Strange behaviour upon restarting a WACCM-CESM 1.0.2 fully coupled run

[swahl@geomar_de]

Dear Support Team

We currently trying to extend a fully coupled CESM-WACCM RCP 8.5 scenario run (using CESM 1.0.2) beyond 2100 by fixing GHGs, aerosol, etc to 2100 conditions by setting the relevant fields in the CAM namelist to 'CYCLICAL'. We do this by restarting the run in 2100-01-01 using a modified CAM namelist. However the model doesn't behave as expected as we get really large changes in e.g. surface Temperature patterns. What worries us a lot more is an increase in stratospheric zonally averaged winds by up to 10m/s. About 40-50% of the differences between the "normal" RCP8.5 scenario run in Jan 2100 and a restarted run can be explained by only fixing the GHGs to year 2100 values taken from the file used by RCP8.5 scenario run in 2100. The only difference in namelist settings in this case cut down to:

"Normal RCP8.5 scenario run reading GHG values from files" (stops at 2100-12-31)
&chem_surfvals_nl
bndtvghg = '$DIN_LOC_ROOT/atm/cam/ggas/ghg_rcp85_1765-2500_c100203.nc'
scenario_ghg = 'RAMPED'
...
...
/

"Restarted run using fixed year 2100 conditions" (restarted in 2100-01-01 using restart files from above run from 2100-01-01)
&chem_surfvals_nl
scenario_ghg = 'FIXED'
co2vmr = 935.9e-6
f11vmr = 1010.0e-12
f12vmr = 167.3e-12
ch4vmr = 3751.0e-9
n2ovmr = 435.1e-9
...
...
/

What we have done so far:
- Carefully checked the fixed values whether they really match the 2100 values used in the "RAMPED" run --> Yes, they do.
- cross-checked whether the way we restart the model works correctly by removing all namelist modifications and only
introducing irrelevant modifications like additional output parameters --> no differences between the two runs.
- We also tested sensitivity against other parameters we want to fix at 2100 conditions (ext_frc_type, sad_type, srf_emis_type, prescribed_aero_type, aerodep_flx_type). It turns out that fixing each of them to year 2100 conditions can significantly influence stratospheric zonal winds already in the first months.

We were expecting to only see random differences in the first month of the restarted run as the FIXED values are taken from the "ghg_rcp85_1765-2500_c100203.nc" file in year 2100.

I now that this setup might be complicated to understand for those who haven't it set up.
Please leave comments if you have an idea what might go wrong here.

Sebastian

 

[santos]

I have very limited knowledge about the scientific issues here, but I want to ask: Are the differences that you are seeing bigger than, or qualitatively different from, typical interannual variability? In most model configurations, even small, random perturbations can cause quite significant divergence from the original within a couple of weeks. After a month, you have to expect some level of natural variability.

 

[swahl@geomar_de]

Dear all,
the issue is solved. It turned out that we had a wrong scaling in our analysis tool making the changes between the runs bigger than they actually are. Stupid me!Thank you anyways for reading through my long postSebastian