bug: Temporal subset error OMOCLO 004 UAT

[ocsmit]

Relevant autotester PR: podaac/l2ss-py-autotest#3920
Version: 3.2.0rc4

Explanation:

The encoding of the time variable causes an error when trying to write out the subsetted netcdf.
This error occurs in 3.2.0rc4 at subset.py#L470.

Specifically the value of encoding['/geolocation']['time'] will be {'_FillValue': None, 'units': 's since 1993-01-01T00:00:00Z', 'dtype': dtype('float64')}. Xarray appears to expect the time units to be in an netcdf compliant format and attempts to find its numpy unit string via a look up table (see times.py#L133). So it is expecting 'seconds' and not 's'.

Since CF compliant unit strings follow a consistent format, a quick fix i put together off the dome and tested is when constructing the time_encoding dictionary (initialized @ subset.py#L326) checking if the unit is a a numpy datetime unit, and replacing it via a lookup table if it is.

from typing import get_args
from xarray.core.types import NPDatetimeUnitOptions

# taken from xarray.coding.times, could just import but it is a private 
# method to the times module
def _numpy_to_netcdf_timeunit(units: NPDatetimeUnitOptions) -> str:
    return {
        "ns": "nanoseconds",
        "us": "microseconds",
        "ms": "milliseconds",
        "s": "seconds",
        "m": "minutes",
        "h": "hours",
        "D": "days",
    }[units]


def check_time_units(unit_str: str) -> str:
    unit_str_list = unit_str.split(" ")

    unit = unit_str_list[0]
    if unit in get_args(NPDatetimeUnitOptions):
        unit_str_list[0] = _numpy_to_netcdf_timeunit(unit)
    return " ".join(unit_str_list)

And then replacing

l2ss-py/podaac/subsetter/subset.py

Line 356 in 6d63fd1

time_encoding[group_path][var_name]['units'] = units

time_encoding[group_path][var_name]["units"] = check_time_units(units)

[ocsmit]

With version 3.3.0a1, this error no longer occurs since the logic of assigning the time encoding items was changed from the following in 3.2.0

l2ss-py/podaac/subsetter/subset.py

Lines 1292 to 1296 in 0fb158e

	if calendar:
	time_encoding[group_path][var_name]['calendar'] = calendar
	if units:
	time_encoding[group_path][var_name]['units'] = units
	time_encoding[group_path][var_name]['dtype'] = dtype

to the current code in 3.3.0a1

l2ss-py/podaac/subsetter/subset.py

Lines 354 to 356 in 1d0f359

	if calendar and units:
	time_encoding[group_path][var_name]['calendar'] = calendar
	time_encoding[group_path][var_name]['units'] = units

For the OMOCLO data there is no calendar provided, only the reference time string in the units, so the units is never populated in the time_encoding dict, and therefore the bug never occurs even though the root of the error still exists.
Looking over the CF-Standards section on time, I do not believe that when the units of the time variable is encoded as a reference time, a calendar needs to be provided. Regardless at first glance the time variable in the data produced by 3.3.0a1 appears to be exactly the same as 3.2.0rc4 (with the above fix).