted values. The performance of the model improves with the increase solids are more than 30 mg/l and 500 mg/l respectively (Nayak et al.,
in the value of Cc. The model having the highest value of Cc during the 2016). Therefore the actual water quality cannot be truly reflected by
testing shows a better correlation between the observed and predicted NSFWQI & VWQI under these cases, due to their limitation of ‘Eclipsing’
values of WQIs and therefore selected as best fitting model. But, if the caused on account of their aggregation function additive in nature.
value of Cc is same for two or more models, then the model having the Whereas at these stations actual water quality status can be truly rep­
lesser value of MSE is selected as best fitting model. The best fitting resented by FWQI, since the fuzzy approach is equally sensitive to a
models for each sampling station has been presented in Table 10. It can minor change in the value of each parameter.
be readily perceived from Tables 9 and 10 that the performance of It is observed from Figs. 5–7 that FWQI-Triangular and FWQI-
models with triangular MF using the bisector method of defuzzification Trapezoidal values are on the relatively lower side as compared to
is better, as compared to the triangular and trapezoidal MF employing NSFWQI and VWQI values. This is because FWQI approach gives equal
other methods of defuzzification. It can also be observed from Table 10 weightage to each parameter. When the value of even a single parameter
that out of fourteen sampling stations in the study area, triangular MF is in the ‘bad’ category (like BOD5 is more than 500 mg/l), then FWQI
performs better for thirteen stations as compared to trapezoidal MF, approach considers that the water quality is ‘bad’, even when the
while trapezoidal MF performs better for one station only. It is evident remaining parameters are in ‘good’ category. This change in water class
from Table 10 that the bisector method of defuzzification performs is not truly reflected by NSFWQI and VWQI approach due to their lim­
better for nine sampling stations, while centroid and mean of maxima itation of ‘Eclipsing’. It is evident from Figs. 5–7 that during the summer
method performs better for three and two stations respectively. The season in all years except 2015, a significant reduction in WQI values
value of Cc at sampling stations 11 to 14 is relatively less, since the valve occurs, which is effectively reflected by FWQI-Triangular approach and
of BOD5 and total solids at these stations is more than 30 mg/l and 500 has not been depicted by NSFWQI and VWQI approaches due to their
mg/l respectively, due to which significant reduction in value of FWQI inherent limitation. KumbhMela was organized during 2015 on the
occurs, while NSFWQI and VWQI values are relatively more as banks of Godavari at Nashik, so waste water was not allowed to be
compared to FWQI values. This significant difference in the values of entered into the Godavari before the prominent bathing ghats at Nashik.
FWQI as compared to values of NSFWQI and VWQI is due to the limi­ Therefore in summer 2015, a significant difference between FWQI and
tations of NSFWQI and VWQI. The limitations of NSFWQI and VWQI are other WQI values is not visible. It can be readily observed from Figs. 5–7

9
J.G. Nayak et al. Groundwater for Sustainable Development 10 (2020) 100350

Fig. 7. Variation of WQIs for sampling stations with ‘very bad’ water quality.

study.
Table 7
It has been observed from the study that from the Gangapur dam to
Water quality as per NSFWQI.
the Someshwar temple only the Godavari river is CPCB ‘Class A’ river
Sr. No. NSFWQI Range FWQI-Triangular MF Range Water Quality and fit to serve as drinking water after disinfection only. Based on study
1 90–100 85–100 Excellent (Tables 7 and 8), it is found that, Godavari is CPCB ‘Class C’ river from
2 70–90 70–85 Good Anandwalli bridge to Ghatgebaba bridge. While from Tapovan STP to
3 50–70 48–70 Medium
Dasak Bridge Godavari is CPCB ‘Class D’ river. It can be understood from
4 25–50 27–48 Bad
5 0–25 0–27 Very Bad
the above deliberation that actual water quality status can not be truly
represented by NSFWQI & VWQI approach at places, where pollution
level increases beyond a certain limit. Therefore a more rational
approach like fuzzy logic approach is required to represent the true
Table 8
picture of the actual water quality status of the Indian rivers, carrying
Class of river as per CPCB & FWQI.
high pollutional loads.
Sr. NSFWQI FWQI-Triangular MF Class of River as per
No Range Range CPCB
4. Conclusions
1 63–100 61–100 A
2 50–63 48–61 B
The study for identification of water quality status of the Godavari
3 38–50 36–48 C
4 <38 <36 D&E river at Nashik was carried out by using the most prevalent NSFWQI and
indigenous VWQI. It is evident that both NSFWQI and VWQI exhibits
similar results and are in close correlation, but both indices suffer from
that the performance of FWQI-Triangular is more representative to ‘Eclipsing’ and can not truly represent the high concentration of pa­
exhibit true water quality status as compared to FWQI-Trapezoidal. This rameters like BOD5 and total solids. Therefore these indices become
is due to the observation that appropriate variation in the values of insensitive towards these parameters. To overcome the limitations of
FWQI-Trapezoidal is not reflected like FWQI-Triangular. This is due to these two indices, a more rational fuzzy logic approach has been
the flat shape of trapezoidal-MF in the central portion, which makes it adopted. The fuzzy models have been developed using triangular and
relatively less representative as compared to triangular MF. As evident trapezoidal membership functions with centroid, bisector and mean of
from Table 10 that FWQI-Triangular model using the Bisector defuzzi­ maxima (MOM) methods for defuzzification. The study revealed that
fication method outperforms other considered FWQI models of the

10
J.G. Nayak et al. Groundwater for Sustainable Development 10 (2020) 100350

Table 9
Stationwise error analysis during training and testing of fuzzy inference system.
Station No. Membership Function Defuzzification Method Training Testing

MSE CC MSE CC

1 Trapezoidal Centroid 162.53 0.60 80.82 0.80

Bisector 172.05 0.60 82.67 0.86
MOM 194.97 0.63 79.83 0.87
Triangular Centroid 85.27 0.69 47.73 0.75
Bisector 86.05 0.65 33.42 0.83
MOM 99.37 0.53 28.98 0.84
2 Trapezoidal Centroid 181.01 0.79 138.25 0.63
Bisector 175.92 0.79 114.75 0.58
MOM 201.60 0.75 117.66 0.80
Triangular Centroid 89.43 0.86 76.67 0.85
Bisector 86.83 0.84 61.17 0.86
MOM 95.61 0.74 57.62 0.80
3 Trapezoidal Centroid 174.06 0.85 154.62 0.46
Bisector 166.92 0.84 153.58 0.41
MOM 184.17 0.77 93.33 0.40
Triangular Centroid 90.61 0.86 82.06 0.93
Bisector 91.16 0.84 75.25 0.89
MOM 104.26 0.79 95.35 0.74
4 Trapezoidal Centroid 322.64 0.79 548.13 0.55
Bisector 332.72 0.79 568.83 0.38
MOM 586.92 0.36 562.58 0.35
Triangular Centroid 73.48 0.96 93.69 0.87
Bisector 77.58 0.96 91.25 0.85
MOM 121.92 0.84 150.03 0.66
5 Trapezoidal Centroid 348.27 0.75 556.8 0.50
Bisector 358.28 0.75 578.08 0.57
MOM 585.25 0.69 581.25 0.51
Triangular Centroid 76.61 0.91 94.49 0.97
Bisector 81.25 0.92 87.67 0.94
MOM 110.93 0.84 113.87 0.84
6 Trapezoidal Centroid 277.95 0.76 447.41 0.69
Bisector 290.11 0.76 458.25 0.63
MOM 450.08 0.60 447.41 0.61
Triangular Centroid 56.54 0.97 85.90 0.94
Bisector 63.92 0.96 71.92 0.94
MOM 90.49 0.83 63.22 0.85
7 Trapezoidal Centroid 230.57 0.72 338.34 0.65
Bisector 206.16 0.75 326.75 0.64
MOM 321.14 0.63 321.42 0.63
Triangular Centroid 77.51 0.94 109.76 0.96
Bisector 72.69 0.95 102.58 0.96
MOM 180.35 0.76 192.27 0.85
8 Trapezoidal Centroid 215.98 0.79 342.03 0.38
Bisector 205.64 0.81 345.08 0.37
MOM 365.36 0.38 341.25 0.38
Triangular Centroid 56.21 0.97 97.87 0.96
Bisector 62.75 0.96 68.83 0.97
MOM 87.62 0.85 111.12 0.79
9 Trapezoidal Centroid 256.23 0.76 395.21 0.40
Bisector 262.72 0.73 403.83 0.40
MOM 384.75 0.39 394.58 0.39
Triangular Centroid 85.30 0.90 94.27 0.93
Bisector 88.27 0.89 77.33 0.97
MOM 115.64 0.82 52.85 0.84
10 Trapezoidal Centroid 301.36 0.61 389.60 0.23
Bisector 329.55 0.61 372.58 0.23
MOM 433.36 0.59 364.75 0.25
Triangular Centroid 99.15 0.90 91.75 0.92
Bisector 149.89 0.91 84.83 0.97
MOM 136.52 0.89 75.18 0.88
11 Trapezoidal Centroid 49.10 0.81 116.94 0.50
Bisector 57.72 0.69 115.33 0.47
MOM 148.15 0.47 142.71 0.30
Triangular Centroid 49.29 0.84 107.40 0.57
Bisector 70.66 0.84 135.75 0.68
MOM 209.36 0.71 241.08 0.58
12 Trapezoidal Centroid 96.22 0.59 218.36 0.20
Bisector 124.44 0.59 218.42 0.25
MOM 288.17 0.49 292.08 0.46
Triangular Centroid 91.27 0.64 158.47 0.36
Bisector 120.17 0.65 167.42 0.69
MOM 287.27 0.60 406.811 0.37
13 Trapezoidal Centroid 251.88 0.41 271.99 0.15
(continued on next page)

11
J.G. Nayak et al. Groundwater for Sustainable Development 10 (2020) 100350

Table 9 (continued )
Station No. Membership Function Defuzzification Method Training Testing

MSE CC MSE CC

Bisector 262.78 0.41 264.17 0.39

MOM 366.72 0.33 335.83 0.38
Triangular Centroid 136.63 0.41 132.55 0.33
Bisector 152.64 0.39 138.08 0.22
MOM 138.22 0.42 245.58 0.47
14 Trapezoidal Centroid 242.55 0.43 280.91 0.24
Bisector 260.69 0.41 276.83 0.20
MOM 360 0.38 31.68 0.21
Triangular Centroid 148.60 0.41 139.55 0.28
Bisector 170.94 0.41 149.50 0.33
MOM 354.79 0.15 20.02 0.28

Table 10
Sampling stationwise best fitting model.
Station Membership Defuzzification Cc Station Membership Defuzzification Cc
No. Function Method No. Function Method

1 Trapezoidal MOM 0.87 8 Triangular Bisector 0.97

2 Triangular Bisector 0.86 9 Triangular Bisector 0.97
3 Triangular Centroid 0.93 10 Triangular Bisector 0.97
4 Triangular Centroid 0.87 11 Triangular Bisector 0.68
5 Triangular Centroid 0.97 12 Triangular Bisector 0.69
6 Triangular Bisector 0.94 13 Triangular MOM 0.47
7 Triangular Bisector 0.96 14 Triangular Bisector 0.33

1) Fuzzy logic approach is equally sensitive to all parameters and pro­ CCME, 2001. Canadian Water Quality Guidelines for the Protection of Aquatic Life.
CCME Water Quality Index 1.0, User’s Manual, Winnipeg, Manitoba, Canada
vides flexibility to choose the range of different parameters and
(Accessed in December 2014). http://www.ccme.
corresponding quality class as per expert’s judgement. ca/files/Resources/calculators/WQI%20User’s%20Manual%20(en).pdf.
2) Fuzzy logic approach is more sensitive for little change in the value of Chau, K., 2006. A review on integration of artificial intelligence into water quality
each parameter and takes cognizance of the water quality minutely. modelling. Mar. Pollut. Bull. 52, 726–733.
Chavan, Ajay D., Sharma, M.P., Bhargava, Renu, 2009. Water quality assessment of the
3) Fuzzy logic approach can truly represent the water quality status in Godavari river. J. Hydro. Nepal 5, 18–23.
the cases of very high pollution of streams, observed in the Indian Cude, C.G., 2001. Oregon water quality index: a tool for evaluating water quality
scenario, which cannot be depicted by NSFWQI and VWQI approach. management effectiveness. J. Am. Water Resour. Assoc. 37 (1), 125–137. https://
doi.org/10.1111/j.1752-1688.2001.tb05480.x.
4) The performance of the fuzzy model changes considerably on Gharibi, Hamed, 2009. A novel approach in water quality assessment based on fuzzy
changing the membership function and defuzzification method. logic. Environ. Manag. 112, 87–95.
5) The fuzzy models with triangular MF and bisector method of Gharibi, Hamed, 2012. Development of a dairy cattle drinking water quality index
(DCWQI) based on fuzzy Inference systems. Ecol. Indicat. 20, 228–237.
defuzzification outperforms other fuzzy models for indian streams. Hanh, P.T.M., Sthiannopkao, S., Ba, D.T., Kim, K.W., 2011. Development of water quality
6) The FWQI index developed by the present study produces reliable indexes to identify pollutants in Vietnam’s surface water. J. Environ. Eng. 137 (4),
and accurate results, therefore its use should be preferred by the 273–283.
Hurley, T., Sadiq, R., Mazumder, A., 2012. Adaptation and evaluation of the Canadian
municipal corporations for assessment of surface water quality, Council of Ministers of the Environment Water Quality Index (CCME WQI) for use as
especially when assessing water for human consumption or use by an effective tool to characterize drinking source water quality. Water Res. 46 (11),
mankind is under consideration. 3544–3552. https://doi.org/10.1016/j.watres.2012.03.061.
Lermontov, A., Yokoyama, L., Lermontov, M., Machado, M.A.S., 2009. River quality
analysis using fuzzy water quality index: Ribeira do Igaupe river watershed, Brazil.
Acknowledgements Ecol. Indicat. 9 (6), 1188–1197.
Li, R., Zhihong, Z., 2016. Water quality assessment in qu river based on fuzzy water
The Authors are grateful to the Management and Principal of Sandip pollution index method. J. Environ. Sci. https://doi.org/10.1016/j.jes.2016.03.030.
Nasiri, Fuzhan, Massod, I., 2007. Water quality index: a fuzzy river pollution decision
Foundation’s Sandip Institute of Technology and Research Centre, support expert system. J. Water Resour. Plann. Manag. 133 (2), 95–105.
Nashik, India for providing all the facilities to carry out this study. Nayak, Jyotiprakash, G., Patil, L.G., 2016. Assessment of water quality of Godavari river
at Nashik, Maharashtra, India. Int. J. Civ. Eng. Technol. 7 (1), 83–92.
Nayak, Jyotiprakash, G., Patil, L.G., 2017. Utilization of prevalent western water quality
References index in Indian scenario: limitations and prospects. In: Proc. Int. Conf. Recent
Advances in Civil and Environmental Engineering.Sangali,Maharashtra,India, 1,
Abbasi, T., Abbasi, S.A., 2011. Water quality indices based on bioasessment: the biotic pp. 160–164.
indices. J. Water Health 9 (2), 330–348. Patki, Vinayak K., 2015. Fuzzy system modeling for forecasting water quality index in
American Public Health Association., APHA, 2012. Standard Methods for the municipal distribution system. Urban Water J. 12 (2), 89–110.
Examination of Water and Wastewater, 22th ed. New York.2-9,2-48,4-87,4-134,5- Ross, S.L., 1977. An index system for classifying river water quality. Water Pollut.
3,9-47. Control 76 (1), 113–122.
Bai, Raman, Mohan, S., 2009. Fuzzy logic water quality index and importance of water Said, A., Stevens, D.K., Sehlke, G., 2004. An innovative index for evaluating water
quality parameters. J. Air Soil Water Res. 2, 51–59. quality in streams. Environ. Manag. 34 (3), 406–414.
Basin, 2001. Water Quality Information References, National Sanitation Foundation, Sargaonkar, A., Deshpande, V., 2003. Development of an overall index of pollution for
Water Quality Index. www.water-h.net/watrqualindex/waterqualityindex.htm. surface water based on a general classification scheme in Indian context. Environ.
Bhargava, D., 1985. Expression for drinking water supply standards. J. Environ. Eng. 111 Monit. Assess. 89 (1), 43–67.
(3), 304–316. https://doi.org/10.1061/(ASCE)0733-9372(1985)111:3(304). Semiromi, B., Hassani, A., 2011. Water quality index development using fuzzy logic: a
Bhutiani, R., Khanna, D.R., Kulkarni, D., Ruhela, M., 2014. Assessment of Ganga river case study of the karoon river of Iran. Afr. J. Biotechnol. 10 (50), 10125–10133.
ecosystem at Haridwar, Uttarakhand, India with reference to water quality indices. Sutadian, A.D., Yilmaz, A., 2016. Development of river water quality indices -A review.
Appl. Water Sci. 1–7. https://doi.org/10.1007/s13201-014-0206-6. Environ. Monit. Assess. 188, 58. https://doi.org/10.1007/s10661-015-5050-0.
Brown, R.M., McClelland, N.I., Deininger, R.A., Tozer, R.G., 1970. A water quality
index—do we dare? Water Sewage Works 117 (10), 339–343.

12
J.G. Nayak et al. Groundwater for Sustainable Development 10 (2020) 100350

Swamee, P., Tyagi, A., 2007. Improved method for aggregation of water quality Tiri, A., Belkhiri, L., 2018. Evaluation of surface water quality for drinking purposes
subindices. J. Environ. Eng. 133 (2), 220–225. https://doi.org/10.1061/(ASCE) using fuzzy inference system. Groundwater Sustain. Dev. 6, 235–244. https://doi.
0733-9372(2007)133:2(220). org/10.1016/j.gsd.2018.01.006.
Zadeh, L.A., 1965. Fuzzy Sets, Information and Control, vol. 8, pp. 338–353. http://
mpcb.gov.in/ereports/pdf/waterqualityreport2014-15TERI.pdf.

13