Source: Times Colonist
Author: Roxanne Egan-Elliott
New artificial intelligence technology that detects orca sounds underwater could help to protect endangered southern resident killer whales in the Salish Sea.
The technology recognizes orca calls and sends real-time alerts to marine-mammal managers, who can use the information to try to prevent harmful collisions between whales and vessels.
“There’s specific areas where vessel strikes are more of a risk, like Active Pass and narrow constrictions, so if we know the whales are travelling into that area and we know there’s vehicle traffic going into that area, we can notify the appropriate folks and they can then wait to go into Active Pass,” said Paul Cottrell, a marine mammal co-ordinator for Fisheries and Oceans Canada.
The declining southern resident killer whale population faces threats to survival that include vessel strikes, a scarcity of prey, contaminants and acoustic and physical disturbance, says Fisheries and Oceans Canada.
Three whales died in 2019, bringing their population down to 73, according to the Center for Whale Research in Washington. The research centre declared another whale missing this week, because the animal hasn’t been seen in more than five months.
The new alert system — a collaboration involving Fisheries and Oceans Canada, Google and California-based non-profit Rainforest Connection — will help track the orcas in real time.
The whale sounds are picked up by a dozen underwater listening stations concentrated in the southern resident killer whales’ critical habitat. Before Google and Rainforest Connection became involved, authorities relied on a livestream of the audio picked up by underwater microphones, which had to be monitored to know where the whales were.
When Cottrell heard about Google’s artificial intelligence that had learned to detect humpback calls in the U.S., he thought: “This is exactly what we need. We need this notification system.”
Cottrell and other key partners now receive alerts to their phones in real time when an orca sound is picked up by one of their listening stations and detected by the technology.
Google used 1,800 hours of annotated audio recordings provided by Fisheries and Oceans Canada to teach a machine model to recognize the sound of an orca and distinguish it from other noises. Google’s technology applies state-of-the-art image classification to spectrogram images, which visually represent sound.
The system feeds audio data to Rainforest Connection’s web interface, which sends notifications directly to Cottrell’s mobile phone.
The non-profit had already developed acoustic monitoring systems to detect the sounds of chainsaws and vehicles in rainforests and send live alerts to their partners in an effort to curb illegal deforestation.
Alerts show where the whales are in real-time and allow marine-mammal managers to determine what direction they’re travelling in when they’re detected by more than one station in the network.
Cottrell said the technology will improve their ability to monitor sick, injured and distressed orcas, which is essential for assessing their health.
It could also prove useful in the event of an oil spill.
“It there was a spill and you knew the trajectory, you could get in front of that and prevent animals from getting into that area,” Cottrell said.
The program might be able to identify the three different pods of southern resident killer whales in the future, something only a few individuals can do, Cottrell said.
“If experts are capable of distinguishing them only by listening to the sounds, then there’s a really good chance that we can as well with AI, because that means there’s some quality of the sound that is distinguishable,” said Julie Cattiau, a product manager at Google headquarters in Mountain View, California.