@hding1981 you would need to make a MCVE like this: https://matthewrocklin.com/blog/work/2018/02/28/minimal-bug-reports in a new issue. It's unlikely someone can help you debug your data.

I am calculating monthly mean anomalies from a CMIP6 file, which is attached and ended by .docx (Please remove .docx from its name after downloading it).

This file has reasonable time coordinate, which has been confirmed by ncdump and cdo. But it shows NaT after a time point by reading the file using xr.open_dataset. I printed out outputs of processing this file using some commands in a jyputer notebook in a pdf file, which is also attached. I really have no idea why xr.open_datase cannot read the time coordinate properly.

Then, I thought maybe I can redefine its time axis by pd.date_range. But it is also not working.

Thank you so much!

Untitled5 - Jupyter Notebook.pdf ts_Amon_CNRM-CM6-1_piControl_r1i1p1f2_gr_185001-234912.tmp.nc.docx

@hding1981 IIUC this isn't supported at the moment. Though with the approaching index refactor, we may be able to support this more easily

I have similar problems with time coordinate. How did you solve your problem in the end?

I also list my problem here.

The following are my commands in a jupyter notebook

import xarray as xr import numpy as np import pandas as pd import sys

dset=xr.open_dataset("input/ts_Amon_CNRM-CM6-1_piControl_r1i1p1f2_gr_185001-234912.nc",decode_times=False) dset['time'] = pd.period_range(start='1850-01-15', end='2349-12-15', freq='M') varname="ts" anom = dset[varname].groupby('time.month')-dset[varname].groupby('time.month').mean('time', keep_attrs=True)

Then, I got the following error message.

KeyError Traceback (most recent call last) ~/miniconda3/envs/python_tutorial/lib/python3.9/site-packages/xarray/core/dataarray.py in _getitem_coord(self, key) 692 try: --> 693 var = self._coords[key] 694 except KeyError:

KeyError: 'time.month'

During handling of the above exception, another exception occurred:

AttributeError Traceback (most recent call last) <ipython-input-1-250a6867750b> in <module> 7 dset['time'] = pd.period_range(start='1850-01-15', end='2349-12-15', freq='M') 8 varname="ts" ----> 9 anom = dset[varname].groupby('time.month')-dset[varname].groupby('time.month').mean('time', keep_attrs=True)

~/miniconda3/envs/python_tutorial/lib/python3.9/site-packages/xarray/core/common.py in groupby(self, group, squeeze, restore_coord_dims) 703 ) 704 --> 705 return self._groupby_cls( 706 self, group, squeeze=squeeze, restore_coord_dims=restore_coord_dims 707 )

~/miniconda3/envs/python_tutorial/lib/python3.9/site-packages/xarray/core/groupby.py in init(self, obj, group, squeeze, grouper, bins, restore_coord_dims, cut_kwargs) 313 f"Received {group!r} instead." 314 ) --> 315 group = obj[group] 316 if len(group) == 0: 317 raise ValueError(f"{group.name} must not be empty")

~/miniconda3/envs/python_tutorial/lib/python3.9/site-packages/xarray/core/dataarray.py in getitem(self, key) 702 def getitem(self, key: Any) -> "DataArray": 703 if isinstance(key, str): --> 704 return self._getitem_coord(key) 705 else: 706 # xarray-style array indexing

~/miniconda3/envs/python_tutorial/lib/python3.9/site-packages/xarray/core/dataarray.py in getitem_coord(self, key) 694 except KeyError: 695 dim_sizes = dict(zip(self.dims, self.shape)) --> 696 , key, var = _get_virtual_variable( 697 self._coords, key, self._level_coords, dim_sizes 698 )

~/miniconda3/envs/python_tutorial/lib/python3.9/site-packages/xarray/core/dataset.py in _get_virtual_variable(variables, key, level_vars, dim_sizes) 179 data = getattr(ref_var.dt, var_name).data 180 else: --> 181 data = getattr(ref_var, var_name).data 182 virtual_var = Variable(ref_var.dims, data) 183

AttributeError: 'IndexVariable' object has no attribute 'month'

In order to maintain a list of currently relevant issues, we mark issues as stale after a period of inactivity

If this issue remains relevant, please comment here or remove the stale label; otherwise it will be marked as closed automatically

Potentially we could some sort of API for surfacing this information, e.g., a pandas_dtype property to xarray.Variable.

It's probably cleaner, because the dtype of PeriodIndex can be several things, i.e. 'period[M]' in my case.

It's only an dtype=object array externally: internally, the data is still stored as a pandas.PeriodIndex. I think you'll find that inside ref_var._data.array .

Potentially we could some sort of API for surfacing this information, e.g., a pandas_dtype property to xarray.Variable.

OK, so the problem is that when given to is_datetime_like here, the pd.PeriodIndex is already transformed into an array of dtype object.

Before @darothen 's https://github.com/pydata/xarray/pull/1356, the piece if code that was executed is date = ref_var.to_index(), which has the nice property to return a PeriodIndex, as shown here from my debugger:

```python

ref_var Out[1]: <xarray.IndexVariable 'time' (time: 12)> array([Period('2000-01', 'M'), Period('2000-02', 'M'), Period('2000-03', 'M'), Period('2000-04', 'M'), Period('2000-05', 'M'), Period('2000-06', 'M'), Period('2000-07', 'M'), Period('2000-08', 'M'), Period('2000-09', 'M'), Period('2000-10', 'M'), Period('2000-11', 'M'), Period('2000-12', 'M')], dtype=object) ref_var.to_index() Out[3]: PeriodIndex(['2000-01', '2000-02', '2000-03', '2000-04', '2000-05', '2000-06', '2000-07', '2000-08', '2000-09', '2000-10', '2000-11', '2000-12'], dtype='period[M]', name='time', freq='M') ```

Indeed! Will look into it.

Looks like https://github.com/pydata/xarray/pull/1356, which introduced a is_datatime_like() check to the dtype of a variable before adding time components.

I was wondering which PR was responsible for the regression, but I guess I don't understand the internals well enough to classify this as "easy fix" for me ;)

I never intended to support time attributes on PeriodIndex objects so there was never any test coverage here, but I guess it has been working! Nonetheless this should be pretty easy to fix.