COSIMA TWG Meeting Minutes 2026

COSIMA TWG
, ,
22

Date: 21 Oct 2026

Attendees: @AndyHoggANU, @aekiss, @MartinDix, @cbull, @dougiesquire, @anton @ezhilsabareesh8 @NoahDay

Minutes: @minghangli

Setting MAXTRUNC

  • @anton reported that the current MAX_TRUNC value is 10000 in most ACCESS-OM3 configurations unless a few 100km configurations. But this is too permissive for routine production.
  • After the meeting, we agreed that truncation should normally be an unusual occurrence, not a routine mechanism replied on to keep an unstable config running.
  • After the meeting, the preferred threshold was 500, which would detect numerical problem earlier but still allowing modest initial transients in most of the configs.

Fast ice configuration

  • The main scientific interest is Antarctic fast ice.
  • Two candidate parameterisations
    • Lateral stress: Dan has developed a parameterisation representing the resistance exerted on sea ice by coastlines and ground icebergs.
    • Seabed stress: CICE already includes a seabed-related stress parameterisation potentially relevant for grounded sea ice and Arctic conditions. But we are not certain how this would interact with the proposed lateral stress when both are enabled.
  • The preferred direction is to focus on Dan’s lateral stress rather than introducing the seabed stress at the same time
  • Dan’s implementation is reported to work in standalone CICE so the work needed for ACCESS-OM3 includes integrating the code changes into ACCESS CICE branch and if suitable proposing the change upstream.
  • A new input dataset is also required - form factors, which would describe the effective lateral drag exerted on seaice by nearby coastlines and grounded icebergs.
  • Kaihong has performed some coupled model experiments which should be examined.
  • Aim to trial it in an upcoming 25km experiment - potentially TEST6?

MOM6 surface pressure issues

@dougiesquire raised a technical issue concerning how the MOM6 nuopc cap passes surface pressure into MOM6.

  • The current ACCESS-OM3 coupling interface passes absolute atmospheric sea-level pressure (SLP) from JRA55/ERA5 directly to MOM6. But MOM6 expects atmospheric surface pressure to be expressed relative to a reference pressure, rather than as absolute atmospheric pressure.

    • In GFDL SIS2 implementation, the atmospheric contribution is optional (APPLY_SLP_TO_OCEAN, off by default and in OM5) and, when included, is expressed relative to 1e5 Pa (approximately one atmospheric pressure). P’_atm = P_atm,abs - P_ref
    • Sea-ice loading is a separate contribution caused by the weight of ice and snow, P_ice = m_{ice+snow} * g
    • So a surface pressure representation that includes both effects would be, P_surf = P’_atm + P_ice. And whether ice loading should directly influence ocean dynamics depends on the chosen treatment of embedded vs levitating sea ice.

  • In GFDL MOM6+SIS2, MOM6 can use this pressure to estimate seaice rigidity because the loading represents physical ice mass rather than the absoluate pressure.

  • In ACCESS-OM3 CDEPS + CMEPS + NUOPC cap, the pathway is different,

  • Our current config enable,

     USE_RIGID_SEA_ICE = True
 DYNAMIC_SURFACE_PRESSURE = True

    where in the nuopc driver, the ice mass used to calculate rigidity is inferred from p_surf_full but since ACCESS-OM3 supplies absolute SLP, this will introduce ~10m of sea ice even in ice-free ocean cells. So MOM6 incorrectly applies sea-ice rigidity across the entire ocean.

  • At the same time, absolute SLP also introduces around 100kPa offset wherever MOM6 expexcts sea pressure or a pressure anomaly, which affects EOS, SSH, frazil freezing and mixing parameterisations.

  • GFDL OM5 they use,

    MAX_P_SURF = 0.0
USE_RIGID_SEA_ICE = True

    This is consistent with SIS2 pressure, but in our models, p_surf_full still contains absolute SLP so setting MAX_P_SURF = 0.0 does not fix the incorrect rigidity.

  • The current workaround after the meeting is to set MAX_P_SURF = 0 while the underlying issue is addressed Surface pressure in MOM6 · Issue #1826 · ACCESS-NRI/access-om3-configs · GitHub. Also set USE_RIGID_SEA_ICE = False to remove the unintended global rigidity damping.

  • The changes would avoid the principal incorrect pressure effects but they are not a complete physical correction at the moment, such as sea-ice rigidity would be disabled rather than calculated correctly, and frazil freezing will still receive the incorrect pressure.

MCW stability issues

  • The baseline config develops an Arctic instability and eventually fails after ~10years with a negative sea ice area error. The failure is located near the Queen Maud Gulf / Canadian Arctic Archipelago region. The instability is most evident around narrow channels and small grid cells.
  • One leading explanation from @anton is the delayed feedback created by the 1hour coupling interval, where CICE computes an ice-ocean stress using velocities and other state info supplied at coupling exchanges. That stress is then used across the subsequent coupling interval. But the ocean can change substantially during the hour while the coupled stress reflects an older state. While the meeting did not establish conclusively that the coupling delay is the unique root cause.
  • @ezhilsabareesh8 had conducted two experiments, starting from a warm restart close to the onset of the known instability. Both can suppress the observed instability over the reported period (1965 - 1980)
    • Reduce coupling step from 3600s to 1800s, while the performance cost can increase up to 74%. With a subsequent load balancing work, the additional cost can be reduced to 37%.
    • Retain 3600s coupling timestep and enable MOM6 DIRECT_STRESS, which distributes the applied momentum forcing over approximately the upper 5m of the ocean. But this change modifies how the stress is applied vertically in MOM6. Rather than relying exclusively on the usual surface-flux pathway, the stress is distributed as a body force through a prescribed upper layer. Unlike the 1st approach, this does not show a similar large computatioanl penalty.
    • Considering the science side effects, we decided to follow reducing the timestep option.
      • @minghangli proposed the problem could be alleviated by shortening the MOM6 dynamic timestep while retaining the 3600s coupling timestep. This was motivated by the possibility that the ocean CFL constraints and numerical integration might be contributing to the truncations. @ezhilsabareesh8 will do a test using DT=1800s that would help distinguish an internal ocean timestep limitation from an instability primarily driven by delayed coupled forcing.
  • Increasing ndtd from 1 to 2 does not resolve the issue - the relevant test still crashed.
  • For MCW config release, the objective for the release is a config that can start from the intended 1958 initial state and run through 1 cycle. Hence dont make the entire release dependent on finding the theoretically perfect combination first but start a full integration using a shorter coupling interval (i.e., more conservative approach cuz it does not introduce difference to vertical momentum forcing) and use that run to establish a realistic release candidate.
  • MCW beta release documentation
    • @ezhilsabareesh8 has prepared a release checklist and a draft description is expected to be circulated for review. But the remaining blocker is - we still needs a stable scientifically acceptable long-run configuration to support the release.

PanAnt update

@AndyHoggANU gave an update on the Panan/iceshelf configuration testing.

  • The test had not been particularly sucessful in eliminating the targeted biases. And the config may be diverging further from the main ACCESS development branches.
  • The runtime performance is still not satisfactory.
  • @dougiesquire was expected to discuss branch synchronisation with Claire and Angus later in the week.