SSH tides cause blow up in corner of big domain

[konsole]

Dear ROMS users,

If I force my whole Antartic application with tidal surface elevation at the northern boundaries, I get this strange response in one of the corners (figure shows surface elevation anomaly):

The figure relates to the following simplified case:

​*Initialization with no stratification and zero velocities
*No surface fluxes
*Open boundaries closed for 3D momentum and tracers
*Chapman (flather) for free surface (2d momentum) open boundary condition
*2d momentum at open boundaries are set to zero
*Forcing just with tidal surface elevation at northern boundaries

Please find the log file attached.

If I close the open boundaries for 2d momentum as well, the model works fine, but it seems to deactivated the tides too.

Do you have any idea what could cause the trouble?

Many thanks,

ole

[Kosa]

Hi I didn't see your log file attached but if it's blowing up at the beginning of the run you could try adding the option

Code: Select all

#define RAMP_TIDES

and recompiling.

[konsole]

Thanks Kosa,

ramping the tides seems just to delay the issue by a couple of timesteps.

Anyway, we tried a test with flat bathymetry and it did not blow up. So I looked at the corner bathy in more detail and found this mountain ridges, which form a little triangle exact in the corner.

modelbathy_corner.png (18.51 KiB) Viewed 3676 times

When I smooth the edge, the model does not blow up there anymore.
Maybe the tides get reflected there and energy can't radiat fast enough out of the domain? Maybe interference between the boundary signals happens in this little triangle?

Anyway, after smoothing this corner, the model blow up a bit later at an other point close to the boundary with similar high bathy gradients.
I smoothed along all the boundaries (all 10 outermost cells, so 100km, with a plusMinus approach to obey rx0 < 0.05) and the model seems to hold with 600s baroclinic timestep (please see h_new - h_old below). However, the highest velocities occur now at the inside boundary of the smoothing area. I processed a second smoothing with rx0 < 0.1 along the next 5 cells and issue seems to vanish.

Now the model blows up inside the domain, but thats an other topic ... I hope.