MOM6 time

[hkershaw-brown]

The default in CESM is to run with no leap year.
For assimilating with real obs, we need to use the actual time.

Todos:

• DART MOM6 documentation, be explicit about the need to set the calendar in CESM.
  ./xmlchange CALENDAR=GREGORIAN

Note on what happens if you don't set CALENDAR=GREGORIAN
mom6 restart file has time in days from year 1.
Starting the run at 2015-02-01:

./xmlchange RUN_STARTDATE=2015-02-01

double Time(Time) ;
                Time:long_name = "Time" ;
                Time:units = "days" ;
                Time:axis = "T" ;
...
// global attributes:
		:filename = "./c.T62_g16.Alper.ens3.mom6.r.2015-02-11-00000._0001.nc" ;
data:

 Time = 735151 ;
}

If you don't have leap years switched on you end up with inconstant time information like this:
Restart filename has the time 2015-02-11-00000.
The Time variable is Time = 735151, which is 2013/10/11

hkershaw@cisl-fisher utilities (main) $ git diff gregorian_time.c 
diff --git a/assimilation_code/modules/utilities/gregorian_time.c b/assimilation_code/modules/utilities/gregorian_time.c
index ee6a72f03..e8fccb87d 100644
--- a/assimilation_code/modules/utilities/gregorian_time.c
+++ b/assimilation_code/modules/utilities/gregorian_time.c
@@ -20,7 +20,7 @@
 #include <stdlib.h>
 
 /* the start of gregorian numbered days */
-#define BASE_YEAR 1601
+#define BASE_YEAR 1
 
 enum todo { TO_YMD, TO_DS };
 enum err  { ERROR, OK };
hkershaw@cisl-fisher utilities (main) $ icc gregorian_time.c ; ./a.out  735151 0
735151     0 == 2013/10/11 00:00:00

• Convert MOM6 time to DART time for read_model_time
read_model_time is used for the move_ahead calculation. MOM6 Model time is days from year 1. This needs to be DART time (days from 1601) since that is what the observation times are.

DART/assimilation_code/modules/assimilation/obs_model_mod.f90

Lines 57 to 65 in 502af78

	subroutine move_ahead(ens_handle, ens_size, seq, last_key_used, window_time, &
	key_bounds, num_obs_in_set, curr_ens_time, next_ens_time)
	! Based on the current ens time and the time of the next available
	! observation, compute whether and how far the ensemble copies of the
	! state need to be advanced. This returns the number of obs in the
	! next assimilation window, and both the current and expected next data
	! times. It NO LONGER advances the model - the calling code must call
	! advance_state() itself in order to do that if next time is /= curr time.

Note the model_time for the whole ensemble is from the first ensemble member only:

DART/assimilation_code/modules/io/state_vector_io_mod.f90

Lines 288 to 292 in 502af78

	! read time from input file if time not set in namelist
	!>@todo Check time constistency across files? This is assuming they are consistent.
	if(use_time_from_file) then
	model_time = read_model_time(get_restart_filename(restart_files, 1, 1)) ! Any of the restarts?
	endif

• write_model_time, is used for dart created files: inflation, mean. Should this be MOM6 time (days from year 1) to be consistent with the model restarts? Or DART time (days from 1601) to be consistent with the observations?

[hkershaw-brown]

Conclusion from standup July 13th 2023: write_model_time should be dart time for files created by dart.

Existing model netcdf files do not have their time overwritten by dart.

Note there is a overwrite_time_in_output_file = .false. in direct_netcdf_mod.f90 but it is hardcoded to false with this comment:

DART/assimilation_code/modules/io/direct_netcdf_mod.f90

Lines 137 to 151 in 1b76f3a

	!>@todo FIXME:
	!> this should be in a namelist somewhere and passed in from
	!> a higher level routine.
	!>
	!> we had this in the ROMS branch but no one can remember why.
	!> we think it's because we copied a template file into place
	!> for filter to overwrite and it might not have had the right
	!> analysis time in the file. it's true that filter doesn't
	!> change the time in the file, but when we're doing direct
	!> updates of a netcdf file it's also true that we rarely
	!> update the file we read from; in case of an error you've
	!> destroyed your input needed to rerun the job.
	!> (but we don't remember for sure if this was the reason.)
	logical :: overwrite_time_in_output_file = .false.