A Subjective Bayesian Framework for Synthesizing Deep Uncertainties in Climate Risk Management. Issue 1 (3rd January 2023)

• Record Type: Journal Article
• Authors: Doss‐Gollin, James; Keller, Klaus
• Is Part Of: Earth's future. Volume 11:Issue 1(2023)
• Page Start: n/a
• Page End: n/a
• Publication Date: 2023-01-03
• Subjects: decision making under deep uncertainty -- climate adaptation -- climate risk management -- house elevation -- Bayesian statistics; Environmental sciences -- Periodicals; Environmental sciences; Periodicals; 550
• DOI: 10.1029/2022EF003044
• Languages: English
• ISSNs: 2328-4277

Abstract

Abstract: Projections of nonstationary climate risks can vary considerably from one source to another, posing considerable communication and decision‐analytical challenges. One such challenge is how to present trade‐offs under deep uncertainty in a salient and interpretable manner. Some common approaches include analyzing a small subset of projections or treating all considered projections as equally likely. These approaches can underestimate risks, hide deep uncertainties, and are mostly silent on which assumptions drive decision‐relevant outcomes. Here we introduce and demonstrate a transparent Bayesian framework for synthesizing deep uncertainties to inform climate risk management. The first step of this workflow is to generate an ensemble of simulations representing possible futures and analyze them through standard exploratory modeling techniques. Next, a small set of probability distributions representing subjective beliefs about the likelihood of possible futures is used to weight the scenarios. Finally, these weights are used to compute and characterize trade‐offs, conduct robustness checks, and reveal implicit assumptions. We demonstrate the framework through a didactic case study analyzing how high to elevate a house to manage coastal flood risks. Plain Language Summary: Identifying sound strategies to manage risks driven by climatic changes is a complex task given the large uncertainties surrounding projections of coupled natural‐human systems. These uncertainties often arise from choices experts have to make, for example, about how to formulate scientific models of future water levels. Different experts can disagree about these choices, leading to different projections. Analyzing decisions in such a situation of deep uncertainty poses nontrivial challenges. For example, picking a single representative projection can under‐estimate risk and result in poor decisions. Similarly, communicating results separately for each projection can overwhelm decision‐makers. To make matters worse, typical approaches to this problem are mostly silent on what assumptions make a difference for the decisions at hand. We develop and demonstrate a framework to address these challenges. The framework provides a transparent approach to (a) combine a large number of deeply uncertain projections to a more interpretable sample set and (b) provide insights about which assumptions and modeling choices influence decisions. We demonstrate the approach with a relatively simple example question of how high to elevate a house in the face of deeply uncertain projections of future water levels. Key Points: We introduce a Bayesian framework to transparently synthesize and characterize deep uncertainties with the goal to support decision‐making We demonstrate the framework using a simple case study of house elevation for coastal flood risk management Estimates of performance or robustness under deep uncertainty necessarily involve subjective judgments