Seismic monitoring using the telecom fiber network

Laser interferometry enables to remotely measure microscopical length changes of deployed telecommunication cables originating from earthquakes. Long range and compatibility with data traffic make it unique to the exploration of remote regions, as well as highly-populated areas where optical networks are pervasive, and its large-scale implementation is attractive for both Earth scientists and telecom operators. However, validation and modeling of its response and sensitivity are still at an early stage and suffer from lack of statistically-significant event catalogs and limited availability of co-located seismometers. We implemented laser interferometry on a land-based telecommunication cable and analyzed 1.5 years of continuous acquisition, with successful detections of events in a broad range of magnitudes, including very weak ones. By comparing fiber and seismometer recordings we determined relations between a cable's detection probability and the magnitude and distance of events, and showed that spectral analysis of recorded data allows considerations on the earthquake dynamics. Our results reveal that quantitative analysis is possible for this sensing technique and support the interpretation of data from the growing amount of interferometric deployments. We anticipate the high integration and scalability of laser interferometry into existing telecommunication grids to be useful for the daily seismicity monitoring, in perspective exploitable for civilian protection use.

probability and the magnitude and distance of events, and showed that spectral analysis of recorded data allows considerations on the earthquake dynamics.Our results reveal that quantitative analysis is possible for this sensing technique and support the interpretation of data from the growing amount of interferometric deployments.We anticipate the high integration and scalability of laser interferometry into existing telecommunication grids to be useful for the daily seismicity monitoring, in perspective exploitable for civilian protection use.
Probing length changes of deployed telecommunication fibers as a way to access ground motion has attracted growing interest from Earth scientists [3][4][5][6][7][8][9][10] and network operators [13][14][15], that foresee the possibility to implement distributed, spatial-aliasing-free sensor grids exploiting the infrastructure already in place for global telecommunications in a scalable and sustainable way.While traditional techniques [16] have a short range and are not compatible with standard data transmission, novel approaches based on coherent laser interferometry [1,2] and stateof-polarization sensing [13] enabled earthquake detection on telecommunication cables laid in the ocean floors [2,13,17] and in land [12,18,19].Laser interferometry measures the variation of the optical path length experienced by a coherent laser radiation as it travels a fiber subjected to strain [1,2,12,18], with higher sensitivity and broader linear range compared to polarization-based sensing [17].Its implementation on land-based fibers allows more insightful interpretation of the fiber response than with sub-sea cables, thanks to a larger number of traditional sensors in cables surroundings [11,12,18].However, laser interferometry on land cables is also attractive per se, for monitoring densely-populated areas: here, the fiber pervasiveness compensates for the poor coverage by traditional seismic networks which privilege sensors installation in quieter regions, and enables applications such as people flows and vehicle traffic monitoring.
We describe the realization of a seismic observatory based on laser interferometry, opera-tional since two years on a telecommunication link in Italy, and present first results of continuous acquisition over a long period, in which we systematically compared fiber data to those from nearby stations of the Italian permanent seismic network (IV network) managed by Istituto Nazionale di Geofisica e Vulcanologia (see Methods).We found recurrent behaviour and underlying scaling rules, which enabled us to make quantitative considerations on the cable response to different classes of events and application of laser interferometry both as a research and a survey tool in different contexts.
The fiber is 36 km long and is part of a regional ring architecture at 100 Gb/s with quadrature phase-shift keying (QPSK) modulation.It is mostly hosted inside a road infrastructure, except for ∼100 m in aerial cable, crosses few bridges and terminates in medium-size towns, thus suffering from diverse and time-varying noise processes.Its average azimuthal direction is 155 • (see Extended Data Fig. 1). Figure 1A shows the cable layout and the location of nearest sensors of the IV network.Specifically, IV.ASCOL was installed at one cable end for this experiment and was used, together with IV.TERO, for comparison throughout the analysis (see Methods).The sensing light signal is produced by a compact narrow-linewidth laser, wavelength-multiplexed to the internet traffic and launched into the network.We measure the interference between the signal travelling the round-trip and a local reference beam (Fig. 1B) or between two remote laser beams launched in opposite direction (Fig. 1C), in a conceptually similar setup modified in technical details (see Methods) [20].

Results
On a standard seismometer, the probability P to detect an event depends on its magnitude M and hypocenter distance to the fiber d.Assuming P is related to the recorded peak value of a  [21] of the deployed observatory and its geographical location.Triangles indicate the location of nearby seismometric stations, the closest of which, IV.ASCOL and IV.TERO (orange), were used for comparison throughout the analysis.IV.ASCOL was installed at the fiber end in Ascoli, IV.TERO is about 9 km west to the fiber.Stars indicate epicenters of events shown in this study.B) Sketch of the optical scheme.In Ascoli, the laser radiation is split into two parts: one serves as a reference beam, the other passes through a frequency shifter, is spectrally-combined with telecom data (TX/RX) using optical Add/Drop multiplexers, amplified and launched towards the remote end in Teramo.Here, it is reflected back along an adjacent fiber and, once back in Ascoli, is recombined to the reference beam to extract an interference signal.C) Modified optical scheme, in which two separate lasers, hosted at opposite cable ends, are launched in opposite directions and interfered with the local laser.Information about the cable deformations is retrieved by combining synchronous recordings gathered at the two cable ends.The frequency shifter is here used to add a periodical pattern to the carrier frequency for synchronizing remote acquisitions.physical quantity (e.g.ground acceleration or velocity), the following relation holds [22]: where A 1 depends on the type of physical measure and A 2 is a generic scaling coefficient.For interferometry-based earthquake detection, no scaling law was reported so far.We performed this analysis, generating an earthquake catalog that included over 600 events with magnitude between 1.5 and 8 and epicentre distances up to 2000 km, spanning a time frame of 1.5 years.
We included all earthquakes located by INGV monitoring system in the considered period and looked for signature of the events on fiber recordings (see Methods).The distribution of detected, unsure and non-detected events as a function of their magnitude and distance to the closest point on the fiber (Fig. 2) well agrees with Eq. 1.
The sensitivity threshold line (blue curve in Fig. 2) is found iteratively by minimizing the cumulative number of detected events below the line and non-detected events above the line and corresponds to A 1 = 0.403 ± 0.001 and A 2 = (3280 ± 10)km (see Methods).Although the fiber suffers from a higher level of integrated noise as compared to a point-like sensor, A 1 is of the same order of magnitude as for IV.ASCOL (see Extended Data Fig. 2) and comparable to the values published in the literature for scaling peak ground accelerations or velocities.
With the better statistics enabled by ongoing data collection, this estimation could be refined accounting for effects such as the cable length or the coupling between M and d, among others.
The fiber shows the capability to detect events with local magnitude M l as small as 1.6 in the local area and a good detection predictability on events with M l > 2.5, in spite of its unfavorable environment.For instance, the background noise recorded by the fiber changes by 10 dB between day and night and shows signature of randomly distributed, non-stationary events which mostly affect detection capability for weakest earthquakes on daily hours.In addition, the two aerial segments of the cable introduce noise in the spectral region around 1 Hz, which is relevant Figure 2: Sensitivity of the fiber as an earthquake detector.A) Map of events considered in the 1.5-year-long catalogue as a function of their magnitude and distance to the fiber.Color indicates if the event was detected (green), unsure (light-green) or non-detected (empty) by the fiber.The line shows the estimated sensitivity threshold for our sensor.Labels (1), ( 2) and (3) indicate specific events which are described in this study.The inset zooms on events in a range of 20 km to 100 km and magnitude 2 to 3.5.B) Same events as the inset, mapped as a function of their angular distribution and distance to the fiber.Circles dimensions indicate magnitude class; the dashed line indicates the average direction of the cable.to seismic detection (see Extended Data Fig. 3 and 4).While most of these impairments cannot be avoided on telecom-integrated sensing systems, we expect their impact to be less detrimental in structured architectures based on separated cables simultaneously probed by the same laser interrogator.
We also analysed the detection probability of average-size earthquakes on local areas as a function of their angular distribution (Fig. 2B) and observe a larger fraction of successful detection on the transverse direction.This suggests that wavefront curvature and angle of incidence with respect to local portions of the cable [11] may have a role in building the cable response and highlights the importance of increasing the catalog size to allow for more conclusive claims.
For a direct comparison of fiber and conventional seismometer measurements, an appropriate conversion metrics between quantities had to be considered, as recordings substantially differ from two points of view: first, fiber sensors measure an optical phase, while seismometric sensors measure ground velocities or accelerations; second, fiber sensor measurements are related to the integral of a distributed deformation, while seismometric sensors measure local quantities.Operationally, we record the time derivative of the phase φ(t) accumulated in a round-trip, i.e. the instantaneous frequency deviations of the returning optical signal with respect to the injected one ∆ν(t) = φ(t)/2π.As other authors [19], we found similarities in the time evolution and spectral composition of this quantity with the ground velocity v(t) recorded by nearby seismometers, and understand them to be consistent with expectations from a functional point of view (see Methods).As an example, Figure 3 compares ∆ν(t) (orange, uppermost panel) with v(t) measured by seismometers along the vertical, North-South and East-West axes (blue, panels 2 to 4) for the three events marked in Fig. 2. For each event, zoom on the P and S-wave arrivals is shown in the right panel.
Figure 3A shows traces of the event occurred in Turkey on Feb. 6th, 2023 (moment magnitude [23] M wp = 7.9, distance to the fiber 2000 km, labelled 1 in Fig. 2).The high signal/noise ratio on the fiber recording allows a sharp picking of the wave arrival time, with similar quality as obtained from seismometer data, and a clear identification of the S-wave arrival.We note that the scaling between the amplitude of the perturbation on the fiber and the vertical and North-South velocity components of IV.ASCOL is about 10 6 Hz/(m/s), which is also in qualitative agreement with the value expected from a simplified configuration in which a straight fiber is solicited at one end [4,11,19,24] (see Methods).This scaling is confirmed for another farfield event, where correlation is also found between the time evolution of fiber and IV.TERO recordings (see Extended Data Fig. 5).
Figure 3B shows traces of an event occurred in Civitella del Tronto on Feb. 19th, 2022 (M l =2, epicenter depth 18 km and hypocenter distance 5 km from the closest point of the fiber, labelled 2 in Fig. 2).Because of the proximity, the seismic phases arrive earlier at the fiber location than at IV.ASCOL.However, the short distance to the cable may entangle the identification of phase arrival times, preventing from precise picking.The scaling coefficient between recordings on the fiber and seismometer is about 2 × 10 7 Hz/(m/s), a factor ∼20 higher than for the Turkey event (this factor is consistent with other observations on a regional scale).This supports the indication of Fig. 2B, that the wavefront curvature and angle of incidence with respect to local portions of the cable [11] may have a non-trivial impact in determining the overall cable response and amplification.This condition is opposite to the previous case (farfield event), in which the seismic components reaching the cable had wavelength comparable to the cable length, thus washing out local effects and giving rise to a response which is more similar to that of a point-like sensor.More quantitative considerations would require integration of local strain, derived from a-priori knowledge of the wave propagation dynamics [12] which was not available for this event.
Fig. 3C shows the recordings of an event occurred in Accumoli (M l =3.4,epicentre at 32 km from the fiber and 27 km from IV.TERO, labelled 3 in Fig. 2).The scaling factor between fiber and seismometer measurements is similar to the previous case.The arrival time of the P-and S-waves is evident, and the lag is consistent with the wave propagation between IV.TERO and the fiber to within the model uncertainty.
These results confirm that interferometric fiber recordings can be used for earthquake localization based on the picking of phase arrival times on a set of cables [2], provided that standard localization algorithms are adapted to take into account the distributed nature of the sensor.Spectral analysis discloses additional information about earthquakes dynamics.In standard seismometers, Fourier amplitude spectra of accelerations ( v) are predicted to follow a standard Brune's model [25] and increase proportionally to f 2 , with f the Fourier frequency, up to a corner frequency f c , above which they flatten.The model also predicts a relation between f c , the source physical parameters and the event magnitude (see Methods).
We analysed the spectral composition of all events detected by both the fiber and IV.ASCOL in a range of 100 km from our observatory, and extrapolated the respective corner frequencies.
Farther events are excluded, as their spectrum is expected to be significantly affected by propagation.As an example, spectra of three events with increasing magnitude are shown in Fig. 4A (spectra of all events are shown in Extended Data Fig. 6).
The spectral content of fiber recordings reproduces that obtained by IV.ASCOL, highlighting the broadband response of the cable as a sensor and corroborating our decision to compare recordings of ∆ν with v, or the corresponding time derivatives.Also, we note that in all spectra the scaling between the cable and seismometer is consistent with that observed for the Civitella and Accumoli events (Fig. 3B and C), and generally higher than observed for the Turkey (Fig. 3A) event.Figure 4B shows the corner frequency values extrapolated from the fiber (orange, left) and IV.ASCOL (blue, right) as a function of magnitude, highlighting in both cases a decreasing trend consistent with the expected one.The Pearson's correlation coefficient between magnitude and f c is r = −0.60 and r = −0.59respectively, indicating that the reason for sta-

A) B) C)
Figure 3: Time traces of relevant case studies.A) Recordings of the Turkey M wp = 7.9 event occurred on Feb. 6th, 2023, with the fiber (panel 1, orange) and IV.ASCOL (panels 2 to 4, blue).On the right, a zoom on the P and S-waves arrivals.B) Recordings of the M l =2 event occurred in Civitella del Tronto on Feb. 19th, 2022 with the fiber (orange) and IV.ASCOL (blue).C) Recordings of the M l =3.4 event occurred in Accumoli on Feb. 27th, 2022 with the fiber (orange) and IV.TERO (blue).Data from both the fiber and the seismometer have been band-pass filtered with a 6th order, bidirectional Butterworth filter to maximize signal/noise ratio on the fiber recording.The pass-band was 0.005 Hz to 0.5 Hz for the Turkey event, and 1.3 Hz to 20 Hz for the Civitella and Accumoli events.Seismometers data were deconvolved from the sensor's response.10 tistical dispersion is not related to the data quality.Instead, we ascribe it to differences in the physical source parameters between various events, as indicated by the predicted limits when typical numbers are plugged into the model (dotted and dashed lines, see Methods).A Pearson's test between the corner frequencies retrieved by the two sensors (Fig. 4C) returns r = 0.81, confirming a good correlation between the two.However, values extrapolated by the fiber appear to be systematically higher than for IV.ASCOL.This is confirmed by a linear fit to the data in Fig. 4C, which returns a slope of (0.9 ± 0.1)Hz/Hz and intercept of (2.0 ± 0.7) Hz.Further inspection from a larger event catalog will help finding an explanation (see also Methods).
This analysis reveals a crucial aspect of fiber-based earthquake detection, i.e. that spectral analysis of fiber deformations can indeed support magnitude estimation for small and mid-size seismic events, even in the impossibility to calibrate the absolute cable response to ground motion due, e.g., to unknown fiber-to-ground coupling parameters or integration effects.

Discussion
With 1.5 years of uninterrupted data collection on a land-based fiber shared with data traffic, we were able to characterize laser interferometry as a tool for earthquake detection, also taking advantage from the presence of co-located seismometers with well-known response.Within the same period, no degradation in data traffic quality metrics was reported, which is a prerequisite for shared use of the existing fiber infrastructure.
We showed that events with M l =2 or larger can be reliably detected at local distances.For the Italian earthquake catalogue [26], M l =2 represents the average magnitude for completeness and in the surveillance system the alert-threshold to the Civil Protection Department is M l = 2.5 [27]: smaller events are not felt unless they have very superficial hypocenters or happen in quietest moments of the day.Detection of such events with a fiber hosted in a road infrastructure, extending to city centers and including aerial segments reveals concrete potential of laser interferometry for supporting traditional monitoring in highly-populated areas.It also highlights the capability to bring data from weak earthquakes that might not be traced unless seismometers are present in a range of few tens of kilometers from the epicentre.Such a dense coverage is in general only available in high-risk areas and is uncommon on the broadest extent of the planet.
Considering the current increase in fiber cable deployment to meet the growing capacity needs for next-era communications, we believe the interesting target for network-integrated sensing to be weak yet close events, rather than teleseismic signals for which the existing network is already adequate.In the perspective of integrating laser interferometry into telecommunication grids, the scaling law we found for the event detection probability can be exploited to draw territorial maps of covered regions by following the hosting infrastructures path and validate small-event detections in combination with spectral analysis, even lacking information on detailed cable routes and mechanical coupling to the ground.
Sharply measuring the arrival time of the P and S seismic phases is a fundamental step for localization.We provided examples of this capability for events diverse in magnitude and distance, showing that event picking is possible, especially with the S-wave often featuring a steep and distinguishable onset.
Availability of large catalogs from land-based observatories, validated by comparison with seismometer data, is crucial in developing advanced signal analysis and training dedicated machine learning routines [28] and allows studying the fiber sensitivity to ultra-low-frequency signals.Clear signature of daily temperature variations was observed on fiber recordings, which could be compared in the future to other distributed temperature sensing methods or extended to other slow-varying phenomena as suggested in [2].
From another perspective, the analysis of background noise confirms that this technology is suitable for monitoring a large variety of phenomena besides earthquakes, such as vehicle traffic and infrastructure mechanical resonances.
For our experiment, we made use of a research-grade, sub-Hz linewidth laser, although results indicate that performances can be relaxed in favour of integration and lower cost, competitive to those of other sensors.Notably, phase detection could be integrated into modern transceivers [14] similarly to what demonstrated for polarization-based sensing [13].Although phase analysis requires more stable lasers than used today in telecommunications, laser integration technologies are progressing [29]; moreover, thanks to the linear relation with the fiber strain, it features lower computational requirements and higher sensitivity than polarization analysis [14].Overall, this may represent in the future the most effective route to the integration in modern smart multi-service grids in a scalable and sustainable approach.

Experimental layout
The fiber optical length is sensed using an inteferometric scheme, where an ultrastable laser beam is split into two branches, one of which travels the link in a round-trip before being recombined to the other.The phase difference between the two arms is attributed to combined variations of the refractive index and length of the deployed link, induced by changing strain in the surrounding environment.A fixed frequency shifter is introduced to enable detection in the radio-frequency domain instead of the low-frequency band, where detection noise would be degraded by the sampling electronics.The beatnote between the round-trip and the reference beam is demodulated, sampled and processed to extract the optical carrier frequency variations at a typical rate of 1 kHz, then stored in a database.Collected data are filtered and decimated to a final rate of 100 Hz for analysis.The instrumentation is housed at the telecom network nodes and installed in compliance to rack standards.Part of the measurements were collected with a modified optical scheme, based on a two-way transmission.In this configuration, a pair of lasers, each housed in a network node, is launched towards the remote end and interfered with the local laser.The frequency shift experienced by the laser while travelling through the fibers is extracted by combining the interference signals recorded at the two cable ends [20].In both schemes, we make the assumption that perturbations are symmetrical on the two directions, as the fibers are laid parallel into the same cable.
Equipment at the two nodes is hardware-synchronized to 5 µs, and referenced to Universal Selection of events for the sensitivity analysis and validation criteria From continuous acquisitions, we automatically extracted windows where to look for signature of seismic events according to the expected arrival times of the seismic wave to fiber, IV.ASCOL and IV.TERO, derived from a-priori knowledge of the event source position and propagation parameters.For each earthquake, the used source parameters (origin time, hypocenter coordinates and magnitude) were extracted from the seismic catalog produced by INGV seismic surveillance center [32].The source-fiber distance was computed with respect to the middle of the cable (42.736 93 • N, 13.674 261 • E).
Waveforms of earthquakes recorded along the fiber and at seismic stations were selected using the theoretical P arrival times computed using the TauP method [33].The travel times were computed in the 1D Earth reference velocity model ak135 [34], that provides a good fit to a wide variety of seismic phases and improves on the previous IASP91 [35] Earth model in terms of data and methodologies.
We considered in our analysis events with: M ≥ 4 closer than 2000 km, or M l ≥ 2.5 closer than 300 km, or M l ≥ 0.4 closer than 30 km, occurred between June 19th, 2021 and Sept. 26th, 2022.To increase statistics in the high-distance/high-magnitude end of the plot, we added events from the recent sequence occurred in Turkey, including those with M wp ≥ 5 occurred between Feb. 6th and March 23th 2023 closer than 300 km to the epicenter of the first event.Also, we included a seismic sequence occurred off-shore Ancona, at a distance of about 130 km from the middle of the fiber (coordinates of main earthquake of the sequence, Mw 5.5: 43.9833 • N, 13.3237 • E) in November 2022, that featured co-localised events in a broad range of magnitudes.To include this sequence, we considered events with M l ≥ 2 closer than 180 km from the fiber, occurred between Nov. 9th, 2022 and Nov. 18th, 2022.
For each considered timeframe, we looked for signature of the wave detection on the fiber according to typical criteria used in seismic analysis and assigned it a score (0 = event not detected; 100 = event clearly visible; intermediate values reflect the confidence on detection and are plotted in light-green color in Fig. 2 of the main text).To make a decision, we considered both the time trace (i.e.clear onset of a phase from background noise; temporal evolution and duration in consideration of epicenter distance; matching of the recording with the expected arrival time of the seismic wave...) and the spectral composition of the signal.
In Fig. 2 of the main text, events marked green are those who featured a score ≥ 60; unsure events are those who featured a score between 5 and 60.

Interpretation of fiber data and comparison to seismometer recordings
The phase accumulated by the optical carrier as it travels a fiber with length L can be related to the integral of local ground strain ϵ along the full cable path, weighted by a geometrical scaling coefficient γ g and a coupling coefficient γ c .The former depends on the relative orientation of the cable and the local strain direction [11,24] and on the ratio between the seismic wavelength and the fiber length projected along the wave direction: for instance, because of the integration effect, we would expect null response from a perfectly straight cable hit by a seismic wave incident from the longitudinal direction when the cable ends happen to be in nodal points; instead, response would be higher when the cable ends happen to be perfectly out of phase [24].
From a practical point of view, this sort of considerations, well-modeled in standard Distributed Acoustic Sensing, cannot be extended to coherent interferometry in a straightforward way, as real cables always feature changing orientation and have dimensions comparable or larger than seismic wavelengths [11].γ c quantifies the anchoring of the fiber to the surrounding ground and depends on construction factors such as the kind of conduit where the fiber is placed, the cable armoring [6,10] or the presence of gel rather than simple air-filling [36].This establishes a general relation between the instantaneous frequency deviations of the optical carrier with respect to the nominal frequency of the injected signal ∆ν = φ/2π , and the time derivative of integral strain: where we parametrized γ g and γ c accounting for the fact that they are local quantities, λ is the optical wavelength (1.5 µm), the factor 2 indicates that the fiber is travelled in a roundtrip and we assumed a constant refractive index n throughout the cable.The relation between strain, or its derivative, and ground motion parameters is, in general, less trivial.Observations with standard Distributed Acoustic Sensing confirm that a linear relation exists between ε and ground velocity components v along the strain direction recorded at the edges of the gauge length g [4, 24, 37], with g of the order of few meters: However, this relation could be extended to the km-scale only in the case of a perfectly straight fiber under homogeneous strain conditions.On deployed fiber layouts with changing orientation and non-homogeneous deformations, knowledge of velocity components at the cable end are not sufficient to quantitatively predict the cable response and propagation models for the seismic wave need to be computed to enable integration of local deformation through the cable path [12].
Still, we expect a linear functional relation between the recorded frequency deviations and the ground velocity to be preserved, which is consistent with the reported measurements for the analysed cases.

Calculation of spectra and corner frequency extrapolation
For all detected events in a range of 100 km from the the fiber we computed the fast Fourier transform (FFT) of the recording over a time interval of 20 s, with Hann window and no av-eraging to enable sufficient resolution at lowest frequencies.The start time of signal windows was automatically set to be 1 s before the calculated arrival time of the P-wave in the center of the cable, except for two cases where it had to be fixed manually due to incorrect arrival time prediction.Similar procedure was followed to calculate the background noise (shaded area): integration intervals had the same duration, and the start time was either automatically set to 60 s before that of the signal or manually adapted to exclude non-stationary noise arising by chance during the considered period.In this latter case we shifted the window by up to few tens of seconds, maintaining the same duration.The modulus of the FFT was normalized by the square-rooth of the channel bandwidth.Results are shown in Extended Data Fig. 6 both for the fiber (orange) and IV.ASCOL East-West component (blue).
To extrapolate the corner frequency we fitted the above spectra with polynomials of the kind The corner frequency is related to the source physical parameters and seismic moment M o according to [38]: with ∆σ the stress-drop, C d the directionality coefficient ranging between 0.1 and 10, β the rupture velocity and coefficient k = 0.37.Using the relation M o = 10 1.5M +9.1 [39], with M the event magnitude, we can derive the relation between f c and M : Clearly, all these parameters may change from an event to the other, although the greatest variability is found for ∆σ, that may vary on almost two orders of magnitude in the range 0.1 MPa to 3 MPa.
The dashed and dotted lines shown in Figure 4B of main text indicate the corner frequency extrapolated from Eq. 5 using β=3 km/s, C d =1 and ∆σ= 0.1 MPa or 3 MPa respectively.A decreasing trend of the corner frequency is observed as the event magnitude increases, with the dispersion of results consistent with inhomogeneities in source parameters.
In our derivation of corner frequencies, we did not take into account propagation effects on the source spectrum, and did not apply any threshold on the event detection quality, based e.g. on signal/noise considerations.These effects may also play a role in explaining the dispersion of experimental results.As highlighted in the Main text, some bias is clearly distinguishable between values derived from the fiber or IV.ASCOL, whose reason deserves further investigation.At this stage, we cannot exclude some bias in the fit of the low-frequency components of spectra with a polynomial of the kind b 2 f 2 component, due to the high noise at 0.75 Hz and 0.95 Hz affecting the cable.
The analysis was repeated with the vertical and North-South components of IV.ASCOL recordings with no substantial differences.Extended Data Fig. 2: Sensitivity of IV.ASCOL.Map of events detected (green), unsure (light-green) and non-detected (empty) by IV.ASCOL, as a function of their magnitude and distance to the sensor; validation criteria were the same as for the fiber.The dashed blue line reproduces the sensitivity threshold of the fiber, shown in the main text; the solid orange line is the one obtained for IV.ASCOL and corresponds to A 1 = 0.552 ± 0.001 and A 2 = (3590 ± 10) km.The sensor suffers from the high amount of anthropogenic noise in the surroundings and has a higher average noise level as compared to other stations of the INGV network.In spite of this, IV.ASCOL is used in the INGV daily routine localization task, for instance during the Nov. 2022 sequence off-shore Ancona (coordinates of main earthquake of the sequence, M wp 5.5: 43.9833 • N, 13.3237 • E).The insets zoom on regions with highest density of events around the threshold.
Extended Data Fig. 3: Background noise of the fiber.Time series of the unfiltered fiber recording over one day (one hour per line), showing increase of the noise at daily hours.We attribute this behaviour to higher vehicle traffic levels on the road where the fiber is hosted, and people activities around the cables, likely in city areas where the cable ends.Occasional, nonstationary solicitations to the cable are also visible.The red area corresponds to the Civitella event described in the main text.

Figure 4 :
Figure 4: Spectral analysis and magnitude determination.A) Spectral composition of the accelerations recorded by IV.ASCOL East-West component (blue, right panel) and fiber differentiated frequency data (orange, left) for three events with epicenters in Civitella (1st and 2nd row) and Accumoli (3rd row) and M l =2.0, 3.0 and 3.9.The spectral response of the fiber well reproduces that of the seismometer.Black dashed lines indicate fits on the low and high end of the detection band, their crossing corresponding to f c .The shaded area indicates the background noise recorded by the sensor 1 minute before the event .B) f c as retrieved from IV.ASCOL (right, blue) and fiber (left, orange) are plotted as a function of magnitude, evidencing a linear relation between the two quantities.Dashed and dotted lines indicate lower and upper limits obtained for typical source parameters; in particular, we let the stress drop ∆σ to vary between 0.1 MPa and 3 MPa (see Methods).C) Dispersion plot of the corner frequency as extrapolated from the fiber and IV.ASCOL, with color scale representing magnitude.Correlation between the response of the two sensors is found (r = 0.81), suggesting that spectral analysis of fiber recordings enables to infer event magnitudes.The dashed line is a linear fit to the data.

Coordinated
Time (UTC) to about 10 ms using Network Time Protocol provided by Ethernet connection.The link implements Dense Wavelength Division Multiplexing (DWDM).All optical carriers, included the one used for this work, are combined through an Optical Add/Drop Mul-tion server, where they can be gathered and distributed via Web Services based on the FDSN specification [31].

FFT(f ) = b 2 f 2
and FFT(f ) = b 0 f 0 in Fourier frequency intervals 1.2 Hz-3 Hz and 10 Hz-30 Hz respectively.The interpolation limits have been chosen to exclude spectral regions where the noise on fiber recordings covers the signal, and especially the discrete noise peaks at 0.75 Hz and 0.95 Hz.The corner frequency is calculated as the crossing point between the two fitted lines, i.e. f c = b 0 /b 2 .

. 1 :
Average azimuth of the cable.We approximate the full cable path to a series of straight segments and for each we calculated the orientation.Polar histogram bars indicate most recurring orientation angles.The red line is the result of a circular mean of all segments and indicates that the cable has an average azimuth of 155 • .