@jbusecke pretty sure this is being caused by the _strip_all_coords function you added in the padding code. I think I understand why you want to strip coords along dimensions along which you are padding, but why also strip them along all other dimensions too?

It was a confusing bug to find because all grid-related coordinates are re-attached at the end of apply_as_grid_ufunc but in this case time is not in the grid, but it is stripped by _strip_all_coords, so it gets lost.

It was a confusing bug to find because all grid-related coordinates are re-attached at the end of apply_as_grid_ufunc but in this case time is not in the grid, but it is stripped by _strip_all_coords, so it gets lost.

I see. That is not desirable. Can you come up with a synthetic test case that reproduces this failure as a PR?

From what I can remember the issue here is that padding with nans caused a bunch of issues, and thus stripping the coordinates helped.

Maybe we can relax the stripping by doing something along these lines?

axis_dims = _axis_names_from_boundary_width(boundary_width)
data = _strip_coords(data, axis_dims) # only stripping the dimension coords along the axes that are actually padded.