An Advanced Framework for Seismic Monitoring: Leveraging Transformer Models and Distributed Acoustic Sensing Technology for Earthquake Detection and Arrival Time Picking

The integration of Artificial Intelligence, particularly foundation models and modern Transformer-based architectures, opens up new frontiers for seismic monitoring. In this work, we propose a comprehensive AI-driven framework for detection and phase picking of seismic events. These models are designed to exploit the capabilities of advanced AI techniques to tackle the challenges posed by high-frequency, high-density data, and noisy environments typically associated with seismic monitoring technologies like Distributed Acoustic Sensing (DAS).

Our method combines the best of two worlds: U-Net's ability to capture high-resolution details and the power of Transformers to model global context. This combination helps the model achieve more accurate segmentation, identifying the phases' arrival times with high accuracy. 

We validate our framework on DAS data acquired from the seismically active area of the Campi Flegrei caldera (Southern Italy), leveraging the dense temporal and spatial sampling offered by DAS technology. The results show that our approach effectively learns seismic wave characteristics: the arrival time picking model demonstrates a notable 5% enhancement in the average F1-score for P and S waves, achieving 90%, surpassing the current state-of-the-art performance.

This study highlights the huge potential of integrating AI-driven methodologies with DAS technology, paving the way for advanced automatic real-time seismic monitoring systems.