search

NOAA-EMC / WW3

WAVEWATCH III
Other
262 stars 530 forks source link

Bug on master, develop and scalability branches (and also V6.07.1) when forcing unstructures grids by currents #257

Open heloisemichaud opened 4 years ago

heloisemichaud commented 4 years ago

Hello, I have a bug when forcing my unstructured grid on Atlantic french coasts with currents on all the WW3 versions since the alpha.6.07.1, and all the branches master, develop and scalability). It works very well with the alpha.5.16 and older versions. I have very strange results from the first time steps (such as problems with the CFL on the shallow waters), the simulation does not crash but runs very very slower, and results are totally wrong at the coastal boundaries. We try to fix the bug with Aron, but for now the problem isn't solved. Thanks for your help Best Regards

Héloïse

aliabdolali commented 4 years ago

@heloisemichaud thanks for reporting the bug, could you provide the switch you used?

heloisemichaud commented 4 years ago

I use switch Ifremer1 and also try with DB0 and TR0, but results are bad. There is something wrong with the frequency advection, because when I removed the frequency shift, I do not observe the bad results at the boundaries.

aliabdolali commented 4 years ago

@heloisemichaud could you change REF1 to REF0 and recheck? I am suspicious about it if it is happening at the coastal boundaries.

ukmo-ansaulter commented 4 years ago

@heloisemichaud when you say the frequency advection, do you mean you are OK when you set FLCK to false? This is also something @ukmo-jianguo-li and I have seen when we use the SMC grid at high resolution - we had an instability in the model where currents are very strong around Ouessant. In our case we felt that it was a combination of the frequency shift being very strong and then the source terms reacting badly to the resulting highly skewed spectrum.

heloisemichaud commented 4 years ago

Hello, @aliabdolali, I have the same problem with REF0. @ukmo-ansaulter yes I try with FLCK to false and i don't have the bug. Did you have the same instabilities with version 5.16 ? because the problem appeared since at least version 6.07.

ukmo-ansaulter commented 4 years ago

We have always had this as an issue. @ukmo-jianguo-li can you confirm which version(s) of your model you ran the FLCK instability tests for?

We had to look at how the spectrum was evolving at different steps in the model to diagnose this. Our problem is that initially the propagation with the k-shift switched on aggregates the energy into only a few spectral bins, subsequently the source/nonlinear terms grow a lot of extra energy - we are guessing because the spectral shape is so skewed toward the top few frequency bins. We have been able to cap the growth a bit, but not satisfactorily so far. Not sure that we've ever tried reducing the timesteps massively to see if this cures the problem....

I'm guessing we might be seeing the same thing, but maybe only if something with the way the k-shift propagation works has been changed for your model between versions??