Adding units/description to netcdf files

[chuaxr]

Currently, the descriptions/units in Var objects are not saved in the output netcdf files. It would be nice if this was the case.

[spencerahill]

Sure, I agree this would be nice, and it also should be relatively straightforward for us to implement.

Initial quick take: this will require adding a 'units' attribute to the DataArray somewhere in the Calc pipeline. One catch is different units whether or not the output type is a vertical integral; that logic is handled here: https://github.com/spencerahill/aospy/blob/develop/aospy/calc.py#L457

[spencerkclark]

One catch is different units whether or not the output type is a vertical integral

Indeed this is true; in the absence of more robust units support (e.g. with pint) there are more or less two options in this case:

1. Simply drop the units attribute
2. Append 'kg m^-2' to the units attribute, which while dimensionally correct, might look a little strange for units that could be simplified.

[spencerahill]

Append 'kg m^-2' to the units attribute, which while dimensionally correct, might look a little strange for units that could be simplified.

True, but I believe that's better than no units. What about being explicit about the vertical integral operation? E.g. '(vertical integral of {0}): {0} kg m^-2)'.format(units).

[spencerkclark]

True, but I believe that's better than no units. What about being explicit about the vertical integral operation? E.g. '(vertical integral of {0}): {0} kg m^-2)'.format(units).

Sounds good to me!

[micahkim23]

I started looking at this issue and was planning on adding attrs to data in the save function https://github.com/spencerahill/aospy/blob/develop/aospy/calc.py#L669. The only weird thing is that data can either be an DataArray or OrderedDict.

OrderedDict([('globe', <xarray.DataArray ()>
array(1.0588634264346013e-05)
Coordinates:
    raw_data_start_date  datetime64[ns] 1678-01-01
    raw_data_end_date    datetime64[ns] 1681-01-01
    subset_start_date    datetime64[ns] 1678-01-01
    subset_end_date      datetime64[ns] 1680-12-31), ('tropics', <xarray.DataArray ()>
array(5.266911604961124e-06)
Coordinates:
    raw_data_start_date  datetime64[ns] 1678-01-01
    raw_data_end_date    datetime64[ns] 1681-01-01
    subset_start_date    datetime64[ns] 1678-01-01
    subset_end_date      datetime64[ns] 1680-12-31)])

In this case I add the units/description to each DataArray inside of the OrderedDict?

[spencerahill]

@micahkim23 I think we can avoid this if we add the metadata earlier in the pipeline. E.g. before this line in Calc.compute. I think the subsequent time and regional reductions will preserve the attrs.

[micahkim23]

@spencerahill When I add attrs to a Dataset, the Dataset gets the attrs but the DataArrays inside the Dataset don't get the attrs. Do you want me to add the attrs to the DataArrays inside the Dataset?

[spencerkclark]

@micahkim23 I think it would make the most sense if these attributes were added to only DataArrays (because the units and description attributes are variable-specific attributes). Attributes of a Dataset can be thought of as more "global" attributes, in the sense that they apply to all the variables it contains.