Add MOM5 kernel

[rabernat]

cc @LaureZanna, as discussed on Slack

[rabernat]

The package is ready to have laplacians added to it. They should be added to this file:

https://github.com/ocean-eddy-cpt/gcm-filters/blob/master/gcm_filters/kernels.py

There are three things that need to be added in that file:

1. A new class (e.g. MOM5Laplacian), which inherits from BaseLaplacian. Any auxiliary grid variables (e.g. tarea, dx, dy, etc.) should be attributes of the class. The actual laplacian should be coded in the __call__ method and have only a single argument.
2. Define a new GridType here:
   

   gcm-filters/gcm_filters/kernels.py

   Line 14 in 91cd579

   GridType = enum.Enum("GridType", ["CARTESIAN", "CARTESIAN_WITH_LAND"])

   
   Maybe something like MOM_GRID
3. Then add the laplacian to the ALL_KERNELS dictionary, like this
   

   gcm-filters/gcm_filters/kernels.py

   Line 89 in 91cd579

   ALL_KERNELS[GridType.CARTESIAN_WITH_LAND] = CartesianLaplacianWithLandMask

Then the kernel will automatically get tested in the test_kernels.py module.
However, it probably won't work unless you define special options for the grid_vars, as here

gcm-filters/tests/test_kernels.py

Lines 15 to 18 in 91cd579

	if grid_type == GridType.CARTESIAN_WITH_LAND:
	mask_data = np.ones_like(data)
	mask_data[: (ny // 2), : (nx // 2)] = 0
	extra_kwargs["wet_mask"] = mask_data

For the tests, we will need to decide about #5. I have some ideas about that I'll share in that issue.

cc @LaureZanna, @ElizabethYankovsky

[rabernat]

@jbusecke maybe you could give a quick overview of the development process

• running tests locally
• pre-commit
• docstrings

[ElizabethYankovsky]

@rabernat - Thanks for cc-ing! Laure, her postdoc Arthur, and I have been discussing the formulation of the Laplacian in MOM5 and will contribute to this

[jbusecke]

Sure thing. Where should that go? As a docs page?

[rabernat]

Sure thing. Where should that go? As a docs page?

I was thinking just a comment on github, but a contributor guide on the docs is probably more robust. Could probably just be copy-pasted from one of our other projects.

[jbusecke]

How detailed should this be with regard to git? Full step by step with add/commit/...?

[jbusecke]

Take a look at my PR #16

[LaureZanna]

Thanks a lot @rabernat and @jbusecke !

[iangrooms]

I think our scalar Laplacians work for MOM5, and #106 is developing a vector Laplacian that should work for MOM5. Can we close this issue?