COSIMA TWG Meeting Minutes 2024

COSIMA TWG
12

Summary of today’s TWG - I didn’t catch everything so please edit to add/correct as needed.

Date: 2024-07-17

Attendees:

DS:

  • Model runs with working group on Gadi
    - The ZV30 project, which is for CMIP7 evaluation, is the right place for data storage, including WOMBAT runs and 0.25deg OM3 testings that ML set up. Currently there is only around 6GB out of a total of 50TB allocated to the ZV30 project, which includes both 0.25-deg OM3 and WOMBAT relevant to CMIP7.
    - AM: No instructions were provided regarding the usage of OL01 storage. There might be a chance to request additional storage if it is used up.

KR:

  • CICE cap initially runs at the same timestep as the coupling timestep. Now, it can run multiple steps per coupling timestep, but currently has the averaging issues.
  • Energy conservation is reasonable, but the ice fractions appear strange, possibly due to improper averaging of ice fluxes before sending them to the ocean.
  • Averaging occurs in the cap before sending data across, rather than in the mediator.
  • DS: Changes have been made to a ~6 month old fork of CICE, well need to harmonise at some point.
  • AS: Current OM3 - CICE version is between 6.5 and 6.5.1, with no significant scientific changes but infrastructure improvements.

DS:

  • Default parameters
    • Every year, MOM developers review and decide on changes to default parameter. And there are a bunch of parameters to change. Our approach of removing default parameters from MOM_input could be risky. Some parameters we rely on as defaults may change without our awareness, potentially causing unexpected results. This is another reason to automatically generate MOM_parameter_doc.* and track them with git so we can detect changes.

ML:

  • Created a preliminary list of parameter changes and identified several scientific tests to start:

      1. tracer timestep: 1,2,4,6,8 baroclinic timestep
      1. mesoscale parameterisation: (1) GM only, (2) MEKE, (3) MEKE + Geometric scaling (Hallberg) probably directly to this instead of MEKE only
      1. Submesoscale parameterisation: (1) Fox-Kemper et al. (2010), (2) Bodner et al. (2023)
      1. Hybrid grid: ZSTAR —> HYCOM1 (Hybrid vertical coords) ?
      1. Lateral friction: (1) isotropic + Biharmonic, (2) via MEKE
      1. Vertical mixing: (1) CVMix - KPP, (2) ePBL+(Langmuir turbulence?)

  • Starting with tracer timestep tests as they significantly improve performance. No specific preference for other parameters; all will be tested.

    • Changes will affect tracer-related diagnostics, including temperature and salinity, such as,
      1. Zonal average temperature and salinity (i.e. depth/latitude maps) (Fig. 12 Kiss et al. 2020) [1993 - 2017]
      2. Time series of global average temperature, salinity and sea surface temperature. (Fig. 3 in Kiss et al. 2020), and sea surface height.
      3. Zonally integrated overturning in density / latitude space (Fig. 7 Kiss et al. 2020) [1993 - 2017]
      4. Time series of Drake Passage zonal transport. (Fig. 4 in Kiss et al. 2020)

  • Comments from Attendees:

    • AK: emphasized the importance of having efficient and stable 1° and 0.25° configs available as soon as we can for CMIP7 development. Suggested starting with well-understood parameters based on OM2 to be used as initial config for CMIP7. We can then refine OM3, starting with refinements that may be valuable for adoption into CM3/ESM3 in time for CMIP7 deadline.
    • AK/AM: suggested starting with extreme parameter values (eg for tracer timestep) to identify the negative consequences to monitor.
    • AM: noted that regional models showed weird behaviors with larger tracer timesteps, not sure about global models.
    • AS: suggested to ask GFDL or NCAR for similar tuning processes.

  • Hybrid coordinates

    • SO: Hybrid coordinates are still experimental for us.
    • AK: Mentioned that CESM3 will use hybrid coordinates.
    • AH: Advocates for z^* if rigorous testing isn’t feasible. Plans to test AG adaptive vertical coordinate but can’t guarantee a better solution at the moment.

  • Parameter change process/protocols

    • It is necessary to work out a process, such as automatically generating the intake catalog to make it easy for community. Aim to simplify access for external users and promote participation through customer meetings.

AK:

  • CMIP7 deliverables
    • working underway on optimising OM3 at 1deg and 0.25deg
    • working underway on tune parameter testings on 1deg and 0.25deg
    • working underway on testing 1deg and 0.25deg with WOMBAT-lite
    • need to determine the speed before testing 1-degree and 0.25-degree models with WOMBAT-mid.
      • DS: WOMBAT-mid is still under development.
      • SO: For the 0.25deg, it will be very expensive with BGC because we have to do pre-industrial controls, pre-emission controls and etc.
    • consider performing offline BGC runs to accelerate parameter tests and spinup. Unclear how much development would be needed for offline BGC - should scope it out before committing, but would be great to have if not infeasibly difficult.

EK

  • Grid update
    • We are now able to generate a grid that closely matches the OM2 grid.
    • Changes we want to make in the new grid (copied from New grids · Issue #172 · COSIMA/access-om3 · GitHub)
        1. leave tripole points where they are
        1. leave longitudinal seam as-is
        1. explore putting C-grid zonal points exactly on the equator
        1. don’t extend 1° or 0.25° closer to the pole, since they won’t have ice cavities, but extend to grounding line to support ice cavities at 0.1° and higher, using a displaced pole (this can be done with ocean_grid_generator.py but can wait until after we’ve handed over 1° and 0.25° OM3 configs for CMIP7)
        1. Refine in the Antarctic by extending the Mercator region to 75°S and then extending with constant dx to the southern edge of the grid. No coarsening in the Arctic.
        1. Quantize double-precision to be exactly representable in single precision
        1. use GEBCO2023 topography
        1. make sure grid allows MOM6 output at half-resolution