Model Improvement by Assimilating Observations of Storm-Induced Coastal Change

Discrete, large scale, meteorological events such as hurricanes can cause wide-spread destruction of coastal islands, habitats, and infrastructure. The effects can vary significantly along the coast depending on the configuration of the coastline, variable dune elevations, changes in geomorphology (sandy beach vs. marshland), and alongshore variations in storm hydrodynamic forcing. There are two primary methods of determining the changing state of a coastal system. Process-based numerical models provide highly resolved (in space and time) representations of the dominant dynamics in a physical system but must employ certain parameterizations due to computational limitations. The predictive capability may also suffer from the lack of reliable initial or boundary conditions. On the other hand, observations of coastal topography before and after the storm allow the direct quantification of cumulative storm impacts. Unfortunately these measurements suffer from instrument noise and a lack of necessary temporal resolution. This research focuses on the combination of these two pieces of information to make more reliable forecasts of storm-induced coastal change. Of primary importance is the development of a data assimilation strategy that is efficient, applicable for use with highly nonlinear models, and able to quantify the remaining forecast uncertainty based on the reliability of each individual piece of information used in the assimilation process. We concentrate on an event time-scale and estimate/update unobserved model information (boundary conditions, free parameters, etc.) by assimilating direct observations of coastal change with those simulated by the model. The data assimilation can help estimate spatially varying quantities (e.g. friction coefficients) that are often modeled as homogeneous and identify processes inadequately characterized in the model.