Unveiling the source of seismic swarms with Coulomb stress imaging: application to the 2025 Santorini-Amorgos, Greece seismic crisis

For civil planning and hazard communication purposes, a central challenge during active seismic swarms is identifying the underlying causative source. This task is challenging because geodetic constraints on deformation at depth, especially in marine settings, are limited or poorly resolved. Therefore, it is essential to exploit the high spatial and temporal resolution provided by modern dense seismicity catalogues.

In early 2025, intense swarm seismicity between Santorini and Amorgos in the southern Aegean Sea triggered evacuations and heightened concern over volcanic and seismic hazards. The unrest occurred near the Santorini and Kolumbo volcanoes, and close to the rupture zone of the 1956 Mw 7.7 Amorgos earthquake, making it critical to determine whether the activity was driven by magmatic intrusion or tectonic fault slip.

We analysed ~25,000 earthquakes recorded over eight weeks using high-precision, machine learning–based relocation of seismic data. The resulting catalogue provides a detailed image of the space–time evolution of the swarm, including short-lived episodic tremor bursts. Relocated earthquakes are treated as virtual probes of stress change at depth to image candidate source processes under the assumption of Coulomb failure stress.

The seismicity defines a complex, migrating swarm at ~10 km depth. Dense swarm seismicity initiated northeast of Santorini and rapidly propagated ~20 km further northeast through mid-February, forming a widening, fan-shaped cloud. Coulomb stress imaging indicates horizontal magmatic dike propagation, rather than tectonic fault slip, as the dominant source of unrest. The intrusion is consistent with pump-like magma injections into newly opened dikes at ~12 km depth, producing multiscale, rebounding episodes of dike opening and triggered seismicity.

These results reveal a dynamic, feedback-driven mechanism for dike emplacement and demonstrate the potential of machine learning–enhanced stress imaging for tracking intrusions and improving eruption forecasting.