issues

Cell boundaries can be either N+1 sized arrays as suggested by xgcm/xmitgcm#15, or (N,2) sized arrays as suggested by the CF convention. However, a DataArray cannot hold both kinds of coordinate variables because they contain a new dimension.

If you try to assign a new coordinate to a DataArray by dr.assign_coords(), you will get ValueError: cannot add coordinates with new dimensions to a DataArray

On the other hand, if your DataSet contains cell boundary variables (for example, #667), the bounds will be dropped when you extract a single variable into a DataArray.

Having cell bounds available in a DataArray is important for a couple of applications:

• Pass cell bounds to DataArray's plotting methods (xgcm/xmitgcm#15). I am aware of the discussion about inferring boundaries (#781). However, for the Cube-Sphere grid or the Lat-Lon-Cap grid (reference) which have tiles covering the poles, I have to explicitly pass cell bounds to the original plt.pcolormesh() to get a good-looking plot. (see this comment for details)

• For conservative (i.e. area-weighted) regridding (mentioned in #486). Cell centers are enough for bilinear interpolation or other simple resamping, but for any Finite-Volume meshes, knowing the boundaries is crucial if you want to conserve the total amount of mass or flux.

Plotting or regridding will work fine if you pass cell bounds as an additional argument to a wrapper function. However, having a single DataArray object containing boundary information seems like a more elegant solution. Is it possible to let DataArray accept N+1 sized coordinate variables, and be able to inherit them from the parent DataSet? If that's too drastic, is it possible to write an accessor to extend DataArray's capability? Say, a "bound" accessor for a new attribute ds.bnd['lat_b'], which can be kept when a DataArray gets extracted (ds['data_var'].bnd['lat_b'] )? Does this make sense?

I am evaluating interp() against xESMF. Here's how xESMF would regrid the built-in 'rasm' data to a regular lat-lon grid.

Seems like interp() can convert rectilinear grids (1D coordinates) to curvilinear grids (2D coordinates), according to the last example. How about the reverse? Can it convert 2D coordinate to 1D, or to another 2D coordinate?

That's the test data: dr = xr.tutorial.load_dataset('rasm')['Tair'] ... Coordinates: * time (time) datetime64[ns] 1980-09-16T12:00:00 1980-10-17 ... xc (y, x) float64 189.2 189.4 189.6 189.7 189.9 190.1 190.2 190.4 ... yc (y, x) float64 16.53 16.78 17.02 17.27 17.51 17.76 18.0 18.25 ...

That's a simple destination grid: lon = xr.DataArray(np.linspace(-180, 180, 120), dims=['lon']) lat = xr.DataArray(np.linspace(-90, 90, 60), dims=['lat'])

I would expect a syntax like: dr_out = dr.interp(xc=lon, yc=lat)

But I got ValueError: dimensions ['xc', 'yc'] do not exist, becauseinterp() only accepts dimensions (which must be 1D), not coordinates.

dr.interp(x=lon, y=lat) runs but the result is not correct. This is expected because x does not mean longitude in the original data.

@crusaderky @fujiisoup

The "Edit on GitHub" button on ReadTheDocs is linked to https://github.com/pydata/xarray/blob/origin/stable/doc/index.rst which doesn't exist. The correct URL would be without "origin": https://github.com/pydata/xarray/blob/stable/doc/index.rst

If func changes the size of the core dimension, apply_ufunc(func, ..) will only change the data variable dimension but keep the coordinate dimension and value unchanged. The resulting DataArray cannot be saved to a NetCDF file due to dimension inconsistency. Please see this GitHub gist for a simple illustration.

Would it be more natural to drop the core coordinate by default? This is safer (will not produce illegal NetCDF file) and also makes more physical sense (the core coordinate is likely to change).

I'd like use apply_ufunc to track metadata in my xESMF package, but the output DataArray will have incorrect lon and lat coordinate dimension and values. It is easy to manually correct this afterwards but I want to bring up this issue.

Output of xr.show_versions()