Sensitivity to Prior and Reliability Measurements for Value of Geophysical Information

To ensure the sustainability of groundwater resources, actions that require spatial decisions may need to be taken although much spatial uncertainty about the aquifer flow properties exists. Geophysical data may be critical to reduce this uncertainty but may be too expensive. Therefore, the value of information (VOI) of such data needs to be assessed before proceeding with the actual survey. We present an example where the decision is whether existing contaminant sources must be relocated by identifying critical surface recharge locations. Hence assessing the aquifer vulnerability is critical. From decision analysis theory, VOI equals value with information minus the prior value. Estimating VOI requires several components. The prior geological uncertainty and a measure for information reliability are two components crucial in the VOI metric. The goal of this work is to assess the sensitivity of VOI to these two components. To address the prior geological uncertainty realistically, multiple-point geostatistical algorithms (ie snesim) stochastically model the patterns of the interpreted geological depositional system (represented by the training image). For this example study, geological concepts for glacial buried valleys are used to develop training images of the valleys. Since properties such as valley width, length and direction are not well known, many possible alternative training images can be built. To assess the most important geological components impacting aquifer vulnerability, we apply a novel distance-based clustering technique to rank the various geological factors. Secondly, to compute VOI, a measure of reliability for the proposed geophysical measurement is needed. For this example, three types of datasets collected in Denmark (in a buried valley system) are used: transient electromagnetic (TEM), DC resistivity and driller’s log. Bayesian calibration is performed to obtain likelihood and posterior functions of electrical resistivity and lithology. Monte Carlo simulation then generate many posterior probabilities of sand or clay occurring. To assess the impact of the geophysical data on aquifer vulnerability, one set of models is constrained to the geophysical surveys, another set is not. The real data set demonstrates a decent electrical resistivity contrast between the two lithologies (sand/clay). However, valley locations may not entirely consist of sand. Thus, another set of soft probabilities is generated from alternative electrical resistivity-lithology likelihood. The two calibrations demonstrate the sensitivity of the VOI measurement to different electrical resistivity contrasts. The work here ensures that the VOI is robust by evaluating the prior geological uncertainty modeling and including several reliability measures.