Deep Learning-Based Earthquakes Localization at Campi Flegrei via Distributed Acoustic Sensing

The current phase of unrest of the Campi Flegrei caldera (Italy), one of the most dangerous volcanic complexes in the world, requires increasingly rapid and high-resolution seismic monitoring solutions. In this context, Distributed Acoustic Sensing (DAS) has recently emerged as a highly innovative technology, enabling existing fiber-optic cables to be repurposed into ultra-dense seismic arrays capable of sampling the seismic wavefield with unprecedented spatial resolution.

In this study, we present a new earthquake-localization method that uses automatically identified P- and S-wave arrivals on DAS data to localize seismic events. Employing Transformer-based architectures designed to process DAS's high-dimensional strain data, our approach simultaneously estimates key source parameters, including hypocentral location, magnitude, and origin time. A comparative analysis against the official seismic catalogue reveals minimal residuals, validating the model's robustness. 

The model therefore represents a significant advancement, as it enables reliable earthquake localization in extremely short time frames using exclusively automatically picked data, while simultaneously overcoming the computational bottlenecks typical of traditional processing workflows. As a result, this methodology establishes a new benchmark for real-time monitoring of magmatic and hydrothermal systems, substantially contributing to improved seismic hazard assessment.