MODULE propagator_mod ! !git $Id$ !svn $Id: propagator_afte.h 1151 2023-02-09 03:08:53Z arango $ !================================================== Hernan G. Arango === ! Copyright (c) 2002-2023 The ROMS/TOMS Group Andrew M. Moore ! ! Licensed under a MIT/X style license ! ! See License_ROMS.md ! !======================================================================= ! ! ! Adjoint Finite Time Eigenvalues Propagator: ! ! ! ! This routine is used during the computation of the eigenvectors of ! ! the adjoint propagator, transpose[R(t,0)]. They are computed in an ! ! analogous way to those of R(t,0). A single integration of an ! ! arbitrary perturbation state vector "u" backward in time over the ! ! interval [t,0] by the adjoint model: transpose[R(t,0)]*u. ! ! ! ! Reference: ! ! ! ! Moore, A.M. et al., 2004: A comprehensive ocean prediction and ! ! analysis system based on the tangent linear and adjoint of a ! ! regional ocean model, Ocean Modelling, 7, 227-258. ! ! ! !======================================================================= ! USE mod_kinds ! implicit none ! PRIVATE PUBLIC :: propagator_afte ! CONTAINS ! !*********************************************************************** SUBROUTINE propagator_afte (RunInterval, state, ad_state) !*********************************************************************** ! USE mod_param USE mod_parallel #ifdef SOLVE3D USE mod_coupling #endif USE mod_iounits USE mod_ocean USE mod_scalars USE mod_stepping ! USE close_io_mod, ONLY : close_inp USE dotproduct_mod, ONLY : ad_statenorm USE packing_mod, ONLY : ad_unpack, ad_pack #ifdef SOLVE3D USE set_depth_mod, ONLY : set_depth #endif USE strings_mod, ONLY : FoundError ! ! Imported variable declarations. ! real(dp), intent(in) :: RunInterval ! TYPE (T_GST), intent(in) :: state(Ngrids) TYPE (T_GST), intent(inout) :: ad_state(Ngrids) ! ! Local variable declarations. ! #ifdef SOLVE3D logical :: FirstPass = .TRUE. #endif ! integer :: ng, tile ! real(r8) :: StateNorm(Ngrids) ! character (len=*), parameter :: MyFile = & & __FILE__ ! !======================================================================= ! Forward integration of the tangent linear model. !======================================================================= ! Nrun=Nrun+1 IF (Master) THEN DO ng=1,Ngrids WRITE (stdout,10) ' PROPAGATOR - Grid: ', ng, & & ', Iteration: ', Nrun, & & ', number converged RITZ values: ', & & Nconv(ng) END DO END IF ! ! Initialize time stepping indices and counters. ! DO ng=1,Ngrids iif(ng)=1 indx1(ng)=1 kstp(ng)=1 krhs(ng)=3 knew(ng)=2 PREDICTOR_2D_STEP(ng)=.FALSE. ! iic(ng)=0 nstp(ng)=1 nrhs(ng)=1 nnew(ng)=2 ! synchro_flag(ng)=.TRUE. tdays(ng)=dstart+dt(ng)*REAL(ntimes(ng),r8)*sec2day time(ng)=tdays(ng)*day2sec ntstart(ng)=ntimes(ng)+1 ntend(ng)=1 ntfirst(ng)=ntend(ng) END DO ! !----------------------------------------------------------------------- ! Clear adjoint state variables. There is not need to clean the basic ! state arrays since they were zeroth out at initialization and bottom ! of previous iteration. !----------------------------------------------------------------------- ! DO ng=1,Ngrids DO tile=first_tile(ng),last_tile(ng),+1 CALL initialize_ocean (ng, tile, iADM) END DO END DO #ifdef SOLVE3D ! !----------------------------------------------------------------------- ! Compute basic state initial level thicknesses used for state norm ! scaling. It uses zero time averaged free-surface (rest state). ! Therefore, the norm scaling is time invariant. !----------------------------------------------------------------------- ! DO ng=1,Ngrids DO tile=last_tile(ng),first_tile(ng),-1 CALL set_depth (ng, tile, iADM) END DO END DO #endif ! !----------------------------------------------------------------------- ! Unpack adjoint initial conditions from state vector. !----------------------------------------------------------------------- ! DO ng=1,Ngrids DO tile=first_tile(ng),last_tile(ng),+1 CALL ad_unpack (ng, tile, Nstr(ng), Nend(ng), & & state(ng)%vector) END DO END DO ! !----------------------------------------------------------------------- ! Compute initial adjoint state dot product norm. !----------------------------------------------------------------------- ! DO ng=1,Ngrids DO tile=last_tile(ng),first_tile(ng),-1 CALL ad_statenorm (ng, tile, knew(ng), nstp(ng), & & StateNorm(ng)) END DO IF (Master) THEN WRITE (stdout,20) ' PROPAGATOR - Grid: ', ng, & & ', Adjoint Initial Norm: ', StateNorm(ng) END IF END DO ! !----------------------------------------------------------------------- ! Read in initial forcing, climatology and assimilation data from ! input NetCDF files. It loads the first relevant data record for ! the time-interpolation between snapshots. !----------------------------------------------------------------------- ! DO ng=1,Ngrids CALL close_inp (ng, iADM) IF (FoundError(exit_flag, NoError, __LINE__, MyFile)) RETURN CALL ad_get_idata (ng) IF (FoundError(exit_flag, NoError, __LINE__, MyFile)) RETURN CALL ad_get_data (ng) IF (FoundError(exit_flag, NoError, __LINE__, MyFile)) RETURN END DO ! !----------------------------------------------------------------------- ! Time-step the adjoint model backwards. !----------------------------------------------------------------------- ! DO ng=1,Ngrids IF (Master) THEN WRITE (stdout,30) 'AD', ng, ntstart(ng), ntend(ng) END IF time(ng)=time(ng)+dt(ng) iic(ng)=ntstart(ng)+1 END DO #ifdef SOLVE3D CALL ad_main3d (RunInterval) #else CALL ad_main2d (RunInterval) #endif IF (FoundError(exit_flag, NoError, __LINE__, MyFile)) RETURN ! !----------------------------------------------------------------------- ! Clear nonlinear state (basic state) variables and insure that the ! time averaged free-surface is zero for scaling below and next ! iteration. !----------------------------------------------------------------------- ! DO ng=1,Ngrids DO tile=first_tile(ng),last_tile(ng),+1 CALL initialize_ocean (ng, tile, iNLM) #ifdef SOLVE3D CALL initialize_coupling (ng, tile, 0) #endif END DO END DO #ifdef SOLVE3D ! !----------------------------------------------------------------------- ! Compute basic state final level thicknesses used for state norm ! scaling. It uses zero time averaged free-surface (rest state). ! Therefore, the norm scaling is time invariant. !----------------------------------------------------------------------- ! DO ng=1,Ngrids DO tile=last_tile(ng),first_tile(ng),-1 CALL set_depth (ng, tile, iADM) END DO END DO #endif ! !----------------------------------------------------------------------- ! Compute final adjoint state dot product norm. !----------------------------------------------------------------------- ! DO ng=1,Ngrids DO tile=first_tile(ng),last_tile(ng),+1 CALL ad_statenorm (ng, tile, knew(ng), nstp(ng), & & StateNorm(ng)) END DO IF (Master) THEN WRITE (stdout,20) ' PROPAGATOR - Grid: ', ng, & & ', Adjoint Final Norm: ', StateNorm(ng) END IF END DO ! !----------------------------------------------------------------------- ! Pack final adjoint solution into adjoint state vector. !----------------------------------------------------------------------- ! DO ng=1,Ngrids DO tile=last_tile(ng),first_tile(ng),-1 CALL ad_pack (ng, tile, Nstr(ng), Nend(ng), & & ad_state(ng)%vector) END DO END DO ! 10 FORMAT (/,a,i2.2,a,i3.3,a,i3.3/) 20 FORMAT (/,a,i2.2,a,1p,e15.6,/) 30 FORMAT (/,1x,a,1x,'ROMS/TOMS: started time-stepping:', & & ' (Grid: ',i2.2,' TimeSteps: ',i8.8,' - ',i8.8,')') ! RETURN END SUBROUTINE propagator_afte END MODULE propagator_mod