Hi all,
I am running some experiments with ensemble members in ACCESS-CM2.
To branch ensemble members I have used the perturbIC_MD.py script in my git repo here: GitHub - SebastianMckenna/perturb_access: perturbing ACCESS-CM2 atmosphere intial conditions
This is the same as @MartinDix perturb_IC.py script but I got rid of the the modifications to stash code 388, so that I am only perturbing theta in the atmosphere.
I have used this script to apply a 10^-5 perturbation to the atmosphere restart file which looks like the plot below. I have confirmed this perturbation is present in the restart file - see this notebook here
The Problem
The perturbation does not affect the model (when the coupled job runs for 1 year) and only affects in month 2 when the coupled job is submitted monthly
-
I perturbed my restart file by 0.00001 (seed = 1234) with the script in the above repo and then used this to run an experiment - suite
u-de695. This is branched from u-dd756 which runs the same experiment with no perturbation. -
I ran this experiment (u-de695) for 2 years, with a yearly job resubmission, and the perturbation (which i check is present in the restart file above) does not seem to have been applied - all the outputs for SST (daily and monthly) and atmospheric temp (monthly, did not save daily output) are identical to the unperturbed version (u-dd756)
-
I also tested this with a monthly re-submission for 2 months to test this method works initially (u-de759). I found that when the coupled job is resubmitted on a monthly basis the experiment begins to branch off on the first day of the second month which is also really strange???
For the visualisations of these problems please see this notebook: here
Is this a known problem with perturbing ACCESS-CM2?
Does the model not couple to restart files correctly?
Is this something else?
Any help understanding what has gone on would be appreciated!
Thanks