Warm starts in ACCESS-ESM1.5

Hi everyone,

I am currently attempting to use the latest release of the ACCESS-ESM1.5 model to perform a series of freshwater perturbation experiments using the run ACCESS-ESM1.5 instructions. I am trying to branch off the preindustrial control at different years using the updated warm-start.sh and other files found here, previously discussed in this post, which have been adapted from the previous release.

However, even though warm-start.sh is building the expected /restart000 directory, the runs still fail. I am currently using the PI-02 restart files in/g/data/p73/archive/CMIP6/ACCESS-ESM1-5/PI-02/restart/ but would ultimately like to use the restart files from PI-01, but I suspect that will require modification of warm-start-csiro.sh.

I am new to using ACCESS-ESM1.5 and unsure if there is a simple step that I am overlooking.

Thanks in advance for any help!

Hi Patrick

Can you please share the error you are getting ?

Also - can you please push your configuration to your github fork of the esm1.5-configs repository, I will have a look.

If you aren’t familiar with git / github, here are some useful instructions

A once-off setting it github commandline client
the actual uploading of a config

1 Like

Hi Anton,

Thanks for your reply, and the git instructions.

The error I’m getting is
FileNotFoundError: CICE restart file not found in /scratch/lo70/pb8549/access-esm/archive/PI-02-byear-701-test2-PI-02-byear-700-test2-665d49cc/restart000/ice. Expected iced.07000101 to exist. Is 'dumpfreq' set in cice_in.nml consistently with the run-length? error.

I have pushed my template config to github here, as well as the failed run, the error file can be found here here.

I suspect this issue is that in archive/restart000/ice the warm start files built iced.7000101 but the model is looking for iced.07000101.

Thanks again!
Pat

Nice one

I think you’ll be fine to rename iced.7000101 to iced.07000101 in your restart dir.

If you are starting a new experiments, it’s neater to specify the initial restart files in restart: line of config.yaml here, instead of linking to an archive directory.

Sorry I am not very familiar with the warm start scripts, so I may have missed something.

2 Likes

Hi @Pat, just a quick update on the warm-start scripts. It looks like scripts were incorrectly omitting the leading zero when renaming the ice restart files. I’ve updated the scripts here to fix this up, and so they should avoid this error in the future.

2 Likes

Thanks @anton @spencerwong for the help! The new warm start script works and I was succesfully able to branch a run from the PI-02 repository.

I would like to branch some runs off the PI-01 restart files, and have modified the section of warm-start-csiro.sh where $yearstart and $pyearend are defined to adjust for the different restart years in PI-01 compared to PI-02;
yearstart="$(printf '%04d' $((source_year + 1)) )0101"
pyearend="$(printf '%04d' $source_year)1231"

Are there any other considerations or differences for PI-01 I should be aware of?

Thanks again for all the assistance.

1 Like

Good news!

PI-01 was run on raijin (e.g. the computer before Gadi) and was used for CMIP6 publications

PI-02 is a newer run than PI-01 but was not published. If you want to compare your new runs to existing output, its better to branch from PI-02. Saying that, I expect that output from any run done now with new esm1.5 software builds would still have rounding error differences to the PI-02 runs. You could do a short run of an unmodified configuration and compare to the PI-01/PI-02 runs to have confidence your run is comparable to existing output. If you don’t need to compare to existing output, then this doesn’t matter.

In general, I suggest doing some sanity checks on your output before running it too long. e.g. is the output you need working and available?

There have been bugs and mix ups with calendars getting out of sync in the past, hopefully we have these resolved now. For example, sometimes the ice calendar and ocean calendar were 100 years out of sync. The payu error you had above was some error checking Spencer added to protect against that sort of error.

1 Like