A BAYESIAN NETWORK MODEL FOR EVALUATING SEA-LEVEL RISE IMPACTS

A Bayesian network model for evaluating sea-level rise impacts

Nathaniel G. Plant¹, E. Robert Thieler², and Benjamin T. Gutierrez²

Assessing the vulnerability of the coastal zone to sea-level rise (SLR) requires integrating a variety of physical, biological, and social factors. These include landscape, habitat, and resource changes, as well as the ability of society and its institutions to adapt. The range of physical and biological responses associated with SLR is poorly understood at some of the critical temporal and spatial scales required for decision making. Here we describe a Bayesian statistical analysis framework developed from a wide range of information on coastal systems and the related uncertainties in physical and process characterizations. Basic data sets characterizing geologic and oceanographic variables are used as inputs to define the initial states of coastal systems, relevant forcing factors, prior behavior, and idealized model simulations. The Bayesian network is used to integrate these data to make probabilistic predictions of the future state of coastal environments, using parameters such as shoreline change, in response to different SLR scenarios (Figure 1). Competing hypotheses regarding the relationships between forcing, the responses, and their interrelationships are evaluated, and their uncertainties are compared. Results from the U.S. mid-Atlantic coastal region are used to explore different scenarios, as well as identify research needed to improve predictive skill. The Bayesian network approach provides an extensible framework to support decision making and to evaluate specific management scenarios for adapting to SLR.

Figure 1. Example of Bayesian network prediction of shoreline change. Each box represents a process variable that is resolved with a finite set of state ranges.  Arrows represent joint correlations that are resolved by the model.