Distributed dynamic strain sensing of very long period and long period events on telecom fiber-optic cables at Vulcano, Italy

Gilda Currenti, Martina Allegra, Flavio Cannavò, Philippe Jousset, Michele Prestifilippo, Rosalba Napoli, Mariangela Sciotto, Giuseppe Di Grazia, E. Privitera, Simone Palazzo, Charlotte M. KrawczykScientific 2023 · DOI

TL;DR

Researchers successfully turned an existing undersea telecommunications cable into a massive seismic sensor array to monitor volcanic unrest at Vulcano Island, Italy. By combining Distributed Acoustic Sensing (DAS) with deep learning, the team detected nearly 1,500 low-frequency events that reveal the "breathing" of the island’s hydrothermal system, demonstrating a cost-effective way to monitor volcanoes in marine environments.

What Problem Does This Paper Address?

Monitoring volcanic activity on small islands is notoriously difficult. While land-based seismometers can track tremors near the summit, they offer little coverage of the submarine flanks where critical plumbing components often reside. Traditional underwater sensors, like Ocean Bottom Seismometers, are prohibitively expensive and difficult to deploy for long-term monitoring.

Furthermore, volcanoes produce subtle, low-frequency signals—Long Period (LP) and Very Long Period (VLP) events—that are precursors to eruptions. These signals are often buried in noise. Because new technologies like DAS generate massive amounts of data (terabytes per month), researchers need automated, highly accurate ways to distinguish volcanic "heartbeats" from the background noise of waves, traffic, and footsteps.

Methods

The researchers utilized Distributed Acoustic Sensing (DAS). By connecting an interrogator to a 40 km-long fiber-optic cable (owned by the telecom company TIM) running from Vulcano to Sicily, they transformed the cable into an array of 3,968 virtual strain sensors spaced just 4 meters apart.

To process the resulting 20 TB of data, they compared two main approaches:

  1. Conventional Algorithms: A standard STA/LTA (Short Time Average/Long Time Average) trigger, which looks for sudden jumps in signal energy.
  2. Deep Learning (DL): The team treated the DAS data as 2D images (channel vs. time) and trained two Convolutional Neural Network (CNN) architectures:
  3. Inception-v3: A heavy-duty model standard in computer vision.
  4. Optimized CNN: A shallower, custom-built network designed to be more efficient for seismic data.

Finally, they used hierarchical clustering to group the detected signals into "families" based on waveform similarity, allowing them to interpret the physical mechanisms behind the noise.

Datasets

  • DAS Dataset: One month (Jan 13 – Feb 14, 2022) of continuous strain-rate data recorded at 1 kHz.
  • Ground Truth: A catalog of 1,488 events (891 LP, 597 VLP) identified by the INGV (Italy’s National Institute of Geophysics and Volcanology) using a permanent network of traditional broadband seismometers.
  • Training Set: 148,800 image snippets (1% volcanic events, 99% background noise) used to train and validate the DL models.

Key Insights and Ideas

  • The "Breathing" Hydrothermal System: The researchers propose a model where gas accumulates in the hydrothermal system (causing VLP signals as the system "inflates") and is then released through resonating fractures (causing LP signals). This "breathing" cycle is a vital indicator of volcanic unrest.
  • Quantity Over Quality: While the fiber cable wasn't perfectly coupled to the ground (some parts were in conduits or lying loose on the seafloor), the spatial density of the sensors compensated for the lower signal-to-noise ratio. Having thousands of sensors allowed the team to "see" the wave propagation in ways a single seismometer cannot.
  • CNNs as Automated Filters: The DL models outperformed traditional triggers because they automatically "learned" to act as frequency filters. The first layer of the Optimized CNN developed kernels that specifically emphasized low, mid, and high frequencies relevant to volcanic activity.

Key Results

  • Superior Detection: The Deep Learning models significantly outperformed the traditional STA/LTA algorithm.
  • Inception-v3 achieved a True Positive Rate (TPR) of 0.97 on continuous data.
  • Optimized CNN achieved a TPR of 0.94.
  • STA/LTA struggled, particularly with VLP events, managing a TPR of only 0.50.
  • Submarine Sensitivity: The DAS system successfully detected volcanic signals even in the submarine sections of the cable, proving its utility for offshore monitoring.
  • Frequency Signatures: The study identified two distinct frequency bands for the events: 0.1 to 0.2 Hz (VLP) and 3 to 5 Hz (LP).

Limitations and Future Work

The authors note that signal quality is highly dependent on cable coupling. Portions of the cable that were "hanging" in the water column or loose on the seabed produced noisier data or lacked signal entirely.

Future work involves moving toward real-time monitoring. The researchers suggest that once a DL model is trained, it can be used for quasi-real-time detection and classification, providing an early warning system for volcanic crises. They also aim to use the high-density DAS data to map "site effects"—how local geology amplifies volcanic shaking—which is currently difficult with sparse land-based stations.

Summary updated

Abstract

Volcano-seismic signals can help for volcanic hazard estimation and eruption forecasting. However, the underlying mechanism for their low frequency components is still a matter of debate. Here, we show signatures of dynamic strain records from Distributed Acoustic Sensing in the low frequencies of volcanic signals at Vulcano Island, Italy. Signs of unrest have been observed since September 2021, with CO<sub>2</sub> degassing and occurrence of long period and very long period events. We interrogated a fiber-optic telecommunication cable on-shore and off-shore linking Vulcano Island to Sicily. We explore various approaches to automatically detect seismo-volcanic events both adapting conventional algorithms and using machine learning techniques. During one month of acquisition, we found 1488 events with a great variety of waveforms composed of two main frequency bands (from 0.1 to 0.2 Hz and from 3 to 5 Hz) with various relative amplitudes. On the basis of spectral signature and family classification, we propose a model in which gas accumulates in the hydrothermal system and is released through a series of resonating fractures until the surface. Our findings demonstrate that fiber optic telecom cables in association with cutting-edge machine learning algorithms contribute to a better understanding and monitoring of volcanic hydrothermal systems.
BibTeX
@inproceedings{currenti2023distributed,
  author = {Gilda Currenti and Martina Allegra and Flavio Cannavò and Philippe Jousset and Michele Prestifilippo and Rosalba Napoli and Mariangela Sciotto and Giuseppe Di Grazia and E. Privitera and Simone Palazzo and Charlotte M. Krawczyk},
  title = {{Distributed dynamic strain sensing of very long period and long period events on telecom fiber-optic cables at Vulcano, Italy}},
  booktitle = {Scientific},
  year = {2023},
  doi = {10.1038/s41598-023-31779-2},
}
Figure