issue_comments

Any updates on this? It would be great to have an option for multiple colorbars for each axis of the FacetGrid, rather than the default of one single colorbar for each axis. The main use case would be when different axes span different orders of magnitude. A global clim hides the variability of the smaller-scale axes.

@shoyer:

If your plots should have different colorbars, then arguably they should be different data variables in a Dataset.

I see the reasoning here. Could the solution be an xr.Dataset.plot.imshow method, which optionally takes a list of cmaps, that automatically produces a new colorbar for each data_var/axis?

In the meantime, manual subplot and colorbar construction with matplotlib works, but is annoying.

I like @shoyer 's suggestion as well. This would be very useful if the data being plotted are equal-sized images, in which case the colorbar could be dropped altogether as it no longer makes sense for them to share the same range.

I tend to agree with you @shoyer. In that case it would be nice to have a Dataset.plot convenience method for plotting all the variables in a Dataset. This method could handle all the annoying boiler plate associated with subplots/AxesGrid.

This does kind of go against the point of FacetGrid, which is to display facets of the same variables in a grid. If your plots should have different colorbars, then arguably they should be different data variables in a Dataset.

That said, this does seem similar in spirit to the sharex and sharey arguments. Maybe sharez or share_vlim would be appropriate arguments for changing this?

Squeezing multiple colorbars in will be a little trickier than just allowing colorbar limits to vary, but if it can be done in a sane fashion that doesn't make the existing code more complex I would be in support.

The AxesGrid API isn't actually that bad. See this

```python from mpl_toolkits.axes_grid1 import ImageGrid

import xarray as xr

fig = plt.figure(figsize=(6,2.5)) grid = ImageGrid(fig, 111, (1,2), axes_pad=.1, cbar_location='bottom', cbar_mode='each', cbar_pad=0.3, aspect=False, cbar_size='3%')

for (lab,data), ax, cax in zip(t.groupby('season'), grid, grid.cbar_axes): im = ax.contourf(data) cax.colorbar(im) ```

whoops. accidentally hit the close button.

@fmaussion I was worried that would be the case. I think matplotlib's AxesGrid module is the appropriate tool to handle this use case. Unfortunately, I find the AxesGrid API pretty unpleasant for interactive use.

Seems reasonable, although this would imply quite a lot of change in the current logic. For example, facetgrids share x and/or y axis:

If there is a colorbar in between, would you like to share axes?

The current way to do what you need is to do it manually:

ds = xr.tutorial.load_dataset('air_temperature') t = ds.air.groupby('time.season').mean(dim='time') f, (ax1, ax2) = plt.subplots(1, 2, figsize=(7, 3)) t.isel(season=0).plot(ax=ax1) t.isel(season=1).plot(ax=ax2)

As you see, this messes up with the layout so a call to tight_layout() is necessary.