error in writing snowage with ORCHIDEE trunk revision 6519

[stafasca]

When running ORCHIDEE offline with GSWP3_v1 on jean-zay, I have an error in writing the variable snowage, on the year 1963 (repeated error for several tests starting on 1960, with 32 and 64 procs, and either one or two procs for XIOS).

Error message:

Error [CNc4DataOutput::writeFieldData_ (CField* field)] : In file '/gpfswork/rech/lmd/rcps009/ORCHIDEE_10022020/modipsl/modeles/XIOS/src/io/nc4_data_output.cpp', line 2676 -> On writing field data: snowage

In the context : orchidee_server
Error when calling function ncPutVaraType(ncid, varId, start, count, data)
NetCDF: Numeric conversion not representable
Unable to write data given the location id: 65536 and the variable whose id: 83 and name: snowage
Abort(-1) on node 63 (rank 63 in comm 0): application called MPI_Abort(MPI_COMM_WORLD, -1) - process 63
srun: Job step aborted: Waiting up to 62 seconds for job step to finish.
slurmstepd: error: * STEP 773956.0 ON r1i1n8 CANCELLED AT 2020-02-10T12:51:07 *

I got the simulations running well when I removed this variable from the output file.

[jgipsl]

The error is due to negative snowfall in the forcing file.

The forcing file IGCM/SRF/METEO/GSWP3/v1_halfdeg/GSWP3_v1_halfdeg_3h_1963.nc contains negative values for SNOWF for L=720 (1st of April).

In explicitsnow_main, following section gives then a floating overflow:

IF (snow(ji) .LE. zero) THEN
     snow_age(ji) = zero
ELSE
     snow_age(ji) =(snow_age(ji) + (un - snow_age(ji)/max_snow_age) * dt_sechiba/one_day) &
          & * EXP(-precip_snow(ji) / snow_trans)
ENDIF

when ji=744, on proc=25 using 63MPI in total (lon, lat=??)
snow(ji)=0.583010534761298
precip_snow(ji)=-268.820589780807
snow_trans=0.2 (parameter)

=> EXP(-precip_snow(ji) / snow_trans) explodes   
=> snow_age explodes  => XIOS fail to write the variable snow_age

[jgipsl]

I added following in the README file at shared repository:
IGCM/SRF/METEO/GSWP3/README

ERRATA: The file GSWP3/v1_halfdeg/GSWP3_v1_halfdeg_3h_1963.nc contains negative values for SNOWF for L=720 (1st of April).
This has not been corrected and might make the model crash. See ticket https://forge.ipsl.jussieu.fr/orchidee/ticket/672

Is this enough and the ticket can be closed?

[jgipsl]

This has also been commented on the wiki:
​https://forge.ipsl.jussieu.fr/orchidee/wiki/Documentation/Forcings?action=diff&version=128&old_version=127

[mmcgrath]

We are going to do some additional analysis of forcings to see where unphysical values of forcing exist and where they need to be corrected (all variables, not just snowfall).