GL7 suite

[rfu]

Hi Patrick
Can you point me to the GL7 WFDEI Rose-Cylc suite? I would like to upgrade from GL6R.
Becky

[pmcguire]

Hi Becky
The suite for the Global Land 7 (GL7) configuration for JULES with WFDEI driving data and 0.5 degree ancillaries is u-bk784. The GL6R suite (the 'R' stands for 'Reading', and it is an informal, unofficial name) that may be similar to the one you have been using is u-as052.

In the u-bk784 suite, I added the WFDEI and the WFDEI ancillaries that we used in GL6R to the original u-bb316 GL7 suite which uses CRU-NCEP driving data and N96 ancillaries (originally from Carolina Duran Rojas and Andy Wiltshire). I had also turned off the time-varying CO2 when going from u-bb316→u-bk784, in order to be consistent with u-as052. Perhaps the time-varying CO2 can be enabled again.

Maybe Omar Müller can provide you with his report looking at the differences in the namelists between the GL6R & GL7 WFDEI suites. I can also point you to an ILAMB comparison of the output of these suites, if you'd like.
Patrick

[rfu]

Hi Patrick,

I am trying to change the GL7 suite to run with 9PFTs (which also means changing some of the ancillary files so they are compatible with 9PFTs). However, I am running into problems, the suite completes initialisation but then it crashes, giving very little information, just a segmentation fault, so I am struggling to work out why it is crashing.

The suite is u-bq898.

I took the following steps to upgrade the suite from u-bk784:
1) I linked to Doug's branch of JULES-vn5.6 - the suite ran ok.
2) I moved to 9PFT and corresponding ancillaries - the suite fails …however
3)…. if I take a bunch of namelists and run it on a linux machine it runs ok, so I think my ancillaries are ok and the namelist settings for 9PFT.

Do you have any ideas?

Thanks

Becky

[pmcguire]

Hi Becky
Can you changes the permissions of your home directory, so that I can look at your roses and cylc-run subdirectories?
Thanks
Patrick McGuire

[rfu]

Hi Patrick,

How do I do change permissions?

Thanks,

Becky

[pmcguire]

Hi Becky
You should make your permissions on jasmin to be drwxr-xr-- and if you use the ls command, the permissions should be visible like this:

[pmcguire@jasmin-sci1 roses]$ ls -ltrd ~
drwxr-xr-- 1 pmcguire users 0 Mar  3 16:59 /home/users/pmcguire

They currently look like this:

[pmcguire@jasmin-sci1 roses]$ ls -ltrd ~rfuller01
drwx------ 1 rfuller01 users 0 Mar  3 16:06 /home/users/rfuller01

You can use the chmod command to change the permissions. I can help further with that, if you'd like.
Patrick McGuire

[rfu]

Hi Patrick,

Yes please, just to be sure I set them correctly and don't end up doing something stupid to my account!

Thanks,

Becky

[pmcguire]

Hi Becky
I think the proper command is:

chmod 754 /home/users/rfuller01

This will give read permission to all users, and read and execute permission to only people in the unix group that you belong to.
And it will only leave the write permission for yourself.

You might want to check that any subdirectories of your home directory that you might want to have as not readable by others have this command applied to them:

chmod 700 /home/users/rfuller01/SUBDIR1

Patrick McGuire

[rfu]

Thanks for that, I have changed the permissions now, so hopefully you can see my roses directory. The suite is u-bq898-Std.

Cheers,

Becky

[pmcguire]

Thanks for changing your permissions.

After a quick look, it looks like your RECON app in your u-bq898-Std suite is failing:
you can see that in:

~rfuller01/cylc-run/u-bq898-Std/log/job/19790101T0000Z/RECON/01/job.err

And in your 5-PFT suite, the RECON app also fails:

~rfuller01/cylc-run/u-bq898-test-5PFT/log/job/19790101T0000Z/RECON/01/job.err

with this error:

{MPI Task 7} [FATAL ERROR] init_ic: Land ice points and soil points are mutually exclusive

Are these the right suites to be comparing? I thought you said you had your 5-PFT version working.
Patrick McGuire

[rfu]

Hi Patrick,

Thanks for looking. The u-bq898-Std suite fails in the RECON step, it completes initialisation, makes the first dump file and then falls over when it starts to run.

The 5PFT version I had running successfully was an earlier revision of the u-bq898-Std suite. The u-bq898-test-5PFT suite is just one I have been playing around with, so ignore that one. It currently fails because it has a different soil file in there, so the land ice points are incompatible with the fractional cover file.

My feeling is that there's something wrong with my ancillary files, but I can' t see what. Or at least something that the intel compiler on JASMIN doesn't like, but the gfortran compiler on our linux boxes is happy with as I can run from a bunch of namelists on our linux boxes.

Cheers,

Becky

Thanks,

Becky

[pmcguire]

Hi Becky
So which suite was the one that was working before you modified it to make the u-bq898-Std suite that fails in RECON?
Do you have the log files and set-up of that working suite?
Patrick

[pmcguire]

Hi Becky:
I copied your u-bq898-test-5PFT suite and reverted from your 3-class soils to the original GL6R version of the soils, and I get a similar error now to the one that you got in your u-bq898-Std suite. The error that I get is in:
~pmcguire/cylc-run/u-bq898-test-5PFT_B1/log/job/19790101T0000Z/RECON/01/job.err:

forrtl: severe (174): SIGSEGV, segmentation fault occurred
Image              PC                Routine            Line        Source             
jules.exe          000000000051D087  model_interface_m       12551  model_interface_mod.F90
jules.exe          00000000004A048D  time_varying_inpu         954  time_varying_input_mod.F90
jules.exe          0000000000409CB5  MAIN__                    106  jules.F90

This refers to this file:
~pmcguire/cylc-run/u-bq898-test-5PFT_B1/share/fcm_make/preprocess/src/jules/src/io/model_interface/model_interface_mod.F90
at these lines:

IF ( PRESENT(cube) ) THEN
    cube_land = map_to_land(cube)
    CALL cube_get_data(cube_land, co2_tmp)

    IF ((MAXVAL(co2_tmp) /= co2_tmp(1)) .OR.                                  &.  <=== the line 12551 with the failure point is here
        (MINVAL(co2_tmp) /= co2_tmp(1))) THEN

This is part of the CALL populate_var(files(i)%fields(j)%var_id, interp_data) in time_varying_input_mod.F90, which is called by CALL update_prescribed_variables() in jules.F90.

I looked at your prescribed co2 data file that you added, and it has the same values globally for each time step, so I think it should work fine. But maybe this part of the JULES code is compiler dependent and hasn't been thoroughly exercised and needs to be debugged and fixed.

I have put in some print statements in another version of the suite ~pmcguire/cylc-run/u-bq898-test-5PFT_B3,
surrounding the 'CALL cube_get_data(cube_land, co2_tmp)' call above, and after re-running the code, it appears that the failure is after the 7th or 8th time that this routine is called.
Patrick McGuire

[pmcguire]

Hi Becky:
Do you need any more help with this? Did you get it working?
What did you do to fix it?
Patrick

[rfu]

Hi Patrick,
Thanks for checking, yes I did get it working. There was a problem with some settings in the initial conditions that related to the snow scheme, that's why the model was crashing.
Thanks,
Becky

[pmcguire]

Hi Becky:
Thanks for the feedback. I wonder why the prescribed co2 data was interacting with the initial conditions for the snow scheme. Or maybe there wasn't this apparent interaction. I am glad it is working now.
I will close the ticket.
Patrick