[Enhancement]: Xarray: Deprecated since version 2025.01.1: Please pass a coders.CFDatetimeCoder instance initialized with use_cftime to the decode_times kwarg instead.

[tomvothecoder]

Is your feature request related to a problem?

xarray=2025.01.1 deprecates the use_cftime kwarg, which affects xcdat.open_dataset() and xcdat.open_mfdataset().

Related to: https://github.com/pydata/xarray/releases/tag/v2025.01.1

It also includes an update to the time decoding infrastructure as a step toward pydata/xarray#9618.

Describe the solution you'd like

1. Add code that checks the Xarray version to determine what to use

	if xarray.__version__ < 2025.01.1: 
		use_cftime = True
    else:
    	decode_times = coders.CFDatimeCoder(use_cftime=True) 

Describe alternatives you've considered

No response

Additional context

[tomvothecoder]

FYI @xCDAT/core-developers the xarray.coders.CFDatetimeCoder has the ability decode and encode the time coordinates variables. I think we've ran into situations where we need the raw float time coordinates, but an encoding feature was not available before.

[tomvothecoder]

Actually, I don't think this deprecation affects xCDAT because we use our own custom functions for decoding time coordinates as cftime objects. No deprecation warnings are raised in the test suite.

• Custom function called decode_time()

  • xcdat/xcdat/dataset.py

    Lines 248 to 386 in c93b5bf

    	def decode_time(dataset: xr.Dataset) -> xr.Dataset:
    	"""Decodes CF and non-CF time coordinates and time bounds using ``cftime``.
    	By default, ``xarray`` only supports decoding time with CF compliant units
    	[5]_. This function enables also decoding time with non-CF compliant units.
    	It skips decoding time coordinates that have already been decoded as
    	``"datetime64[ns]"`` or ``cftime.datetime``.
    	For time coordinates to be decodable, they must have a "calendar" attribute
    	set to a CF calendar type supported by ``cftime``. CF calendar types
    	include "noleap", "360_day", "365_day", "366_day", "gregorian",
    	"proleptic_gregorian", "julian", "all_leap", or "standard". They must also
    	have a "units" attribute set to a format supported by xCDAT ("months since
    	..." or "years since ...").
    	Parameters
    	----------
    	dataset : xr.Dataset
    	Dataset with numerically encoded time coordinates and time bounds (if
    	they exist). If the time coordinates cannot be decoded then the original
    	dataset is returned.
    	Returns
    	-------
    	xr.Dataset
    	Dataset with decoded time coordinates and time bounds (if they exist) as
    	``cftime`` objects.
    	Raises
    	------
    	KeyError
    	If time coordinates were not detected in the dataset, either because they
    	don't exist at all or their CF attributes (e.g., 'axis' or
    	'standard_name') are not set.
    	Notes
    	-----
    	Time coordinates are represented by ``cftime.datetime`` objects because
    	it is not restricted by the ``pandas.Timestamp`` range (years 1678 through
    	2262). Refer to [6]_ and [7]_ for more information on this limitation.
    	References
    	-----
    	.. [5] https://cfconventions.org/cf-conventions/cf-conventions.html#time-coordinate
    	.. [6] https://docs.xarray.dev/en/stable/user-guide/weather-climate.html#non-standard-calendars-and-dates-outside-the-timestamp-valid-range
    	.. [7] https://pandas.pydata.org/pandas-docs/stable/user_guide/timeseries.html#timestamp-limitations
    	Examples
    	--------
    	Decode the time coordinates in a Dataset:
    	>>> from xcdat.dataset import decode_time
    	>>>
    	>>> ds.time
    	<xarray.DataArray 'time' (time: 3)>
    	array([0, 1, 2])
    	Coordinates:
    	* time (time) int64 0 1 2
    	Attributes:
    	units: years since 2000-01-01
    	bounds: time_bnds
    	axis: T
    	long_name: time
    	standard_name: time
    	calendar: noleap
    	>>>
    	>>> ds_decoded = decode_time(ds)
    	>>> ds_decoded.time
    	<xarray.DataArray 'time' (time: 3)>
    	array([cftime.DatetimeNoLeap(1850, 1, 1, 0, 0, 0, 0, has_year_zero=True),
    	cftime.DatetimeNoLeap(1850, 1, 1, 0, 0, 0, 0, has_year_zero=True),
    	cftime.DatetimeNoLeap(1850, 1, 1, 0, 0, 0, 0, has_year_zero=True)],
    	dtype='object')
    	Coordinates:
    	* time (time) datetime64[ns] 2000-01-01 2001-01-01 2002-01-01
    	Attributes:
    	units: years since 2000-01-01
    	bounds: time_bnds
    	axis: T
    	long_name: time
    	standard_name: time
    	calendar: noleap
    	View time encoding information:
    	>>> ds_decoded.time.encoding
    	{'source': None,
    	'dtype': dtype('int64'),
    	'original_shape': (3,),
    	'units': 'years since 2000-01-01',
    	'calendar': 'noleap'}
    	"""
    	ds = dataset.copy()
    	coord_keys = _get_all_coord_keys(ds, "T")
    	if len(coord_keys) == 0:
    	raise KeyError(
    	"No time coordinates were found in this dataset to decode. If time "
    	"coordinates were expected to exist, make sure they are detectable by "
    	"setting the CF 'axis' or 'standard_name' attribute (e.g., "
    	"ds['time'].attrs['axis'] = 'T' or "
    	"ds['time'].attrs['standard_name'] = 'time'). Afterwards, try decoding "
    	"again with `xcdat.decode_time`."
    	)
    	for key in coord_keys:
    	coords = ds[key].copy()
    	if _is_decodable(coords) and not _is_decoded(coords):
    	if coords.attrs.get("calendar") is None:
    	coords.attrs["calendar"] = "standard"
    	logger.warning(
    	f"'{coords.name}' does not have a calendar attribute set. "
    	"Defaulting to CF 'standard' calendar."
    	)
    	decoded_time = _decode_time(coords)
    	ds = ds.assign_coords({coords.name: decoded_time})
    	try:
    	bounds = ds.bounds.get_bounds("T", var_key=coords.name)
    	except KeyError:
    	bounds = None
    	if bounds is not None and not _is_decoded(bounds):
    	# Bounds don't typically store the "units" and "calendar"
    	# attributes required for decoding, so these attributes need to be
    	# copied from the coordinates.
    	bounds.attrs.update(
    	{
    	"units": coords.attrs["units"],
    	"calendar": coords.attrs["calendar"],
    	}
    	)
    	decoded_bounds = _decode_time(bounds)
    	ds = ds.assign({bounds.name: decoded_bounds})
    	return ds

• use_cftime=True is passed to xarray.coding.decode_cf_datetime but it is not deprecated here.

  • xcdat/xcdat/dataset.py

    Line 648 in c93b5bf

    return decode_cf_datetime(offsets, units, calendar=calendar, use_cftime=True)

  • https://github.com/pydata/xarray/blob/609412d8544217247ddf2f72f988da1b38ef01bc/xarray/coding/times.py#L492-L565