Vertical geometry mismatch between field and file during a writefield

[Yves-meteo]

In an Olive experiment I try to take some fields from an historical global Arpege file from an experiment in 32 bit and to write them in an analysis file from the e-suite in 64 bit

The code used :

fic_init_ress = epygram.formats.resource(filename=fic_init, openmode='a')
fic_guess_ress = epygram.formats.resource(filename=fic_guess, openmode='r')
level = 1
level_max = 105
while level <= level_max :
level_str = '{:03d}'.format(level)
field_name = 'S' + level_str + 'MO_ORGANISAT'
print(field_name)
field = fic_guess_ress.readfield(field_name)
fic_init_ress.writefield(field)
level+=1

fic_init_ress.close()
fic_guess_ress.close()

The analysis file can be found on hendrix here : /home/m/mxpt/mxpt001/vortex/arpege/4dvarfr/DBLE/2025/08/01/T0000P/4dupd2/analysis.atm-arpege.tl1798-c22.fa
The guess file can be found here : /home/boutelou/vortex/arpege/4dvarfr/H5PG/20250731T1800P/forecast/historic.arpege.tl1798-c22+0006:00.fa

The error message :

====================================================================================================
= HANDLING EXCEPTION =

Exception type: <class 'AssertionError'>
Exception info: vertical geometry mismatch between field and file.

---

Vortex v-1.12.1: Traceback Error / BEGIN

None
File "/var/spool/slurmd/job6055043/slurm_script", line 956, in
fic_init_ress.writefield(field)

File "/home/mf/dp/marp/verolive/public/mtool-2.3.5/public/epygram-default_py3.7/epygram/formats/FA.py", line 947, in writefield
"vertical geometry mismatch between field and file."

Regards
Yves

[SebastienRietteMTO]

Vertical geometries are actually different because A and B values are different.
When writing a field in an already existing file, we allow a small difference between A and B values but, in your case, the difference is way larger than the threshold (0.001 against 2E-16).

I don't know how to modify epygram to allow what you are trying to do while keeping a relevant safety control. So I only propose two possible workarounds:

• enhance the value of the threshold with epygram.config.epsilon = 0.002 in your script
• update the field geometry with field.geometry.vcoordinate.grid = fic_init_ress.geometry.vcoordinate.grid

@AlexandreMary Would you have a better idea?

[AlexandreMary]

Indeed I think we have to workaround these kinds of issues at the moment.
The underlying issue (which is not an epygram one) is that the FA format should be precision-agnostic so that files written from a 32 or 64-bit forecast end up with the same meta-data in the FA frame.

[SebastienRietteMTO]

Yes but in IAL, VAH and VBH defined in yomvert.F90 are typed as JPRB, the precision is lost before writing to the FA file. Or maybe we could always write 32 bit values in the FA frame?