imshow RGB plot overlaps map borders when using projection

[bask0]

xarry version: '0.14.0'
matplotlib version: '3.1.1'

When I create an RGB plot using ds.plot.imshow with a procection, the plot overlaps the map borders:

import cartopy.crs as ccrs
import xarray
import matplotlib.pyplot as plt

ds = xr.load_dataset(...)

ax = plt.subplot(projection=ccrs.Robinson())
(ds.isel(time=[0, 4, 8])).plot.imshow(robust=True, ax=ax, transform=ccrs.PlateCarree())

When using ds.plot.imshow with a single layer, the plot looks as expected:

import cartopy.crs as ccrs
import xarray
import matplotlib.pyplot as plt

ds = xr.load_dataset(...)

ax = plt.subplot(projection=ccrs.Robinson())
(ds.isel(time=0)).plot.imshow(robust=True, ax=ax, transform=ccrs.PlateCarree(), add_colorbar=False)

[dcherian]

Does this work if you use matplotlib directly? i.e. plt.plot(ds.isel(time=[0, 4, 8]).values, ...)

[bask0]

Does this work if you use matplotlib directly? i.e. plt.plot(ds.isel(time=[0, 4, 8]).values, ...)

@dcherian Thanks, yes, this works as expected:

ax = plt.subplot(projection=ccrs.Robinson())
ax.imshow(ds.isel(time=[0, 4, 8]).transpose('lat', 'lon', 'time').values[-1::-1, :, :], transform=ccrs.PlateCarree())

[dcherian]

Thanks, I'm marking this as a bug.

If you have the time, it would be nice if you could look into this and send in a PR.

We do some RGB stuff here:

xarray/xarray/plot/plot.py

Lines 632 to 642 in 8a650a1

	imshow_rgb = plotfunc.__name__ == "imshow" and darray.ndim == (
	3 + (row is not None) + (col is not None)
	)
	if imshow_rgb:
	# Don't add a colorbar when showing an image with explicit colors
	add_colorbar = False
	# Matplotlib does not support normalising RGB data, so do it here.
	# See eg. https://github.com/matplotlib/matplotlib/pull/10220
	if robust or vmax is not None or vmin is not None:
	darray = _rescale_imshow_rgb(darray, vmin, vmax, robust)
	vmin, vmax, robust = None, None, False

and imshow is here:

xarray/xarray/plot/plot.py

Lines 843 to 915 in 8a650a1

	@_plot2d
	def imshow(x, y, z, ax, **kwargs):
	"""
	Image plot of 2d DataArray using matplotlib.pyplot
	Wraps :func:`matplotlib:matplotlib.pyplot.imshow`
	While other plot methods require the DataArray to be strictly
	two-dimensional, ``imshow`` also accepts a 3D array where some
	dimension can be interpreted as RGB or RGBA color channels and
	allows this dimension to be specified via the kwarg ``rgb=``.
	Unlike matplotlib, Xarray can apply ``vmin`` and ``vmax`` to RGB or RGBA
	data, by applying a single scaling factor and offset to all bands.
	Passing ``robust=True`` infers ``vmin`` and ``vmax``
	:ref:`in the usual way <robust-plotting>`.
	.. note::
	This function needs uniformly spaced coordinates to
	properly label the axes. Call DataArray.plot() to check.
	The pixels are centered on the coordinates values. Ie, if the coordinate
	value is 3.2 then the pixels for those coordinates will be centered on 3.2.
	"""
	if x.ndim != 1 or y.ndim != 1:
	raise ValueError(
	"imshow requires 1D coordinates, try using " "pcolormesh or contour(f)"
	)
	# Centering the pixels- Assumes uniform spacing
	try:
	xstep = (x[1] - x[0]) / 2.0
	except IndexError:
	# Arbitrary default value, similar to matplotlib behaviour
	xstep = 0.1
	try:
	ystep = (y[1] - y[0]) / 2.0
	except IndexError:
	ystep = 0.1
	left, right = x[0] - xstep, x[-1] + xstep
	bottom, top = y[-1] + ystep, y[0] - ystep
	defaults = {"origin": "upper", "interpolation": "nearest"}
	if not hasattr(ax, "projection"):
	# not for cartopy geoaxes
	defaults["aspect"] = "auto"
	# Allow user to override these defaults
	defaults.update(kwargs)
	if defaults["origin"] == "upper":
	defaults["extent"] = [left, right, bottom, top]
	else:
	defaults["extent"] = [left, right, top, bottom]
	if z.ndim == 3:
	# matplotlib imshow uses black for missing data, but Xarray makes
	# missing data transparent. We therefore add an alpha channel if
	# there isn't one, and set it to transparent where data is masked.
	if z.shape[-1] == 3:
	alpha = np.ma.ones(z.shape[:2] + (1,), dtype=z.dtype)
	if np.issubdtype(z.dtype, np.integer):
	alpha *= 255
	z = np.ma.concatenate((z, alpha), axis=2)
	else:
	z = z.copy()
	z[np.any(z.mask, axis=-1), -1] = 0
	primitive = ax.imshow(z, **defaults)
	return primitive

[bask0]

Thanks, I'll give it a try

[mathause]

This happens in matplotlib when you pass an RGBA array. Thus the following should cause the same problem.

ax = plt.subplot(projection=ccrs.Robinson())
ax.imshow(ds.isel(time=[0, 1, 4, 8]).transpose('lat', 'lon', 'time').values[-1::-1, :, :], transform=ccrs.PlateCarree())

I.e. it is an matplotlib or cartopy error. However, xarray always adds an alpha channel internally, see

xarray/xarray/plot/plot.py

Lines 901 to 903 in 8a650a1

	# matplotlib imshow uses black for missing data, but Xarray makes
	# missing data transparent. We therefore add an alpha channel if
	# there isn't one, and set it to transparent where data is masked.