Machine learning-based high-resolution data set for the 2009 L'Aquila earthquake sequence

SUMMARY In recent years, machine learning (ML) techniques have emerged as a powerful tool in seismology, enabling the detection of small-magnitude seismic events that typically go unnoticed by traditional methods. Here, we apply ML-based methods to improve the characterization of normal fault systems and aftershock activity in the Central Apennines, using data from the 2009 L'Aquila seismic sequence. By processing data from both permanent and temporary seismic stations, we identified approximately 191 000 events—with a local magnitude range of −1.83 and 5.96, recorded during January-December 2009—nearly ten times more than the standard catalogue maintained by INGV. These events were relocated using a combination of absolute and relative location techniques, resulting in a high-resolution catalogue of 148 000 earthquakes. This catalogue is distinguished by an increased number of S-wave pickings, which significantly reduces localization errors and enhances the accuracy of fault geometry reconstruction. Compared to an existing semi-automatic catalogue, we observe a full recovery of seismic events, and a significant improvement of new events identified and well-located by the ML approach, with a marked increase in the quality and quantity of P- and S-wave arrivals. The refined seismic catalogue not only provides a more detailed and accurate definition of the fault architecture but also offers new insights into the distribution of aftershocks, unrolling the complex pattern of faulting that normally remains masked during standard analyses. This work highlights the potential of ML methods in advancing our understanding of complex fault systems and seismic sequences.