python-stratify still needed?

[bouweandela]

It looks like iris supports vertical regridding: https://scitools.org.uk/iris/docs/latest/userguide/interpolation_and_regridding.html#interpolating-non-horizontal-coordinates

@bjlittle Do we still need python-stratify for vertical interpolation? Additionally, do we still need to do cube slicing if the levels are already available, is this shortcut not taken by the iris vertical interpolator?

There are two problems with python-stratify:

• it does not seem to support lazy data
• it cannot be installed from PyPI -> this is apparently solved, but see also https://github.com/SciTools-incubator/python-stratify/issues/25

Here is a link to the relevant code: https://github.com/ESMValGroup/ESMValTool/blob/3e68d1593f544694618617368f8a3c88d0683f33/esmvaltool/preprocessor/_regrid.py#L327-L434

[bjlittle]

@bouweandela I highly suspect that the generic core iris.analysis.Linear interpolator scheme will not be performant enough for you. Not even close.

The attraction of python-stratify is that it is cythonized, therefore closer to bare metal speed. However, you are right, it isn't lazy. I'm keen to do a couple of things here...

• [x] Move the python-stratify package from our experimental SciTools-incubator organisation into our core SciTools organisation, as it's stable an used in production code
• [ ] Move the _vertical_interpolate from esmvaltool into the iris core
• [x] Resolve issue https://github.com/SciTools-incubator/python-stratify/issues/25
• [ ] Investigate whether we can make _vertical_interpolate lazy
• [ ] By it's nature, the function of the _vertical_interpolate is embarrassingly parallel, as each vertical column is completely independent, so investigate whether we can parallelize using dask

[valeriupredoi]

will this be included (or excluded) in iris=2.3 @bjlittle ? I am starting to think we should make a list of ESMValTool/Core iris-related issues with fixes in 2.3 :beer:

[valeriupredoi]

noticed in #776 that stratify accepts and works happily with cube.core_data() - any way to make it fully lazy and output a lazy data array @bjlittle ? :beer: