Enhanced Seismicity Monitoring in the Rapid Scientific Response to the 2025 Santorini Crisis

By applying a deep learning workflow to seismic data from the 2025 Santorini-Amorgos crisis, researchers expanded the earthquake catalogue from 4,000 to 80,000 events, revealing unprecedented fluid-driven volcanic-tectonic swarms and identifying a new deep magmatic reservoir beneath Anydros Islet.

The Gist

Here is an explanation of the paper, translated into simple, everyday language using analogies to help visualize what happened.

The Big Picture: A Volcano's "Heart Attack"

Imagine the island of Santorini (famous for its stunning sunsets and ancient history) as a giant, sleeping beast. In early 2025, this beast started having a very intense, rapid heartbeat. The ground began shaking violently, causing panic among the 15,000 residents and the millions of tourists who visit every year.

The Greek government declared a state of emergency. The big question on everyone's mind was: "Is the volcano about to erupt, or is this just a tectonic muscle cramp?"

To answer this, a team of international scientists (like a "super-team" of doctors) rushed in. But they faced a problem: the shaking was so fast and chaotic that their standard tools were like trying to count raindrops in a storm with a bucket. They were missing almost everything.

The Solution: The "AI Super-Scanner"

The team decided to use Artificial Intelligence (Deep Learning) to act as a super-sensitive stethoscope.

• The Old Way (The Bucket): Traditional methods could only "hear" about 4,000 earthquakes during the crisis. It was like trying to count a crowd of people by only looking at the ones wearing bright red hats.
• The New Way (The AI Scanner): The AI listened to the raw sound waves of the earth. It found 80,000 earthquakes. It didn't just see the "red hats"; it saw the people in the back, the people whispering, and the people moving in the shadows.

The Result: By finding 20 times more earthquakes than usual, the scientists could finally see the pattern of the shaking, not just the noise.

What the AI Revealed: The "Spasms"

Once they had the full list of 80,000 events, a clear picture emerged. The shaking wasn't random. It happened in explosive bursts, like a series of violent hiccups or spasms.

• The Analogy: Imagine a balloon being slowly inflated. Suddenly, the air pushes hard against a weak spot, causing a loud pop, then another pop, then a rapid series of pops as the air rushes through a crack.
• The Science: These "hiccups" (seismic bursts) told the scientists that fluids were moving underground. It wasn't just rocks grinding against rocks (which is normal tectonic shaking); it was magma and high-pressure water/steam pushing through cracks in the Earth's crust.

The Diagnosis: "It's the Plumbing, Not the Engine"

The scientists used three main tools to figure out exactly what was happening:

1. The "Fingerprint" (Moment Tensors):
   Every earthquake leaves a unique "fingerprint" showing how the rocks broke. The scientists found that many of these fingerprints were weird and complex. They didn't look like a simple rock sliding past another rock. They looked like a crack opening up or a cavity collapsing. This confirmed that magma or super-heated fluids were forcing their way through the ground.

2. The "X-Ray" (Tomography):
   Using the 80,000 earthquakes as flashlights, the team built a 3D X-ray of the underground.

   • They found a deep magma reservoir (a pool of molten rock) under a small island called Anydros.
   • They found that the "plumbing" connecting the main Santorini volcano and the nearby Kolumbo submarine volcano was active.
   • Crucially: The magma was deep (about 8km down). It was pushing up, but it wasn't close enough to the surface to cause an eruption right now.

3. The "Map" (Migration):
   The earthquakes didn't stay in one spot. They started near the Kolumbo volcano and then migrated (moved) like a wave, traveling 20 kilometers northeast along a fault line, passing under Anydros Island. This movement proved that the pressure was traveling through a specific channel, like water flowing through a pipe.

The Verdict: A Scary "False Alarm" (But a Real Warning)

By mid-March, the shaking slowed down. The "hiccups" stopped.

• Did it erupt? No. The magma stayed deep underground.
• Was it dangerous? Yes, but not because of an eruption. The danger was the shaking itself (which could damage buildings) and the potential for a tsunami if a large fault broke (like the massive 1956 earthquake in the same area).
• The Takeaway: The crisis was a volcanic-tectonic event. Deep magma pushed up, pressurized the rocks, and caused a massive swarm of earthquakes. It was a warning shot from the Earth, showing that the plumbing system is still very active.

Why This Matters for the Future

This event was a test run for the future.

• Before: We were blind to small earthquakes. We only saw the big ones, so we couldn't predict the pattern.
• Now: We have an AI-powered early warning system. We can see the "hiccups" before they become a "heart attack."

The paper concludes that we need dedicated volcano observatories in places like Santorini. We need permanent underwater sensors and real-time AI monitoring to catch these "hiccups" instantly, so we can tell the difference between a harmless tremor and a prelude to a disaster.

In short: The Earth gave Santorini a massive scare in 2025. Thanks to AI, scientists could listen to the Earth's "heartbeat" clearly enough to say, "It's a plumbing issue, not an explosion. Stay calm, but keep watching."

Technical Summary

Here is a detailed technical summary of the paper "Enhanced Seismicity Monitoring in the Rapid Scientific Response to the 2025 Santorini Crisis."

1. Problem Statement

In early 2025, the Santorini volcanic field (Greece) experienced a rapid and intense seismic crisis characterized by a swarm of earthquakes between Santorini, Kolumbo, and Amorgos islands. The event posed significant challenges for traditional monitoring:

• Data Limitations: Standard seismic catalogues (relying on STA/LTA algorithms) detected only ~4,000 events, missing the vast majority of microseismicity due to high cultural noise, overlapping waveforms, and uneven station distribution.
• Urgency: The crisis triggered a state of emergency, causing public panic and economic disruption. Rapid, high-resolution characterization of the seismicity was critical to distinguish between tectonic faulting and magmatic intrusion (potential eruption risk).
• Complexity: The region involves a complex interaction between the Santorini caldera, the submarine Kolumbo volcano, and the Anydros Islet, making it difficult to determine the source mechanism and migration path of the unrest using standard methods.

2. Methodology

The study employed a multi-stage, near real-time workflow integrating deep learning (DL) and advanced seismological techniques:

• Deep Learning Detection (QuakeFlow):
  • Utilized PhaseNet (a deep neural network) for P- and S-wave phase picking on continuous waveforms from 24 seismic stations.
  • Used GaMMA (Bayesian Gaussian Mixture Model) to associate picks into earthquake events.
  • Processed data daily from January 25 to March 3, 2025, via cloud computing (AWS) to enable near real-time analysis.
• Catalogue Enhancement (Template Matching):
  • Applied EQCorrscan using the DL-detected events as templates to scan continuous waveforms.
  • This similarity-based approach resolved overlapping events and detected low-magnitude earthquakes that DL alone missed, effectively doubling the detection rate during peak swarm activity.
• Source Characterization:
  • Moment Tensor Inversion: Calculated full and deviatoric moment tensors for $M > 3.8$ events using the FOCI approach (P-wave first-motion amplitudes) to identify non-double-couple (non-DC) components.
  • Long-Period (LP) Search: Used Frequency Index (FI) analysis and waveform stacking to search for LP events and volcanic tremor.
  • Global Analogues: Compared the seismicity rates and patterns to historical dyke intrusion events (e.g., Kilauea, Mayotte, Bardarbunga).
• Tomographic Imaging:
  • Performed a non-linear Bayesian trans-dimensional inversion to jointly estimate P-wave velocity ($V_p$), $V_p/V_s$ ratio, and earthquake locations.
  • Used ~13,000 well-located events from the enhanced catalogue to image crustal structures and identify melt/fluid reservoirs.

3. Key Contributions

• Scale of Detection: Enhanced the earthquake catalogue from 4,000 (standard) to **80,000 events** (DL + Template Matching), a ~20x increase in detection capability.
• Real-Time Scientific Response: Demonstrated the feasibility of using cloud-based DL workflows to provide actionable scientific data to civil protection authorities within hours of data acquisition during a crisis.
• Methodological Integration: Successfully combined DL detection with template matching to overcome the "overlap problem" in high-rate seismic swarms, significantly lowering the magnitude of completeness ($M_c$) from 2.7 to 1.3.
• Discovery of a Third Reservoir: Identified a previously unknown deep magmatic reservoir beneath Anydros Islet, refining the understanding of the Santorini-Kolumbo plumbing system.

4. Key Results

Seismicity Patterns and Evolution

• Burst-Like Swarms: The crisis was defined by spasmodic, burst-like seismicity (banded tremor) rather than a standard Omori-law decay. These bursts contained >200 $M_L > 4$ events within weeks.
• Migration: Seismicity migrated rapidly (~20 km) along a NE-SW trend from Kolumbo toward Anydros and Amorgos.
• Phases: The crisis evolved through distinct phases: Onset (Jan 25), Peak (Feb 1–12), Post-Peak, and Deceleration. The peak featured hourly rates of up to 150 events.

Source Mechanisms

• Fluid-Driven Processes: Moment tensor analysis revealed significant non-double-couple (non-DC) components (up to 60% CLVD) in early events, indicating fluid injection, crack opening, or cavity collapse.
• Stress Regime: The dominant mechanism was normal faulting consistent with the regional extensional tectonic fabric, but with a strong influence of pore pressure changes.
• LP Events: Only 61 LP events were detected, and no clear shallow volcanic tremor was observed, suggesting magma remained deep and an immediate eruption was unlikely.

Tomographic Findings

• Three Reservoirs: The study identified three distinct low-$V_p$ zones:
  1. Beneath the Santorini caldera (fluids/hydrothermal).
  2. Beneath Kolumbo volcano (melt reservoir, high $V_p/V_s$).
  3. Beneath Anydros Islet: A new discovery of a deep magmatic reservoir (>8 km depth) with variable $V_p/V_s$, interpreted as the source of the dyke intrusion.
• Dyke Intrusion: The rapid migration of seismicity and the presence of high non-DC components strongly suggest a deep basaltic dyke intrusion that pressurized the crust, triggering fault slip and fluid migration along the Anydros horst.

5. Significance

• Crisis Management: The study provided critical evidence that the unrest was volcanic-tectonic but not imminently eruptive, helping to guide civil protection decisions and mitigate unnecessary evacuations while maintaining vigilance.
• Scientific Paradigm: It challenges the assumption that purely tectonic swarms cannot produce such high rates of moderate-to-large earthquakes ($M > 4$) without an eruption, highlighting the role of high-pressure fluids in volcanic-tectonic zones.
• Future Monitoring: The paper argues for the establishment of dedicated volcanic observatories in the South Aegean equipped with real-time DL processing, offshore seismic networks, and integrated geodetic data to handle future crises effectively.
• Global Analogue: The Santorini 2025 sequence serves as a unique global analogue for fluid-driven seismicity in mature island arc systems, distinct from the basaltic rift zones typically associated with such swarms.

In conclusion, the paper demonstrates that integrating deep learning with traditional seismology transforms the ability to monitor volcanic crises, providing the resolution necessary to distinguish between tectonic and magmatic processes in near real-time.