ESM working group: Meeting notes 2024

Earth System Working Group
, ,
46

Date: 25/09/2025
Chair: @dkhutch
Participants: 16

1. Admin and resource usage

  • Science Presentations: There are currently no science presentations scheduled. If you have any work you would be interested in presenting, contact any of the ESM WG co-chairs or @spencerwong. Presentations from recent conferences and workshops are welcome, as are informal updates on in progress work. Students and early career researchers are especially encouraged to give presentations.

  • Change of working group co-chairs: @eunpalim will be completing her role as a ESM WG co-chair and the position will be taken up by @ctychung.

  • LG87 Resource Usage:

    • The ESM WG compute allocation was increased for quarter 2 to 1.3MSU. This has been used up, and the allocation will be reset to the default 875KSU on 1 Oct.
    • If you have any experiments you are interested in running with ESM WG resources, follow the proposal guidelines here to apply to use WG compute. Members are encouraged to apply for resources early in the quarter, as there may be opportunities to obtain more resources later in the quarter.
    • Storage usage at 65 TB of 100 TB allocation. In the future, we may require long term planning to manage the storage resources.

2. Proposed ESM1.5 infrastructure changes

  • @spencerwong and @manodeep discussed recent work on optimising ESM1.5 and 1.6.

  • Infrastructure updates to ESM1.6 include compiler changes, queue and layout changes, and updates to communication libraries. These changes improved ESM1.6’s walltime and running cost.

  • Back-porting the updates to ESM1.5 would reduce walltimes for a year of simulation from 65.9 to 58.4 minutes (21.9 years/day → 24.7 years/day) and service unit cost from 950 to 810 SU/year.

  • These infrastructure updates would change answers produced by the model. These have not been specifically tested with ESM1.5, however tests using ESM1.6 suggest that answer changes fall within the model’s internal variability.

  • Model determinism is not impacted: repeating a simulation multiple times with the new version will still produce identical answers.

  • Discussion:

  • Members were interested in seeing these changes added to a new release, as long as previous releases remain available for continuing older experiments.

  • Members were interested in further opportunities for optimisation, including vectorisation which may be able to improve performance. This would lead to further numerical changes, however should still maintain determinism. Scaling studies are also being worked on to assess how changing the number of CPUs affects model speeds.

  • Producing multiple versions with different optimisations, with different levels of reproducibility were also discussed.

  • Switching off the carbon cycle may be an additional way to improve performance.

Discussion on collaborative CM3 evaluation

  • @wghuneke presented on a proposed framework for CM3 evaluation.

  • The long term goal is to produce a peer reviewed evaluation publication, with an interim goal of helping with model development and tuning. Evaluation work on CM3 can then be leveraged in the future for CM3.

  • The aim is to work collaboratively, where everyone is welcome to contribute to the evaluation. A GitHub repository will be set up where members can contribute analysis notebooks create GitHub issues for discussing different diagnostics. ESM Working Group meetings can also be used to discuss progress and the analysis.

  • Aiming to hold a hackathon around CM3 evaluation next year.

  • Existing OM3 analysis scripts will be a useful starting point.

  • Technical points around data availability, standardisation, and CMORisation were discussed.

  • There will be a meeting to plan the evaluation framework on Zoom on Wed 8 Oct at 11:30. Anyone is welcome to join – meeting details are available in this topic

Additional information

Please feel free to correct any mistakes in these notes directly, or message @spencerwong with corrections.

1 Like