The story is made together with DMI.
A new interdisciplinary project, IcyAlert, aims to develop an advanced forecasting model to predict ice-free Arctic summers. The system will combine artificial intelligence, advanced climate modelling, and causal analysis to determine when the Arctic will become ice-free during summer – and assess the global climate impacts. The project is supported by the Novo Nordisk Foundation with a grant of DKK 39.5 million.
The system will be developed and trained on the Danish AI supercomputer Gefion. The goal is to provide early warnings to societies, authorities, and businesses – especially those dependent on stable climate conditions, sea ice, and marine ecosystems.
The project brings together three leading research teams from the Danish Meteorological Institute (DMI), DTU Compute | DTU, and the Royal Meteorological Institute of Belgium (RMI).
Ice-free Arctic summers could occur as early as the 2030s – well before global temperatures reach a 2°C increase.
“Sea ice is not merely a local phenomenon. Its disappearance can amplify global climate feedback, alter atmospheric circulation, and intensify extreme weather patterns far beyond the Arctic region. IcyAlert is designed to deliver knowledge and early warnings that society can act upon,” says Tian Tian, Project Lead from the National Center for Climate Research at the Danish Meteorological Institute (DMI).
Tian Tian serves as scientific lead, with Head of Section Stéphane Vannitsem from the Royal Meteorological Institute of Belgium (RMI) and Associate Professor Tommy Sonne Alstrøm from DTU Compute as co-leads.
Gefion opens a whole new world for us
“Access to the AI supercomputer Gefion enables us to train models that are significantly larger than before. There is an entire area of research we have previously been unable to explore, simply because we lacked the computational power to conduct that research. But now, we can begin to develop generative AI-models of such scale that allow us to pursue new climate research. It opens a whole new world for us in terms of creating much more tailored AI-driven climate models for the Northern Hemisphere,” says Tommy Sonne Alstrøm.
Some of the best AI-climate models have been developed by Google’s DeepMind; extremely large models that take several weeks to train, even on a supercomputer. These models are typically trained to forecast the weather fifteen days ahead.
In the IcyAlert project, the researchers aim to expand on DeepMind’s models and combine them with causal methods so that they may be able to make reliable predictions up to six months into the future.
Climate disasters can be traced back to Arctic warming
This is important because many climate disasters that affect Denmark and Europe can be traced back to Arctic warming, as the Earth’s climate is interconnected in a vast, closed system.
As temperatures continue to rise, ice-free conditions may extend across multiple seasons, amplifying climate impacts beyond the polar region. Although melting sea ice does not directly raise sea levels, it accelerates Arctic warming – up to four times faster than the global average – by reducing surface reflectivity and increasing solar absorption. White ice reflects sunlight; dark ocean water absorbs it.
This intensified warming disrupts atmospheric circulation and increases the likelihood of extreme weather events across Europe and other regions far from the Arctic. These include more frequent and intense rainfall, droughts, storms, and heatwaves.
DMI also underlines the importance of the project:
“We are pleased that the Novo Nordisk Foundation is supporting our vital Arctic research. The prospect of an ice-free Arctic Ocean has major implications for security policy, international shipping routes, wildlife, and – most importantly – our climate,” says Ulrik Smith Korsholm, Deputy Director at DMI.
Together, the researchers will develop a climate forecasting tool capable of delivering high-resolution predictions of ice-free conditions and associated climate patterns – both in the short term (upcoming seasons) and long term (post-2030).
NNF: A major step forward in applying AI to climate science
“Each year, we launch the Challenge Programme to engage with the research community and support leading scientists in realizing their most promising, ambitious, and societally relevant projects,” says Lene Oddershede, Chief Scientific Officer for Planetary Science & Technology at the Novo Nordisk Foundation.
“Accurately predicting when the Arctic will be ice-free in summer is difficult due to the complex interplay between sea ice, ocean, and atmosphere. But with IcyAlert, DMI is taking a major step forward in applying AI to climate science—and in developing better strategies to mitigate the effects of global warming.”
The IcyAlert project will run until 2031, with the first early warnings and analyses expected by 2028. Initially, the system will be used for research and policy advice, but the long-term ambition is to integrate it into emergency preparedness and planning systems. Once complete, the system will be able to run on a standard computer.
The six-year project gives Tommy Sonne Alstrøm the opportunity to hire four PhD students.