From numerical models to AI-assisted seasonal forecasts, BG4Sea demonstrates Mercator Ocean’s cutting-edge approach to marine ecosystem prediction.
Marine ecosystems are under increasing pressure from climate change, affecting the global carbon cycle, ocean productivity, biodiversity, and human livelihoods. Anticipating these changes is essential for monitoring, sustainable management, and climate-informed decision-making.
At Mercator Ocean International, this challenge is addressed through state-of-the-art numerical models, complemented by a comprehensive AI strategy spanning physics and biogeochemistry, developed jointly to serve multiple forecasting horizons and scientific objectives.
“Marine ecosystems are at the heart of our mission. Consolidating and scaling up our work beyond ocean physics to the full scope of marine ecosystems is essential, both for science and for policy. By embracing a broader vision of the ocean and leveraging AI, digital innovation, and strategic partnerships, we can better understand and predict ecosystem changes, supporting sustainable management and informed decision-making.” — Stefano Ciavatta, Head of Marine Ecosystems, Mercator Ocean International.
From numerical models to AI-based seasonal forecasts
Mercator Ocean relies on high-resolution numerical models, including the coupled NEMO/PISCES framework, which simulates ocean physics and biogeochemistry. These models support forecasts ranging from short-term operational products to climate projections.
Running high-resolution biogeochemical models over long-time horizons is computationally demanding, limiting routine seasonal forecasting.
BG4Sea addresses this challenge by using machine learning (ML) to emulate BIO4, Mercator Ocean’s global biogeochemical forecasting system. Rather than replacing numerical models, BG4Sea reproduces their outputs at a fraction of the computational cost, enabling:
- Seasonal-scale biogeochemical forecasts (beyond the typical 10-day horizon)
- Competitive forecast skill with limited degradation over longer lead times
- Flexibility in spatial resolution, from global to regional applications
BG4Sea in practice
Developed by PhD researcher Gabriela Martinez Balbontin, BG4Sea learns from Mercator Ocean’s physical (GLORYS12) and biogeochemical model (BIO4) to produce seasonal forecasts at a fraction of the usual computational cost. Its flexible design allows forecasts at different geographic scales, from global to regional, and could be enhanced in the future by incorporating real-world observations.
” Developing BG4Sea has been an exciting opportunity to explore how AI can extend our ability to predict marine biogeochemistry at seasonal timescales. By learning from trusted numerical models, we can produce faster, flexible forecasts that support monitoring and management, while staying grounded in rigorous science” said Gabriela.
Recognition from the AI for Science community
BG4Sea was recently recognised with a poster award from the NeurIPS@Cam community at the AI for Science Summit, reflecting growing interest from the AI research community in robust, impact-driven ocean applications. The award highlights both the scientific relevance of seasonal marine ecosystem prediction and the importance of developing AI approaches that remain closely linked to trusted numerical models.
BG4Sea within Mercator Ocean’s multi-domain AI capability
BG4Sea is a key element of Mercator Ocean’s comprehensive AI strategy, which spans physics, biogeochemistry, and multiple forecasting horizons:
- AI for Physics – GLONET: emulates physical ocean models for short-term forecasts, benchmarked against operational systems like GLO12.
- AI for Biogeochemistry – BG4Sea: targets biogeochemical processes and seasonal prediction, emulating BIO4 rather than physical models.
- Validation framework – OceanBench: currently benchmarks AI models for short-term physical forecasts; soon expanding to biogeochemistry, regional applications, and longer horizons.
- Digital infrastructure – EU Digital Twin Ocean: provides the computational and operational framework to develop, test, and deploy AI-driven ocean forecasting at scale.
BG4Sea’s award underscores its role as a bright element of this AI ecosystem, illustrating how AI complements numerical models to deliver faster, extended, and scientifically robust forecasts.
Looking ahead
BG4Sea demonstrates how AI can extend marine ecosystem forecasting into new time horizons, complementing numerical models to deliver faster, more efficient, and scientifically rigorous forecasts.
This approach supports ocean monitoring, environmental management, and climate-informed decision-making, while illustrating Mercator Ocean’s coherent AI strategy.
Together with GLONET and OceanBench, BG4Sea highlights the organization’s commitment to advancing operational oceanography through a science-based, multi-domain AI capability.
“Mercator Ocean is developing an integrated, science-based multi-domain AI capability that complements numerical ocean models to deliver trustworthy ocean forecasts across physics, biogeochemistry, and multiple time horizons, at global scale,” concludes Yann Drillet, Head of Research & Development department.