An efficient seismic wavefield recording is achieved if the network has sufficient sensors dynamic range, frequency response and spatial resolution (Evans et al., 2005). The first two aspects can be tackled by the installation of modern seismographs with very large dynamic range (> 120 dB) and broadband response (10⁻² to 10² Hz). The weak point in the experimental seismic response analysis still remains the limited number of sensors installed throughout the territory. Densely distributed arrays would lead to better earthquake localization and local seismic response definition, but the use of high-quality sensors makes this option prohibitively expensive, in terms of both purchase and maintenance costs. An alternative approach can be based on the use of cheaper sensors such as Micro Electro-Mechanical Systems sensors known as MEMS (Fleming et al., 2009; Evans et al., 2014). Analogic and digital MEMS technology has proven to reach a performance efficient enough to be adopted as ground motion accelerometers for moderate (M > 4) to strong seismic events (D'Alessandro et al., 2014; Lawrence et al., 2014; Boaga et al., 2019). Nowadays, MEMS can easily meet the moderate-to-strong motion seismological needs, having instrumental noise of ≈100 µg Hz${}^{-\mathrm{1}/\mathrm{2}}$, a fairly broad frequency response and an acceleration range up to 1–4 g.

In the spring of 2016 we installed 20 prototypes of MEMS accelerometers in the Perugia – Foligno area in the Central Apennines (Fig. 2). The MEMS network was installed in the local telecommunication infrastructures (owned by Telecom Italia Mobile – TIM S.p.A.), originally for infrastructure monitoring purposes. The MEMS accelerometers are new prototypes designed and built by AD.EL, a company specialised in distributed sensors and telecommunication supplies. The MEMS prototypes, named D600003A, were first tested under excitations at sweeping frequencies in a certified shaking table, giving deviation compared to the reference of less than 0.3 % for frequencies below 20 Hz in the acceleration range 0.1–10 m s⁻². The MEMS sensor has small size ($\mathrm{110}\mathrm{mm}\times \mathrm{60}\mathrm{mm}\times \mathrm{40}\mathrm{mm}$) and includes: 3 component accelerometers, a local storage card and several possible communications options from cable Ethernet transmission to GPRS communication tool. Timing was provided by an external GPS antenna. The technical specifications of D600003A are provided in Table 1.

Table 1Technical specifications of the ADEL D600003A MEMS accelerometer.

Download Print Version | Download XLSX

The prototype MEMS sensor arrays recorded the main shocks of 24 August and 30 October of the Central-Italy 2016 sequence. The resulting data are of good quality, especially if compared to the closest high-quality National Accelerometers Network stations (RAN see Fig. 2) – see Boaga et al. (2018) for details. The National Accelerometers Network stations (RAN, ran.protezionecivile.it) adopt conventional high quality strong motion sensor such as the Kinemetrics Episensor ES-T (kinemetrics.com/post_products/episensor-es-t). Figure 3 shows, as an example, the comparison between MEMS and the high-quality RAN station close to the city of Gubbio and Spello for the 2 main earthquakes of 24 August and 30 October. The stations inter-distances were less than 0.2 km and both are located in the same soil condition. Both peak ground acceleration and spectral response are comparable, especially in the range of interest for a-seismic design (0–15 Hz).

Figure 3Comparison between MEMS based prototypes (black) and closest RAN sensors (grey) for the 24 August “Accumoli” earthquake (a, c, e) and the 30 October “Norcia” earthquake (b, d, f). (a, b) “Spello” and “Padule” MEMS time series are compared to national stations RAN “CSA” and “GBC”; (c, d) frequency domain responses; (e, f) Elastic response spectra in the period range 0–5 s (5 % damping).

Download

The earthquake sequence occurred in an area where a number of scientific and technical continuous GPS (CGPS) stations are active (Fig. 2). An accurate study of the CGPS time series can provide information about co-seismic and post seismic displacements, as proven by many authors (e.g. Cheloni et al., 2017; Cenni et al., 2012; Savage and Langbein, 2008).

The GPS observations, acquired with a sampling rate of 30s, were initially analysed using the GAMIT/GLOBK software, 10.7 release (Herring et al., 2018a, b) adopting a distributed approach (Dong et al., 1998; Cenni et al., 2012, 2013). This procedure provides the daily time series of the local geodetic components North, East and Vertical for each position, referred to the international frame ITRF2014 (Altamimi et al., 2016). The estimated time series were then analysed separately by the procedure suggested by Cenni et al. (2012, 2013) in order to estimate: (i) the tectonic velocity, (ii) the discontinuities due to seismic events and instrumental features, (iii) seasonal signals and noise. The daily positions were firstly processed to detect and remove outliers, allowing a preliminary estimation of velocity and displacement values. A theoretical model of the time series is then estimated and subtracted from the observed data. The residual time series are then analysed adopting the Lomb (1976) and Scargle (1982) approach in order to evaluate the first 5 frequencies with the most powerful peaks in the interval between one month and half of the observation time span. These frequencies have been successively considered for a more detailed model of the coordinate time series:

where k is the local geodetic component of the position (North = 1, East = 2 and Vertical = 3), A_k and v_k are the intercept and tectonic velocities, $D_{\mathrm{lk}}=\sqrt{B_{lk}^{\mathrm{2}}+C_{lk}^{\mathrm{2}}}$ is the amplitude of the l seasonal signal with period P_l, g_kj terms are the estimated offset magnitudes for the N discontinuities due to instrumental changes or seismic events eventually occurred at the Tj epochs, while H is the Heaviside step function, ε_k (t_i) is the error term.

The parameters can be estimate by the Weight Least Squared Method (WLSM), highlighting the discontinuities due to events or instrumental changes which occurred in a time span of few days. The lack of data between two discontinuities can limit the reliability the displacements estimation. This problem could manifest itself when two or more seismic events occur in a few days, as for example in the central Italy sequence where the main event of the 30 October (M_w = 6.5) occurred only 4 d after another shock of M_w = 5.9. During these few days the operations of several CGPS sites could have been interrupted and the observations lost. For these reasons, here we focus on the 24 August earthquake main shock data, in order to use all available stations data. The geodetic displacements for the 24 August earthquake are shown in Fig. 4. The estimated co-sesmic displacements are also reported in Table A1 (Appendix A1).

Figure 4Horizontal (a) and Vertical (b) co-seismic displacements due to the 24 August 2016 main event. The displacements (millimetres) are estimated analysing the coordinate time series of the continuous GPS permanent stations located at a distance smaller than 100 km from the epicentral area. The white diamonds indicate the locations of the CGPS sites. The epicentres of the events occurring during the period 1 June 2016–26 December 2017 are also reported in the Figure, with the circles' size of proportional to the moment magnitude. T_s =  hanging-wall Tyrrhenian sector; A_s =  footwall Adriatic sector. (Built with the Generic Mapping Tool software GMT – Creative Commons Attribution 4.0 License.)

The co-seismic patterns shown in Fig. 4 are consistent with the activation of a SW dipping normal fault system and the results obtained by other research groups (e.g. Cheloni et al., 2016, 2017). The horizontal field is characterized by a general SW-NE oriented extension, while the vertical movements show a more heterogeneous pattern. In particular, the stations located within few tens of km from the epicentre and placed on the hanging-wall Tyrrhenian sector (Fig. 4) present an uplift not completely coherent with a normal fault source as described by the focal mechanism. The recorded displacements are anyway in good agreement with InSAR observations analysed by Bonini et al. (2016). The vertical field on the footwall Adriatic sector (Fig. 4) shows a more coherent pattern, even if some sites located at few tens of km from the event show a non-negligible subsidence in contrast with the movements of the entire structure. These differences could be due to local movements of the building where the CGPS stations are installed, or to local displacements.

Over the past decade, several authors (Avallone et al., 2009, 2012; Bock et al., 2004; Cheloni et al., 2016; Wang et al., 2007; Xiang et al., 2019) demonstrated that the GPS instruments are able to infer the dynamic features of fault rupture processes, when they are set up with an high frequency sampling interval (≤1 Hz). These dynamic observations are several orders of magnitude less sensitive than the common high quality seismometers readings, but do not suffer from drift, clipping, or instrument tilting (Avallone et al., 2009). The so-called High-Rate GPS (HRGPS) time series have been recently compared with the strong motion data acquired from high-quality accelerometric networks (Cheloni et al., 2016). In this work we want to compare the HRGPS series with the data acquired from the low-cost MEMS prototype distributed network, in order to understand if such not-conventional networks can be also adopted to retrieve seismic motion parameters. We used the available high rate recordings of the CGPS stations compared to the closest available MEMS data (see Table 2). We selected the area close to Foligno (Fig. 5) that hosts a CGPS station, with observations acquired during the events with a sampling rate of 1 Hz, and 4 MEMS sensors in the surrounding (named respectively B00393, B00396, B00399 and B00402).

Table 2Distances in km between the MEMS accelerometers and the Foligno CGPS site (FOL1, Fig. 5), and the earthquake epicentre of 24 August 2016 event. The North and East columns contain the distance along these horizontal components, while in the Dist. columns we report the modulus of the horizontal distance.

Download Print Version | Download XLSX

Figure 5Location of the Foligno CGPS site (FOL1, diamond) equipped with a sampling rate of 1 Hz and the closest (distance < 10 km) MEMS sensors in the area (inverted triangle). The thick back lines show the principal tectonic lineaments from the DISS database of Individual Seismogenic Sources (DISS, Basili et al., 2008). The epicenters of the events occurring during the period 1 June 2016–26 December 2017 are indicated with gray circles, the size of the symbols are proportionally to the moment magnitude (M_w) following the scale on the figure. The star indicates the location of the 24 August 2016 main shock. (Built with the Generic Mapping Tool software GMT – Creative Commons Attribution 4.0 License.)

The GPS observations at 1 Hz sampling have been analysed using the GAMIT/GLOBK software (Herring et al., 2018a, b), using the kinematic module TRACK. This module performs a relative kinematic positioning, where it is necessary to define a reference station. The reference observations should be not affected to the dynamic displacement due to the seismic rupture. Therefore, we assumed as a reference station the Cassino site (CASS, Figs. 1 and 2), that is located outside the strike area of the earthquake. The distances between the GPS station and the MEMS sensors are smaller than 10 km, and two accelerometers are particularly close to the Foligno CGPS station with distance within 1.5 km range. In order to get a fair comparison between data, the MEMS observations have been decimated at the same frequency (1Hz) of the GPS data.

The noise level in the time series must be taken into account in order to investigate whether dynamic displacements due to the seismic event can be detected by the HRGPS station. As a representative level of the noise we have taken the Root Mean Square (RMS) value estimated during the time interval ranging from 90 s before the origin time of the event (01:36:32 UTC, Marchetti et al., 2016) and 10 s afterwards (we considered the data up to 10 s after the UTC main shock because the MEMS and HRGPS Foligno arrays are located about 50 km from the epicentral area, so not including the event). The estimated RMS values of the HRGPS components and MEMS time series are reported in Table 3.

Table 3Root Mean Square (RMS) values of the HRGPS and MEMS time series estimated using the data available from 90 s before and 10 s after the main UTC time of the event, during quite acquisition time. The GPS and MEMS values are respectively in mm and m s⁻². The columns reported the values estimated analysing the time series of the North, East and Vertical components.

Download Print Version | Download XLSX

4.1 Results

Figure 6 shows the HRGPS displacement time series and the MEMS accelerations acquired during the 24 August earthquake. The horizontal axis in Fig. 6 is scaled in seconds from the main shock origin time (01:36:32 UTC, Marchetti et al., 2016). The North component of Fol1 GPS station (thick dashed line in Fig. 6) shows an initial displacement of about 30 mm peak-to-peak, while the displacement along the East coordinate is about 20 mm peak-to-peak. These values are about twice the RMS values reported in Table 3, assumed as the noise amplitude of the time series. The co-seismic dynamic displacements observed on the vertical component are conversely similar to the RMS value (about 10 mm), therefore it is difficult to identify the first arrival of the seismic sequence and the other seismic signals. The inelastic co-seismic movements observed at the same station are shown in Fig. 4 and are of −3.4 and −0.6 mm (Table A1) along respectively the East-West and North-South directions, while the vertical component shows a subsidence of −2.1 mm (Table A1). These values are one order of magnitude lower than the initial elastic displacements observed in the HRGPS time series during the event.

Figure 6GPS-derived displacement waveform recorded at the Foligno permanent station (thick dashed line) and low-cost MEMS array signals (thin solid line). Observations are presented for the three local geodetic components: North, East and Vertical, with the same sampling rate of 1 Hz. The MEMS instrument code is reported in each figure. The horizontal axis is given in seconds after the main shock origin time: 01:36:32 UTC of 24 August 2016. The right vertical axis denotes GPS displacements, while accelerations refer to the left axis. According to the epicentral distances, the MEMS signals of the four stations are arranged in decreasing order from top to bottom. (Built with the Generic Mapping Tool software GMT – Creative Commons Attribution 4.0 License.)

Download

The local geodetic coordinate system adopted to show the results in Fig. 6 is probably non optimal to compare the high rate GPS time series against the data from MEMS. Therefore, the horizontal components (North and East) of GPS and MEMS sensors have been rotated with respect to the epicenter and presented in the Radial-Transverse (RT) coordinate system. The azimuth angle adopted to rotate the horizontal components in the RT system is 303^∘ as suggested to other authors (Avallone et al., 2016). The radial component should contain mainly P and SV-wave arrivals, while the SH-wave modes should be predominant in the transverse component.

To analyse the similarity between MEMS and HRGPS observation we have transformed in displacements the MEMS recording by applying a double-integration. We have firstly filtered the MEMS and HRGPS observations by a Butterworth high pass filter. Figures 8 and 9 show respectively the MEMS and HRGPS displacements times series of the local geodetic components (North, East, Vertical) and the ones in the RT coordinate system.

Figure 7Acceleration time series from MEMS and HRGPS displacements in the Radial-Transverse coordinate system. The thick dashed lines represent displacement estimated using the HRGPS displacements of the Foligno (FOL1) permanent site. The thin solid lines show acceleration time series recorded at four different MEMS stations – the code of each instrument is reported. The MEMS signals of the four stations are arranged in decreasing order from top to bottom according to epicentral distances. The horizontal axis denotes seconds from the origin time of the main shock: 01:36:32 UTC of 24 August 2016. The right vertical axis denotes GPS displacements, while accelerations refer to the left axis. (Built with the Generic Mapping Tool software GMT – Creative Commons Attribution 4.0 License.)

Download

Figure 8MEMS (thin solid line) and HRGPS (thick dashed line) displacements time series obtained after the filtering and the double-integration. Observations are presented for the three local geodetic components: North, East and Vertical, with the same sampling rate of 1 Hz. The MEMS instrument code is reported in each figure. The horizontal axis is given in seconds after the main shock origin time: 01:36:32 UTC of 24 August 2016. According to the epicentral distances, the MEMS signals of the four stations are arranged in decreasing order from top to bottom. (Built with the Generic Mapping Tool software GMT – Creative Commons Attribution 4.0 License.)

Download

Figure 9MEMS and HRGPS (thick dashed line) displacements in the Radial-Transverse coordinate system. Thick dashed lines and thin solid lines represent respectively the HRGPS and MEMS displacements. The MEMS signals of the four stations are arranged in decreasing order from top to bottom according to epicentral distances, and the code of each instrument is reported on the figures. The horizontal axis denotes seconds from the origin time of the main shock: 01:36:32 UTC of 24 August 2016. (Built with the Generic Mapping Tool software GMT – Creative Commons Attribution 4.0 License.)

Download

The displacement waveforms of the MEMS network and GPS station are qualitatively in agreement (Figs. 8 and 9). The displacement amplitude observed by the MEMS instruments are in fact similar to ones registered on GPS sites. Only the B00396 MEMS radial component presents greater displacement compared to the HRGPS station ant others MEMS stations. These differences are probably due to the large distance between the B00396 and GPS (about 9 km, Table 2) and the other accelerometers (Table 2). Also, the B00396 site is the closest to the earthquake respect the other MEMS and the FOL1 station. It can be noted as the displacements observed by MEMS and GPS at greater distance from the epicenter (about 50–55 km, Table 2) present similar displacement values (Figs. 8 and 9).

In order to quantify the similarity between the MEMS and GPS displacements we have estimated the correlation coefficients between the time series (Table 4). In particular, we have estimated the values analysing the displacements observed in the 30–40 s period after to the main shock event (01:36:32 UTC). It can be noted in fact in Figs. 8 and 9 as the first arrivals in the time series appears about 30 s after the main shock origin time.

Table 4Absolute Correlation coefficients estimated comparing the double-integrated MEMS data and GPS displacements filtered time series. The values refer to the time interval between 30 and 40 s after the origin time of the main shock (01:36:32 UTC).

Download Print Version | Download XLSX

The average absolute values of the correlation coefficients is 0.5, if we consider the 30–40 s period after the main shock in radial and transverse components (Fig. 9). This indicates a quite agreement between the waveforms compared, in particular between the components of the CGPS station and the MEMS sensors located at a relative distance lower than 3 km (Table 2). The worst values are observed in the comparison between FOL1 GPS station and the B00393 MEMS site, probably due to the relatively weak signals observed in this MEMS site (Figs. 8 and 9). B00393 is located at a distance of about 56 Km from the epicenter (Table 2), and the noise in the time series could be comparable with the amplitude of the seismic signals.

GNSS DATA are available at “High-Rate GPS data archive of the 2016 Central Italy seismic sequence” Ring-INGV repository (ftp://gpsfree.gm.ingv.it/amatrice2016/hrgps/data/, last access: 1 August 2019). MEMS data are property of the company owner of the prototype network and are not public available. Private request can be sent to AD.EL srl, Martellago, Venezia , Italy.

The supplement related to this article is available online at: https://doi.org/10.5194/adgeo-51-1-2019-supplement.

NC extracted and processed the GNSS data. He has also collaborated to comparison between the series. JB, FC and MRV processed the MEMS data. GdM furnished the MEMS prototype characteristics and data. GC evaluated the comparison and wrote the manuscript.

The authors declare that they have no conflict of interest.

This article is part of the special issue “Improving seismic networks performances: from site selection to data integration (EGU2019 SM5.2 session)”. It is a result of the EGU General Assembly 2019, Vienna, Austria, 7–12 April 2019.

This paper was edited by Damiano Pesaresi and reviewed by two anonymous referees.