AI detects earthquake surge beneath Europe’s most dangerous volcano

Italy’s Campi Flegrei has been restless again when it comes to earthquakes, and a new AI approach is helping scientists see what the ground is really doing.

In a peer-reviewed study, researchers from Stanford University, along with colleagues at INGV Osservatorio Vesuviano and the University of Naples Federico II, reanalyzed continuous seismic records. The team found more than four times as many small earthquakes between January 21, 2022, and March 20, 2025.

The researchers did not detect any direct sign of magma moving upward. Instead, they highlighted very shallow hybrid quakes tied to the hydrothermal system.

That fuller picture matters because small events reveal where the crust is stressed and how energy is moving through the volcanic field. With cleaner detections and sharper locations, scientists can track changes faster and judge when activity starts to cluster in risky ways.

Clear signals, urgent stakes

“Seismicity could change at any time, and that may be the most important thing about this study: this capability of getting a clear view is now operational,” said study co-author Greg Beroza, a geophysics professor at Stanford University.

The caldera sits within the Neapolitan area, where more than half a million people live and work. That density means even modest shaking or ground movement can disrupt daily life, strain buildings, and complicate evacuations.

History shows why clarity is essential. During the 1982-1984 crisis, uplift and persistent earthquakes damaged structures across Pozzuoli. A major analysis documented that the events triggered the evacuation of about 40,000 people.

AI turns noise into earthquake signals

The team trained a machine learning workflow on continuous waveforms and used it to scan years of data consistently. That created one catalog rather than a patchwork of methods, so patterns stand out instead of becoming blurred.

AI did more than count earthquakes. The workflow sharpened event locations and magnitudes, turning noisy curves into clean signals that line up with mapped features and known hot zones.

That is the kind of precision emergency managers need when activity jumps in hours, not weeks. One payoff is a much clearer view of the caldera’s thin ring shaped structures and shallow faults.

Those features organize the small quakes and help separate background rumble from changes that could matter for public safety.

Volcano under quiet strain

Campi Flegrei is a large caldera, the broad depression left by past eruptions, ringed by faults and dotted with old vents and fuming ground. It is not a single cone, it is a field where different parts can wake up at different times.

The region breathes over years through uplift and subsidence known as bradyseism. Those slow vertical motions have been recorded for centuries and often line up with swarms of small earthquakes.

In the new catalog, most activity sits at very shallow depths and traces a narrow ring fault system consistent with surface features and the zone now rising.

The authors reported no direct signature of magma rising toward the surface in this interval. This lowers near-term eruption concern but does not remove the need for close watching.

Faults reveal hidden stress

Faults are breaks in rock that slip when stress overcomes strength, and their length and orientation constrain how big an earthquake could be. When small quakes line up along a longer fault, planners can bound the upper size of plausible events rather than guess in the dark.

Seeing faults crisply also helps compare present unrest with past episodes. Shaking that migrates onto a new segment or clusters at the edge of the ring may signal a change in how the system stores and releases pressure.

Those signals are not predictions by themselves. They are inputs to models that weigh where stress is building, how close rocks are to failure, and what level of shaking the built environment can handle.

AI refines risk planning

Near real-time detections let observatories update maps of active zones as conditions evolve. That supports decisions like where to inspect bridges, how to stage ambulances, or when to test public alerts before a busy weekend.

Sharper locations and magnitudes also help engineers validate expected shaking levels against what instruments record in town. With that feedback loop, officials can refine microzonation, revisit building priorities, and tune maintenance schedules before the next swarm.

Past experience is a warning here. During 1982-1984, the harbor shallowed and commerce paused, schools closed, and the government approved new housing to relocate residents.

Planning for unrest is not only about eruptions, it is about everyday disruptions that stack up when the ground keeps moving.

AI helps, not predicts

The present results point to pressure within a shallow system, not direct magma rise, during 2022-2025. That aligns with other independent work that has argued for pressure buildup and evolving stress in the crust during long stretches without eruption.

Unrest can still change quickly. More small quakes could organize into tighter clusters, or uplift rates could shift, and those changes would be visible fast with the new tools.

AI does not replace dense earthquake monitoring or local expertise. It helps analysts keep up when data surge, flag subtle changes that a human might miss after a long night, and focus attention on the signals that matter most.

The study is published in the journal Science.

—–

Like what you read? Subscribe to our newsletter for engaging articles, exclusive content, and the latest updates. 

Check us out on EarthSnap, a free app brought to you by Eric Ralls and Earth.com.

—–