Title: Rapid Evaluation of Hurricane-driven Storm Surge Using a Response Surface Method

Urgent and on-demand simulations of storm surge and wave impacts from hurricanes are needed to help create forecasts in the days and hours before such events. If simulations are delayed beyond the forecast window, then the information is effectively outdated and no longer relevant to decision-making. An alternative to computing new dynamic solutions is to select a close match to the forecast hurricane track from an existing data set of storm surge solutions. While fast, uncertainty in the resulting surge estimate is difficult to evaluate, and the low-order nature of the close match approach ignores important nonlinear coupling among parameters. In order to be both rapid and accurate, higher-order interpolation methods can be used that harness pre-computed solutions of storm surge and wave responses and the parameters that define the storm population. Such predictions can be available long before a fully dynamical solution is completed. Since the dataset of pre-computed solutions can be large (order 10-100 Tb) for comprehensive spatial coverage of flooded land areas, efficient statistical methods are essential to enabling this capability.

Our AdcircLite-NC project (funded by the DHS S&T Coastal Hazards Center of Excellence at UNC) implements a second-order moving least squares response surface method (RSM) to compute statistical predictions of water level and wave heights for coastal North Carolina. The RSM approach has been successfully demonstrated for wave height predictions in Hawaii. The goal is to produce predictions rapidly and well in advance of dynamically computed deterministic surge/wave simulations. A surrogate model for the surge and wave fields is computed using the RSM approach from a database of high-resolution storm surge and wave solutions. These solutions were computed using the tide, storm surge, and wave model ADCIRC for a recent FEMA-funded coastal flood insurance study (FIS) for North Carolina, and are comprised of a 675-storm population representing probable hurricanes. The rapid evaluation of individual storm tracks allows the computing of large ensembles of tracks that represent uncertainty in the hurricane track and parameters, thus enabling probabilistic surge and wave predictions at high-resolution. We demonstrate the rapid nature of the method and compare results to predictions from an existing forecast system for recent hurricane events impacting North Carolina.