On catching the preparatory phase of damaging earthquakes: an example from central Italy

Abstract

How, when and where large earthquakes are generated remain fundamental unsolved scientific questions.Here, we demonstrate that intercepting when a fault system starts deviating from its steady behavior by monitoring the spatio-temporal evolution and dynamic source properties of micro-to-small earthquakescan have high potential as tool for identifying the preparatory phase of large earthquakes. We re-analyzed the seismic activity that preceded the Mw 6.3 earthquake that hit L’Aquila on 6 April 2009 in central Italy. We show that the seismic catalog information can be transformed into features allowing us to track in a statistical framework the evolution of seismicity. Features associated to foreshocks show different patterns from the background seismicity that occurred in the previous years. While foreshocks share similar clustering properties of previous seismic sequences not culminating in large earthquakes, we show that their ensemble allows to clearly capture the activation phase of the main event.

Empowering Machine Learning Forecasting of Labquake Using Event‐Based Features and Clustering Characteristics

“…Such a separation was also used by Picozzi et al. (2023) who found different patterns in the features relative to background and clustered seismicity that occurred prior to the L’Aquila earthquake on 6 April 2009 in central Italy.…”

“…Traditional feature computation is based on selecting a time window and computing the features on regular time intervals Picozzi et al, 2023). Subsequently, the computed features are assumed as time series and could be introduced into, for example, an LSTM network for either TTF prediction (Karimpouli et al, 2023;Laurenti et al, 2022) or foreshock-aftershock classification (Picozzi & Iaccarino, 2021).…”

“…All AE events located inside the time-space window are used to compute a feature. We start from the feature pool developed and discussed in and used in Karimpouli et al (2023) for TTF prediction as well as additional features described in Picozzi et al (2023). As the features are now computed per-event, we only use those for which the calculations can be performed reliably and efficiently while having limited input data (see Text S2 in Supporting Information S1 and .…”

“…(2023) for TTF prediction as well as additional features described in Picozzi et al. (2023). As the features are now computed per‐event, we only use those for which the calculations can be performed reliably and efficiently while having limited input data (see Text S2 in Supporting Information and Kwiatek et al., 2024).…”

“…This is technically very similar, in this study, to the separation of all events into background and cluster populations. Such a separation was also used byPicozzi et al (2023) who found different patterns in the features relative to background and clustered seismicity that occurred prior to the L'Aquila earthquake on 6 April 2009 in central Italy.…”