We demonstrate the capability of distributed acoustic sensing (DAS) to record volcano-related dynamic strain at Etna (Italy). In summer 2019, we gathered DAS measurements from a 1.5 km long fibre in a
shallow trench and seismic records from a conventional dense array
comprised of 26 broadband sensors that was deployed in Piano delle Concazze close to
the summit area. Etna activity during the acquisition period gives the
extraordinary opportunity to record dynamic strain changes (
In the recent past, direct measurements of the strain field have been hampered by the complex installation and high maintenance cost of conventional strainmeters. In the best case, a few instruments are deployed in borehole, thus sensing the strain at only a few points (Currenti and Bonaccorso, 2019). Numerical investigations have clearly shown that seismic velocities, gradient displacements and strain dramatically changes at sharp boundaries and in the presence of steep topography (Kumagai et al., 2011; Jousset et al., 2004; Cao and Mavroeidis, 2019). This poses challenges in accurately interpreting strain observations when relying on only a few measurement points.
Nowadays, advances in the direct measurement of strain at an unprecedented high spatial and temporal sampling over a broad frequency range have become possible due to the growing use of distributed acoustic sensing (DAS) technology. Since the application of DAS in geoscience is an emerging field, many open questions still exist about the DAS device response and the coupling effect between the fibre and the ground, which strongly depends on the conditions of the fibre installation (Ajo-Franklin et al., 2019; Reinsch et al., 2017). A few field experiments in various environments have been designed to compare DAS strain measurements and indirect strain estimates from co-located or nearby traditional sensors, such as geophones and broadband seismometers (Jousset et al., 2018; Wang et al., 2018; Yu et al., 2019; Lindsey et al., 2020).
In this paper we compare array-derived strain with direct-strain DAS data acquired at the Etna volcano (Italy). We selected Etna as a test site for its various and frequent activity. Lava flows, explosive eruptions with ash plumes and strombolian lava fountains commonly occur from the Etna summit craters, and these multiple episodes of different styles result in a wide variety of signals (Bonaccorso et al., 2004). A dense seismic array and a 1.5 km long fibre optic cable connected to a DAS interrogator were jointly installed a few kilometres away from the summit craters to assess the reliability of DAS in recording volcano-related strain changes. To validate the DAS measurements, we explore several methods for the indirect estimates of strain field from dense seismic array data.
Several methods exist in the literature for estimating strain from high to quasi-static frequencies that are relevant for seismologic and geodetic investigations. These include the evaluation of seismic rotational components (Basu et al., 2013), the analysis of the performance of rotational seismometers (Suryanto et al., 2006), the estimation of strainmeter response and calibration (Donner et al., 2017; Currenti et al., 2017), the computation of strain rate maps (Teza et al., 2004; Shen et al., 2015), the estimation of the stress field induced by the passage of seismic waves (Spudich et al., 1995), the determination of seismic-phase velocity (Gomberg and Agnew, 1996; Spudich and Fletcher, 2008), and the estimates of wave attributes (Langston and Liang, 2008).
In general, those methods can be grouped into two main families: those relying on single- or multiple-station methods. On the one hand, the single-station method, pertinent only to seismology, has been widely used to estimate dynamic gradient displacements and strain tensors from the translational components of a 3C seismometer (Gomberg and Agnew, 1996). It assumes that seismic energy is carried by plane waves with a known horizontal velocity in a laterally homogeneous medium. On the other hand, multiple-station procedures involve the use of displacements or velocity recordings from a number of close sensors (at least three; Basu et al., 2013; Currenti et al., 2017) or dense arrays (from tens to thousands of sensors, with short spacing covering tens to hundreds of metres). The density and number of stations depends on the objectives of the array (Basu et al., 2017). In such cases, the measurements are processed using spatial interpolation approaches (Sandwell, 1987; Paolucci and Smerzini, 2008; Sandwell and Wessel, 2016) or the well-known seismo-geodetic method (Spudich et al., 1995; Shen et al., 2015; Teza et al., 2004; Basu et al., 2013). Aside from the use of geostatistical approaches, which are generally applied for optimal interpolation of a discrete scalar field, several interpolation schemes have been formulated using the solutions of suitable elasticity problems (Sandwell, 1987; Wessel and Bercovici, 1998; Paolucci and Smerzini, 2008). A complete review of these latter methods can be found in Sandwell and Wessel (2016). In the same paper, the authors, following the biharmonic spline method (Sandwell, 1987), proposed a new solution in which the interpolation functions originate from the computations of the analytic Green's solutions of an elastic body subjected to in-plane forces. Finally, the seismo-geodetic method is based on the Taylor expansions of the displacements field of first- (Spudich et al., 1995) or higher-order derivatives (Basu et al., 2013). Both interpolation and seismo-geodetic methods have been used for seismic dynamic and geodetic quasi-static strain estimates. While the single-station procedure estimates the strain at the single position of the collocated sensor, the other two methods allow deriving a map of strain on irregularly distributed points by making use of sensor array measurements. This makes the multiple station approach appropriate to estimate strain using dense seismic array methods and compare it with values derived from DAS along the fibre cable.
Here, we extend the interpolation method proposed by Sandwell and Wessel (2016) and use the seismo-geodetic method in the formulation by Shen et al. (2015) in order to derive strain estimates from the dense seismic array data acquired at Etna. The aim of the paper is twofold: (i) exploring the performance of interpolation and seismo-geodetic methods in deriving the strain field along a fibre optic cable and (ii) validating the DAS measurements acquired during the experiment at Etna volcano.
We designed the experiment in Piano delle Concazze at the Etna summit (Fig. 1) in order to test the potential of the DAS technology in a volcano-seismology application.
Digital terrain model of Piano delle Concazze on the north-eastern flank of the Etna volcano (Palaseanu-Lovejoy et al., 2020) with the dense broadband seismic array (26 stations; Bb01–Bb26) and distributed acoustic sensing (DAS) cable layout. The DAS cable geometry is designed in order to record dynamic strain changes along several directions. No data were recorded by station Bb05 because of a technical problem. The DAS interrogator was hosted inside the Pizzi Deneri Observatory, which is about 2 km away from the five summit craters (North East: NEC; Voragine: VOR; Bocca Nuova: BN; South East: SEC; New South East: NSEC). The investigated area, which is almost flat, is crossed by a sub-vertical fault system (redrawn following Azzaro et al., 2012). Geographic coordinates (in kilometres) are in the UTM33S system.
Deployment of equipment on active and dangerous volcanoes is challenging due to the harsh environment and the danger associated with the volcanic activity. Piano delle Concazze is a large flat area (elevation of about 2800 m) on the northern upper flank of the Etna volcano that is dominated by the North East Crater. It is bounded by the upper extremity of the North-East Rift, a preferential pathway for magma intrusions due to its structural weakness (Andronico and Lodato, 2005), and by the rim of the depression of the Valle del Leone. The area is affected by several north–south-trending faults that result from the accommodation of the extension exerted by the North-East Rift (Azzaro et al., 2012; Napoli et al., 2021). Therefore, Piano delle Concazze is an area where this new methodology can be tested and is close enough to the active craters to study volcanic processes (ca. 1.8 km away from the Etna summit craters) while still being safe.
In order to record strain changes related to volcanic activity, we jointly
deployed the following instruments in Piano delle Concazze (Fig. 1).
A 1.5 km long fibre connected to a DAS interrogator (“Ella”, an iDAS® from Silixa) set up in Pizzi Deneri Observatory. We interrogated the fibre from 1 July to 23 September 2019. A long trench was dug to deploy the fibre cable at a depth of about 40 cm. At a sampling rate of 1 kHz, the DAS acquired the strain rate along the axial direction of the cable with a spatial resolution of 2 m and a gauge length of 10 m. This results in a dataset of 824 channels distributed along the fibre path. The spatial calibration and locations of the fibre channels were determined by jumping at several points near the fibre with jumping locations determined by GPS. The positions of the channels between the two successive jump locations were computed by linear interpolation. A dense seismic array comprised of 26 Trilium Compact 120 s broadband sensors (inter-station distance of about 70 m) distributed over an area of about 0.2 km
We designed the broadband sensor distribution and the path of the fibre optic cable such that we could compare both records with several methods.
We estimate the strain field from the seismic array data using two different algorithms. The first is a new spatial interpolation method, in which we extend the work presented in Sandwell and Wessel (2016). The second is the seismo-geodetic method in the formulation proposed by Shen et al. (1996, 2015). For both methods the description and the analysis are limited to the 2D domain, since the investigated area is almost flat. However, extension to 3D is straightforward.
In the spatial interpolation method (SIM) proposed by Sandwell and Wessel
(2016), the general solution of the horizontal displacement components
The seismo-geodetic method (SGM) is based on the Taylor expansion of the
displacement field and has been applied using different assumptions and
strategies that lead to very similar formulations (Spudich et al., 1995;
Shen et al., 1996; Teza et al., 2004; Langston and Liang, 2008; Basu et al., 2013; Langston, 2018). Here, we follow the formulation proposed in Teza et al. (2004). The displacement components
Using one of the above methods, the strain tensor is computed at all channels of the fibre and is projected along the local fibre direction to compute the local axial strain. Both procedures are iterated at each time step to obtain the time series of dynamic strain at all the channels of the fibre using the time series of the seismic array.
The accuracy of the array-derived strain estimates is limited by the
inter-station distance
The SIM and SGM depend on the choice of some parameters that need to be
tuned in order to obtain optimal solutions. They are the Poisson ratio and
the
During the acquisition period, Etna volcanic activity was mainly characterised by discontinuous strombolian explosions and isolated ash emissions from most of the active summit craters (North East, New South East, Bocca Nuova and Voragine) at a distance of about 2 km from Piano delle Concazze (Fig. 1). These activities, occurring at fluctuating intensity, preceded and accompanied the short-lived effusive eruptions on 18 and 27 July 2019 from the New South East Crater (NSEC) and continued until the end of the experiment. A wide variety of signals have been recorded, e.g. volcanic tremors, LP events, volcanic explosions, teleseismic and local seismic events. In order to validate the DAS records with the approaches described above, here we focus our analysis on classes of events with a frequency content less than 6 Hz. Among the several signals, we selected two types of events (Fig. 2): (i) a volcanic explosion (VE) accompanying the strombolian activity at NSEC on 6 July 2019 and (ii) a long-period event (LP) on 27 August 2019 preceding the intensification of the eruptive activity at the summit craters in early September 2019. In agreement with other similar events recorded at Etna (Cannata et al., 2009), the spectra show frequency contents in the range 0.1 to 1 and 1 to 5 Hz for the LP and the VE events, respectively. Both broadband and DAS signals are filtered with a third-order Butterworth filter. The integration of DAS data over time provides the strain along the fibre optic cable. Data are down-sampled from 1 kHz to 200 Hz for direct comparison with strains derived from broadband array signals.
Time series of DAS strain during a small volcanic explosion (VE)
at Etna on 6 July 2019
Some of the broadband seismometers (e.g. Bb02, Bb03, Bb04) are co-located
with the fibre optic cable (distance of less than 1 m). This configuration
also enables the application of the single-station method, which under the plane wave
assumption relates the DAS strain at the nearest channel with the broadband
particle velocity as
Velocity data are first integrated over time to derive the displacement field, which is then used in the SIM and SGM methods. Residuals, in terms of RMSE (root-mean-square error) misfits, are computed between DAS strain measurements and the strain derived from the seismic array. The RMSE values computed along the fibre give a local estimate of their respective discrepancy (Figs. 3 and 4).
RMSE residuals between DAS strain measurements (Fig. 2a)
and the strain derived from the seismic array along the fibre for the VE
event.
RMSE residuals between DAS strain measurements (Fig. 2c) and the strain derived from the seismic array along the fibre for the LP event. Black lines and open circles indicate faults and seismometers shown in Fig. 1, respectively.
The SGM depends strongly on the spatial smoothing parameter
Cumulative residuals over all fibre channels between DAS
strain measurements and strain derived from SGM (blue line) and from SIM
(red line) for different smoothing parameters (D for SGM and
An overall good match is achieved for both methods (SGM and SIM) along the fibre. Higher RMSE misfits concentrate at the corner points, where the cable direction turns abruptly and the axial strain is locally disturbed. Larger misfits are also visible along the two nearly EW (east–west) branches (channel 134–302; 449–787), where the strain wave field (Figs. 2 and S1–S2 in the Supplement) is more complex and amplified with respect to the nearly NS (north–south) branch (channel 302–449). Furthermore, discrepancies are also observed in regions where the fibre crosses fault zones (Figs. 3 and 4). The amplitude of the array-derived strain estimates in the fault zone is mostly underestimated.
We investigated several methods for indirect strain estimates from dense seismic array data, aiming at the assessment of DAS records. After a straightforward recasting and derivation of equations, we compared the SIM and the SGM methods for the estimates of strain fields from dense array data, which, in similar forms, have been applied in several fields of seismology and geodesy. To our knowledge, these approaches are adapted and tested on DAS data for the first time here.
SIM and SGM offer some advantages over single-station methods. Both provide a direct comparison between strain data and their estimates from velocity data, without any assumption on the local-phase seismic velocity as required by single-station procedure. Specifically, when media are dispersive, the assumption of a constant phase velocity is possibly prone with errors. Usually, and especially in volcanic areas, the estimate of a phase seismic velocity is challenging because of the presence of strong heterogeneity and fractured zones. Specifically, at Piano delle Concazze the phase seismic velocity varies markedly along the fibre cable because of the local complex structural geology that is characterised by a lava flow succession interbedded with volcaniclastic products (Branca et al., 2011) and by sub-vertical north–south trending faults affecting the superficial layers up to a maximum depth of about 40 m (Azzaro et al., 2012; Napoli et al., 2021).
Indeed, our results highlight strong discrepancies between direct DAS
measures and indirect strain estimates in coincidence with fault zones (Figs. 3, 4). Thanks to the high spatial resolution of the DAS records, it is
possible to observe how small-scale soil structure heterogeneities affect
strain. Local strain perturbations are much shorter in wavelength (a few tens
of metres) with respect to the wavelength resolution of the array (
Array-derived strain estimates at station Bb04 (Fig. 1)
using the SIM (green line) and the SGM (blue line) in comparison with DAS
strain data (red line). Computations are performed for the VE
The discrepancies are higher for the shorter wavelength signal recorded during the VE event with respect to the LP event (Figs. 6, 7), for which good estimates are obtained at almost all channels (Figs. 4, 8). Indeed, as already noted in Currenti et al. (2020), explosive events excite more phases due to scattering and reflection on faults and layered geology (Figs. 2 and S1). The discrepancies are larger in the nearly EW branches, possibly due to the relative direction between the main structural geology and the cable branches. For the longer wavelengths (e.g. LP event), SIM and SGM (Fig. 6) perform better than the single-station approach (Fig. 2).
Strain estimates over all fibre channels using SGM
Strain estimates over all fibre channels using SGM
We also attempt to investigate how the gauge length may average out the effect of the small-scale heterogeneities by virtually increasing the gauge length with a spatial average of DAS data. We report the analysis on the VE event (Figs. S3–S8), where the effect of increasing the gauge length from 10 to 30 and 100 m is more significant due to the higher-frequency content. Indeed, as expected, at higher gauge lengths the shorter wavelengths are filtered out (Fig. S3, S6). We used two averaging approaches: (i) a simple average over channels (Figs. S4, S7) and (ii) a moving average with a shift of 1 channel (Figs. S5, S8). With a gauge length of 30 m, the simple average degrades the signals and the main phases are already lost. On the other hand, the moving average preserves the main signal but smooths out local scattering and reflections that are no longer visible. When computing the misfit with the array-derived strain estimates (Figs. S4–S8), the localised anomalies (Fig. 3) in coincidence of the faults due to the small-scale heterogeneities are flattened and broaden. These findings confirm that the distortion of the strain field is very localised and difficult to observe via traditional seismic array methods, which require deployment of very dense network along “well-chosen active faults and a good amount of luck” (Cao and Mavroeidis, 2019).
Both methods allow for estimating strain on points distributed irregularly exploiting all the available dataset instead of relying on a single-point measure (Figs. 7, 8). Moreover, the derivation of the analytical strain solutions in the SIM (i) avoids using the finite-difference scheme to derive strain from regular grid point distribution of displacements and hence (ii) provides the strain at any point of the investigated area once the body force coefficients have been estimated by solving the linear system of equations. This results in a greater accuracy.
Finally, the proposed approaches offer the possibility to combine seismic array data with DAS measures to derive a 2D map of the local strain of the investigated area. By recasting the system of linear equations, it is straightforward to include the DAS strain measures and perform a joint inversion.
The joint deployment of a DAS device and of a dense seismic array at Etna summit offers a unique opportunity to observe and accurately quantify strain changes related to volcanic activity. The dataset recorded during summer 2019 showed the great potential of distributed fibre optic sensing in a volcanic environment. To our knowledge this is the first time that tiny strain changes related to volcanic explosions and LP events have been clearly recorded by DAS technology, opening new perspectives for its use in volcano monitoring. The high spatial sampling of DAS measurements confirms the high variability of strain variations in complex geology and may offer a great opportunity to study soil response. These findings also contribute in explaining the difficulties often encountered in interpreting local strain changes from single strainmeter observations. The indirect strain estimates derived by the dense seismic array match quite well with the direct DAS measurements. Our findings validate both the proposed methods and the accuracy of DAS measurements in sensing strain changes produced by volcanic processes.
MATLAB scripts and data are available upon request to the corresponding author.
The supplement related to this article is available online at:
PJ, CK and MW conceived and supervised the project. PJ, GC and RN were involved with the experiment planning. GC conceptualised this study and performed the analyses. All authors contributed to the acquisition of the field data, the writing of the manuscript and the discussion of the results.
The authors declare that they have no conflict of interest.
This article is part of the special issue “Fibre-optic sensing in Earth sciences”. It is not associated with a conference.
Broadband seismometers and data logger equipment are from the Geophysical Instrument Pool Potsdam (GIPP). Thanks are due to Valentin Parra and the INGV staff composed of Salvatore Consoli, Danilo Contrafatto, Graziano Larocca, Daniele Pellegrino and Mario Pulvirenti for their great help during the field work.
This research has been supported by the INGV, GeoForschungZentrum Potsdam, and Helmholtz Association. The experiment was also financially supported through the Trans National Activity “FAME” within the EUROVOLC project (EU grant agreement ID: 731070).
This paper was edited by Zack Spica and reviewed by Corentin Caudron and one anonymous referee.