Deep Uncertainty in Multi-Objective Optimization

Abstract

Transboundary river basins are increasingly subjected to pressures from climate change, economic expansion, and population growth. These challenges are compounded by Deep Uncertainties and the complexities of managing water resources that cross administrative borders. The thesis aims to improve decision support for river basin management under Deep Uncertainty. Robustness Analysis is leveraged to improve the adaptive policy design of reservoir operating policies. Open Exploration generates future states of the world. Using Feature Scoring, PRIM, and Logistic Regression Modeling, Scenario Discovery pinpoints vulnerabilities that result in Adaptation Tipping Points described by streamflow patterns. Results identified decreasing precipitation and low seasonal amplitudes as the most significant uncertainty factors influencing system performance. In combination with the evaporation rate, they accurately predict policy failure. A precipitation threshold of 0.965 and a seasonal amplitude threshold of 1.05 effectively describe streamflow patterns that describe the Adaptation Tipping Point across the Best Hydropower, Best Environment, and Best Tradeoff policy. Specifically, a resultant streamflow pattern at these thresholds accurately signals imminent policy inefficiencies that necessitate policy adaptation.