The loudest part of an iceberg’s birth may come after it has already broken free.
A newly calved block can be tall, narrow and unstable. As it rolls onto its side, the enormous mass of ice pushes against the glacier front and the seafloor through the water. That slow shove can send surface waves through Earth that instruments register hundreds of kilometres away.
These are glacial earthquakes. They are not conventional earthquakes caused by rock abruptly slipping on a fault, and the strongest evidence so far suggests that they happen far more often along part of West Antarctica than standard catalogues showed.
In a new Geophysical Research Letters study, seismologist Thanh-Son Phạm searched Antarctic recordings from 2010 through 2023. The search produced a catalogue of 362 small seismic events around the Amundsen Sea Embayment, most of them previously uncatalogued. Of those, 245 were concentrated near the ocean edge of Thwaites Glacier.
The number is striking, but it needs careful wording. Researchers did not watch 245 icebergs capsize. Only a few detections were directly paired with satellite evidence. The others are most likely capsizing events because their locations, wave character, force directions and timing fit that explanation.
A different kind of earthquake
A tectonic earthquake begins with rapid rupture. Two sides of a fault move, releasing stored elastic strain in rock. The sharp beginning produces familiar compressional and shear waves, followed by surface waves.
An iceberg capsize is a slower mechanical event. When a column-like berg turns, its contact with the remaining ice produces a force that is mostly horizontal and can last for many seconds. The resulting waves are strongest at longer periods than the brittle cracking of ice, but these Antarctic events are still quicker and smaller than many of their better-known Greenland counterparts.
Most events in the new catalogue had broadband surface-wave magnitudes between 2 and 3. Their reconstructed force histories were around 15 seconds. Large Greenland glacial earthquakes detected across the world can have magnitudes near 5 and force histories lasting roughly 50 seconds.
That difference is one reason the Antarctic signals remained hidden. They were not absent. The searches had not been tuned to hear this particular population.
How the hidden signals were found
Phạm used recordings from permanent and temporary seismic stations operating across West Antarctica. The detector looked for coherent Rayleigh surface waves with periods from 17 to 25 seconds at stations within 1,100 kilometres of a possible source.
The method is called delay and stack. A wave leaving one point reaches each station at a slightly different time. The algorithm shifts the recordings by the travel times expected from a candidate location, then adds them together. Random noise tends to cancel while a wave shared across the network becomes clearer.
The analysis used a regional surface-wave speed of about 2.95 kilometres per second. That is substantially slower than the value used in some earlier searches. It also focused on shorter periods than a recent 30-to-100-second search.
Both choices mattered. Waves in the 17-to-25-second range attenuate over long distances, so a global network may struggle to detect them. A detector built around the wrong wave speed can also misalign the signals instead of reinforcing them.
After automatic detection, Phạm excluded poorly covered offshore candidates and manually assessed the remaining records. The final set contained 362 events. Only five were listed as earthquakes in the International Seismological Centre catalogue, and only nine appeared in another recent low-frequency catalogue.
Why 245 point to capsizing at Thwaites
The strongest clue is geography. Most of the Thwaites events sat close to its terminal coastline, especially the central ice tongue and the western ice mélange. These are places where glacier ice reaches the ocean, fractures and produces new bergs.
The central tongue is not a smooth, single plate. It contains conglomerated icebergs held together and constrained by fast sea ice, with contact from an ice shelf to the east and small submerged islands to the northwest. That crowded geometry can keep an unstable block upright until the surrounding support changes.
A second clue came from modelling the force behind each signal. For about 60 percent of the catalogue, seven stations supplied enough three-component data to attempt a centroid single-force inversion. At Thwaites, the inferred horizontal directions generally aligned with the glacier’s flow.
That is the pattern expected from a capsizing iceberg pushing on its parent glacier. Similar alignments have been measured in Greenland and in the few Antarctic glacial earthquakes already confirmed.
The force modelling is not decisive on its own. Source shape and recovered parameters trade off against one another, and 40 percent of events lacked the data required for an inversion. The study therefore treats orientation as supporting evidence, not a fingerprint that identifies every capsize beyond doubt.
Three events that satellites could check
A detection on 5 November 2016 offered a rare visual test. Satellite images acquired around the event revealed at least four new bergs near the seismic location. Two had remained upright and two had overturned, exposing fresh ice with a different reflectance and infrared temperature.
The larger overturned berg was the likely source. After refining the seismic location, the two estimates agreed to within about 10 kilometres. Given the sparse station geometry and scale of the region, that was a useful validation of the travel speed and relocation method.
The catalogue also recovered two glacial earthquakes from 8 November 2018. Those had been tied to satellite-observed capsizing in a previous study of Thwaites calving. That earlier work also found six days of higher-frequency cracking before the capsizes.
These checks establish that the detection method can find real capsizing events. They do not turn every nearby signal into a visually confirmed capsize. Clouds, darkness and the spacing of satellite passes leave many events without a matching image.
The busy years matched a faster ice tongue
The Thwaites detections did not follow a clear seasonal rhythm. Instead, the frequency climbed during 2018 through 2020, with the event count peaking in 2018.
More instruments were not the simple explanation. Seismic station coverage within the study’s regional radius was greatest in 2016, two years before the event peak.
The timing instead tracked an independent change seen by satellites. During the roughly three-year busy interval, the floating central tongue was moving about 20 to 40 percent faster than in the periods immediately before and after it. The inland sampling point did not show a significant speed change.
The satellite analysis had already described this as episodic dynamic change linked to damage on the Thwaites ice tongue. Its authors found that speed-ups between 2000 and 2018 coincided with structural weakening and progressively greater fragmentation.
Phạm’s seismic catalogue adds a second, independent record. When the tongue sped up, the number of long-period events near its edge also rose. That agreement is part of the case that the signals record bergs overturning rather than unrelated earthquakes beneath the glacier.
How an ice mélange can hold a berg upright
The proposed mechanism begins with back pressure. Sea ice and fragmented glacier ice packed against a terminus form a mélange. When it is strong and tightly confined, the mixture presses against the glacier front.
That pressure can deter fractures from opening and help restrain a newly detached iceberg. Even a top-heavy berg may remain standing if the surrounding ice gives it nowhere to rotate.
A faster tongue could weaken the fast sea ice binding the conglomerated blocks, reduce the back force and increase fragmentation. Smaller pieces would have more freedom to move. Newly separated tall bergs could then roll, each producing the horizontal shove that generates a glacial earthquake.
This is a plausible physical account of the correlation, not a completed chain of proof. The external ocean forcing behind the 2018-to-2020 speed-ups is still not well understood. The catalogue shows that the seismic burst and tongue acceleration occurred together; it does not isolate which process initiated each change.
What 245 events do not measure
A seismic magnitude cannot be converted cleanly into tonnes of calved ice. The signal depends on the berg’s shape, how far it rotates, what it strikes and how the force is transmitted. A smaller mass moving violently may not resemble a larger mass turning slowly.
The study explicitly stops short of calculating total ice loss from the 245 detections. Its algorithm also used empirically chosen thresholds and template lengths, while local weather and seasonal noise can change how easily weak signals are detected. The catalogue is systematic, but it cannot be assumed complete.
There is another distinction worth keeping clear. Ice that calves from a floating tongue is already displacing seawater, so the capsize itself does not directly raise sea level. As ScienceBlog noted in coverage of the Petermann Glacier breakup, the greater concern is what the loss of floating ice may allow land-based ice to do next.
Floating tongues and shelves can buttress the glacier behind them. If weakening reduces that resistance, ice resting on land can flow into the ocean more quickly and add to sea level. The new earthquake count is therefore a record of active fragmentation, not a direct sea-level gauge and not a forecast that Thwaites is about to collapse.
Pine Island produced a different mystery
The same method found 66 events around nearby Pine Island Glacier, the catalogue’s second-largest cluster. At first glance, their magnitudes and force orientations resembled the Thwaites events.
Their locations did not. They consistently clustered near Pine Island’s grounding line, where the glacier leaves the bed and starts to float, roughly 60 to 80 kilometres from the waterfront. The median location error was about 16 kilometres.
That gap is too large to dismiss as ordinary uncertainty. An independent high-frequency study had also found events in a similar inland area. Capsizing icebergs cannot plausibly explain signals so far from the calving front.
Basal sliding or another transient process near the grounding zone may be responsible, but the paper leaves the physical source unresolved. The Pine Island result is a useful restraint on the broader interpretation: not every long-period glacier signal has the same cause.
A new way to listen between satellite passes
Satellite imagery remains essential because it can show where ice broke, rotated and drifted. Radar and velocity products reveal deformation and movement across a wide area. Seismology supplies something different: continuous timing, including during polar night and between usable images.
The recordings used here were not collected specifically to count capsizing icebergs. Much of the value came from searching an existing regional archive with a detector matched to shorter Antarctic surface waves.
That leaves room for improvement. Denser stations could sharpen locations and force estimates. A catalogue built with better-calibrated completeness could test changes through time more confidently. The physics-based detections may also provide examples for machine-learning searches of a much larger polar archive.
For now, the study’s contribution is narrower and sturdy. It turns two previously confirmed Thwaites glacial earthquakes into evidence for a much more active class of small events. It also shows where certainty ends: three satellite-supported capsizes, 245 coastal detections with a strong common explanation, and a correlation with ice-tongue speed that is revealing without yet being causal proof.