Site icon Conservation news

Scientists develop AI that can listen to the pulse of a reef being restored

A healthy coral reef in Tahiti.

A healthy coral reef in Tahiti. Image by Jayne Jenkins / Ocean Image Bank.

  • Scientists have developed a machine-learning algorithm that can distinguish healthy coral reefs from less healthy ones by the soundscape in the ecosystem.
  • Previous studies had established that the sounds of life in a successfully recovered reef are similar to those from a healthy reef, but parsing all the acoustic data was slow and labor-intensive.
  • The new algorithm has been hailed as “an important milestone” for efficiently processing acoustic data to answer the basic question of how to determine the progress of a reef restoration program.
  • Researchers say follow-up work is still needed, including to check whether the algorithm, tested in the Pacific Coral Triangle, also works in reefs in other parts of the world.

Healthy coral reefs, with their brightly colored corals and bustling schools of fish, are easy to spot underwater. New research into the marine soundscapes of reefs in the Sangkarang Archipelago in Sulawesi, Indonesia, shows they’re easy to hear too — for the right algorithm.

The study, led by University College London and Zoological Society of London doctoral student Ben Williams, was published in July in the journal Ecological Indicators.

Williams’s research builds on a 2021 paper led by his collaborator, Tim Lamont, which demonstrated that the diversity of sounds from marine life — the pops, grunts, scrapes, whistles and rattles that fish and other animals make, detectable underwater by underwater microphones called hydrophones — on mature restored reefs were similar to those heard on healthy reefs.

According to Williams, that study was a proof of concept that soundscapes can be used to track reef recovery. But manually listening to the recordings was a slow and labor-intensive process.

A coral reef in the Marsa Ghozlani dive site in Egypt’s portion of the Red Sea. Image by Renata Romeo / Ocean Image Bank.

Inspired by an assignment from his undergraduate studies, in which Williams was tasked with identifying people diagnosed with heart disease from a complex data set, he turned to machine learning to make sense of the recordings.

“I realized this string of numbers of recordings from a healthy and of a degraded reef was really similar to this string of numbers we had of patients with or without heart disease,” he said.

So he created an algorithm to classify the soundscape data as belonging to either healthy or degraded reefs. Williams then tested his model on a separate group of recordings from three restored reefs in the Sangkarang Archipelago (also known as the Spermonde or Pabbiring Archipelago): two established more than two years ago and one recently restored site. He discovered that most recordings from the older restoration sites had the soundscape of a healthy reef, whereas the majority of recordings from the newly restored site were classified as degraded.

To Aaron Rice, a research scientist at Cornell University’s K. Lisa Yang Center for Conservation Bioacoustics, who was not involved in the study, these findings are an exciting step forward for coral reef restoration.

“Extensive resources and time are spent trying to restore different reef sites,” he said. “The fundamental question is, how do you measure progress and how do you know when it’s done?”

The ability to use acoustic data, which are relatively easy to collect, to answer that question is thus “an important milestone,” Rice said.

Plenty of non-human species have their ears tuned to the sounds of these coral reefs as well. Juvenile fish and coral larvae listen for healthy reef sounds before selecting where to make their homes. If restored sites sound like nice places to settle down, they can attract a new generation of fish, corals, and other marine animals, returning the reef to its former glory. Busy, diverse soundscapes beget busy, diverse ecosystems teeming with marine life.

A reef restoration project in the South China Sea. Image by Ansrx via Wikimedia Commons (CC BY-SA 4.0).

Williams observed this process up close while collecting the data for this project.

“The most inspiring thing I’ve ever seen on a coral reef was to go and swim on one of the sites which had been restored several years ago, and it was completely back to brand new. It looks like a really healthy thriving reef,” he said.

The only hint that it hadn’t always been that way was that if he looked closely, he could see the metal frames that had originally been seeded with coral fragments peeking through gaps in the reef. These frames, called reef stars, are a technique used by the Mars Coral Reef Restoration Program to spur coral growth in barren rubble beds.

Both Williams and Rice say that figuring out exactly what sounds distinguish healthy from degraded reefs is the logical next step to this study. Machine learning is often described as a black box because the models are complex and difficult to interpret. Coral reefs are also busy and jumbled audio environments (Rice likens them to the cacophony of New York City). The difference between a healthy and degraded coral reef might be the crackle of a snapping shrimp, the crunch of parrotfish ripping algae off the reef, or even just the way individual sounds propagate through the physical environment — it will take more time and experimentation for scientists to narrow down which are important.

In the meantime, the research team Williams belongs to is sending hydrophones to other restoration sites in places like the Great Barrier Reef, Mexico, and the Maldives, to learn what reef recovery sounds like in other places. The Pacific Coral Triangle, where the Sangkarang Archipelago is located, has the highest marine biodiversity in the world, so the range of sounds that signify healthy reefs there may be different from the Caribbean or elsewhere. In the team’s quest to understand the world’s coral soundscapes, Williams said he’s interested in testing whether one global or several regional machine-learning models will be needed to separate healthy reefs from degraded ones.

Hydrophone deployment at a reef site. Williams created an algorithm to classify the soundscape data as belonging to either healthy or degraded reefs. Image by National Marine Sanctuaries via Wikimedia Commons (Public domain).

Regardless of the outcome of Williams’s follow-up research, bioacoustic data will remain an important and efficient way to understand marine ecosystems. This same research team has created a relatively inexpensive and open-source hydrophone called a HydroMoth for distribution to those other sites. And hydrophones in general are easy to install and can collect data much longer than researchers can stay underwater. They also record at night and in other ocean conditions that might keep scuba-diving scientists away.

“The idea of listening to coral reefs is either seen as really, really niche, experimental, or too uncertain, and now it’s like — well, no,” Rice said. “These are the kinds of papers that show this is really viable and is ready for prime time.”

But, as Williams is quick to note, successful restoration alone will not save the world’s coral reefs; climate change remains an urgent threat to these vital ecosystems.

“Restoration is brilliant because it can bring back habitat that we’ve lost,” he said. “But if we don’t get a handle on climate change, on greenhouse gas emissions, it’s not going to matter because we’re still forecasted to lose almost all of the world’s reefs by the end of the century if we don’t do that.”

Banner image: A healthy coral reef in Tahiti. Image by Jayne Jenkins / Ocean Image Bank.

Citations:

Williams, B., Lamont, T. A., Chapuis, L., Harding, H. R., May, E. B., Prasetya, M. E., … Simpson, S. D. (2022). Enhancing automated analysis of marine soundscapes using ecoacoustic indices and machine learning. Ecological Indicators140, 108986. doi:10.1016/j.ecolind.2022.108986

Lamont, T. A., Williams, B., Chapuis, L., Prasetya, M. E., Seraphim, M. J., Harding, H. R., … Simpson, S. D. (2021). The sound of recovery: Coral reef restoration success is detectable in the soundscape. Journal of Applied Ecology59(3), 742-756. doi:10.1111/1365-2664.14089

Gordon, T. A., Harding, H. R., Wong, K. E., Merchant, N. D., Meekan, M. G., McCormick, M. I., … Simpson, S. D. (2018). Habitat degradation negatively affects auditory settlement behavior of coral reef fishes. Proceedings of the National Academy of Sciences115(20), 5193-5198. doi:10.1073/pnas.1719291115

Vermeij, M. J., Marhaver, K. L., Huijbers, C. M., Nagelkerken, I., & Simpson, S. D. (2010). Coral larvae move toward reef sounds. PLOS ONE5(5), e10660. doi:10.1371/journal.pone.0010660

Roberts, C. M., McClean, C. J., Veron, J. E., Hawkins, J. P., Allen, G. R., McAllister, D. E., … Werner, T. B. (2002). Marine biodiversity hotspots and conservation priorities for tropical reefs. Science295(5558), 1280-1284. doi:10.1126/science.1067728