Modelling the ocean in the Earth system has undergone significant improvements over the years. Nevertheless, many challenges remain, ranging from uncertainties in carbon and heat transport to accurate simulations of large-scale ocean circulation. Addressing these issues is essential for reliable long-term climate projections and for improving models used for early warning systems. Machine learning offers new opportunities by complementing traditional physical-based approaches, enhancing predictive skill, and accelerating simulations. Further opportunities lie in developing physical predictions of the ocean system based entirely on AI/ML methods.

The merging of these two fields brings with it new challenges. Some of these range from: developing and choosing an appropriate set of training data; agreement on the benchmarking process; ensuring numerical stability and physical consistency in new parameterisations.

Our objective in this session is to bring together researchers across machine learning and ocean model development to synthesise the community's progress and foster new ideas.

This session invites contributions from any of the below points (examples are included but are not limited to):

- Physics emulation: development of ocean parameterisations including their methods and architectures, offline and online performance in idealised and realistic (global or coupled) simulations, AI/ML only models, ocean forecasting.

- Training data: development and availability of new fit-for-purpose datasets, preprocessing techniques, accounting for out of distribution events such as marine heatwaves or system collapse.

- Benchmarking: defining a framework, establishing evaluation datasets and metrics, new tools.

- Calibration and optimisation: use of ML techniques to improve model fidelity, for example through optimisation of parameters and inputs or through online bias reduction.