Simpler horizontal remapping methods for diagnostics

[akshaysridhar]

Enables simpler bilinear interpolation option (see ClimaInterpolations.jl) within the Remapping module.

Example:

[akshaysridhar]

Nq = 2 ; e.g.

[ph-kev]

I think the tests in ClimaDiagnostics are failing because a default wasn't provided for horizontal_method in the Remapper object?

I think someone who know more about the math of interpolation should look at this too.

[akshaysridhar]

I think the tests in ClimaDiagnostics are failing because a default wasn't provided for horizontal_method in the Remapper object?

I think someone who know more about the math of interpolation should look at this too.

Thanks @ph-kev : API has been updated to fix downstream failures - and struct properties updated in BilinearRemapping to maintain the current API (+ minor docs fixes following your suggestions)

[akshaysridhar]

Adding some context from ClimaAtmos runs via @oalcabes :
The default remapping method result in the following outcome for some representative diagnostic variable (cloud top height)

julia> cltz.data |> extrema
(-2279.2927f0, 13606.046f0)`

wherein the -2.2km minimum is clearly unphysical, compared with the simpler bilinear remapping (both calculations on identical target meshes) -

julia> cltz.data |> extrema
(0.0f0, 12421.474f0)

[juliasloan25]

Thanks Akshay! The results look really nice. Just to clarify - the default remapping remains unchanged (spectral element), and now we have an option for bilinear remapping which can be used for e.g. diagnsotics?

[akshaysridhar]

Thanks Akshay! The results look really nice. Just to clarify - the default remapping remains unchanged (spectral element), and now we have an option for bilinear remapping which can be used for e.g. diagnsotics?

The intent here was to retain the existing SEM interpolation and add a simpler option that people can invoke elsewhere if necessary. Bilinear interpolation will "smooth" the result (but will not generate spurious negative values in the output due to ringing from higher polynomial orders). For diagnostic outputs only, I'd like to make the bilinear interpolation the default choice. I expect that, where conservation is critical we would be using the ConservativeRegridding tools rather than this Remapping module.

[imreddyTeja]

This looks mostly good to me. I'm a bit confused by the interpolate_array changes though

[imreddyTeja]

Is this needed? The constructor can also be dispatched to with the types

[juliasloan25]

Btw the downstream ClimaLand tests are failing because of a separate issue that will be fixed today

[akshaysridhar]

Btw the downstream ClimaLand tests are failing because of a separate issue that will be fixed today

Thanks - the build history issue also appears to be a broader permissions problem