Optimal pumping policy for a public supply wellfield using computational neural network with decision-making methodology

Effective management of groundwater resources is at the forefront of environmental challenges confronting mankind in the 21st century. In many regions of the world, growing human populations coupled with decades of improper use and disposal of chemical contaminants have diminished the quantity and quality of this irreplaceable resource. It is projected that by the year 2025, thirty-five percent of the world population will experience chronic water shortages [Ahfeld et al, 2000]. As much of the world relies upon groundwater as its drinking water source (e.g. 51.7% of the United States population), optimal management will become increasingly important. Complicating the problem is that human use considerations must be balanced with environmental and economic concerns. Balancing multiple concerns such as these constitutes a multiobjective and conflict resolution problem, where tradeoffs among non-commensurable objectives must be identified to select the best compromise solution. In this research, a Computational Neural Network (CNN) methodology has been developed for identifying pumping policies for public supply wells that effectively balance risk of contamination with supply objectives. Utilizing simulation results from MODFLOW, monthly CNN's were developed to predict groundwater elevations at select locations for a hypothetical but realistic unconfined, heterogeneous aquifer under variable monthly pumping and recharge rates. The resulting CNN architecture, a simplified linear approximation to the finite-difference flow equations, was embedded into a linear optimization program, and an objective function that quantified both risk and supply was solved for using different weight preferences. The resulting Pareto frontier served as the basis for multiobjective and conflict resolution analyses. The CNN and decision-making methodologies were then applied to a real-world test case in Toms River, New Jersey, where contaminated public supply wells, a suspected cancer cluster, and few alternative water sources motivated the need for a formal and rigorous analysis that identified the best compromise solution. The new CNN methodology achieved a very high degree of accuracy in both simulation and optimization, and, once trained, is computationally more efficient than traditional methods. Perhaps most importantly, this research demonstrates the theoretical possibility of training a CNN with real-world data, allowing direct optimization of the actual groundwater system