92nd American Meteorological Society Annual Meeting (January 22-26, 2012)

Thursday, 26 January 2012: 3:30 PM
Numerical Optimization and Sensitivity Evaluation of the National Weather Service Southern Region Simulating Waves Nearshore (SR-SWAN) Modeling System
Room 242 (New Orleans Convention Center )
Alex Gibbs, NOAA, Miami, FL; and P. Santos, A. J. Van der Westhuysen, R. Padilla, and B. A. Mroczka
Manuscript (1.5 MB)

Coastal offices in the NWS Southern Region have been running the Simulating WAves Nearshore (SWAN) model in order to provide more precise near shore wave forecasts. This presentation will include a description of a series of tests that were performed to improve the accuracy of the model, and the resulting optimization of the model within the constraints of an operational forecast environment.

The tests included results from a non-stationary hindcast simulation of the historic 1993 Superstorm and several smaller tests involving various flow regimes across the nearshore coastal waters. Each test involved comparisons of various model configurations (based on a 0.05-1.5 Hz frequency range), considering different integration time steps, directional resolutions, and the number of iteration steps. The tests revealed the most important modification was a shorter model time step to improve the response of the model during rapidly evolving high-end marine events. Increases in directional resolution and iteration steps did not result in substantial improvements to the model forecasts.

Based on these results, and after consideration of impact on operational performance while maintaining a reasonable response to rapidly changing forcing mechanisms, the development team settled on an integration time step of 600s with a directional resolution of 15⁰ (24 bins) while maintaining one iteration per integration time step. To compensate for the overall increased computational expense with this change, SWAN's hot start functionality was programmed into the system. This additional feature allows the forecaster the option to define the initial conditions based on the previous model run, thus eliminating an initial 12 hour model spin-up that would otherwise be required.

Supplementary URL: