PyNGL dependency should be removed

[zklaus]

Right now, ece2cmor3 depends on PyNGL. As far as I could figure, this is only for a linear interpolation in tm52cmor. In light of the fact that PyNGL is now in "maintenance mode", and the clarification by NCAR in NCAR/pyngl#45 that this means that no future versions of Python will be supported, it would be good to replace the present use of NGL with an alternative implementation for linear interpolation.

[tommibergman]

Do you have an idea how urgent this is? At the moment I have no time to work on this. Also do you have any suggestions to use in place of PyNGL interpolation routine?

[etiennesky]

Looks like the cdo operator ml2pl does the trick? However it's unclear how to specify the vertical coordinate

[zklaus]

Edit: I wrote the following comment before @etiennesky's. If ml2pl does the trick, great!

I think it is not super urgent. I fixed the conda-forge build of PyNGL; if everything works as it should we should be good now until Python 3.10. But I had to fix a bug in PyNGL itself to do it and it is clear now that NCAR takes "maintenance mode" to mean extremely little maintenance. The other issue is that it's not so clear if everything works because the tests are rather limited. Plus, we really use a tiny fraction of PyNGL and incur a rather heavy dependency.

In terms of possible replacements, I am not sure. Ultimately what happens is super simple, but I have no direct replacement in mind right now.

[fserva]

Hi, I think scipy is in a dependency and it has several options https://docs.scipy.org/doc/scipy/reference/interpolate.html, otherwise for the vertical ml2pl works too

[tommibergman]

This is in the code I created. However, I don't think I will have time for this until fall. But if someone else has the time please go ahead. I have no issue using other ways of doing it, however, if possible I would try stay away from cdo. So if scipy routines can do it sounds the best.