time axis challenges: xcdat -> numpy

[gleckler1]

I am testing the use of xcdat to read in a large dataset via xcdat.open_mfdataset. My goal is to re-write this data using CMOR, but to do that I need to isolate the time axis as a numpy array with index values (with cdms this is rel time instead of component time). I am struggling to find a reasonable/simple path. Any hints?

[pochedls]

Does this work: time = ds.time.values?

[gleckler1]

type(time)
<class 'xarray.core.dataarray.DataArray'>
t4 = time.time.values
type(t4)
<class 'numpy.ndarray'>
t4[0:4]
array([cftime.DatetimeGregorian(1982, 1, 1, 0, 0, 0, 0, has_year_zero=False),
cftime.DatetimeGregorian(1982, 1, 2, 0, 0, 0, 0, has_year_zero=False),
cftime.DatetimeGregorian(1982, 1, 3, 0, 0, 0, 0, has_year_zero=False),
cftime.DatetimeGregorian(1982, 1, 4, 0, 0, 0, 0, has_year_zero=False)],
dtype=object)

SP, Thanks for the rapid response! I am trying to translate the cftime.DatetimeGregorian values to an integer (days since datum).

[pochedls]

@gleckler1 – I think you have two options:

1. Open the dataset with ds = xc.open_mfdataset(fn, decode_times=False) or

2. try to re-encode the time axis after the fact:

from xarray.coding.times import encode_cf_datetime

units = ds.time.encoding['units']
calendar = ds.time.encoding['calendar']
time_decoded = encode_cf_datetime(ds.time, units, calendar)
print(time_decoded)

(array([711399.5, 711429. , 711458.5, 711489. , 711519.5, 711550. , ...
734940.5, 734971.5, 735002. , 735032.5, 735063. , 735093.5]), 'days since 0001-01-01', 'noleap')

[gleckler1]

@pochedls That works great! Now I need to do the same with the time_bnds - not obvious how to extend the solution. I'll try striping out the lower bounds into one array and the upper into another, apply encode to both, and then reconstruct as tuple numpy.array. Many thanks!

[tomvothecoder]

Thanks for providing a solution Steve. I'll mark this as the answer.

@pochedls That works great! Now I need to do the same with the time_bnds - not obvious how to extend the solution. I'll try striping out the lower bounds into one array and the upper into another, apply encode to both, and then reconstruct as tuple numpy.array. Many thanks!

@gleckler1 Solution 1 will leave time and time bounds encoded, but you won't be able to perform temporal operations in xCDAT unless they are decoded. In Solution 2, encode_cf_datetime() seems to only operate on a 1-d array so your workaround sounds like the way to go.

One thing to note is that encode_cf_datetime() probably won't work with non-CF compliant time units ("months since ..." or "years since ...").

Given an array of datetime objects, returns the tuple (num, units, calendar) suitable for a CF compliant time variable.
Source - xarray.coding.times.encode_cf_datetime docstring