Technical seminar Presentation on

“AI-BASED MODELS FOR ESTIMATING GROUNDWATER QUALITY

PARAMETER”

Presented by

Harish K L 1SI17CV023

Under the guidance of ,

Dr. K H Mamatha
Assistant Professor
Department of Civil
Engineering
SIT, Tumakuru– 572103

11/07/2021 Department of CIVIL Engineering, 1

Siddaganga Institute of Technology,
Tumakuru, Karnataka, India
Sl. No. Topic Slide No.
1 INTRODUCTION 3~4
2 LITERATURE SUMMARY 5~6
3 CASE STUDY 7~9
4 DATA SETS & INFORMATION 10~11
5 METHODOLOGY 12~13
6 PROS & CONS & LIMITATIONS 14
7 RESULTS AND DISCUSSION 15~16
8 CONCLUSIONS 17
9 REFERENCES 18~19

11/07/2021 Department of CIVIL Engineering, 2

Siddaganga Institute of Technology,
Tumakuru, Karnataka, India
 INTRODUCTION
 Water is essential for life and a crucial resource for drinking, irrigation and industrial uses.

 In the semi-arid and arid regions with annual rainfall of generally less than 250 mm where surface
water resources are limited and groundwater is the most important source of water.

 Changes in groundwater quality can be caused by altering groundwater flow characteristics which
can lead to dissolution and transportation of different minerals within aquifers.

 The geochemical processes influence groundwater quality for the effective management of water resources and
for the management and protection of aquatic environments

11/07/2021 Department of CIVIL Engineering, 3

Siddaganga Institute of Technology,
Tumakuru, Karnataka, India
 INTRODUCTION
 Artificial intelligence (AI) is a branch of computer sciences that, in its broadest sense, would mean the
ability of a machine to mimic the operations of human brain to understand and apply thinking
methodologies.
 AI methods are greatly suited to dynamic nonlinear-system modeling and are widely used in various fields
of science.
 It is important to determine the processes that influence water quality parameters by means of effective
assessment tools, including the use of modelling techniques Modeling of water quality parameters has many
benefits.
 Artificial Neural Networks (ANN), Fuzzy Logic (FL) and Adaptive Neuro Fuzzy Inference System (ANFIS)
are some intelligent methods which have been successfully used in water engineering and for estimating
water quality.
11/07/2021 Department of CIVIL Engineering, 4
Siddaganga Institute of Technology,
Tumakuru, Karnataka, India
 LITERATURE SUMMARY
1. ANN consists of an information-processing paradigm and a pattern recognition tool inspired by how the biological
nervous systems process information.

2. MNN has the ability to learn a disintegrated convoluted function faster than a typical multilayer ANN Since the input
output response patterns between the distribution of nitrate concentration in ground water and the relevant processes are
not well behaved, MNN is expected to be more suitable for such problems.

3. MNN consists of modules called expert networks that compete to learn different aspects of a problem. In addition,
MNN has an integrating unit called a gating network that assigns different features of the input space to the different
expert networks.

4. The quality of ground water defines its usefulness for drinking, irrigation, and industrial uses.
11/07/2021 Department of CIVIL Engineering, 5
Siddaganga Institute of Technology,
 LITERATURE SUMMARY
5. To standardize the model, the entire data set is segregated into three phases. The first phase is the learning
phase which is utilized to train the network. The goal of training is to guarantee that the network replicates the
inherent characteristic of the information availed in the ANN modelling. ANN weights and biases are fixed during
the training procedure. The input variables and already ascertained output parameters decide the associated weights
in such a way that the predicted and observed values are in agreement.

6 Three statistical evaluation criteria,namely,the determination coefficient(R2),the scatter index(SI) and the Nash-Sutcliffe
coefficient (NS) were used to assess the employed models

11/07/2021 Department of CIVIL Engineering, 6

Siddaganga Institute of Technology,
Tumakuru, Karnataka, India
 CASE STUDY
The management and protection of groundwater resources is of great importance in countries such as Iran, located in arid and
semiarid regions lacking surface water resources. Thus, preservation in addition to short and long-term planning of resources, is
necessary to optimize their efficiency. Baha bad is located in Iran’s central plateau river basin, and groundwater is the main water
source. In this region, the quality of water has been threatened in recent years by agricultural and industrial activities, and
researchers have given more attention to quality parameters like EC, Cl, SO4 and NO3. The utilization, management and protection
of ground water should be a basic principle of Country planning. In this regard, the model can be used as an efficient tool for
managers. A neural network model is a conceptual model and in fact, is a simplified image of the mathematical model. The biggest
problem faced by users and suppliers of mathematical models, is the need of these models to exact various input data. Artificial
neural networks (ANN) which are driven by biological neural networks can help in solving such problems. These networks are a
part of the intelligent systems, developed with various spread structures. This study used an artificial neural network in estimating
the quality parameters and compared them with the measured values, in order to evaluate the influence of key input parameters to
the neural network.

11/07/2021 Department of CIVIL Engineering, 7

Siddaganga Institute of Technology,
Study Area: -Baha bad plain is one of the plains of Yazd Province in Central Iran. The geographical location of
this plain is 56º20' to 56º56' east longitude and 31º40' to 32º16' north latitude . Its center is Baha bad city, located
85 km northeast of Bafq City. This city covers an area of 8 km2 and has an altitude of 1398 m above sea level.
The annual long term precipitation rate of the region is 55.1 mm; average temperatures are maximum (26ºC) and
minimum (10ºC). Baha bad plane is located between the heights in the north west and east south, with
remarkable ground tables which are used for agricultural development, animal breeding and drinking water. Due
to the paucity of precipitation and over use of water in the region, the annual loss of groundwater amounted to 25
cm withdrawal, and resulted in an annual regional ground water deficit of 8-12 million m 3.

As above similar Case Study of AI-based models for estimating groundwater quality parameters of the Khezri
plain, Eastern Iran taken as another example.
11/07/2021 Department of CIVIL Engineering, 8
Siddaganga Institute of Technology,
Tumakuru, Karnataka, India
 A case study of Baha bad plain, Yazd, Iran

Location of study region (Bahabad Case Study) Relationship between Observed and simulated data for
the best ANN model for Cl, EC,So4,No3 parameters

11/07/2021 Department of CIVIL Engineering, 9

Siddaganga Institute of Technology,
Tumakuru, Karnataka, India
 DATA SETS
The processing phase is very important in data analysis to improve the data quality. In this research, it was necessary to gather
data for training purpose, therefore data gathered from Yazd Regional Water Company and 260 samples of 13 wells were
assessed in the Baha bad plain from 2003 to 2013. Model input parameters included: (Cl, EC, NO3, SO4, HCO3, Anion,
Cation, TDS, TH, pH, SAR, % Na, K, Mg, Ca, Evaporation, Water level, Q) and the model output parameters included: (Cl,
EC, NO3, SO4). Artificial neural network modeling was developed using MATLAB R2013 software package. In this
research, 70% of inputs were used for training and the remaining 30% were used for test. At first, input data were normalized
to the range 0–1, then the best number of epochs were estimated for each output parameter, and the root mean square error
value used for the stopping criterion was set as 0.01. The best combination of inputs was estimated for Cl, EC, NO3, and
SO4. The Coefficient of determination (R2), root mean squared error (RMSE) and Nash–Sutcliffe model efficiency
coefficients were used for modeling Cl, EC, NO3, SO4 for different ANN models with the best combination of inputs and
best number of neurons in the hidden layer.
29/01/2020 Department of CIVIL Engineering, 10
Siddaganga Institute of Technology,
Tumakuru, Karnataka, India
and the output layer by modifying the number of hidden layers and the learning rate.
Feed forward back propagation neural networks (FFN-BP) are relatively new tools in
the earth sciences.

INFORMATION: Genetic programming (GP) was used to determine relationships

between groundwater quality parameters including total hardness (TH), total dissolved
solids (TDS) and electrical conductivity (EC) for 240 groundwater samples collected
from 12 wells in the Khezri plain, eastern Iran. The artificial neural networks (ANN)
and adaptive neuro-fuzzy inference system (ANFIS) methods were also developed to
further verify the estimation capability of GP by a comparison of estimated and
observed values of chemical parameters

11/07/2021 Department of CIVIL Engineering, 11

Siddaganga Institute of Technology,
Tumakuru, Karnataka, India
 Artificial intelligence method
METHODOLOGY
A)Network-based models’ method
1. Artificial Neural Network (ANN).
2. Adaptive Neuro-fuzzy Inference System (ANFIS).
3. Genetic Programming (GP).

 B) Classification-based analysis models method

4. Support Vector Machine (SVM).
  C) Regression-based analysis models method
1. Multivariate Adaptive Spline Regression (MARS).

  D) Ensemble tree-based technique (Random Forests, RF). method

  E) Integrative models method

1. Embedded Firefly Algorithm with ANN and SVM, ANN-FA and SVM-FA).
11/07/2021 Department of CIVIL Engineering, 12
  Siddaganga Institute of Technology,
Tumakuru, Karnataka, India
 NUMERICAL FORMULA
Co–efficient of Determination (R2)
R2 describes the degree of association between experimental data and those obtained by the model

Root mean squared Error (RMSE)

RMSE indicates the goodness-of-fit between predicted and observed water quality parameter values

 
Mean Absolute Relative Error (MARE)
MARE measures the average relative discrepancy and not only gives the performance index in terms of
estimating hydro chemical parameters but also the distribution of the estimation errors.

11/07/2021 Department of CIVIL Engineering, 13

Siddaganga Institute of Technology,
Tumakuru, Karnataka, India
 ADVANTAGES
1. ANN uses input output response patterns to map a function approximation to the underlying governing rules of the
output responses corresponding to specific inputs in a convoluted physical space.
2. The capability to adapt to recurrent changes and detect patterns in a complex natural system is found to be one of
the most important feature of ANN models.
DISADVANTAGES
1. Depletion of groundwater is also linked with the vulnerability of an aquifer to chemical contamination.
2. For example, groundwater depletion is one of the causes for arsenic contamination in some parts of the world.
3. The Artificial Intelligence models are not economical.
LIMITATIONS
1. Data mining with machine learning approaches has been effectively popularized due to their ability to analyze
the multifarious relationship between predictors and response.
2. Based on the results considering other models, RF model has the best reliability during the test phase.
11/07/2021 Department of CIVIL Engineering, 14
Siddaganga Institute of Technology,
Tumakuru, Karnataka, India
 RESULTS & DISCUSSION
• The results of RMSE, COREL, N and R2 for different ANN models with the best combination of inputs and number of
neurons in hidden layer were shown. The number of epochs was set as 200 for Cl, and the optimum combination of inputs
was a combination of seven parameters: anionic, Cation, TDS, TH, SAR, Na, and Mg, the best number of neurons in the
hidden layer was 19. The number of epochs was set as 100 for EC, and the optimal combination of inputs was a
combination of nine parameters: Anion, Cation, TDS, TH, SAR, K, Na, Mg, and Ca, and the best number of neurons in the
hidden layer was 15. The number of epochs was set as 400 for NO3, and the best combination of inputs was a combination
of two parameters: HCO3 and Evaporation, and the best number of neurons in the hidden layer was 7. The number of
epochs was set as 400 for SO4. The best combination of inputs was a combination of four parameters: Anion, Cation, TH,
Na and the optimum number of neurons in the hidden layer was 10. The relationship between the estimated and observed
values of the parameters Cl, EC, NO3, and SO4, are modeled

11/07/2021 Department of CIVIL Engineering, 15

Siddaganga Institute of Technology,
Tumakuru, Karnataka, India
 RESULTS & DISCUSSION
•Eighteen different parameters were used as input and four output parameters were examined. Five training algorithms were examined, namely train

GD, train GDA, train GDX, train SCG and train LM with different numbers of neurons in the hidden layer. The optimal training algorithm for Cl

was train SCG and the values of RMSE, COREL, N and R2 were 0.04 0.98, 0.96 and 0.96, respectively by the best combination of inputs. For EC,

train SCG was the optimum training algorithms used and the values of RMSE, COREL, N and R2 were 0.033, 0.99, 0.98 and 0.98, respectively.

For NO3, train LM was the desirable training algorithm and this result is similar to the research of Venkat Kumar et al. (2010), in which the values

of RMSE, COREL N and R2 were 1.25, 0.3, -2.1 and 0.096, respectively. The number of neurons in the hidden layer was equal to double the

number of inputs plus one for all parameters and this result is in line with the research. e. The maximum desired EC parameter is 1500 μmohs/cm,

and the maximum allowable value is 2000 μmohs/cm, so the increase of this parameter, which has led to changes, has increased the salinity of

water in recent decades. Therefore, underground aquifer changes can cause instability, and low groundwater level also affects water quality, and

could be a subject for discussion at the Earth Summit.

11/07/2021 Department of CIVIL Engineering, 16

Siddaganga Institute of Technology,
Tumakuru, Karnataka, India
 CONCLUSION
AI Models improved the precision of water quality parameter simulation with minimal computation compared to the
FNN and CFN. The obtained results are capable of providing constraints for implementation in future management,
planning, and remedial measures, and the monitoring of natural groundwater where groundwater qualities have been
threatened by various anthropogenic activities.

The geostatistical methods are based on the spatial location of variables, but the accuracy in estimating the variable
depends on the number of samples in the region, so in areas such as the present study, which is limited to the
number of samples, it is difficult to find out the relationship between spatial location of sampling and variable,
while intelligent models were more capable of extracting this relationship. Finally, NF-GP can be introduced as an
appropriate and efficient tool for deter- mining the quality parameters in the study area in order to monitor quality

11/07/2021 Department of CIVIL Engineering, 17

Siddaganga Institute of Technology,
Tumakuru, Karnataka, India
 REFERENCES
 
• Arya far, A., Khosravi, V., Zarepourfard, H., & Rooki, R. (2019). Evolving genetic programming and other AI-based models for estimating
groundwater quality parameters of the Khezri plain, Eastern Iran. Environmental earth sciences, 78(3), 69.

 
• Maroufpoor, S., Fakheri-Fard, A., & Shiri, J. (2019). Study of the spatial distribution of groundwater quality using soft computing and
geostatistical models. ISH Journal of Hydraulic Engineering, 25(2), 232-238.
 
• Haghbin, M., Sharafati, A., Dixon, B., & Kumar, V. (2020). Application of soft computing models for simulating nitrate contamination in
groundwater: comprehensive review, assessment and future opportunities. Archives of Computational Methods in Engineering, 1-23.
 
• Maroufpoor, S., Jalali, M., Nikmehr, S., Shiri, N., Shiri, J., & Maroufpoor, E. (2020). Modeling groundwater quality by using hybrid
intelligent and geostatistical methods. Environmental Science & Pollution Research, 27(22).

11/07/2021 Department of CIVIL Engineering, 18

Siddaganga Institute of Technology,
Tumakuru, Karnataka, India
 REFERENCES  
• Yeh, J. R., Shieh, J. S., & Huang, N. E. (2010). Complementary ensemble empirical mode decomposition: A novel noise enhanced data analysis method. Advances in adaptive data
analysis, 2(02), 135-156.

• Hussein, E. A., Thron, C., Ghaziasgar, M., Bagula, A., & Vaccari, M. (2020). Groundwater prediction using machine-learning tools. Algorithms, 13(11), 300.

• Nathan, N. S., Saravanane, R., & Sundararajan, T. (2017). Application of ANN and MLR models on groundwater quality using CWQI at Lawspet, Puducherry in India. Journal of
Geoscience and Environment Protection, 5(03), 99.
 
• Shiri, N., Shiri, J., Yaseen, Z. M., Kim, S., Chung, I. M., Nourani, V., & Zounemat-Kermani, M. (2021). Development of artificial intelligence models for well groundwater quality
simulation: Different modeling scenarios. Plos one, 16(5), e0251510.
 
• Khaki, M., Yusoff, I., & Islami, N. (2015). Application of the artificial neural network and neuro ‐fuzzy system for assessment of groundwater quality. CLEAN–Soil, Air, Water, 43(4), 551-
560.
 
• Machiwal, D., Cloutier, V., Güler, C., & Kazakis, N. (2018). A review of GIS-integrated statistical techniques for groundwater quality evaluation and protection. Environmental Earth
Sciences, 77(19), 1-30.

 
11/07/2021 Department of CIVIL Engineering, 19
Siddaganga Institute of Technology,
Tumakuru, Karnataka, India
 THANK YOU

11/07/2021 Department of CIVIL Engineering, 20

Siddaganga Institute of Technology,
Tumakuru, Karnataka, India