should pint-xarray return the units attr?

[raybellwaves]

I noticed pint-xarray converts to the desired units but along the way it loses the units attribute. I can see in the html repr it keeps track of the unit within the Quantity. Is it worth keeping a copy in the attrs as well as these are in used in plotting?

import pint_xarray
import xarray as xr

var = "precipitation_amount_1hour_Accumulation"
ds = xr.open_mfdataset(
    "s3://era5-pds/zarr/2020/01/data/{}.zarr".format(var),
    engine="zarr",
)
ds = ds.drop("time1_bounds")
ds.lat.attrs["units"] = "degree"
ds.lon.attrs["units"] = "degree"
da = ds.sel(lat=slice(2, 0), lon=slice(0, 2), time1="2020-01-01T00")[var]
da
<xarray.DataArray 'precipitation_amount_1hour_Accumulation' (lat: 9, lon: 9)>
dask.array<getitem, shape=(9, 9), dtype=float32, chunksize=(9, 9), chunktype=numpy.ndarray>
Coordinates:
  * lat      (lat) float32 2.0 1.75 1.5 1.25 1.0 0.75 0.5 0.25 0.0
  * lon      (lon) float32 0.0 0.25 0.5 0.75 1.0 1.25 1.5 1.75 2.0
    time1    datetime64[ns] 2020-01-01
Attributes:
    long_name:       Total precipitation
    nameCDM:         Total_precipitation_1hour_Accumulation
    nameECMWF:       Total precipitation
    product_type:    forecast
    shortNameECMWF:  tp
    standard_name:   precipitation_amount
    units:           m

q = da.pint.quantify(da.attrs["units"])
q
<xarray.DataArray 'precipitation_amount_1hour_Accumulation' (lat: 9, lon: 9)>
<Quantity(dask.array<getitem, shape=(9, 9), dtype=float32, chunksize=(9, 9), chunktype=numpy.ndarray>, 'meter')>
Coordinates:
  * lat      (lat) float32 2.0 1.75 1.5 1.25 1.0 0.75 0.5 0.25 0.0
  * lon      (lon) float32 0.0 0.25 0.5 0.75 1.0 1.25 1.5 1.75 2.0
    time1    datetime64[ns] 2020-01-01
Attributes:
    long_name:       Total precipitation
    nameCDM:         Total_precipitation_1hour_Accumulation
    nameECMWF:       Total precipitation
    product_type:    forecast
    shortNameECMWF:  tp
    standard_name:   precipitation_amount

c = q.pint.to("mm")
c
<xarray.DataArray 'precipitation_amount_1hour_Accumulation' (lat: 9, lon: 9)>
<Quantity(dask.array<mul, shape=(9, 9), dtype=float32, chunksize=(9, 9), chunktype=numpy.ndarray>, 'millimeter')>
Coordinates:
  * lat      (lat) float32 2.0 1.75 1.5 1.25 1.0 0.75 0.5 0.25 0.0
  * lon      (lon) float32 0.0 0.25 0.5 0.75 1.0 1.25 1.5 1.75 2.0
    time1    datetime64[ns] 2020-01-01
Attributes:
    long_name:       Total precipitation
    nameCDM:         Total_precipitation_1hour_Accumulation
    nameECMWF:       Total precipitation
    product_type:    forecast
    shortNameECMWF:  tp
    standard_name:   precipitation_amount

c.plot()

Desired:

c.attrs["units"] = "mm" # User specified in `pint.to` or the term returned by pint ("millimeter")
c.plot()

[keewis]

thanks for the issue, @raybellwaves. Keeping the units in the attributes is difficult because it would have to be synchronized with the units on the data, and most operations in xarray don't have hooks for that.

For now, the best workaround is to dequantify immediately before calling the plotting functions. In the future, I hope that we can just have this work automatically (you set the default format: ureg.default_format = "~P" (or pint-xarray does set a good default) and the matplotlib.units functionality does the rest). However, that currently doesn't work because xarray's plotting functions convert to ndarray (see #91 for the issue).

[raybellwaves]

Thanks. That works. And thanks a lot for you all your work here!