Researchers at the GEOMAR Helmholtz Centre for Ocean Research Kiel and Kiel University have shown that artificial intelligence, paired with data from the global Argo float programme, can track key characteristics of the Atlantic Meridional Overturning Circulation (AMOC) - a vast current system that moves heat around the planet and directly shapes Europe's climate. The study, published July 31, 2026, in the journal Ocean Science, introduces a method that makes it possible to infer large-scale circulation patterns from scattered point measurements, reducing reliance on the handful of fixed, costly monitoring stations that have anchored AMOC observation for decades.
The international Argo programme has operated for more than twenty years, with roughly four thousand autonomous profiling floats drifting through the world's oceans. Each float periodically descends to 2,000 metres and then rises, recording temperature, salinity and pressure along the way before surfacing to transmit data via satellite. The resulting dataset is freely available to researchers worldwide and has become one of the most consequential observation networks in ocean science.
Combining machine learning and physical models
"We wanted to know whether we could use the scattered Argo measurements to gain insights into large-scale circulation systems that, until now, could only be recorded through very extensive measurement campaigns," said Dr Yannick Wölker, lead author of the study and until recently a PhD student in the Ocean Dynamics research unit at GEOMAR and the Archaeoinformatics - Data Science group at Kiel University. "Artificial intelligence opens up new possibilities here. Combining machine learning with established physical models allows us to get more out of existing measurement data and better understand how key circulation systems work."
The AMOC functions like a conveyor belt: warm surface water travels north, cools, sinks to great depths, and returns south as cold deep water. This movement transports enormous quantities of heat and drives weather and climate patterns, including those affecting Europe. How stable the AMOC remains under climate change is one of the most urgent questions in climate research, yet direct observation has depended on only a few fixed measurement arrays in the Atlantic - all of them technically demanding and expensive to maintain.
Closing the observation gap
To address this gap, the research team combined observational data with model calculations. They trained a machine learning algorithm on high-resolution ocean simulations, teaching it to recognise how typical ocean current patterns relate to temperature and salinity profiles. The trained model was then applied to real Argo data from the Atlantic, allowing the team to estimate large-scale current strengths from individual point measurements. In particular, the method captures the geostrophic component of the circulation - governed by temperature and salinity distribution - which has been persistently difficult to measure on a continuous basis.
The resulting estimates align well with established observational series and model simulations. Such applications of AI for Science & Research are becoming more practical as computational methods mature and observational datasets grow in size and quality. The work was carried out through the Helmholtz School for Marine Data Science (MarDATA), a doctoral programme in which PhD students receive joint supervision from marine scientists and computer scientists.
A complement, not a replacement
The authors are clear about the method's limits. It depends on model assumptions and captures only specific time windows, meaning very short-term fluctuations and the AMOC's long-term decline can only be tracked to a limited degree. The new methodology is designed to strengthen existing measurement systems rather than replace them.
"Our approach is no substitute for direct measurements in the ocean," said Wölker. "But it can help to make better use of existing data and bridge gaps in observations."
The method could influence how future observation networks are designed. It demonstrates that global programmes such as Argo can yield even greater value through modern data analysis, and it suggests ways to deploy additional ocean infrastructure more strategically. The full study is available in Ocean Science.
Why this matters for science and research professionals
For researchers managing long-term environmental monitoring programmes, this study offers a practical template: machine learning can extract information about large-scale systems from data that is already being collected, without requiring new hardware or additional field campaigns. The approach does not eliminate the need for direct measurement, but it can fill temporal and spatial gaps in observation records. For teams facing constrained budgets and rising equipment costs, methods that increase the utility of existing infrastructure are worth close attention. The paper also reinforces the value of cross-disciplinary training - the MarDATA programme's structure, pairing marine scientists with computer scientists, produced a result that neither discipline would have reached alone.
Your membership also unlocks:
