A Volcanic Seismic Crisis Comes Into Sharper Focus
The intense seismic sequence that affected Santorini and neighboring islands between late 2024 and mid-2025 was far larger and more structured than conventional monitoring had shown. According to research presented at the 2026 annual meeting of the Seismological Society of America, a machine learning-based analysis identified more than 60,000 earthquakes during the episode, creating a high-resolution catalog that allowed scientists to follow the crisis as it unfolded.
The headline number alone is striking, but the broader significance lies in what this kind of near-real-time analysis makes possible. Instead of reviewing data months later, the research team used machine learning tools while the event was ongoing, detecting bursts of seismicity, tracing how activity migrated through fault networks and capturing details that standard workflows would likely have missed or delivered too slowly to guide operational decisions.
Why the Santorini Sequence Was Hard to Track
Santorini’s seismic crisis was unusually intense. Researchers said some periods contained hundreds of earthquakes within a single hour, a scale that makes conventional cataloging difficult under time pressure. That volume matters because when earthquakes come in dense clusters, the challenge is not only to measure magnitude or location but to separate individual events from overlapping waveforms and to do so quickly enough for the information to remain actionable.
Using machine learning pipelines running in parallel, the Stanford-led team was able to process large quantities of waveform data and identify thousands upon thousands of events during the crisis itself. The resulting dataset stretched from December 2024 through June 2025 and offered a much more detailed picture of how the sequence evolved over time.
This is an important operational shift. Machine learning in seismology is often used retrospectively, after a crisis has passed. In Santorini, the methods were deployed in a way that approached real-time monitoring. That makes the work notable not just as a study of one earthquake sequence, but as a demonstration of how volcano-related seismic crises could be handled differently in the future.
Bursts, Migration and Evidence Pointing to Magma Movement
The catalog identified 46 recurring bursts of seismicity, each involving hundreds of earthquakes over one to two hours. During some bursts, seismic migration moved along fault zones at speeds of up to 2 kilometers per hour. Those patterns are more than descriptive curiosities. They help scientists evaluate the underlying process driving the swarm.
According to the researchers, both the speed and the migration pattern strengthen the interpretation that the sequence was linked to magma intrusion associated with the region’s volcanoes. In other words, the earthquakes were not simply scattered tectonic noise. They appear to have traced the movement of material and stress through an active volcanic system.
That distinction matters for hazard assessment. In volcanic settings, whether a swarm is driven mainly by fault slip, fluid movement or magma intrusion changes how scientists think about escalation risk and public communication. A richer catalog does not remove uncertainty, but it can narrow the range of plausible explanations and help authorities build a clearer situational picture.
From Research Tool to Operational Expectation
One of the strongest messages from the study is institutional rather than purely geological: researchers argue that these methods should move from limited use into routine operational practice. That is a consequential claim. Monitoring agencies are often cautious about adopting new analytic methods in real-time workflows because reliability, speed and interpretability all matter when public safety is involved.
But events like Santorini expose the limitations of status quo approaches. When a crisis is evolving rapidly, delays in analysis are not academic inconveniences. They can affect forecasting, warnings and emergency planning. The researchers’ position is that machine learning has matured enough to become part of the standard monitoring toolkit, especially in high-tempo volcanic crises where human analysts alone may struggle to keep up with event volume.
If that transition happens, the practical implications could extend far beyond the Aegean. Volcano observatories and seismic monitoring networks worldwide face similar challenges during swarms, intrusions and earthquake cascades. Faster and denser event catalogs could improve how agencies interpret unfolding hazards and communicate uncertainty to the public.
What This Study Changes
The Santorini sequence has become a case study in how computation can change observational science during a live event. The value was not just that machine learning found more earthquakes. It found structure: repeated bursts, migrating activity and fault-network detail that together produced a more coherent story about what was happening underground.
That is the deeper lesson. In hazard science, better resolution can change the meaning of the event itself. A diffuse and overwhelming swarm becomes a mapped process with rhythms, pathways and likely drivers. That does not make prediction easy, and it does not eliminate the possibility of surprise. But it does improve the quality of information available when decisions have to be made in real time.
For Santorini, the outcome is a clearer record of a remarkable seismic crisis. For the field more broadly, it is a sign that operational seismology may be entering a new phase in which machine learning is no longer a post-event research assistant, but a front-line analytical tool.
Key Takeaways
- Researchers identified more than 60,000 earthquakes during the 2025 Santorini sequence using machine learning.
- The study detected 46 recurring bursts of seismicity and migration along faults at speeds up to 2 kilometers per hour.
- The observed patterns support the interpretation that magma intrusion played a central role in the crisis.
- The team argues that such machine learning methods should become part of routine real-time monitoring during volcanic events.
This article is based on reporting by Phys.org. Read the original article.
