Orthographic threshold reduction

[ludwigVonKoopa]

Related issue : #1907

This PR changes Orthographic projection threshold to solve tiles loading error.
With actual threshold value, when using tiles with enough precision, the projection boundaries are eroded too much, and result in some tiles data not loaded :

import matplotlib.pyplot as plt
import cartopy.crs as ccrs
from cartopy.io.img_tiles import Stamen, GoogleWTS, GoogleTiles
import matplotlib.patches as mpatches
import shapely.geometry as sgeom

tiler = Stamen(style="terrain-background", cache="/tmp/tiles/")

fig = plt.figure(figsize=(12,7), dpi=120)

proj1 = ccrs.Orthographic(0, 45)
proj2 = ccrs.Orthographic(0, 45)
proj2.threshold /= 20
titles = [f"threshold = {p.threshold:,.2f}" for p in (proj1, proj2)]
for p, proj, title in zip(range(2), [proj1, proj2], titles):

    ax_pla = fig.add_subplot(2, 2, 1+p, projection=ccrs.PlateCarree(), frameon=False)
    ax_ort = fig.add_axes([0.5*p, 0.1, 0.5, 0.5], projection=proj)

    bounds_proj = sgeom.Polygon(proj.boundary)
    geom = ax_ort._get_extent_geom(tiler.crs)
    img, extent, origin = tiler.image_for_domain(geom, 4)

    for ax in [ax_pla, ax_ort]:
        ax.imshow(img, extent=extent, origin=origin, transform=tiler.crs, regrid_shape=500)
        c_geom = ax.add_geometries([geom], crs=tiler.crs, color="red", alpha=0.3, zorder=30)
        c_bounds = ax.add_geometries([bounds_proj], crs=proj, color="green", alpha=0.3, zorder=40)
    ax_pla.set_title(title)
    
geo2 = mpatches.Patch(color='green', label="ortho boundaries")
geo1 = mpatches.Patch(color='red', label='eroded boundaries')
ax_pla.legend(handles=[geo1, geo2], loc=(-0.3,-0.5))

I did not add tests with this PR, because i don't see how i could easily tests this.

I could add a test which

• create a black image
• paint tiles which pass through correct extent
• project this tiles on orthographic
• and check in the final image if there is black.

But this may be too much. What do you think ?

Thanks

[greglucas]

Looks like you'll have to update some images that are causing the test failures. I think all of the new images look like improvements to me, which is good :)

For the multiple projections one, we may want to just increase the tolerance a bit rather than updating that image, since it is only a small difference for one image test. (We should really break this out into separate images later on)

[QuLogic]

We should really break this out into separate images later on

Oh, wait, have I not PR'd this already? I should. #1964

[greglucas]

@ludwigVonKoopa, the images were just updated, so you're going to have to rebase this and likely need to update the Orthographic test image. Can you also remove the 0.9999*a in the projection definition while you're at it?

[greglucas]

@ludwigVonKoopa, I still think this looks like a good update if you're interested in getting it in. I'm guessing you'll need to update the Orthographic image tests already present in the test suite. One additional request would be to put a comment above, saying something about why this should be the threshold so someone else doesn't come and change it later unknowingly.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}