We used continuous three-component seismic data recorded at 71 permanent broadband stations in the Eastern and Southern Alps from the Seismic Network of Austria , National Seismic Network of Switzerland , Italian Seismic Network , Province Südtirol , German Regional Seismic Network , BayernNetz, Germany , Slovenian Seismic Network , Hungarian National Seismological Network , and Slovak National Network of Seismic Station . In order to improve ray coverage, we completed our dataset with 19 temporary broadband stations of the AlpArray-EASI project . The Eastern Alpine Seismic Investigation (EASI) is a collaborative seismological project that was the first AlpArray collaborative experiment between the Swiss Federal Institute of Technology, University of Vienna (Austria), and the Academy of Sciences of the Czech Republic; it ran between July 2014 and July 2015. In that project, seismic stations were deployed along a north–south profile, roughly along longitude 13.5∘ E, from the internal Bohemian Massif to the Adriatic Sea with inter-station distances between 10 and 15 km. Using the 19 EASI stations improved station coverage, especially in the central Eastern Alps where the station density is relatively low. After applying several selection criteria (see Sect. 2.3.) to the computed cross-correlation functions, 79 stations were selected for the tomography (see Fig. ).

The first step of any noise-based analysis requires preprocessing the continuous waveform data, which strongly affects the quality of cross-correlation functions, dispersion curves, and the resulting velocity maps. Because noise characteristics and station configurations differ for each study, no universal preprocessing methodology exists. The best methodology and the various processing steps have to be tested for each dataset and are usually evaluated with basic parameters such as the symmetry of the cross-correlation function, signal-to-noise ratio, and frequency bandwidth e.g.,. Here, we tested several classical preprocessing methods including windowing e.g.,, whitening e.g.,, and 1 bit normalization e.g.,. The processing scheme chosen maximizes the signal-to-noise ratio (SNR), defined here as the peak amplitude divided by the standard deviation of the noise. Our final preprocessing methodology follows and consists of the following steps: (1) removing the instrument responses, high-pass filtering at 125 s, and glitch correction by clipping the data at 15 standard deviations; (2) removal of the transient signals (e.g., earthquakes) by cutting the daily records into 2 h segments on which we perform an energy test: when the energy of a segment is greater than twice the standard deviation of the energy of the 24 h (daily) record, the 2 h segment is removed. (3) Ambient noise is not spectrally white, which may induce an amplitude bias in the resulting cross-correlations e.g.,. The noise spectrum is therefore normalized using a whitening function by dividing the amplitude by its absolute value between 1 and 100 s periods without changing the phase. (4) We also perform a second clipping step in order to ensure that all the energy from transient sources that were not previously deleted by the energy test, such as small earthquakes, is properly removed from the waveforms. This is done by clipping amplitudes larger than 4 standard deviations of the whitened records. (5) Finally, the data are down-sampled to 4 Hz to reduce computational costs.

After preprocessing of the waveform, we compute cross-correlation functions for all station pairs, which resulted in 4005 cross-correlations from 90 stations. The cross-correlation for each daily record and each station pair is computed over all possible combinations of three-component data, vertical (Z), north–south (N), and east–west (E). This yields nine inter-components, ZE, ZN, ZZ, EE, EN, EZ, NE, NN, and NZ, constituting the correlation tensor. The ambient noise in the microseism frequency band is dominated by surface waves ; using all nine of these inter-components of cross-correlation enables us to construct both Rayleigh and Love waves from the computed cross-correlation. The correlation tensor consists of nine inter-components: RR, RT, RZ, TR, TT, TZ, ZR, ZT, and ZZ. Rayleigh waves emerge from the RR, RZ, ZR, and ZZ, and Love waves emerge from the TT. The cross-terms (TR, RT, ZT, TZ) also carry some weak diffuse energy without any clear arrivals, which confirms the quality of the correlation component rotations. The correlation tensor representing the propagation of surface waves through all station pairs is shown in Supplement Fig. S1.

Cross-correlations are computed in the frequency domain and then returned to the time domain to be stacked over the 2 years of data in order to reduce the seasonal effect of the noise source in the cross-correlation. Rotation of the stacked cross-correlation was then performed based on the azimuth between the station pairs to obtain the cross-correlation on the radial (R), transverse (T), and vertical (Z) components, which form a nine-inter-component correlation tensor including RR, RT, RZ, TR, TT, TZ, ZR, ZT, and ZZ. In order to select the clearer cross-correlation, we picked those that have signal-to-noise ratio (SNR) larger than 4. The SNR is calculated as the maximum amplitude of cross-correlation divided by the standard deviation of a noise window. We found that cross-correlations with a low SNR can be associated primarily with certain stations, marked in grey in Fig. . We therefore removed those 11 stations, which resulted in 23 % less cross-correlation. Figure  shows examples of 2-year (2014–2015) stacked and rotated cross-correlations, with the corresponding station path shown in the top panel. Examples present long inter-station distances between CEY-DAVOUX and CONA-KNDS, as well as short station distances between FETA-ROSI and AAE26-AAE31. Only ZZ, ZR, RZ, RR, and TT are shown in the figure as Rayleigh and Love waves are extracted from these five inter-components .

The examples shown in Fig.  were chosen based on the station geometry and location with respect to the North Atlantic and the North Sea coastline, which should be the dominant noise source for our study region e.g.,. We observe that for pairs parallel to the North Sea coastline, the cross-correlations are slightly asymmetric, with larger amplitudes for propagation directions arriving from the north and northeast. This can be seen at CONA-KNDS (Fig. b) and AAE26-AAE31 (Fig. d), for which the surface-wave amplitude is larger on the causal side. When station pairs are nearly normal to the North Sea coastline, such as CEY-DAVOX (Fig. a) and FETA-ROSI (Fig. c), we obtained strongly asymmetric cross-correlation. The differences between causal and acausal parts in the coast-normal station pairs are larger than the coast-parallel cases. Propagation directions from the North Sea coastline (from the northwest) result in large amplitudes, as can be seen in the propagation direction from DAVOX to CEY and from FETA to ROSI (Fig. ). This supports the hypothesis that seismic ambient noise is mainly generated by the interaction of the ocean swells with the seafloor located north and northwest of Europe. It also suggests that less energy is coming from southerly directions with respect to the northerly directions.

We assessed the spatial averaging of the tomography inversion results in two ways.

The first is through the path density at each cell used for the inversion. Figure  presents the path density map at 8 and 16 s for both Rayleigh and Love waves (path density maps at larger periods are also presented in Supplement Fig. S3). The path coverage of the region is good for most of the cells, with a path density above 20. The path coverage reaches 120 paths per cell in the Southern Alps and in the Tauern Window, particularly for periods shorter than 15 s. At the edges of the study region the resolution decreases rapidly due to fewer stations and less ray coverage (see Fig. ).

Figures  and  show the group velocity maps at periods of 5, 10, 15, and 20 s. In general, both Rayleigh- (Fig. ) and Love-wave (Fig. ) group velocity maps present similar features and correlate well with surface geology, particularly at upper crustal depths. To assess the velocity pattern with respect to the geological units, we extracted their borders from the geological map of Austria Supplement Fig. S4 and the tectonic map of the Alps Fig. 1,. The borders are shown as dashed lines in Figs.  and  , representing the margins of the two dolomite units to the south (Southern Limestone Alps, SLA) and to the north (carbonates of the Northern Calcareous Alps, NCA), the crystalline core zone of the Alps (CZA) in between (Fig. ), and the Tauern Window (TW) of the Eastern Alps (yellow dashed lines in Figs.  and  ). The CZA is well-marked by the broad high-velocity anomalies extending from the west to the east of the region. The SLA and the NCA are also marked by low velocities. At 5 and 10 s periods, the eastern border of the CZA matches the group velocity contrast (no. 1 in Fig. a). The southern margin of the CZA, particularly its western part, is clearly fitted by the edge of the high-velocity anomaly in that area at 5 to 10 s (no. 2 in Fig. a).

A high-velocity zone is featured in the easternmost part of the CZA at 5, 10, and 15 s. This feature might be associated with the Koralpe–Wölz high-pressure nappe system, the area on both sides of the Lavant Valley transform fault, which consists of eclogite facies and has the age of the Alpine tectono-metamorphic event of 90–110 Ma . The late Cretaceous (Eoalpine) Ötztal–Bundschuh and Silvretta metamorphic basement nappes are also perfectly imaged by high-velocity zones (shown as OTZ on Fig. a and b). At periods greater than 10–12 s, the OTZ no longer appears on the group velocity map, which may indicate the depth extent of the OTZ. A high-velocity anomaly is observed in the western part of the TW, while its eastern part shows lower velocities.

Figure  shows Love-wave group velocity maps. As expected, the Love wave presents higher velocities than the Rayleigh wave, as noted by the difference between color scales in Figs.  and . Similar to the pattern of the Rayleigh-wave group velocity, the CZA is well-marked by the Love-wave high-velocity anomaly bounded by the two dolomite provinces (Fig. a) to the north and to the south. The TW is also marked by high-velocity anomalies at most of the periods. However, similar to Rayleigh waves, the western part of the TW shows higher velocity. There is a velocity increase to the west of the Po Basin at ∼ 11∘E (no. 3 in Fig. a), which becomes more pronounced at the 10 s period. This might be associated with the depth extent of the magmatic rocks under the southern Alpine sediments (Dolomites in the Trentino region, Italy). At the 15 s period, a pronounced high-velocity anomaly occurs in the easternmost part of the NCA (no. 4 in Fig. c). A similar anomaly appears in the velocity model of . The Molasse and Po Basin (Fig. ) can be clearly located on both Rayleigh- and Love-wave group velocity maps (Figs.  and 9). The OTZ metamorphic units are marked partly by the Love waves at the 5 s period (Fig. a). Further discussion of this feature will be provided in Sect. 6 when describing the shear-wave velocity model.

Similarly to the Rayleigh-wave group velocity maps, the Rayleigh-wave shear velocity (hereafter RVs) model (Fig. ) shows a large high-velocity zone corresponding at the surface to basement units of the Tauern Window and Austroalpine units just north of the Periadriatic fault. The high-velocity area is bounded by two lower-velocity zones at all depth slices, which are associated with surface exposures of Mesozoic carbonates in the nappes of the Northern Calcareous Alps (NCA) and the Southern Alps (SA). The NCA corresponds to low shear velocities (< 3.1 kms-1) down to 10 km of depth (Fig. ). To the south, the velocity model clearly separates the SLA from the CZA. At 4 to 10 km of depth, the Dolomites of the Southern Alps show velocities of < 3.1 kms-1. The northern margin of the SA (dashed line in Fig. a, b, c) clearly matches the boundary between high and low velocities. The same pattern is observed in the Love-wave shear velocity (hereafter LVs) model (Fig. ).

A low-velocity anomaly (see the position labeled I1 in Figs. a and  a) is found under the Molasse Basin, with velocities for basinal sediments ranging from 2.4 to 2.9 kms-1. At depths of > 4 km, this low-velocity zone extends beneath the NCA in accordance with the occurrence of a south-dipping northern Alpine thrust fault that emplaced Mesozoic nappes of the NCA onto Neogene Molasse sediments e.g.,. The anomaly appears to extend down to 10 km of depth, indicating that the Neogene basinal sediments in the footwall of this fault form a wedge some 8–9 km thick . Alternatively, the anomaly extends no deeper than 8 km; the deeper part of this anomaly may be produced by downward smearing in connection with the prominent low velocity of the Molasse Basin. In the LVs model, the dominant low-velocity zones associated with the Molasse Basin (Fig. ) extend down to about 8 km.

Another low-velocity anomaly in the northwestern part of the study area (labeled I2 in Figs. a and  a) is more pronounced in the LVs model (Fig. ). At shallow depths, it could be related to the Swiss Molasse Basin. Note that this anomaly is located at the edge of the study area where the ray coverage is relatively poor, particularly at longer periods. This leads to a decrease in the lateral resolution and the related Vs values. In addition, the low-velocity zone extends down to 20 to 22 km of depth, which may indicate a smearing effect with low velocity in the lower crust not necessarily reflecting the structures. A clear velocity contrast, labeled I3 in Figs. a and 12a, marks the eastern boarder of the CZA and the transition to the Styrian basin. However, this is also located at the edged of the study area, and care should be taken in interpreting this anomaly.

A low-velocity anomaly in the Southern Alps, labeled II in Fig. a and 12a, with values of less than 3 kms-1 can be seen down to 10 km of depth beneath the Po Basin in northern Italy. This basin contains several kilometers of Mio-Pliocene clastic sediments derived from the retro-wedge of the Alps and the pro-wedge of the northern Apennines . The Po Basin is also easily identified in the LVs model (Fig. ) in which the velocity values and depth of the low-velocity zone are more or less as they are in the RVs model. Even though the low-velocity anomalies I2 and II match the location of the Swiss Molasse Basin and the Po Basin well, we should consider the fact that the anomalies are located at the edge of our study region with a lower ray-path density. This might affect the velocity values and the depth extent of the anomalies.

At the eastern part of the NCA, a small high-velocity anomaly (anomaly III; Figs. a and  a) is observed at 4 and 6 km of depth in the RVs model. It shows up along the southern margin of the NCA and might be associated with the eastern greywacke zone, consisting primarily of Paleozoic low-grade metamorphic rocks. In the greywacke zone the model exhibits velocities of more than 3.5 kms-1. Such high velocities are no longer visible at 10 km of depth. Similar high velocities have already been observed by between the SEMP fault and the northern Alpine thrust fault as well as towards the Bohemian Massif. They interpreted that feature as a southeastern tip of the Bohemian Massif dipping under the Alps. In the LVs model, this high-velocity zone is observed in a wider area. It can be traced down to 20 km of depth as the anomaly becomes smaller. Anomaly III extends south at 10 and 14 km of depth in RVs and becomes more prominent in LVs.

One of the most notable features of the shear-wave velocity model (anomaly IV; Figs. b and  b) is the higher velocity in the western part of the Tauern Window compared to the eastern part. The velocity contrast in the TW is more visible in the LVs model (Fig. a, b, c). This anomaly is found at depths down to about 10 km in both RVs and LVs models. The TW exposes both Penninic basement and underlying Subpenninic units , with the latter containing high-grade basement (Central Gneiss) of the Venediger complex exposed in two domes at the W and E ends of the TW Fig. S1. The western Tauern dome clearly corresponds to high velocities in our model. In contrast, the eastern Tauern dome exhibits lower velocities. The Central Gneiss is subdivided geologically into subunits such as Granatspitz in the west and the Hochalm nappes in the east e.g.,. However, it is not yet clear how such different lithologies could produce different shear-wave velocity in the TW region.

The shear velocity models clearly illustrate the Silvretta and Ötztal–Bundschuh nappes (OTZ) to the west of the Tauern Window (anomaly V; Figs. a and  a). These nappes are bounded by the Giudicarie fault and the Engadin Window at its eastern and western edges, as well as by the Inntal and Periadriatic faults at its northern and southern margins (see Fig.  for fault locations). Rocks of the Ötztal nappe underwent polymetamorphic metamorphism and deformation in Variscan and pre-Variscan time e.g.,. High shear velocity observed in this area might be associated with the Eoalpine high-grade metamorphic nappes, as seen in both the Rayleigh- and Love-wave shear velocity models. The high velocity related to the OTZ is observed down to 6 km of depth (Fig. b).

From depths 14–18 km downwards, we observe a clear separation between the high velocity under the TW and the low velocity beneath the OTZ. The high-velocity zone is cut by the Giudicarie fault, which sinistrally offsets the Periadriatic fault system (Fig. 11e). The Giudicarie fault can be traced down to 40 km, confirming that it affects almost the entire crust e.g.,and references therein. However, the crust–mantle discontinuity (Moho) does not appear to be offset beneath the Giudicarie fault , suggesting that this fault is a crustal feature that does not penetrate down to the mantle lithosphere.

Towards the south, the Periadriatic fault system (PAL) separates the Austroalpine nappes to the north from the Southern Alps, including the Dolomites, to the south . In the upper crust down to a depth of 14 km, most of the high-velocity anomalies lie to the north of this fault. The PAL does not seem to separate units with different velocity structures. This may indicate that units with similar physical properties are located on either side of the fault.

The low velocity associated with the Neogene sediments of the Po Basin is quite clear at 4 and 6 km of depth. However, at 10 km and deeper, the velocity reduction appears to deepen northward and extend to the lower crust (marked VI in Fig. h). Since this feature is not located exactly under the Po Basin, it is unlikely that we have a smearing effect due to the dominant low velocity of the Po Basin. Such a smearing effect can also be ruled out because it cannot be observed at such depths away from the Po Basin. Beneath the Po Basin, a high-velocity zone at >22 km of depth can be seen in both the RVs model and the LVs model (marked VII in Figs. h and h). The margins of this high-velocity domain are marked with red lines in Fig. . They may show the boundary between intermediate and lower crust. More particularly, the southern margin might indicate the intermediate–lower crust boundary within the transition from thinned Dinaric crust to the Pannonian basin.

In Fig. , we show depth slices at 30 and 40 km of our RVs and LVs models in comparison to the Vs model derived from surface-wave phase velocity using a combination of ambient noise and earthquake data , as well as a Rayleigh-wave Vs model derived from ambient noise data . The anomaly VII observed at 30 and 40 km is marked by the red lines. It can also be observed at 30 and 40 km in , Fig. e and g. To the south of the PAL, the pattern of velocity changes in our model (Fig. a, b, c, d) and the Kästle model (Fig. e, g) are more or less similar; however, they do not show a similar pattern to the north of the PAL. Note that the Vs model of has been derived jointly from Rayleigh and Love waves, while that presented here has separate Rayleigh-wave Vs and Love-wave Vs models. This could explain some discrepancies in the pattern of the anomalies between our models and to the north of the PAL. Differences in station geometry could also be considered regarding some discrepancies between our model and . The anomaly VII can somehow be seen in in the 40 km depth slice (Fig. h). The relatively low-velocity anomaly VI, between the PAL and the anomaly VII (Fig. h), can also be observed in the Lu model (Fig. f, h). North of the PAL, our RVs model shows a low-velocity area at 30 and 40 km at the eastern part of the region. This can somewhat be seen in at 30 km. In addition, we see a clear high-velocity anomaly under the Tauern Window at 40 km of depth, while in we observe a broad low-velocity area at 40 km of depth. It seems that to the north of PAL where we have complex structures due to the interplay between orogen-normal shortening and orogen-parallel motion e.g.,, our models better resolve small-scale velocity contrasts and features.

Figure  shows cross sections of the RVs model presented in Fig. . Profile AA’ crosses the Austroalpine (Silvretta and Ötztal–Bundschuh) nappe system and the Giudicarie fault. The high velocity associated with the Silvretta and Ötztal–Bundschuh basement units is observed down to about 6 km of depth, and then a layer of lower-velocity anomaly can also be seen underneath this high-velocity anomaly. Profile CC' (Fig. ) crosses the easternmost part of the TW. This profile at longitude 13.3∘ E is roughly parallel to the EASI profile . South of the Periadriatic fault in the Southern Alps, a velocity change at about 20–25 km of depth overlies the Moho depth at > 40 km . To the north of the TW and under the Molasse Basin, the velocity change is also observed, which may correspond to the boundary between the upper granitic and the lower mafic crust. Profile DD' (Fig. ) crosses from the NCA across the Mur–Mürz fault to the Styrian basin. The high velocity in the vicinity of the Mur–Mürz fault in the uppermost 6 km seems to not be connected to any structure to the north. We also find a positive velocity change at about 10–15 km of depth along the southern part of the profile. The crust–mantle boundary (Moho) beneath this area is shallow and becomes shallower to the east, toward the Pannonian basin . Previously, a crustal thinning was proposed for this area and previous studies. also suggested a Moho jump of about 10 km under this area, which results in a Moho depth of 30 km. Such a sharp positive velocity change can be seen at 25–35 km of depth on the DD' profile, but it is not clear enough to be discussed with regards to a sharp change in the Moho depth along the profile. Profile EE' (Fig. ) crosses the length of the Tauern Window from E to W. The high-velocity zone beneath the TW is probably associated with the European basement and can be tracked at depth eastward along the profile. A lower-velocity zone under part of the TW is also visible in the uppermost 15 km of depth on this profile.

Figure  shows profile BB'. It is oriented N–S, coincident with the TRANSALP profile TRANSALP Working Group;. Here the basement under the Tauern Window can be imaged as a high-velocity anomaly. A relatively low-velocity anomaly is observed beneath the Periadriatic fault zone at < 15 km of depth. A geological interpretation of the TRANSALP profile (; ) is also shown in Fig. . The pattern of the high-velocity zone associated with the sub-Tauern basement (marked with red dashed line) is in good agreement with the geological interpretation. A clear velocity contrast is observed at 20 km of depth under the Inntal fault and to the north. This could reflect the contrast between European upper crust and the lower crust. To the south of the Periadriatic fault, there is a south-dipping velocity contrast reaching 20 km of depth that cuts across the nappe contacts in the Southern Alps (Fig. ). We speculate that this may correspond to a post-nappe metamorphic front that is below the current erosional level.

The shear velocity models extracted from Rayleigh and Love waves correlate well with most crustal geological and tectonic units down to 20 km. However, we observe some inconsistencies; for example, in the vicinity of the Mur–Mürz fault the LVs model shows higher shear velocity than the RVs model at most of the depths (Figs. , ). Since we typically have very good ray coverage for both Rayleigh and Love waves at lower periods (meaning shallower depths) and therefore greater resolution, the difference between the Rayleigh and Love waves, especially in the uppermost 20 km of crust, is noteworthy. We have also assessed the possible effect of inversion parameters on the inverted velocity values and found that it has no significant effect on the inverted Rayleigh- and Love-wave velocities, more specifically on the Love–Rayleigh velocity ratio.

As Love and Rayleigh waves are sensitive to shear displacement in different orientations (horizontal versus vertical), different velocity anomalies between Rayleigh and Love, particularly high velocities of Love waves, may indicate seismic anisotropy. The observed velocity difference may be attributed to the preferred alignment of the main schistosity and shear zones subparallel to the shearing plane of the Mur–Mürz fault. The velocity difference between the western and the eastern parts of the Tauern Window (anomaly IV in Figs.  and ) might be related to the difference in the orientations of the anisotropy. Structural data from deeply exhumed Penninic and Subpenninic units indicate that the main schistosity strikes NE–SW and is subvertical in upright, post-nappe folds of the west, whereas it is variably oriented to subhorizontal in folds in the east TW .

We also found a striking velocity difference in the RVs and LVs models to the east of the Tauern Window at > 20 km of depth (Fig. ). This occurs in the eastwardly extruded block of orogenic crust (Alcapa) bounded by the aforementioned sinistral SEMP and dextral Periadriatic faults. Eastward, orogen-parallel escape of the Alcapa in Miocene time is attributed to a combination of indentation of the Adriatic Plate and pull in the upper plate of the retreating Carpathian orogen e.g.,and references therein. The velocity contrast at the western margin of the high-velocity domain (dark green line in Fig. a) is quite discordant to the trend of Moho depth contours and might represent the boundary between thinned intermediate and lower crust of Alcapa. We interpret this velocity contrast as a possible zone of intracrustal decoupling at the base of the laterally eastward-extruded Alcapa unit. The observed velocity difference between the Love-wave Vs model and Rayleigh-wave model may reflect shear-induced anisotropy originating from eastward motion of the Alcapa block above the subducting lithosphere during Miocene Adria–Europe convergence. Such high velocity is not seen in the Vs model, possibly due to the fact that their model is isotropic and jointly inverted from Rayleigh and Love dispersions. This inversion may have averaged any anisotropic effects due to contrasting Rayleigh–Love Vs differences. We do observe a slightly low-velocity anomaly to the east of the TW in the Rayleigh-wave model of . Since they only presented a Rayleigh-wave Vs model, it is not possible to discuss and compare the anisotropy effect on the RVs and LVs difference from .

However, here we have separately inverted Rayleigh and Love dispersions, which present well-resolved isotropic shear velocity models with observed Rayleigh and Love velocity inconsistency in some parts of the region. It may highlight a notable anisotropy signal, but we would need a joint inversion of Rayleigh and Love dispersions in order to build an anisotropic velocity model to confirm this observation.