Reducing the predictive uncertainty associated with groundwater management decision-making in the Perth regional aquifer system of Western Australia

Abstract:

The Perth Regional Aquifer Model (PRAMS) framework has been used for about a decade now to evaluate the potential anthropogenic impacts associated with management decisions that affect Perth’s groundwater resources. A great wealth of data, expertise and numerical analysis have gone into the development of PRAMS over the years. However, there has been little quantitative work conducted on systemically addressing the uncertainty in the model’s structure and predictions. PRAMS is designed to make a variety of regional and local-scale predictions and, both the nature and magnitude of the uncertainty associated with these predictions can vary significantly. A primary prediction to be addressed using the PRAMS framework, will be the effects of various deep-aquifer groundwater management scenarios on both the environmental and social concerns surrounding the superficial aquifer, which supports sensitive wetlands, and the negative impacts of seawater intrusion into the deep aquifers. A particular model-structure component that greatly affects the predictions associated with deep-aquifer groundwater extraction is the characterization of the local fault structure, i.e., whether or not faults are acting as barriers to groundwater flow. Therefore, uncertainty in fault characterization can subsequently lead to significant predictive uncertainty. However, new observation data can be obtained to reduce this uncertainty. In this study, an experimental design methodology is employed to optimally acquire new observations of state in such a way as to maximize the information obtained about the hydraulic properties of faults. Various information criteria are employed to develop optimal locations of new observation wells. The A-optimality criterion was found to be the most effective for comparing sampling strategies given the design assumptions, which include the parameter sets employed, hydraulic forcing, temporal considerations, and the use of the existing observation network. A binary-integer formulation of the particle swarm optimization technique is employed to evaluate the optimal design.