Santorini’s Shifting Ground: AI Reveals Volcanic Secrets

by

Author: Denis Avetisyan

A new study details how advanced seismic monitoring, powered by deep learning, provided crucial real-time insights during a period of intense volcanic unrest at Santorini, Greece.

Deep learning-enhanced tomography and earthquake catalogue analysis characterized a potential dyke intrusion and improved understanding of the 2025 Santorini crisis.

Rapidly characterizing volcanic unrest demands timely and comprehensive seismicity data, yet traditional methods often struggle with the volume and complexity of signals. This is addressed in ‘Enhanced Seismicity Monitoring in the Rapid Scientific Response to the 2025 Santorini Crisis’, which details a deep learning workflow applied to seismic waveforms during a period of intense activity between Santorini and Amorgos. The analysis expanded the initial earthquake catalogue from approximately 4,000 to 80,000 events, revealing burst-like seismicity swarms and evidence of deep magmatic reservoirs-patterns indicative of fluid-driven processes. Could this approach redefine real-time volcanic hazard assessment and improve our understanding of dyke intrusion dynamics globally?

Beneath the Aegean: Unraveling the 2025 Santorini Seismic Enigma

In 2025, a dramatic escalation of seismic activity gripped the region between the iconic caldera of Santorini and the neighboring island of Amorgos, prompting widespread concern over potential volcanic unrest. This surge wasn’t characterized by a single, major event, but rather a sustained increase in the frequency and intensity of earthquakes, deviating from the historically stable background seismicity of the area. The concentration of activity, coupled with the region’s complex geological history – a consequence of the Hellenic Volcanic Arc – immediately raised the possibility of renewed volcanic hazards, including potential eruptions or large-scale landslides. Authorities initiated heightened monitoring and preparedness protocols, recognizing the densely populated islands and critical infrastructure within the potential impact zone, and the need to swiftly assess the evolving threat posed by this escalating geological phenomenon.

Conventional earthquake monitoring networks, designed for typical tectonic events, proved inadequate in characterizing the 2025 Aegean unrest. The region’s complex geology, coupled with the unusual distribution of seismic events – bursts concentrated in both space and time – overwhelmed the capacity of standard algorithms to accurately locate and assess the magnitude of each tremor. This inability to resolve the spatiotemporal patterns of activity created a significant challenge for risk assessment; standard hazard maps failed to capture the localized intensity of the shaking, and estimations of potential volcanic impact remained imprecise. Consequently, emergency preparedness efforts were hampered by uncertainty, and the potential for a rapidly escalating crisis remained difficult to quantify with existing tools.

The seismic unrest between Santorini and Amorgos in 2025 presented a particularly challenging analytical puzzle due to the atypical nature of the observed earthquakes. Unlike typical tectonic events, the region experienced frequent, short-lived bursts of seismic energy, punctuated by the occurrence of long-period earthquakes – events characterized by unusually long wavelengths and durations. These long-period signals, often associated with fluid movement within the volcanic system, complicated interpretations based on conventional seismological techniques. Existing models, designed to interpret more standard fault-driven seismicity, proved inadequate in resolving the complex interplay between these distinct event types, necessitating the development of advanced analytical tools capable of disentangling the various contributing mechanisms and providing a more nuanced understanding of the underlying volcanic processes. This demand spurred innovation in seismic data processing and modeling, ultimately leading to a more refined approach to hazard assessment in this volatile region.

Augmenting Detection: A Deep Learning Approach to Seismic Signals

A deep learning-based earthquake detection pipeline was implemented, incorporating algorithms such as PhaseNet and Template Matching to substantially improve catalogue completeness. This resulted in a nearly 20-fold enhancement in the number of detected earthquakes compared to standard techniques. The pipeline leverages the strengths of both PhaseNet, which excels at identifying seismic phases, and Template Matching, which identifies events similar to known waveforms. Combined, these algorithms facilitate the detection of lower magnitude events and contribute to a more comprehensive seismic catalogue.

The implemented deep learning pipeline detected a total of 79,245 earthquake events. This represents a substantial increase in catalogue completeness when contrasted with conventional earthquake detection techniques. Specifically, template matching, as integrated within the pipeline, identified approximately twice the number of events compared to other established methods. This improved detection rate contributes to a more comprehensive seismic record and allows for the characterization of previously undetected seismic activity.

The development and validation of the deep learning earthquake detection pipeline relied heavily on data access from the European Integrated Data Archive (EIDA) and collaborative contributions from the National Observatory of Athens. This data facilitated the training of algorithms, specifically enabling a reduction in the magnitude of completeness – the smallest earthquake reliably detected – to ML 1.3. Prior to this enhancement, standard techniques yielded a higher magnitude of completeness, indicating a significant improvement in the catalogue’s ability to identify smaller seismic events. The availability of a comprehensive and well-curated dataset was therefore crucial for achieving this increased sensitivity and improving the overall completeness of the earthquake catalogue.

Illuminating the Magmatic System: A Subsurface Portrait through Seismic Tomography

Seismic tomography, and specifically the TomoDD (Tomographic Diffraction Diffraction) method, was utilized to generate a three-dimensional velocity structure of the subsurface beneath Santorini. TomoDD improves upon traditional seismic tomography by incorporating diffracted seismic waves, allowing for higher resolution imaging, particularly of complex geological structures. This technique involves analyzing the travel times of seismic waves generated by local earthquakes and back-projecting them to determine variations in seismic wave velocity within the Earth. Slower velocities generally indicate the presence of magma, hydrothermal fluids, or partially molten rock, while faster velocities represent denser, cooler material. The resulting 3D model provides detailed insight into the geometry and composition of the Earth’s interior beneath Santorini, revealing previously unresolved features of the magmatic system.

Seismic tomography imaging beneath Santorini revealed a substantial magmatic reservoir located approximately 4-6 kilometers below the Anydros Islet. This reservoir, estimated to contain between 1 and 7 cubic kilometers of magma, is characterized by low seismic velocities consistent with the presence of molten rock. Furthermore, the imaging data indicates the pervasive influence of hydrothermal fluids within the volcanic system; these fluids are concentrated around the magma reservoir and along fault structures, and are strongly correlated with the observed distribution of microseismicity, suggesting a link between fluid migration and the triggering of small earthquakes.

Analysis of the seismic tomography data revealed a complex subsurface structure directly correlated with the Christiana-Santorini-Kolumbo Rift Zone. This rift zone acts as a primary control on both stress distribution and magma emplacement; the model indicates magma is not distributed uniformly, but is concentrated and channeled along the rift’s structural weaknesses. Specifically, the zone’s geometry dictates areas of high stress concentration, influencing where magma accumulates and propagates. Variations in seismic velocity within the model directly correspond to the rift’s fault network, demonstrating that the rift’s architecture governs the pattern of magma distribution and the location of associated hydrothermal activity beneath Santorini.

Beyond Simple Fractures: Decoding the Complexities of Kolumbo’s Faulting

Seismic activity around Kolumbo Volcano isn’t characterized by simple fractures, but rather by complex faulting mechanisms revealed through the analysis of moment tensor solutions. These solutions demonstrate the presence of Non-Double Couple Moment Tensors, indicating forces beyond typical shear stress – specifically, the influence of forces like those exerted by fluid pressure. This suggests that magma and hydrothermal fluids are actively involved in driving the observed seismicity, potentially weakening the volcanic structure and contributing to the complex patterns of ground deformation. The identification of these non-double couple components is critical, as it moves beyond simplistic models of fault rupture and highlights the intricate interplay between tectonic stresses and fluid dynamics beneath the volcano.

Recent observations at Kolumbo Volcano reveal a confluence of factors suggesting escalating volcanic risk. An increase in volcanic tremor – sustained vibrations caused by magma movement – has coincided with discrete seismic bursts, indicating periods of intensified stress and potential fracturing within the volcanic system. Critically, these signals originate in proximity to a well-defined magmatic reservoir beneath the volcano. This combination – heightened tremor, episodic bursts, and a confirmed magma source – strongly implies that Kolumbo is undergoing increased internal pressure and demonstrates a heightened potential for future eruptive activity. The convergence of these indicators warrants continued, vigilant monitoring to assess the evolving hazard and inform appropriate mitigation strategies.

The intricate interplay between complex faulting and magmatic processes at Kolumbo Volcano necessitates a sustained commitment to comprehensive monitoring and sophisticated analytical approaches. Detailed analysis of seismic data, coupled with observations of volcanic tremor and reservoir behavior, highlights the limitations of relying solely on traditional hazard assessments. Continuous, real-time data streams – incorporating seismology, geodetics, and gas monitoring – allow for the early detection of subtle changes that may precede future eruptions or significant seismic events. Furthermore, the application of advanced techniques, such as moment tensor inversion and detailed waveform modeling, provides critical insights into the underlying physical mechanisms driving these hazards, ultimately enabling more accurate risk evaluation and effective mitigation strategies for the surrounding population.

The study’s success hinges on a harmonious interplay of advanced techniques – deep learning algorithms processing complex seismic data to illuminate previously obscured patterns of volcanic unrest. This echoes Erwin Schrödinger’s observation: “The task is not simply to replace what we already know with something new, but to integrate the new knowledge with the old.” The researchers didn’t merely apply deep learning; they integrated it with established tomographic methods and earthquake catalogue analysis, achieving a more complete understanding of the potential dyke intrusion at Santorini. The resulting interface, so to speak, sings with clarity, revealing the subtle nuances of the earth’s behavior, even when those signals are faint and complex. Every detail, meticulously examined, contributes to a richer, more resonant understanding of the unfolding crisis.

The Horizon Beckons

The rapid response system detailed within reveals not a destination, but a sharpening of the gaze. The success at Santorini hinged on a pre-existing, meticulously curated earthquake catalogue – a foundation rarely afforded when confronting genuine novelty. Future iterations must wrestle with the problem of ‘first light’ – how to distill signal from chaos when historical context is scant, or misleading. Deep learning, for all its power, remains reliant on the ghosts of past events; true predictive capability demands an ability to discern the truly unprecedented.

Tomography, too, is a dance between resolution and robustness. The current approach offered glimpses into the subsurface, but the intrusion’s precise geometry remains somewhat obscured. The pursuit of higher resolution must be tempered by an understanding that increased complexity doesn’t necessarily equate to increased understanding. Simplicity, elegantly achieved, will always outweigh baroque detail. The goal isn’t merely to map the magma, but to infer its intentions.

The work presented here isn’t an ending, but rather a rigorous editing of the tools at hand. It is a refinement, not a rebuilding. The true test will come not with another crisis like Santorini, but with the one that defies comparison. Beauty scales – clutter doesn’t. And the most valuable innovation will be the one that quietly disappears into the background, leaving only clarity in its wake.

Original article: https://arxiv.org/pdf/2603.11108.pdf

Contact the author: https://www.linkedin.com/in/avetisyan/

See also:

2026-03-13 11:56