[Geos update] Yppm-Xppm

[FlorianDeconinck]

Description

Update YPPM/XPPM to GEOS-FP requirements:

• Allow computation for ord = -6
• Add an edge fix to deal with imprint (per William Putman)
• Change the checks to explicit white list of available ord and grid

Validity is checked with a temporary script bringing data and running the translate test.

** Difference between NOAA & NASA to reconcile **

The edge fix is unique to the GEOS code. It is meant to solve an imprint issue.

        # Fix edge issues
        with horizontal(region[i_start - 1, :], region[i_start, :]):
            smt5 = bl * br < 0.0
        with horizontal(region[i_end, :], region[i_end + 1, :]):
            smt5 = bl * br < 0.0

Tag @oelbert @lharris4

[lharris4]

@FlorianDeconinck thank you for porting this. I think it would be better to introduce this as hord=6, which is what we did when @JosephMouallem added it to the Python FV3. I think absolute reproducibility of older cases is less important than improving upon this advection scheme.

[FlorianDeconinck]

ord=6 is covered by the code (and the test, though I'll double check that). The -6 is used within GEOS still so we need it for now.

Available in code with this PR is ord = 5, 6, -6, 8

[lharris4]

Hi, Florian. My recommendation was that instead of adding a new -6 option, which people may confuse as a positive-definite scheme (akin to -5), we update the existing hord=6 to use the improved edge handling; ie what you now call -6 would be implemented as a new version of hord=6.

…

On Thu, Nov 14, 2024 at 1:56 PM Florian Deconinck ***@***.***> wrote: ord=6 is covered by the code (and the test, though I'll double check that). The -6 is used within GEOS still so we need it for now. Available in code with this PR is ord = 5, 6, -6, 8 — Reply to this email directly, view it on GitHub <#29 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AMUQRVBP5OU32LYVZGK7FXL2ATW6NAVCNFSM6AAAAABRZHKPXSVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDINZXGE4DKOBUG4> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[FlorianDeconinck]

I see! Sorry I misunderstood, ans also I was a bit confused myself about it!

So the only difference I see in the code is that -6 vs 6 will trigger al to be max(0, al). Can you point to the updated code ?

[JosephMouallem]

Hi @FlorianDeconinck , we have implemented the changes in our internal branches for testing purposes. This updated code will be included in our upcoming public release, which is expected to be out shortly, thanks!

[bensonr]

You say iord must be positive (line 329), but have a value of -6 here?

[FlorianDeconinck]

Yes the docs are not corrected there. I also need to have XPPM docs point here, as they share a same conceptual limitation.

[bensonr]

It would be great if we could use something more descriptive than integers here, but I understand that might not be possible.

[FlorianDeconinck]

Allowing Enum to be parsed should be do-able. We just need to boil down the enum to an AST unique and then either we can convert it silently to an integer in code for portability

[FlorianDeconinck]

Moving this back to draft. The updated dynamics for 32-bit float is unstable for now - so any further numerical change should remain un-merged until we stabilized the dynamics