Apply xcdat to xarray-opened dataset

[lee1043]

Question criteria

• I added a descriptive title here.
• I searched the xCDAT GitHub Discussions to find a similar question and didn't find it.
• I searched the xCDAT documentation and/or Xarray documentation.

Describe your question

Do we have a capability to enable xcdat capability to already opened xarray dataset?

For example

import xarray as xr
import xcdat as xc
ds = xr.open_dataset(MYFILE)
ds_new = DO_SOMETHING(ds)
ds_new_ave = ds_new.spatial.average(data_var, axis=['X', 'Y'])

I am curious because if there is some capability like this, it might be possible to easily use sample data from xarray.tutorial.open_dataset or xarray.tutorial.load_dataset.

This might also enable use of xcdat combined with intake or intake-esm capability. Any idea or opinion will be appreciated!

[lee1043]

I guess this might be technically possible by making the _postprocess_dataset function as an importable API?

@tomvothecoder is that sounds a reasonable thing to do?

[pochedls]

@lee1043 – shouldn't this code work as is? The only reason I think this wouldn't work is if you didn't have bounds (which would have been added using xc.open_dataset but are not added with xr.open_dataset).

[lee1043]

@pochedls I was surprised it does work! I thought xcdat capability is only enabled when dataset was opened using xcdat, not xarray, but it looks like as long as xcdat is imported, xcdat capability is working for xarray Dataset. Is that correct? It is nice to learn the new thing, thanks!

[pochedls]

You're right. xcdat functionality will operate on xarray dataset objects after import (no need to open datasets with xcdat). Note that xc.open_dataset just uses/wraps xr.open_dataset with some custom features (e.g., adding bounds, handling a directory with mfdataset, dealing with months since... time units, and things like this).

[tomvothecoder]

Each of the xCDAT post-processing options applied in xc.open_dataset() and xc.open_mfdataset (via _postprocess_dataset) are available as standalone APIs, with the exception of _keep_single_var().

This includes:

• decode_time
• center_times
• swap_lon_axis
• add_missing_bounds

This means if you use Xarray's open_dataset() or open_mfdataset(), you can still apply xCDAT post-processing operations downstream.

[tomvothecoder]

The only reason I think this wouldn't work is if you didn't have bounds (which would have been added using xc.open_dataset but are not added with xr.open_dataset).

Yup, xCDAT computational APIs require bounds. Also xCDAT temporal APIs require decoded time coordinates.

[lee1043]

@tomvothecoder @pochedls thank you for the discussion. I think it would be still convenient to have the _postprocess_dataset as an API that wrapping up the decode_time, center_times, swap_lon_axis, and add_missing_bounds, so user does not have to call the 4 different processes individually, if this kind of workflow is used in their analysis when collectively process multiple models. What do you think?

[lee1043]

An example use case I can think of is with intake-esm. User opens multiple dataset through intake, which will open via xarray, then want to use xcdat but different models may have deficiencies in different part, so user want to systematically process them to comply with xcdat requirement to apply xcdat capability.