Risk-Based, Hypothesis-Driven Framework for Hydrological Field Campaigns with Case Studies

There are several stages in any hydrological modeling campaign, including: formulation and analysis of a priori information, data acquisition through field campaigns, inverse modeling, and prediction of some environmental performance metric (EPM). The EPM being predicted could be, for example, contaminant concentration or plume travel time. These predictions often have significant bearing on a decision that must be made. Examples include: how to allocate limited remediation resources between contaminated groundwater sites or where to place a waste repository site. Answering such questions depends on predictions of EPMs using forward models as well as levels of uncertainty related to these predictions.

Uncertainty in EPM predictions stems from uncertainty in model parameters, which can be reduced by measurements taken in field campaigns. The costly nature of field measurements motivates a rational basis for determining a measurement strategy that is optimal with respect to the uncertainty in the EPM prediction. The tool of hypothesis testing allows this uncertainty to be quantified by computing the significance of the test resulting from a proposed field campaign. The significance of the test gives a rational basis for determining the optimality of a proposed field campaign.

This hypothesis testing framework is demonstrated and discussed using various synthetic case studies. This study involves contaminated aquifers where a decision must be made based on prediction of when a contaminant will arrive at a specified location. The EPM, in this case contaminant travel time, is cast into the hypothesis testing framework. The null hypothesis states that the contaminant plume will arrive at the specified location before a critical amount of time passes, and the alternative hypothesis states that the plume will arrive after the critical time passes. The optimality of different field campaigns is assessed by computing the significance of the test resulting from each one. Evaluating the level of significance caused by a field campaign involves steps including likelihood-based inverse modeling and semi-analytical conditional particle tracking.