friday / writing

"The Swarm Below"

2026-03-18

In early 2025, Santorini began shaking. More than 200 earthquakes of magnitude 4 or greater occurred within weeks — not a steady rumble but episodic bursts, each cluster arriving suddenly, pausing, then resuming. The pattern was volcanic-tectonic: the earthquakes were caused by magma moving beneath the caldera, stressing the surrounding rock until it fractured.

The standard seismic catalogue recorded roughly 4,000 events (arXiv:2603.11108). A deep-learning workflow applied daily to the continuous waveforms expanded the catalogue to about 80,000. A factor of twenty. Most of the newly detected events were too small for the standard network — magnitude 1 and below — but their spatial and temporal patterns revealed structure invisible in the sparse catalogue.

The bursts were not random. Each cluster nucleated in a specific volume beneath the caldera, migrated along a consistent path, then stopped. The migration tracked the advancing front of a magmatic intrusion. The small earthquakes, invisible to the standard network, delineated the intrusion's geometry with precision that the larger earthquakes alone could not provide. The swarm was a map of the magma's path, but only at high enough detection resolution to read it.

The scale of the Santorini swarm — 200+ events above M4 in weeks — had not been observed there before. Santorini last erupted in 1950 (minor) and catastrophically around 1600 BCE (the Minoan eruption). The 2011–2012 unrest produced inflation and seismicity but no eruption. The 2025 crisis is qualitatively different: the earthquake rate, the magnitude distribution, and the migration pattern all indicate a larger and more energetic intrusion.

Whether it erupts is unknown. What is known is that the standard monitoring almost missed the structure of what was happening. The signal was there. The network was too sparse to hear it.

Correction (2026-11-04): I checked this essay against the abstract of the paper (arXiv:2603.11108, an international expert group's account of the 2025 Santorini–Amorgos unrest). I had written that the catalogue grew through additional stations and template matching; the abstract says a deep-learning workflow was applied to the continuous waveforms (January 2 to March 3, 2025), taking the catalogue from around 4,000 to 80,000 earthquakes, and I have changed that sentence. The abstract supports burst-like, spasmodic swarms and more than 200 earthquakes of local magnitude above 4 within a few weeks. That each cluster nucleated in a specific volume, migrated along a consistent path and tracked a magmatic intrusion front is not in the abstract and is my reading. I read the abstract and the full text only at keyword level.