Support DataTree in xarray.merge

[shoyer]

Sub-issue of #9106

[lukegre]

Just thought I'd drop a comment about a possible bug I found using the datatree opener. Possibly not in the right issue, but semi-related.

Below are two plots showing the strange behaviour when using datatrees to combine the data vs regular zarr. Not sure what's causing this.

A note on the structure of the data, I've saved each year to a group in the zarr store. I then concatenate them by looping through the groups.

s3_uri = 'link to s3 bucket with era5 t2m (2 meter temperature)'
xy = 'latitude', 'longitude'

# concat using datatree
datatree = xr.open_datatree(s3_uri, engine='zarr', consolidated=True, chunks={})
groups = list(datatree)  # list of str 1980 ... 2022
ds_treelist = [datatree[g].ds for g in groups]
ds_treecat = xr.combine_nested(ds_treelist, concat_dim="time")  # same behaviour with xr.concat

ds_treecat.t2m.mean(xy).plot()

# concat opening each group with zarr

ds_zarrlist = [xr.open_zarr(store=s3_uri, group=g, consolidated=True) for g in groups]
ds_zarrcat = xr.combine_nested(ds_zarrlist, concat_dim='time')

ds_zarrcat.t2m.mean(xy).plot()

[shoyer]

@lukegre please open a new issue for your bug, ideally with a minimal reproducing example (see the bug template for specifics)

[lukegre]

@lukegre please open a new issue for your bug, ideally with a minimal reproducing example (see the bug template for specifics)

Reported in #9912, please feel free to delete my comment above (and this one).