Use pint for units

[TomNicholas]

Uses pint-xarray to first add units to the dataset during pre-processing, then convert depth to metres. Might help close #31

Doesn't seem to break any tests (I had one fail locally but that also failed before I made any changes).

Not really sure what other changes might be required to use pint in this package properly but this is a start.

[TomNicholas]

Installs and uses cf_xarray.units so possibly related to #59

[jbusecke]

It seems like this actually works quite well. The only thing we need now is to ignore models that provide no unit attrs (there are a few 😭). I feel that should be an upstream option though?
If we want to be more fancy, we could 'guess the units', but that seems risky.

[jbusecke]

Should we do this inside correct_units?

[TomNicholas]

That would fix your bug, but whether or not you want to conceptually combine "adding codified physical units to the data" and "potentially change what units the data is expressed in" is up to you.

[jbusecke]

So now that I have put in another if statement to not run on the datasets that are missing units it complains about this:

incompatible units for variable 'lev': cannot use 'centimeters' as a unit

Full error here.

I first thought that centimeters might be a weird spelling and it might not be in the pint registry, but then I checked:

from pint import UnitRegistry
ureg = UnitRegistry()
'centimeters' in ureg

and got True...

not sure what is going on here.

[keewis]

it is a bit difficult to see, but the unit attribute is currently a str. To get the conversion of dimension coordinates to work properly you have to call .pint.quantify first.

[TomNicholas]

But we are calling ds = ds.pint.quantify() first?

[keewis]

that does not seem to be the case for the traceback posted in #160 (comment)?

[TomNicholas]

Right I see what you mean - in the failing test it only imports correct_units(), without first calling .quantify.

[jbusecke]

Thanks a lot for the pointers. I think I get it now (still very new to the whole pint thing).

Ill fiddle some more...

[keewis]

just checking: if you want to get the string "meter" after calling dequantify, use "P" as the format. If you want "m" instead, use "~P".

[jbusecke]

Ah thanks a lot for checking in @keewis, I misunderstood that one. Lets see if the tests pass this time. Ill try to get this merged today.

[jbusecke]

Thanks a lot @keewis and @TomNicholas for getting this started and all the detailed pointers.