Persistence · the daily reading
Built to hear Soviet submarines, the U.S. Navy's seafloor ears spent forty years recording the loudest biological signal in the ocean — and filed it under noise. The story of how that record was finally read is also, it turns out, the story of how we were quietly closing the ocean it described.
There is a kind of archive that is most complete precisely when no one is reading it. The Navy's Sound Surveillance System — SOSUS, a chain of hydrophones bolted to the deep seafloor and wired back to coastal listening rooms — was such a thing. For roughly four decades it heard, recorded, and catalogued the low moan of the largest animals that have ever lived, and for nearly all that time the people doing the listening were trained specifically not to hear them. The whales were in the data the whole time. They were the part you were supposed to ignore.1
That distinction — between a record being made and a record being read — is the one I cannot let go of in this story, the one the MIT Press Reader's fine adaptation of David Rothenberg's Whale Music gestures at but does not quite stop on.18 An archivist spends a working life on the gap between those two verbs, because almost everything that is ever lost is lost in that gap: not destroyed, just unread, sitting legible in a box that no one with the right eyes ever opens. What is unusual about SOSUS is that the box was eventually opened, by the right eyes, and we can date the morning it happened.
The morning was in 1992. Christopher Clark, a bioacoustician from the Cornell Laboratory of Ornithology — a building devoted to birds, a few hundred miles from the nearest sea — was led into a room "the size of a gym," he recalls, lined with dot-matrix printers spooling out paper covered in dashes and dots: spectrograms, the visual shorthand of sound, scrolling off the hydrophone arrays of the world's oceans in something close to real time.1 Clark looked at one scroll and saw, near the bottom of the frequency scale, a shape he knew in his body: the signature of a blue whale. Then he walked down the row, comparing printouts from arrays moored hundreds of miles apart, and felt the back of his neck go cold. The separate machines were drawing the same whale. The Navy's net could locate a single animal singing across an entire ocean basin, hour by hour.1
Here is the part that rearranges the story. When Clark told the Navy's analysts what he was seeing, they were not surprised. They had been hearing it for forty years. "In every training manual that I've found for Navy technicians," he says, "every image I've seen, every spectrogram, in the background, there were always whales."1 The sonar crews had a working vocabulary for the animals they were taught to discard — the "jezz monster" was their name for the summer crescendo when the fin whales of a whole sea came into voice at once, a chorus dense enough to make the submarines harder to find. It was, to them, a nuisance in the signal. They filed all of it, ocean by ocean, under a single dismissive heading: biologicals. Sounds of no strategic value.1
So the recording was never the problem. The recording was immaculate — a continuous, basin-scale, multi-decade acoustic census of the deep ocean, far longer and wider than any instrument science could have afforded to build for its own purposes. What was missing was not data but attention: a reader who would look at the discarded layer and see a text. For forty years the archive was perfect and unread, sealed twice over — once by classification, and once by the simple fact that the men holding it had been told what it meant and the telling was wrong.
To see why the Navy could hear a whale from a thousand miles away — and why the whale, arguably, built its whole life around being heard from a thousand miles away — you have to go down to a particular layer of the sea. Sound speed in water is not constant; it falls as the water cools with depth, then rises again as pressure mounts toward the bottom. Somewhere in between, in temperate and tropical seas roughly a kilometre down, there is a depth where sound travels slowest. A ray of sound that strays above that depth is bent back down; a ray that strays below is bent back up. The layer behaves as a waveguide — a horizontal pipe with no walls — and low-frequency sound trapped inside it can travel for thousands of kilometres with almost no loss.2
This layer has a name that records its dual life. To the physicists who found it during the Second World War — Maurice Ewing and J. Lamar Worzel, who in the spring of 1944 dropped explosive charges from one ship and heard them, sharp and late, on a hydrophone three hundred miles off — it was the SOFAR channel, for SOund Fixing And Ranging. The original idea was a rescue tool: a downed airman could drop a small charge into the channel and shore stations a thousand miles apart would each hear the boom and, from the timing, fix his position.2 A few years later the same physics was turned the other way. If a small charge in the channel could be heard a thousand miles off, so could a Soviet submarine. Out of the 1950 "Project Hartwell" summer study came the recommendation to wire the channel for surveillance; the Office of Naval Research contracted Western Electric, the system was built under the cover name "Project Caesar," and the first full array — forty hydrophones in a thousand-foot line — went onto the bottom off Eleuthera in January 1952. The shore stations were disguised as oceanographic research posts. They kept that cover for thirty-nine years. The mission was declassified in 1991.3
The thing I want to mark is the symmetry. The whale did not stumble into the Navy's channel. The fin whale's 20-hertz pulse, the blue whale's long moan — these are tuned, at their frequency and their depth, to the one waveguide in the ocean that carries sound the farthest. Whatever the song is for, the animals are using the same physical fact the Navy spent billions to exploit, and they were using it for several million years before Ewing dropped his first charge. When SOSUS heard the whales, it was not overhearing a private signal. It was two listeners discovering they had moved into the same room.
An archive comes with a provenance — a record of where each item was made, with what instrument, under what conditions — and the provenance is what lets you trust the contents. Strip it, and even a perfect document becomes hard to read: you can see the marks but you cannot be sure what they measure. The SOSUS corpus is the strangest case of broken provenance I know, because the break is deliberate and, in part, permanent.
When the dual-use programs of the early 1990s — Clark credits a push by senators Al Gore, Sam Nunn, and Ted Kennedy to turn Cold War instruments toward civilian science — let biologists into the data, the biologists inherited a record whose own specifications were still secret.1 Rothenberg puts the load-bearing caveat in a single sentence: just how precisely the system can locate a sound "is one aspect of the technology that remains classified."1 Sit with what that does to a scientific result. A biologist can publish that a particular blue whale was tracked for weeks across a basin — Clark describes following one for 43 days, from five hundred miles northeast of Bermuda down to within a hundred miles of Cuba and partway back, roughly 2,200 miles in all1 — but the error bars on that track depend on array parameters the analyst is not cleared to state. The conclusion is real. The full provenance of the conclusion is sealed. You are asked to trust a ruler whose markings you are not allowed to see.
This is not a complaint about secrecy so much as a note about what kind of knowledge results. A great deal of marine bioacoustics from the 1990s rests, in part, on an instrument whose calibration is a state secret — which means the field carries, structurally, a footnote it can never fully discharge. That is rarer than it sounds. Most broken provenance in an archive is an accident: a label fell off, a logbook burned. Here the missing line is held back on purpose, by an owner who is still alive and still declines to supply it. The record can be read. It cannot be fully audited. Those are different things, and an honest catalogue has to say which one it is offering.
There is a second reason the whales hid in plain sight for forty years, and it has nothing to do with classification. The signal is pitched below us, and clocked slower than us, on both axes at once. The fin whale's call is a pulse near 20 hertz — at or beneath the floor of human hearing — repeated about every three seconds; its biological origin was finally pinned down by William Schevill, William Watkins, and Richard Backus in 1964, after years in which the regular thuds had been blamed on geology, on equipment, even on a suspected Soviet scheme to seed the ocean with standing waves.4 The blue whale's note is lower still and far slower: one long moan of up to half a minute, then a wait — Rothenberg gives 70 seconds in the Atlantic, 140 in the Indian Ocean — then the same note again, held to a rhythm so lax that, as he says, no human musician could keep time counting that slowly.1 To hear it as a song at all, you have to speed the tape up ten or thirty times and lift the pitch into a range your ear admits. Only then does the structure surface: a phrase, a silence, the same phrase.
I find this the most quietly radical part of the whole episode, because it is the same problem the deep-time sciences were built to solve. A stratum, a varve couplet, an ice-core layer — these are records written too slowly for an unaided human to perceive as a record at all; you need an instrument and a change of timescale to read them as text rather than landscape. The thousand-mile song is that problem moved into sound. The whales are writing on a scale of time and frequency outside the human window, and reading them requires exactly what reading a sediment core requires: an apparatus that re-scales the signal until it falls inside a human sense. We did not lack the whale's song for forty years because it was faint. We lacked it because it was slow, and low, and we had not yet built — or been let near — the instrument that makes the slow legible.
The hypothesis that organizes all of it is older than the access. As early as 1971 Roger Payne — who with Scott McVay had just shown the world that humpbacks sing structured songs — worked out, with the physicist Douglas Webb, that the lowest calls of the largest whales could in principle carry across an entire ocean basin, hundreds to thousands of kilometres, and proposed that this long-range signalling might be the point of the calls.5 The SOSUS data, opened twenty years later, is what let anyone test the idea against real animals at real range. The "thousand-mile song" is Payne and Webb's 1971 arithmetic, finally given an ear wide enough to check it.
And here is the turn that makes this, to me, more than a good declassification story. Payne and Webb's 1971 calculation came with a condition written into the physics: the song travels a thousand miles in a quiet ocean. The range of any signal is set by how far it stays above the background noise, and their numbers assumed the low-frequency background of a pre-industrial sea.5 That sea no longer existed by the time the data came out — and we were the ones who ended it.
The dominant source of low-frequency sound in the modern ocean is not weather or geology. It is propellers. Across the second half of the twentieth century, as the world's merchant fleet grew and grew, the ambient noise of the deep ocean in exactly the band the great whales use — tens of hertz — rose steadily. Repeat measurements off California, the cleanest long record we have, put the increase at roughly three decibels per decade through the latter part of the century: in that 20-to-50-hertz band, something like a ten-to-hundredfold rise in noise power over four decades, attributed chiefly to commercial shipping.6 The whales' channel — the SOFAR waveguide that carries their voices so efficiently — carries ship noise just as efficiently, and from far more sources. We did not just add noise to the ocean. We added it to the precise pipe the song travels down.
So put the two facts side by side. The instrument that finally let us hear the thousand-mile song was a Cold War surveillance net switched to science in 1991. The ocean it let us hear into was, by 1991, already tens of decibels louder in the relevant band than the ocean Payne and Webb had assumed in 1971. We obtained the means to document a basin-scale acoustic behaviour at roughly the same historical moment we were, with our own engines, shrinking the range over which that behaviour can work. The closing of the loop is almost too neat to bear: the very scientist who walked the printout room in 1992, Christopher Clark, is lead author of the 2009 paper that put numbers to the loss — quantifying how rising ship noise contracts the "communication space" of baleen whales, the volume of ocean across which a call can still be heard.7 The man who first read the archive went on to measure the world the archive had recorded, and found it smaller.
This is the form of value an archive has that nothing else does. It is not only a store of what happened; it is, sometimes, the only surviving witness to conditions that no longer obtain. The SOSUS tapes from the 1950s and '60s are recordings of a quieter ocean — a sound field we cannot remeasure, because it is gone, because we replaced it. Whatever the thousand-mile song was at full strength, in a sea without container ships, those reels may be the closest thing to a record of it that will ever exist. The Navy kept them to hunt submarines. They turn out to be a baseline for a silence we did not know we were spending.
The question the biologists are still left with is whether the whales were ever really listening at those ranges — whether the thousand-mile song is communication, or navigation, or a slow sonar bounced off seamounts, or simply a loud call that happens to carry far whether or not anyone a thousand miles away attends to it. Clark is honest that the romance outruns the data here, and so am I.1 But the archivist's share of the story does not depend on settling it. The record was made, kept, and nearly never read; it was opened by accident of policy; it can be read but not fully audited; and it preserves a condition of the world we have already foreclosed. That is four different ways for knowledge to be fragile, stacked in one box of tapes — and the only reason we can name any of them is that somebody, against the whole training of the institution that built it, finally looked at the part that was supposed to be noise.