health & science
It takes a village to save Antarctica’s seals
Is that a seal or a shadow? Hundreds of thousands of volunteers are scouring satellite images to help save Antarctica's seals, reports Eloise Gibson.
Crabeater seals do not do what they say on the label. Instead of eating crabs, they filter krill from the oceans around Antarctica using special, sieve-like teeth. When not eating, they rest on free-floating pack ice, resting, raising pups and hissing at any researcher who bothers them in the name of conservation.
University of Canterbury ecologist Michelle LaRue first encountered a crabeater when she was trying to study a different, more laid-back seal species. "I was out on the ice with colleagues in McMurdo Sound and we were looking for Weddell seals when we came across this lone crabeater seal," she says. "We were taking blood samples from this seal and he was not very happy about it. I just happened to be walking around behind him to give him space [when] he reared his head up and growled and hissed and huffed."
The resulting photo shows a huge and grumpy crabeater towering over a small LaRue. But size and bad temper are not the only reasons it is often better to study crabeaters from a distance - from as far away, in fact, as a satellite orbiting Earth. Ice expeditions during October, when crabeaters are easiest to accurately count, are “difficult and dangerous", says LaRue. "It's impossible to get in there with a ship and even flying in is probably impossible because the weather so difficult and dangerous," she says. Counting seals by scouring satellite images is safer and cheaper and can cover much more ground, she says. But inspecting kilometres of ice using aerial photographs also has a drawback: it takes a lot of time. In 2016, LaRue enlisted 330,000 volunteers to help count Weddell seals using the internet. Anyone could visit a website and help sort the seals from shadows using their home computer. Between them, citizen scientists scanned a chunk of ice the size of New Zealand, speeding up the research process. Now LaRue wants people to help count crabeaters.
The research is important because, in October, delegates from 24 countries and the European Union will meet to discuss whether to designate the Weddell Sea, east of the Antarctic Peninsula, as a Marine Protected Area. It will take impeccable research and skilled diplomacy for countries that favour ocean protection to convince other nations - including those that operate commercial krill fisheries - to set aside havens for crabeater seals and other Antarctic marine species. One important question is the abundance of krill, which are the staple food for many larger ocean creatures. Krill are almost impossible to count, but species like seals give a good indication of the overall health of the ecosystem. "Research is supposed to inform the management of fisheries and one way to do that is to understand how many krill there are," says LaRue. "Counting krill is hard but we can study indicator species that eat krill - for example seals. What we want to do in the Weddell sea is figure out where the crabeater seals are, and how many there are to get a baseline number."
Using the crowd to help count seals isn't always accurate, but the project is designed so it doesn't greatly matter if people see seals when they're really looking at rocks. The campaign website asks people to look at images one by one and cast a vote: are there seals in the picture, or not? From high up it can be difficult to tell what is a crabeater and what is a smudge or a bump in the ice. But LaRue says that, just by ruling out empty ice swathes, citizen scientists are narrowing the search for the researchers. "In our experience, folks are really good at identifying when seals are not there," she says. "We end up with lots of false positives, lots of images where people think it's a seal but it's a actually a shadow, but that’s okay because we don't want people to count the seals, we do that. But when there’s nothing there, there's no mistaking it, so we very rarely get a false negative. People are putting the empty images off to the side for us. It might seem boring to go through lots of images without seals but it’s really important to scientists and we really appreciate it."
In Wellington, researchers are trying to make it even easier for wildlife researchers to identify and count species. They are using the results from crowd-sourced research to train artificial intelligence programmes to differentiate between species. Wildlife researcher Victor Anton employed citizen science to identify images from camera traps he'd set up in reserves around Wellington. He wanted to know which pictures contained invasive mammals that might hurt native species, and whether the mammal in question was a cat, possum, rat, mouse or other creature. To work out whether the volunteers' answers were accurate, he got at least two professional ecologists and three citizen scientists to view the same set of images, and compared the results. He noticed the accuracy was good, but it depended on the type of animal.
"If you go to multiple citizen scientists you can get accuracy to same level as the professionals, but it changes a lot depending on the animal," says Anton. "If it was a cat, for example, maybe I only need two or three volunteers, and if they agree it’s a cat there is a high chance they're accurate. But sometimes people say the image is empty and there is a mouse. People struggled more to identify a mouse."
"[Now] we are looking at weighting people’s responses depending on how well they they did before [at identifying the right species] and how confident people were in their answer. If you’re good, the model will rely on you more," he says. "We can also can pick up if people need more training."
Tweaking the citizen science model may make it more accurate, but it still relies on people doing all the hard work. The next step for Anton and other digitally-minded biologists is trying to create algorithms that can sort images, and other data like audio recordings, for scientists. "It’s still in the early stages but that is the next step," says Anton. The data supplied by ordinary people helps, because software designers can use human answers to help train the machines in what to look for.
For now, wildlife biology still relies heavily on professionals and even promoters of AI say that it probably always will. But enthusiastic helpers - and, maybe, one day, new software - can make the process shorter and allow more research to be done.
Newsroom is powered by the generosity of readers like you, who support our mission to produce fearless, independent and provocative journalism.