`sflux` time shifts if it's not hour `00` based

[SorooshMani-NOAA]

While working on setting up sflux ensemble based on HAFS and GEFS I realized that if the sflux "input" files (e.g. what I generate from HAFS or GEFS) then pyschism incorrectly shifts the time vector:

pyschism/pyschism/forcing/nws/nws2/sflux.py

Lines 331 to 338 in 00b1137

	dst['time'].base_date = (
	nc_start_date.year,
	nc_start_date.month,
	nc_start_date.day,
	0)
	dst['time'][:] = [
	(localize_datetime(x) - nc_start_date) / timedelta(days=1)
	for x in variable.datetime_array]

This is because the schism sflux time hour 00 based, but the vector differences for sflux_*.nc file are calculated based on the actual input .nc file's base time. So for example if I have the following time vector in my custom input nc file:

array(['2023-08-14T18:00:00.000000000', '2023-08-15T00:00:00.000000000',
       '2023-08-15T06:00:00.000000000', '2023-08-15T12:00:00.000000000',
       '2023-08-15T18:00:00.000000000', '2023-08-16T00:00:00.000000000',
       '2023-08-16T06:00:00.000000000', '2023-08-16T12:00:00.000000000',
       '2023-08-16T18:00:00.000000000', '2023-08-17T00:00:00.000000000',
       '2023-08-17T06:00:00.000000000', '2023-08-17T12:00:00.000000000',
       '2023-08-17T18:00:00.000000000', '2023-08-18T00:00:00.000000000',
       '2023-08-18T06:00:00.000000000', '2023-08-18T12:00:00.000000000',
       '2023-08-18T18:00:00.000000000', '2023-08-19T00:00:00.000000000',
       '2023-08-19T06:00:00.000000000', '2023-08-19T12:00:00.000000000',
       '2023-08-19T18:00:00.000000000', '2023-08-20T00:00:00.000000000',
       '2023-08-20T06:00:00.000000000', '2023-08-20T12:00:00.000000000',
       '2023-08-20T18:00:00.000000000', '2023-08-21T00:00:00.000000000',
       '2023-08-21T06:00:00.000000000', '2023-08-21T12:00:00.000000000',
       '2023-08-21T18:00:00.000000000', '2023-08-22T00:00:00.000000000',
       '2023-08-22T06:00:00.000000000', '2023-08-22T12:00:00.000000000',
       '2023-08-22T18:00:00.000000000', '2023-08-23T00:00:00.000000000',
       '2023-08-23T06:00:00.000000000', '2023-08-23T12:00:00.000000000',
       '2023-08-23T18:00:00.000000000', '2023-08-24T00:00:00.000000000',
       '2023-08-24T06:00:00.000000000'], dtype='datetime64[ns]')

then I get the following time in the final sflux file sflux_air_1.0001.nc written by pyschism:

array(['2023-08-14T00:00:00.000000000', '2023-08-14T06:00:00.000000000',
       '2023-08-14T12:00:00.000000000', '2023-08-14T18:00:00.000000000',
       '2023-08-15T00:00:00.000000000', '2023-08-15T06:00:00.000000000',
       '2023-08-15T12:00:00.000000000', '2023-08-15T18:00:00.000000000',
       '2023-08-16T00:00:00.000000000', '2023-08-16T06:00:00.000000000',
       '2023-08-16T12:00:00.000000000', '2023-08-16T18:00:00.000000000',
       '2023-08-17T00:00:00.000000000', '2023-08-17T06:00:00.000000000',
       '2023-08-17T12:00:00.000000000', '2023-08-17T18:00:00.000000000',
       '2023-08-18T00:00:00.000000000', '2023-08-18T06:00:00.000000000',
       '2023-08-18T12:00:00.000000000', '2023-08-18T18:00:00.000000000',
       '2023-08-19T00:00:00.000000000', '2023-08-19T06:00:00.000000000',
       '2023-08-19T12:00:00.000000000', '2023-08-19T18:00:00.000000000',
       '2023-08-20T00:00:00.000000000', '2023-08-20T06:00:00.000000000',
       '2023-08-20T12:00:00.000000000', '2023-08-20T18:00:00.000000000',
       '2023-08-21T00:00:00.000000000', '2023-08-21T06:00:00.000000000',
       '2023-08-21T12:00:00.000000000', '2023-08-21T18:00:00.000000000',
       '2023-08-22T00:00:00.000000000', '2023-08-22T06:00:00.000000000',
       '2023-08-22T12:00:00.000000000', '2023-08-22T18:00:00.000000000',
       '2023-08-23T00:00:00.000000000', '2023-08-23T06:00:00.000000000',
       '2023-08-23T12:00:00.000000000'], dtype='datetime64[ns]')

I know that I'm using custom input .nc file as SfluxDataset, but still I'm not sure if this is intentional (i.e. I need to make sure my input .nc file dates are 00 based) or a bug!

Note that I call this function to read the custom .nc file I generate from HAFS or GEFS:

pyschism/pyschism/forcing/nws/nws2/nws2.py

Lines 35 to 41 in 00b1137

	def read(cls, path, sflux_1_glob="*_1.*", sflux_2_glob="*_2.*"):
	path = pathlib.Path(path)
	sflux_2 = list(path.glob(sflux_2_glob))
	return cls(
	sflux_1=SfluxDataset(list(path.glob(sflux_1_glob))),
	sflux_2=SfluxDataset(sflux_2) if len(sflux_2) > 0 else None,
	)

Is this not meant to be used on a generic .nc file? My .nc file has all the required variables, it just doesn't have hour 00 based time.

[josephzhang8]

It's a potential trap for users for sure and I've been thinking about removing the 'hour' in base_date, because it's actually not read in; see:

https://schism-dev.github.io/schism/master/input-output/sflux.html

Instead the hour offset is specified in param.nml (utc_start).