Plotting datasets with variable coordinates

[smillerc]

I'm trying to plot a dataset that has variable coordinates, and am getting behavior that is unexpected, and I wonder if it is a design issue or a user issue (i.e. I don't grok xarray). I've posted the question on SO (https://stackoverflow.com/questions/45474153/plotting-xarray-datasets-with-variable-coordinates).

Is the behavior that xarray is showing by design or is it a corner case that I'm getting because I'm not using xarray in the way it was originally intended?

[fmaussion]

On 08/06/2017 11:37 PM, Sam Miller wrote: Is the behavior that xarray is showing by design or is it a corner case that I'm getting because I'm not using xarray in the way it was originally intended?

Xarray follows the CF conventions as far as possible, and the convention states that dimension coordinates should have the same name as their dimension. See the docs for terminology: http://xarray.pydata.org/en/latest/data-structures.html#coordinates What we could do however is to add a "x" kwarg to plot.line so that one can specify which coordinate to use for the plot, like plot.pcolormesh i.e. it would be possible to do: ds.variable.plot(x='r') Instead of renaming the dimension for it to become a dimension coordinate. Thoughts? Cheers, Fabien

[smillerc]

I think that would work. I've just been trying to understand what the CF conventions mean for me in my use case, where I have a grid that moves over time, but keeps the same number of nodes and cells. I feel like xarray could be a big help, but I'm still trying to figure it out.

Thanks for the quick responses here and on SO.

[dcherian]

Does this work now?

[fmaussion]

No it doesn't! Your most recent PR still checks for the kwarg in dims, not in coordinates.

[dcherian]

Shall I send in a PR to check da.variables then?

[rabernat]

@kthyng & @hetland: this issue is related to our lunch discussion today about how to automatically use multidimensional coordinates when plotting.

[hetland]

To follow up with this, plotting using 'x' and 'y' kwargs works fine for now, even when the variable coordinates do not share the same number of dimensions. (i.e., a common case of plotting a cross section with sigma- or s-coordinates where the horizontal variable is 1D, but the vertical variable is 2D). So, I think this great, and should cover most use cases.

The only approach I can think of to do this automatically would be to find variable coordinates that have matching dimensions. This would probably work in lots of cases, but unexpectedly break in others. I think I would prefer explicit over clever, unless we can think of a way to do it that won't break.