A recent study by OGS and INGV integrates a machine learning algorithm into the analysis of seismic sequences to evaluate the probability of strong aftershocks.
An algorithm for the probabilistic evaluation of strong aftershocks based on data and information from the California seismic catalogs. This is the focus of the new study by Stefania Gentili of the National Institute of Oceanography and Experimental Geophysics - OGS and Rita Di Giovambattista of the National Institute of Geophysics and Volcanology (INGV), recently published in Physics of the Earth and Planetary Interiors.
Earthquakes do not occur in a homogeneous way either in time or in space: a first particularly strong seismic shock, in fact, is often followed by a series of successive aftershocks, even after weeks or months in the same area. Sometimes it may happen that after a shock of high magnitude, similar or greater magnitude aftershocks follow. Machine learning algorithms, a branch of artificial intelligence, have been applied to evaluate the probability that an event of magnitude greater than 4 is followed by a strong event.
"Machine learning algorithms work by learning and need a large amount of data to train. What we have proposed, called NESTOR, from the first hours after the first strong event provides indications on the probability of similar or greater intensity aftershocks“ tells Stefania Gentili of the OGS Seismological Research Center. “In this study, we used catalogs of earthquakes that occurred in California, a seismically very active area and therefore very well monitored and analysed. NESTORE was able to predict the occurrence of strong earthquakes well in advance in eighty percent of the cases analysed, with less than 20% of false alarms. Aftershocks of significant magnitude can have further impacts on buildings, structures and infrastructures already damaged by previous earthquakes and pose new risks for the population”, continues the researcher, specifying that “having possible probabilistic indications on their occurrence would be extremely useful".
To statistically validate the method and favor its application to a large number of events in different tectonic areas, the software will be made available to the scientific community.
The paper, entitled Forecasting strong subsequent earthquakes in California clusters by machine learning, is the result of a long research that is part of the project "Analysis of seismic sequences for the prediction of strong aftershocks", coordinated by the National Institute of Oceanography and Experimental Geophysics - OGS, and in which INGV and the Japanese research organization The Institute of Statistical Mathematics (ISM).
The project is included in the 2021-2023 Executive Protocol of bilateral scientific-technological cooperation between Italy and Japan and is included among the eleven major projects admitted by the agreement signed in January last year.
The goal is to improve the ability to estimate the probability of future strong aftershocks, starting from a few hours after the first major shock, making the algorithm increasingly robust, i.e. training it so that it provides increasingly reliable probability estimates.
Link to the original paper: https://www.sciencedirect.com/science/article/pii/S0031920122000401
The research is part of an Italy-Japan scientific-technological cooperation agreement