HR: 17:00h
AN: H54B-05    [Abstracts]
TI: Locating Groundwater Pollution Source using Breakthrough Curve Characteristics and Artificial Neural Networks
AU: Kumar, J
EM: jitendra@iitk.ac.in
AF: Master's Student, Department of Civil Engineering, Indian Institute of Technology Kanpur, Kanpur, UP 208 016 India
AU: * Jain, A
EM: ashujain@iitk.ac.in
AF: Assistant Professor, Department of Civil Engineering, Indian Institute of Technology Kanpur, Kanpur, UP 208 016 India
AU: Srivastava, R
EM:
AF: Associate Professor, Department of Civil Engineering, Indian Institute of Technology Kanpur, Kanpur, UP 208 016 India
AB: The identification of pollution sources in aquifers is an important area of research not only for the hydrologists but also for the local and Federal agencies and defense organizations. Once the data in terms of pollutant concentration measurements at observation wells become known, it is important to identify the polluting industry in order to implement punitive or remedial measures. Traditionally, hydrologists have relied on the conceptual methods for the identification of groundwater pollution sources. The problem of identification of groundwater pollution sources using the conceptual methods requires a thorough understanding of the groundwater flow and contaminant transport processes and inverse modeling procedures that are highly complex and difficult to implement. Recently, the soft computing techniques, such as artificial neural networks (ANNs) and genetic algorithms, have provided an attractive and easy to implement alternative to solve complex problems efficiently. Some researchers have used ANNs for the identification of pollution sources in aquifers. A major problem with most previous studies using ANNs has been the large size of the neural networks that are needed to model the inverse problem. The breakthrough curves at an observation well may consist of hundreds of concentration measurements, and presenting all of them to the input layer of an ANN not only results in humongous networks but also requires large amount of training and testing data sets to develop the ANN models. This paper presents the results of a study aimed at using certain characteristics of the breakthrough curves and ANNs for determining the distance of the pollution source from a given observation well. Two different neural network models are developed that differ in the manner of characterizing the breakthrough curves. The first ANN model uses five parameters, similar to the synthetic unit hydrograph parameters, to characterize the breakthrough curves. The five parameters employed are peak concentration, time to peak concentration, the widths of the breakthrough curves at 50% and 75% of the peak concentration, and the time base of the breakthrough curve. The second ANN model employs only the first four parameters leaving out the time base. The measurement of breakthrough curve at an observation well involves very high costs in sample collection at suitable time intervals and analysis for various contaminants. The receding portions of the breakthrough curves are normally very long and excluding the time base from modeling would result in considerable cost savings. The feed-forward multi-layer perceptron (MLP) type neural networks trained using the back-propagation algorithm, are employed in this study. The ANN models for the two approaches were developed using simulated data generated for conservative pollutant transport through a homogeneous aquifer. A new approach for ANN training using back-propagation is employed that considers two different error statistics to prevent over-training and under-training of the ANNs. The preliminary results indicate that the ANNs are able to identify the location of the pollution source very efficiently from both the methods of the breakthrough curves characterization.
DE: 1805 Computational hydrology
DE: 1829 Groundwater hydrology
DE: 1831 Groundwater quality
DE: 1832 Groundwater transport
SC: Hydrology [H]
MN: Fall Meeting 2005