GC53A-1185
Using unknown knowns to predict coastal response to future climate

Friday, 18 December 2015
Poster Hall (Moscone South)
Nathaniel G Plant, U.S Geological Survey, Coastal and Marine Science Center, Saint Petersburg, FL, United States
Abstract:
The coastal zone, including its bathymetry, topography, ecosystem, and communities, depends on and responds to a wide array of natural and engineered processes associated with climate variability. Climate affects the frequency of coastal storms, which are only resolved probabilistically for future conditions, as well as setting the pace for persistent processes (e.g., waves driving daily alongshore transport; beach nourishment). It is not clear whether persistent processes or extreme events contribute most to the integrated evolution of the coast. Yet, observations of coastal change record the integration of persistent and extreme processes. When these observations span a large spatial domain and/or temporal range they may reflect a wide range of forcing and boundary conditions that include different levels of sea-level rise, storminess, sediment input, engineering activities, and elevation distributions. We have been using a statistical approach to characterize the interrelationships between oceanographic, ecological, and geomorphic processes—including the role played by human activities via coastal protection, beach nourishment, and other forms of coastal management.

 

The statistical approach, Bayesian networks, incorporates existing information to establish underlying prior expectations for the distributions and inter-correlations of variables most relevant to coastal geomorphic evolution. This underlying information can then be used to make predictions. We demonstrate several examples of the utility of this approach using data as constraints and then propagating the constraints and uncertainty to make predictions of unobserved variables that include changes in shorelines, dunes, and overwash deposits. We draw on data from the Gulf and Atlantic Coasts of the United States, resolving time scales of years to a century. The examples include both short-term storm impacts and long-term evolution associated with sea-level rise. We show that the Bayesian network can solve problems of inference when the predictions require some form of assimilation or inverse modeling. And we illustrate some prediction scenarios where we are interested in simulating future conditions when constraints are more uncertain. The predictions are used to assess societal and ecological risks.