This study presents a methodology for quantifying the tradeoffs between sampling costs and local concentration estimation errors in an existing groundwater monitoring network. The method utilizes historical data at a single snapshot in time to identify potential spatial redundancies within a monitoring network. Spatially redundant points are defined to be monitoring locations that do not appreciably increase local estimation errors if they are not sampled. The study combines nonlinear spatial interpolation with the nondominated sorted genetic algorithm (NSGA) to identify the tradeoff curve (or Pareto frontier) between sampling costs and local concentration estimation errors. Guidelines are given for using theoretical relationships from the field of genetic and evolutionary computation for population sizing and niching to ensure that the NSGA is competently designed to navigate the problem's decision space. Additionally, both a selection pressure analysis and a niching-based elitist enhancement of the NSGA are presented, which were integral to the algorithm's efficiency in quantifying the Pareto frontier for costs and estimation errors. The elitist NSGA identified 34 of 36 members of the Pareto optimal set attained from enumerating the monitoring application's decision space; this represents a substantial improvement over the standard NSGA, which found at most 21 of 36 members.
A multiobjective approach to cost effective long-term groundwater monitoring using an elitist nondominated sorted genetic algorithm with historical data