The AniMaLS seismic network had been deployed between October 2017 and January 2018 and was operated for a period of about 2 years, until October 2019. Two institutions contributed to the temporary seismic network – the Institute of Geophysics, Polish Academy of Sciences (IG PAS) provided 10 Güralp CMG-6T (30 s corner period) seismometers with Güralp DM24S3EAM data acquisition units and one CMG-6TD 30s seismometer and data acquisition unit. The Institute of Geophysics of the University of Warsaw (IG UW) supplied 12 Reftek-130B data acquisition systems with broadband seismometers Reftek 151–120 “Observer” with bandwidth of 0.0083–50 Hz (120–0.02 s). Additionally, for observations of local seismicity, IG PAS deployed six units with short-period (1 s corner period) Mark L-4C sensors. All stations had a 130 dB dynamic range and used 100 Hz sampling frequency. Timing was provided by GPS receivers. The average inter-station distance in the array was about 25–30 km.

As several permanent seismic stations were operated in the study area, it was possible to enlarge the dataset with recordings of stations: KSP (Polish Seismological Network) and CHVC, DPC, KRLC, MORC, OKC, OSTC and UPC (Czech Regional Seismic Network), all equipped with 120 s sensors. The short-period (1–5 s sensors) LUMINEOS network, designed by IG PAS for monitoring of the induced seismicity in Legnica-Głogów Copper District (LGCD) (Mirek and Rudziński, 2017), is also in the area we are investigating. The distribution of the AniMaLS stations and permanent seismic stations used in the experiment is shown in detail in Fig. 2. The coordinates of the stations, location names and technical details are summarized in Table 1.

Usually, the sites for permanent broadband seismic stations are carefully selected in areas with extremely low noise. The sensors are located in vaults designed to minimize noise resulting from thermal and atmospheric variations. Such a careful site preparation and installation is often not possible in the case of temporary seismic projects, where selection of the location, installation and formal issues (permissions, rental contracts) have to be done in a short time and with limited resources. Additionally, to form a more or less uniform network, the sites should be located at similar inter-station distances, which is another constraint for the site location. When deploying the array, we attempted to obtain a compromise between several factors: low seismic noise, site availability, continuous power supply and high signal level of the mobile telecommunications network (UMTS/LTE). An important issue was a high level of security, in order to avoid the damage or loss of the equipment. Meeting all these requirements was not straightforward, since in most of the locations the level of anthropogenic noise was elevated due to high population density and industrial activities. In these areas, fulfilling both constraints (station spacing and low noise) has been extremely hard. When possible, we placed the units at the unused basements of buildings, in the outbuildings or in rarely used public utility buildings. The sensors were placed on a hard surface – concrete or tiled floor, and in some cases a 5 cm thick granite slab was used for this.

At the sites, a thermal insulation of the sensor was ensured in the form of a styrofoam box covering the sensor. A few pictures from installation of a typical station are shown in Fig. 3. Each station was powered by a power grid system. The 12 V power supply was buffered with 40–60 Ah batteries in order to ensure continuous operation of the units in the case of the power outages. Near-real-time data transfer was done with the use of UMTS/LTE mobile network connection.

A precise orientation of seismic sensor axes with respect to geographical north direction is of great importance during installation of a three-component seismic station. Incorrect orientation of the seismometer can result in substantial errors when using three-component methods of interpretation, e.g. in the case of shear-wave-splitting analysis (Ekström and Busby, 2008; Vecsey et al., 2014; Wang et al., 2016). The simplest method of geographical north determination, using a magnetic compass, often results in uncertainty exceeding 5∘ (Vecsey et al., 2017), which is not satisfactory for some interpretation methods. The modern approach, involving the use of an optical gyrocompass, allows for much higher precision but requires expensive equipment.

Taking these limitations into account, we have designed our own low-cost system for precise orientation of the seismometers deployed in the project. For the determination of the geographical north direction in the field, we used a global navigation satellite system (GNSS) unit with a real-time kinematic (RTK) positioning technology unit and ASG-EUPOS network (Ryczywolski et al., 2008) for receiving location corrections. Two ways for transferring the north direction to the seismometer location were considered. The first method was geodetic tacheometry. However, this method is not only time-consuming but also causes problems in less accessible locations such as basements. To solve this problem, we developed a simple device for azimuth transfer which makes the process more time efficient while retaining satisfactory precision.

The core of the device is a MEMS (micro-electro-mechanical system) triple-axis accelerometer and gyroscope unit MPU-6050, controlled by a single-board Raspberry Pi microcomputer. The data communication between device modules is based on I2C serial protocol over the general-purpose input/output (GPIO) ports. The code for processing the data from the gyroscope unit was written as a Python script. Raw data from the unit are converted into stable values of the rotation angle of the device. The problem of the gyroscope drift was solved by calibration of the immobile device prior to the measurement phase. During calibration the drift is evaluated, and, based on this, corrections for drift are continuously applied during the measurement.

Orienting the seismometer towards the geographic north with this method is done in two stages. First, GNSS RTK unit is used to obtain precise positions of two points of a baseline outside of the station site and to calculate the azimuth of the baseline as a reference. Next, the azimuth is transferred to the place where the seismometer will be installed. To this end, the gyroscope device is aligned parallel to the baseline using laser pointer, and a GNSS-measured reference azimuth value is used as input to the device. The device is then moved indoor to the station site where it is rotated to the north, according to displayed current azimuth, and the N–S line is marked on the floor at the location of the sensor. Finally, to check if the device readings were stable during the north measurement at the site, the device is moved back to the baseline and oriented along it, where, ideally, reference azimuth value should be again displayed. If the value differs substantially from the reference, it indicates excessive/variable drift or other errors, and the measurement is considered to be invalid. The procedure is repeated until two to three stable (negligible drift) and consistent measurements are obtained. Assuming availability of the GNSS RTK unit, this method is an affordable solution which allows for orientation of the sensor with the error determined in field tests to be at the level of ±2∘. Additionally, a data-based verification of the orientations was done with use of polarization analysis. The results are discussed in Sect. 3.2.

The seismic data were written in the internal storage of the data acquisition units (Güralps – 16 GB flash memory, Reftek 2 × 16 GB CF cards) and, simultaneously, they were transmitted in near-real time to dedicated acquisition servers at the IG PAS. Additionally, state of health (SOH) information including temperature, voltage and mass positions were transmitted. The data transmission was done using UMTS internet connection, with all devices running an IPsec VPN system to securely connect all the stations to the data acquisition server and to protect the system from unauthorized access. The Güralp units were connected to the network using Mikrotik routers with LTE modems. The data transfer to a dedicated CMG-NAM data hub was based on GDI protocol with a back-fill buffer, which allows for handling temporary loss of internet connection and retransmission of missing data packets after the connection is re-established. Connection loss and router/modem hang-up situations were handled by a data acquisition unit by an implemented software watchdog, which allowed for three levels of action: (1) soft reset of the modem, (2) power cycling of the modem and (3) power cycling of the unit and of the modem. For data transmission from Reftek units, a modified system designed at IG UW (Polkowski, 2016), based on Raspberry Pi Linux microcomputers with UMTS or LTE wireless modems was used. The Raspberry Pi units served both as routers and as devices scheduling the data transmission – collecting data from acquisition units and sending them to server (FTP, rsync and SSH protocols). The control scripts (PHP, bash) were designed to check for gaps in transferred data (due to, e.g. network connection loss, server or device hang-up) and to schedule data retransmission, if necessary. Hardware watchdog devices, designed at IG UW (Polkowski, 2016), were used to assure automatic restart of the transmitting unit on no connection or hang-up.

Both the Güralp and Reftek stations were remotely controlled and monitored using their proprietary software providing a WWW control interface. It allowed for checking the status of the individual units, mass positions, timing, voltages, temperature, as well as setting various recording parameters. For Reftek units, the control interface allowed for monitoring of mass positions and mass centring. Also, automatic mass centring could be triggered if mass voltages exceeded a threshold after a user-defined time.

The near-real-time data transfer has well-known advantages – inspection of the current data flow and access to current SOH information is useful for monitoring of the data quality and allows for fast detection of failures, such as power supply malfunctions or timing problems. Also, an increase of the noise level or the signal distortion due to an inadvertent moving or tilting of the sensor can be detected, and necessary station maintenance can be planned. This saves the number of the field trips needed for servicing the stations and helps to quickly re-establish proper acquisition of the seismic data.

The gaps in transmitted data resulting from the lack of UMTS connection were filled by periodic retrieving of the recorded data directly from the data acquisition system memory during stations maintenance in the field, if needed. The transmitted and patched data were stored in miniSEED format. After unification of information in headers, the daily miniSEED files were finally stored in the form of SeisComP data structure (SDS) – a hierarchical structure with file and directory naming convention which allows for easy access to the data, e.g. with the ObsPy package.

The seismic studies require exact measurement of absolute time of the seismogram to be able to determine the arrival times of the analysed phases. An incorrect timing may lead to erroneous identification of the phases or incorrect travel-time determination. Currently, the seismic acquisition systems use GPS/GNSS receivers that allow for the synchronization of the internal clock with a high accuracy (±10 µs). However, in practice, technical malfunctions or loss of GNSS signal can introduce timing errors, and such problems should be recognized. If possible, incorrect timing should be corrected during initial data processing, or reported, to avoid using badly timed data for the interpretation. During the data acquisition for the project, an important problem with timing occurred for five Reftek acquisition units due to the “week number roll-over” (WNRO) issue in the GPS system in 2019, which affected the GPS receivers with older hardware that were not designed to cope with this issue. As a result, in July 2019, some of the stations started to report the date with wrong year (e.g. 2099) and incorrect day of year. However, the correct time of the day was preserved; therefore, it was easy to obtain the proper date by shifting the time by a fixed amount of full days. The corrected date/time was then written into miniSEED headers. Nevertheless, the wrong date caused malfunction of the online data transmission system, which expected a correct date in the transmitted file names and in the headers. The transmission system software had to be temporarily modified in order to avoid the problem. A permanent solution of the problem was later achieved by updating recorders' firmware with a patched, WNRO-aware version.

Other problem was detected at the AG23 station, equipped with CMG-6TD data logger: after a few weeks, the internal clock lost synchronization with GPS time, in spite of a properly working and locked GPS receiver. This resulted in a linear increase of the time difference, which reached ∼ 20 s after few months of recording. In this case, only an approximate time correction was possible. By comparing the timing of good-quality arrivals in seismograms from AG23 and neighbouring, correctly timed stations, it was possible to measure the time differences over the recording period and to apply appropriate corrections. Here, the accuracy of time determination after the correction was estimated to ∼ 1 s. This is a relatively large value, and it prevents such data from being used for modelling methods which require exact knowledge of the absolute time, as, e.g. seismic tomography. Nevertheless, such data can still be used in methods based on the relative time of the seismogram components, as receiver function method or shear-wave splitting. More precise determination of timing corrections for this station is planned with the use of a method based on the noise correlation between recordings from incorrectly timed station and neighbouring, correctly timed ones (Sens-Schönfelder, 2008).

The geology of the near-surface sequences varies considerably over the study area, ranging from Proterozoic crystalline rocks to unconsolidated Quaternary sequences. The geological structure of the basement at the observation site can heavily affect the character of the recorded seismograms; therefore, we summarize the differences in the near-surface lithology and discuss their possible influence on the seismic data. Table 1 presents locations, technical information (sensor type, operation time), lithology and stratigraphy at the site for temporary and permanent stations. Geological information is based on the Geological Map of Poland 1 : 500 000 (Państwowy Instytut Geologiczny – Państwowy Instytut Badawczy, 2021) and the Geological Map of Czech Republic 1 : 50 000 (Czech Geological Survey, 2021).

In the SW part of the study area (roughly corresponding to the Sudetes mountains), the observation sites are located directly on consolidated rocks of Palaeozoic or Proterozoic basement (except AG01 and OSTC, positioned on Cretaceous rocks). The stations in the NE (the less elevated region of Sudetic Foreland) are located on a layer of Cenozoic unconsolidated sediments, overlying the Palaeozoic basement. This area is marked in Fig. 3 with a dotted green line. The presence of the low-velocity Cenozoic deposits at these sites has a distinct influence on the seismic records, and a more detailed discussion of these effects is presented in Sect. 3.1.

To estimate the level of the ambient noise at various frequencies, we calculated the probabilistic power spectral density (PPSD) distributions (McNamara and Buland, 2004) for the data recorded at each station with the use of ObsPy package (Krischer et al., 2015). The PPSDs was calculated for continuous recordings from the period 1 December 2018–1 October 2019 (22 months).

The PPSD calculation was based on analysis of 1 h long windows of continuous seismic data (with 0.5 h overlap). The processing sequence consisted of demeaning, tapering, fast Fourier transform (FFT) computation and instrument response removal. The obtained frequency spectra for all windows were smoothed and summed to form a histogram representing the frequency distribution of noise amplitudes at various period ranges. The result shows which amplitudes are observed for a given period. The PPSD medians were also calculated.

Figure 11 shows a comparison of the PPSDs for three types of stations: AG10 (with 30 s CMG-6T sensor), AR06 (RT 151–120 s sensor) and permanent station UPC (STS-2 sensor). Diagrams for three components are presented. Figure 12 shows the PPSDs of Z component for 12 selected temporary and permanent stations used in this study (PPSDs for all stations are presented in Fig. S1 in the Supplement). Figure 13 shows a comparison of PPSD median curves for all sites used, including permanent and temporary stations. There is a systematic difference in the noise level between permanent and temporary sites. The difference is notable for long-period range (> 10 s) and is particularly large for the horizontal components. High amplitude of the noise for the long periods of the horizontal components is often experienced in the case of temporary stations, mainly due to an imperfect protection from environmental thermal/pressure changes or the sensor base tilt (Wilson et al., 2002). Another factor contributing to higher long-period amplitudes on the horizontal component with respect to vertical amplitudes, in particular for stations located on young/low velocity sediments, could be the ellipticity of the Rayleigh waves. In the presence of a low-velocity layer, the Rayleigh waves exhibit horizontally flattened particle motion, whereas at hard-rock sites on consolidated/crystalline basement, the particle motion is vertically elongated (Tanimoto et al., 2013). However, here, this factor seems to have a minor influence, considering relatively small (< 1 km) thickness of the low-velocity layer in this area, which should not affect the ellipticity of long-period (> 10 s) waves in question.

The highest amplitude of the long-period noise, often exceeding the new high noise model (NHNM) level, characterizes all sites with 30 s CMG-6T sensors. Similar behaviour of these sensors, independently of the actual noise at the site, was reported by Tilmann (2006). This is most likely due to high self-noise of this device type and, partially (for Tnoise>30 s), to a lower corner period of the instrument (30 s vs. 120 s for other units). It is worth noting that the CMG-6T high long-period noise is at a very similar level to that for the ocean-bottom seismometer (OBS) version of the Güralp CMG-40T (30 s) sensor (Stähler et al., 2018), while the land version of CMG-40T shows substantially lower (by about 20 dB) self-noise in this period range (Custódio et al., 2014; Tasič and Runovc, 2012).

The short-period (SP) parts of the all PPSD medians (Fig. 13) show amplitude differences independent of the station type and can be subdivided into two groups. Stations located on Palaeozoic, or older, consolidated basement (solid lines in Fig. 13) show much lower noise in this part of the spectrum than the stations on the basement covered by unconsolidated, alluvial Cenozoic sequences (dotted lines). This area represents NE part of the network, marked with a dotted green line in Fig. 3. The stations with high amplitude of the short-period noise, marked with light blue colour, mostly fit into this region, which suggests a high correlation of this effect with the basement type. When attempting to interpret these differences in terms of the near-surface geology, care must be taken, because the high-noise sites installed on the Quaternary cover are, in the same time, located in the area with higher population density, denser network of roads, expressways and railroads, with typically higher anthropogenic noise. To check if the anthropogenic effects are responsible for these differences in short-period noise level, two variants of the PPSD medians were calculated for the same time span: only for daytime hours – from 12:00 to 16:00 LT, and only for nighttime hours – from 00:00 to 04:00 LT. Comparison of results (Fig. 13c, d) shows that the SP noise level during the daytime generally exceeds the nighttime noise by 5–15 dB for all stations, irrespective of their location. In the same time, differences in the short-period noise level between the sites located on old Palaeozoic rocks and the sites on the young Quaternary cover are of similar amplitude (∼ 25 dB) for day- and nighttime PPSDs, suggesting that they are indeed related to the basement type at the sites.

The presence of the low-velocity sediments in the area of Sudetic Foreland is also related to another effect, affecting the character of the P-phase onsets. The P-wave pulses on the horizontal components are followed by a prominent, high-amplitude coda/reverberations, extending over up to several hundreds of seconds (Fig. 14). The coda is characterized by a narrow frequency range, with a central frequency of 0.25–0.40 Hz (periods of 2.5–4 s), depending on the station location. The corresponding P pulses on the vertical component are much shorter and seem to be only weakly affected (or not affected) by the coda. In contrast, for the stations located on the consolidated basement, such reverberations are not observed on any component (Fig. 14a).

Such phenomenon is well known for a long time and described by several authors, e.g. by Zelt and Ellis (1999) or Yu et al. (2015), as it may heavily distort the results of the three-component interpretation methods. A layer of low-velocity sediments, with a strong impedance contrast relative to the consolidated or crystalline basement, produces multiple P-to-S conversions and reflections between the free surface and the base of the sediments. This results in high-amplitude reverberations in a narrow frequency range, mostly visible on the horizontal components. The frequency of the multiples is directly related to the seismic velocity and the thickness of the low-velocity layer. A systematic determination of the properties of the near-surface layer is out of scope of this paper. However, these observations can be compared with studies of the northeastern part of the study area (LGCD), where the properties of the low-velocity layer were studied by Mendecki et al. (2016). They used the horizontal-to-vertical spectral ratio (HVSR) method to analyse the resonance frequencies and amplification factors based on the data collected by a broadband station in Tarnówek (Fig. 3), located ∼ 15 km to the east of AR20 station. The HVSR peaks at 3.6–4.2 s were found, and Vs of ∼ 0.4 km/s was estimated for a ∼ 380 m thick Cenozoic layer at this location. In our study, the Fig. 14b shows a shorter (3.3 s) main period of the coda for AR20 station, which most likely corresponds to the thinning of the sedimentary layer or higher S-wave velocity.

The reverberations related to a low-velocity layer pose significant problems for the interpretation of the data, e.g. with the receiver function (RF) technique, as they overprint Ps conversion pulses on the radial component. One of the methods to overcome this problem was presented by Yu et al. (2015). As the reverberations exhibit a resonant frequency related to the two-way travel time of the wave in the sediment layer, the approach is based on designing a resonance removal filter in the frequency domain with filter parameters derived from the properties of the autocorrelation of the calculated RF. Our first tests showed that such a filter, applied to the data from Sudetic Foreland, is quite effective and significantly reduces the effect of reverberations.

During the installation of stations in the field, to assure correct orientation, an azimuth measurement system with a GNSS RTK unit and a MEMS gyroscope was used, as described in Sect. 2.3. According to our estimates, such a system allows for determination of the north direction at the sensor location with ±2∘ accuracy, if appropriate care is taken by the operator during all steps of the procedure. In order to additionally check for possible misorientation of the sensors after deployment, using the acquired data, a method based on the analysis of the P-wave polarization described by Fontaine et al. (2009) was applied. These estimates were verified with the use of a method proposed by Braunmiller et al. (2020), based on the P-wave polarization, and with the approach of Doran and Laske (2017) based on polarization of the Rayleigh waves. The two latter methods are implemented in the OrientPy package (Audet, 2020).

For a correctly oriented sensor and a homogeneous, isotropic medium, the polarization of the P-wave and of the Rayleigh wave particle motion is expected to be confined to the ray plane, and its horizontal component to be polarized parallel to the event back azimuth. The misorientation of the seismometer (deviation of the north seismometer axis from the geographical North by A degrees – equivalent to rotation of the coordinate frame of the measurement system) will obviously result in an apparent deviation of polarization of the P wave from the ray direction by an angle -A, independently of the event back azimuth. However, in a real medium, this deviation can be superimposed by the effects of the heterogeneity (dipping velocity discontinuities) or anisotropy of the medium under the station (Crampin et al., 1982; Schulte-Pelkum et al., 2001; Fontaine et al., 2009). These effects show a specific azimuthal dependence of resulting deviation angles (periodic with 180 or 360∘ period); therefore, it is often possible to separate these factors, if data from a wide range of back azimuths are available. The total directional variability of the polarization deviation can be decomposed as (Schulte-Pelkum et al., 2001) 1Dpolα=A+Bsin2α+Ccos2α+Dsinα+Ecosα, where particular terms reflect the magnitude of various factors: A – the constant (azimuth-independent) component of polarization deviation, directly related to the incorrect sensor orientation; B and C – effect of anisotropy with horizontal symmetry axis; D and E – effect of anisotropy with inclined axis or effect of an inclined discontinuity; α – the event back azimuth.

For the analysis, from 165 events in the epicentral distance range of 5–100∘, the recordings with high signal-to-noise ratio (SNR) on the vertical component of the P phase (SNR > 5) were selected for each station. Selected data were filtered (various sub-bands of 2–16 s period band were used) and 3-D particle motion at the P onset was analysed with the use of the orthogonal distance regression (ODR) method implemented in the ObsPy package, providing the azimuthal angle of the motion in the horizontal plane and incidence angle. Also, rectilinearity as defined by Fontaine et al. (2009) was calculated and was used to reject arrivals with poor rectilinearity of the particle motion, as contaminated by noise or other effects, and likely to produce distorted results. The error of the azimuthal angle was determined based on calculated eigenvalues of the particle motion (Fontaine et al., 2009). In order to improve stability of final results, individual Dpol values were sorted into back-azimuthal bins of 30∘ width and averaged. Subsequently, these mean values were used for fitting the curve based on Eq. (1) and for calculation of A-E parameters. The constant parameter A corresponds to the sensor misorientation.

To verify the results, we also analysed the same dataset with a recently released software package OrientPy (Audet, 2020). The package implements two methods of determination of sensor orientation. The method described by Braunmiller et al. (2020) (BNG) determines the direction of P-wave polarization by minimizing the energy on the transverse component in a selected window around the P-wave onset (Wang et al., 2016). Subsequently, polarizations for all events are averaged. The averaged value represents the constant component of the azimuth-dependent deviations and is related to the misorientation angle for given station. It should be noted that the BNG method relies on relatively uniform back-azimuthal coverage of the analysed data – averaging of a non-uniformly sampled sinusoidal curve is likely to result in a biased estimate of the mean value. Obtaining the mean value A by fitting the function (Eq. 1) to the points as proposed by Fontaine et al. (2009) should produce a more reliable result if the azimuthal distribution of the data is highly inhomogeneous.

The Doran and Laske (2017) method (DL) is based on Rayleigh-wave polarization analysis. For each event, a search is done for an angle α (defined relatively to theoretical back azimuth) which maximizes the cross-correlation between the Hilbert transform of the vertical component and the radial component rotated by α. As in the BNG method, calculated individual deviations for all the analysed events are averaged to get a value of the misorientation of individual stations.

Figure 15 shows values of misorientations of all stations in the study area obtained with the use of three described methods; the permanent station GKP is outside the study area, but it is shown for comparison, as previous studies also reported its significant misorientation – Vecsey et al. (2014) reported 41∘, Wilde-Piórko et al. (2017) reported 39 and 45∘, and the result of this study is 34–37∘. For many stations, the results derived from the three methods are more or less consistent but with some conspicuous exceptions. It can result from a small amount of recordings used for analysis because of low SNR for several stations.

For some permanent stations of the Czech Regional Seismic Network (CHVC, DPC, KRLC, OKC, OSTC and UPC), the orientation angles obtained from direct, high-accuracy gyrocompass measurements in field were available (Luděk Vecsey, Institute of Geophysics of Czech Academy of Sciences, personal communication, 2020). They are presented as a reference in Fig. 15. For almost all these stations (except CHVC) our results are in a good agreement (in a ± 2–3∘ range) to the gyrocompass measurements.

It must be noted that the results of the indirect, polarization-based methods are not as precise as direct orientation measurements, e.g. with the optical gyrocompass. According to Rueda and Mezcua (2015), the Rayleigh wave polarization method achieves 1–5∘ uncertainty in the case for long time spans of observations, e.g. at permanent stations, while for shorter time intervals the uncertainty can exceed 10∘. Therefore, as pointed out by Vecsey et al. (2017), in the case of temporary arrays with limited period of data acquisition, the methods based on polarization analysis are able to detect only substantial (>∼ 10∘) misorientation of seismometers.

For most of the stations, the orientation values obtained from polarization analysis agree, within the error bounds, with the orientations measured directly at the sites with a GPS/gyroscope system, as can be seen in the Fig. 15 and in Table 2 (the estimated error bounds for both methods are ∼± 3–7∘ (largely) and ±2∘, respectively). Therefore, we assume that the orientation of these stations determined by GPS/gyroscope can be considered as correct (0∘ misorientation). However, for five other stations, the polarization analysis results differ significantly from the orientations measured at the sites – AR07, OSTC and GKP (absolute orientation values of ∼ 20–37∘), CHVC and OKC (∼ 9∘), suggesting that these sensors were incorrectly oriented during installation. The seismograms from these stations need to be rotated to a correct NE coordinate frame before use, and orientation codes in the headers of original (unrotated) data need to be set to Z, 1 and 2 instead of Z, N and E, according to the Standard for the Exchange of Earthquake Data (SEED) definition.