DataArrays to/from Zarr Arrays

[rabernat]

Right now, open_zarr and Dataset.to_zarr only work with Zarr groups. Zarr Groups can contain multiple Array objects.

It would be nice if we could open Zarr Arrays directly as xarray DataArrays and write xarray DataArrays directly to Zarr Arrays.

However, this might not make sense, because, unlike xarray DataArrays, zarr Arrays can't hold any coordinates.

Just raising this idea for discussion.

[shoyer]

Take a look at DataArray.to_netcdf() for one approach. We save the DataArray into a netCDF group, saved into a variable with the special name __xarray_dataarray_variable__.

[maxrjones]

We created https://github.com/carbonplan/xrefcoord to generate coordinates from GeoTiff metadata using kerchunk reference files and xarray without needing to store the coordinates. As discussed in fsspec/kerchunk#313, the ability to open Zarr arrays in xarray would allow using kerchunk on a broader set of GeoTIFFs but would require experience with the backend code and/or non-trivial amounts of time.

Do you think using xarray's flexible backends with zarrita could provide a way to both solve this problem and experiment with the most recent v3 spec in xarray without requiring as much experience with xarray's current backend code?

[krokosik]

The functionality to call to_zarr has been added and they can be then opened using `open_dataset('...', engine='zarr'), so I think this issues could be closed?
#7692

[maxrjones]

The functionality to call to_zarr has been added and they can be then opened using `open_dataset('...', engine='zarr'), so I think this issues could be closed? #7692

They primary functionality we found missing in fsspec/kerchunk#313 was the ability to open Zarr arrays as xarray DataArrays. afaik #7692 allows dataarrays to be written to Zarr groups, and Zarr groups can be read as datasets, but Zarr arrays still cannot be read by Xarray. The ability to directly read Zarr arrays would be extremely helpful for reading data objects created outside Xarray.

[LaurentBerder]

Hi, is there any update on this?
I also am saving a DataArray to zarr, which I would then need to read as a DataArray (right now it only reads as DataSet).

[rabernat]

For everyone interested in this feature: we would welcome a PR! It should be a straightforward extension of open_dataaaray. The xarray core dev team would be happy to provide support and advice along the way. To get started, check out the contributing guide.

[LaurentBerder]

For everyone interested in this feature: we would welcome a PR! It should be a straightforward extension of open_dataaaray. The xarray core dev team would be happy to provide support and advice along the way. To get started, check out the contributing guide.

Hi @rabernat,
Unhappily, I'm not (in any short-medium time) going to have bandwidth for a PR.
However, I managed to get this code running for myself:

import xarray as xr

ds = xr.open_zarr(path)
array = ds.to_array(dim='extra').squeeze('extra')

That seems to work easily enough.