British Antarctic Survey puts AI into ArctIc to aid caribou conservation

The research, led by scientists from Cambridge-headquartered British Antarctic Survey (BAS) in partnership with The Alan Turing Institute, WWF and the Government of Nunavut, demonstrates how this technology could assist local agencies in protecting critical migration routes which cross areas of land and sea ice.

The technology could help prevent ice-breaking vessels from disrupting these routes at key moments in the migration of caribou, ensuring they can safely travel south across areas of sea ice from their summer calving grounds.

The study, published in Ecological Solutions and Evidence combines satellite observations, GPS tracking, AI forecasting, and local expertise to create AI-informed migration early-warning systems for Dolphin and Union caribou (Rangifer tarandus groenlandicus x pearyi).

Dr Ellen Bowler, a machine learning research scientist at BAS and lead author of the study, explained: “Many Arctic animals’ life histories and movement patterns are intrinsically linked to sea ice, which grows and retreats with the seasons. Climate change is causing Arctic sea ice to recede and become less stable – which also means the region is becoming more accessible for ships. This combination of threats could have fatal consequences for iconic animals like caribou.

“This new technique combines our AI sea-ice forecasts with historic tracking data from the caribou, so we can predict when they are likely to start migrating based on when the sea ice forms. This means that local teams can take short-term action to support the migration by protecting sea ice migration routes from ice-breaking vessels. It’s exciting that AI could give such a positive real-world outcome.”

Dolphin and Union caribou migrate twice a year across sea ice which forms seasonally in the waterways between mainland Canada and Victoria Island. This transit sees the caribou gather on the banks of Victoria Island every autumn, waiting for sea ice to reform before they can migrate back south. They begin crossing once sufficient ice has formed, which varies annually.

Unfortunately, increasing numbers of caribou perish during these crossings due to lower sea ice quality, and ice-breaking vessels which pose a further threat by creating open-water leads, potentially blocking crucial migration routes. Local organisations are exploring how they can mitigate shipping impacts to protect the herd, which are vital to the lives and livelihoods of local Inuit and Inuvialuit people.

The authors developed a model combining caribou tracking data and satellite observations of sea ice to establish connections between sea ice formation and the timing of the autumn migration. The AI then forecasts when sea ice conditions triggering the autumn migration will be met, to give early-warnings of peak migration times which can be updated on an annual basis.

This ecological research builds on an existing machine learning model, IceNet, a probabilistic model shown to accurately predict sea ice extents up to three months in advance. IceNet is trained on satellite-derived observations and outperforms state-of-the-art traditional models, including in seasonal forecasts of summer sea ice. Developed by BAS and The Alan Turing Institute, IceNet is also set to aid navigation by the British Antarctic Survey research ship RRS Sir David Attenborough.

The information provided by the AI model can support local agencies in making more informed and dynamic decisions - such as when and where to recommend short-term limits on ice-breaking activity by nearby ships. The AI model’s development and operation benefit greatly from invaluable local knowledge.

Dr Scott Hosking, co-author and Mission Director for Environmental Forecasting at The Alan Turing Institute said: “The Arctic is warming at four times the rate of the rest of the planet, resulting in rapid changes. In the hands of local conservation experts, and combined with their existing knowledge and practices, this tool could help mitigate the threats posed to wildlife and communities in the Arctic.”

The team are working to explore applications for other Arctic species, including iconic species such as polar bears and walrus. This could include predicting when polar bears are most likely to arrive on land near to communities, to mitigate against human-polar bear conflict.