Title: Elucidating drivers of Southern Ocean circulation change: A blueprint for interpretable and explainable machine learning

Abstract: The global ocean is central to the health of the planet, but open questions about what drives the circulation hinder our understanding and ability to monitor changes even in key regions undergoing rapid change. Climate models suggest that the circulation is changing, but the physical drivers are poorly constrained. Here, machine learning is used to both construct hypotheses to gain new theoretical understanding of the circulation and to design a monitoring framework and assess sensitivity to climate change. Focusing on the Southern Ocean (surrounding Antarctica), we use a machine learning guided objective leading order analysis to determine dynamical regimes as unique balances between driving terms. Challenging the conventional framework within which basin scale upwelling processes are understood, we propose a novel theory of scaffolding by bathymetry. Upwelling is important, as it governs outgassing of CO2, biological activity and heat available to melt ice. Yet, it remains poorly sampled and difficult to represent in models due to the extreme physical inaccessibility and dynamics resisting classical means of analysis.