GEOS-Chem Adjoint v33

From Geos-chem
Revision as of 23:44, 6 January 2012 by Daven (Talk | contribs) (Update comments for STT_ADJ unit conversion)

Jump to: navigation, search

What's new in this version

Updates to forward model

Bug fixes in forward model

Ionic strength below zero in rpmares (adj33_001)

On prospero, we often get the message: Ionic strength below zero...negative concentrations.

Despite the setting of negative roots to zero, sometimes the code still ends up with negative CRUTES. Trap for it here to prevent mysterious crashing later. Also, add the following

          !IF ( CRUTES(1) .ne. ABS(CRUTES(1) ) ) CRUTES(1)=1d9
          !IF ( CRUTES(2) .ne. ABS(CRUTES(2) ) ) CRUTES(2)=1d9
          !IF ( CRUTES(3) .ne. ABS(CRUTES(3) ) ) CRUTES(3)=1d9

which for some reason (I have no idea why) prevents the code from returning a negative CRUTES on prospero.

--Daven 14:58, 6 January 2012 (EST)

Updated lightning parameterization and fix for cloud-top-height algorithm (v9-01-01)

Updates to a6_read_mod.f and lightning_nox_mod.f. Details in Lee's description:

--Daven 15:18, 6 January 2012 (EST)

Fix for GEIA emissions scaling factor over Botswana (v9-01-02h)

Updates to scale_anthro_mod.f. See GEOS-Chem_v9-01-02#Fix_for_GEIA_emissions_scaling_factor_over_Botswana

--Daven 16:16, 6 January 2012 (EST)

TINY parameter in convection_mod.f (v8-03-02)

Update to convection_mod.f GEOS-Chem_v8-03-02#TINY_parameter_in_convection_mod.f

--Daven 16:16, 6 January 2012 (EST)

Updates to adjoint model

Bug fixes in adjoint model

Mover error checking for O3 assimilation (adj33_002)

Nicolas Bousserez wrote:

the code crashes in "input_adj_mod.f" when doing the checking of observation settings (see line 2027 in my code). It is apparently due to "IDCSPEC_ADJ", which is not allocated/defined yet at the time the program executes those lines. I'm skipping these checks for now then. Note: The code crashes also when using "TES_O3_OBS" instead of "SCIA_DAL_NO2_OBS", so it is not related to what I implemented apparently.

The following block of code has been moved from input_adj_mod.f to subroutine INIT_CSPEC_ADJ in adj_arrays_mod.f, as it can only be called after the CSPEC variables have been initialized.

#elif defined ( TES_O3_OBS ) 
     ! Since the O3 obs operators will pass adjoints back 
     ! to CSPEC via CSPEC_AFTER_CHEM_ADJ, we need to make sure that 
     ! these species are listed as observed species
     FOUND = .FALSE.
     DO N = 1, NOBS_CSPEC

        IF ( TRIM( NAMEGAS( IDCSPEC_ADJ(N) ) ) == 'O3' ) THEN
           FOUND = .TRUE.

     IF ( .not. FOUND ) THEN

        CALL ERROR_STOP( ' Need to list O3 as observed species',
    &                    ' input_adj_mod ' )

--Daven 16:38, 6 January 2012 (EST)

Retire CSPEC_NO2_ADJ (adj33_003)

Nicolas Bousserez wrote:

In "chemdr_adj.F" I removed this part:
#if defined( SCIA_KNMI_NO2_OBS ) || defined( SCIA_DAL_NO2_OBS )
! Apply forcing from satellite observations
CSPEC_NO2_ADJ(:) = 0d0
because it seems like I'm already updating the adjoint of NO2 in "sciadal_no2_obs_mod.f" through CSPEC_AFTER_CHEM_ADJ. I would like to be sure that it is correct though.

Daven Henze wrote:

Yes, that is correct. I didn't want to take it out incase you had built your obs operator using CSPEC_NO2_ADJ. But since you used the more generic CSPEC_AFTER_CHEM_ADJ, then we can delete this section entirely and use the section higher up in this file ( IF (LCSPEC_OBS) ...) to apply the forcing from your obs operator to CSPEC_ADJ.

Now use CSPEC_AFTER_CHEM_ADJ instead of CSPEC_NO2_ADJ in chemdr_adj.f.

--Daven 18:04, 6 January 2012 (EST)

Fix to NaNs in STT_ADJ everywhere after adjoint of partitioning (adj33_004)

The forward gas-phase chemical solver may sometimes need to repeat the internal integration of the reaction rate ODEs within a single grid cell. When this happens, we now only redo the model calculation during the forward model run, not the adjoint, as this can lead to a corrupt CSPEC_ADJ array globally (not sure how that happens, but it does), eventually triggering a check for NaNs after the partitioning adjoint. See updates to chemistry_mod.f and gckpp_adj_Integrator.f90. Thanks to Richard Xu for helping pinpoint this bug.

--Daven 18:03, 6 January 2012 (EST)

Fix to NaNs in sulfate aerosol adjoint tracers (adj33_005)

Owing to cancelation of errors, it's possible to seemingly randomly end up with NaNs coming out of CHEM_SO2_ADJ . This can happen if L3 ends up equaling 1d0 exactly. We now trap for this and skip in the rate event this happens.

--Daven 18:03, 6 January 2012 (EST)

Remove superfluous debugging which had LAVHRRLAI set to FALSE

Noted by Fabien. Upated in input_mod.f.

--Daven 18:07, 6 January 2012 (EST)

Fix for reseting deposition adjoints (adj33_006)

Fabien Paulot wrote:

It seems it's "a bug" in the code. To keep track of the dry deposition flux of HNO3 I use dryhno3 (in cspec array). The CSPEC_ADJ(DRYHNO3,:) should be reset to zero at every time step but it is not. That's because you are saving in chk_cspec before it is reset to zero. Easy fix, just force to zero in chemdr_adj before the forcing is applied et voila, it works (the adjoint does not diverge any more after one hour). It should not affect anything as long as there is no forcing on dryhno3 (or any dry dep counters for that matter).

We add the following to reset the dry dep adjoints in chemdr_adj.f:

      DO N = 1,NUMDEP
         NK = NTDEP(N)
         IF (NK.NE.0) THEN
            JJ = IRM(NPRODLO+1,NK,NCS)
            IF (JJ.GT.0) THEN
               CSPEC_ADJ(:,JJ) = 0.0D0

--Daven 18:34, 6 January 2012 (EST)

Update comments for STT_ADJ unit conversion

Martin Keller wrote:

I'm currently guessing that CONVERT_UNITS(1,etc.) does something like:
STT:     kg -> v/v

STT_ADJ: J/[v/v] -> J/kg
and similarly CONVERT_UNITS(2,etc.)
STT:     v/v -> kg

STT_ADJ: J/kg -> J/[v/v]
From a dimensional analysis this would seem plausible. It would also explain some of the errors I'm currently fighting with. If this is correct, then a short comment should perhaps be added to the code.

The comments in geos_chem_adj_mod.f have been updated surrounding calls to CONVERT_UNITS.

--Daven 18:43, 6 January 2012 (EST)

Previous issues now resolved in v33

Previous issues now resolved in v32

CO2 emissions update to allow for both monthly and years emissions

Update to co2_mod.f (dkh, 03/07/11)

Updates to lidort_mod.f

  • Reinstate NCOUNT in CALC_RF_FORCE so that aod files are made at the right time. (dkh, 03/27/11)
  • now check to make sure that MAXTHREADS matches JJPAR in lidort_mod.f (dkh, 03/27/11)

Update to Mie table

Update mie_mod.f to use finite difference values for adjoints. (dkh, 03/27/11)

Remove bug in SET_SF

Remove erroneous debugging code in subroutine SET_SF in inverse_mod.f. (dkh, 03/27/11)

Outstanding issues not yet resolved in v33