How to run CESM with finer timesteps

[wangxiong]

Hello everyone,

I want to ask about how to run CESM with finer timesteps (for example, with STOP_N choice is 6h), the CESM version I use is CESM 1.2.1. I understand the CESM 1.2.1 is out of support for a long time, but this is latest version DART(data assimilation research testbed) can support. so I still want to post a thread, and maybe someone has same experience before.

Since the atmospheric observation at available at 6 hour intervals (NCEP DS090.0). I need to stop the model after a 6 hour run, do the assimilation, and run the model again. But I got some issues if I set the STOP_N less than 1 day. To be specific, I use the B_2000 compset with STOP_N = 12HOUR, after a few resubmits, the timestamp of restart files in rpointer file may differ. (like cam and other component restart file has timestamp 2014-01-09-43200, but pop restart file has timestamp 2014-01-10-00000), the timestamps are different and the model can not run any further. To solve this problem, I change the ocn_ncpl in env_run.xml from 1 to 4, and the timestamp problem won't happen anymore. But model turns to unstable and may corrupt with error message "ZM_COMV IENTROPY: Tmix did not converge" . After some simple search, I decrease the pop timestep by changing dt_count from 23 to 29 (26 not working). Also, I increase the nspltvrm from 2 to 4. After all these changes, the model can run a few more times and still may abort (too fragile to do the assimilation).

So I want to ask how to run CESM 1.2.1 with finer timestep (like 6h or 12h). Is there any additional configuration I missed? Any suggestion may help. Thank you.

Best wishes.
Wang Xiong

 

[sacks]

It should work to run:

./xmlchange STOP_OPTION=nhours
./xmlchange STOP_N=12

(I can't remember if CESM1.2 supports exactly that xmlchange syntax or if you need to adjust the syntax for that older model version.)

I also think that DART has been run with more recent versions of CESM – specifically the CESM2.1.z release series. I'll check with the DART developers and see if they have anything to add.

 

[wangxiong]

It should work to run:

./xmlchange STOP_OPTION=nhours
./xmlchange STOP_N=12

(I can't remember if CESM1.2 supports exactly that xmlchange syntax or if you need to adjust the syntax for that older model version.)

I also think that DART has been run with more recent versions of CESM – specifically the CESM2.1.z release series. I'll check with the DART developers and see if they have anything to add.

Thank you. Bill

Yes, the CESM1.2.1 do support the xmlchange syntax, and that are the exact moves I have done trying to do a 12 hours gap assimilation.
Now I am wondering may the frequent assimilation makes the model unstable, since the model run smoothly once I increase the timestep to 1day or more.

And DART do support more recent version of CESM(for example cam-fv has CESM2.0 DART interface), but accoding to the warning part of this page (Multi-Component CESM+DART Setup — DART 9.11.7 documentation)
CESM 1.2.1 is the latest version DART can support if you want to do the multi-component assimilation (in this case,I want to assimilate CAM CICE AND POP at the same time).

Best wishes,
Wang Xiong

 

[raeder]

CESM1.2.1 appears to be the only version with has interfaces for CAM, POP, and CLM
committed to DART. But I've asked Dan Amrhein about that, and the stability issues,
since his research focus is on coupled assimilation.
He's out until Monday, but might respond sooner.
Kevin

 

[raeder]

Correction; Dan is out until Wednesday.

 

[damrhein]

Apologies for the delay. Yes, 1.2.1 is the only version committed for multi-component including CAM. CESM2.1.1 is committed with POP and CICE (documented here), with plans in to commit a version including CAM. Regarding the stability, do I understand correctly that the model is stable on its own, and only unstable when using DA? Perhaps you could try very large values of obs uncertainty or reducing the number of observations to see where the instability is arising and verify it's something from adding increments?