Improve accuracy of the SCM "Replay" mode #3051
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
wlin7
reviewed
Aug 9, 2019
| @@ -1887,7 +1887,7 @@ subroutine cime_init() | |||
| !---------------------------------------------------------- | |||
|
|
|||
| areafact_samegrid = .false. | |||
| #if (defined BFB_CAM_SCAM_IOP ) | |||
| #if (defined E3SM_SCM_REPLAY ) | |||
| if (.not.samegrid_alo) then | |||
| call shr_sys_abort(subname//' ERROR: samegrid_alo is false - Must run with same atm/ocn/lnd grids when configured for scam iop') | |||
There was a problem hiding this comment.
Not essential but just to have a more consistent message. change "when configured for scam iop" to " ... for scm replay". Similarly for line 1702.
wlin7
reviewed
Aug 9, 2019
| @@ -3979,7 +3979,7 @@ if ($cfg->get('scam')) { | |||
| } | |||
|
|
|||
| # CAM generates IOP file for SCAM | |||
| if ($cfg->get('camiop')) { | |||
| if ($cfg->get('e3smreplay')) { | |||
| add_default($nl, 'inithist', 'val'=>'CAMIOP'); | |||
There was a problem hiding this comment.
About the 'CAMIOP' type of inithist: do we want to rename this type to reflect the current naming convention for this feature? If changed, it would trigger changes in several other files (including configure, build-namelist, namelist_definition.xml, and cam_history.F90). It can certainly be considered in a future PR rather than now.
wlin7
reviewed
Aug 9, 2019
| ! if the dataset is a CAM generated dataset set use_camiop to true | ||
| ! CAM IOP datasets have a global attribute called CAM_GENERATED_IOP | ||
| ! if the dataset is a CAM generated dataset set use_replay to true | ||
| ! E3SM IOP datasets have a global attribute called E3SM_GENERATED_IOP |
There was a problem hiding this comment.
line 682 global attribute "E3SM_GENERATED_FORCING" instead of "E3SM_GENERATED_IOP"
wlin7
reviewed
Aug 9, 2019
| call outfld('shflx ',cam_in%shf, pcols, lchnk) | ||
| call outfld('lhflx ',cam_in%lhf, pcols, lchnk) | ||
| call outfld('trefht ',cam_in%tref, pcols, lchnk) | ||
| call outfld('Tg', cam_in%ts, pcols, lchnk) | ||
| call outfld('Tsair',cam_in%ts, pcols, lchnk) |
There was a problem hiding this comment.
I think cam_in%tref is more suitable as Tsair here. If PS is not exactly reproduced, tsair could be used for vertical interpolation of T.
|
Hi @bogensch , the PR looks good. The suggested changes are really not essential. Can you take a look if you would like to get this PR into master as is for now, and probably consider the suggested changes in a future PR? |
|
Hi @wlin7 thanks for the careful review, as always. Sounds good to me, yes let's get the PR in for now and I'll consider your comments in another SCM branch I'm working on. |
|
Thanks @bogensch . your plan sounds good to me. Not sure when regression tests are going to start tonight. Look to merge it tomorrow. |
wlin7
approved these changes
Aug 14, 2019
wlin7
added a commit
that referenced
this pull request
Aug 14, 2019
Improve accuracy of the SCM "Replay" mode This PR improves the accuracy of the SCM Replay mode. Previous implementation allowed the user to "replicate the behavior" of an E3SM run in SCM, but with temperature errors on the order of ~0.1 K after a day. This PR will provide the user with capability to produce "quasi-bit-for-bit" results with Replay mode, with temperature errors on the order of ~1.e-5 K. The reason why Replay mode is unable to provide fully B4B results is because the dynamics tendency computed for Replay is not identical to the way that the full model computes its dynamics tendencies (which involves sub-cycling in the SE dynamical core). Thus, there is issue with roundoff. Note that in the past, CAM Replay mode with Eulerian dy-core could produce "B4B" results but only because if the flag to generate Replay mode output was turned on, they updated their T and Q based on the tendency computed for Replay purposes. Thus the run to generate Replay forcing was NOT b4b with a normal production run. We are working on a truly B4B Replay mode for E3SM and a future PR will support this which will involve higher precision computation of the dynamical tendencies. This PR also renames the old "-camiop" flag to "-e3smreplay" flag to be more current and consistent with E3SM. Other flags are renamed accordingly throughout the code. Note that in this PR the e3sm_developer test "SMS_R_Ld5.ne4_ne4.FSCM5A97" is an expected fail when comparing results to baseline. This is because where the large-scale vertical velocity is read in and updated had to be moved so that CLUBB could see it at the consistent timestep. It was validated that results for selected tested cases (ARM97, GOAMAZON, DYCOMS-RF01, BOMEX) have very minimal impacts with this fix included. All other e3sm_developer tests pass. [BFB] except for SCM test.
|
Merged to next. |
wlin7
added a commit
that referenced
this pull request
Aug 15, 2019
Improve accuracy of the SCM "Replay" mode This PR improves the accuracy of the SCM Replay mode. Previous implementation allowed the user to "replicate the behavior" of an E3SM run in SCM, but with temperature errors on the order of ~0.1 K after a day. This PR will provide the user with capability to produce "quasi-bit-for-bit" results with Replay mode, with temperature errors on the order of ~1.e-5 K. The reason why Replay mode is unable to provide fully B4B results is because the dynamics tendency computed for Replay is not identical to the way that the full model computes its dynamics tendencies (which involves sub-cycling in the SE dynamical core). Thus, there is issue with roundoff. Note that in the past, CAM Replay mode with Eulerian dy-core could produce "B4B" results but only because if the flag to generate Replay mode output was turned on, they updated their T and Q based on the tendency computed for Replay purposes. Thus the run to generate Replay forcing was NOT b4b with a normal production run. We are working on a truly B4B Replay mode for E3SM and a future PR will support this which will involve higher precision computation of the dynamical tendencies. This PR also renames the old "-camiop" flag to "-e3smreplay" flag to be more current and consistent with E3SM. Other flags are renamed accordingly throughout the code. Note that in this PR the e3sm_developer test "SMS_R_Ld5.ne4_ne4.FSCM5A97" is an expected fail when comparing results to baseline. This is because where the large-scale vertical velocity is read in and updated had to be moved so that CLUBB could see it at the consistent timestep. It was validated that results for selected tested cases (ARM97, GOAMAZON, DYCOMS-RF01, BOMEX) have very minimal impacts with this fix included. All other e3sm_developer tests pass. [BFB] except for the SCM test.
|
Merged. Request to bless SCM test on sandiatoss3 and melvin. |
rljacob
pushed a commit
that referenced
this pull request
Apr 12, 2021
Improve accuracy of the SCM "Replay" mode This PR improves the accuracy of the SCM Replay mode. Previous implementation allowed the user to "replicate the behavior" of an E3SM run in SCM, but with temperature errors on the order of ~0.1 K after a day. This PR will provide the user with capability to produce "quasi-bit-for-bit" results with Replay mode, with temperature errors on the order of ~1.e-5 K. The reason why Replay mode is unable to provide fully B4B results is because the dynamics tendency computed for Replay is not identical to the way that the full model computes its dynamics tendencies (which involves sub-cycling in the SE dynamical core). Thus, there is issue with roundoff. Note that in the past, CAM Replay mode with Eulerian dy-core could produce "B4B" results but only because if the flag to generate Replay mode output was turned on, they updated their T and Q based on the tendency computed for Replay purposes. Thus the run to generate Replay forcing was NOT b4b with a normal production run. We are working on a truly B4B Replay mode for E3SM and a future PR will support this which will involve higher precision computation of the dynamical tendencies. This PR also renames the old "-camiop" flag to "-e3smreplay" flag to be more current and consistent with E3SM. Other flags are renamed accordingly throughout the code. Note that in this PR the e3sm_developer test "SMS_R_Ld5.ne4_ne4.FSCM5A97" is an expected fail when comparing results to baseline. This is because where the large-scale vertical velocity is read in and updated had to be moved so that CLUBB could see it at the consistent timestep. It was validated that results for selected tested cases (ARM97, GOAMAZON, DYCOMS-RF01, BOMEX) have very minimal impacts with this fix included. All other e3sm_developer tests pass. [BFB] except for the SCM test.
rljacob
pushed a commit
that referenced
this pull request
Apr 21, 2021
Improve accuracy of the SCM "Replay" mode This PR improves the accuracy of the SCM Replay mode. Previous implementation allowed the user to "replicate the behavior" of an E3SM run in SCM, but with temperature errors on the order of ~0.1 K after a day. This PR will provide the user with capability to produce "quasi-bit-for-bit" results with Replay mode, with temperature errors on the order of ~1.e-5 K. The reason why Replay mode is unable to provide fully B4B results is because the dynamics tendency computed for Replay is not identical to the way that the full model computes its dynamics tendencies (which involves sub-cycling in the SE dynamical core). Thus, there is issue with roundoff. Note that in the past, CAM Replay mode with Eulerian dy-core could produce "B4B" results but only because if the flag to generate Replay mode output was turned on, they updated their T and Q based on the tendency computed for Replay purposes. Thus the run to generate Replay forcing was NOT b4b with a normal production run. We are working on a truly B4B Replay mode for E3SM and a future PR will support this which will involve higher precision computation of the dynamical tendencies. This PR also renames the old "-camiop" flag to "-e3smreplay" flag to be more current and consistent with E3SM. Other flags are renamed accordingly throughout the code. Note that in this PR the e3sm_developer test "SMS_R_Ld5.ne4_ne4.FSCM5A97" is an expected fail when comparing results to baseline. This is because where the large-scale vertical velocity is read in and updated had to be moved so that CLUBB could see it at the consistent timestep. It was validated that results for selected tested cases (ARM97, GOAMAZON, DYCOMS-RF01, BOMEX) have very minimal impacts with this fix included. All other e3sm_developer tests pass. [BFB] except for the SCM test.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
This PR improves the accuracy of the SCM Replay mode. Previous implementation allowed the user to “replicate the behavior” of an E3SM run in SCM, but with temperature errors on the order of ~0.1 K after a day. This PR will provide the user with capability to produce “quasi-bit-for-bit” results with Replay mode, with temperature errors on the order of ~1.e-5 K.
The reason why Replay mode is unable to provide fully B4B results is because the dynamics tendency computed for Replay is not identical to the way that the full model computes its dynamics tendencies (which involves sub-cycling in the SE dynamical core). Thus, there is issue with roundoff. Note that in the past, CAM Replay mode with Eulerian dy-core could produce “B4B” results but only because if the flag to generate Replay mode output was turned on, they updated their T and Q based on the tendency computed for Replay purposes. Thus the run to generate Replay forcing was NOT b4b with a normal production run. We are working on a truly B4B Replay mode for E3SM and a future PR will support this which will involve higher precision computation of the dynamical tendencies.
This PR also renames the old “-camiop” flag to “-e3smreplay” flag to be more current and consistent with E3SM. Other flags are renamed accordingly throughout the code.
Note that in this PR the e3sm_developer test “SMS_R_Ld5.ne4_ne4.FSCM5A97” is an expected fail when comparing results to baseline. This is because where the large-scale vertical velocity is read in and updated had to be moved so that CLUBB could see it at the consistent timestep. It was validated that results for selected tested cases (ARM97, GOAMAZON, DYCOMS-RF01, BOMEX) have very minimal impacts with this fix included.
All other e3sm_developer tests pass.