datetime64 data type in arrays is always 'datetime64[ns]'

[sjsmith757]

I think this is probably an enhancement request, but there doesn't seem to be any documentation related to this and it doesn't really match the behavior I would expect. From what I can tell, xarray doesn't support datetime arrays with units other than [ns]. For example,

times = np.arange(np.datetime64('2005-01'),np.datetime64('2007-01'),np.timedelta64(1,'M'))
xrtimes = xr.DataArray(times,dims=['time'],coords={'time':times},name='time')

I would expect xrtimes.dtype to return dtype('<M8[M]') as times.dtype does, but instead it returns dtype('<M8[ns]'). Even looking at xrtimes.indexes['time'] to ensure it isn't just a problem with the repr, the dtype is listed as 'datetime64[ns]'. Trying to do xrtimes.astype('datetime64[M]') does nothing. It would be nice if this didn't simply silently fail, or if there were some documentation explaining that datetime64 objects have to be in units of ns.

In an ideal world, it would be nice if xarray supported datetime units other than ns, but I understand with internals that may be more trouble than its worth...

I'm using
xarray=0.19.0
pandas=1.3.1
numpy=1.21.1

[sjsmith757]

Actually I am guessing this is related to the underlying pandas.DatetimeIndex, which I guess only supports units of 'ns'. So perhaps there isn't much that can be done at this time, but it is not ideal to have to go to pandas documentation to see that. It is sort of suggested here, but perhaps it could be made more explicit

[dcherian]

Thanks @sjsmith757 .

Can you send in a PR to improve the documentation on that points?

We have documentation on contributing to the docs here: https://xarray.pydata.org/en/stable/contributing.html#contributing-to-the-documentation

[kmuehlbauer]

This has been treated by #9618. The example returns now dtype('<M8[s]'), the documentation has been updated, too.