Wednesday, 8 October 2008: 5:00 PM
George R. Brown Convention Center, 352DEF
The design of a monitoring network to assess groundwater quality is complicated by uncertainties in the subsurface characterization as well as in the nature of the contaminant sources. A method is presented that explicitly treats system uncertainty in identifying sampling strategies that most efficiently allocate sampling resources. Uncertainty is addressed using Monte Carlo simulation of ground-water flow in conjunction with particle tracking. The uncertain inputs to the simulation are the hydraulic conductivity field and the location and timing of contaminant sources. The approach here focuses on reducing the uncertainty associated with a fundamental quantity: the proportion of a subsurface water resource in which a contaminant is present. The design objectives considered are to (1) minimize the uncertainty in the estimate of current contaminant proportion (“status” objective), and (2) minimize the uncertainty in the estimate of the temporal change in contaminant proportion (“trend” objective). The network design problem is formulated as an integer programming problem and is solved using a genetic algorithm. The goal is to identify the location and timing of sampling, subject to a constraint on sampling costs. An application to a synthetic problem illustrates the trade-offs between sampling objectives. A sensitivity analysis demonstrates that the network designed for meeting the status (trend) objective will provide a suboptimal result when applied to the trend (status) objective. However, the approach presented here can identify an alternative optimal sampling strategy (and the additional cost associated with that strategy) that balances the status-versus-trend trade-off and meets both assessment objectives with an acceptable level of uncertainty.