The Repository Abyss Report: Into the Dark

The Upsweep: The Pacific Sound Nobody Went to Check

October 19, 2026 · Runtime 16:10 · Watch on YouTube

Since August 1991 NOAA has had a recording of a train of rising tones from the far South Pacific, near 54 S, 140 W. Its official page is six sentences long, and the Internet Archive shows they have not changed since February 2001. We measured the recording, checked what is at the coordinate, and listened for 220 hours on a public NOAA hydrophone.

Chapters

  1. 0:00 A sound on the Navy's tape
  2. 1:22 A page unchanged since 2001
  3. 2:16 The machine that heard it
  4. 2:40 Sped up twenty times
  5. 3:29 One tone, measured
  6. 4:19 54 South, 140 West
  7. 5:08 No ship has mapped it
  8. 5:37 The earthquakes that are not there
  9. 6:55 The equatorial hydrophones
  10. 7:43 We listened: station NRS04
  11. 8:54 220 hours, no train
  12. 9:56 Vailulu'u: a sound that got a visit
  13. 11:17 Okeanos Explorer, dive 09
  14. 14:18 What would close the Upsweep
  15. 15:45 Next: reading a seafloor map

Transcript

A sound on the Navy's tape 0:00

In August of nineteen ninety one, the U.S. Navy let a NOAA lab listen to its secret underwater microphones. One sound was already on the tape. A train of rising tones, one every fifteen seconds, coming from the far south of the Pacific. This is it, sped up twenty times, as NOAA published it.

They called it the Upsweep. The official page that describes it is six sentences long. It says the sound is loud enough to be recorded across the whole Pacific. It says it peaks in spring and in fall. It gives a position, fifty four degrees south, one hundred and forty degrees west. And it ends with four words: the origin is unresolved.

This episode does three things nobody else seems to have done with that page. We measure the recording itself, slowed back to real time. We check what is actually at that coordinate. And we go and listen, on a NOAA hydrophone whose data is public today. The footage you will see comes from a different volcano in the same ocean, one that somebody did go to see. That difference is the episode.

A page unchanged since 2001 1:22

Start with the page. The Internet Archive first captured it on the twenty fourth of February, two thousand and one. We compared that copy with captures from two thousand and two, two thousand and four, two thousand and seven, two thousand and thirteen, and with the live page today. The text is the same, word for word. The website around it was redesigned. The six sentences never changed.

That matters, because two of those sentences are in the present tense. The source level has been declining since nineteen ninety one. The sounds can still be detected. Every article that tells you the Upsweep is fading is quoting a sentence that was already on the web before two thousand and one. Nobody has updated it in twenty five years. So the public record does not say whether it is still there.

The machine that heard it 2:16

The machine that first heard it was built to hunt submarines. SOSUS is a chain of hydrophone arrays the Navy laid on the seafloor during the Cold War. In nineteen ninety one, NOAA was given access to the arrays in the northeast Pacific to listen for something else: underwater earthquakes and eruptions on the seafloor spreading centres. The Upsweep came with the job.

Sped up twenty times 2:40

The file NOAA published lasts twenty eight point eight seconds. It is sped up twenty times, so the real event took nine minutes and thirty six seconds. Speeding up moves both axes. Time gets shorter, and every frequency gets twenty times higher. What you heard in the cold open is a sound pushed into your range. The original is not.

Here is the same recording, slowed back down, drawn by this channel from NOAA's own file. Nine and a half minutes, ten to forty five hertz. Almost all of the energy sits between twenty and thirty five hertz. Twenty hertz is roughly the floor of human hearing. The Upsweep lives right on the edge of what a person could hear, and mostly just under the threshold of noticing.

One tone, measured 3:29

Zoom into the first eighty seconds and the shape is clear. Each tone starts a little above twenty hertz and climbs about ten hertz in about ten seconds. Then silence, and the next one. We counted them with a detector built for exactly that shape. Thirty two rising tones in nine and a half minutes, a median of fifteen point six seconds apart.

Put it next to the other sounds on this channel. The Bloop, measured in episode two, sat between thirteen and twenty six hertz. The northeast Pacific blue whale, sixteen to seventeen and a half. The fifty two hertz whale, forty nine to fifty three. The Upsweep is in the same low band as all of them, which is why the first guesses were always animals or ice.

54 South, 140 West 4:19

Now the position. Fifty four south, one forty west is in the Southern Ocean, south of French Polynesia and far west of Chile. The page says the source can be roughly located there. Roughly is doing a lot of work. Every instrument that ever heard it was thousands of kilometres away.

So what is at that coordinate? We asked GMRT, the global seafloor synthesis hosted at Columbia's Lamont observatory. Its depth at fifty four south, one forty west is one thousand five hundred and sixty six metres. The seafloor around it is three thousand. The point lands on the flank of a narrow ridge, and five kilometres northeast the grid puts its crest at seven hundred and twenty five metres below the surface.

No ship has mapped it 5:08

Before that ridge goes on a thumbnail, one check. GMRT marks which cells came from a ship's sonar. In a box twelve degrees wide and eight degrees tall around the coordinate, it has none. Not one cell. The ridge is predicted from satellite measurements of the sea surface, which bulges over mass on the seafloor. It is real enough to show up from orbit. Its exact depth could be off by hundreds of metres.

The earthquakes that are not there 5:37

The page also says the position is near inferred volcanic seismicity. We checked the global earthquake catalog of the U.S. Geological Survey, from nineteen seventy three to this week. Within a hundred kilometres of the point, zero earthquakes. The nearest is a hundred and twenty seven kilometres away, a magnitude five on the Pacific Antarctic Ridge in twenty twenty. Within three hundred kilometres, a hundred and fifty eight, almost all of them along that ridge.

So where does the seismicity come from? From the hydrophones themselves. A NOAA paper from nineteen ninety five explains why: sound travels so well in the ocean that hydrophones record seafloor quakes too small for the land networks. The volcano near the Upsweep was never seen. It was inferred, most likely from more sound.

Then the seasons. Spring and fall peaks, and the page is careful. It says it is unclear whether that comes from the source, or from the propagation environment. The ocean's sound channel changes with the seasons. A steady source can look seasonal if the road it travels on changes. That sentence is still the most honest one about the Upsweep.

The equatorial hydrophones 6:55

And the last check the page mentions: the equatorial arrays. From May of nineteen ninety six, NOAA moored six hydrophones in the eastern tropical Pacific, at eight north, the equator and eight south, along ninety five and a hundred and ten west. They recorded from zero to forty hertz, about six and a half thousand kilometres from the source. That is the network the page means when it says, can still be detected.

Online you will also read that whales were ruled out in nineteen ninety six. We could not find a NOAA document that says so. The page itself rules nothing in and nothing out. It says unresolved. When the only primary source is six sentences long, everything else is a retelling.

We listened: station NRS04 7:43

So we listened. NOAA and the National Park Service run a network of ocean noise reference stations, and the raw audio is in NOAA's public archive. Station four sits off O'ahu, Hawaii, its hydrophone nine hundred metres down in water nearly five kilometres deep. It recorded without a break from July twenty fifteen to December twenty sixteen. It is eight thousand six hundred and fifty kilometres from fifty four south.

A person cannot read a year and a half of spectrogram. So we built a detector from the Upsweep's own shape. It scores a rising sweep, twenty two to thirty two hertz in nine seconds, and subtracts the same sweep falling. An earthquake lights up both and cancels. On NOAA's file, it finds thirty two rising tones in nine and a half minutes.

Then we ran it on the station. One four hour file every five days, through the spring and the fall of twenty fifteen and twenty sixteen, when the page says the Upsweep peaks. Fifty five files. Two hundred and twenty hours of the Pacific.

220 hours, no train 8:54

No train. Not one ten minute window looks like NOAA's file. The best score of the whole search was a pulse every ten seconds, broadband, top to bottom of the plot. That pattern is what a ship's seismic air guns look like.

The runner up came on the twenty second of November twenty sixteen. Sweeps around twenty hertz, regular, strong. But they fall instead of rising, and each lasts about a second. That is the shape of the fin whale's twenty hertz pulse. The Upsweep is its mirror image: rising, and ten times longer.

So is the Upsweep gone? This search cannot say that. One omnidirectional hydrophone, eight and a half thousand kilometres away, against Navy arrays that combine many sensors to pull faint sounds out of noise. What it shows is narrower. In two hundred and twenty hours of public data from the middle of the Pacific, the Upsweep is not loud enough to see.

Vailulu'u: a sound that got a visit 9:56

Now look at what it takes to actually close a sound like this. Five thousand one hundred kilometres north of fifty four south, east of Samoa, there is a seamount called Vailulu'u. Its story starts exactly like the Upsweep's.

On the tenth of July, nineteen seventy three, hydrophones in the ocean's sound channel recorded a series of explosions from that area. A sound, and a rough position. Nobody knew yet that there was a volcano there.

Two years later, on the eighteenth of October, nineteen seventy five, a scientist named Rockne Johnson found the seamount. In nineteen ninety five, a swarm of earthquakes, up to magnitude four point nine, came from the same area. And in two thousand, a team mapped it and published what it was: an active volcano, its summit five hundred and ninety metres down. They called the paper The New Samoa.

Then it started doing things fast. Inside its crater a new cone appeared. They named it Nafanua. It grew three hundred metres, from a crater floor a kilometre deep, in less than four years. The paper in the Proceedings of the National Academy of Sciences says it could reach the surface within decades.

Okeanos Explorer, dive 09 11:17

In February twenty seventeen, NOAA's Okeanos Explorer sent its robot into that crater. Dive nine, the twenty fourth of February, down to nine hundred and twenty eight metres. On the science list for the dive was Bill Chadwick, of NOAA's Pacific Marine Environmental Laboratory. The same lab whose page still carries the Upsweep.

Before the robot reached the bottom, a voice came in from shore. Craig Young, of the University of Oregon, one of the authors of the two thousand and six paper.

This is the crater floor. The water was so cloudy that the camera sled above the robot could not see it. The sensors picked up the plume at about seven hundred and twenty metres. Hydrothermal particles, hanging in the crater like smoke in a room. The dive's science lead explained it to everyone watching from shore.

Then the robot climbed Nafanua. Pillow lava, the shape basalt takes when it erupts into cold water. Fresh, almost no sediment, almost no animals. The dive's geologist logged that these flows were less than five years old. They had formed since the last time anyone mapped the crater, in twenty twelve. This was the plan, read out before the dive.

That is what visiting buys. Rock samples you can date. Vents you can measure. A new cone you can compare against the old map. Every sentence that is a guess on the Upsweep's page became a measurement on Vailulu'u, because a ship went there.

The dive ended on the highest point of Nafanua, six hundred and seventy seven metres down. Older pillows here, more sediment, more life. The part of the volcano that had stopped growing.

What would close the Upsweep 14:18

So here is exactly what is missing for the Upsweep. First, a ship's sonar over fifty four south, one forty west, to replace a ridge drawn from orbit with a ridge that was measured. Second, a hydrophone or a seismometer within a few hundred kilometres, for at least one spring and one fall. Third, if there is a vent, a camera on it. None of the three has a public record.

Somebody probably does still hear it. The treaty that bans nuclear tests runs its own hydrophones, one of them off the Juan Fernandez Islands, about five thousand two hundred kilometres from the point. They send sound from one to a hundred hertz to Vienna in real time. Those recordings are not in the public archive we used.

Which leaves the page. Six sentences, unchanged for twenty five years. This is the sound it describes, one more time, as it was published.

The Upsweep is not a monster and it is not solved. It is a position nobody has visited, and a page nobody has updated. The most honest sentence about it is still the last one NOAA wrote: the origin is unresolved.

Next: reading a seafloor map 15:45

Next time, the skill this episode needed. How to read a seafloor map and tell which parts of it were ever measured. Only twenty eight point seven percent of the ocean floor has been mapped to modern standards, by the official count of twenty twenty six. Every map you have seen fills in the rest. We will show you how to find the seams.

Description and sources

Measured by this channel from NOAA PMEL's upsweep20x.wav (28.8 s, sped up 20x = 9 min 36 s): 32 rising tones, a median of 15.6 s apart, each climbing about 10 Hz in about 10 s from just above 20 Hz, the floor of human hearing. At 54 S 140 W the GMRT grid has a narrow ridge to about 725 m below the surface, but not one cell of ship sonar in a 12 x 8 degree box: it is satellite-predicted. The USGS catalog has no earthquake within 100 km since 1973. On station NRS04 off Oahu, 8,650 km away, 220 hours of spring and fall 2015-2016 show no Upsweep train. The contrast is Vailulu'u, Samoa: heard by SOFAR hydrophones in 1973, found in 1975, visited by NOAA's Okeanos Explorer in 2017.

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

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

  • This channel (scripts/find_upsweep.py) — 1991-2001 | NOAA PMEL, spectrogram computed by this channel | public-domain

... and 7 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. The Upsweep audio is NOAA PMEL's published file (sped up 20x), played unaltered. Spectrograms marked 'this channel' are computed by us from that file and from NOAA/NPS station NRS04 (NCEI, public domain); the detector and the 220-hour search are ours (find_upsweep.py). The relief map and profile are the GMRT grid (Ryan et al. 2009, CC BY 4.0). Every moving frame is NOAA Ocean Exploration video from EX1702 dive 09, Vailulu'u, American Samoa, 24-25 February 2017, public domain and credited on screen: a different volcano, never the Upsweep's source. Three cuts play the original control room audio; each window was re-transcribed in isolation twice and used only where both passes matched. Diagrams are drawn in code from the sources listed. Nothing is generated.

#DeepSea #NOAA #OceanSounds #Upsweep #OceanExploration