Artificial intelligence used to count tens of thousands of puffins

How exactly do you count the tens of thousands of puffins who call one of Scotland’s most famous nature reserves home?

For years, the answer was by hard graft, with rangers checking burrows and nests for birds and eggs, and observers forced to sit for hours at a time armed with clipboards and no little patience.

But in a marriage of nature and cutting edge technology, the arduous task of establishing the puffin population on the Isle of May is being carried out using artificial intelligence, machine learning, and image recognition software.

Those behind the project believe it could help minimise disruption to birds’ breeding and feeding habits, particularly when faced with developments such as offshore windfarms.

The initiative uses four cameras placed in stainless steel boxes at various points of the island in the Firth of Forth in order to capture live footage of the puffins. Each box has a condensation heater as well as a backup power supply.

The footage is then stored and processed using an artificial intelligence program which is capable of spotting the puffins and tracking them frame by frame. Each bird is assigned a unique identifier, allowing the software to follow its movements and establish the overall number of puffins.

The scheme is the brainchild of SSE Renewables, which wants to find out if its Beatrice windfarm, situated eight miles off the coast of Wick, is impacting on the flight paths of the birds as they travel to gather food to take back to their burrows.

The AI project is tracking the puffins on the Isle of May national nature reserve. PIcture: Lone Gill/NatureScot

It teamed up with the tech giant, Microsoft, and Avanade, a US-based artificial intelligence specialist, to roll out the tracking program, which monitored the birds as they landed to breed in late March and early April, before returning to sea in August.

With around 80,000 puffins recorded on the island in March last year, the data is currently being analysed, but there are hopes the approach could be replicated in order to monitor the habitats of other species.

Simon Turner, chief technology officer of data and AI at Avanade, said the use of technology made the counting process more efficient and less invasive.

“Using cameras and AI, we are now able to count the number of birds and monitor their burrows all day, every day, without going near them,” he said.

“The AI will draw a box around each puffin it spots and give them unique tags like ‘001’, ‘002’, or ‘003’. When the camera moves to the next frame, it understands that the puffin closest to a particular box is the same one.”

James Scobie, SSE Renewables’ digital delivery lead for the project, said: “We’re still looking at the data but through the trial, we have already made some interesting findings.

“The land was barren when we first visited the island in February. However, as spring progressed into summer, the array of flowers that had grown tricked the AI. We learned that it would be important to have seasonal training of the AI to maintain the level of accuracy we expect.

“We found that on average, the highest level of puffin activity in a colony is observed at dawn and at dusk, but this is dependent on tidal times and fishing conditions. Adult puffins will not turn down a good opportunity to catch food for their young.”

A message from the editor:

Thank you for reading this article. We're more reliant on your support than ever as the shift in consumer habits brought about by coronavirus impacts our advertisers. If you haven't already, please consider supporting our trusted, fact-checked journalism by taking out a digital subscription.