[TST] Upcoming dependency test failures

[github-actions]

The build with nightly wheels from matplotlib, pyproj, scipy, shapely and
their dependencies has failed. Check the logs for any updates that need to
be made in cartopy.
https://github.com/SciTools/cartopy/actions/runs/22881338352

[rcomer]

These are image test failures. I don't see a difference flicking between them and the diffs look like

lib/cartopy/tests/mpl/test_examples.py::test_global_map

Reference image

lib/cartopy/tests/mpl/test_images.py::test_imshow_projected

Reference image

Maybe a result of matplotlib/matplotlib#31021?

[dopplershift]

We are getting a bunch of failures on conda now due to PROJ 9.8.0. I assume anyway, since we're seeing a bunch of image failures on MetPy with it now too. Unfortunately, we don't upload the results of the conda tests. PR incoming to address the lack of images...

[dopplershift]

See #2636 for PR adding conda image artifact.

[dopplershift]

So I'm going to guess that most of the problems with PROJ 9.8 are caused by them adding ellipsoid support for Equidistant Cylindrical, i.e. our Plate Carree. Easy to get back the previous behavior by making sure we set up a spherical datum.

I've not yet convinced myself whether using ellipsoid support for this simplified coordinate system is good or bad.

[greglucas]

It does look like most coastline images are now shifted with PROJ 9.8. Link to PROJ PR implementing this for xref: OSGeo/PROJ#4656

I think that the ellipse should be taken into account. I think the NaturalEarth coastline data is defined based on the WGS84 ellipse, so anything that uses that as the default ellipse should probably use the ellipsoidal formulas when transforming to other coordinate systems. I'm not positive on that statement at all, that is just a quick thought.

I am somewhat curious what will happen to this example and if we were poorly referencing the ellipse ourselves!
https://cartopy.readthedocs.io/stable/gallery/lines_and_polygons/effects_of_the_ellipse.html#sphx-glr-gallery-lines-and-polygons-effects-of-the-ellipse-py

[rcomer]

New proj is also breaking the tick_labels example in the docs. Testing locally in a new conda environment, the script will complete if I comment out this line

cartopy/examples/gridlines_and_labels/tick_labels.py

Line 24 in 7dcf56c

ax1.set_yticks([-90, -60, -30, 0, 30, 60, 90], crs=ccrs.PlateCarree())

[aulemahal]

Hi,

I wanted to pitch in as a user of cartopy. I stumbled on the issue of PROJ 9.8 changing the definition of proj=eqc and I opened an issue there : OSGeo/PROJ#4714. It seems that it is not really a PROJ issue so I closed it.

However, this means that code written with cartopy's PlateCarre will change behaviour with this change. My question is : do you plan to force PlateCarre to use a spherical definition and avoid most breaking changes, or should we adapt our code ?

EDIT : To get back the same "PlateCarree" projection with PROJ 9.8 that we had in PROJ 9.7.1, I think one needs to do:

WGS84_SEMIMAJOR_AXIS = 6378137
PC = PlateCarree(globe=Globe(ellipse='sphere', semimajor_axis=WGS84_SEMIMAJOR_AXIS, semiminor_axis=WGS84_SEMIMAJOR_AXIS))