MOM6 regional modelling: Choosing stable timestep for higher res domain?

Hello! @john_reilly and I are trying to get a regional modelling run at 30th degree. The steps we’re following are:

  1. Choose a hgrid from a 10th degree mom5 run
  2. Increase hgrid to 30th degree resolution
  3. Interpolate all the inputs from a mom5 global run onto this new hgrid
  4. Interpolate Gebco topography onto hgrid & fill in holes / lakes with ‘deseas’ F90 script
  5. Use Angus’ brushcutter tools to interpolate & generate the boundary forcing
  6. Cross fingers and payu run
    All these steps are in this notebook

The above steps all work for the ‘default’ 10th degree grid but we’re running into issues with the higher res runs. Namely, there are large velocities that appear at the boundary which mess everything else up

We’re thinking it’s the timestepping and are currently trial-and-erroring different DT, DT_THERMS and OBC IN/OUT timescales. Attached is an image of the velocity field. The outputs live at /scratch/x77/ab8992/mom6/archive/tassie_30th

Currently I’ve scaled everything back by a factor of 18 which does seem like overkill for a factor of 3 increase to resolution. Larger timesteps break for the first 5 days, but I seem to increase them on a restart after 5 days, albeit still with the ridiculous velocities.
DT 900 → 50
DT_THERM 1800 → 100
OBC inflow and outflow remain at DT

update It seems like the Panan folk must have figured things out for their 20th degree run. I was about to message them directly but figured I should still post here in case others run into the same issues

1 Like

The panan 1/20th ran fine with DT = 360 from rest, DT_THERM = 900. I am no expert, but this looks perhaps like a problem with the boundary conditions not the time step?

Thanks Adele, yes I think you’re right! I think something has gone wrong in cutting out the ICs

It was the initial conditions right at the boundary. In interpolating I didn’t provide a ‘buffer’ zone on the original grid so interpolation didn’t work right at the extremes of domain. This gave a discrete step in all IC fields at the boundary which gave us the instability. A testament to MOM6’s stability that even with such a broken initial condition it ran on a short timestep!