Custom Query (986 matches)

We have been working on the tangent linear and adjoint version of the nesting algorithms for a while. The backward time-stepping logic for the adjoint of nesting grid is very tricky, and it is not entirely correct. We will continue debugging.

Accidentally, I modified other files that are needed for this development in this repository. It needed to go to one of my research repositories. I am restoring those files to the previous version r959 to avoid any confusion and interference with other algorithms. We need further testing to check that other algorithms work well with the changes. I apologize for the inconvenience. Those files are:

ROMS/Modules/mod_nesting.F
ROMS/Drivers/tl_ocean.h
ROMS/Drivers/ad_ocean.h
ROMS/Adjoint/ad_step3d_t.F
ROMS/Adjoint/ad_initial.F
ROMS/Adjoint/ad_u2dbc_im.F
ROMS/Adjoint/ad_v2dbc_im.F
ROMS/Adjoint/ad_main3d.F
ROMS/Tangent/tl_main3d.F
ROMS/Tangent/tl_step3d_t.F
ROMS/Tangent/tl_u2dbc_im.F
ROMS/Tangent/tl_initial.F
ROMS/Tangent/tl_v2dbc_im.F
ROMS/Utility/set_contact.F
ROMS/Utility/checkadj.F
ROMS/Utility/ntimestep.F

In the ROMS Forum, Julia Levin ​proposed some enhancements to the Point Source algorithm to model river runoff using either the standard input switches LuvSrc or LwSrc. The LuvSrc option is used to model river runoff transport by modifying the horizontal advection transport fluxes whereas the LwSrc option is used as a vertical mass volume inflow.

I updated inp_decode.F to be more robust with the gfortran compiler during debugging. Really weird behavior in the DO-loop counter. It continued looping after the specified range. No idea where that comes from but this update force a strict looping. For example, in load_1d_i, we now have:

!  If not all values are provided for the variable, assume the last value
!  for the rest of the array.
!
      ic=0
      IF (Ninp.le.Nout) THEN
        DO i=1,Ninp
          ic=ic+1
          Vout(i)=INT(Vinp(i))
        END DO
        IF (Nout.gt.Ninp) THEN
          Nstr=Ninp+1
          DO i=Nstr,Nout
            ic=ic+1
            Vout(i)=INT(Vinp(Ninp))
          END DO
        END IF
      ELSE
        DO i=1,Nout
          ic=ic+1
          Vout(i)=INT(Vinp(i))
        END DO
      END IF
      Nval=ic

I only implemented the changes associated with LuvSrc. The suggested changes for LwSrc are unstable and blow-up for the TS_MPDATA advection option. It requires further work. The old algorithm implemented the vertical mass inflow by modifying the vertical advection fluxes in the tracer equations and by modifying the omega equation that is associated with the 3D continuity equation. The proposed change is to introduce a forcing term to the rate of change of tracer. At the moment, such treatment is problematic and unstable for a simple Euler forward step. There are other issues to consider like conservation and variance of the tracer.

For the LuvSrc, we now have in pre_step3d.F:

!
!  Apply tracers point sources to the horizontal advection terms,
!  if any.
!
          IF (LuvSrc(ng)) THEN
            DO is=1,Nsrc(ng)
              Isrc=SOURCES(ng)%Isrc(is)
              Jsrc=SOURCES(ng)%Jsrc(is)
              IF (((Istr.le.Isrc).and.(Isrc.le.Iend+1)).and.            &
     &            ((Jstr.le.Jsrc).and.(Jsrc.le.Jend+1))) THEN
                IF (INT(SOURCES(ng)%Dsrc(is)).eq.0) THEN
                  IF (LtracerSrc(itrc,ng)) THEN
                    FX(Isrc,Jsrc)=Huon(Isrc,Jsrc,k)*                    &
     &                            SOURCES(ng)%Tsrc(is,k,itrc)
                  ELSE
                    FX(Isrc,Jsrc)=0.0_r8
                  END IF
                ELSE
                  IF (LtracerSrc(itrc,ng)) THEN
                    FE(Isrc,Jsrc)=Hvom(Isrc,Jsrc,k)*                    &
     &                            SOURCES(ng)%Tsrc(is,k,itrc)
                  ELSE
                     FE(Isrc,Jsrc)=0.0_r8
                  END IF
                END IF
              END IF
            END DO
          END IF

Changes were made to the tangent, representer, and adjoint versions of pre_step3d.F and step3d_t.F.

Many thanks to Julia Levin, John Wilkin, and Chuning Wang (Jupiter Intelligence INC) for looking and testing the formulation of Point Sources.

Updated the EcoSim model to include diagnostics of the optical terms when DIAGNOSTICS_BIO and BIO_OPTIC to write out into output NetCDF file DIANAME various underwater light spectral properties.

• The standard input script ecosim.in now has several logical switches to process:

  ! Logical switches (TRUE/FALSE) to activate writing of time-averaged,
  ! underwater optical diagnostics terms into DIAGNOSTIC output file [Ngrids].
  
    Dout(idSpIr) == T      ! surface_irradiance      surface spectral irradiance
    Dout(iddIrr) == T      ! downward_irradiance     downward spectral irradiance
    Dout(idsIrr) == T      ! scalar_irradiance       scalar spectral irradiance
    Dout(idAcos) == T      ! light_cosine            light spectral average cosine
    Dout(idLatt) == T      ! light_attenuation       diffuse light attenuation
    Dout(idaPHY) == T      ! Phy_absorption          phytoplankton absorption
    Dout(idaDET) == T      ! detrital_absorption     detrital absorption
    Dout(idaCDC) == T      ! CDC_absorption          CDC absorption
    Dout(idsPHY) == T      ! Phy_scattering          phytoplankton scattering
    Dout(idbPHY) == T      ! Phy_backscattering      phytoplankton backscattering
    Dout(idsTOT) == T      ! total_scattering        total scattering
    Dout(idbTOT) == T      ! total_backscattering    total backscattering
  

• The varinfo.dat was updated to include the metadata for the new light spectral properties.

• Added new routines bc_4d.F and exchange_4d.F to process the open boundary conditions for 4D spectral optical fields.

• If ECOSIM is activated, all output NetCDF files contain the light coordinate for all optical spectral variables:

         double light(Nbands) ;
                 light:long_name = "center wavelength of spectral irradiance bands" ;
                 light:units = "nanometer" ;
  

  Currently, ECOSIM is configured with 60 wavelenghts of light spectral bands ranging from 402:5:697 nanometers. Notice that all the def_*.F routines that creates ROMS output NetCDF files were updated to include this light coordinate.

• A new C-preprocessing option BIO_OPTIC is introduced to compute the underwater light spectral properties that are essential for any biogeochemical model. All the EcoSim model routines (ecosim.h, ecosim_def.h, ecosim_inp.h, ecosim_mod.h, ecosim_var.h, and ecosim_wrt.h) were updated.

• All the changes to the ROMS code were tested with the BIO_TOY test case:

• The BIO_TOY application was also updated in the test repository. It now includes Matlab script ecosim_optics.m used to generate the above plots.

Many thanks to Bronwyn Cahill for her help coding and testing this update to the EcoSim model.