Earthquakes: New Study on the Use of Machine Learning to Understand the Evolution of Seismicity

Using machine learning algorithms to recognize specific patterns in seismic activity data and study how fault activity evolves over time: this is the objective of a study recently published in Nature Communications involving the National Institute of Oceanography and Applied Geophysics – OGS together with the University of Genoa and other international scientific institutions, including the Helmholtz Centre for Geosciences, RWTH Aachen University, University of Potsdam, Free University of Berlin, and Stanford University.

The study analyzed five major earthquakes from the past: the 2023 Kahramanmaraş earthquakes, the 2009 L'Aquila earthquake, the 2014 Iquique earthquake, the 2016 Central Italy earthquake, and the 2024 Noto earthquake. In some of these events, transient variations in seismic activity — such as earthquake swarms or increases in seismicity — had been observed in the months or weeks preceding the main shock, while in others such signals were not evident or were highly complex to identify.

The results show that the method successfully identified patterns associated with phases of transient seismicity in cases where such phenomena had already been documented in the scientific literature, while it did not reveal similar signals in earthquakes where such variations were not clearly observable.

Image credits: Angelo Giordano/Pixabay