Differences between SWAN v41.31 and v41.10

Updated 2020/06/23: Adjusted for new SWAN setting with NSWEEP=1.

In late May 2019, the SWAN developers released a new version. Whenever this happens, the new version needs to be implemented into the coupled SWAN+ADCIRC, thus replacing an older version in the coupled model.

Starting with the upcoming release version 55 of ADCIRC, the coupled SWAN has been upgraded to its latest release version 41.31. It replaces the older version 41.10.

This upgrade is mostly a benefit to users of SWAN+ADCIRC. It has been almost 4 years since the last upgrade, and we had skipped a new SWAN version (41.20) during that time. Thus, this upgrade is adding features and bug fixes from two newer versions (41.20 and 41.31). SWAN has added several capabilities that will be advantageous to users of SWAN+ADCIRC.

However, a few of its changes will cause differences in the wave predictions, as described below. Users will likely need to re-calibrate their input settings for SWAN.

Modifications

The full lists of additions, changes, and bug fixes for SWAN versions 41.20 and 41.31 can be found in its online user manual. Here are the most important:

  • The so-called ST6 physics were added. This is a major change to the wind input and white-capping to be “observation-consistent,” and this physics package has become popular in wave models including SWAN and WAVEWATCH III. Please see Rogers et al. (2012).
  • For ambient currents on unstructured grids (like the water velocities passed from ADCIRC), the derivatives are now computed with a Green-Gauss formula.

Both of these changes will be important to users of SWAN+ADCIRC. However, these and other changes will cause differences in the SWAN wave predictions, relative to older versions.

Differences in the Un-Coupled SWAN

It is important to note that there are no differences in the un-coupled SWAN, i.e. without any information (water levels, water velocities, modified friction) passed from ADCIRC. The differences occur in how the ambient currents are handled, which only occur with the coupling.

To prove this, we can consider an un-coupled test case. This un-coupling can be set by removing the corresponding READINP commands from the SWAN control file (fort.26). Only the wind velocities should be passed from ADCIRC; otherwise, it is like SWAN is running by itself.

Figure 1: Significant wave heights in the Albemarle and Pamlico Sounds during Hurricane Irene: (left) predicted by SWAN v41.31, and (right) differences from v41.10.

In Figure 1, we show predictions of significant wave heights on a small model of the Albemarle and Pamlico Sounds during Hurricane Irene (2011). This model has been shared previously as an example of the coupled SWAN+ADCIRC, but now we are disabling the coupling to SWAN. The differences are zero everywhere.

Thus, without any coupling to water levels and currents, the SWAN predictions will be the same for the new v41.31.

Differences in the Coupled SWAN+ADCIRC

However, the new SWAN v41.31 will cause differences in the wave predictions in the coupled SWAN+ADCIRC. These differences are due to a change in how SWAN computes derivatives of the ambient currents on unstructured meshes. Previously, these derivatives were computed as backward differences.

In v41.31, these derivatives are computed by using the Green-Gauss formula. This change will improve the accuracy of the solution, but it will cause differences in the wave predictions relative to previous versions. In this section, we show these differences for three examples.

Figure 2: Significant wave heights in the Albemarle and Pamlico Sounds during Hurricane Irene: (left) predicted by SWAN v41.31, and (right) differences between v41.31 and v41.10.

In Figure 2, we show predictions of significant wave heights on the same example of the coupled SWAN+ADCIRC, but now with the full coupling enabled. Winds, water levels, currents, and friction roughness lengths are passed from ADCIRC, and wave radiation stress gradients are passed from SWAN.

The significant wave heights are different between the two versions. Ignoring the differences at the wet/dry front (shown as brief red flashes in the animation), we see the significant wave heights are generally smaller in the new v41.31. The maximum differences are about 50 cm in the river estuaries, and the wave heights are smaller by 10 to 20 cm throughout the sounds as the storm passes.

Figure 3: Significant wave heights in the Gulf of Mexico during Hurricane Gustav: (left) predicted by SWAN v41.31, and (right) differences between v41.31 and v41.10.

In Figure 3, we show predictions of significant wave heights on a larger (but still coarse) model of the Gulf of Mexico during Hurricane Gustav (2008). This model has also been shared previously as an example of the coupled SWAN+ADCIRC, and we are running it here with the full coupling.

The significant wave heights are different between the two versions, with the largest differences of about 50 cm on the leeward side of Cuba as the storm enters the Gulf. These differences propagate away from the storm and become large again as they enter shallower water on the continental shelf.

For these first two examples, it is emphasized that: (a) these meshes are extremely coarse — e.g. the mesh spacings are about 42 km in the central Gulf; and (b) these wave predictions are not accurate — e.g. the significant wave heights are overpredicted along the Louisiana coast by almost 2 m (as shown in the page linked above). These videos are shown only to illustrate the differences in the new version.

Figure 4: Significant wave heights near North Carolina during Hurricane Irene: (left) predicted by SWAN v41.31, and (right) differences between v41.31 and v41.10.

In Figure 4, we again show predictions of significant wave heights during Hurricane Irene (2011), but now on a larger, higher-resolution mesh. The NC9 mesh was developed for floodplain mapping studies for FEMA and is now used for real-time forecasting for North Carolina. Its coverage extends into the Atlantic Ocean, so it can show the evolution of waves in deep water.

The significant wave heights are again different, especially in regions where they interact with currents on the continental shelf. In the new v41.31, the wave heights are smaller by more than 1 m near the shelf break, but then larger by close to 1 m in regions closer to shore. These differences in the waves will then translate into differences in the currents and water levels, via the feedbacks in the coupled SWAN+ADCIRC.

Thus, the new v41.31 can have large differences in its wave predictions, due to its updated method of computing derivatives for ambient currents on unstructured meshes. Users will need to be careful to do a complete re-calibration, so that they are not surprised by changes in their model results.

Summary

For the next release version 55 of the coupled SWAN+ADCIRC, the SWAN component has been upgraded to its newest release version 41.31. This new SWAN version has several new features (including the so-called ST6 physics package), but its predictions will also differ from previous versions of the coupled models. Predictions of significant wave heights can differ by as much as 50 cm.

When switching to this new version of SWAN+ADCIRC, users should compare and become comfortable with the differences relative to older versions.