Nathan BonhamNathan Bonham successfully defended his Ph.D. dissertation “Empowering Participatory Decision Making Under Deep Uncertainty: Novel Algorithms and Interactive Tools applied to Colorado River Post-2026 Negotiations” on Monday 11/6. His research advances methods for DMDU with an emphasis on water resources management. Dr. Joseph Kasprzyk and Dr. Edith Zagona were his Faculty Advisors. Congratulations to Nathan who will graduate December 20, 2023 with a doctoral degree.


Decision Making Under Deep Uncertainty (DMDU) is an emerging field of model-based decision support methods. Deep uncertainty exists when decision-makers disagree on how to prioritize conflicting performance goals and when there is no consensus on which assumptions to make about uncertain future conditions, such as climate and human population. DMDU tests policy alternatives in many plausible futures, prioritizes policies that perform well in many futures (robust policies), and discovers conditions causing poor performance (vulnerability analysis).

There is growing demand for participation by stakeholders in decision-making. In DMDU-based decision support, participation includes the modelling process since methodological decisions can have unexpected (and undesirable) impacts on policy recommendations. Participation requires that stakeholders and decision-makers can interpret complex relationships between policies, plausible futures, and system performance. Participation also requires that the analysis can be iterated based on their feedback.

The goal of this thesis is to address four barriers to participatory DMDU: large computing requirements, choosing robustness metrics considering tradeoffs, choosing policies considering conflicting goals, and choosing methods for vulnerability analysis that are interpretable for decision-making.

This thesis contributes novel algorithms and interactive tools to address these barriers. Chapter 2 contributes a framework to test the number of futures required to accurately identify the most robust policies compared to testing more futures, potentially reducing computing requirements. Chapter 3 contributes a framework for posteriori choosing of robustness metrics, which enables stakeholders to refine their choice of robustness metrics after seeing robustness tradeoffs. Chapter 4 uses the Self-Organizing Map, a machine learning method, to create a negotiation framework that helps decision-makers identify compromise policies. Chapter 5 contributes a review of vulnerability analysis methods and identifies best practices for choosing interpretable methods.

The research contributions are demonstrated using a case study of reservoir operation policy in the Colorado River Basin (CRB). Since 2000, extended periods of low inflow have depleted storage in Lakes Mead and Powell, threatening hydropower infrastructure and water deliveries. Current policies expire in 2026, thereafter a new policy take effect. This research evaluates a set of Lake Mead policies in 500 futures of streamflow and demand conditions. The robustness of these policies are discussed in Chapters 2-4, and Chapter 5 uses the CRB to illustrate purposes and methods for vulnerability analysis.