use time without date as one dimension in xarray

Question

I have an xarray.DataArray that I successfully set one dim as time of one day.

tmp.dims
('ITEM', 'DATE', 'TIME', 'CODE')
tmp.TIME
<xarray.DataArray 'TIME' (TIME: 15)>
array([datetime.time(14, 15), datetime.time(14, 16), datetime.time(14, 17),
       datetime.time(14, 18), datetime.time(14, 19), datetime.time(14, 20),
       datetime.time(14, 21), datetime.time(14, 22), datetime.time(14, 23),
       datetime.time(14, 24), datetime.time(14, 25), datetime.time(14, 26),
       datetime.time(14, 27), datetime.time(14, 28), datetime.time(14, 29)],
      dtype=object)
tmp.TIME.values[0]
datetime.time(14, 15)

but I cannot save this xarray do to the this error:

tmp.to_netcdf('/sdata/user/tsu/tmp/srpd.nc')
        *** ValueError: unable to infer dtype on variable 'TIME'; xarray cannot serialize arbitrary Python objects

Is it wrong to use datetime.time in this way?

Answer 1

Insteads of using lists of python datetime objects, use timedelta arrays from numpy or pandas. See the pandas docs on working with time series data .

For data indexed by hour of day, but not a specific date, I'd recommend using pd.to_timedelta , as in this example:

In [9]: da = xr.DataArray(
   ...:     np.ones(16).reshape(4, 4),
   ...:     dims=['x', 'hour'],
   ...:     coords=[range(4), pd.to_timedelta(range(4), unit='h')],
   ...: )
   ...:

In [10]: da
Out[10]:
<xarray.DataArray (x: 4, hour: 4)>
array([[1., 1., 1., 1.],
       [1., 1., 1., 1.],
       [1., 1., 1., 1.],
       [1., 1., 1., 1.]])
Coordinates:
  * x        (x) int64 0 1 2 3
  * hour     (hour) timedelta64[ns] 00:00:00 01:00:00 02:00:00 03:00:00

In [11]: da.hour
Out[11]:
<xarray.DataArray 'hour' (hour: 4)>
array([             0,  3600000000000,  7200000000000, 10800000000000],
      dtype='timedelta64[ns]')
Coordinates:
  * hour     (hour) timedelta64[ns] 00:00:00 01:00:00 02:00:00 03:00:00

Note that this can be written to netCDF without issue:

In [12]: da.to_dataset(name='myarr').to_netcdf('sample.nc')

In [13]: xr.open_dataset('sample.nc')
Out[13]:
<xarray.Dataset>
Dimensions:  (x: 4, hour: 4)
Coordinates:
  * x        (x) int64 0 1 2 3
  * hour     (hour) timedelta64[ns] 00:00:00 01:00:00 02:00:00 03:00:00
Data variables:
    myarr    (x, hour) float64 ...