Regional modelling

Copying the google doc here so we can continue the discussion started at the COSIMA2022 workshop.

Starter questions:

  1. Pain points with MOM6 regional - what is the most important documentation to get done first?
  2. How to get the most balanced open boundary forcing and initial conditions?
  3. Recipes for setting up regional models
  4. Tides? Bathymetry? Esp. around Antarctic
  5. How can we learn from other regional systems and communities (eg MITGcm & ROMS)?
  6. What are the priority regions?
  7. What synergies are there between research interests/projects? Is there scope for piggybacking multiple projects on the same model runs?
  8. How best to prepare for coupled (ocean-atmosphere) regional modelling?
  9. The MOM6 regional simulations look pretty good … but there are a couple of surprising results and we are not certain that the OBCs are performing as they should. Especially, we seem to get differences with prescribed tidal forcing … Can we do a more in-depth evaluation of how well these OBCs are working?
  10. Is there scope for data assimilation? If so, how can IMOS help to make this happen?
  11. How can ACCESS-NRI help with the above?
  12. What about CICE6 for the regional modelling?

Model configuration

  • should we settle on a single model?
    • How do we meet the needs of a large, varied community with an infinite variety of of use cases;
    • Boundary inputs need to be fit for purpose (frequency, .
    • We need a set of well-described “configurations” that we support - then people can tailor these to their own needs.

Single model or multiple models?

MOM6 has existing user and knowledge base in Australia - has a lot of what community wants in regional model. Contrasted to ROMS

What about forcing? Can we generalise this to make it easy for users to switch out different forcing fields?

Application of forcing will be model specific.

Looking for a “one-touch” solution - opens doors to many applications

Need to develop understanding of what different forcing products can be used to force the model. Can experiences are that can get unexpected behaviour with specific products.

Chris mentioned Herzfeld et al. 2008 paper that provides rules of thumb for driving regional models: every case is different.

Current configs all within GFDL suite. Will be moving from MOM+SIS to MOM+CICE. Models have different requirements and there is need for a coupler. Could complicate things.

One reason for MITgcm success is it’s very large number of well-documented examples on website - could we do this - ie instead of developing a shared repo of code, develop shared repo of examples.

This is sort of the way things are developing at the moment. Ashley Barnes has put together a “cookie cutter” example notebook

There are lot’s of different subgroups within this group - scope for knowledge sharing, extending cookbook. Also being clear about plans for next generation of models is important. Many things to plans/consider: e.g. what boundary condition, tides

Ashley has tried adding the GFDL tidal frequencies at the boundary in his notebook.

How precise does one need to be about specifying boundaries? We think don’t need to be that precise.

There is a lot of sensitivity for where you put your boundaries

What’s really important is to set up a framework for people to assess how good their regional model.

Performance benchmarking and profiling - help from NCI? - scaling, I/O cost of boundary forcing, load balancing when coupled - NCI would like a timeline and idea of workload required