Create projections from CF attributes

[aulemahal]

This fixes #2099.

Rationale

As detailed in #2099, the from_cf class method inherited from pyproj does not work with cartopy as it creates a pyproj class, and not the cartopy class it was called from.

The following was the closest I could get:

import cartopy.crs
import pyproj

p = cartopy.crs.Projection(pyproj.CRS.from_cf(cf_attrs))

However, as p.bounds is None here, plotting with matplotlib fails if you pass this as the projection of the axes.

Thus, I thought it would be useful to be able to create the exact cartopy projection needed from a set of CF convention attributes.

Implications

The only "breaking" change here is that calling from_cf on a projection that does not map to a CF grid mapping will fail with a NotImplementedError instead of returning a pyproj object. But that behaviour was already broken I would say.

This PR exposes cartopy.crs.from_cf that takes in all attributes of the grid mapping variable. It uses a lookup dict to map from the CF grid_mapping_name to cartopy's Projection and calls that projection's from_cf constructor.

The PR does not try to validate CF attributes, except that it will raise a KeyError if an attribute that the conventions does not flag as optional is missing. The mappings of grid mapping names and attributes is based on the description from the conventions themselves and on this (old) note from the CF conventions wiki.

Checklist

• Example of use:

import cartopy.crs as ccrs
import xarray as xr

ds = xr.open_dataset('rotated_pole_example.nc')

proj = ccrs.from_cf(**ds.rotated_pole.attrs)

• Tests added (minimal)

• I did sign the CLA.

[CLAassistant]

All committers have signed the CLA.