Advective transport diagnostics are incorrect when using certain tracer advection schemes

Context

• Branches impacted: main
• Reference configuration/test case (chosen or used as template): ORCA2
• Computing architecture: compiler, MPI & NetCDF libs (name and version)
• Dependencies: since 16443a0b
• Any other relevant information
  • Affected diagnostics
    • sophtadv- advective poleward heat transport
    • sopstadv- advective poleward salt transport
    • uadv_heattr/vadv_heattr- U/V advective heat transport
    • uadv_salttr/vadv_salttr- U/V advective salt transport
  • Affected modules
    • traadv_cen.F90
    • traadv_mus.F90

Analysis

Following #68 (closed) (specifically #80 (closed)), calls to dia_ptr_hst and dia_ar5_hst in tra_adv_cen and tra_adv_mus have been commented out due to refactoring of the 3D arrays (to 2D) passed to these subroutines. This means the related diagnostics are incorrect.

tra_ldf_iso and tra_ldf_lev were refactored in the same way, but are not affected by this issue as a workaround was used (storing the diagnostic result in 3D arrays, then using this in the dia_ptr_hst/dia_ar5_hst calls).

NOTE: A similar issue exists for the TRD diagnostics- this is not resolved here

Fix

Refactor dia_ptr_hst and dia_ar5_hst to allow 2D (i-j) arrays as arguments. When using 2D arrays, the diagnostic must be constructed by calling these subroutines for each level before finalising the diagnostic (calling MPI, XIOS subroutines).

The approach to this finalisation step differs for each subroutine:

• dia_ptr_hst
  • An existing finalisation subroutine, dia_ptr_iom (called by dia_ptr), is already called from traadv.F90 and the main timestepping subroutine
  • The mpp_sum calls in ptr_sum are moved into ptr_mpp_sum, which is called from dia_ptr_iom
• dia_ar5_hst
  • There is no explicit finalisation step, so we must tell the subroutine when the diagnostics are ready to be sent to XIOS
  • The calculations are stored in 3D arrays (with the SAVE attribute)
  • A logical input argument is used to control the iom_put calls

The workarounds in tra_ldf_iso and tra_ldf_lev will be removed.

As a further optimisation, a couple of diaptr.F90 module arrays (hstr_ove, hstr_btr) are replaced by a single local working array.