COSIMA TWG Meeting Minutes 2025

COSIMA TWG
, , , , ,
9

Meeting Summary & Minutes

Date: April 24, 2025
Attendees: @cbull, @dougiesquire, @helen, @AndyHoggAnu, @ezhilabareesh8, @minghangli
Minutes: @ezhilabareesh8

  1. GFDL PR Update: (Dougie)
  • The upcoming PR from GFD includes around 30,000 lines of changes. These changes need to be reviewed to ensure they integrate well with ACCESS-OM3.
  • Marshall is revisiting the Fortran do concurrent construct, which had issues a few years ago but has since seen improvements. This construct could make GPU porting more seamless.
  • A hybrid approach is being considered, using OpenMP for memory management and do concurrent for loop parallelization. This approach aims to manage memory more efficiently by specifying which arrays need to be passed to the GPU.
  • Testing is required to ensure the changes work do not alter results, and do not impact CPU runtime (likely that it will improve runtime performance given the barotropic solver performance changes). Could be worth considering using for the beta June release.
  1. Restart Reproducibility Issues: (Dougie)
  • Challenges with restart reproducibility were discussed, particularly with the LEITH parameterization and the use of the use_leithy parameter, which turns on harmonic backscatter.
  • The combination of LEITH_AH = True and USE_LEITHY = False breaks restart reproducibility. One potential solution is to keep the LEITH parameterization and turn on harmonic backscatter, which Dougie has shown fixes reproducibility.
  • Another option is to move to Smagorinsky, which also fixes the restart reproducibility issue.
  • Dougie discovered that the checksums written into the MOM restart file are not reliable for checking if answers have changed, due to differences in negative and positive zero representations.
  1. Truncation Errors: (Minghang, Ezhil and Chris)
  • Minghang Li reported ongoing truncation errors, particularly near the Kara strait, Kerguelen Plateau and West Antarctica.
  • Various tests have been conducted, including changes to atmospheric forcing and bathymetry adjustments, and using GFDL’s OM5 MOM_input parameters.
  • Reducing the time step to 900s is recommended to see if fixes the truncations.
  • The truncation errors tend to occur at high latitudes with small grid cells. The team is considering whether to tolerate some level of truncations or implement local engineering fixes.
  1. Configuration Updates and Documentation: (Chris)
  • Chris mentioned the deployment of new configuration documentation and requested no further additions to the old Wiki.
  • Plans for a more permanent URL and integration with ACCESS-NRI documentation are underway.
  • The new documentation will include general information about ACCESS-OM3 configurations, with specific configuration details provided in separate sections.
  1. GitHub backup: (Chris)
    Chris outlined strategy that the team have discussed and are planning on implementing. @AndyHoggANU suggested that it would be okay if COSIMA was included in the routine backup but is yet to be implemented by Micael, Aidan etc.