Data variable CRS

[benbovy]

Currently Dataset.proj("coord_name").crs can be used to access the CRS of either a spatial reference (scalar) coordinate or a CRS-aware indexed coordinate.

This could probably be extended to data variables as well, which may also be CRS-aware if they have spatial dimensions. Dataset.proj("data_var_name").crs could either return:

• The CRS returned by Dataset.proj("spatial_ref_name").crs if Dataset["data_var_name"].attrs["grid_mapping"] == "spatial_ref_name" (as inspired by CF conventions)
• Implicitly fallback to Dataset.proj.crs if there is a unique CRS

To set the CRS of a data variable we could reuse Dataset.map_crs(spatial_ref_name=["data_var_name"]), which will basically (re)set the "grip_mapping" attribute of the given data variable(s).

[dcherian]

I agree with setting the grid_mapping attribute. My general preference is to encourage users to annotate their data properly rather than rely on implicit behaviour.

[benbovy]

I agree with your preference, instead of a fallback we could simply raise an error with some clear instructions in the message. Manually annotating data may be tedious, though. It would be nice if we could provide some API to help doing that in an automated but explicit way.