How do I check if a Xarray Dataarray has units assigned

[axelschweiger]

I have come across the problem that during filling of "nans" the previously assigned units are getting lost. Prior to the call units are assigned using good_var= good_var*units.degC

good_var=good_var.interpolate_na(dim='time',method='slinear',max_gap='3H') #

In order to reassign units my program needs to check if units have been assigned or not. How can I do this?

When I use the metpy.quantify() is there a way to specify the units? How are units determined when calling the method?

Thanks
Axel

[dopplershift]

For data being read into xarray, the units usually come from metadata stored in a 'units' attribute, which is what var.metpy.quantify() uses to create the appropriate Quantity.

In this particular case, it looks like interpolate_na does not preserve the internal "duck array" that's used to store data. I'm not sure if that's a bug or not. (@jthielen ?)

The direct answer to your question is you can look at var.metpy.units and see if it returns 'dimensionless', but that's unsatisfying. What you can do is use dequantify, which changes from using a unit-aware array and puts the unit information back into attribute metadata, then tell interpolate_na to keep the metadata:

var = ds.Temperature_surface * units.K
var = var.metpy.dequantify().interpolate_na(dim='time', method='slinear', max_gap='3H', keep_attrs=True)
# Can "requantify" if necessary at this point
var = var.metpy.quantify()

EDIT: Fix keep_attrs to conform to docs.

[jthielen]

In this particular case, it looks like interpolate_na does not preserve the internal "duck array" that's used to store data. I'm not sure if that's a bug or not. (@jthielen ?)

Not really a bug, just a known missing feature that's a bit hidden away in the documentation (see https://docs.xarray.dev/en/stable/user-guide/duckarrays.html#missing-features).

[axelschweiger]

Thanks much Ryan
The xarray (its a DataArray) is created in the program and I didn't have units assigned. Kind of surprising how quantify strips the attribute for the DataArray but I suppose that information is transferred into the Quantity. This is good to know!

Correction in your code suggestion (replace "units" with "True"... which threw me off for a moment in case somebody else runs into this.

var=var.metpy.dequantify().interpolate_na(dim='time',method='slinear',max_gap='3H',keep_attrs=True)

I was also able to this to solve the problem

 if isinstance(var.data,units.Quantity)==False:
      var=var*units.degC 

I noticed that when I don't do this check and the variable already has a Quantity assigned. I get Kelvin^2 ...

Axel

[dopplershift]

Good catch about keep_attrs being a boolean. Apparently I can't read docs.

[dopplershift]

Also, yeah .quantify() removes the attribute so that there's only one canonical source for information about the current units. It's also easier for that to propagate through xarray operations (though obviously interpolate_na is one of a few exceptions).

[jthielen]

You may also be interested in using pint-xarray for this: https://pint-xarray.readthedocs.io/en/latest/generated/xarray.DataArray.pint.interpolate_na.html

[axelschweiger]

Aha! That I think should take care of this problem. So many ways to xarray, it should be a verb! Not that I have any idea if and how this could be done, but wouldn't it make sense to somehow get the standard xarray aware of this and use the pint version when needed? If that's laughable have a pint on me!
Best
Axel

[jthielen]

Not that I have any idea if and how this could be done, but wouldn't it make sense to somehow get the standard xarray aware of this and use the pint version when needed?

That's indeed the long-term goal! Unfortunately, for features like this (that use np.vectorize or other NumPy components not handled by existing protocols), it's a part of ongoing big-picture developments in the ecosystem like NEP 37, so not something that's possible at the moment. Hence, the need for work-arounds like the pint-xarray accessor methods.