Continuous DAS Acquisition for Seismic Event Detection in an Underground Environment

Distributed Acoustic Sensing (DAS) has recently emerged in seismological monitoring, converting standard telecommunication fiber optic cables into dense linear arrays of virtual seismic sensors. This capability is particularly relevant in underground environments, where existing infrastructure can be reused.

This study presents results from a continuous experiment, conducted at the Low-Noise Underground Laboratory (LSBB, Laboratoire Souterrain à Bas Bruit, https://www.lsbb.eu/ located in Rustrel, southern France). Multiple fiber optic cables with varying characteristics and installation configurations were deployed throughout the gallery network, including cables laid directly on the ground, weighted with sandbags, sealed in concrete trenches, or structurally attached to gallery walls. These fibers were interrogated by a Febus Optics DAS A1 interrogator. A dense array of seismometers is already in place at the LSBB site, inside the galleries and also at the surface. The geometry of the seismic and the DAS networks is detailed in the Figure 1.

Figure 1 : Map of the LSBB galleries with seismic stations (red stations inside the galleries, green stations at the surface). On the right, map of the fiber optic cable deployment. We deployed two different cable types Telecom and MultiSens.

A key objective of the study is to develop and evaluate automated workflows for the detection and characterization of teleseismic and regional events recorded on DAS arrays. For development we used the Xdas[i] python library. Two complementary detection strategies were applied. The first relies on the classical STA/LTA algorithm. The detection is based on statistics of STA/LTA among all channels (see Figure 2).

The second detection strategy leverages deep-learning phase pickers. The standard PhaseNet[ii] model, applied channel by channel, and the PhaseNet-DAS[iii] model, designed to exploit the multi-channels geometry, were evaluated. The Figure 2, shows a representative regional event, a magnitude 2.6 earthquake at an epicentral distance of 147 km. PhaseNet-DAS, exploiting the spatial continuity of the array, found coherent picks collocated with the predicted P and S arrival times. Both approaches confirmed the capability of the DAS system to detect regional events at the LSBB, even at moderate magnitudes. The goal is to create a catalog detection and to develop an automatic detection workflow.

Figure 2 : Strain-rate data with PhaseNet-DAS picks (top). PhaseNet-DAS detection (middle). Seismic detection using STA/LTA (bottom). The PhaseNet-DAS detection enlightening the better detection of P waves for AI algorithms.

The results demonstrate that DAS fiber networks deployed in underground galleries can reliably detect regional seismic events. Machine-learning phase pickers improve the detection of P waves and offer a better phase discrimination compared to energy-based STA/LTA detectors.

[i] Trabattoni, A., Baillet, M., Ende, M. van den, Rivet, D., Stutzman, E., Strumia, C., & Biagioli, F. (2025). Xdas: A Python Framework for Distributed Acoustic Sensing. Seismological Research Letters. https://doi.org/10.1785/0220240366

[ii] Weiqiang Zhu, Gregory C Beroza, PhaseNet: a deep-neural-network-based seismic arrival-time picking method, Geophysical Journal International, Volume 216, Issue 1, January 2019, Pages 261–273, https://doi.org/10.1093/gji/ggy423

[iii] Zhu, W., Biondi, E., Li, J. et al. Seismic arrival-time picking on distributed acoustic sensing data using semi-supervised learning. Nat Commun 14, 8192 (2023). https://doi.org/10.1038/s41467-023-43355-3