Help with weird uxarray, matplotlib, and orthographic plot

[erogluorhan]

Today I was trying out different things with uxarray and some data and found an...oddity. With the bokeh backend, I get a nice pretty image with an Orthographic projection:

but when I try and use the matplotlib backend:

Does anyone know what I might be doing wrong? My "relevant" code is:

import uxarray as ux
import geoviews as gv
import geoviews.feature as gf
import cartopy
import cartopy.feature as cf
import cartopy.crs as ccrs
import holoviews as hv

graticules = cf.NaturalEarthFeature(
    category='physical',
    name='graticules_15',
    scale='50m')

ofeatures = gf.coastline(
    projection=ccrs.Orthographic(), line_width=1, scale="50m"
) * gf.borders(
    projection=ccrs.Orthographic(), line_width=1, scale="50m"
) * gv.Feature(graticules, group='Lines')

ortho_plot = uxds['SLP'][0].plot.polygons(
    rasterize=True, 
    cmap='viridis', 
    backend='matplotlib', 
    projection=ccrs.Orthographic(),
).opts(
    global_extent=True,
    xaxis=None,
    yaxis=None 
) * ofeatures

Just the backend change in ortho_plot is the differences between these two. Note: This is a mix of uxarray examples and Gemini/ChatGPT/whatever so I freely admit I probably did something stupid!

Originally posted by @mathomp4 in #1488

[mathomp4]

@erogluorhan Here is a link to the file

It's pretty small to make things easy.

[mathomp4]

Whoops. I also forgot one sort of crucial thing in the Python snippet:

file_path = r"c24.nc4"
uxds = ux.open_dataset(file_path, file_path)

I mean, you need to open the file 😄

[alioacar]

The problem seems related to gv.Feature(graticules, group="Lines") using the Natural Earth graticules_15 feature. The base UXarray polygon plot, gf.coastline(), and gf.borders() all render correctly with the matplotlib backend but adding the graticules feature causes the rendering to collapse.

field = uxds["SLP"][0].plot.polygons(
    rasterize=True,
    cmap="viridis",
    backend="matplotlib",
    projection=ccrs.Orthographic()
).opts(
    xaxis=None,
    yaxis=None,
)
field
field * gf.coastline(scale="50m")
field * gf.borders(scale="50m")
field * gv.Feature(graticules, group="Lines")

As a workaround, I was able to get the expected result by avoiding gv.Feature(graticules_15) and instead adding gridlines directly through the Cartopy axis using ax.gridlines(...) after rendering:

renderer = hv.renderer("matplotlib")
overlay = (
    field
    * gf.coastline(scale="50m")
    * gf.borders(scale="50m")
)
mpl_obj = renderer.get_plot(overlay)

mpl_obj.handles["axis"].gridlines(
    draw_labels=False,
    linewidth=0.6
)

This suggests the issue may be somewhere in the GeoViews/Cartopy feature overlay or clipping pipeline rather than in UXarray itself. Hopefully this helps @mathomp4 :)

[erogluorhan]

@alioacar thanks a lot for looking into this and suggesting a workaround! I'll try to think about this case a bit more when I have some time.

[Sevans711]

Confirming, this looks like an upstream issue, I was able to reproduce it without using UXarray at all. See holoviz/geoviews#845 for more details!

[rajeeja]

Confirming, this looks like an upstream issue, I was able to reproduce it without using UXarray at all. See holoviz/geoviews#845 for more details!

Thanks for pointing this out. It seems like our fix still holds

[erogluorhan]

Confirming, this looks like an upstream issue, I was able to reproduce it without using UXarray at all. See holoviz/geoviews#845 for more details!

Thanks for pointing this out. It seems like our fix still holds

So, are you suggesting that we should still put together a workaround, maybe something close to what @alioacar suggested? I think I lean more towards waiting for the fix on the holoviz/geoviews/issues/845.

[alioacar]

I initially thought this was a more complex overlay issue. @erogluorhan I agree that it's better waiting for the fix to be merged

[rajeeja]

I misspoke — I was thinking of #1515 which fixed Bokeh Orthographic plots, but that's unrelated. The matplotlib+graticules issue is purely upstream in geoviews (holoviz/geoviews#845). Agree with you: wait for holoviz/geoviews#846 to merge.

[erogluorhan]

Okay, perfect, thanks for clarifying @rajeeja and @alioacar !

Let's keep this open until the geoviews fix is released, and then we can give this a second try and hopefully close it once confirmed things work.