NOAA Fisheries identified that Bryde’s whales are the source of a new whale call in the western North Pacific. Working with Google AI, we used machine learning to sort through thousands of hours of acoustic recordings and identify when and where these calls occur.
In 2014, an Oregon State University autonomous seaglider survey in the Mariana Archipelago heard a new sound: a mechanical-sounding whale vocalization. They called it a “Biotwang.” Without visual observers to identify the source of the sound, they guessed that it might be made by a baleen whale, but could not determine the species. In 2018, NOAA scientists were conducting a visual and acoustic survey for marine mammals in the Mariana Archipelago. During the month-long survey, they spotted 10 different groups of Bryde’s whales. In nine of those groups they also recorded Biotwangs—solving the mystery of the source of the unique call.
Examining the Data with AI
NOAA scientists use long-term passive acoustic recorders to monitor the health and status of marine mammal populations—particularly in hard to reach, remote locations. In the North Pacific they have been deploying bottom-mounted data recorders at 13 locations since 2005. These recorders have captured more than 500 TB of data, or almost 200,000 hours of recordings. If you were to listen to all of that audio straight through it would take you almost 23 years! Because the Biotwang had not been identified before, scientists had no idea where they might find it in this sea of data.
That’s where Google stepped in. NOAA scientists worked with Google to use AI and machine learning to analyze the immense passive acoustic monitoring datasets. With the help of AI and machine learning, we were able to identify where and when Biotwangs were recorded in past acoustic data. Thanks to AI we were able to do it in a matter of hours, rather than years. The results of this research were recently published in Frontiers in Marine Science.
This model was part of an open-source multi-species marine mammal vocalization machine learning classification model. It allowed us to analyze the recordings quickly and efficiently and with a higher level of accuracy than available using traditional methods.
We identified a consistent seasonal presence of Biotwangs only in the Mariana Archipelago and to the east at Wake Island. This suggests the Biotwang may be a call specific to a western North Pacific population of Bryde’s whales. Knowing that Biotwangs are produced by a specific population helps us to monitor the distribution of that population.
The seasonal occurrence of Biotwangs is consistent with Bryde’s whales migrating between low and mid-latitudes. There is a small peak between February and April, and a larger peak between August and November, as the whales travel past the recording sites. These peaks vary from year to year. There were a lot of Biotwangs in 2016, during a strong El Niño year, and almost none heard in 2021, a La Niña year. We think the whales may be feeding along ocean currents that are strongly influenced by the Pacific Decadal Oscillation. As climate change advances, more frequent and extreme El Niño and La Niña years may cause the productive current to shift poleward and be more variable. This means the whales will have to travel further and work harder to find their food, which can impact the health of the population.
We also partnered with Google in 2018, when we used machine learning to identify humpback whale song in our long-term passive acoustic dataset. That was the first time a large audio dataset was analyzed in a comprehensive way.
Next Steps for Bryde’s Whale Research
All Bryde’s whales are protected under the Marine Mammal Protection Act. We now know that Bryde’s whales produce Biotwangs, and the geographical and seasonal patterns of the calls. This gives our scientists a better idea of their population distribution and movement pattern, and will help us better protect this species.
