Incorporating Deeply Uncertain Factors into the Many Objective Search Process: Improving Adaptation to Environmental Change

Monday, 15 December 2014
Joseph R Kasprzyk and Abigail A Watson, University of Colorado at Boulder, Boulder, CO, United States
Deep uncertainty refers to situations in which decision makers or stakeholders do not know, or cannot fully agree upon, the full suite of risk factors within a planning problem. This phenomenon is especially important when considering scenarios of future environmental change, since there exist multiple trajectories of environmental forcings (e.g., streamflow timing and magnitude) and socioeconomic factors (e.g., population growth). This presentation first briefly reviews robust optimization and scenario approaches that have been proposed to plan for systems under deep uncertainty. One recently introduced framework is Many Objective Robust Decision Making (MORDM). MORDM combines two techniques: evolutionary algorithm search is used to generate planning alternatives, and robust decision making methods are used to sample performance over a large range of plausible factors and, subsequently, choose a robust solution. Within MORDM, Pareto approximate tradeoff sets of solutions are used to balance objectives and examine alternatives.

However, MORDM does not currently incorporate the deeply uncertain scenario information into the search process itself. In this presentation, we suggest several avenues for doing so, that are focused on modifying the suite of uncertain data that is selected within the search process. Visualizations that compare tradeoff sets across different sets of assumptions can be used to guide decision makers’ learning and, ultimately, their selection of several candidate solutions for further planning. For example, the baseline assumptions about probability distributions can be compared to optimization results under severe events to determine adaptive management strategies. A case study of water planning in the Lower Rio Grande Valley (LRGV) in Texas is used to demonstrate the approach. Our LRGV results compare baseline optimization with new solution sets that examine optimal management strategies under scenarios characterized by lower than average streamflow and higher evaporation that mimic possible scenarios of water management under climate change. By examining how planning strategies change under the new optimization runs, we show the impact of deep uncertainty assumptions on the best strategies for mitigating environmental change in the LRGV problem.