A new AI (artificial intelligence) tool is set to allow scientists to more accurately predict Arctic sea ice conditions over a distance of months. Improved forecasts could favor new early warning systems that protect Arctic wildlife and coastal communities from the impacts of sea ice loss.
Published this week in Nature Communications, an international team of researchers led by the British Antarctic Survey (BAS) and The Alan Turing Institute describes how the AI system, IceNet, tackles the challenge of producing accurate predictions of Arctic sea ice for the next season — something that has eluded scientists for decades. Sea ice, a vast layer of frozen seawater that appears at the North and South poles, is notoriously difficult to predict because of its complex relationship to the atmosphere above and the ocean below.
The sensitivity of sea ice to rising temperatures has caused the Arctic’s summer sea ice area to halve over the past four decades, the equivalent of losing an area roughly 25 times the size of Britain. These accelerated changes have dramatic consequences for our climate, arctic ecosystems, and indigenous and local communities whose livelihoods are linked to the seasonal cycle of sea ice. IceNet, the AI prediction tool, is nearly 95% accurate in predicting whether sea ice will be present two months earlier — better than the leading physics-based model.
Lead author Tom Andersson, BAS AI Lab data scientist funded by The Alan Turing Institute, explains: “The Arctic is a region at the forefront of climate change and has seen substantial warming over the past 40 years. IceNet has the potential to fill an urgent gap in sea ice forecasting for Arctic sustainability efforts and work thousands of times faster than traditional methods.”
Dr. Scott Hosking, Principal Investigator, BAS AI Lab co-lead and Senior Researcher at The Alan Turing Institute, says: “I’m excited to see how AI is making us rethink how we do environmental research. Our new sea ice prediction framework merges data from satellite sensors with the output of climate models in a way that traditional systems simply could not achieve. “
Unlike conventional prediction systems that try to model the laws of physics directly, the authors developed IceNet based on a concept called deep learning. Through this approach, the model ‘learns’ how sea ice changes from thousands of years of climate simulation data, along with decades of observational data to forecast months in the extent of Arctic sea ice.
Tom Andersson concludes: “Now that we’ve demonstrated that AI can accurately predict sea ice, our next goal is to develop a daily version of the model and run it publicly in real time, just like the weather forecasts. This can act as an early warning system for risks associated with rapid sea ice loss.”