Journal of Environmental Protection, 2010, 1, 136-142
doi:10.4236/jep.2010.12018 Published Online June 2010 (
Copyright © 2010 SciRes. JEP
Analysis of Groundwater for Potability from Tiruchirappalli
City Using Backpropagation ANN Model and GIS
Natarajan Venkat Kumar*, Samson Mathew, Ganapathiram Swaminathan
Civil Engineering Department, National Institute of Technology, Tiruchirappalli, India.
Received February 28th, 2010; revised April 9th, 2010; accepted April 11th, 2010.
Monitoring groundwater quality by cost-effective techniques is important as the aquifers are vulnerable to contamina-
tion due to point sources and non point sources. This paper presents Artificial neural Network (ANN) Models that might
be used to predict water parameters from a few known parameters. The sample data from 112 hand pumps and hand
operated tube well water samples used for drinking purposes by the local population was used. The ANN model features
a back propagation algorithm and neuron members were determined for optimization of the model architecture by trial
and error method. The model simulations show that the optimum network of 4-50-50-6 has mean error of –0.023% on
complete data was utilized. This demonstrated that the developed model has high accuracy for predicting. Thus it has
been established that the two hidden layers neural network has more efficiency than asymptotic regression in the pre-
sent. This model can be used for analysis and prediction of subsurface water quality prediction.
Keywords: Groundwater Quality, ANN, Ec, TDS, Sulphates, pH, Tiruchirappalli
1. Introduction
Water Quality is the physical, Chemical and biological
characteristics of water in relationship to a set of stan-
dards. Water quality standards are created for different
types of water bodies and water body locations as per
desired uses. There are separate quality standards for
portable water, agriculture needs, industrial needs and
construction needs. The primary uses considered for such
contact and for health of ecosystems. The methods of
hygrometry are used to quantify water characteristics.
Water quality depends on the local geology and ecosys-
tem, as well as human uses such as sewage dispersion,
industrial pollution, use of water bodies’ heat sink and
overuse. References [1-3], which emphasized the work
initiated from Hill’s general theory of uniqueness and
bifurcation [4]. Environmental engineers and researchers
have paid much attention to the behavior of water quality
in portable operations over the last decade. It is difficult
to define criteria for describing the stability in Water
quality. In general, the contamination of groundwater
could occur from non point and point sources. Among
the major contaminants linked to non point sources are
nitrates, heavy metals and pesticides. Heavy metals con-
taminate groundwater from anthropogenic sources as
well as natural sources. Some of the major anthropogenic
sources of heavy metals are mining, fertilizers and pesti-
cides and industrial wastes [5]. The high toxicity of hu-
man health is manifested in the low maximum contami-
nant levels (MCL). High concentrations of heavy metals
occur due to a complex interaction between factors that
include land use practices and hydro geochemical condi-
tions prevailing in soils [6].
Groundwater monitoring is defined as testing of
groundwater over a next ended time period in oder to
document groundwater conditions, including the collec-
tion of chemicals such as contaminant concentrations [7].
Only limited monitoring schemes considered the conse-
quences of different concentrations and varying pumping
flows among wells as a major factor. One of the most
interesting aspects of hydrochemistry is the occurrence of
water bodies with different water chemistries is very
close proximity to each other. This has been variously
attributed to the surface and subsurface geology [8]. Wa-
ter quality assessment in Akpabuyo, south eastern Nige-
ria indicated waters are acidic, soft and characterized by
low sodium absorption ratio, and waters are classified
into four chemical facies: Ca2+-Cl-1, Na+-Cl-1, CaSO4
- [9]. Studies conducted on the chemical quality
of groundwater of Mangalore city in Karnataka state and
madras in Tamilnadu, India revealed that the groundwa-
ter quality has been deterioted due to over exploitation of
groundwater [10].
Replenishment the groundwater aquifers through arti-
ficial recharge were carried out tin various parts of the
worlds for the last six decades. This motivated the study
Analysis of Groundwater for Potability from Tiruchirappalli City Using Backpropagation Ann Model and GIS
of [5] which set the basics for the current study area. The
chemical composition of water is an important factor to
be considered before it is used for domestic or irrigation
purpose [11]. This present study has been conducted to
assess and evaluate the groundwater quality conditions
for domestic purposes. In this paper an attempt has been
made to evaluate the quality indices of groundwater to
understand the hydro geochemical relationships of the
water quality parameters for the suitability of groundwa-
ter resources.
2. Study Area, Materials and Methods
2.1 Study Area
Tiruchirappalli City of Tamil Nadu, India is selected for
the study. The general topology of Tiruchirappalli is flat
and lies at an altitude of 78 m above sea level. Tiruchi-
rappalli is fed by the rivers Cauvery and Kollidam. There
are reserve forests along the river Cauvery. Golden Rock
and the Rock Fort are the prominent hills. The south-
ern/south-western part of the district is dotted by several
hills which are thought to be an offset of the Western
Ghats Mountain range and the soil is considered to be
very fertile. Figure 1 shows the study area.
The following data that was collected from 112 from
which seventy nine hand pumps and electrically operated
tube well water samples used for drinking purposes by
the local population of the Trichy town of Tiruchirappalli
district of Southern Tamil Nadu Shown in Figure 1. The
samples collected were during 2006 and 2008 on Pre
monsoon and Post monsoon were used in our project for
modeling and prediction purposes. The samples were ana-
Figure 1. Study area of Tiruchirappalli city
Copyright © 2010 SciRes. JEP
Analysis of Groundwater for Potability from Tiruchirappalli City Using Backpropagation Ann Model and GIS
Copyright © 2010 SciRes. JEP
lyzed for the following physico chemical water quality
parameters as per standard methods [1].
The Laboratory tests are conducted on these samples
for 16 different physico-chemical water quality parame-
ters as per the standard procedure [1]. [2,12] criteria are
adopted for testing these samples. The groundwater hy-
drochemistry records of the study area are used for the
preparation of maps. These maps are obtained by geosta-
tistical (Kriging) methodology and the results are pre-
sented in the form of equal ion concentration lines [7,10,
13]. The MATLAB software was also used to analyze the
data. The groundwater quality data are used as the hidden
layer for the preparation of base maps. These features are
the boundary lines between mapping units, other linear
features (streets, rivers, roads, etc.) and point features
(bore well points, etc.). The contours are developed for
pH, EC, Cl-, Na+, Ca2+, Mg2+, Total Hardness, Alkalinity,
F, SO4
-, Coliform and NO3
- values for the pre monsoon
and post monsoon values. The monitoring and sampling
program was initiated in 2006 and finalized the year 2008.
A total of seventy nine monitoring stations were estab-
lished of them represented groundwater conditions.
2.2 Artificial Neural Networks (ANNs)
An artificial neural network was then developed using
part of the experimental data for training and testing.
Finally, the neural network model was applied to Collec-
tion of available data about the various water quality pa-
rameters of the groundwater sources in Tiruchirappalli
town of Tiruchirappalli district in Tamil Nadu, India. The
reserved part of the experimental data to perform the
investigation of the occurrence of water quality constitu-
ents and to examine the effectiveness and robustness of
the neural net work model. The effect of Sulphates, TDS,
Ec, Chlorides, pH, Creation of an ANN model for the
prediction of Electrical Conductivity when Sulphates,
Chlorides, Total Dissolved Solids and pH data are avail-
able for studying wrinkling behavior in ANN model. The
literature reveals that a very little effort is reported on the
use of ANNs in water quality modeling.
Neural networks, as used in artificial intelligence, have
traditionally been viewed as simplified models of neural
processing in the human brain. It is accepted by most
scientists that the human brain is a type of computer. The
origins of neural networks are based on efforts to model
information processing in biological systems, which may
rely largely on parallel processing as well as implicit
instructions based on recognition of patterns of “sensory”
input from external sources [14].
2.2.1 Back Propagation Network
The back propagation algorithm has made it possible to
design multi-layer neural networks for numerous appli-
cations, such as adaptive control, classification of sonar
targets, stock market prediction and speech recognition
[14]. Also, BPNN has the advantage of fast response and
high learning accuracy [14]. Hence an ANN with back
propagation algorithm (BP) has been adopted here to
model the potability behavior of subsurface water. One
of the advantages of using the neural network approach is
that a model can be constructed very easily based on the
given input and output and trained to accurately predict
process dynamics. This technique is especially valuable
in processes where a complete understanding of the
physical mechanisms is very difficult, or even impossible
to acquire, as in the case of porous powder performs
during upsetting. Neural network is a logical structure
with multi-processing elements, which are connected
through interconnection weights.
The knowledge is presented by the interconnection
weights, which are adjusted during the learning phase.
There are several algorithms available among which the
Levenberg-Marquardt algorithm (trainlm) will have the
fastest convergence [15]. In many cases, trainlm is able to
obtain lower mean square errors than any of the other
algorithms tested. This BP network is a multi layer of the
network architecture including the input layer, hidden
layer(s) and output layer. Layers include several process-
ing units known as neurons. They are connected with
each other by variable weights to be determined. In the
network, the input layer receives information from exter-
nal source and passes this information to the network for
processing. The hidden layer receives from the input layer,
and does all information processing. The output layer
receives processed information from the network, and
sends the results to an external receptor [15]. The algo-
rithm for the back propagation program is described be-
low with the help of flow diagram as shown in Figure 2.
2.2.2 Model Description
In the development of a multi layer neural network mod-
el, several decisions regarding number of neuron(s) in the
input layer, number of hidden layer(s), number of neu-
ron(s) in the hidden layer(s), and number of neuron(s) in
the output layer and optimum architectures have to be
decided. Based on the experimental investigation by
venkatkumar et al. [7,10], the important input parameters
such as the total hardness. Alkalinity, chlorides and Sul-
phates are given as input parameters to the present ANN
model. The output parameters are the pH, the Ec, the
Ca2+, Mg2+, Na+ and K. The input/output dataset of the
model is illustrated schematically in Figure 3.
2.2.3 Approach towards Groundwater Classification
A fuzzy rule based system is generated in which users
classify the water according to given data in Desirable,
Acceptable, Not acceptable, Rejected quality with re-
spect to different parameters, all connected using AND
operator. The user’s feedback is also taken with respect
to overall quality for different parameters connected by
AND operator. For example, one of the feedbacks taken
Analysis of Groundwater for Potability from Tiruchirappalli City Using Backpropagation Ann Model and GIS 139
Figure 2. Model for feed forward network (source: Matlab user guide)
Figure 3. Schematically input/output data set model for prediction of Chlorides and Sulphates
may be like this, If TDS = good AND pH = medium and
Sulphate = good then, overall water quality = What! Af-
ter this, Delphi’s technique is applied to converge the
feedback of various users to a single value. A degree of
match is computed between the user’s perception and
field data for different parameters and for every type of
water quality viz. good, (Desirable) medium (Acceptable)
or bad (Not Desirable). The water quality for which de-
gree of match is the highest is considered to represent the
quality of the water sample (see Figure 4).
3. Results and Discussions
Physio-chemical Groundwater quality assessment by de-
terministic method for drinking groundwater usage on the
basis of 8 water quality parameters were compared with
the concentration in the water with point value prescribed
limits. In case Groundwater quality model approach, these
8 parameters were divided in the four categories on the
basis of expert opinion having their importance with re-
spect to drinking water quality criteria.
The hydro chemical analyses revealed that water sam-
ples in the study area is characterized by hard to very
hard, fresh to brackish and alkaline in nature. The highly
turbid water may cause health risk as excessive turbidity
can protect pathogenic microorganisms from the effects
of disinfectants and also stimulate the growth of bacteria
Copyright © 2010 SciRes. JEP
140 Analysis of Groundwater for Potability from Tiruchirappalli City Using Backpropagation Ann Model and GIS
010 20 3040 50 60 7080
Number of Samples
% of Groundwater quality Classification
FIP model out put for Premonsoon '06
qty min
qty max
qty mean
010 20 30 40 50 60 70 80
Number of samples
% of Groundwater quality Classification
FIP model output for Premonsoon '07
OUTPUT(1:79, 1)
qty min
qty max
qty mean
010 20 304050 607080
Number of samples
% of Groundwater quality classification
FIP model output for Premonsoon '08
qty min
qty max
qty mean
Figure 4. Subsurface water potable frequency during pre-
monsoon periods. (a) 2006; (b) 2007; (c) 2008
during storage.
The performance capability of each network has been
examined based on the correlation coefficient, error dis-
tribution, and convergence of entire dataset within speci-
fied error range between the network predictions and the
experimental values using the test and entire dataset. In
order to decide the optimum structure of neural network,
the rate of error convergence was checked by changing
the number of hidden neurons and number of hidden lay-
ers. From out results, it is identified that the network with
single hidden layer of 50 neurons has given correlation
coefficient of 0.9978 and mean error of 0.0289% which
means the error distribution was uniform. But it was ob-
served that only 78.68% of entire dataset are within ± 4%
error further increasing of neuron in the single hidden
layer (beyond 50); the error distribution was not uniform.
Hence, it has been decided to select two hidden layers
and varied the number of neurons in each hidden layer to
get an optimum one. It was observed that the network
with 50 neurons in each hidden layer has produced the
best performance for each of the output parameters
(4-50-50-6). It is also observed that the architectures
4-40-40-6, 4-46-46-6, and 4-52-52-6, have not more dif-
ference in the mean correlation coefficient than previous
architecture; it had not been selected as an optimum ar-
chitecture, because the mean prediction error as well as
the error distribution, maximum value of error, minimum
value of error were observed as high.
It was evaluated and presented the optimum network
model of 4-50-50-6 has mean prediction error of –0.023%
on the entire dataset. The result shows that 91.69%,
95.78%, and 98.63% of the entire dataset were the per-
centage error ranging between ± 4%, ± 8, and ± 12%
respectively. This was demonstrated that the developed
model has high accuracy for predicting. Characteristic by
pH values, most of the water samples were alkaline in
nature which are well within permissible limit (6.5-8.5)
and some of the samples have been found acceptable for
usage and the ranges were between 6.5 and 9.2 meeting
BIS standards of IS:10500:1991 and WHO (2006) guide-
lines. Based on Electrical Conductivity (Ec) values
measured all water samples Zone-I (Srirangam) were
desirable (< 1 mS/cm) for potability. Potability maps for
the Premonsoon period are shown as Figures 5(a), 5(b)
and 5(c) for years 2006, 2007 and 2008 respectively.
4. Conclusions
The quality of the groundwater of the Tiruchirappalli city
was monitored in 79 sampling wells for 3 years and re-
corded data revealed that the concentrations of cations
and anions were above the maximum, desirable for hu-
man consumption. The Electrical Conductivity was
found to be the most significant parameter within input
parameters used in the modeling. The developed model
was enabled well to test the data obtained from 79 sam-
ples of bore wells of Tiruchirappalli city. Therefore, with
the proposed model applications, it was possible to man-
Copyright © 2010 SciRes. JEP
Analysis of Groundwater for Potability from Tiruchirappalli City Using Backpropagation Ann Model and GIS
age groundwater resources in a more cost-effective and
easy way. Based upon the correlation coefficient, error
distribution, and convergence 46 different BPNN archi-
tectures were trained/analyzed using the experimental
data until an optimum architecture was identified. Out of
the different multilayer BPNN architecture trained, the
BPNN with two hidden layers having 50 neurons trained
with Levenberg-Marquardt algorithm was found to be the
78.64 78.66 78.6878.778.72 78.74
78.64 78.66 78.6878.778.72 78.74
10.74 10.76 10.7810.810.82 10.84 10.86
Suitability Map - winter'06 - Tiruchirappalli City
Zone I
Zone IV
Zone II
Zone III
86 - 91
91 - 96
81 - 86
76 - 81
71 - 76
66 - 71
< 66
78.64 78.66 78.6878.778.72 78.74
78.64 78.66 78.6878.778.72 78.74
10.74 10.76 10.7810.810.82 10.84 10.86
Suitability Map - winter'07 - Tiruchirappalli City
Zone I
Zone IV
Zone II
Zone III
76 - 84
84 - 92
68 - 76
60 - 68
52 - 60
44 - 52
< 44
78.64 78.6678.6878.778.72 78.74
78.64 78.6678.6878.778.72 78.74
10.74 10.76 10.7810.810.82 10.84 10.86
Suitability Map - winter'08 - Tiruchirappalli City
Zone I
Zone IV
Zone II
Zone III
85 - 95
> 95
75 - 85
65 - 75
55 - 65
45 - 55
< 45
Figure 5. (a) Potability map of Winter 2006; (b) Potability
map of Winter 2007; (c) Potability map of Winter 2008
optimum network model. A sound performance was
achieved with the neural network model, with good cor-
relation coefficient (between the predicted and experi-
mental values), high uniform error distribution and the
convergence of the entire dataset within the permissible
error range.
5. Acknowledgements
The First Author would like to express his sincere appre-
ciation to TEQIP Scholarship during his Ph.D., work.
Special thanks go to S.Sivasankaran (NITT) J. Nagesh
Gupta (TCS) and M. Kannan (SASTRA) in the GIS field,
ANN, Simulink activities and in the laboratory meas-
urement of this study. Thanks to Prof. K. Palanichamy,
Civil Engineering Department, NITT for providing con-
stant encouragement throughout his research study. The
authors wish to express their sincere thanks to the ano-
nymous reviewers for their valuable comments to en-
hance the quality of this paper.
[1] American Public Health Association, “Standard Method
for Examination of Water and Waste Water,” 21st Edition,
American Public Health Association, Washington, D.C.,
[2] Bureau of Indian Standard, “Indian Standard Specifica-
tion For Drinking Water,” BIS Publication No. IS: 10501,
New Delhi, 1991.
Copyright © 2010 SciRes. JEP
142 Analysis of Groundwater for Potability from Tiruchirappalli City Using Backpropagation Ann Model and GIS
Copyright © 2010 SciRes. JEP
[3] Z. Chen, G. H. Huan and A. Chakma, “Hybrid Fuzzy-Sto-
chastic Modeling Approach for Assessing Environmental
Risks at Contaminated Groundwater Systems,” Journal of
Environmental Engineering, Vol. 129, No. 1, 2003, pp.
[4] C. Okoliand and S. D. Pawlowski, “The Delphi Method
as a Research Tool an Example, Design Considerations
and Applications,” Information and Management, Vol. 42,
No. 1, 2004, pp. 15-29.
[5] R. D. Deshpande and S. K. Gupta, “Water for India in
2050: First Order Assessment of Available Options,”
Current Science, Vol. 86, No. 9, 2004, pp. 1216-1224.
[6] T. Subramani, L. Elango and S. R. Damodarasamy,
“Groundwater Quality and its Suitability for Drinking and
Agricultural Use in Chithar River Basin, Tamil Nadu, In-
dia,” Environmental Geology, Vol. 47, No. 8, 2005, pp.
[7] N. V. Kumar, S. Mathew and G. Swaminathan, “A Pre-
liminary Investigation for Groundwater Quality and
Health Effects—A Case Study,” Asian Journal of Water,
Environment and Pollution, Vol. 5, No. 4, 2008, pp. 99-
[8] K. Sivasankar and R. Gomathi, “Fluoride and Other
Quality Parameters in the Groundwater Samples of Pet-
taivaithalai and Kulithalai Areas of Tamil Nadu, Southern
India,” Water Quality Exposure Health, Vol. 1, No. 2,
2009, pp. 123-134.
[9] R. Khaiwal and V. K. Garg, “Distribution of Fluoride in
Groundwater and its Suitability Assessment for Drinking
Purposes,” International Journal of Environmental
Health Research, Vol. 16, No. 2, 2006, pp. 163-166.
[10] N. V. Kumar, S. Mathew and G. Swaminathan, “Fuzzy
Information Processing for Assessment of Groundwater
Quality,” International Journal of Soft Computing, Vol. 4,
No. 1, 2009, pp 1-9.
[11] S. Dahiya, B. Singh, S. Gaur, V. K. Garg and H. S.
Kushwaha, “Analysis of Groundwater Quality Using
Fuzzy Synthetic Evaluation,” Journal of Hazard Materi-
als, Vol. 147, No. 3, 2007, pp. 938-946.
[12] World Health Organisation, “Guidelines for Drinking
Water Quality Recommendation,” Vol. 2, World Health
Organisation, Geneva, 1984.
[13] Z. Sen, “Fuzzy Groundwater Classification Rule Deriva-
tion from Quality Maps,” Water Quality Exposure Health,
Vol. 1, 2009, pp. 115-112.
[14] W. C. Chen, G. L. Fu, P. H. Tai and W. J. Deng, “Process
Parameter Optimization for MIMO Plastic Injection
Molding via Soft Computing,” Expert System with Ap-
plications, Vol. 36, No. 2, 2009, pp. 1114-1122
[15] H. Demuth and M. Beale, “Neural Network Toolbox
User’s Guide,” Version 4 (Release 12), The Mathworks,
Inc., 2000.