Using neural networks for parameter estimation in ground water

• Published in: Journal of hydrology (Amsterdam) Vol. 318; no. 1; pp. 215 - 231
• Main Authors: Garcia, Luis A., Shigidi, Abdalla
• Format: Journal Article
• Language: English
• Published: Amsterdam Elsevier B.V 01-03-2006; Elsevier Science
• Subjects: Aquifer parameters; aquifers; Artificial neural network; Earth sciences; Earth, ocean, space; equations; Exact sciences and technology; Ground water; groundwater; groundwater flow; Hydrogeology; hydrologic models; Hydrology. Hydrogeology; Inverse problem; inverse process; mathematical models; neural networks; Numerical models; transmissivitiy; Aquifer parameters; Numerical models; Artificial neural network; Ground water; Inverse problem; stress; inverse problem; hydraulic conductivity; hydraulics; aquifers; numerical models; neural networks; ground water; hydrodynamics; transmissivity
• ISSN: 0022-1694; 1879-2707
• DOI: 10.1016/j.jhydrol.2005.05.028

Artificial neural networks act as universal function approximators. This makes them useful in modeling problems in which the relation between dependent and independent variables is poorly understood. In this paper, the ability of an artificial neural network (ANN) to provide a data-driven approximation of the explicit relation between transmissivity and hydraulic head as described by the ground water flow equation is demonstrated. This approximation can be easily solved for the inverse problem and is capable of simulating aquifer response to additional stresses. The methodology developed as part of the research presented in this paper is comprised of two major tasks. The first task is to successfully train the ANN to approximate the relation between any possible transmissivity field of the aquifer being modeled and the hydraulic head values as described by a ground water flow model. This, in effect, will produce a data-driven model capable of mapping the relation between any realization of tranmissivity of a specific ground water aquifer and the resulting hydraulic head values as computed by the ground water flow equations. The second task is to invert this model to solve the inverse problem so as to produce a transmissivity field that will honor the known transmissivity values and reproduce the known hydraulic head values when used in a ground water flow model. This paper explains the ANN training and the inversion process and demonstrates that the process works using a hypothetical two-dimensional aquifer problem where the input and outputs are assumed known and therefore the performance of the inversion process can be quantified.