Date: 26/10/2023
Participants: 19
Chair: @tiloz
Admin
@tiloz: ESM Working Group clashes: ACEAS Meeting on November 9th, CLEX 23rd. Please answer poll for meeting times to assist with deciding if the meetings should go ahead.
Working Group Resources
Shared Experiments
New NCRIS funding. 50TB for lg87. Currently no community proposals for how to use that storage.
@tiloz: Still no proposals for shared experiments for this quarter. Can use lg87 storage for shared experiments. Usually struggle for storage.
@eunpalim: Waiting for SAC to approve proposed ACCESS-S2 experiment. @tiloz will mention as part of Working Group update to SAC today. Minimum requirement was storage as join project with F&P WG. 10TB from each. Would be happy to utilise computing resources.
@LaurieM: Interest in running a standardised CMIP6 future scenario with higher resolution model @wghuneke is working on?
@wghuneke: Can run an experiment if required. Maybe defer to @eunpalim to run first. Currently running 0.25 deg present day and plan to do pre-industrial. For the standard (no BGC) CM2-025: cost 12.5 kSU/yr and walltime ~4 yr/d
@sofarrell definitely want to be able to share runs. Had some issues around Antarctica but that should be fixed.
@dhb599 Interested in new runs. Want to run with ocean-bgc. Previous run wasn’t clean due to crashes. Will run on CSIRO project. Important delivery for NESP project, as a base for ACCESS-CM2 0.25 run.
@tiloz Model crashes affect reproducibility? Long enough period would still be useful to the community. Use for initial analysis. @dhb599 can make available. Ran over 275 years with ocean bgc. Be careful using it because crashes means we can’t reproduce it. Runs at less than 3 model years/day. Not sure about compute and storage requirement.
@dhb599 need help to scale ocean. Maybe need some COSIMA help. Could maybe run 4 years/day. So no work done on efficiency and tuning. BGC adds a lot of time and resource use. Need to talk to @matthew.chamberlain
@AndyHoggANU ocean will scale, should be able to match the atmosphere.
@wghuneke Runs through accessdev rose/cylc on accessdev. About to start running on ARE on persistent session.
@AndyHoggANU Would it be useful for ACCESS-NRI (@paulleopardi) to do some model load-balancing optimisation? Answer: Yes!
@tiloz would be great to have faster throughput, 5-6 yr/day vs 2-3. Also useful for future CMIP7 work, but can benefit immediately.
@AndyHoggANU mentioned that new long Rose/Cylc based experiments should be run from persistent sessions rather than Accessdev.
@Aidan need to put shared information on the forum rather than in emails. Aidan will make important topics more visible.
ACCESS-NRI Update
@AndyHoggANU ACCESS-NRI has been slow to release compute time and storage because agreement ran out in June. Have been waiting for NCRIS funding announcement. Now been announced. Received most of requested funding.
Total data allocation is 3-4PB over 4 years. This includes data collections, working group storage and ACCESS-NRI storage.
If there is a run that is of interest for whole working group can keep it in /g/data/lg87. Can index, make it easy to access. If there is sufficient meta-data, can sustain and manage the storage.
Building up expertise with ESM2 type model is critical for the community.
If there is something the WG chairs think will benefit for community, ok to say yes. For @eunpalim’s experiment collaboration with another working group is fine, but better if all under one project. ESM is probably the better fit.
Major aspect funded: efforts to support CMIP7 and advances in CABLE4 LSM. Asked for 4 years of ($5M) gave us $2.5M. Interpreted as funded for 2 years. Re-apply for 2 more years. Feeling is NCRIS want to see how CMIP7 consortium comes together.
Will be hiring more people in Claire’s land team. Also in model evaluation, atmosphere, ocean and coupling. Will have to be fixed term appointments. Will help CMIP7 development effort. Also help to fund people working on CMIP7 who are currently funded off recurrent funding.
Have a current recruitment round just closed. Will have another early next tear.
Science talk
Cancelled.
What’s going on?
Changing coupled model land-sea mask
@paulleopardi What has been happening for ancillary generations for paleo runs?
@dkhutch recently got Miocene land-sea mask working in ACCESS-ESM-1.5. Have a number of scripts that are accessible/readable but will make available.
@paulleopardi want to know how it relates to other work, particularly what @atteggiani has been working on. Might be some possibility for rationalisation.
@dkhutch used older CMIP5 generation tools. Worked for me.
@tiloz understand there is a more long-term fix, but we do need to be able to work in the mean-time. Davide has been very helpful. Need quick fixes but do need a well documented tool.
@dkhutch is your work related to the request that Dietmar had for N48 land-sea mask? It’s related. Have had some discussions. No entirely sure. He is using rose-cylc, some of the interpolation steps are being auto-generated, and might be going wrong. Will give him the routine that worked and see if that will work for his problem.
@tiloz want to make sure people stay connected. Would not be a good outcome to let it drop. Small changes work, larger scale falls over. Similar to my experience.
@dhb599 reconfiguring models is critical part of workflow. Looking forward to this work. Want to change CM2 and easily land-sea mask and change related. Want to see that script working for everyone, not just specific case.
@tiloz want flexibility to support multiple resolutions and configs. @dkhutch great to know others are interested.
Issues with different run systems
@LaurieM followed issues on forum. Due to different systems (script/payu/rose-cylc) there are issues with information exchange. We want to run ssp2.6 run. Currently not possible with forcing. Could run 8.5 no problem. 2.6 has a problem.
Emissions driven scenarios
@tiloz: strong focus on emissions driven simulations in CMIP7. Trying to prepare for that through new international multi-model experiment. Some experiments we can run with ACCESS CMIP6 version already. The CMIP6 DECK includes the 1% concentration driven CO2 increase run which is the basis for estimating TCRE and for other experiments around ZEC and reversibility as well. We need equivalent emissions driven run with constant emissions (e.g. 10 PgC/yr). This will become part of CMIP7 and will allow for a better estimation of TCRE and basis for ZEC and reversibility assessment without switching between concentration and emissions driven mode. Currently running ACCESS-ESM1.5 with constant emissions over 100 year period and then switch to zero emissions. Happy to share results. (Thanks to @atteggiani for helping to generate the required CO2 emissions ancillary files)
This is a wiki post , if you want to update any of the details in these meeting notes, or to add your own recollection of what was discussed then edit me rather than replying.