Run type
I can’t find the run type documentation for CESM2, but the analagous documentation from CESM1 can be found on this Running CESM page.
Note
There was some confusion on my part about whether the KiloCAM experiment
should be a branch or a hybrid run. To extend the Reanalysis from 2019 to
2020, Kevin set up a branch run because it allowed him to change the SST
forcing for the reanalysis while, at the same time, using restart files.
For the KiloCAM experiment, we require using perturb_single_instance
to
generate the thousand members, which only works on initial files rather than
restart files, so it should be set up as a hybrid run.
Initial files
The initial files from the CAM6 reanalysis are here:
/lustre/project/k1421/Reanalyses/f.e21.FHIST_BGC.f09_025.CAM6assim.011/rest/2018-01-01-00000
The individual filenames are of the form:
f.e21.FHIST_BGC.f09_025.CAM6assim.011.cam_00??.i.2018-01-01-00000.nc
Using perturb_single_instance
The &perturb_single_instance_nml
namelist in input.nml
has this
structure:
&perturb_single_instance_nml
ens_size = 3
input_files = 'caminput.nc'
output_files = 'cam_pert1.nc','cam_pert2.nc','cam_pert3.nc'
output_file_list = ''
perturbation_amplitude = 0.2
/
The simplest way to generate the initial ensemble is to use a symbolic link
for caminput.nc
and write the names of the desired perturbed initial files
into the file declared in output_file_list
.