Contextily basemap tiles not displaying correctly when using a projected CRS

[chudlerk]

I am trying to use contextily tiles to display a basemap on an axis.

If I set up like this:

import matplotlib.pyplot as plt
import ultraplot as uplt
import contextily as cx
import cartopy.crs as ccrs

#set up max extent
lon_min, lon_max = -123.0, -121.5
lat_min, lat_max = 36.8, 38.5
c_lon = (lon_min + lon_max)/2
c_lat = (lat_min + lat_max)/2

#set up unprojected CRS
crs = ccrs.PlateCarree()

I can make a map with a basemap using Cartopy like this

#make unprojected map with Cartopy
fig, ax = plt.subplots(subplot_kw={"projection": crs})
ax.set_extent([lon_min, lon_max, lat_min, lat_max], crs=ccrs.PlateCarree())
cx.add_basemap(ax=ax, crs=crs)
plt.show()

And the Ultraplot equivalent to this is

#make unprojected map with Ultraplot
fig,ax = uplt.subplots(proj=crs, lonlim=(lon_min, lon_max), latlim=(lat_min, lat_max))
cx.add_basemap(ax=ax, crs=crs)
uplt.show()

These two code snippets make nearly identical maps.

However, issues arise when I try to use a projected CRS

#set up projected CRS
proj_crs = ccrs.LambertConformal(central_latitude=c_lat, central_longitude=c_lon)

#make projected map with Cartopy
fig, ax = plt.subplots(subplot_kw={"projection": proj_crs})
ax.set_extent([lon_min, lon_max, lat_min, lat_max], crs=ccrs.PlateCarree())
cx.add_basemap(ax=ax, crs=proj_crs)
plt.show()

#make projected map with Ultraplot
fig,ax = uplt.subplots(proj=proj_crs, lonlim=(lon_min, lon_max), latlim=(lat_min, lat_max))
cx.add_basemap(ax=ax, crs=proj_crs)
uplt.show()

The cartopy map looks fine:

But the Ultraplot version is all wonky:

I noticed the issue appear suddenly when using the legacy version of this package, proplot. Interestingly, the above code used to all work fine, projected CRS and all. To my knowledge, it just suddenly stopped working, without me updating any packages or anything. Love it when that happens... (although I'm willing to concede I may have updated something without knowing/remembering it).

I was unsure if this should go in this GitHub, or the contextily one. Let me know if it belongs in the other one!

[cvanelteren]

Hi @chudlerk thanks for reporting! Let's see if we can get to the source together. I can see that indeed in ultraplot 1.0 this issue is also present so the bug must be rather old. I can confirm your issue. There are probably some changes in both packages that we need to align.

I did some sanity checks. The projection does seem to work, i.e. we are in the correct area on the map, and the projection is "straightened":

The image to me looks like it is being interpolated (which I can confirm is the default behavior of contextily) but the interpolation itself won't be enough to not display a map at all.
I suspect that the tile is not setting the proper zoom level which is triggered for matplotlib. I think it is quicker to also take this up with contextily with a cross-reference here to see if we can align and trigger the proper zoom.

[cvanelteren]

Just checked with proplot and contextily seems to be broken there too; there are probable some changes on the contextily side we need to account for.

[cvanelteren]

Dunno if this is related: geopandas/contextily#261

@chudlerk could you pick it up on their end to ask for

• how they handle adding the basemap and how it interacts with an matplotlib axis object?

[cvanelteren]

Found the issue! You have to pass on the transform to imshow. We check for this in our backend. This works:

# make projected map with Cartopy
fig, ax = uplt.subplots(proj=proj_crs)
ax.format(
    lonlabels=True,
    latlabels=True,
    latlim=(lat_min, lat_max),
    lonlim=(lon_min, lon_max),
    # land=True,
    # ocean=True,
)
ax.set_extent([lon_min, lon_max, lat_min, lat_max], crs=ccrs.PlateCarree())
cx.add_basemap(ax=ax, crs=proj_crs, transform=proj_crs)
uplt.show()

[cvanelteren]

[cvanelteren]

Our backend infers transform keyword from special functions on geo to allow for plotting inside a coordinate system. Contextily does not pass this on in their source by default but you can pass it on manually as above.

I think contextily by default uses matplotlib's default projection. We are actively projecting data within the CRS to avoid having to manually specify it each time -- making it "just work". By default it will use PlateCarree when it does not see it within the inputs, which in this case does not work due to the LCC projection.

[cvanelteren]

I wonder if it would be better for us to just default to the projection or not..

[cvanelteren]

@chudlerk I changed the ordering of inference for the projection such that explicit input of the transform is not necessary. If the tests pass, and the reviews are cleared the issue will not exist anymore moving forward.

[cvanelteren]

@beckermr @chudlerk the issue between the mismatch is the default reference frame used by both libraries. We use PlateCarree for internal reference whereas contextily uses EPSG:3857. Therefore when interfacing with contextily the transform parameter has to be given.

[cvanelteren]

Closing the issue due to^