MIPT develops AI that predicts where major earthquakes will strike

Scientists from the Moscow Institute of Physics and Technology and Innopolis University have developed an algorithm capable of identifying the most dangerous sections of tectonic faults in advance — the places where the most destructive earthquakes originate.

The new system could significantly improve early warning technologies and help assess construction risks in seismically active regions.

The algorithm is called SPAD (Seismogenic Patches Detection). Instead of analyzing powerful earthquakes, it focuses on background seismicity — the weak, almost imperceptible tremors that occur constantly.

Why it matters

Tectonic faults have a highly uneven structure. Their surfaces contain strong “patches” — zones with high frictional resistance. Elastic energy can accumulate there for years before eventually being released as a major earthquake.

Until now, such zones could usually only be identified reliably after a catastrophic quake had already occurred.

Alexey Ostapchuk, associate professor at the Department of Theoretical and Experimental Physics of Geosystems at Moscow Institute of Physics and Technology, explained that the new research challenges this understanding by showing that a large number of weak earthquakes originate in the same zones where the strongest earthquakes later begin. He said the method now makes it possible to detect these dangerous regions many years before a major event.

How the SPAD algorithm works

The algorithm operates fully automatically, without requiring expert guidance or historical data from past major earthquakes.

First, SPAD “cleans” seismic catalogs by removing aftershocks and interconnected tremors. What remains is background seismicity, which reflects the slow deformation of Earth’s crust.

The algorithm then searches for areas where these weak events cluster particularly densely and clearly defines their boundaries. These regions correspond to the seismogenic “patches” where stress capable of producing major earthquakes accumulates.

SPAD relies on two physical criteria: the maximum density of clustered events and the characteristic distance between them.

Successful testing in Kamchatka

The researchers tested the system using 35 years of seismic data from Kamchatka.

The results were impressive: the algorithm successfully identified 13 of the 16 major earthquakes that occurred before 1990 and 7 of the 8 that occurred after 1990.

This suggests that seismogenic zones are long-lived structures that can remain active along tectonic faults for decades.

Future applications

Scientists are now working on improving the algorithm’s precision and creating more detailed maps of the internal structure of seismogenic zones.

The method has already demonstrated effectiveness not only in Kamchatka, but also in the Kuril Islands, Japan, and along California’s San Andreas Fault.

In the future, SPAD could become the foundation for global mapping of dangerous tectonic fault zones and significantly improve seismic hazard assessment worldwide.

In brief

Researchers from the Moscow Institute of Physics and Technology and Innopolis University have developed the SPAD algorithm, which uses background seismic activity to identify dangerous seismogenic zones along tectonic faults before major earthquakes occur. The technology was successfully tested on 35 years of data from Kamchatka and has already shown promising results in other seismic regions. The new system could greatly improve earthquake forecasting and increase safety in high-risk areas.

Follow NEWS.am Tech on Facebook and Twitter