Support CF Aggregation Variables

[aulemahal]

Is your feature request related to a problem?

New and fresh from the 1.13 CF conventions released in December 2025 : the « Aggregation variable », a new syntax for creating ncml-like aggregation within valid files (mostly netCDF).

https://cfconventions.org/Data/cf-conventions/cf-conventions-1.13/cf-conventions.html#aggregation-variables

The idea is to explicitly map external variables (i.e. in another file) to section of a "virtual" variable in the main file.

Describe the solution you'd like

Would be nice if xarray was able to open these transparently, with the correct combination of combine_by_coords and such.

Describe alternatives you've considered

As this is a I/O thing, I thought it would be best to be done here, instead of cf-xarray which concerns itself with more advance/high-level stuff. And making xr.open_dataset transparent to this seems a more user-friendly idea than having an extra function from an external package handle this.

Additional context

This new thing was a standalone extension to the CF conventions (under the name CFA) since 2023 (at least) and is already implemented by package cf-python.

[TomNicholas]

Interesting - so this is basically like virtual zarr but in netcdf?

Could a Zarr store follow the same aggregation convention?

the correct combination of combine_by_coords and such.

What combining is required? Does the netCDF file not already completely specify how to interpret an aggregated variable?

(i.e. in another file)

I wonder how much more complex the netcdf backend has to get to generalize to handling many file handles...

We could add this in xarray, but it could also be added to virtualizarr's HDFParser (or in a new CF-specific parser).

[aulemahal]

Could a Zarr store follow the same aggregation convention?

Probably! This is a CF convention thing, it is independent from the netCDF format.

What combining is required?

Maybe I'm not using the best words, I meant "combining" as in the low-level concatenation of (chunked?) arrays from other files. Indeed, combine_by_coords might not be useful here.