Breakout: Physics-informed / Hybrid Machine Learning (Day 3 12:30-1:30)
This breakout will discuss hybrid climate modelling in Austraila, including efforts underway and opportunities. Hybrid modelling can broadly include efforts to incorporate AI/ML elements into traditional plysical models, to use AI to discover physical principles (i.e. “interpretive AI”), or to incorporate physical constraints into ML (e.g. PINN). A particular emphasis will be the use of ML surrogates either to replace physical parameterizations (for speed or accuracy), or to produce new model outputs. We would like to include discussion of a recent proposal to enable the creation of training data needed to develop ML physics surrogates: see https://forum.access-hive.org.au/t/experiment-proposal-coarsening-of-the-fine-scale-atmosphere-for-cross-scale-learning/6035/4 so please have a look at this in advance if you can, and think about what might be achievable or not. However we can discuss any other topics or initiatives that attendees bring up!