Chinstrap penguins stand around 76cm tall. They are easily recognisable by the thin black band around their heads. Credit: Chris Oosthuizen

Lire en français

Researchers have trained deep learning algorithms to calculate how much chinstrap penguins eat.

The project, funded by the Antarctic Wildlife Research Fund and the Research Council of Norway, was led by South African researchers, and included ecologists and data scientists from France, Australia and Norway. Published in a Royal Society Open Science paper, it is the first time penguin feeding has been studied using deep learning techniques. Similar methods were previously tested on larger marine animals, including turtles and whales.

Chinstrap penguins (Pygoscelis antarcticus), named for the narrow straplike black band under their heads, feed almost exclusively on the small marine crustaceans Antarctic krill (Euphausia superba). A paper in Global Change Biology this year listed them — along with Humboldt (Spheniscus humboldti) and African (Spheniscus demersus) penguins — as increasingly the most vulnerable of the world’s 18 penguin species. Climate-driven environmental changes at sea, industrial fisheries, and other human pressures are their greatest threats.

There are approximately 3,42 million breeding pairs of chinstraps across Antarctica, with the greatest concentrations in the West Antarctic Peninsula, according to the first global assessment of the species, reported in 2020 in Scientific Reports.

“Although abundant there, the regional population in West Antarctica has been declining since at least the 1980s. This is probably linked to reduced food availability,” says Chris Oosthuizen, a research fellow of the Centre for Statistics in Ecology, the Environment and Conservation (SEEC) at the University of Cape Town, South Africa, who led the Royal Society Open Science paper.

Gauging how much chinstraps eat has been challenging, given that the birds hunt underwater in icy conditions, says lead author Stefan Schoombie, an associate of SEEC and South Africa’s National Institute for Theoretical and Computational Sciences (NITheCS).

Tiny video cameras strapped with waterproof tape to the backs of chinstrap penguins were used to shoot footage of how they hunt krill off the South Orkney Islands in the Western Antarctic Peninsula. Credit: Chris Oosthuizen
Historically, to estimate prey consumption, some researchers have resorted to pumping out the stomach contents of birds. Less invasive data gathering methods are now available. One method that has provided valuable data involves capturing the birds and attaching small video cameras to record their hunts. Biologging technology such as accelerometers and depth sensors are used to gather information on their body movement and the depth at which the seabirds swim.

“With video, you can only collect a few hours of footage before the battery dies, but accelerometers last for days and can collect vast volumes of data,” Schoombie says.Deep learning methods automate the otherwise time-consuming manual analysis of months of field data and terabytes of raw video, audio or accelerometer data.

As part of the Royal Society Open Science study, Schoombie built tiny video cameras and used waterproof tape to attach them, together with an accelerometer and depth sensor tag, on chinstrap penguins feeding off the South Orkney Islands in the Western Antarctic Peninsula. He then reviewed all video footage and recorded when the penguins captured krill. The information gathered was used to train two deep learning algorithms based on two technologies known to AI specialists as a Convolutional Neural Network (CNN) and a V-Net model, to “recognise” prey capture events, based only on body movement and depth data.

Schoombie says their work will be presented in August in Chile at the Open Science Conference of the Scientific Committee on Antarctic Research (SCAR). This will hopefully open the door for their work to strengthen the ongoing revision process of the Ecosystem Monitoring Program (CEMP) of the Commission for the Conservation of Antarctic Marine Living Resources (CCAMLR), the international commission responsible for the conservation of Antarctic marine ecosystems. Chinstraps count among the numerous krill-dependent marine predators CEMP has monitored since the 1980s.

“The advent of new technology and methods could potentially enhance the capacity of CEMP to attribute changes in ecosystem functioning to the effects of fishing pressure or climate-driven environmental change. Newer methods, such as remote-operating time-lapse camera networks, are already used. Such efforts can be expanded more broadly to enhance marine predator monitoring,” says Oosthuizen.

He believes his project team’s algorithms could also be trained to better understand the prey consumption of Gentoo, Adélie and African penguins.

Co-authors Loréne Jeantet and Emmanuel Dufourq of the African Institute for Mathematical Sciences (AIMS South Africa and Research and Innovation Centre, Rwanda), Stellenbosch University, and NITheCS, hope that their efforts will help guide the conservation of this Antarctic species, and help set reasonable limits on commercial krill fishing in the Southern Ocean.