cam6_4_200: Remove PS noise in SE-CSLAM, WACCM stability and other stuff - #1664
Conversation
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
…om pg3 to GLL, hypervis bug fix
…ion sponge in physics
# Conflicts: # src/dynamics/se/dycore/fvm_consistent_se_cslam.F90
ne30pg3 model_top=mt keeps -1 (operational) only for non-WACCM; the waccm_phys=1 variant gets the validated filter default 1.0.
…subcycle fix New namelist switch se_gll_advect_q (default off, b4b with production): under CSLAM, restore cam6_3_151-style GLL advection of thermodynamic tracers alongside CSLAM transport (2-timelevel Qdp, FQ forcing coupling, advection buffers) instead of the cslam2gll overwrite. Also: - Fix unstable auto-selection of hypervis_subcycle (drop 1.2 factor). NOT b4b vs baselines run with the old auto value; set se_hypervis_subcycle explicitly to reproduce those. Note that noise in abs_dps_dt was fixed with this (with gll double-advection setting)
Conflicts resolved in favor of this branch, which supersedes the cam6_4_199 (PR ESCOMP#1656) changes it already contains: - global_norms_mod.F90: keep the automatic hypervis_subcycle formula without the 1.2 fudge factor (removed on this branch, ESCOMP#1595) - namelist_defaults_cam.xml: keep the removal of the WACCM se_hypervis_subcycle overrides (automatic -1 default) - prim_driver_mod.F90: keep the gll_advect_q path and the unconditional cslam2gll (the .not.last_step guard is incompatible with hypervis_Qdp) - control_mod/fvm_mod/dyn_comp: keep the CSLAM Q filter and gll_advect_q declarations/plumbing - doc/ChangeLog: replace this branch's draft entry for PR ESCOMP#1656 with the official cam6_4_199 entry, below the pending entry Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
cacraigucar
changed the title
nusbaume
left a comment
nusbaume
left a comment
There was a problem hiding this comment.
Thanks For the dycore updates and fixes @PeterHjortLauritzen! I have some change requests, but hopefully nothing too difficult (and if there is a problem then just let me know). I should also note that I did not review fvm_filter_mod.F90, as I believe @jtruesdal was going to look over that file as part of his review.
jtruesdal
left a comment
jtruesdal
left a comment
There was a problem hiding this comment.
Nice job Peter. I did my best to check the stenciling and algorithms they all look good. I did notice that the mpi exchanges could be made more efficient by pulling a static nlev arrays out of the isub loop and exchanging proper sized buffers. I ran the mpi improvements and it was BFB on Izumi. If you agree with the suggestions I could update your branch if I have permission or push a PR with the changes if you'd like.
Co-authored-by: Jesse Nusbaumer <nusbaume@ucar.edu>
Co-authored-by: Jesse Nusbaumer <nusbaume@ucar.edu>
Co-authored-by: Jesse Nusbaumer <nusbaume@ucar.edu>
Co-authored-by: Jesse Nusbaumer <nusbaume@ucar.edu>
|
Thanks Jesse and John! I think the issues (thus far) have been addressed (but since Derecho is down I have not been able to check regression test impact of changes) |
cacraigucar
changed the title
nusbaume
left a comment
nusbaume
left a comment
There was a problem hiding this comment.
Thanks for addressing my comments @PeterHjortLauritzen! I just had one final comment (mostly to double-check if the latest changes are correct).
| <se_sponge_del4_lev waccm_phys="1" model_top="mt"> -1 </se_sponge_del4_lev> | ||
| <se_sponge_del4_lev waccm_phys="1">-1</se_sponge_del4_lev> | ||
| <se_sponge_del4_lev model_top="ht">-1</se_sponge_del4_lev> | ||
| <se_sponge_del4_lev model_top="xt"> 30 </se_sponge_del4_lev> |
There was a problem hiding this comment.
Do we want to keep the value at 30 when using the WACCM-X top?
There was a problem hiding this comment.
Good catch (I don't think it is needed anymore but it will need to be tested). Restored for now ...
|
@PeterHjortLauritzen I can run the regression tests later tonight if derecho comes up. @nusbaume I'm helping Peter get the mpi exchange mods correct and I'll test the new version just to make sure its BFB. |
jtruesdal
left a comment
jtruesdal
left a comment
There was a problem hiding this comment.
My issues have been addressed and the code is BFB with what came before the fvm_filter_mod.F90 improvements.
4 participants
closes #1595 Remove subtle noise in PS in SE-CSLAM
closes #1648 Wrong dycore settings when running WACCM physics with LT
closes #1523 optimize SE dycore for WACCM ne30pg3 and ne16pg3
closes #1487 2 degree (ne16pg3) out-of-the-box LT and MT functionality
closes #574 Setting time-steps based on model top location
Also other changes like default topo file for ne120pg3 etc. See ChangeLog for details