SWANFAR® -- Example from Hurricane Irene, Duck, NC

(Note: The following tests are described in greater detail in: Flampouris et al. (2013), Validation of a wave data assimilation system based on SWAN. Geophysical Research Abstracts, (15), EGU2013-5951-1, EGU General Assembly, April 2013.)

SWANFAR® is applied for the re-analysis of the wave field at the Field Research Facility (FRF) in Duck, NC.

In the figure below, the blue line corresponds to the measurement, the green line corresponds to the model forecast and the red line corresponds to the SWANFAR® (4DVar) result

The simulated significant wave height (Hs) from SWAN-F is nevertheless accurate, with the only exception the Hs during the peak of the storm (line ar case 2). For these 20h SWAN-F underestimates the Hs significantly, by assimilating the available measurements, the accuracy is increased in the order of 50%.

It is known that SWAN underestimates the wave mean period (Tm) systematically (1-2 s). The accuracy of Tm after the assimilation is increased by approximately 60%. Still, it is apparent also from the Tm graph that for wave fields with Tm lower than 5 s, the error remains significant.

SWANFAR® increased the error of wave direction. This is mainly due to the low directional resolution of the assimilation which cause problems during the run of the SWAN-Adjoint.

On the right side of the plot, you can see the corresponding scatter plots of the 3 quantities.

But still, our objective is to assimilate wave spectra. Looking at the following plot:

The left column of spectra (Case 1) corresponds to low wave/wind conditions; the right (Case 2) corresponds to the moment of the peak of Irene.

The top row shows the measured spectra, the second row shows the forecasted spectra and the third row presents the results of the assimilation.

The shape (directional and frequency spread) of the assimilation output is very similar to the measured, which is not the case for the output of the forward model.

For the high energy case (Case 2), the order of magnitude of the energy density in individual frequency/direction bins is more accurately captured by the assimilation spectrum than by the SWAN-F forecast. In this case, the measured energy (top panel) is spread over a range of frequencies from 0.04Hz up to 0.18Hz. In contrast, the forecast (middle panel) has concentrated almost all the energy in two frequency bins. The assimilation output (bottom panel) has a frequency spread that is much more similar to the measurement. Consequently, the maxima of the SWAN-FAR energy densities are also much closer to those of the measured spectrum (as can be seen by comparing color scales on the right-hand side of each panel).