Seismic monitoring using the telecom fiber network

Probing length changes of deployed telecommunication fibers to assess ground motion has garnered interest as a scalable approach that leverages existing global optical infrastructure. While Distributed Acoustic Sensing (DAS) is powerful, it has limited range and typically cannot operate on fibers simultaneously carrying internet traffic. Interferometry of continuous-wave coherent lasers and state-of-polarization sensing have enabled earthquake detection on operational links on land and subsea. Laser interferometry measures variations of optical path length due to strain with high sensitivity and broad linearity, and land-based fibers provide better interpretability thanks to nearby conventional seismic stations. This study reports a seismic observatory based on laser interferometry operated for about two years on a wavelength-multiplexed telecom fiber in Italy. Over 1.5 years of continuous acquisition were systematically compared with co-located stations of the Italian permanent seismic network (INGV). The goals are to quantify detection capabilities across magnitude and distance, establish empirical scaling relations for interferometry-based detection, and assess spectral analysis of fiber data as a tool for investigating earthquake dynamics and supporting daily monitoring in populated areas.

Prior work has shown that DAS on dark fibers can capture earthquake wavefields and near-surface dynamics but is limited in range and often incompatible with live traffic. Polarization-based sensing on transoceanic cables has detected seismic and ocean signals. Long-range fiber-optic earthquake sensing via active phase noise cancellation and interferometry has been demonstrated on both land and seafloor cables. Compared to polarization sensing, laser interferometry offers higher sensitivity and a broader linear range. Theory for phase-transmission fiber-optic deformation sensing has been developed, and field deployments have shown feasibility. However, quantitative scaling laws linking detection probability to earthquake magnitude and distance for interferometry on operational land cables have been lacking, and comprehensive datasets for validation against standard seismometers remain limited.

Site and fiber: A 39 km telecom fiber in Italy, mostly in conduits along a road (≈100 m aerial), crosses several bridges and terminates in medium-sized towns. Average azimuth ≈125°. The link is part of a regional DWDM ring carrying 100 Gb/s QPSK traffic. Interferometric schemes: Two configurations were used. - Self-heterodyne (single-end) scheme: An ultrastable laser is split; one arm serves as a reference, the other passes through a 40 MHz frequency shifter and traverses the link out-and-back (adjacent fiber for return), then recombines to produce a beatnote. Instantaneous beat frequency deviations Δν(t)=ϕ̇(t)/2π reflect integrated fiber strain-induced optical path changes. Sampling at 1 kHz, with demodulation and storage; data volume ≈26 MB/h at 1 kHz, 14 KB/h at 10 Hz. Data filtered and downsampled to 100 Hz for analysis. Heterodyne detection is used to avoid low-frequency electronic noise. - Dual-end (paired lasers) scheme: A laser at each end is launched toward the opposite node and interfered with the local laser, yielding beatnotes near 1.1 GHz at both ends. By synchronously combining the demodulated signals with appropriate signs, common-mode laser noise is rejected while fiber-induced frequency deviations remain. Synchronization between nodes is required to ≤5 μs, achieved by applying a frequency step to the shifter to trigger acquisitions and aligning to UTC via NTP (≈10 ms), then within the demodulation window. Assumption: Perturbations are symmetric bidirectionally since fibers run parallel in the same cable. Instrumentation is rack-mounted at telecom nodes. DWDM add/drop multiplexers combine the sensing carrier with traffic; an optical preamplifier compensates ≈10 dB loss. Seismic stations: Two nearby INGV stations were used for comparison. IV.TERO: accelerometer(s) and velocimeter connected to a 6-channel digitizer, installed in a pit ~2 m below ground; site VS30 ≈981 m/s. IV.ASCOL: velocimeter and accelerometer at the telecom PoP hosting the interrogator, in a high-noise urban site on fluvial deposits; both stations stream near real-time data to INGV. Event selection for sensitivity and validation: Built an earthquake catalog spanning 1.5 years, including >600 events with magnitudes ~1–8 and epicentral distances up to ~2000 km. Primary period: June 19, 2021 to Sept 26, 2022 with criteria: M ≥ 2 within 30 km; M ≥ 2 within 300 km; M ≥ 2.5 within 300 km (as stated). To boost statistics at larger distances/magnitudes, included Turkey sequence (Mw ≥ 5, Feb 6–Mar 23, 2023) and the Ancona offshore sequence (~100 km away, Nov 9–18, 2022; M ≥ 2 within 180 km). For each event, theoretical P/S arrivals at the fiber midpoint and stations were computed using TauP and a 1D Earth velocity model; time windows were extracted from continuous recordings. Each event’s visibility on the fiber was scored (0 not detected; 1 clearly detected; intermediate unsure), considering time traces and alignment with expected arrival times. Signal comparison and analysis: Fiber measures Δν(t), the time derivative of accumulated optical phase along the link, effectively integrating distributed deformation and coupling geometry; seismometers record point ground velocity/acceleration along axes. Despite different observables, similarities in temporal evolution and spectra between Δν(t) and ground velocity v(t) enable comparative analysis. Time-domain comparisons included P/S picking relative to IV.ASCOL/IV.TERO; spectral analysis focused on events within 100 km to minimize propagation effects. Corner frequencies were estimated from Fourier amplitude spectra and analyzed versus magnitude. Data were band-pass filtered (fiber: 1.2–20 Hz; seismometers: 1.2–40 Hz) with bidirectional Butterworth filters; seismometer responses were deconvolved.

- Detection catalog: >600 earthquakes (M ~1–8; distances up to ~2000 km) analyzed over 1.5 years. The distribution of detected/unsure/non-detected events versus magnitude M and closest distance d to the fiber follows an empirical scaling consistent with P ∝ 10^{-M}/d. - Empirical sensitivity law: The fiber’s sensitivity threshold was determined iteratively by minimizing detections below a threshold line in M–d space. Parameters yielding best agreement: A1 ≈ 0.4–0.8 and A2 ≈ (3280 ± 10) (as reported for the threshold estimation). Despite higher integrated noise, the effective A1 for the fiber is of the same order as co-located seismometer IV.ASCOL and comparable to literature values for peak ground motion scaling. - Weak event detection: Weakest observed event had M = 1.4, depth 17.5 km, epicenter 2.4 km from the fiber; recorded at night when noise was lower. - Noise environment: Background noise on the fiber varies by ~10 dB between day and night due to anthropogenic activity; additional spectral noise near ~1 Hz from two cable segments restricts detection in that band. Nevertheless, weak-event detection is achievable in this unfavorable environment. - Directional dependence: For local, moderate events, detection probability was higher for sources located transversely to the average cable azimuth (~125°), suggesting cable orientation and wavefront curvature/incidence angle influence response. - Time-domain case studies: For the Feb 6, 2023 Turkey event (Mw 7.9; ~200 km), fiber recordings had high SNR with sharp P and S picks, comparable to seismometers; timing offsets between fiber Δν(t) and IV.ASCOL vertical/N-S velocities were ~10 ms. For a local M2.2 event (epicentral distance ~5 km), P/S were clearly identified on the fiber; geometry may enhance phase picking. For an Accumoli event (M1.6; fiber 32 km; IV.TERO 27 km), the fiber P-wave lagged IV.TERO by ~0.9 s, consistent with a P-wave speed ~6 km/s and ~15 km station–epicenter distance difference. - Spectral analysis: For events within 100 km, fiber spectra reproduced seismometer spectra, indicating broadband fiber response. Corner frequency f_c decreased with magnitude for both sensors. Pearson correlations between magnitude and corner frequency: r ≈ −0.60 (fiber) and r ≈ −0.59 (IV.ASCOL). Corner frequencies derived from fiber vs seismometer correlated well (r ≈ 0.81) but were systematically higher for the fiber; linear fit slope 0.9 ± 0.1 Hz/Hz, intercept 0.2 ± 0.7 Hz. - Operational data volume: Interferometric setup produced ~2.1 GB/day at 1 kHz, far less than typical commercial DAS deployments (>100 GB/day at ~250 Hz), suggesting practicality for continuous monitoring. - Practical utility: Arrival-time picking of P/S on fiber enables localization when combining multiple cables and/or seismometers; the empirical detection scaling can inform coverage maps along existing telecom infrastructure, especially for weak, nearby events relevant for urban monitoring.

The study establishes that laser interferometry on operational telecom fibers can reliably detect a broad range of earthquakes, including small local events (around M ≈ 2) at local distances, even under challenging urban-noise conditions. The empirical detection scaling with magnitude and distance provides a quantitative framework to assess coverage and sensitivity along fiber routes and to complement sparse traditional networks, particularly in populated areas where fiber is pervasive but seismometer density is low. Directional dependence indicates that cable orientation and wavefront incidence influence detection, motivating multi-segment or multi-cable deployments to average geometry effects. Time-domain comparisons demonstrate that P and S arrivals can be picked on fiber data with quality comparable to nearby seismometers, supporting integration of fiber-derived picks into standard localization workflows after accounting for the distributed nature of the sensor. Spectral analyses show consistent corner-frequency–magnitude trends between fiber and seismometer data, validating the use of fiber recordings for source parameter studies; the observed systematic bias in corner frequency suggests differences in coupling or integration effects that can be calibrated with larger datasets. The significantly lower data throughput compared with DAS enhances the feasibility of continuous, long-term operation in existing networks. Beyond earthquakes, the sensitivity to anthropogenic and environmental noise indicates opportunities for monitoring traffic and infrastructure dynamics.

This work realizes and operates a laser-interferometry-based seismic observatory on a live, DWDM telecom fiber and validates its performance against co-located seismometers over 1.5 years. It demonstrates successful detection from very weak local events to large regional earthquakes, derives an empirical detection probability scaling with magnitude and distance, and shows that spectral content and corner frequency trends match those from conventional sensors. These results position telecom-fiber interferometry as a practical tool for daily seismic monitoring, particularly in urban and suburban regions underserved by dense seismic arrays, and as a contributor to global monitoring. Future work should expand event catalogs for improved statistics, refine physical models linking distributed fiber strain to ground motion, quantify and mitigate orientation and coupling effects, integrate multiple cables for array processing and localization, and explore hardware advances (compact, low-noise lasers and coherent transceivers) to scale deployments and costs. Combining interferometry with DAS and polarization-based sensing could enable comprehensive, multi-phenomena environmental monitoring.

- Single-fiber geometry and orientation limit azimuthal coverage and introduce directional sensitivity; transverse incidence was favored. - Strong day–night anthropogenic noise variation (~10 dB) and specific spectral noise near ~1 Hz from cable segments reduce sensitivity, especially to weak signals and in certain bands. - Unknown and spatially varying fiber-to-ground coupling and housing conditions complicate absolute calibration; the fiber measures integrated distributed deformation, not a point quantity. - Spectral analysis did not correct for propagation effects and applied no SNR-based quality thresholds, contributing to dispersion in corner frequency estimates and potential bias between fiber- and seismometer-derived values. - Some inconsistencies/typos in event descriptors and station naming reflect operational complexity; precise analytical modeling of Δν(t) to ground motion is beyond the present scope. - Results are from a single 39 km land link; generalization to different terrains, cable constructions, and urban settings requires broader testing. - Dual-end scheme requires microsecond-level synchronization; network timing via NTP provides millisecond-level UTC reference and additional in-band triggering was needed.