Oral C3 Al, Automation & Optimization

4:00 PM - 4:15 PM JST(7:00 AM - 7:15 AM UTC)

[GW9BIT7e-01] How ocean reanalysis are used to train Machine Learning Emulators

Yann Drillet (1. Mercator Ocean International (France))

Mercator Ocean International operates the Copernicus Marine Service and develops some of the world’s most comprehensive ocean reanalysis systems, providing consistent reconstructions of the past ocean state by combining numerical modelling, observations, and data assimilation. These long-term, quality-controlled datasets include the global physical ocean reanalysis GLORYS, the global biogeochemical reanalysis BIORYS, the global wave reanalysis WAVERYS, the regional Iberia–Biscay–Ireland physical and biogeochemical reanalysis IBIRYS, and the global mid-trophic ecosystem reanalysis MICRORYS. Together, these products describe the evolution of the ocean circulation, temperature, salinity, sea level, biogeochemistry, waves, and marine ecosystems over several decades, providing an unparalleled digital representation of the ocean system. Beyond their traditional applications in ocean monitoring, climate assessment, marine ecosystem studies, and operational forecasting, these reanalyses constitute a unique foundation for the development of machine learning (ML) and artificial intelligence (AI) models. Their global coverage, long temporal extent, physical consistency, and multidimensional nature make them ideal training datasets for data-driven ocean prediction systems. Within the EDITO ecosystem, several AI models available through the AI Gallery are already leveraging Copernicus Marine datasets to demonstrate new approaches for ocean forecasting. Current developments include emulators of the global and regional physical forecasting system, emulator of the biogeochemistry and micronekton. These approaches aim to significantly reduce computational costs while preserving the predictive skill of physics-based models, opening new opportunities for rapid scenario exploration, uncertainty quantification, data assimilation, and Digital Twin Ocean applications. By combining decades of Copernicus Marine reanalyses with advanced AI methods, Mercator Ocean is building a new generation of hybrid ocean prediction systems where numerical modelling and machine learning work together to enhance ocean monitoring, forecasting, and decision support for marine services