Grid description in case of variable dependent remapping methods

[sol1105]

I have a question regarding the global attribute grid. We plan to remap extensive quantities conservatively, and else with the nearest neighbor method. What do you suggest as the grid description? I could think of 2 options:

1

Rotated-pole latitude-longitude with 0.11 degree grid spacing, interpolated by nearest neighbor remapping from the original unstructured icosahedral ICON grid R13B05 (~12.1 km).

And specifying a variable comment for all variables where this is not the case.

2

Rotated-pole latitude-longitude with 0.11 degree grid spacing, interpolated by either second order conservative (extensive quantities) or nearest neighbor remapping (intensive quantities) from the original unstructured icosahedral ICON grid R13B05 (~12.1 km).

That would be my preference. If this is not sufficient, we could additionally include a variable comment denoting the actually used remapping method for each variable.

[larsbuntemeyer]

Both should be fine with the archiving specs, so i suggest, you can go with your preference. Of course, if your workflow allows it, you are welcome to include variable specific interpolation info, e.g., if it's an extensive or intensive quantity.

[sol1105]

Thank you for the quick reply :)

We will discuss whether to additionally include the specific interpolation info as variable comment. Workflow-wise it is possible, though a little more effort.

[jesusff]

Maybe I'm missing something here but, why not have for each variable the corresponding grid attribute with the interpolation that was really applied?

[larsbuntemeyer]

It's just that I am also used to having one dataset json file per experiment as described in the cmor_dataset_json API and not making it dependent on the variable_id. However, there is also set_cur_dataset_attribute which might be useful in this case to set some variable_id specific attributes during the workflow.

[sol1105]

Yes, it would be possible, but in our workflow more effort than the general grid description and specification via variable comment (which is also persistent when eg. merging multiple variables into one xarray dataset later on).

Update: There was the final decision to stick with nearest neighbour remapping for all variables.