As a state of emergency is declared on the Greek island of Santorini, seismologists are increasingly turning to artificial intelligence technology to provide high-resolution images of the ongoing seismic activity to enhance short-term forecasting accuracy.
Since the start of the crisis, a team from BGS comprising Margarita Segou, Brian Baptie, Rajat Choudhary, Wayne Shelley, and Foteini Dervisi has been employing machine learning algorithms to detect ten times as many earthquakes as standard techniques. Since 1 December 2024, over 20,000 tremors have been accurately predicted in the Santorini area alone. This approach allows geologists to identify small-magnitude earthquakes previously undetected.
BGS Seismologist Margarita Segou is leading the development of groundbreaking research, which has revolutionised how scientists learn from seismic activity and predict patterns.
Dr Margarita Segou, BGS Seismologist, said: “This machine learning technique results in far richer data feeding into short-term forecasts, allowing experts to track the evolution of events and better advise emergency services and at-risk communities.
These algorithms allowed researchers to first note increased seismic activity across the Santorini region on 26 January 2025. Standard detection schemes did not register the same increase until 31 January. They only picked up around 2000 seismic events in the Santorini area, ten times less than the new approach has detected.
Dr Segou says the heart of advancement is the ability to combine different sources of information more quickly.
“Through strong international partnerships, we can reprocess past and present data through machine learning and gain a new and priceless insight into the seismic activity in Santorini in previous phases of unrest and its links to the volcanic system.”
Santorini is on the Hellenic volcanic arc at the convergence of the African and Eurasian plates—at a complex tectonic boundary. Currently, seismic events around the island show that seismicity bursts occur almost twice a day, with the tremors lasting one to two hours.
Dr Segou adds that the data reveals some unique features.
“We have evidence that this is fluid-driven, swarm-type seismicity that comes in pulses. This is not unheard of in other volcanic regions; however, this time, it is evolving on top of active faults that complicate the expression of seismicity.
“It is easy to get a disconnected story when we just look at moderate magnitude seismic events. It is only when we investigate the smaller magnitude events that occur between that we learn of the hidden mechanisms that take place between the large earthquakes.
“It is critical that we track whether those pulses become more frequent and how they migrate in space and depth. So far, the largest quake in this swarm has been a 5.2 magnitude.”
Pic Caption: Machine learning algorithms track the evolving seismicity crisis on the island of Santorini in Greece (Segou et al., in prep.). As expected, machine-learning earthquake detection uncovers 10 times more earthquakes (in red) than standard seismic catalogues from the National Observatory of Athens (in grey). This is the first high-resolution image of the complex volcanic/seismic system [30/11/24 to 8/2/25]. Prepared by Margarita Segou, Rajat Choudhary, Wayne Shelley, Foteini Dervisi and Brian Baptie. BGS © UKRI 2025.
