If you liked this story, share it with other people.
What does an eagle ray’s menu look like?
An artificial intelligence model can now answer that question by listening to sounds of the animal chewing on food.
Scientists developed the machine learning algorithm to detect the sound of shells being crushed by predators when they feed on mollusks. According to a study published in the journal Ecological Informatics, the model can also identify the prey based on the sounds.
“A lot of animals out there, particularly marine animals, have the unique ability to crush shells open,” Matt Ajemian, assistant research professor at the Harbor Branch Oceanographic Institute at Florida Atlantic University in the U.S. who was part of the research, told Mongabay in a video interview. “But we don’t know how much they eat and what they feed on. So we wanted to see if we could remotely detect an animal feeding on a clam versus a gastropod.”
Keeping track of predator-prey interactions is crucial, especially in the face of rapidly changing marine habitats. Monitoring what and how much larger predators are eating is important to understand the resources they depend on and subsequently plan effective conservation action.
Conversely, it’s also critical to have data on how much pressure there is on shellfish populations that serve as prey. “For example, in a clam bed or seagrass bed, we want to know how much prey is removed by a predator over the course of a year,” Ajemian said.
However, gathering this data is not an easy task. Tracking predators underwater is a challenge. Even with the use of sensors and cameras mounted on animals, scientists often find it difficult to see what they are eating amidst the digging up of sand that obscures their vision. In the past, scientists have also captured the animals to flush out the content in their stomach. But this method is intrusive, and “you don’t get everything or the contents are very degraded,” Ajemian said.
That’s what led Ajemian and his team to turn the focus to the signature sounds that animals make while breaking open shells.
The team conducted the experiment in a controlled tank setting. They focused on whitespotted eagle rays (genus Aetobatus) that are adept at crushing shells of their prey. They fed the predators known amounts and sizes of prey, and gathered the corresponding audio. In order to validate their research, they took the experiment into the ocean off the Florida coast where they set up clams and snails in front of a camera and audio recorder. On analyzing the data, the team found that they were able to tell between the prey that was being eaten based on the sounds as well as the amount of time it took for the predator to process its food.
For example, for an eagle ray, crushing and eating a clam takes a bit of effort, because it will need to sift through the food to spit out the shells and pull the meat out. With snails, however, it takes much less time since there’s just one point of attachment to the shell. “If you break that attachment point, the meat comes loose,” Ajemian said.
They also attached biologgers on the animals, which allowed them to hear and see the animal pick up its prey and crush it.
The model was then trained on all this data to first detect shell-crushing sounds, and filter out those sounds from ocean noise. It was then trained to identify the prey based on that sound.
The biggest surprise from the research, Ajemian said, was that advanced AI was not required to make these detections. “The method that required the most computing power wasn’t necessarily the method or approach that yielded the best results,” he said. “There were ones that would take a fraction of the computing power that were pretty darn close.”
This, he said, indicated that the tool could be made accessible to more scientists and that it could be used in a way that is much less resource-intensive.
Going forward, Ajemian said the team is planning to expand beyond whitespotted eagle rays and train the model for other predators like crabs. They are also planning to use the methodology to apply it to data gathered from animal-borne tags.
“We have long-term recordings at stations that are in habitats for months,” he said. “And we want to strip out all of this information to see where and when shellfish get cracked open.”
Banner image: A whitespotted eagle ray in its natural habitat. Understanding the diet of predators is crucial to figure out the resources they depend on and the pressure it puts on prey. Image courtesy of Cat Nickell.
Abhishyant Kidangoor is a staff writer at Mongabay. Find him on 𝕏 @AbhishyantPK.
Citation:
Ibrahim, A., Cherubin, L., Hampton, C., DeGroot, B., Zhuang, H., & Ajemian, M. (2026). Evaluation of a signal processing and machine learning framework to detect and classify shell-crushing predation events. Ecological Informatics, 103795. doi:10.2139/ssrn.5547167
Suriname remains an outlier in the Amazon Basin: more than 90% of the country is still covered by rainforest, making it one of the few nations in the world that remains a net carbon sink. But a wave of development proposals — from large-scale agriculture and Mennonite farming settlements, to mining projects and new carbon […]
© 2026 Copyright Conservation news. Mongabay is a U.S.-based non-profit conservation and environmental science news platform. Our EIN or tax ID is 45-3714703.
you're currently offline
Leave a Reply