Custom Query (964 matches)

Several bugs were corrected in the 4DVar algorithms when running with multiple outer loops. This included the computation of the cost function. We also corrected the MPI exchanges in the adjoint convolution (ad_convolution.F and ad_variability.F) to follow adjoint rules. This was not really a bug now because the values in the ghost point were zero any way. However, this may not the case in the future.

Many thanks to Andy Moore for reporting and correcting these problems.

The bulk flux (BULK_FLUXES) formulation in ROMS is complicated and highly nonlinear. Although we have the tangent linear, representer, and adjoint versions of bulk_flux.F, we have found out stability issues with the linearization in the adjoint-based algorithms.

In the past, we have run the nonlinear model first and saved the fluxes in the history file to be used as a forcing file in the adjoint-based applications. The problem that we had with this strategy is that we output the surface salinity flux as (E-P)*Ssur (variable ssflux) and the model at input just needs E-P (variable swflux). In order to avoid conflicts, a new variable is introduced EminusP that allows some flexibility. So varinfo.dat has now the following entry:

'EminusP'                                          ! Input/Output
  'bulk_flux surface net freswater flux, (E-P)'
  'meter second-1'                                 ! computed by NLM ROMS
  'EminusP, scalar, series'                        ! bulk_flux.F and
  'ocean_time'                                     ! needed in adjoint-based
  'idEmPf'                                         ! applications
  'r2dvar'
  1.0d0

Also, all the input script were modified to add a new I/O variable switch:

Hout(idEmPf) == F                          ! E-P flux

A new C-preprocessing option NL_BULK_FLUXES was introduced to handle the reading of surface fluxes computed by the nonlinear model (bulk_flux.F) in the representer model. See rp_get_data.F and rp_set_data.F.

This option can be also used in the nonlinear model when BULK_FLUXES is not defined to read such surface salinity flux if needed:

#  if defined NL_BULK_FLUXES && !defined BULK_FLUXES
!
!-----------------------------------------------------------------------
!  Surface net freshwater flux (E-P) from NLM bulk flux computation.
!-----------------------------------------------------------------------
!
      CALL get_2dfld (ng, iRPM, idEmPf, ncFRCid(idEmPf,ng),             &
     &                nFfiles(ng), FRCname(1,ng), update(1),            &
     &                LBi, UBi, LBj, UBj, 2, 1,                         &
#   ifdef MASKING
     &                GRID(ng) % rmask(LBi,LBj),                        &
#   endif
     &                FORCES(ng) % stflxG(LBi,LBj,1,isalt))

#  elif !(defined ANA_SSFLUX || defined EMINUSP || defined SRELAXATION)
!
!-----------------------------------------------------------------------
!  Surface net freshwater flux: E-P.
!-----------------------------------------------------------------------
!
      CALL get_2dfld (ng, iNLM, idsfwf, ncFRCid(idsfwf,ng),             &
     &                nFfiles(ng), FRCname(1,ng), update(1),            &
     &                LBi, UBi, LBj, UBj, 2, 1,                         &
#   ifdef MASKING
     &                GRID(ng) % rmask(LBi,LBj),                        &
#   endif
     &                FORCES(ng) % stflxG(LBi,LBj,1,isalt))
#  endif

Many thanks to Andy Moore for suggesting this capability.

Added a new C-preprocessing option AD_IMPULSE to control the forcing of the adjoint model as an intermittent impulse (on) or a continuous force term (off). Many thanks to Andy Moore for coding this option.

I also added the missing code for the CLIMA_TS_MIX option in the tangent linear, representer, and adjoint models. Recall that this option is used for the horizontal diffusion of tracer perturbation (t-tclm).