In our Research Highlights blog series, we debut newly published fisheries research by our women of fisheries colleagues. If you have research you would like to highlight and share with our readers, submit a nomination form here.

---

This Month’s Research Highlight:
Munger, J.E., D.P. Herrera, S.M. Haver, L. Waterhouse, M.F. McKenna, R.P. Dziak, J. Gedamke, S.A. Heppell, and J.H. Haxel. 2022. Machine learning analysis reveals relationship between pomacentrid calls and environmental cues. Marine Ecology Progress Series 681:197–210.

This is a story of innovation, family collaboration, and of course fish. It embodies so much of what makes our field so exciting; by thinking outside the box, we can move fisheries science forward. 

The story begins at the Cooperative Institute of Marine Ecology and Resource Studies (CIMERS) at Oregon State University. As a fellow there, Jill Munger began using acoustics to assess the effect of changing environmental conditions, such as those related to climate change, and management practices on reef fishes within the National Park of American Samoa. To do this work, she and her team analyzed the output of hydrophone recordings.

Hydrophones are specialized microphones used for passive acoustic monitoring, or PAM for short. Although there are different types, Jill analyzed data from a fixed hydrophone platform maintained by NOAA and the National Park Service that was deployed at 33 meters on the ocean floor off Tutuila Island. The hydrophone  recorded sound over a period of four years. PAM allows acoustic information to be collected with minimal disturbance by people and with limited influence of maritime conditions. 

In this case, Jill focused her work on damselfishes. They are part of a group of fishes known as pomacentrids that are prolific producers of sound, making them prime candidates for study. However, as Jill notes, “Hypothetically, this technique could work for anything that generates sound in any environment including other species of soniferous fish (freshwater included), vessels, invertebrates, mammals, and anything else you can think of!”

Jill began by subsampling damselfish calls from about 18,000 hours of acoustic data. It was an extremely time-consuming process, resulting in only a small proportion of the acoustic data being processed. There had to be a better way, and she didn’t have to look far for a solution. It turns out her brother, Daniel Herrera, a machine learning engineer who would become a co-author on the publication, proposed an idea. Jill recalls, “I was lamenting the loss of potentially important time series data from my subsampling strategy and he offered to collaborate with me on the machine learning front. This was my introduction to machine learning and as a bonus, he helped me learn to code in Python. I obviously won the sibling lottery!”

So, what is machine learning? It essentially involves a computer program that “learns” from known data to then process unknown data. This can be very complicated, because the ocean is, well, very noisy. Jill and her brother used visual representations of the sound called spectrograms, like the one you see below, to train the computer model. Once trained, the model was used to process the entire record in 2 second intervals. This allowed researchers to calculate the percent of time damselfish calls were detected and then relate that information to factors such as time of day, water temperature, and tidal amplitude. 

And it turns out this sibling duo was onto something. Their model was able to identify pomacentrids with 94% accuracy.  

Jill is currently a research assistant working for CIMERS on another marine acoustic project with Dr. Samara Haver and looks forward to additional opportunities for using this technology to answer important scientific questions. “I’m excited about advances in machine learning techniques and how they impact our ability to get as much as we can out of datasets that span multiple years. Instead of getting a glimpse of what is happening in an ecosystem, we can now see the ecological stories unfold. That gives us a scalable and efficient tool for research, environmental monitoring, and conservation planning.”

This paper represents the interdisciplinary nature of fisheries science. Crossing over into the realms of other disciplines, such as engineering in this case, can bring about new perspectives and new ways of doing science. Thinking outside the box isn’t new for Jill, though. This month, she officially ends her 22-year career with accounting and finance to pursue her marine science career full-time. With her diverse experience and innovative, out-of-the-box thinking, we are excited to see where this work takes her next. 

The full manuscript can be found and downloaded here:
https://doi.org/10.3354/meps13912

Original press release:
https://today.oregonstate.edu/news/researchers-develop-automated-method-identify-fish-calls-underwater