Tracing the Central Italy 2016-2017 seismic sequence fault system: insights from unsupervised Machine Learning and Principal Component Analysis

In recent years, we have witnessed the rise of Machine Learning (ML) in popularity and adoption across most scientific disciplines. The reasons behind this success are partly its versatility to adapt to different problems and types of data sets, the automatization of time-consuming repetitive tasks or its ability to learn complex relationships between observed variables. All of these make ML indispensable to the scientific discovery. In Seismology, ML has been applied to problems as different as earthquake detection and phase picking, signal classification, ground motion prediction or early warning systems development. In this work, we investigate a rich deep learning seismic catalogue from the Central Italy 2016-2017 seismic sequence (Tan et al., 2021) with the aim of identifying active faults and study their distribution and evolution over the duration of the sequence. The catalogue, built using a deep-neural-network based phase picker, includes over 900 000 earthquakes with moment magnitudes ranging from 0.5 to 6.2, of which 72 000 contain focal mechanism information (p.c. Meier, 2023). For our analysis, we combine unsupervised clustering algorithms such as DBSCAN, HDBSCAN or OPTICS with Principal Component Analysis (PCA). Our preliminary clustering results of the full, year-long, catalogue, as well as extracted month-, and week-long catalogues, with and without focal mechanisms, reveal the presence of high-density clusters of earthquakes of varying extent within a cloud of diffuse seismicity. Through PCA, we associate some of these high-density clusters to individual faults, highlighting the complexity of the fault system and showing how a multitude of faults, often small-scale, became active at different points of the seismic sequence.