Ensemble simulations of nearshore hydrodynamics and morphologic evolution

In this study, we used an ensemble model system and observations to dynamically calibrate model parameters. This approach differs from the heuristic approach frequently taken in model calibration, where models are set up at a particular site, and then calibrated to match a single set of observations. In contrast, this approach takes into account the uncertainty in model inputs and parameters to produce a robust probabilistic forecast of waves, currents, and sediment transport in the nearshore.

The ensemble simulations of hydro- and morphodynamics were performed using a nested Delft3D setup at the U.S. Army Corps of Engineers Field Research Facility (FRF) in Duck, NC. The nested simulation setup runs Delft3D-FLOW forced with modeled winds and tides on a 1-km resolution US East Coast grid that extends from Cape Cod, MA to Charleston, SC. This regional grid provides boundary conditions to another nested set of simulations that increase in resolution from 500-m to 10-m at the inner-most grid. Only in the inner-most grid at the highest resolution were ensembles of the fully coupled Delft3D wave-flow-morphology model run.

The sensitivity of the model parameters to the simulated hydro- and morphodynamics at the FRF site was tested first by running over 2000 hindcast ensemble simulations with varying parameter values. The parameter ranges that contributed the most variation in the flow and sediment transport results were used in the ensemble system to produce outputs of waves, currents, and sediment transport. The system also incorporates remotely sensed and in situ observations to optimize the model parameters by choosing the ensemble member that minimizes the error between the simulations and observations. The optimization can be performed any time new observations are made. Once the optimized model parameters have been determined, the system runs using ensembles of model inputs (e.g., bathymetry, wind) in order to capture the uncertainty from both model parameters and inputs. The system can be utilized in either mode depending on amount and type of observations. We present comparisons of the results to in situ observations and the skill of the system for a week-long hindcast simulation at the FRF during Hurricane Joaquin in 2015.