You are on page 1of 13

REVIEW - ARTIFICIAL INTELLIGENCE BASED MODELLING OF

HYDROLOGICAL PROCESSES

Jawad S. Alagha1, Md Azlin Md Said2, Yunes Mogheir3

1: Graduate Student, School of Civil Engineering, Universiti Sains Malaysia (USM), Malaysia,
jawad_s78@yahoo.com.
2: Professor, School of Civil Engineering, Universiti Sains Malaysia (USM), Malaysia,
azlin@eng.usm.my.
3: Associate Professor, Environmental Engineering Department, Islamic University of Gaza (IUG),
Palestine, ymogheir@iugaza.edu.ps.

ABSTRACT

Hydrological processes such as runoff and contaminant transport are usually affected by
various complex interrelated variables. Moreover, uncertainties in variables estimate are
the common stamp of these processes. Due to this complex nature, Physical modeling of
any hydrological system requires availability of large, accurate and detailed data related to
all influencing variables, which are not always available due to financial and technical
constraints. This may lead to deficiencies in model’s performance which in turn, negatively
affect hydrological planning and policy drawing. To address these shortcomings, artificial
intelligence (AI) based techniques have been recently used as alternative tools to
traditional physical hydrological models. These techniques have been proved to be
successful and effective in tackling wide spectrum of challenging hydrological processes.
This article is intended to serve as an introductory review of application of two AI
techniques namely, artificial neural networks (ANNs) and support vector machine (SVM)
in various hydrological applications. In this article, ANNs and SVM theoretical
background coupled with their strength points that make them suitable for hydrological
modeling were briefly described. Moreover, various examples of successful applications of
ANNs and SVM for modeling different hydrological processes were also provided.

KEYWORDS: Artificial Neural Networks (ANNs), Hydrology, Prediction, Support


Vector Machine (SVM).
I. INTRODUCTION

Hydrology can be defined as the scientific study of water and its properties, distribution,
and effects on the earth’s surface, soil, and atmosphere [1]. Most hydrological processes
are characterized by high degree of spatial and temporal variability. Moreover,
uncertainties in hydrological variables’ estimates are usually the main feature of these
processes [2]. A lot of hydrological processes are not clearly understood as well, where
these processes are affected by various interrelated variables [3]. Hydrologists are often
confronted by these characteristics when they are required to provide rational answers to

Copyright © 2012 IUG. The 4th International Engineering Conference –Towards engineering of 21st century

1
problems related to different hydrological features such as flooding, precipitation,
contaminant transport, and so on [4]. For almost all hydrological problems, physical
models were the default modeling tools worldwide during the last decades [5]. However,
the accuracy of these models highly depends on the availability of detailed and accurate
data about the complex hydrological system properties, which are not usually available due
to cost and time limitations especially in the developing countries resulting in model
uncertainties and unsatisfied performance, which in turn result in insufficient water
resources management decisions [6, 7].

The aforementioned limitations led to use artificial intelligence (AI) as alternative


modeling tools. Many techniques can be categorized under this newly available approach
such as artificial neural networks (ANNs) and support vector machine (SVM). These
techniques have become very popular and efficient for modelling complicated hydrological
processes using relatively less cost, effort and data These approaches have proven to be
highly efficient in solving the problems when the rules to solve such problems are difficult
to be expressed which is the case in most hydrological processes [8]. Therefore, AI
techniques have become attractive alternatives to traditional physical modeling techniques
for many hydrological applications [8, 9].

The objective of this review paper is to highlight the utilization of two AI techniques
namely, artificial neural networks (ANNs) and support vector machine (SVM) as modeling
and prediction tools in various hydrological applications.

II. THEORETICAL BACKGROUND OF ANN AND SVM


Artificial intelligence (AI) is an artificial system that mimics the internal behavior of
human’s brain such as perception, learning and reasoning [8]. Many techniques can be
categorized as AI techniques, each of which is based on different theoretical background.
In this study the light will be concentrated on ANNs and SVM. A brief overview of these
two techniques is presented as follows.
A. Artificial Neural Networks (ANNs)

• Overview

ANNs are computing systems whose compositions are inspired from the analogy of
biological neural networks in the human brain [۱۰, 1۱]. ANNs are simply nonparametric
statistical tools that can perform many tasks including pattern classification, clustering,
function approximation, prediction, optimization, and control [12, 13].

As shown in Figure 1, ANNs’ model is usually composed of three layers: input layer,
output layer, and one or more hidden layer. Each layer consists of one or more basic simple
interconnected elements called neurons or nodes. A neuron is a processing element that
takes a number of inputs, weighs them, sums them up, and adds a bias; the result is
modified by the transfer function (also called the activation function), and the final result

Copyright © 2012 IUG. The 4th International Engineering Conference –Towards engineering of 21st century

2
of this process is the neuron’s output. Each neuron in any layer is connected to all neurons
in the next layer; each link is given a weight representing its connection strength [14].

Input Hidden Output


Layer Layer Layer

Input # 1

Input # 2
Output
Input # 3

Input # 4

Figure 1: ANNs’ structure with one hidden layer [8]

The main difference between various ANNs types is the network architecture which is
related to how ANNs’ components are put together [15,16]. Network architecture
determines the connection pattern, number of connection weights, and the way information
flow through the network, and others. Determination of the appropriate network
architecture significantly influences the performance of ANNs, thus it is considered the
most important point in ANNs’ model development. However, it is also the most difficult
task in the development of ANNs’ model [17,18].

• ANNs Training Process


Training is the main process that enables ANNs’ model to capture the complex
relationships between the known input data and the desired output data which cannot be
easily analyzed using other approaches [19]. Instead of following a set of rules specified by
the modeler as the case of physical models, ANNs are trained to solve the problem by
learning the underlying rules or pattern (input–output relationships) from the given
representative examples [13]. Then, a new mathematical relationship specific to the
problem under question is developed by ANNs [16].

Model generalization is achieved through repeatedly modification of the network


architecture using a training algorithm or learning rules [18]. The training phase usually
terminates when achieving one of various stopping criteria which are [20]: (1) when a
fixed number of training cycles is presented to the network, or (2) when the training error,
(the difference between ANNs’ model calculated output and the actual measured output),
reaches a predetermined small value, or (3) when there is no further improvement in the
model’s prediction.

The development of ANNs requires partitioning of the input database into three subsets:
training, test, and validation. The training subset is used in the training phase to update the
weights of the network [21]. The test subset is used during the training process to check

Copyright © 2012 IUG. The 4th International Engineering Conference –Towards engineering of 21st century

3
the network response of untrained (new) data. The data used in the test subset should be
distinct from those used in the training, but the data should lie within the training data
domain (boundaries). Based on the performance of the ANNs’ model on the test subset, the
model architecture may be changed. The third subset of data is the validation subset, which
should include examples that are different from those in the other two subsets. This subset
is used after selecting the best network architecture to further examine the network or
confirm its accuracy before it is used [9].

In hydrology, the inputs of ANNs’ model can be any causal variable, such as rainfall,
temperature, water levels, elevation, contaminant loads, and meteorological data. The
output can be the hydrological responses, such as runoff and contaminant concentrations
[2]. The choice of input variables is generally based on a priori knowledge on causal
variables, inspections of time series plots, and statistical analysis of potential inputs and
outputs [15].

• Characteristics of ANNs’ Models

ANNs have many characteristics that enable them to be effective tools for modeling
different processes especially in hydrology. These characteristics include the following [2,
9, 10, 12, 22, 23]:
• ANNs have the ability to recognize relationships between input and output variables
without deep knowledge of the underlying physical principles.
• ANNs are nonlinear, this allows better fit to data, and provides ANNs the capability of
approximating almost any input–output relationship.
• ANNs are insensitive to noise; hence, relatively accurate predictions could be
achieved in the presence of uncertain data and measurement errors.
• ANNs can interpret information from multiple variables or parameters.
• ANNs learning and adaptability allow systems to update (modify) their internal
structure in response to changing environments.
• ANNs have generalization ability that enables application of the model to unlearned
data.
Despite the aforementioned tremendous advantages, strong theoretical and practical
capabilities, and successful applications in a wide variety of tasks in many fields, ANNs
sometimes have numbers of limitations and drawbacks that should not be ignored; these
include [2, 24, 25, 26, 27]:
• The performance of ANNs’ models is dependent on the quality and quantity of
available data. This requirement cannot be easily met in many cases.
• There is no standardized way of selecting model parameters, such as network
architecture and training algorithm.
• The generalization ability of ANNs is dependent on network topology and the
selection of key network parameters, such as activation function, learning rate, and
momentum, which mainly depend on trial and error.

Copyright © 2012 IUG. The 4th International Engineering Conference –Towards engineering of 21st century

4
The aforementioned drawbacks can be treated by using general guidelines set by ANNs
experts and by associating ANNs with suitable hybrid systems that compensate for the
their deficiencies.

• ANNs Applications in Hydrology

A lot of examples that discuss different applications of ANNs for hydrological prediction
and forecasting can be found in the literature. For example, Maier et al. [28] mentioned
that 516 ANNs related papers had been published in 19 widely recognized hydrology and
water resources international journals during the period from 1999 to 2007. Tables (1) and
(2) illustrate very few examples of various applications of ANNs in the field of
groundwater and surface water respectively. These examples are related to both water
quantity and quality.

• Comments on ANNs Applications in Hydrology

The presented examples are very limited compared with the actual ANNs applications in
different hydrological fields, but these are considered as samples that demonstrate the
ability of ANNs in tackling complex hydrologic problems.
As obvious from Table (1), time-series input data records that describe the real system are
the main requirement for developing a reliable ANNs’ model. It could also be noticed that
different ANNs’ types, architectures, and training algorithms have been used for each
application; this is related to the varying natural of each hydrological problem. Hybrid
models that combine ANNs with other techniques have also effectively used due to their
ability to overcome some ANNs’ shortcomings.

Copyright © 2012 IUG. The 4th International Engineering Conference –Towards engineering of 21st century

5
Table 1: Examples of ANNs applications in groundwater modelling

Description of the Model Type and


Research Topic Input variables Model Output Ref.
application Training Algorithm
ANNs based model was Precipitation, temperature, Prediction Feed-forward neural [3]
Forecasting of developed to simulate the groundwater level, stream flow. of the well level at network trained with the
groundwater decreasing trend of the Each one had 5 monthly time- time step t+1 up to 18 Levenberg–Marquardt
level groundwater level in Greece. lags which are t, t-1, t-2, t-3 and months ahead algorithm
t-4.
Prediction of ANNs’ model was used to Aquifer depth, pesticide leaching Estimation of Feed-forward with back- [29]
predict the concentrations of class, aquifer sensitivity to propagation training
pesticide, sampling time (month), pesticide
pesticide
pesticide in groundwater algorithm
leaching in monitoring wells in state of well and well screen depths and concentrations in
groundwater Illinois, USA water level
the sampled wells
Groundwater

ANNs’ model was applied for Permeability and thickness of Estimation the depth Feed-forward with back- [21]
Assessment of
estimating the depth of the cover layer, water table depth, of the contaminated propagation training
groundwater contaminated zone in the soil and soil surface–pollutant contact zone algorithm
contamination after a pollutant discharge at the time.
in road soil surface in the north of
projects France.

Prediction of ANNs’ model was developed to The easily measurable Prediction of nitrate Training algorithm is [14]
predict concentration of nitrate parameters such as temperature, concentration Levenberg–Marquardt.
nitrate in in groundwater monitoring wells electrical conductivity,
groundwater in of Harran Plain in Turkey. groundwater level and pH
ANNs have been utilized to Rainfall from 6 previous days, Estimation of the Multilayer perceptron [25]
Forecasting of simulate hydraulic head change temperature, day number, and hydraulic head change with two hidden layers
groundwater at an observation well in pumping rates from nearby wells of the observation with Back-propagation
level Edward’s karstic aquifer in during 4 previous days. well algorithm
Texas – USA

Copyright © 2012 IUG. The 4th International Engineering Conference –Towards engineering of 21st century

6
Table 2: Examples of ANNs applications in surface water modelling

Description of the Model Type and Training


Research Topic Input variables Model Output Ref.
application Algorithm
Modelling of ANNs’ model was developed for The daily average rainfall and Runoff prediction Comparison between various [30]
rainfall–runoff modelling rainfall–runoff stream flow data training algorithms. Real-
process process at the Kentucky River– coded genetic algorithm
USA (RGA) outperformed other
algorithms.
Stream flow A hybrid model composed of The past stream flow Forecasting Feed-forward multilayer [23]
time series traditional time series models stream flow perceptron.
forecasting and ANNs was used in stream
flow forecasting for Colorado
River, USA
Coastal water ANNs models were used for Station location (longitude, Salinity, General regression neural [15]
Surface Water

quality temporally and spatially latitude), past salinity, temperature, network and back propagation
prediction prediction of selected water temperature, dissolved oxygen, dissolved oxygen, training algorithm
quality variables of coastal and chlorophyll-a in and chlorophyll-a
waters bodies in Singapore. neighbouring stations at any location
River water Two ANNs’ models for 11 eleven input water quality Prediction of DO Three layer feed-forward [31]
quality computing the dissolved oxygen variables (such as pH, TS, COD, & BOD neural network with back
modelling (DO) and biochemical oxygen etc..) propagation training algorithm
demand (BOD) were developed
in Gomti river - India
Modelling of Different ANNs’ models were Rainfall and evaporation, Water level Comparison between different [32]
water level in developed for estimation of change models. Adaptive neuro-fuzzy
lake water level change in Beysehir inference system (ANFIS) had
Lake in Turkey the best results.
Modelling of ANNs model was developed for Rainfall durations, average Runoff peak flow Multilayer perceptron (MLP) [33]
rainfall–runoff modelling rainfall–runoff intensities and season index of
process process in 2 catchments in Iran. over 100 events

Copyright © 2012 IUG. The 4th International Engineering Conference –Towards engineering of 21st century

7
B. Support Vector Machines (SVM)

• Overview

Support Vector Machine (SVM) is a machine learning technique proposed by V. Vapnik, a


Russian mathematician, and his team in the early 1960s [34]. It is a new technique
compared with ANNs, whose the main idea is to use a function, called kernel function, that
projects input data from a low-dimensional space to a space of higher dimension, called a
feature space, such that the new high-dimensional feature space will become easily and
linearly separable by simple regression [35]. SVM has been developed based on the
statistical learning theory (SLT). The basic strength point of SLT is its dependence of the
structural risk minimization (SRM) principle that aims at minimizing the generalized
model error (which is sometimes called risk), rather than just minimizing the training set
error, which consequently highly increases SVM generalization ability [36, 37].

The main advantage of SVM is that both model complexity and prediction error are
simultaneously minimized [38]. In addition SVM has excellent generalization capability
and can achieve unique and globally optimal solution with limited information. These
features result in that SVM has gained popularity in many traditionally ANNs research
areas [36, 37, 39].

For developing SVM’s model, various parameters have to be determined and optimized in
advance. These parameters, which are called hyperparameters, include error function
parameter (ε), regularization parameter (c), and RBF parameter ( γ ) [37].

SVM constructs a new mathematical model specific to the problem under question in the
form:

Y = F(x) = w . Ф(x) + b (1)

Where x is an input vector, Y is a corresponding output value, w is a weight vector, and b is


a bias. Ф is a kernel nonlinear transfer function. Detailed theoretical background and basic
SVM’s formula can be found in many references especially [40].

• Characteristics of SVM

The aforementioned ANNs’ advantages are also valid for SVM. Furthermore, it has been
reported that SVM has extra advantages over ANNs in relation to generalization ability,
tackling complex variables interrelationships and filtering out the noise. High
generalization ability of SVM over ANNs may be related to the learning principles
embedded in both techniques. Where SRM principle of SVM minimizes error and model
complexity simultaneously which improve the generalization ability. On the other hand the
empirical risk minimization (ERM) learning theory embedded in ANNs can minimize only

Copyright © 2012 IUG. The 4th International Engineering Conference –Towards engineering of 21st century
8
training data’ error and cause the model to be over-trained which reduce the model
generalization ability [41]. However, SVM has some shortcomings, for instance, choosing
of the suitable kernel function and hyperparameters are heuristic and depend on trial and
error which is a time-consuming approach [35, 36].

Many studies have compared the performance of both SVM and ANNs, and most of them
reported that SVM’s models are comparable or even outperforms ANNs in many
hydrological applications such as surface water problems [41].

• SVM Application in Hydrology

Studies that present SVM’s application in hydrological processes are relatively less than
that for ANNs because SVM is relatively new approach. Despite that there are growing
successful applications of SVM in the many hydrological fields, SVM’s application in
other fields are still few [41, 42]. Table (3) illustrates some examples of various
applications of SVM for hydrological related modeling.

III. CONCLUSIONS

Few examples of successful application of ANNs and SVM as two AI techniques for
hydrological modeling are presented in this paper just to highlight the efficiency of this
approach as a promising modeling approach for wide hydrological applications. The
characteristics of AI techniques make them highly suit tackling different hydrological
processes as presented in Table (4). Most studies concluded that SVM slightly
outperformed ANNs especially in its generalization ability due to its theoretical basis.
Like any scientific technique, AI techniques have some shortcomings that sometimes
constraint their performance. Therefore, hybrid models that combine one of AI techniques
with other tools seem to be the most promising future direction in hydrological modelling
field. In addition, enhancement of AI techniques and their optimization procedures are
currently a research focus point.
Finally, it could be concluded that AI techniques can be considered as appropriate
alternatives to traditional physical techniques for dealing with hydrological problems
especially in the developing countries such as Palestine. In such cases, detailed and
accurate data about the variables expressing hydrological processes are not usually
available due to technical and financial constraints. Therefore, related institutions have to
encourage and support researches that tackle the applications of various AI techniques in
hydrological and environmental fields. In this context, this review paper is a part of an
ongoing research aims at developing a groundwater management model using AI
techniques that could be effectively applied in Gaza strip.

Copyright © 2012 IUG. The 4th International Engineering Conference –Towards engineering of 21st century
9
Table 3: Examples of SVM’s applications in hydrological modelling

Research
Description of the application Input variables Comments Ref.
Topic
SVM’s model was developed to forecast Local-climatological data with SVM is considered as a
both 6-month ahead annual flow volume different time lags and past flow promising tool for solving site-
Stream flow
and 24-hour ahead hourly stream flow. The volumes and stream flow specific, real-time water [36]
predictions models were applied in Sevier River Basin, resources management problems.
Utah, USA.
SVM’s model was applied for forecasting Rainfall and flow data The real problem of SVM’s
Flood flood events over the Bird Creek catchment model development is the
[34]
forecasting in the USA. selection the input variables
combinations
Identification The suitability of ANN, SVM and logistic GW depth, recharge, aquifer, ANNs and SVM gave
regression integrated with GIS was assessed
vadose zone and soil media satisfactory performance with
of wells
for identifying nitrate contamination of characteristics, topography, slight advantage of ANNs that [42]
contaminated wells in Polk County, Florida, USA. hydraulic conductivity, landuse, outperformed SVM
by nitrate pH, hydrologic group and pedality.
Groundwater SVM’s model was applied for classification Groundwater quality classification SVM offers high prediction
of 8 water quality parameters into 5 grades indicators accuracy and is a reasonable and
quality at the Niangziguan fountain, China. feasible assessment method.
[39]
assessment
SVM and ANNs’ models were developed Various inputs and their time lags Both models were comparable,
for prediction of transient groundwater such as mean daily and cumulative however SVM slightly
Groundwater
levels in a complex groundwater system for pumping rate for 3 production outperformed ANNs particularly
level (GWL) [37]
various prediction horizons (daily, weekly, wells, precipitation, mean daily for longer prediction horizons,
forecasting biweekly, monthly, and bimonthly) in temperature, and initial water level especially when less input data
Towaco aquifer, New Jersy, USA. of monitoring wells examples were available.
Time-series models were developed using Past ground water level (GWL), Generalization ability of the
Groundwater SVMs and ANNs that predict (GWL) precipitation, and tide level SVMs model was better than the
level (GWL) fluctuations in 2 wells at a coastal aquifer in ANNs’ model. [41]
forecasting Korea.

Copyright © 2012 IUG. The 4th International Engineering Conference –Towards engineering of 21st century
10
Table 4: Matching between the characteristics of hydrological processes and AI capabilities

Factor Characteristics of Hydrological processes AI capabilities


Number of input
Influenced by large number of input variables Able to deal with any number of input variables
variables
Nature of input variables Input data are space and time dependent Able to deal with any type of data.
Relationship between Prove powerful capabilities in modeling nonlinear
Variables have non-linear interrelationships each other
variables relationships and complex systems
Missing and incomplete Input data may be incomplete and some data may be Illustrate satisfactory results in modeling
input data missing due to monitoring irregularity relationships with missing input data
Deal with the problem as a black box where AI allow
modeling of any system without prior understanding
Nature of the problem This type of problems are not clearly understood
of theoretical background of the problem or
relationships between variables
Models’ structure is flexible enough to allow the
system to update (modify) its internal structure in
Nature of the problem Variables are subject to change
response to changing environment and accommodate
additional constraints.
No need to make assumptions about the
Due to uncertainty and lack of theoretical background,
Need for assumptions assumptions are inherent in the hydrological processes.
mathematical form of the relationship between input
and output
Data collection is complex, periodic, costly, and time Have powerful generalization capabilities based on
Data needed consuming process less data than physical models
Output results need to be presented using clear
Output can be presented and visualized by multi
Output presentation presentation to be utilized for decision making related
form presentations
measures
Integrations between May require integration of more than one technique to
Easily linked and integrated with other techniques
techniques utilize the strength points of various techniques

Copyright © 2012 IUG. The 4th International Engineering Conference –Towards engineering of 21st century
11
IV. REFERENCES
[1] McCuen, R. H.: Hydrologic analysis and design, 2nd Ed., Prentice Hall, Upper Saddle River, N.J.
1997.
[2] ASCE-I (American Society of Civil Engineers): Task Committee on Application of Artificial Neural
Networks in Hydrology, Artificial neural networks in hydrology, I: Preliminary concepts, Journal of
Hydrological Engineering, 5(2), 2000.
[3] Daliakopoulos, I. N.; Coulibaly, P.; Tsanis, I. K.: Groundwater level forecasting using artificial
neural networks. Journal of Hydrology, 309: 229–240, 2005.
[4] ASCE-II (American Society of Civil Engineers): Task Committee on Application of Artificial Neural
Networks in Hydrology, Artificial neural networks in hydrology, II: Hydrologic Applications,
Journal of Hydrological Engineering, 5(2), 2000.
[5] Javadi, A.; AL-Najjar, M.: Finite element modelling of contaminant transport in soils including the
effect of chemical reactions, Journal of hazardous materials, 143(3), p. 690-701, 2007.
[6] Krishna, B.; Satyajirao, Y.; Vijaya, T.: Modelling groundwater levels in an urban coastal aquifer
using artificial neural networks, Hydrological Processes, 22(8), p. 1180-1188, 2008.
[7] Coppola, Jr. E.A.; Rana, AJ.; Poulton, MM.; Szidarovszky, F.; Uhl, VW.: A neural network model
for predicting aquifer water level elevations, Ground Water, 43(2), p. 231-241, 2005.
[8] Chen, S.H.; Jakemana, A.J.; Norton, J.P.: Artificial intelligence techniques: an introduction to their
use for modelling environmental systems, Mathematics and computers in simulation, 78(2), p. 379-
400, 2008.
[9] Basheer, I.; Hajmeer, M.: Artificial neural networks: fundamentals, computing, design, and
application, Journal of Microbiological Methods, 43(1), p. 3-31, 2000.
[10] Jain, S. K.; Singh, V. P.; ASCE, F.; Genuchten, M. T. V.: Analysis of Soil Water Retention Data
Using Artificial Neural Networks, Journal of Hydrologic Engineering, 9: 415-420. DOI:
10.1061/(ASCE)1084-0699(2004)9:5(415), 2004.
[11] Deswal, S.; Pal, M.: Artificial Neural Network based Modeling of Evaporation Losses in Reservoirs,
World Academy of Science, Engineering and Technology, 39. On-line at:
http://www.waset.org/journals/waset/v39/v39-51.pdf, 2008.
[12] Besaw, L.; Rizzo, D.: Counterpropagation Neural Network for Stochastic Conditional Simulation:
An Application with Berea Sandstone, Seventh IEEE International Conference on Data Mining,
USA, DOI: 10.1109/ICDMW.2007.54, 2007.
[13] Jain, A.K.; Mao, J.; Mohiuddin, K.M.: Artificial neural networks: a tutorial, Comput, IEEE March,
31–44, 1996.
[14] Yesilnacar, M. I.; Sahinkaya, E.; Naz, M.; Ozkaya, B.: Neural network prediction of nitrate in
groundwater of Harran Plain, Turkey, Environ Geol, 56: 19–25, 2008.
[15] Palani, S.; Liong, S. Y.; Tkalich, P.: An ANN application for water quality forecasting. Marine
Pollution Bulletin, 56: 1586–1597, 2008.
[16] Starrett, S. K.; Najjar, Y.; Adams, G.; Hill, J.: Modeling pesticide leaching from golf courses using
artificial neural networks, Communications in Soil Science and Plant Analysis, 29 (19): 3093-3106,
1998.
[17] Chen, Y.; Chang, F.: Evolutionary artificial neural networks for hydrological systems forecasting,
Journal of Hydrology, 367: 125–137, 2009.
[18] Singh, R. M.; Datta, B.: Artificial neural network modeling for identification of unknown pollution
sources in groundwater with partially missing concentration observation data, Journal of Water
Resources Management, 21: 557–572, 2007.
[19] Lin, G. F.; Chen, G. R.: An improved neural network approach to the determination of aquifer
parameters, Journal of Hydrology, 316: 281–289, 2006.
[20] Maier, H. R.; Dandy, G. C.: Neural Network Based Modeling of Environmental Variables: A
Systematic Approach, Mathematical and Computer Modeling, 33: 669-682, 2001.
[21] Tabach, E. E.; Lancelot, L.; Shahrour, I.; Najjar, Y.: Use of artificial neural network simulation
meta-modeling to assess groundwater contamination in a road project, Journal of Mathematical and
Computer Modeling, 45: 766–776, 2007.

Copyright © 2012 IUG. The 4th International Engineering Conference –Towards engineering of 21st century
12
[22] Rajanayak, C.; Samarasinghe, S.; Kulasiri, D.: Solving the Inverse Problem in Stochastic
Groundwater Modeling with Artificial Neural Networks, Journal of Hydraulic Engineering, 127(5):
154-159, 2001.
[23] Jain, A.; Kumar, A. M.: Hybrid neural network models for hydrologic time series forecasting,
Applied Soft Computing, 7: 585–592, 2007.
[24] May, D. B.; Sivakumar, M.: Prediction of urban stormwater quality using artificial neural networks,
Environmental Modeling & Software, 24: 296–302, 2009.
[25] Trichakis, I. C.; Nikolos, I. K.; Karatzas, G. P.: Artificial Neural Network (ANN) Based Modeling
for Karstic Groundwater Level Simulation, Water Resources Management, DOI 10.1007/s11269-
010-9628-6, 2010.
[26] Wong, B. K.; Bodnovich, T. A.; Selvi, Y.: Neural network applications in business: A review and
analysis of the literature (1988-95), Decision Support Systems, 19: 301-320, 1997.
[27] Abraham, A.: Meta learning evolutionary artificial neural networks, Neurocomputing Journal, 56: 1
– 38, 2004.
[28] Maier, H. R.; Jain, A.; Dandy, G. C.; Sudheer, K.P.: Methods used for the development of neural
networks for the prediction of water resource variables in river systems: Current status and future
directions, Environmental Modelling & Software, 25: pp 891-909, 2010.
[29] Sahoo, G. B.; Ray, C.; Mehnert, E.; Keefer, D., A.: Application of artificial neural networks to assess
pesticide contamination in shallow groundwater, Science of the Total Environment, Vol. 367, pp
234–251, 2006.
[30] Srinivasulu, S.; Jain, A.: A comparative analysis of training methods for artificial neural network
rainfall–runoff models, Applied Soft Computing, 6: 295–306, 2006.
[31] Singh, K.P.; Basant, A.; Malik, A.; Jain, G.: Artificial neural network modeling of the river water
quality—A case study, Journal of Ecological Modeling, 220: 888–895, 2009.
[32] Yarar, A.; Onucyıldız, M.; Copty, N. K.: Modelling level change in lakes using neuro-fuzzy and
artificial neural networks, Journal of Hydrology, 365: 329–334, 2009.
[33] Rakhshandehroo, G., R.; Vaghefi, M.; Shafiee, M. M.: Flood forecasting in similar catchments using
neural networks, Turkish J. Eng. Env. Sci, 34: pp 57 – 65, 2010.
[34] Han, D.; Chan, L.; Zhu, N.: Flood forecasting using support vector machines, Journal of
Hydroinformatics 09.4, IWA Publishing. doi: 10.2166/hydro.2007.027, 2007.
[35] Noble, W. S.: What is a support vector machine?, Nature Biotechnology, Vol. 24, No. 12. Nature
Publishing Group, 2006.
[36] Asefa, T.; Kemblowski, M.; McKee, M.; Khalil, A.: Multi-time scale stream flow prediction: The
support vector machines approach, Journal of Hydrology, 318, 7-16. 2006.
[37] Behzad, M.; Asghari, K.; Coppola, E.: Comparative Study of SVMs and ANNs in Aquifer
Water Level Prediction, Journal of Computing in Civil Engineering, ASCE, Vol. 24, No. 5,
pp.408-413, 2010.
[38] Kanevski, M.; Parkin, R.; Pozdnukhov, A.; Timonin, V.; Maignan, M.; Demyanov, V.; Canu,
S.: Environmental data mining and modeling based on machine learning algorithms and
geostatistics, Environmental Modelling & Software, Vol. 19, pp. 845–855, 2004.
[39] Junping, L.; Mingqi, C.; Xiaoyan, M. A.: Groundwater Quality Assessment Based on Support
Vector Machine, Paper funded by Global Environment Fund (GEF) Integral Water Resource
and Environment Management of Haihe River basin (MWR-9-2-1),”111”Introducing
Intelligence Project(B08039), 2009, www.seiofbluemountain.com/search/detail.php?id=5037.
[40] Vapnik, V.: The Nature of Statistical Learning Theory, Springer-Verlag, New York, 1995.
[41] Yoon, H.; Jun, S.; Hyun, Y.; Bae, G.; Lee, K.: A comparative study of artificial neural networks
and support vector machines for predicting groundwater levels in a coastal aquifer, Journal of
Hydrology, Vol. 396, pp 128–138. doi:10.1016/j.jhydrol.2010.11.002, 2011.
[42] Dixon, B.: A case study using support vector machines, neural networks and logistic regression
in a GIS to identify wells contaminated with nitrate-N, Hydrology Journal, 17, 1507-1520,
2009.

Copyright © 2012 IUG. The 4th International Engineering Conference –Towards engineering of 21st century
13

You might also like