Antialiased contourf plot shows unexpected artifacts

[jamesp]

🐛 Bug Report

When plotting an iris cube with contourf in some circumstances an additional contour is shown. This appears to be related to the logic included by the antialiased option in iris.plot.contouf.

How To Reproduce

Steps to reproduce the behaviour:

1. Contact @jamesp for data and reproduction script

Expected behaviour

Antialiased data should not show additional artifacts when plotting.

Screenshots

Environment

• OS & Version: Linux
• Iris Version: 3.0.1. Affects Iris >= 2.3

[bjlittle]

@jamesp What version of mpl are you using?

[jamesp]

The antialiased artifact (additional contour) is present on all stack versions since Iris 1.13.

I've tested on matplotlib 2.2.3 and 3.3.4, both show the same additional contour. Setting antialiased=False restores the same behaviour as 1.13, which is what the user was running on previously.

I've tried to create a standardised test that reproduces the issue, but haven't achieved it yet, there does appear to be something pathological about the input data

[rcomer]

Here is a minimal example:

import iris
import iris.quickplot as qplt
import numpy as np
import matplotlib.pyplot as plt


lats = np.linspace(-10, 10, 20)
lons_2d, lats_2d = np.meshgrid(lats, lats)

data = np.cos(lons_2d * np.pi/10.) + np.cos(lats_2d * np.pi/10.)
np.place(data, data < 0., 0.)

cube = iris.cube.Cube(data)
for dim, name in enumerate(['longitude', 'latitude']):
    cube.add_dim_coord(iris.coords.DimCoord(lats, standard_name=name, units='degrees'), dim)

levels = [0.2, 2, 20, 200]
colors = ['b', 'r', 'y']

qplt.contourf(cube, levels=levels, colors=colors)

The problem seems to be triggered by a combination of

• having zeros everywhere other than our desired feature
• data values all being below the second level

If we pick level so that all the colours are used, all seems well

levels = [0.2, 0.5, 1, 2]
qplt.contourf(cube, levels=levels, colors=colors)

Since iris.plot.contourf does some subsetting of levels and colors, the problem boils down to

sub_levels = [2, 20]
sub_colors = ['b', 'r']

qplt.contour(cube, levels=sub_levels, colors=sub_colors)

which we also see if we plot directly with matplotlib

plt.contour(cube.data, levels=sub_levels, colors=sub_colors)

[rcomer]

matplotlib/matplotlib#20203