Identification of Unknown Groundwater Pollution Sources Using Artificial Neural Networks

Abstract

The temporal and spatial characterization of unknown groundwater pollution sources remains an important problem in effective aquifer remediation and assessment of associated health risks. The characterization of contaminated source involves identifying spatially and temporally varying source locations, injection rates, and release periods. The proposed methodology exploits the universal function approximation capability of a feed forward multilayer artificial neural network (ANN) to identify the unknown pollution sources. The ANN is trained to identify source characteristics based on simulated contaminant concentration measurement data at specified observation locations in the aquifer. These concentrations are simulated for a large set of randomly generated pollution source fluxes. The back-propagation algorithm is used for training the ANN, with each corresponding set of source fluxes and resulting concentration measurement constituting a pattern for training the ANN. Performance of this methodology is evaluated for various data availability, measurement error, and source location scenarios. The developed ANNs are capable of identifying unknown groundwater pollution sources at multiple locations using erroneous measurement data.