Fibre optic distributed acoustic sensing of volcanic events

The study addresses how to better detect, locate, and interpret subtle volcanic processes that precede and accompany eruptions. Although volcano seismology and infrasound observations have advanced understanding of magma migration, hydrothermal circulation, and explosive activity, sparse sensor networks and incomplete knowledge of near-surface structure limit detection of small but potentially important signals and hinder robust source interpretation. The authors hypothesize that dense spatial sampling with fibre-optic distributed acoustic sensing (DAS) can discriminate tiny volcanic events embedded within tremor, locate explosion sources, and reveal hidden near-surface structures by resolving the coupled seismo-acoustic wavefield and its interaction with heterogeneous volcanic deposits.

Prior work in volcano seismology has catalogued diverse signal types (VT, LP, VLP, tremor, explosion quakes) and linked many to fluid-rock interactions, while alternative mechanisms (e.g., slow rupture) have been proposed. Infrasound has been leveraged to characterize explosion sources, energy partitioning, and conduit conditions, with multiparametric approaches improving eruption understanding. Dense seismic deployments increase sensitivity to small events and enhance tomographic imaging, but near-surface heterogeneity and limited array density still impede decoding of subtle mechanisms. Recent advances show DAS can image seismological and structural features on land and seafloor using existing fibre infrastructure, offering long-range coverage and high spatial sampling. These developments motivate applying DAS to volcano monitoring to overcome limitations of conventional arrays.

Field deployment: From 30 Aug to 16 Sep 2018, an iDAS interrogator was connected to a 1.3 km standard telecommunication multimode fibre optic cable buried ~15–25 cm in scoria deposits at ~2–2.5 km from Etna’s summit craters. The cable geometry comprised two branches (B1 and B2) and both deep and shallow sections. DAS recorded dynamic axial strain rate with 2 m channel spacing. Co-located instruments included a broadband seismometer/geophone array and an infrasound array (CARB). Position accuracy of channels was ±3 m. Data and validation: DAS strain-rate measurements were validated against strain-rate estimates from broadband seismometers and geophones; infrasound records provided reference for acoustic arrivals. Weak volcanic activity during the period included explosions, small transients (degassing), local VT events, hail/thunderstorms. Wavefield analysis: The authors applied a coherent wavefield separation and enhancement scheme based on semblance (slowness grid p ∈ [−0.008, 0.008] s/m, Δp = 0.0002 s/m) to reconstruct coherent components and iteratively subtract them, enabling separation of (i) seismic surface waves and (ii) infrasound-induced ground response. Transmission and reflection components of the infrasound-induced wavefield were separated by constraining slowness to the positive branch for transmission and subtracting to reveal back-propagating reflections. Event location: Acoustic arrival times along the cable were automatically picked on the enhanced wavefield and used in least-squares beamforming and 2D template matching against theoretical travel times (straight rays due to short source–receiver distance) to estimate back-azimuth and apparent velocity. Near-surface structure: The interaction of explosion infrasound with the scoria layer was analyzed via resonance features (16–21 Hz). The relation between resonance frequency and scoria thickness was inferred using shear-wave velocities from two approaches: (1) inter-channel apparent velocities from explosion-induced Rayleigh waves and (2) MASW from controlled source “jumps,” producing multimodal dispersion curves inverted via MCMC for 1D Vs profiles (first layer Vs ~200 m/s, thickness 3–5 m; deeper layers up to ~600 m/s at 20–25 m). Continuous detection: Three multichannel detection methods were applied to two weeks of DAS data: average STA/LTA (STA=0.7 s, LTA=10 s), absolute amplitude stacking, and local similarity (average normalized cross-correlation across channels ~60 m apart; 10 s window, 0.5 s step). Event declarations used median + 3×MAD thresholds in sliding 5-min windows. Frequency bands: 0.1–5 Hz for STA/LTA and stacking, 0.1–0.6 Hz for similarity to suppress persistent tremor. Inter-event time distributions were analyzed and rescaled by average event rates to compare with gamma distributions for tectonic and LP volcanic activity.

• Two-component wavefield from explosions: A low-frequency (LF) 1–10 Hz seismic sequence (dominated by Rayleigh waves) and a superimposed high-frequency (HF) 16–21 Hz sequence induced by acoustic waves coupling into the scoria.
• Nonlinear ground response: HF energy (16–21 Hz) was strong in DAS and seismometers but absent in infrasound sensors, indicating it does not originate from direct air–cable coupling. HF amplitudes were absent where the scoria layer was thin or absent (channels 1–200; geophones C664–C666), implicating nonlinear resonance of the scoria layer driven by infrasound.
• Threshold behavior: A large NSEC explosion (5 Sep 2018, 10:54:11 UTC) with peak pressure ~107 Pa (140 Pa peak-to-peak; energy ~2.5×10^11 J) produced maximum strain-rate amplitudes ~5×10^-5 s^-1 and clear HF resonance. A later explosion (14:04 UTC) with 23.5 Pa produced total strain-rate amplitudes ~2×10^-6 s^-1 (only ~1/5 of linear expectation, ~1.1×10^-5 s^-1) and no HF resonance. Smaller infrasound events (2.7 Pa and 4.8 Pa; 15–16 Sep 2018) should yield ~1.3×10^-6 and ~2.2×10^-6 s^-1 but showed no detectable resonance above the ~1×10^-8 s^-1 noise floor, supporting a high-strain/strain-rate threshold for nonlinear excitation.
• Spatial variability of resonance: Along branch B2, resonance frequency varies from ~16 Hz (channel ~425) to ~21 Hz (channel ~510), consistent with variable scoria thickness. Using Vs ranges (apparent 400–1100 m/s; MASW 200–600 m/s), estimated scoria thickness spans ~2.5 m (f=21 Hz, Vs=200 m/s) to ~17 m (f=16 Hz, Vs=1100 m/s). MASW inversions indicate a near-surface layer Vs ~200 m/s, 3–5 m thick; resonance likely involves layers within the upper 4–5 m (Vs <300 m/s).
• Source location and local structure: Beamforming and 2D template matching of acoustic arrivals yield a back-azimuth ~201° pointing to NSEC and an apparent air-wave velocity of 355±13 m/s along branch B2 (azimuth ~24°). A weak back-propagating reflection shows apparent velocity ~432±17 m/s, attributed to a planar subsurface structure with azimuth 145°±5°, consistent with a local magnetic contrast interpreted as a lava flow margin.
• Wavefield decoding: Coherent wavefield separation enabled clear discrimination of seismic vs infrasound-induced components and isolation of weak reflections, and improved picking for event location.
• Continuous monitoring sensitivity: DAS revealed frequent small transients (~30–35 events/hour; durations 5–15 s; amplitudes ~5×10^-7 s^-1) largely invisible in the sparse seismic array and not seen in infrasound. Two classes were identified: STP (Single Tremor Pulses, 0.1–6 Hz, high inter-channel coherence) and DG (Degassing events, 1–15 Hz, low coherence). DG events were corroborated visually at NEC; STPs likely originate at greater depth.
• Inter-event statistics: Inter-event times fit gamma distributions with parameters differing from tectonic “universal” values and resembling LP events at Etna. Example parameters (R is average rate): STA-LTA (R≈30.9 h^-1, γ≈1.677, α≈0.533); stacking (R≈28.3 h^-1, γ≈1.297, α≈0.690); similarity (R≈35.9 h^-1, γ≈1.766, α≈0.503). This supports intermittent fluid-driven processes (e.g., bubble migration) as sources of the small events.

The dense spatial sampling of DAS enables decomposition of complex volcanic wavefields into constituent components, allowing event detection and structural characterization beyond the capability of sparse conventional arrays. By separating seismic and infrasound-induced responses, the study located explosions, mapped near-surface resonance, and identified a shallow reflector aligned with independent magnetic anomalies. The observed nonlinear scoria response to infrasound implies that near-surface volcanic sediments can act as selective resonators, with excitation thresholds dependent on pressure amplitude and rate of change; this has implications for interpreting seismo-acoustic coupling in explosive eruptions and in atmospheric sources (e.g., thunderstorms). The detection of numerous weak STP and DG events, invisible to infrasound and barely visible on nearby seismometers, demonstrates DAS’s potential for continuous monitoring of subtle fluid-driven processes, offering insights into degassing dynamics and tremor generation. Leveraging existing dark-fibre infrastructure and long-range interrogation, DAS can expand network coverage in hazardous or logistically challenging volcanic settings, while big-data processing and coherent wavefield analysis facilitate robust weak-signal extraction and source characterization.

This work demonstrates that fibre-optic DAS can (1) remotely detect and locate volcanic explosions, (2) reveal hidden near-surface structures and non-linear site effects (scoria resonance and shallow reflector), and (3) identify frequent weak transients associated with degassing and deeper fluid motion within volcanic conduits. The approach deciphers complex seismo-acoustic wavefields via coherent separation and enhances structural and source analyses not feasible with sparse instruments. Future research should deploy more extensive fibre networks for improved 2D/3D location accuracy, integrate DAS with multiparametric observations (e.g., thermal/UV degassing imaging), exploit submarine telecom fibres near volcanic islands, and develop full-waveform inversion frameworks for source and path effects. These advances can improve real-time volcano monitoring, early warning, and hazard assessment.

• Spatial coverage and geometry: The 1.3 km cable with two branches provides limited azimuthal coverage; accurate 3D locations of weak events (e.g., STPs) are not possible with the current layout. Straight-ray assumptions for near-field infrasound may introduce small biases.
• Site dependence: HF resonance requires sufficient scoria thickness and strong/rapid pressure changes; results are sensitive to local near-surface conditions and may not generalize where scoria is absent or thin.
• Instrumental coupling and heterogeneity: Variations in burial depth, shallow vs deep cable sections, and fault-zone crossing affect amplitudes and apparent velocities, complicating uniform interpretation.
• Detection scope: STP and DG events lacked infrasound signatures and were barely visible on conventional sensors, limiting independent validation.
• Infrastructure and data volume: DAS deployment may be constrained by fibre availability on volcanoes; large data volumes demand substantial processing and data management resources.