UC Berkeley

Groundwater resources are an important part of any water resources management strategy, but must be managed with care to ensure the availability of water in sufficient quantity and quality to meet the ever growing demand from industrial, agricultural, and municipal uses. Hydrogeological models have a great potential to support the successful management of groundwater resources by making predictions regarding water availability and contaminant migration. However, complications arise due to the uncertainty stemming from incomplete characterization and natural variability in subsurface hydraulic properties. The uncertainty relating to these properties and model selection propagates through the modeling process, producing uncertain hydrogeologic predictions. Many formal methods exist for coping with this uncertainty in a theoretically sound manner which could improve decision making, but widespread adoption of these methods has been slow or nonexistent. This dissertation explores potential reasons for this hindered adoption of these methods in hydrogeological practice, despite success of similar methods in other industries. Issues examined include practicability of stochastic methods, the role of regulations, and the translation of uncertainty to risk at the knowledge-decision interface.

This dissertation presents a framework which addresses these challenges. The framework utilizes statistical hypothesis testing and an integrated approach to the planning of hydrogeological site characterization, modeling prediction, and water resources decision making. Benefits of this framework include aggregated uncertainty quantification and risk evaluation, simplified communication of risk between various stakeholders, and improved defensibility of decisions. The framework acknowledges that 100% certainty in decision making is impossible to obtain--rather, the focus is on providing a systematic way to make decisions in light of this inevitable uncertainty and on determining the amount of field information needed to make a decision under conditions meeting predefined criteria. In this manner, quantitative evaluation of any field campaign design is possible before data is collected. This can be done beginning from any knowledge state, and updating as more information becomes available is also possible.

The framework is presented in general and demonstrated in two synthetic case studies predicting 1) contaminant arrival time and 2) enhanced cancer risk due to groundwater contamination. Results from the arrival time study indicate that the effectiveness of field campaigns depends not only on the environmental performance metric being predicted but also its threshold value in decision making. Results also demonstrate that improved parameter estimation does not necessarily lead to better decision making, emphasizing the need for goal-oriented characterization design. The case study in predicting enhanced cancer risk involves hydrogeological characterization as well as population characterization. Population characterization can involve physiological and behavioral parameters. This case study explores the relationship between hydrogeological characterization, population characterization, cancer risk, and water resources decision making.