NRI Antarctic ice shelf development minutes

COSIMA TWG
1

Feb 7th /2025 minutes
Present: @cbull @Wilton_Aguiar @claireyung @fabiobdias @adele157 @helen @aekiss @angus-g @AndyHoggANU
Not present: @ezhilsabareesh8

Four main topics were discussed in this last meeting. Here is a summary of them.

Pole displacement:
-Claire is working on displacing the grid’s South Pole and extending the grid using the ocean_model_grid_generator
- Issue: One issue has been avoiding discontinuities in dx, dy, and aspect ratio when displacing the pole. With the currently generated grid, it seems that there are discontinuities in dy and aspect ratio in the Ross and Weddell Seas, while the borders of the displaced pole over land don’t have the same issue.
- Next steps: We decided to move on and just use an extended grid with no South Pole displacement. If numerical instabilities arise, it might be worth trying to refine the Weddell/Ross dy to solve the discontinuity issue in away that allows the South Pole displacement to significantly increase grounding line grid size.

Ice shelf thickness product to use:

  • Issue: Currently we are using GEBCO2024, which contains the old BedMachine v2 for ice thickness.
  • Two new ice thickness products could be used instead: BedMachine v3, and AntGG2021. It would be nice to use these newer products, however they are not stitched into the GEBCO bathymetry. That means using the latest products would require us to reproduce the stitching method used in GEBCO2024.
  • Next steps: Evaluate if the newer products represent a significant improvement relative to GEBCO2024/BedMachine v2. Also, contact the GEBCO group to see if they are working on releasing a new product that incorporates BedMachine v3: this would allow us to eliminate the effort of reproducing their stitching method.

Initial conditions interpolation:

  • Issue 1: There is some leakage of the TS properties from Amundsen to the Weddell sea.
  • Issue 2: Currently, we use offshore WOA data to interpolate the data into the ice shelf cavities. This offshore data is likely too warm and could create spurious melting if used as initial conditions.
  • Steps to solve issue 1: Play around with the current interpolation tools, and try to figure out a way method that properly separates the Amundsen and Weddell shelves. @Wilton_Aguiar 's task
  • Steps to solve issue 2: Test other observational products containing data closer to the ice shelf, such as Schmidko, Pauthenet and Kaihe’s Climatology.

Vertical coordinate choice:
- Currently, we are debating between using SIGMA_SHELF_ZSTAR and ZSTAR.
- Issue: SIGMA_SHELF_ZSTAR have strong discontinuities along the ice shelf front. ZSTAR don’t have these discontinuities, but it doesn’t continuously represent the ice shelf ocean boundary layer. A more ideal coordinate would combine the best parts of each, but requires mapping where the ice shelf caving front is.
- Steps to solve: Being debated on GH

I probably missed a lot here, so please edit to add any missing info.

3 Likes
2

March 12th /2025 minutes
Present: @cbull @Wilton_Aguiar @claireyung @fabiobdias @adele157 @helen @aekiss @ezhilsabareesh8
Not present: @angus-g @AndyHoggANU

The discussion focused on three main topics (mostly updates):

Ice shelf thickness product to use:

· Background: It seems that the ice sheet modelling community widely agrees that AntGG2021 (Charrassin et al., 2025) provides the best representation of ice shelf cavities, with BedMachine v3 being the second-best. The general consensus is that we should avoid using BedMachine v2.

· Issue: GEBCO2024 is currently being used for the topography and incorporates BedMachine v2. Using either AntGG2021 or BedMachine v3 will require significant testing to ensure that the ice shelves are properly connected/stitched to the bathymetry. Waiting for the stitched product to be ready could slow down the workflow for producing model bathymetry, ice shelf thicknesses, and initial conditions.

· Next steps: Continue the model workflow using the current GEBCO2024, while the proper ice shelf thickness/bathymetry dataset is being developed. This decision will allow us to advance the workflows and model development, providing more time for thorough debugging.

· PS: We will likely revisit this discussion at the next meeting, where we aim to get input from other ice sheet modelers

Extrapolation of Initial conditions:

· Background: In MOM6-Panan, we used a backward filling method to propagate offshore temperature-salinity (TS) fields southward, where ice shelf cavities would be. This method undesirably leaked properties from the Amundsen Sea into the Weddell Sea cavities. Additionally, previous insights from ocean-ice sheet modeling with ROMS suggest that the sharp density gradients created by backward filling can be a challenge for the ocean model. We attempted to extrapolate the surface offshore TS using Inverse Radius Weighting (IRW) interpolation instead, which seems to fix the leakage problem, and reduce sharp gradients.

· Issue: While this method resolves the leakage issue, it does not fix the warmer surface conditions near the Amery Ice Shelf. Furthermore, the evaluation of the IRW propagation is not ideal, as ice shelf cavities are much deeper ( z~ 1000m).

· Next steps:

  • Evaluate how the IRW extrapolation performs at the depth of ice shelf cavities (~1000 m?).
  • Compare the TS on the shelf with previous ocean-ice shelf model outputs (e.g., are the warmer waters in Amery reasonable).
  • Test other observational products containing data closer to the ice shelf (~1000 m), such as Pauthenet and Kaihe’s climatology.

Target horizontal grid:

· Issue: We do not yet have the target horizontal grid (with the southern boundary extended to 86.5°S) that will be used for the model testing phase or the final version of the model. Currently, the workflows are using grids from Panan 1/10°, ACCESS 1/10°, or ACCESS 1/4°.

· Next steps: We have decided to start with a 1/12° grid for testing, which is still computationally inexpensive for testing in a regional, Pan-Antarctic domain. @angus-g volunteered to create a global 1/12° grid that we can cut to the desired domain. Once testing is complete, we will transition to the final 1/20° (or 1/24°) grid.

There were no updates on the vertical coordinate choice.