Using cartopy projections sometimes results in garbled plots

Question

Using cartopy projections sometimes results in garbled plots

dennissergeev opened this issue 2 years ago · comments

🐛 Bug Report

I'm very grateful for geovista supporting cartopy projections, but the interface seems to contain a few bugs 😃

One bug leads to the data not displayed correctly for some values of the central_longitude parameter in the Plate Carree projection. For example, the plot looks OK for central_longitude=0 or 90 or 180, but not for 45, 135, etc. (see screenshots below). This is likely related to the arcsin issue, because I sometimes see that error too.

I also noticed that some of the projections are not displayed correctly, such as Mercator.

How to Reproduce

Click to expand this section...

Data source: LFRic C48 simulation.

import cartopy.crs as ccrs
import geovista as gv
import iris
from geovista.pantry import capitalise
from iris.experimental.ugrid import PARSE_UGRID_ON_LOAD

fname = "lfric_diag.nc"
with PARSE_UGRID_ON_LOAD.context():
    my_cube = iris.util.squeeze(iris.load_cube(fname, "toa_upward_longwave_flux"))
face_node = my_cube.mesh.face_node_connectivity
lons, lats = my_cube.mesh.node_coords
indices = face_node.indices_by_location()
mesh0 = gv.Transform.from_unstructured(
    lons.points,
    lats.points,
    indices,
    data=my_cube.data,
    start_index=face_node.start_index,
)

file_prefix = __file__.lower().replace(".py", "")

# Point no. 1: central longitude causes the mesh to display incorrectly
for cen_lon in [0, 45, 90, 135, 180]:
    mesh = mesh0.copy()
    plotter = gv.GeoPlotter(
        off_screen=True, crs=ccrs.PlateCarree(central_longitude=cen_lon), border=True
    )
    sargs = dict(title=f"{capitalise(my_cube.name())} / {my_cube.units}")
    plotter.add_title(f"LFRic C48\nPlateCarree(central_longitude={cen_lon})")
    plotter.add_mesh(
        mesh, cmap="thermal", show_edges=True, edge_color="grey", scalar_bar_args=sargs
    )
    # plotter.add_base_layer(texture=gv.natural_earth_1())
    # plotter.add_coastlines(resolution="10m", color="white")
    plotter.view_xy()
    # plotter.add_axes()
    plotter.camera.zoom(1.5)
    plotter.show(screenshot=f"{file_prefix}__platecarree_cen_lon{cen_lon}.png")

Expected Behaviour

Robust and valid plots.

Screenshots

Point no. 1: central longitude causes the mesh to display incorrectly

Environment

OS & Version: Ubuntu 22.04.1 LTS 64-bit
GeoVista Version: 0.1a1.dev460

Bill Little · Answer 1 · Thu Oct 20 2022 21:10:21 GMT+0800 (China Standard Time)

@dennissergeev Thanks for raising this 👍

There's some technical debt that I need to service which I think is related to this issue (and a few others) i.e., basically for geovista to take a more robust approach prior to projecting when ripping the unstructured mesh.

Bill Little · Answer 2 · Fri Jan 13 2023 21:21:41 GMT+0800 (China Standard Time)

@dennissergeev Just to let you know that I've been making good progress on this issue and essentially resolved the problem of cells smearing across the projection when ripping the mesh...

I've been stress testing the basic geovista projection support with more complex mesh geometries and uncovered a few other interesting cases where this can occur. So I'm going to tackle those also, then I'll update you here went you can pick up the changes from main and test at your end i.e., this fix isn't available just yet.

Additionally, I want to revisit pole singularity handling as there is obvious incorrect behaviour that requires to be flushed out... but once that's done, this would be a reasonable first stable step forwards.

Noted about Mercator support 👍 I'll create a separate issue for that. Over time the projection support will become more mature (gotta start somewhere 😉)... but in this particular case for Mercator I specifically require to honour max_latitude and min_latitude.

BTW It would be really great to know the basic projections that you tend to use for your workflow. This'll help me prioritise effort on those projections first, ensuring that they're feature complete and with most wrinkles ironed-out - flushing these out just needs mileage, so keep reporting issues to me when you find them.

Much appreciated 🤩🍻