Scientists automatically detect rare marine animals from the air
Knowing how many animals live there is vital information for conservationists but it is not always easy to find this out. When you are attempting to monitor shy, elusive animals who are not only extremely rare but can also range across hundreds of kilometres of coastal waters, getting hold of this data can get really tricky. With support from the Marie Skłodowska-Curie Actions programme, the scientists from MEGAFAUNA have pioneered a new, more efficient way of mapping where the sharks, sea turtles, rays and dugongs live in the coastal waters of New Caledonia. “We don’t currently know where the most critically endangered animals live. So the goal is not only to increase the coverage of protected areas but to find out where the key places to protect are,” explains David Mouillot, project coordinator and professor of Marine Ecology at the University of Montpellier, France.
Alternative solutions
In March 2020, MEGAFAUNA’s main researcher Laura Mannocci was preparing to travel to the south-west Pacific to begin the work of surveying when the pandemic struck. Grounded, she quickly came up with an alternative involving fitting a GoPro camera to a light aircraft owned by local company Air Paradise, thus taking advantage of flights for tourists over the Poé Lagoon. Concentrating on two vulnerable species, dugong and eagle rays, the MEGAFAUNA team localised and identified images of these animals in the 50 hours of footage they obtained. They used this to train artificial intelligence (AI) algorithms to automatically detect these species in aerial images. Using AI to automate the process has significant advantages. “Until now, you would need to watch hours of videos and, since humans get tired and bored, you would miss some of the animals,” remarks Mannocci. A computer scientist on the team used the data to build an algorithm capable of detecting animals even 2 metres underwater, something the human eye cannot detect, thereby improving detection rates still further.
Helped by citizen science
Training an algorithm calls for plenty of data but dugongs are extremely rare. The scientists harvested footage of dugong sightings from social media as one way of overcoming the scarcity of data. With the advent of affordable drones, social media has become a rich source of information. “We were in touch with people from NGOs in Indonesia and amateurs from Australia who are passionate about wildlife,” says Mannocci. “Social media is an incredible source, so it was like indirect citizen science,” adds Mouillot. The MEGAFAUNA team completed the mapping of biodiversity hotspots in the lagoon and demonstrated their technique works, representing an advance in the application of AI to monitoring marine biodiversity. “You can scan vast areas and there is low disturbance [to animals] compared to doing it by boat or underwater,” notes Mouillot. “We now have a groundbreaking data set of marine megafauna in New Caledonia – we literally have hundreds of images for each species,” says Mannocci. The technique could be useful in other island environments, where marine wildlife is less well-protected. The team is applying for funding to do similar work in Mayotte.
Keywords
MEGAFAUNA, dugong, eagle ray, rare animals, biodiversity, biodiversity hotspots, aerial surveys, deep learning, marine ecology, New Caledonia
