Convert decimal years to datetime64 data

[mgorfer]

Dear everyone,

I am trying to use the gridded Monthly Land + Ocean dataset from Berkeley earth.
http://berkeleyearth.org/data/

The time steps are saved as decimal years. I have tried various approaches to convert these to datetime64 data, but I have no success so far. Is there an implemented version which could perform this conversion?

These are the time steps I would like to convert.

<xarray.DataArray 'time' (time: 2070)>
array([1850.041667, 1850.125   , 1850.208333, ..., 2022.291667, 2022.375   ,
       2022.458333])
Coordinates:
  * time     (time) float64 1.85e+03 1.85e+03 1.85e+03 ... 2.022e+03 2.022e+03
Attributes:
    units:          year A.D.
    standard_name:  time
    long_name:      Time
    calendar:       gregorian

Best regards
Max

[andersy005]

@mgorfer,

it's my understanding that the time units units: year A.D. don't follow CF conventions. Consequently, xarray doesn't know how to decode these values out of the box. One solution is to convert your decimal values into datetime objects. The following code snippet is adapted from this Stackoverflow post

In [30]: def decimal_to_datetime(value):
    ...:     year = int(value)
    ...:     remainder = value - year
    ...:     base = datetime.datetime(year , 1, 1)
    ...:     result = base + datetime.timedelta(seconds=(base.replace(year=base.year + 1) - base).total_seconds() * remainder)
    ...:     return result
    ...: 

In [31]: decimal_to_datetime(1850.041667)
Out[31]: datetime.datetime(1850, 1, 16, 5, 0, 10, 511999)

In [32]: times
Out[32]: 
<xarray.DataArray (time: 6)>
array([1850.041667, 1850.125   , 1850.208333, 2022.291667, 2022.375   ,
       2022.458333])
Dimensions without coordinates: time

In [33]: times.data
Out[33]: 
array([1850.041667, 1850.125   , 1850.208333, 2022.291667, 2022.375   ,
       2022.458333])

In [34]: dates = [decimal_to_datetime(value) for value in times.data]

In [35]: dates
Out[35]: 
[datetime.datetime(1850, 1, 16, 5, 0, 10, 511999),
 datetime.datetime(1850, 2, 15, 15, 0),
 datetime.datetime(1850, 3, 18, 0, 59, 49, 488001),
 datetime.datetime(2022, 4, 17, 11, 0, 10, 511999),
 datetime.datetime(2022, 5, 17, 21, 0),
 datetime.datetime(2022, 6, 17, 6, 59, 49, 488001)]

In [36]: times.data = dates

In [37]: times
Out[37]: 
<xarray.DataArray (time: 6)>
array(['1850-01-16T05:00:10.511999000', '1850-02-15T15:00:00.000000000',
       '1850-03-18T00:59:49.488001000', '2022-04-17T11:00:10.511999000',
       '2022-05-17T21:00:00.000000000', '2022-06-17T06:59:49.488001000'],
      dtype='datetime64[ns]')
Dimensions without coordinates: time

would this suffice for your use case?

[mgorfer]

Dear @andersy005, thank you very much for your fast reply!
Yes, that solution is totally sufficient for my use case.

Do you know if there is any time unit that can be used for automatic conversion?

[andersy005]

You bet!

Do you know if there is any time unit that can be used for automatic conversion?

As long as the time units follow the CF-conventions¹, xarray automatically tries to decode the time coordinate via cftime or pandas when using xr.open_dataset() function.

Footnotes

1. http://cfconventions.org/cf-conventions/cf-conventions#time-coordinate ↩