ESM working group: Meeting notes 2024

Earth System Working Group
, ,
41

Date: 05/06/2025
Chair: @spencerwong
Participants: 9

1. Admin and resource usage

  • Science Presentations: No upcoming presentations scheduled. Members encouraged contact the ESM WG co-chairs or @spencerwong to propose a topic.
  • LG87 Resource Usage:
    • 836 of 875 KSUs used this quarter.
    • Storage usage at 55.2 TB of 100 TB allocation.

2. 2025 ACCESS Community Workshop – Working Group Day and ESM WG organisation

  • Scheduled for Friday, 12 September 2025.
  • Currently unclear who is responsible for organising the ESM WG meeting at the Community Workshop. ACCESS-NRI will provide rooms and catering; content planning is up to the group. @jemmajeffree to follow up with @dkhutch on how last year’s meeting was planned, however additional people are needed to lead the planning.
  • The format for last year’s workshop, with a single track session of presentations, was seen as successful and a model to follow for this year.
  • Lack of clear message that the ESM WG day is for anyone involved in coupled system research rather than just technical model development.
  • Discussion on organisation of ESM WG, noting that doesn’t have the same level of organisation or engagement as COSIMA despite overlapping with a wide range of interests. Noted that it would be helpful to advertise the group more widely with a clearer message of who it is for.

4. NetCDF Conversion and Output Format Updates

Presented by Martin Dix

  • Proposed changes to atmospheric model output:
    • Move from stash code-based names to CMIP-style short names (e.g., ts for surface temperature).
    • Shift to single-variable files for better integration with intake catalogs and diagnostics.
  • Discussion Points:
    • Pros: Improved usability, consistency with ocean model outputs, easier data access.
    • Concerns: Increased number of files, naming inconsistencies for variables without standard names.
    • Suggestions:
      • Group related variables (e.g., carbon pools, radiation variables).
      • Maintain user-friendliness and performance balance.
      • Post detailed proposal on Hive for community feedback.

5. Science Presentation

Presenter: Dietmar Dommenget
Title: How to Train Your Dragon – Tuning for Coupled Model Development with the MPI ICON-XPP Model

Introduction:

  • Tuning is the process of finding model parameters which optimise the behaviour of the model. Tuning is often complicated, and it can be challenging to achieve the desired model improvements.
  • This work focuses on tuning the MPI ICON-XPP model to optimise it’s ENSO in an atmosphere-only configuration, aiming to show that a more systematic approach to tuning can work in a simplified setup.
    Methodology:
  • A metric based on the Planton metric is used to evaluate the ENSO performance, incorporating information on climatology and processes.
  • 30 year simulations were run while varying individual parameters, producing an estimate of error sensitivity to changes in each parameter. Sensitivities for some pairs of parameters were highly correlated.
  • Sensitivities calculated from the low resolution model were found to also extend to the higher resolution model.
  • The ENSO metric for different parameter combinations is estimated as a linear superposition of the individual parameter sensitivities.

Results

  • Optimising 21 parameters using the estimated ENSO metric substantially reduces the biases, and additionally reduces biases in the higher resolution model. Biases are still correlated with the original unoptimised parameters.
  • Applying the AMIP optimised parameters to the coupled model however was unsuccessful.

Outlook

  • Tuning in a systematic way can improve the model without increasing resolution.
  • With more resources and better optimisation schemes, better outcomes may be achievable.
  • Applying this form of tuning to a coupled model risks making model components artificially dependent.

NOTE: These minutes were partially generated using AI. Please feel free to correct any mistakes directly, or message @spencerwong with corrections.

Additional information