Predicting earthquakes has always been a major challenge for scientists, as there is currently no reliable method to determine when and where a large earthquake will strike. However, a new study published in Nature Communications opens up a new avenue after artificial intelligence (AI) was able to detect hidden patterns within thousands of small tremors that may indicate some faults are preparing for a more dangerous phase before the main earthquake occurs.

Researchers from the German Research Center for Geosciences (GFZ) in Germany, in collaboration with international scientists, have developed a technique based on unsupervised machine learning. This method allows AI to analyze large amounts of seismic data and search for previously unknown patterns, rather than relying on indicators predefined by scientists.

The team analyzed what are known as "earthquake families." Instead of treating each earthquake as a separate event, they studied the relationships between small earthquakes in terms of location, time, and magnitude. These events can influence one another and reveal how stress changes within the Earth's crust.

The researchers tested this methodology on several major earthquakes that occurred in recent years, including the 7.8-magnitude earthquake in Kahramanmaraş, Turkey in 2023, the 8.1-magnitude earthquake in Iquique, Chile in 2014, and the 6.3-magnitude earthquake in L'Aquila, Italy in 2009.

The results showed that these earthquakes were preceded by subtle changes in seismic activity behavior, characterized by an increase in the clustering of small tremors in specific areas and a higher level of interaction between them. Additionally, there are indicators suggesting changes in how stress accumulates within faults before the major collapse.

Scientists believe that these patterns may represent a transitional phase where the region approaches instability, as the fault begins to behave differently from its usual activity. However, these signals are not a fixed rule, as each earthquake has its own unique characteristics that determine how it develops.

Conversely, the study revealed an important aspect of the limitations of this technology. The same patterns were not observed before some major earthquakes, such as the 2016 Amatrice earthquake in Italy and the 2024 Noto earthquake in Japan, confirming that not all earthquakes leave clear signs that can be detected beforehand.

To determine the feasibility of using this technology for future seismic monitoring, researchers did not just analyze past earthquakes but also tested artificial intelligence’s ability to monitor seismic activity as it develops. The system compared new tremors with typical patterns in each area, seeking the moment when the fault begins to behave in an unfamiliar way. Scientists believe that the emergence of a new and unexpected seismic pattern may indicate that the region is transitioning into a more disordered state, but it does not represent a confirmed prediction of an earthquake.

Researchers emphasize that the goal is not to achieve a “seismic clock” that tells humans when disaster will strike, but rather to develop smarter tools that help scientists understand fault behavior and monitor changes that may precede earthquakes when such signals exist and can be detected.

This study combines seismology with artificial intelligence in a new way, giving scientists greater ability to see patterns that were previously hidden amidst the noise of thousands of small tremors.