Artificial Neural Network Application for Groundwater Flow and Contaminant Transport Under Uncertainty of Hydraulic Conductivity

Abstract

The uncertainty related to the spatial variability of hydraulic conductivity (K) is an important aspect to be considered in simulations of contaminant plume migration and, consequently in the environmental management of contaminated areas. The use of stochastic K models with flow simulation and the transport of contaminants is frequent in researches to predict the behavior of the plume as well as to remediation projects under the uncertainty of this hydrogeological parameter. However, traditional flow modeling and contaminant techniques, such as MODFLOW and MT3D, based on the finite difference method, are usually slow and require considerable computational effort. Thus, this research applied artificial neural networks (ANNs) in 100 scenarios of hydraulic conductivity, performed by stochastic methods, in order to evaluate the uncertainty of K in the migration of the plume. In addition, it was possible to validate the use of ANNs as a decision tool for estimating the average concentration of contaminants in groundwater over three years of simulation. Based on the results obtained, it was possible to evaluate remediation techniques for the research area due to the migration of the plume under the uncertainty of the hydraulic conductivity.