Update ocean grid inputs to mosaic format

[dougiesquire]

These configurations use an old FMS format to specify the ocean grid. Can/should we update to use the new mosaic format? Can we just use the mosaic from the ACCESS-OM2 1 deg configurations?

This notebook includes some checks to help us answer these questions.

To summarise:

• The ACCESS-ESM1.5 and ACCESS-OM2 grid locations are the same, but the cell areas are different.
• The differences in the cell areas are small and at the north pole
  
• It looks like the differences in area are introduced by the FRE-NCTools tool used to update to the mosaic format, transfer_to_mosaic_grid. Converting the ESM1.5 grid to the mosaic format using transfer_to_mosaic_grid gives areas that match OM2.

So, we could update the ocean grid to use the same file as the OM2 1 deg configuration, but this will change cell areas slightly which will presumably change results. Does this mean we should leave this for a subsequent release?

[anton-seaice]

It looks like the ESM1.5 ice + ocean grids are consistent with each other, so that's good? Whats the advantage in the new format?

(EDIT: we should be able to get 1e-13 accuracy between ice & ocn grids by regenerating)

[dougiesquire]

It seems that calculating the areas near a pole is... complex. See

• this description for the poly_area function used in FRE-NCTools' transfer_to_mosaic_grid to calculate the areas
• the discussion in this issue: CubedSphere grid cell area calculated by exchange grid tool has discontinuous values around the two poles NOAA-GFDL/FRE-NCtools#44

[anton-seaice]

It looks like a fix was made in FRE-NCtools.

Does one of the versions of the grids have an obvious jump / discontinuity in areas around the true north pole?

[dougiesquire]

It looks like a fix NOAA-GFDL/FRE-NCtools#60 in FRE-NCtools.

I built FRE-NCTools from source yesterday (i.e. including that bug fix) and converted the ESM1.5 grid to a mosaic spec that had areas that match the OM2 grid. So that fix doesn't resolve our issue.

Both the ESM1.5 and OM2 grid areas have a small jump across the north pole, but it's slightly smaller in the OM2 grid...? See the notebook above for details.

[ezhilsabareesh8]

This plot shows the area of OM2 grid (orig) and the newly generated (new) using make_hgrid grid at the north pole along the longitude direction. The left plot shows the area symmetry error.

plt.plot(Orig05['area'][-1,:], label='orig' )

A big spike at the tripoles shows the abnormalities in cell area in the existing OM2 grid and new grid. I tried generating the grid using --rotate_poly, which calculates the area of polygons near the pole using a copy of the polygon that has been rotated away from the pole. It resolves the abnormal area spike however it Introduces a big area symmetry error compared to the original one without --rotate_poly (left plot). Even though the error is of the magnitude of ~2e-6% (3m2 error on ~1.5e8 m2) still it we need to think about what can be done, @ofa001 and @DaveBi any thoughts on this?

PS - Area symmetry error is calculated at the north pole as following:

plt.plot( New05['area'][-1,:] - New05['area'][-1,::-1], label='new' );
plt.plot( Orig05['area'][-1,:] - Orig05['area'][-1,::-1], label='orig' );

[dougiesquire]

Note, --rotate-poly is not an option for transfer_to_mosaic_grid but I modified FRE-NCtools to add it. As @ezhilsabareesh8 shows using make_hgrid, this gets rid of the spike at the pole. See the notebook linked in my first post for details.

Good catch re the increase in symmetry errors @ezhilsabareesh8.

[aidanheerdegen]

Great work @ACCESS-NRI/ocean!

So is there a recommendation for how to proceed?

[dougiesquire]

So is there a recommendation for how to proceed?

We are unable to reproduce the old grid exactly to correct the issue at the pole. So I think we should release ESM1.5 with the original "pre-mosaic" grid and then update all the ice and ocean grids (and related files) for a subsequent major release.