use CF conventions to enhance plot labels

[rabernat]

Elsewhere in xarray we use CF conventions to help with automatic decoding of datasets. Here I propose we consider using CF metadata conventions to improve the automatic labelling of plots. If datasets declare long_name, standard_name, and units attributes, we could use these instead of the variable name to label the relevant axes / colorbars. This feature would have helped me avoid several past mistakes due to my failure to examine the units attribute (e.g. data given in cm when I assumed m).

Code Sample, a copy-pastable example if possible

Here I create some data with relevant attributes

import xarray as xr
import numpy as np
ds = xr.Dataset({'foo': ('x', np.random.rand(10),
                         {'long_name': 'height',
                          'units': 'm'})},
                coords={'x': ('x', np.arange(10),
                              {'long_name': 'distance',
                               'units': 'km'})})
ds.foo.plot()

Problem description

We have neglected the variable attributes, which would provide better axis labels.

Expected Output

Consider this instead:

def label_from_attrs(da):
    attrs = da.attrs
    if 'long_name' in attrs:
        name = attrs['long_name']
    elif 'standard_name' in attrs:
        name = attrs['standard_name']
    else:
        name = da.name
    if 'units' in da.attrs:
        units = ' [{}]'.format(da.attrs['units'])
    label = name + units
    return label
ds.foo.plot()
plt.xlabel(label_from_attrs(ds.x))
plt.ylabel(label_from_attrs(ds.foo))

I feel like this would be a sensible default. But it would be a breaking change. We could make it optional with a keyword like labels_from_attrs=True.

Output of xr.show_versions()

INSTALLED VERSIONS ------------------ commit: None python: 3.6.5.final.0 python-bits: 64 OS: Linux OS-release: 4.4.111+ machine: x86_64 processor: x86_64 byteorder: little LC_ALL: en_US.UTF-8 LANG: en_US.UTF-8 LOCALE: en_US.UTF-8

xarray: 0.10.3+dev13.g98373f0
pandas: 0.22.0
numpy: 1.14.3
scipy: 1.0.1
netCDF4: 1.3.1
h5netcdf: 0.5.1
h5py: 2.7.1
Nio: None
zarr: 2.2.1.dev2
bottleneck: 1.2.1
cyordereddict: None
dask: 0.17.4
distributed: 1.21.8
matplotlib: 2.2.2
cartopy: None
seaborn: None
setuptools: 39.1.0
pip: 9.0.1
conda: 4.3.29
pytest: 3.5.1
IPython: 6.3.1
sphinx: None

[shoyer]

Default labels are intended more for convenience than anything else. For publication/production purposes, I suspect labels are almost always been explicitly set, anyways.

So I think this convention makes sense. Let's just document it directly in the plotting docs, maybe even by including your label_from_attrs() function to make it fully explicit. That way non-weather/climate users can also make use of it, without referring to the full CF conventions spec.

[rabernat]

It would take me some time to wrap my head around the plotting code again, and as a result I don't anticipate having time to work on this any time soon.

I know @dcherian has been digging into the plotting functions a lot. I imagine that he would be able to implement this much faster than I would. Just leaving that suggestion out there... ;)

Obviously this is not an urgent issue.

[fmaussion]

This comes out regularly. See also:

• Include units (if set) in plot labels #1787
• Option to include units in .plot() #1630

[dcherian]

I can do it. are we agreed that long_name, standard_name, and units are the attributes we want to follow? I think we should also add support for upper-case versions of these (I think I've seen those in the wild)