Biotwang: The Mariana Trench Sound Was a Whale Near the Surface

Chapters

1. 0:00 A recorder off Pagan, 5 August 2016
2. 0:54 Three claims
3. 1:15 The glider, September 2014
4. 2:15 Five parts, named for its synthetic sound
5. 3:37 Claim one: the Mariana Trench
6. 4:05 Sirena Canyon, April 2016, from the control room
7. 5:24 What it was not
8. 6:07 Bryde's whale, almost
9. 6:30 Claim two: most likely a minke
10. 7:18 Why a sound alone cannot name a species
11. 7:39 July 2018: the ship and the sonobuoys
12. 8:38 Nine of ten
13. 9:48 The animal, for scale
14. 10:15 Claim three: 200,000 hours
15. 11:22 The detector
16. 12:03 Where, and when
17. 13:13 Every four minutes
18. 13:33 Our turn: the public archive
19. 13:51 A clock time in a figure caption
20. 15:05 The loudest thing in the file
21. 15:30 Our count
22. 16:29 What stays open
23. 17:09 Three claims, closed
24. 17:37 Next: read this hydrophone

Transcript

A recorder off Pagan, 5 August 2016 0:00

Eight hundred and thirty metres down, off Pagan Island in the Marianas, a recorder has been sitting on the seafloor since May of twenty fifteen. On the fifth of August, twenty sixteen, at two minutes past five in the afternoon, U T C, it recorded this.

That is a Biotwang, played ten times faster than it happened. At real speed, most of it sits below what your ears can do. A glider first recorded it in twenty fourteen, over the Mariana Trench, and for four years nobody could say what was making it. This episode is about how that was closed. And about a number in a figure caption that let us pull this exact call out of a public archive of two hundred thousand hours.

Three claims 0:54

You have probably heard three things about this sound. That it came from the Mariana Trench. That it was probably a minke whale. And that it took artificial intelligence to solve it. One is misleading, one was wrong, and one is half true. The record for all three is public, and we read it.

The glider, September 2014 1:15

The recorder was not a ship. It was a Seaglider: a torpedo-shaped robot with no propeller, which changes its buoyancy to sink and rise, and flies a saw-tooth path down to a thousand metres and back. One hydrophone, sampled at one hundred and ninety-four kilohertz. Each dive took four to six hours. It was launched off the west coast of Guam on the twenty-ninth of September, twenty fourteen, for a U S Navy monitoring programme.

It was picked up on the fourteenth of November, twenty-four nautical miles from where it went in. Between the fourteenth of October and the sixth of November it had recorded three hundred and twenty-six of these calls, on thirty-eight dives. Typically five to six minutes apart, often in long sequences. In March and April of twenty fifteen two more gliders went back. One heard twenty-nine calls. The other, eighty-one.

Five parts, named for its synthetic sound 2:15

The team at Oregon State and NOAA's Pacific Marine Environmental Laboratory named it the Western Pacific Biotwang, in their words, because of its unusual synthetic sound. To measure it they picked twenty calls with a signal to noise ratio above twenty decibels, hours apart, so that they were probably not all the same animal. The whole thing lasts two and a half to three and a half seconds. And it has five parts.

Part A is a short tone, four tenths of a second, centred near sixty hertz. Part B is the moan: one point six seconds, sweeping down from about forty-four hertz to about thirty-one. It carries strong harmonics, and the one near four hundred and ten hertz was often the loudest thing in the call. That harmonic is how the analysts found Biotwangs in days of recording.

Then two short upsweeps, sixty to a hundred and fifty hertz, half a second each. Then a noisy burst, two tenths of a second, with energy up to seven and a half kilohertz. And the ending: metallic upsweeps, two thirds of a second, most of the energy between seven hundred and eight hundred hertz. When the call is faint, only part B survives. Remember that. It matters later.

Claim one: the Mariana Trench 3:37

Now the first claim: that this came from the Mariana Trench. Here is what the paper actually says. The survey area was the Mariana Trench Marine National Monument, two hundred and forty-six thousand square kilometres of ocean east of the islands. The glider never went below a thousand metres. And the calls were recorded, quote, in shallow and deep water, and during all portions of a dive.

Sirena Canyon, April 2016, from the control room 4:05

This is the same monument, eighteen months later. NOAA Ship Okeanos Explorer, April twenty sixteen, dive three of the Deepwater Exploration of the Marianas, into Sirena Canyon. Five thousand metres. That is the trench. It is also less than half the way down.

The science lead that morning was Jeff Drazen, a deep-sea biologist at the University of Hawaii. Here is why the robot was going down there at all.

The Challenger Deep is close to eleven kilometres. A Bryde's whale spends most of its day within fifteen metres of the surface and dives to about three hundred. The trench was underneath the glider. It was never part of the sound. Saying the Biotwang came from the Mariana Trench is like saying a birdsong came from the Grand Canyon because the microphone was on the rim.

What it was not 5:24

So what was it? The paper worked by elimination. Not ships, not seismic airguns: nothing man-made matched. Not earthquakes, not ice, not wind or rain. Biological, then. And with a moan that low, a baleen whale. Six of them pass through the Marianas.

Blue whale calls last ten to thirty seconds. Too long. Fin whales make twenty hertz pulses, sharp and short. Too short. Sei whale downsweeps sound different. Humpback song is complex, but it does not repeat one stereotyped call like this. That left two.

Bryde's whale, almost 6:07

The first was Bryde's whale. Two NOAA reports from these same islands had already logged an unidentified tonal call that looked like parts A and B, and compared it to a known Bryde's whale call from the eastern Pacific. But no published Bryde's whale call had anything like parts D and E. Nothing above two hundred hertz. Nothing metallic.

Claim two: most likely a minke 6:30

The second was the minke. Dwarf minke whales on the Great Barrier Reef make a call nicknamed Star Wars: complex, sweeping, with a metallic finish. Aurally, the authors wrote, the Biotwang most resembled it. And minke boings were recorded on the southern survey in March twenty fifteen. The paper's last paragraph: most likely a minke whale.

That is how it went out in December twenty sixteen. A mysterious sound from the Mariana Trench, probably a minke. The lead author, Sharon Nieukirk, was more careful than the headlines. If it is a mating call, she asked, why are we getting it year round? The glider had heard it in autumn, winter and early spring. That, she said, is a mystery.

Why a sound alone cannot name a species 7:18

Here is why the paper could not do better. You cannot assign a species to a sound from the sound alone. Somebody has to see the whale and record the call at the same moment. The authors said so: more data are needed, including genetic, acoustic and visual identification of the source. That took a ship.

July 2018: the ship and the sonobuoys 7:39

July twenty eighteen. The NOAA research vessel Oscar Elton Sette surveyed from Guam north to Pagan and west to the West Mariana Ridge, the ninth of July to the first of August. Observers on line transects. Whenever a baleen whale was approached within a nautical mile, they dropped a sonobuoy, a Navy DIFAR buoy, and listened live to everything below three kilohertz.

The same cruise also dropped drifting buoys. Eight spar buoys in twenty eighteen, each with two hydrophones hanging at a hundred and thirty-five and a hundred and fifty metres, left to drift west for up to eleven and a half days. Eighteen hundred hours of recording. A hundred and ninety-one Biotwangs, most of them on two drifts. Three years later, twenty-two buoys, forty-four hundred hours, eighty calls. The whales were there both years. There were fewer in twenty twenty-one.

Nine of ten 8:38

Thirteen large whale sightings had a sonobuoy in the water. Ten were confirmed Bryde's whales. Nine of those ten had Biotwangs on the buoy. Single animals. Mother and calf pairs. Groups of up to four.

The one Bryde's whale sighting with no Biotwang was a group of five adults and no calf. Two more sightings were large whales nobody could identify to species. Both had Biotwangs. Take one line from the table: seventeenth of July, three fourteen in the morning, fourteen point five north, one forty-four point six east. One Bryde's whale, closest approach fifty metres. Biotwang: yes.

That closes it. Bryde's whales make the Biotwang. It is only the third call type ever attributed to Bryde's whales in the western North Pacific, and the first with anything metallic in it. The minke guess was reasonable and wrong. And the authors added a line worth keeping: if Bryde's whales can do this, they may be responsible for other unknown complex calls.

The animal, for scale 9:48

The animal, for scale. Twelve to seventeen metres. Around forty tonnes. Three ridges on the top of the head where every other rorqual has one. It likes water between sixteen and twenty-two degrees, and lives between forty south and forty north. It does not migrate to the poles to feed. It eats where it is, and drifts between low and middle latitudes with the food.

Claim three: 200,000 hours 10:15

Now the third claim, the artificial intelligence. NOAA's Pacific Islands network has kept bottom recorders at thirteen sites since two thousand and five, at depths from a hundred and eleven to twelve hundred and sixty-six metres. Five hundred terabytes. Almost two hundred thousand hours. To listen to it once would take about twenty-three years.

The instrument is a HARP, a high-frequency acoustic recording package built at Scripps. One hydrophone, flat from ten hertz to a hundred kilohertz, on a frame on the seafloor with a float to bring it back. For this search the recordings were filtered down to ten thousand samples a second, so nothing above five kilohertz survives. The top of the twang is gone before anyone listens. What is left, at eight hundred metres, is the moan.

So people annotated three deployments by hand. Saipan for two years, Tinian for one summer, scrolling through spectrograms and marking every Biotwang they saw. Those became the training set.

The detector 11:22

Google engineers built a convolutional network on it. Five second windows, a log mel spectrogram from twenty hertz to twelve kilohertz, and a normalisation step that flattens differences between recorders. On held-out data it scored point nine nine two at Wake, point nine six at Saipan, with a threshold set for three percent false alarms.

So did A I solve the Biotwang? No. A ship and a Navy sonobuoy solved it in July twenty eighteen. The model did something else, something no person could do. It read the other hundred and ninety-nine thousand hours.

Where, and when 12:03

This is what it found. Biotwangs at four sites, consistently: Saipan, Tinian, Pagan, and Wake Island, twenty-two hundred kilometres to the east. A few at Pearl and Hermes Reef in the north-western Hawaiian Islands, mostly in twenty fifteen and sixteen. A few at Howland Island, near the equator. Nowhere else in the network.

And when. At all four sites, calling rises from July and peaks between August and November. A smaller peak in February to April, some years. On the busiest days, seventy-eight percent of the recording time at Wake had a Biotwang in it. Sixty at Pagan. Thirty at Saipan. Twenty sixteen was loud everywhere. Twenty twenty-one was nearly silent.

That pattern is a whale passing. Mark and recapture data from Japanese surveys had these whales at the equator from January to March and in middle latitudes from April to September. The recorders sit on the road between the two. They hear the traffic going north in spring and coming back in autumn.

Every four minutes 13:13

The calls come three and a half to five minutes apart, then the spacing thins out after nine. Off Kauai, whales tracked while calling kept in step with each other, and the interval stretched when one stopped. Nobody has shown what the Biotwang is for. The best guess on the record is that it keeps a group together.

Our turn: the public archive 13:33

Everything above is other people's work. Here is ours. The recordings are public, on a cloud bucket run by NOAA's data centre. We downloaded three point four gigabytes of it. About a hundred and sixty hours from Saipan, Pagan and Wake.

A clock time in a figure caption 13:51

The twenty twenty-four paper has a supplementary figure with one example Biotwang, and the caption gives a clock time. Pagan, the fifth of August, twenty sixteen, seventeen oh two and eight seconds. The recorder writes a table into every file: when each seventy-five second chunk began. Read the table, and that time stamp is second fifty-three thousand three hundred and three of one file. This is what is there.

Measured from that file, the moan runs from thirty-six point six hertz down to thirty-three, over one point seven seconds, twenty-nine decibels above the background. A harmonic at a hundred and twelve hertz. The tone before it, near sixty. And the metallic twang, faint, between seven hundred and eight hundred hertz, three seconds after the start. At the bottom recorder the moan is lower and flatter than the glider heard it. Allen and colleagues call it the thirty-seven hertz long downsweep.

The loudest thing in the file 15:05

One thing the detector taught us before it found any whale. The loudest sound in a hydrophone file is the hydrophone. Every seventy-five seconds the recorder's hard disk spins up, and it leaves a rising harmonic chirp up to five kilohertz. Our first version called that the twang and found it every seventy-five seconds, like clockwork. Which is exactly what it was.

Our count 15:30

With the detector calibrated on the real call, here is the count. Pagan, third to sixth of August, twenty sixteen: thirty-six Biotwangs in twenty-two hours of recording. Pagan, late September: twenty-seven. Wake Island, mid September: twenty-nine. Saipan, seventy hours sampled across twelve months: seventeen, every one of them between the twenty-second of September and the twenty-fourth of November. None in the other eight months.

That is the paper's seasonal curve, redrawn from less than a tenth of a percent of the archive. And the whale is never alone in the file. At Saipan, on the eighteenth of November, this one came in under a cloud of dolphin whistles.

What stays open 16:29

What stays open. No Bryde's whale has ever been filmed making a Biotwang; the link is sightings and buoys, nine times. Which population makes it is not settled: the genetics split near the date line, and the calls are on both sides. And nobody knows why a forty tonne baleen whale ends its call with a sound like a struck wire.

There is also a second unidentified call from these islands, heard at night, that the same group suspects is Bryde's. What would close it all: an acoustic tag on a Bryde's whale in the Marianas, in October, recording the animal's own body while it calls.

Three claims, closed 17:09

Three claims. The Mariana Trench had nothing to do with it; the whale was near the surface and the trench was five kilometres under the microphone. The minke was a fair guess that a ship overturned. And the artificial intelligence did not solve the mystery; it read the archive after the mystery was solved. The call is still down there, on a public server, with a clock time. We went and got it.

Next: read this hydrophone 17:37

Next time, this channel stops telling you about a hydrophone and teaches you to read one. How a clock table hides inside an audio file. Why the recorder is the loudest animal in the ocean. What a ten times speed-up does to both axes. And how a time stamp in a figure caption is a coordinate. Every example from the public archive and the four sounds already on this channel. No mystery. Just the skill.

Description and sources

The Western Pacific Biotwang was recorded by a Seaglider over the Mariana Trench Marine National Monument between 14 October and 6 November 2014: 326 calls on 38 dives, 2.5 to 3.5 seconds each, five parts, from a 60 Hz tone through a moan near 40 Hz to metallic upsweeps reaching 7.5 kHz. Nieukirk et al. (2016) called it, by elimination, most likely a minke whale. In July 2018 NOAA's Mariana Archipelago survey dropped a sonobuoy at thirteen large-whale sightings: ten were Bryde's whales, nine of those had Biotwangs on the buoy. Allen et al. (2024) then ran a Google-built detector over NOAA's Pacific Islands Passive Acoustic Network: Biotwangs at Saipan, Tinian, Pagan and Wake, peaking August to November. The trench had nothing to do with it: the glider never went below 1,000 m and a Bryde's whale lives near the surface. The archive is public. Using the clock time in the 2024 paper's supplementary figure and the timing table inside each file, we located that exact call in a Pagan recording from 5 August 2016 (moan 36.6 to 33.0 Hz over 1.7 s, 29 dB above background, harmonic at 112 Hz, twang near 750 Hz) and counted calls in about 160 hours from Pagan, Wake and Saipan: 36, 27 and 29 in three multi-day files, and 17 at Saipan across a year, all between 22 September and 24 November.

PRINT-READY, FROM THIS CHANNEL

Read This Dive, Vol. 01 — the method, and five dives read with it

https://therepository.gumroad.com/l/read-this-dive

PRIMARY SOURCES

• Nieukirk, S.L. et al. (2016). A complex baleen whale call recorded in the Mariana Trench Marine National Monument. J. Acoust. Soc. Am. 140(3): EL274. doi:10.1121/1.4962377 (CC BY): https://repository.library.noaa.gov/view/noaa/14697
• Allen, A.N. et al. (2024). Bryde's whales produce Biotwang calls, which occur seasonally in long-term acoustic recordings from the central and western North Pacific. Front. Mar. Sci. 11: 1394695 (CC BY; Supplementary Figure 1 gives the time stamp used here): https://doi.org/10.3389/fmars.2024.1394695
• NOAA PIFSC (2021). PIPAN 10 kHz Data. NCEI, doi:10.25921/Z787-9Y54, gs://noaa-passive-bioacoustic/pifsc/audio/pipan_10/ (Pagan, Wake and Saipan files named on screen)
• NOAA Fisheries (2024). Hey, Google: Find this New Whale Sound: https://www.fisheries.noaa.gov/feature-story/hey-google-find-new-whale-sound
• Google Research (2024). Whistles, songs, boings, and biotwangs: https://research.google/blog/whistles-songs-boings-and-biotwangs-recognizing-whale-vocalizations-with-ai/
• NOAA Fisheries, Bryde's whale species profile: https://www.fisheries.noaa.gov/species/brydes-whale
• NOAA Fisheries (2018). Photo Journal: Whales and Dolphins in the Marianas: https://www.fisheries.noaa.gov/science-blog/photo-journal-whales-and-dolphins-marianas
• Szesciorka, A.R. et al. (2023). An unknown nocturnal call type in the Mariana Archipelago. JASA Express Lett. 3: 011201: https://repository.library.noaa.gov/view/noaa/49197

ARCHIVE PLATES — public-domain documents, with author, date, archive and licence.

• 2018-07 | NOAA Fisheries | public-domain

... and 5 more plates, each credited on screen in the video and listed in this project's manifest.

Abyss Report reads the deep ocean through the record that resolved it -- instrument, date, position and the primary sources above.

About the material. Every moving frame is NOAA Ocean Exploration video from EX1605L1 dive 03, Sirena Canyon, April 2016, public domain, credited on screen: the place, not the animal, which has never been filmed making this call. Photographs: NOAA Fisheries, July 2018. The whale calls are unaltered windows of NOAA PIPAN recordings (NCEI, public domain), one sped up 10x and labelled; two cuts play the Okeanos control room. Spectrograms marked 'this channel' and the counts are ours. Diagrams are drawn in code from published values. Nothing is generated.

#Biotwang #MarianaTrench #BrydesWhale #NOAA #OceanSounds