Operationalizing Bayesian Model Checking for Robust Decision Making: Insights from House Elevation

Many homeowners in the coastal zone elevate their houses to manage flood risk. Federal guidance for this decision relies on floodplain maps that are silent on projected future changes and neglect the deep and dynamic uncertainties surrounding the projected hazards. This uncertainty poses challenges for the design of decision-support systems. Many methods for decision making under deep uncertainty first evaluate candidate decisions over many plausible states of the world, then aggregate performance over these possible futures. Through a didactic case study of determining how high to elevate a single home in Norfolk, VA, we demonstrate that the common practice of weighting all scenarios equally in this aggregation creates a tension between (a) fully exploring the parameter space, including unlikely regions, and (b) accurately representing available scientific knowledge. We introduce an approach to bridge this divide through a computationally efficient method that weights each state of the world based on a probability distribution over possible futures. Since the distribution of deeply uncertain variables is necessarily subjective, we turn to frameworks for model critique from applied Bayesian statistics to compare and contrast different modeling assumptions. This approach can help to improve decision-making by facilitating iterative and collaborative model improvement.