What is this about???
Dear reader,
This is an important enhancement of LIM3 to improve its stability in ORCA2 configuration at least. By default, the horizontal gradient of sea surface height is not activated. It should.

How to activate that???

1. Uncomment the two lines starting with zdsshx and zdsshy in limrhg.F90.
   
2. Define ssh_m in sbc_oce.F90:
   

REAL(wp), PUBLIC, DIMENSION(jpi,jpj) :: ssh_m !: mean (nn_fsbc time-step) sea surface height [m]

3. Compute ssh_m in sbc_ssm.F90 (see attached file)

Bur is there any effect ???
Of course, yes!!! It definitely stabilizes the sea ice near the Canadian Archipelago and reduces the chances of model crash.

Merci pour votre collaboration!
Salutations distinguées,
Martin!

Context

The ice initialisation options includes a nn_iceini_file = 2 option which is meant to allow ice initial conditions to be read from a restart file (even when ln_rstart is false) but this option is not working in v4.0.3. The option is not present in the trunk version.

Analysis

The relevant code in icestp.F90 is:

      IF ( ln_rstart .OR. nn_iceini_file == 2 ) THEN 
         CALL ice_rst_read                      ! start from a restart file
      ELSE
         CALL ice_istate_init
         CALL ice_istate( nit000 )              ! start from rest or read a file
      ENDIF

but this won't work because nn_iceini_file isn't set until ice_istate_init, so this needs to be:

      CALL ice_istate_init
      IF ( ln_rstart .OR. nn_iceini_file == 2 ) THEN
         CALL ice_rst_read                      ! start from a restart file
      ELSE
         CALL ice_istate( nit000 )              ! start from rest or read a file
      ENDIF

Recommendation

In 4.0-HEAD, Make the simple change above and there is also an opportunity to tidy code inside icerst.F90 to remove a (now) redundant call to ice_istate_init.

Activating the adaptive EVP for sea-ice rheology (instead of the regular EVP) results in a crash in ORCA2 after some iterations. The lower nn_fsbc is, the sooner the model crashes. For nn_fsbc=3, it occurs after about 6 months simulation. It is way shorter for nn_fsbc=1 and it does not crash for nn_fsbc=5.

I do not know why. Maybe aEVP just reveals an issue with the ice-ocean drags, or maybe there is a problem with the implementation of aEVP. In order to have time to investigate the origin of this behavior, I will set ln_aEVP to false for the time being.