Open wwieder opened 3 weeks ago
Are the b4b parameter file changes the ones from this PR?
Yes, I'll link above too for clarity
I made new ctsm60
parameter files , starting with @slevis-lmwg's most recent (ctsm60_params.c240814.nc
), then applying Linnia's changes, then applying Fang's.
I checked for conflicts at both steps. Sam L. and Linnia modified different parameters, so nothing there. Linnia and Fang both modified froot_leaf
, but fortunately different PFTs, so no conflicts there either.
The files are in /glade/work/samrabin/ctsm_fire-li2024-pr2_paramfiles/
:
ctsm60_params.c240822.li2021gswpfrc.nc
: The version with just Linnia's changes added to Sam L.'s. This should be the default for clm6_0
runs with the li2021gswpfrc
fire_method
.ctsm60_params.c240822.li2024gswpfrc.nc
: Applied Fang's new parameterizations for GSWP3 forcings onto the above paramfile. Should be the default for clm6_0
runs with the li2024gswpfrc
fire_method
.ctsm60_params.c240822.li2024crujra.nc
: Applied Fang's new parameterizations for CRU-JRA forcings onto the above paramfile. Should be the default for all other clm6_0
runs.I didn't think I should apply Linnia's and Fang's changes to any of the other paramfiles (clm45_params
, clm50_params
, ctsm51_params
, ctsm60_params_cn30
), but let me know if I'm wrong there. If I'm not, then the c240814
versions of those parameter files can just be copied to c240822
versions.
Thanks for merging these, Sam. I'm curious what changes made to feoot_leaf (and for what reason)?
I don't think we want to change parameter files for older physics versions.
We'll likely have to discuss how best to handle options for different datm options, but that can happen over the fall.
Left-hand side of <>
gives the values Linnia and Fang started with; right gives their new values:
$ nccmp -dfs -v froot_leaf $linnia_base $linnia
DIFFER : VARIABLE : froot_leaf : POSITION : [1] : VALUES : 1.5 <> 1.7693
DIFFER : VARIABLE : froot_leaf : POSITION : [2] : VALUES : 1.5 <> 1.7945
DIFFER : VARIABLE : froot_leaf : POSITION : [11] : VALUES : 1.2 <> 1.7915
DIFFER : VARIABLE : froot_leaf : POSITION : [12] : VALUES : 1.2 <> 1.7924
$ nccmp -dfs -v froot_leaf $fang_base $fang_gswp
DIFFER : VARIABLE : froot_leaf : POSITION : [6] : VALUES : 1.5 <> 0.5
DIFFER : VARIABLE : froot_leaf : POSITION : [14] : VALUES : 1.5 <> 0.5
$ nccmp -dfs -v froot_leaf $fang_base $fang_crujra
DIFFER : VARIABLE : froot_leaf : POSITION : [6] : VALUES : 1.5 <> 0.5
DIFFER : VARIABLE : froot_leaf : POSITION : [14] : VALUES : 1.5 <> 0.5
My note from Fang says, "Changed froot_leaf
of tropical BDT and C4 grass to 0.5, so more new carbon goes to the leaves to represent faster post-fire recovery as observed."
Thanks. Let's see how things look in the initial 5.3 simulations and we can evaluate
@wwieder @samsrabin I modified froot_leaf for BDT tropical (PFT=6) and C4 grass (PFT=14) in the parameter file because in tropical savannas, where fires are frequent, observations show that root burning and damage during fires are much less than leaf burning for these two PFTs and leaves recover quickly after fires. This means new carbon should be allocated more to leaves than to roots for the 2 PFTs
I expect Fang's reduction of froot_leaf in BDT tropical & C4 grass to actually help the low LAI bias in tropical savannas (independent of fire feedbacks). Our CLM5.1 LAI tuning suggested reductions for these two PFTs as well.
I made new
ctsm60
parameter files , starting with @slevis-lmwg's most recent (ctsm60_params.c240814.nc
), then applying Linnia's changes, then applying Fang's.I checked for conflicts at both steps. Sam L. and Linnia modified different parameters, so nothing there. Linnia and Fang both modified
froot_leaf
, but fortunately different PFTs, so no conflicts there either.The files are in
/glade/work/samrabin/ctsm_fire-li2024-pr2_paramfiles/
:
ctsm60_params.c240822.li2021gswpfrc.nc
: The version with just Linnia's changes added to Sam L.'s. This should be the default forclm6_0
runs with theli2021gswpfrc
fire_method
.ctsm60_params.c240822.li2024gswpfrc.nc
: Applied Fang's new parameterizations for GSWP3 forcings onto the above paramfile. Should be the default forclm6_0
runs with theli2024gswpfrc
fire_method
.ctsm60_params.c240822.li2024crujra.nc
: Applied Fang's new parameterizations for CRU-JRA forcings onto the above paramfile. Should be the default for all otherclm6_0
runs.I didn't think I should apply Linnia's and Fang's changes to any of the other paramfiles (
clm45_params
,clm50_params
,ctsm51_params
,ctsm60_params_cn30
), but let me know if I'm wrong there. If I'm not, then thec240814
versions of those parameter files can just be copied toc240822
versions.
@samsrabin I am looking in /glade/work/samrabin/ctsm_fire-li2024-pr2_paramfiles, and I do not see the crujra.nc param file, which is the one that I need for this simulation.
Oh sorry, that one is just ctsm60_params.c240822.nc
.
@slevis-lmwg @samsrabin I want to clarify that the new set of parameters I provided is specifically for CTSM6 and should not be used with earlier CLM versions. The file located at /glade/derecho/scratch/fangli/firedata/ctsm60_params.fire.final-CRUJRA.nc is for CTSM6 when driven by CRUJRA data. The only difference from the one for GSWP3 is rswf_max because CRUJRA is around 15% wetter than GSWP3 during the fire season.
Diagnostics for last 20 years of pSASU compared to last 20 years of pSASU for ctsm52018_f09_xsiceON_acON_pSASU (Issue https://github.com/NCAR/LMWG_dev/issues/62 ):
JJA TLAI for last 20 years of pSASU for ctsm52026_f09_pSASU and ctsm52018_f09_xsiceON_acON_pSASU:
Seems encouraging? LAI is reduced but not dead!
Agreed!
Even more encouraging are the TOTVEGC totals 758 vs 962 PgC with new vs. old parameterizations! Can likely even take this lower with a full PPE calibration, but the LAI values look much improved, which was the aim for coupled model initial conditions. Thanks @linniahawkins!
Fire C fluxes and burned area are much higher, especially in tropical Savannas, once we have historical runs maybe @lifang0209 or @samsrabin can help us evaluate the fire behavior we're seeing?
@wwieder sure, I could evaluate the fire simulations. In GFED5, burned area is double that of GFED4s, and fire carbon emissions are 50% higher.
Thanks Fang, I like a factor of 2 uncertainty in different versions of the same observational product. Hopefully that is a target we can hit! My larger concern is if we fall well above gfed5 estimates
@wwieder before I take off for the weekend... Status of the historical simulation: Latest history written ctsm52026_f09_hist.clm2.h0.1900-02.nc I currently have this stopping in 1919 BUT if I get around to it, I will stop it when it writes the next restart file in 1904, so that I can kick off the next phase before I go.
UPDATE: Simulation not progressing fast enough for me to start the next phase before I go. It will wait for Friday evening or Saturday morning.
@wwieder do we want this run to include the extra output starting in 2000 for assessing pft survival? For now I will assume yes.
Yes, let's add the additional history output for the 21st century.
@wwieder:
Fire C fluxes and burned area are much higher, especially in tropical Savannas, once we have historical runs maybe @lifang0209 or @samsrabin can help us evaluate the fire behavior we're seeing?
The increased burned area in tropical savannas makes sense given the increased allocation to leaves for tropical BDT and C4 grass from @lifang0209's new parameters.
This spin-up and hist have completed.
Standard diagnostics for historical compared to https://github.com/NCAR/LMWG_dev/issues/62
ILAMB diagnostics:
https://webext.cgd.ucar.edu/I20TR/ctsm52026_f09_hist/lnd/_build_ctsm52026_f09_hist/ Note that the labeling refers to issue number, e.g, this issue number, I65.
NOTE: fire_method = li2024crujra not actually implemented in I65 due to namelist problem.
Lot's of moving pieces to get a 5.3 spinup started. I'll take notes here.
Description:
5.3 surface data (just f09 for now), including:
Code:
[x] Sam L's CTSM5.3 branch CTSM#2500 updated to master + PPE B4B dev work #2689
Answer changing tag defaults turned on (excess ice, AC & dust)
New fire code (default
fire_method
changed toli2024crujra
)Will not include alternative Sturm parameterization (Keith's experiment in #63, maybe less critical and potentially conflict with the hand tuned parameterization), we'll consider this as tuning for later
Parameter file that combines:. @samsrabin has this done
/glade/u/home/linnia/clm5ppe/pyth/CLM6_dev_handtuning/final_paramfiles/ctsm60_params.c240208_lh.nc
Case directory: Locally: /glade/u/home/slevis/cases_LMWG_dev/ctsm52026_f09_AD DONE /glade/u/home/slevis/cases_LMWG_dev/ctsm52026_f09_SASU DONE /glade/u/home/slevis/cases_LMWG_dev/ctsm52026_f09_pSASU DONE /glade/u/home/slevis/cases_LMWG_dev/ctsm52026_f09_hist DONE
Sandbox: Locally: /glade/work/slevis/git/ans_chging_tag_1 (sorry if the directory name is confusing) git branch: new_rawpftlai git describe: ctsm5.2.026-28-g40aacf6e9 (before making and pushing new branch tag) git describe: branch_tags/ctsm5.3.n01_ctsm5.2.026
usernl changes: AD
SASU difference from AD
pSASU diff from SASU
hist diff from pSASU
SourceMods: NONE
Diagnostics: Diagnostics for last 20 years of pSASU compared to last 20 years of pSASU for ctsm52018_f09_xsiceON_acON_pSASU (Issue #62 ):
https://webext.cgd.ucar.edu/I1850/ctsm52026_f09_pSASU/lnd/ctsm52026_f09_pSASU_401_420-ctsm52018_f09_xsiceON_acON_pSASUl_221_240/setsIndex.html
Standard diagnostics for historical compared to #62
https://webext.cgd.ucar.edu/I20TR/ctsm52026_f09_hist/lnd/ctsm52026_f09_hist_2004_2023-ctsm52018_f09_xsiceON_acON_hist_2004_2023/setsIndex.html
ILAMB diagnostics:
https://webext.cgd.ucar.edu/I20TR/ctsm52026_f09_hist/lnd/_build_ctsm52026_f09_hist/ Note that the labeling refers to issue number, e.g, this issue number, I65.
Output: Output (if still available): /glade/scratch/$USER/archive/$CASE/
Contacts: @slevis-lmwg
Extra details: We will cycle 1901-1920 of datm, starting with finidat from #62, and completing this sequence: AD 40 yrs --> SASU 270 yrs (~16% land area in disequilibrium in TOTECOSYSC) --> pSASU 420 yrs (3.43%) --> hist modeled after Daniel's approach described here: We ran a long spinup for the default parameters and used that as initial conditions for all ensemble members. Then for each new parameter set we ran, 20year AD, 80 years SASU, 40 years postSASU cycling a ten year forcing dataset.