how to make ensemble runs using CESM1.2

[hannay]

I don't run large ensemble so I usually run one experiment at a time.
I believe that @jedwards could answer your question about ensemble.

 

[jedwards]

CESM2 has capability to generate ensemble members and cylc workflow control files for running ensembles, however there is nothing like this in cesm1.2.

 

[wxalpha]

CESM2 has capability to generate ensemble members and cylc workflow control files for running ensembles, however there is nothing like this in cesm1.2.

Hi @jedwards, Thank you for your guidance! I am trying to conduct ensemble runs using CESM2.1. I am wondering could you direct me to the instructions? I could not find related threads in the forum.

 

[jedwards]

In cesm2.1.3 ./create_clone has options --ensemble and --keep-exe which will aid in creating an ensemble of cases.
cime/scripts/Tools/generate_cylc_workflow.py will create a suite.rc file that can be used with the workflow control tool cylc

 

[wxalpha]

In cesm2.1.3 ./create_clone has options --ensemble and --keep-exe which will aid in creating an ensemble of cases.
cime/scripts/Tools/generate_cylc_workflow.py will create a suite.rc file that can be used with the workflow control tool cylc

Many thanks for the quick reply! I am wondering if I need to add different pertlim values to each of the user_cam_nl files created by ./creat_clone --ensemble?

 

[jedwards]

Yes - that would be the normal process for creating an ensemble.

 

[wxalpha]

Yes - that would be the normal process for creating an ensemble.

Thanks again for the guidance! I guess I still have to add the pertlim to individual cases manually. :)

 

[wxalpha]

H

Yes - that would be the normal process for creating an ensemble.

Hi @jedwards,

Because I am new to CESM, I am wondering whether two hybrid runs with the same reference case, the same start date, and the same pertlim value produce the same outputs (bit-to-bit results)?

Thank you so much for your help!

 

[jedwards]

On the same machine with the same compiler and pe-layout, yes they should give bfb results.

 

[AndrewY]

On the same machine with the same compiler and pe-layout, yes they should give bfb results.

Hi jedwards,

Do ensemble members have to be hybrid run? I wonder if it is okay to clone an existing case and do another startup run with pertlim specified in user_nl_cam? Thanks!

 

[jedwards]

No they do not have to be hybrid runs. Yes what you are suggesting is doable.

 

[Li1108]

The first method I pointed you to (varying pertlim in the CAM namelist) is a very common an recognized method to create ensemble runs.For instance, it what was used to create the Large ensemble done at NCAR.See: http://www2.cesm.ucar.edu/models/experiments/LENSIt is an ensemble of 30 runs created using round off error introduced through the CAM namelist (i.e. varying pertlim)http://www2.cesm.ucar.edu/models/experiments/LENS/instructionsThey varied pertlim from pertlim = 1.d-14, 2.d-14, ..., 30.d-14to create their 30 ensemble members. I am not very familiar with the multi instance. To my knowledge, it is used for data assimilation but not really to create traditionnal ensemble.See: http://www.cesm.ucar.edu/models/cesm1.2/cesm/doc/usersguide/x834.html

Hi Hannay,

Thanks for your reply and information! I am also learning how to do the ensemble simulations using CESM1.2 but I still do not understand the values assigned to pertlim. For example, 1.d-14, what does "1" "d-14" mean here? Does the former refer to the realization of the ensemble member?

Best,
Li

 

[jedwards]

Li that is a representation of the value 0.00000000000001 - and that is the perturbation to be added to the initial condition of the T field.

 

[Li1108]

Li that is a representation of the value 0.00000000000001 - and that is the perturbation to be added to the initial condition of the T field.

Ok. Thanks jedwards!