We image anisotropic structure of the upper mantle beneath the Northern Apennines based on body-wave data collected during the RETREAT experiment (2003–2006). Joint analysis of anisotropic parameters evaluated from independent data sets—teleseismic P-wave travel times and shear-wave splitting—allows us to identify regions of different fabrics both in the mantle lithosphere and in the sublithospheric mantle. We recognize three regions—the Tyrrhenian, Adriatic, and Transition—with their own anisotropic characteristics. The slab-parallel flow prevails in the sublithospheric mantle beneath the thin Tyrrhenian plate, while nearly slab-perpendicular orientation of the high velocities dominates on the Adriatic side of the region. The asthenospheric-flow pattern excludes a simple corner-flow model to fit the fabric of the upper mantle in the syn-convergent extensional tectonics and suggests the end of the subduction roll-back. We map fossil anisotropy with inclined symmetry axes within two domains of the thick continental Adriatic lithosphere. We estimate the lithosphere thickness of the Tyrrhenian and Adriatic plates at ~50 km and ~80 km, respectively, the latter being subducted down to no more than ~200 km with indications of inherited frozen-in anisotropic fabric. If a potential detachment at the northern end of the Apennine slab exists, then it would have to be narrow and in its initial stage. Synthetic tests of the well-known trade-off between isotropic heterogeneity and evaluated anisotropic parameters, along with combining independent data sets, document a sufficient separation of both effects.
 The north-west/south-east trending Apennine orogenic belt, located in the central Mediterranean region, is separated into two geodynamical different arcs—the Northern Apennine (NA) arc and the arc-shaped region beginning in the Southern Apennines, continuing to the Calabrian Arc and ending in Sicily (Figure 1). The Calabrian Arc exhibits characteristics of an active subduction zone, e.g., the volcanism in Aeolian Islands and seismicity down to several hundreds of kilometers, while none of these two features is present in the NA. In comparison with the southern part, there is only shallow seismicity possibly related to an eastward retreating subduction zone beneath the NA [e.g., Malinverno and Ryan, 1986; Chiarabba et al., 2005; Pondrelli et al., 2006]. Different geodynamic development of the two tectonic arcs of the Apennine belt is also evident in GPS measurements [e.g., Serpelloni et al., 2005]. While Sicily and southern Italy move northward at speed as much as 10 mm/yr, the velocities of the northern part of Italy are one order smaller or even statistically insignificant [Oldow et al., 2002].
 The Apennine mountain belt started developing in the Late Cretaceous due to slow collision of the African and European plates [Stampfli and Borel, 2002]. The collision led to westward subductions of the Adriatic and Ionian microplates beneath the European plate [Dewey et al., 1989; Doglioni, 1991; Rosenbaum and Lister, 2004; Vignaroli et al., 2008; Handy et al., 2010]. It is estimated that rates as large as 5 cm/yr occurred in the Apennine convergence zone in isolated periods during the last 30 My [Faccenna et al., 2003]. Evolution of the region was controlled by processes of syn-convergent extension related to a roll-back of the Adriatic slab and extension of the Tyrrhenian back-arc basin. As a consequence of these processes, an eastward retreat of the subducting slab started to form the geometry of the Apennine trench in the Late Miocene. The Apennine orogeny was thus moved to its present-day position, i.e., the north-west/south-east orientation. Serpelloni et al.  estimate the present-day convergent movement in the NA of less than 1 mm/yr. According to Wortel and Spakman , such low rates might indicate either that the subduction stopped there, or that some portions of the Apennine slab begun to detach from the surface plate. The end of subduction in the NA is also inferred from observations of quasi-Love (QL) waves from the great 2004 Sumatra-Andaman earthquake [Levin et al., 2007] and the recent teleseismic tomography of Benoit et al. .
 To bring new information on the lithosphere-asthenosphere structure beneath the NA and on tectonic development of the region, international multidisciplinary project RETREAT was organized between 2003 and 2006 [Margheriti et al., 2006]. The RETREAT project aimed at developing a self-consistent dynamic model of the syn-convergent-extension in the NA. Results of seismological investigations indicated significant lateral differences in the upper-mantle structure (for a review, see http://www.rci.rutgers.edu/~vlevin/vadims-research/RETREAT-research.html). Studies of seismic anisotropy that constitute a strong research component of the project clearly identified different structures of the upper mantle on the opposite sides of the Apennines [Plomerová et al., 2006; Salimbeni et al., 2008]. Measurements along the NA Profile passive experiment (NAP) show orogen-parallel polarizations of the fast split shear waves in the mountain crest, on average, that tend to change to orogen-normal further south-west in the Tyrrhenian zone [Margheriti et al., 1996]. Such results could imply a 2D orogen-parallel sublithospheric flow beneath the Adriatic plate due to the slab roll-back and orogen-normal flow beneath the Tyrrhenian plate associated with the extension in the back-arc basin [Park and Levin, 2002]. However, results of the shear-wave splitting from the RETREAT data in an area northward of the NAP profile indicate more complex structures beneath the NA [Plomerová et al., 2006; Salimbeni et al., 2008]. Geographical variations of polarization azimuths both in the Adriatic and Tyrrhenian plates along with a convincing dependence on back-azimuth do not permit the simple explanation of the observed anisotropy exclusively by the mantle flow. The authors relate the variations of shear-wave splitting also to a fossil anisotropy in the mantle lithosphere with generally inclined symmetry axes, or to a combination of both the effects.
 Though the splitting of teleseismic shear waves clearly evidences anisotropic structure of the upper mantle, complementary information on the upper-mantle fabrics can be exploited from variations of directional terms of relative P-wave travel-time residuals [e.g., Babuška and Plomerová, 1992]. P waves illuminate the upper-mantle volume at broader fans of incidences and azimuths than core-mantle refracted shear waves (SKS), which allows us to retrieve orientation of symmetry axes more reliably. Particularly anisotropy with symmetry axes of inclination between 30º and 60º is usually explicitly manifested in the so-called “bipolar” pattern of directional variations of P-wave travel-time residuals at a station [Babuška et al., 1993]. Due to their shorter wavelength in comparison with shear waves, the P waves also can detect boundaries of the mantle lithosphere domains with different fabrics in a greater detail than shear waves [Plomerová et al., 2011].
 Several tomographic models of velocity perturbations in the upper mantle beneath the NA image the steeply inclined Adriatic slab [e.g., Amato et al., 1993; Babuška and Plomerová, 1990; Lucente et al., 1999; Piromallo and Morelli, 2003; Spakman and Wortel, 2004; Cimini and Marchetti, 2006; Koulakov et al., 2009; Benoit et al., 2011] The slab represents a high-velocity subvertical heterogeneity which contributes to the directional variations of the P-wave travel-time residuals and thus might complicate the interpretation of regional changes in patterns of directional terms of relative P residuals. On the other hand, neglecting anisotropic propagation in isotropic tomography might cause misinterpretations of parts of the travel-time residuals generated by anisotropic structures. Therefore, we try to carefully consider the well-known effects related to the trade-off between anisotropy and velocity heterogeneities.
 This paper is a follow-up of Plomerová et al.  where the results of P-wave travel-time residuals from several regional arrays and from the national network in Italy [Plomerová, 1997] were summarized and where the shear-wave splitting parameters evaluated from data collected during the initial stage of the RETREAT experiment were analyzed. Our aim is to present a more detailed image of anisotropic structure of the lithosphere and the sublithospheric mantle beneath the NA based on an enhanced set of body-wave data collected during the RETREAT experiment. We jointly analyzed P-wave velocity variations and shear-wave splitting and discuss the effects caused by seismic anisotropy and heterogeneity. The joint interpretation of body-wave anisotropic parameters allows us to image individual domains of the mantle lithosphere characterized by their own fabrics generally oriented in 3D. We discuss our results in the context of the up-to-date published results on the crustal structure [Di Stefano et al., 2011], the isotropic tomography [Benoit et al., 2011], and other recent findings on the large-scale tectonics of the region [Levin et al., 2007; Salimbeni et al., 2008; Piccinini et al., 2010; Miller and Piana Agostinetti, 2012].
2 Data and Methods
 Permanent stations of the Italian National Network (INGV Roma) and a MedNet station (VLC) together with 10 temporary stations from the Institute of Geophysics of the Czech Academy of Sciences (IG Prague) formed a backbone of the RETREAT array during its three-year operation. Additional 25 stations of the IRIS PASSCAL Instrument Center were installed on a profile perpendicular to the NA mountain belt. In total, 50 stations (Table S1, Supporting Information) provided data for the passive seismic experiment oriented towards studying the upper-mantle structure [Plomerová et al., 2006; Margheriti et al., 2006].
 Propagation of both shear and longitudinal (P) waves are affected by anisotropic structures of the upper mantle. In the case of P waves, large-scale anisotropy in the upper mantle is well expressed in patterns of P-residual spheres showing azimuth-incidence angle-dependent terms of relative residuals at individual stations [e.g., Babuška et al., 1993]. Polarizations of the fast split-shear waves and delay times of the slow split waves measure orientation and strength of anisotropy. Joint interpretation of the shear-wave splitting parameters and directional terms of P-wave travel-time residuals allows us to infer 3D self-consistent anisotropic models of the upper mantle [Šílený and Plomerová, 1996].
 To analyze directional dependences of P-wave travel-time residuals at individual stations [e.g., Babuška et al., 1992], we measured arrival times of teleseismic P waves from epicentral distances between 20° and 100°. We picked the P-wave arrival times on recordings from the RETREAT event-oriented database sampled with 20 Hz. To gain as much as possible arrival-time measurements at highest accuracy (+/− 0.05 s) and to process huge amount of seismograms effectively, first we tested on a subset of seismograms arrival times measured manually with three semi-automatic pickers [Munzarová, 2011]. A picker developed by F.P. Lucente and D. Piccinini (personal communication within the RETREAT-experiment cooperation) and another one based on the adaptive-stacking method designed by Rawlinson and Kennett  were found unsatisfactory for the RETREAT data. Therefore, a new semi-automatic picking procedure based on Seismic Handler software by Stammler  complemented with several shell scripts [L. Vecsey, personal communication] was finally assessed to be the most precise and reliable of the tested pickers.
 High accuracy of the measurements was reached by a new two-step picking procedure, in which: (1) a selected part of a reference P-wave seismogram (Figure S3a, Supporting Information) is automatically cross-correlated with the waveforms at other stations in order to find time differences between them. Then, the traces are shifted to become aligned (Figure S3b, Supporting Information); (2) in an arbitrary seismogram, we select a distinct extreme which can be traced across all the waveforms and which is the closest to the first P-wave onset. Time of this extreme, called the relative P-wave arrival time, is then automatically picked in all seismograms of the event (Figure S3c, Supporting Information). We also measure the absolute arrival time for each event on at least one seismogram, where the first onset is clear. Then, we recalculate the absolute arrival times from the relative ones at all stations.
 We measured P-wave arrival times on recordings of 312 teleseismic events (see upper right inset of Figure 2 and Table S2, Supporting Information), which provided 7378 measurements at individual stations. Total number of measurements gives us also an estimate of approximately 150 arrival times per station on average. Station SFIR exhibits the highest number of 302 observations for a single station. We checked the time stability of the travel-time residuals computed according to radial reference Earth model IASP91 [Kennett, 1991] (the so-called absolute residuals) to prevent including incorrect data due to technical problems at some stations, e.g., due to temporary failure of time synchronization.
 As our study is focused on the upper-mantle structure, we corrected the P-wave travel times for effects originated in the crust with the use of Di Stefano et al.  crustal model. For stations located above the transition between the thin Tyrrhenian and thick Adriatic lithosphere [see also Bianchi et al., 2010], where a distinct step in Moho depths of about 20 km exists, we considered azimuths of arriving waves and applied appropriate corrections according to the two different crusts. Effects originated outside the volume studied, e.g., due to foci mislocations, source-region structures, or reflecting heterogeneities along deep-mantle paths, are minimized by calculating relative residuals. An event mean calculated from residuals of all stations which recorded the event was subtracted from individual measurements. The procedure allows us to separate the relative travel-time residuals into a station directional mean, i.e., an average of relative residuals at a station, and into directional terms, which represent azimuth-incidence angle-dependent components of the relative residuals [e.g., Babuška and Plomerová, 1992]. Data binning according to source regions is used to balance uneven foci distribution (see upper right inset of Figure 2). The directional terms at a station are visualized in the lower hemisphere stereographic projection as a function of azimuth, from 0º to 360º, and angle of propagation within the mantle lithosphere—from 15º and 45º measured from vertical, called “P spheres” (see lower right inset of Figure 2). Positive directional terms, i.e., delayed arrivals relative to the station directional mean, indicate low-velocity directions, while negative values, relatively early arrivals, signify high-velocity directions beneath a station.
 To better constrain the well-known trade-off between the effects of anisotropy and heterogeneities in P-wave propagation, we also analyze results of shear-wave splitting [Plomerová et al., 2006; Salimbeni et al., 2008], which is considered as indisputable evidence of anisotropy. Shear waves entering an anisotropic medium split into two quasi-shear waves, each propagating with a different velocity and with mutually perpendicular polarizations. Salimbeni et al.  studied splitting of the core-mantle refracted shear waves (SKS), i.e., the delay times between the fast and slow split waves and the polarization directions of the fast waves, for 27 teleseismic earthquakes recorded during the RETREAT experiment (see upper right inset of Figure 4). For each station-event pair, they determined the splitting parameters according to the method of minimizing energy on the transverse component [Vecsey et al., 2008], using computer code SPLITshear available at http://www.ig.cas.cz/en/personal-pages/ludek-vecsey/split . The method used is a 3D modification of procedure designed by Silver and Chan . The splitting parameters—the fast polarization directions ψ and split delay time δt—are sought in the LQT ray-parameter coordinate system in the modified version.
 We search for individual groups of stations characterized by similar patterns of the P spheres and/or exhibiting similar variations of the SKS splitting parameters with back-azimuth, which indicates regions with similar structure of the upper mantle. Sharp changes of the P-sphere patterns and the shear-wave splitting parameters mark boundaries of domains with consistent fossil anisotropy in the mantle lithosphere. Anisotropy related to a present-day flow in the sublithospheric upper mantle usually relates to large regions and changes are smooth.
 Stations with similar P-sphere pattern approximately delimit a lateral extent of the upper-mantle domains with consistent anisotropy. To estimate thickness of the mantle lithosphere, we employ, with some modifications, the method of evaluating the lithosphere-asthenosphere boundary (LAB) depths beneath the stations from static terms of relative P-wave travel-time residuals in central Europe [Babuška and Plomerová, 1992]. There, the static terms are calculated from steep waves approaching a station from all back-azimuth ranges. However, having in mind the Adriatic subduction, we modified the method and separated waves approaching the stations into two azimuthal segments (from −45° to 135°; and from 135° to 315°). The separation of the azimuthal segments is governed by a strike of ~135º of the NA mountain range. We model the LAB relief according to the empirical residual-depth relation with a gradient of 9.4 km/0.1 s [Babuška and Plomerová, 1992]. A reference level for the residual-depth estimates is set by linking the RETREAT relative residuals with the European and Italian data sets [Plomerová and Babuška, 2010]. We plot the LAB depths at piercing points of the rays and the LAB boundary.
3.1 Directional Dependence of P-wave Travel-time Residuals
 Areas with consistent P-sphere patterns at stations deployed in different tectonic settings indicate uniform anisotropic structures in a volume of the upper mantle beneath the stations [e.g., Babuška and Plomerová, 2006]. Stations with similar P-sphere patterns form groups in the NA region and divide it into three major parts—the Tyrrhenian and the Adriatic, and a Transition zone in between—and into six subregions altogether (Figure 2 and Figure S4, Supporting Information). In order to demonstrate characteristic P-sphere patterns, we show an average P sphere for each subregion (Figure 2) where each directional term is calculated as an average of respective directional terms from all stations in a subregion. P spheres of individual stations are shown in Figure S4 (Supporting Information) of the electronic supplement.
 Waves arriving to the stations in the Tyrrhenian region (TYR) from the south-west are delayed (positive terms, red diamonds in Figure 2) relatively to a directional mean at the station, while waves coming from the north-east arrive earlier (negative terms, blue diamonds). Such a type of P-sphere pattern—one half positive and the other negative—is called a “bipolar pattern” [e.g., Babuška and Plomerová, 1992].
 Stations east of the NA crest, in the Adriatic region (ADR), exhibit characteristic bipolar P pattern, but reversed in comparison with the TYR pattern (Figure 2 and Figure S4, Supporting Information). The ADR region can be divided into two subregions (ADR1 and ADR2), which slightly differ for waves propagating from the north-east—positive terms dominate in the ADR1 subregion for all angles of incidence, while negative directional terms (faster directions) occur at shallow angles in the ADR2.
 Approximately a half of the RETREAT stations were located in the NA mountain range, above the Tyrrhenian- and Adriatic-plate contact. Abrupt changes of the P patterns at these stations relative to the TYR and ADR regions delimit the Transition zone (TR). Additional changes of the P-sphere patterns in the TR allowed us to divide the zone into three subregions (TR1, TR2, and TR3 in Figure 2). The directional terms of waves arriving from south-west and north-east azimuths at shallow angles are positive at stations of the TR1 group thus indicating relatively low-velocity directions, while negative terms (relatively high-velocity directions) are evaluated for the north-west and south-east azimuths, regardless of incidences. The TR3 pattern tends to be a reversed TR1 pattern, and absolute values of the directional terms in both the TR1 and TR3 regions are lower than that in the small region TR2 (Figure 2 and Figure S4, Supporting Information) which contains only three stations. Their P patterns, with larger values, tend to the bipolar ADR1 pattern. The TR2 thus seems to form a separate region within the heterogeneous TR. In comparison with the consistent P patterns in the TYR and ADR regions, the P pattern above the transition of both plates is less consistent internally (see Figure S4, Supporting Information). However, we admit that the P pattern of the TR region reflects a complex structure of the contact of the Tyrrhenian and Adriatic crusts, and the applied crustal corrections can hardly eliminate all effects of the real structure of the crust.
 To show clearly lateral changes of the P-sphere patterns at densely spaced stations located along profile AB crossing the NA (see Figure 1), we plot a vertical cross section down to a depth of ~100 km (Figure 3). The P spheres represent each subregion along the profile. To what extent this overall P-velocity pattern reflects anisotropy of the lithosphere-asthenosphere system, or, is affected by the subducted slab or inaccuracies of the crustal model, will be discussed in Chapter 4.
3.2 Regional and Directional Variations of SKS Splitting Parameters
 As the shear-wave splitting is considered as an indisputable proof of anisotropy, we also analyze characteristic features of SKS splitting parameters evaluated by Salimbeni et al.  from the RETREAT station seismograms in combination with the P-wave travel-time residuals that form the core data in this paper. The SKS splitting parameters were evaluated by the method of minimizing energy on the transverse component [Silver and Chan, 1991; Vecsey et al., 2008]. We plot the fast shear-wave polarization azimuths and split delay times in piercing points at depth of 80 km (Figure 4) as main sources of the anisotropy are in the upper mantle, and we expect its lateral changes around this depth. We examine the geographical distribution of the splitting parameters and how much they vary with back-azimuth at individual stations. For that purpose, we plot the splitting parameters in different colors assigned according to their back-azimuth [see also Plomerová et al., 2006].
 Considering the fast shear-wave polarization azimuths, the whole area of the RETREAT experiment can be roughly divided into two main regions: the Tyrrhenian and the Adriatic (see the black dashed line in Figure 4). The TYR is characterized by highly consistent polarization azimuths regardless of back-azimuths of the steeply incident SKS waves. Polarizations in the ADR exhibit both a distinct dependence on back-azimuth and different regional characteristics, according to which Plomerová et al.  delimited three additional subregions in the Adriatic plate (dotted black curves in Figure 4). Clustering the stations with similar splitting parameters leads to regionalization of the NA area into regions which differ, at a first glance, from the six subregions identified in compliance with the P-residual analysis (Figure 2).
 To appraise variability of polarization azimuths at individual regions, we plot rose diagrams, which illustrate a frequency of occurrence of the fast shear-wave polarization azimuths in 18° wide fans. First, we plot the diagrams for the six subregions delimited according to the P-sphere patterns (Figure 5). To emphasize the azimuthal dependence of the splitting parameters, we consider the full azimuth range, i.e., the φ from 0° to 360° (retrieved from the polarization directions ψ in the LQT ray-parameter system) and plot rose diagrams of the fast shear-wave polarization azimuths for the south-western back-azimuths (from 135° to 315°) separately from those for the north-eastern back-azimuths (from −45° to 135°). Such separation of the polarization azimuth is related to the strike of the subduction zone beneath the NA (see also Figure 4). To be compatible with the commonly used presentations and to emphasize differences in the variations of the splitting azimuths, we also present the opposite azimuths in the rose diagrams (light colors). Behavior of the polarizations indicates a complex anisotropic structure of the upper mantle beneath the NA region (Figure 5). We observe both distinct regional variations of the polarization azimuths and different characteristics of the back-azimuth dependences in individual subregions.
 There is only a weak dependence on back-azimuth in the westernmost region (TYR). The fast S polarization azimuths are almost independent of the two back-azimuth intervals, (left and right columns in Figure 5). Further to the north-east, in the TR1 region, the rose diagram for the north-eastern back-azimuths splits into two dominant polarizations. One of them coincides with the dominant polarizations in the rose diagram for the south-western back-azimuths, which are slightly clockwise rotated relative to the TYR region. The TR2 region contains only few data, but with the same trend. The back-azimuth dependence of polarizations becomes significant in the diagrams of the TR3, ADR1, and ADR2 subregions, in which the distinct polarization azimuths for the waves coming from the north-east do not correspond to the most frequent polarization azimuths for the south-west (Figure 5). Moreover, moving from the Tyrrhenian coast through the TR to the Adriatic coast, a general clock-wise rotation of the polarization azimuths is observed (Figures 4 and 5).
 Particularly, the diffuse pattern of the fast S polarizations in the ADR1 for waves from the north-east led us to regroup the splitting parameters in accordance with the clustering introduced by Plomerová et al. . According to the geographical variations of the shear-wave splitting parameters, the authors identified the Tyrrhenian domain and three subregions in the ADR—the Southern domain of the NA (SD), the Ferrara arc domain (FER) in the central part of the region, and the Alps-Apennines Transition domain in the north characterized with null splitting (see limits of these regions in Figure 4). We regrouped the stations in the ADR1, ADR2, and a part of the TR3 regions similarly to that and searched whether the new rose diagrams differ distinctly for the FER and SD regions (Figure 6). Indeed, the shear-wave polarization azimuths differ, both regionally and in their back-azimuth dependences. Two components of the anisotropic signal are clearly evident, particularly in the SD domain. The north-south component of the apparent fast polarizations occurs in both regions (FER, SD) with only a weak back-azimuth dependence in FER (see Figure 6). On the other hand, the second component in the diagrams differs substantially in both regions. Polarizations at ~300° prevail in the SD domain, while the second component in the FER rose diagram is perpendicular (~210°) to that direction. Distinct back-azimuth variations of the polarizations in the ADR indicate non-negligible contributions reflecting anisotropy of individual domains comprising the continental Adriatic mantle lithosphere. The observed apparent upper-mantle anisotropy thus likely consists of an anisotropic signal from the mantle lithosphere superimposed on an anisotropic signal related to a sublithospheric mantle flow.
 With the use of the standard double-layer formula of Savage and Silver  for fitting back-azimuth variations of the apparent polarizations of SKS phases, we grid searched for a double-layer model by varying azimuths φ1,2 of horizontal fast symmetry axes a and time delays δt1,2 for the lower and upper layer, respectively. The main task was to fit the rose diagrams of the FER and SD regions for the polarization azimuths of waves propagating through the Adriatic plate domains (see Figure 6). We show results of the modeling for waves from the north-eastern back-azimuths (7), for which the propagations are not affected by structure of the subducting slab. The solution is not unique. The two components to the apparent anisotropic signal in the FER and SD subregions can be simulated best by about three four models of two anisotropic layers (Table 1 and Figure 7). The best solution for the FER region was obtained for the high velocities in azimuths of 80° and 20° and in the SD region in 60° and 0° for the lower and upper layer, respectively. Three of the four suitable models for the FER subregion require larger contributions from the lower layer than from the upper, which means either a thicker lower layer or a stronger anisotropy, or both. The preferable model for FER (Table 1, in bold) with δt2 = 0.6 s is compatible with ~5 % anisotropy of the mantle lithosphere and a thickness of ~50 km [Silver, 1996]. In the SD region, the synthetic polarizations are less scattered, but fitting the two components of the observed polarizations is more difficult. The north-western polarizations prevail in the observed diagrams, while the northward polarizations dominate in the synthetics calculated for the same set of waves.
Table 1. Parameters of Double-layer Models Fitting the Rose Diagrams of the FER and SD Subregions (See Figure 4 for their Location)
3.3 Comparison of Observed and Synthetic P-residual Spheres Generated for Isotropic Tomography to Evaluate Effects of Subducted Slab on the Directional Dependence of Relative P Residuals
 The NA lie in an active tectonic region of the subducted Adriatic plate and assumed retreating trench accompanied by horizontal flows in the sublithospheric mantle [e.g., Margheriti et al., 2003]. Both the upper-mantle anisotropy and velocity heterogeneities, manifested mainly by the subduction, affect propagation of seismic waves. Standard methods of imaging velocities or velocity perturbations in the upper mantle consider only isotropic propagations. It is generally accepted that velocity perturbations due to heterogeneities and due to anisotropy are comparable in their amplitudes, though difficult to separate. However, neglecting anisotropy can cause false or distorted artifacts in the isotropic tomography, e.g., wrong amplitudes of heterogeneities, or, false heterogeneities [Eken et al., 2012]. On the other hand, also the P-residual spheres, which bear information about anisotropy, might be affected by uncorrected heterogeneities.
 To estimate possible effects of the major velocity heterogeneity in the upper mantle beneath the NA on the directional terms in the P spheres, we computed synthetic travel times for P waves with event-station locations of our measurements propagating through the recent tomographic model of the subducted Adriatic slab by Benoit et al. . Comparison of the synthetic directional terms plotted in the P spheres (Figure 8 and Figure S5, Supporting Information) with the observed P spheres (Figure 2 and Figure S4, Supporting Information) reveals significant differences in the geographical distribution of the P-sphere patterns from both methods. The spheres differ both in the distribution pattern of the negative and positive terms, as well as in their absolute values.
 According to the distribution of the P-wave directional terms in the synthetic P spheres, the stations form only two distinct groups, each of them with its own “synthetic” P pattern (Figure 8 and Figure S5, Supporting Information). Stations clustered into the pattern Synthetic 1 are located in the TYR. With exception of station PIIR, the lateral extent of the Synthetic 1 subregion is the same as for the TYR subregion defined according to the “observed” P-sphere pattern. However, there are differences in characteristics of both types of the P-patterns. The “observed” pattern is clearly bipolar (Figure 2), while the “synthetic” pattern only weakly tends to that. Moreover, the “synthetic” P-pattern is less consistent than the observed one, and the absolute values of the synthetic terms gradually decrease with increasing distance south-westward from the boundary between the Synthetic 1 and Synthetic 2 regions, which is not the case of the “observed” P patterns in the TYR subregion (Figures S4 and S5, Supporting Information).
 Other stations form one large group—Synthetic 2 (Figures 8 and S5, Supporting Information) with a clear bipolar synthetic P pattern with negative terms from the south-western directions and positive terms from the north-eastern directions. This is, however, different from the observed P-sphere patterns, clustered into the five subregions (see TR1, TR2, TR3, ADR1, and ADR2 in Figure 2). The synthetic P-spheres thus do not allow any detailed delimitation of the TR and the ADR.
 To demonstrate quantitatively the differences between the synthetic and the observed P spheres, we have calculated average directional terms in six azimuth bins, for which we show characteristic examples (Figure 9). To avoid additional smoothing of the pattern, always one of the segment boundaries in each P sphere follows the azimuth separating the early and delayed arrivals.
 Some of the observed and synthetic P spheres, e.g., those at stations MURB and RAVR, seem to be similar, but the amplitudes of the directional terms are much larger in the synthetics (Figure 9a). On the other hand, at some stations, the observed and synthetic patterns are completely reversed (station PIIR in Figure 9b), or tend to be reversed (station in FOSR Figure 9b). A clear bipolar pattern from the observed data might also change into a quadruple pattern in the synthetics (see SACS and VOLR Figure 9c). Finally, a bipolar pattern turned into “no pattern” in some cases (Figure 9d). The “no pattern” is well reflected in the bin values of the directional terms close to zero for the synthetics of MAON and the observed pattern at CAIR, while the differences of the bin values attain ~1 s in the case of the observed bipolar pattern at MAON and the synthetics of CAIR.
 Schematic ray tracing with the use of TauP [Crotwell et al., 1999] from the epicentral distances of 40° and 80° to four RETREAT stations is imaged against a vertical cross section through the tomography (Figure 10; for location of the profile see Figure 8). Station VOLR is located just above the high-velocity heterogeneity imaged by tomography, which results in the quadruple pattern of the synthetic P sphere, while the “observed” pattern is clearly bipolar. Stations CSNR and SFIR further to the north-east show similar bipolar patterns in the synthetics, but observed and synthetic patterns for the CSNR station differ. Less data at station BARR located in the north-easternmost part of the profile resulted in similar patterns.
3.4 Lithosphere Thickness
 Figures 11 and 12 show the LAB depths estimated from the modified static terms calculated from the relative P-wave residuals recorded during the RETRAT experiment. The LAB relief is smooth and indicates a depth of the NA subduction. The LAB estimates contour the most prominent high-velocity perturbations in the tomography by Benoit et al.  along both profiles (Figures 11 and 12). The Adriatic slab probably sinks no more than to 200 km. The lithosphere of the over-riding Tyrrhenian plate is thinner (e.g., Figure 12). A thinning of the Adriatic plate occurs also towards the Po Plain (cf. the Adriatic LAB in Figures 11 and 12).
 The LAB models strongly depend on our knowledge of the crust structure. To correct for crustal effects, we used the Moho-depth model by Di Stefano et al.  and incorporated also corrections for sediments [Vuan et al., 2011]. Good knowledge of the Moho depth around the contact of the Tyrrhenian and Adriatic crusts is crucial. Our results, calculated with the model characterized by a simple step at the Moho, did not fit results from surrounding stations (Figure 11). Therefore, we modified the crustal model around the contact in analogy with the model of Finetti et al.  and the recent results from harmonic decomposition of receiver functions by Bianchi et al. . Instead of a simple step we considered an under-thrusting of the Adriatic crust beneath the Tyrrhenian plate. Under-thrusting of the ~16 km thick crustal material near the step in Moho depths deepens the anomalous values of the LAB by ~60 km and matches them with surrounding LAB depths.
 Seismic anisotropy provides key information for understanding the tectonic fabric of the whole lithosphere-asthenosphere system. Mapping structural changes through both geographical and directional variations of anisotropic parameters evaluated in 3D represents a powerful tool, especially in continental provinces, where relatively thick mantle lithosphere consists of domains with their own fossil structure [Babuška and Plomerová, 2006]. Moreover, in regions of colliding plates accompanied by a plate subduction, distinct anisotropy related to a present-day flow in the sublithospheric mantle can mask the lithospheric contribution to the observed apparent anisotropy, particularly if only azimuthal anisotropy is searched for. The NA result from intermittently independent motion of up to five microplates between a collision of the major plates of Eurasia and Africa [Handy et al., 2010]. At present, the collision is expressed by a south-westward subduction of the Adriatic plate, and the region can serve as a natural laboratory to test the upper-mantle structure, its formation, and evolution [e.g., Faccenna et al., 2001; Lucente et al., 2006].
4.1 Heterogeneity Versus Anisotropy
 Three main regions, the TYR, the TR, and the ADR, delimited according to body-wave anisotropy, form bands approximately parallel to the mountain range (see Figure 2). The bipolar P-sphere patterns in the TYR and ADR regions are mutually reversed, each with negative terms from the directions where the subducted slab is located. This feature naturally leads to a possibility that the observed bipolar signal might be significantly controlled by the high-velocity heterogeneity represented by the subvertical slab.
 The synthetic P spheres (see Figure 8) calculated for waves propagating through the isotropic velocity model of the upper mantle from the RETREAT data [Benoit et al., 2011] differ from the observed P spheres associated mainly with anisotropic structure of the mantle lithosphere (see Figures 9 and 10). Though an imprint of the high-velocity slab can remain in the P spheres even after the careful P-residual processing, the slab effect is weaker in comparison with effects of the plate fabrics at majority of stations (see Figure 8). Moreover, the shear-wave splitting exhibits similar geographical variations (see Figures 11 and 12) and thus indicates that also in this tectonically complex region, we are able to detect the P-wave anisotropy and relate it to the mantle lithosphere. Tiny differences in delimiting the regions of similar characteristics of anisotropic parameters determined from the SKS splitting and P residuals are restricted to the margins of individual domains (see Figures 4 and 11). We explain this “discrepancy” by differences in wavelengths and ray paths of both types of waves. P waves, having only about ¼ of the SKS wavelength, are more sensitive to lateral structural changes than the SKS waves.
 There are differences in tomographic models of the subducted Adriatic slab [e.g., Amato et al., 1993; Lucente et al., 1999; Piromallo and Morelli, 2003; Spakman and Wortel, 2004; Giacomuzzi et al., 2011; Benoit et al., 2011]. A question how exactly the position of the steep high-velocity slab is imaged by seismic tomography remains open, as well as the depth extent of the subduction. Most of the tomography studies detect high velocities down to about 200–400 km. However, Benoit et al.  admit ~100–150 km depth smearing of the steep heterogeneity. Nevertheless, extent of the positive perturbations, related to the subduction, is also in accord with our estimates of the depth of the slab bottom to be around 200 km (see Figures 11, 12, and 13). Piromallo and Morelli  included into their tomography also regional events, which broadened the fan of P-wave ray paths and resulted in modeling a less steep Adriatic subduction. However, neither the broadening, nor the deepening of the high-velocity heterogeneity can alternatively explain the anisotropic signal retrieved from body waves [Plomerová et al., 2006].
 One has to admit that real slab geometry is more complex than that retrieved from any tomography, which might also lead to differences between the synthetic and observed P-spheres. This is also evident in the TR zone, where the heterogeneity effect of the slab is combined with the fabric of the flat-lying mantle lithosphere. Having in mind the complex crustal structure, effects due to the significant (~20 km) step of Moho depths could explain the weak P patterns in the TR. However, the heterogeneous signal observed in the TR should not be assigned solely to the complex Moho geometry, because (1) we corrected the travel times for propagation of the P waves through the crust [Di Stefano et al., 2011] and (2) the step in Moho depths located beneath the TR3 (Figure 11) would make the hypothetical bipolar pattern caused by the slab heterogeneity even stronger. Nevertheless, the bipolarity of the observed P-sphere pattern is suppressed in this region. Therefore, a source of the observed directional dependences of the P-wave residuals in the TR must be located deeper below the crust, i.e., in the upper mantle. We consider this heterogeneous zone to be formed by blocks of mantle lithosphere with frozen anisotropy. The blocks might have been sheared during the movement of the Adriatic promontory to the north, at the contact zone with the European plate, before the opening of the Tyrrhenian Sea [e.g., Handy et al., 2010].
 The shear-wave splitting evaluated in the region attains more than 1 s at the most of the stations [Margheriti et al., 1996; Plomerová et al., 2006; Salimbeni et al., 2008]. Average value of ~1 s is typical for most of tectonically different continental regions [Wuestefeld et al., 2009]. The heterogeneous crust, with schistose metamorphic rocks and imbricated layers like those in the Adriatic upper crust [Finetti et al., 2001] representing anisotropic media, can only locally affect the observed splitting delays by a weak contribution at values below a detectable limit of teleseismic shear-wave splitting, for which usually 0.3 s is accepted.
 An intrinsic anisotropy of the upper mantle, caused mainly by systematic lattice preferred orientation of olivine crystals, is at least partly constrained by anisotropy of mantle xenoliths [e.g., Ben-Ismail and Mainprice, 1998]. Formation of frozen-in anisotropic fabric can be related to the stress field of the last mantle-lithosphere deformation [Silver and Chan, 1991; Savage, 1999], or, the fabric has preserved information about the olivine preferred orientation formed in the stress field in time of the lithosphere origin [Babuška and Plomerová, 1989]. The overall percentage of the preferred orientation of olivine crystals in an anisotropic volume is decisive for the apparent large-scale anisotropy detected through the shear-wave splitting. Surface-wave studies [e.g., Montagner, 1994] detect significant decrease of anisotropy below 200 km. Similarly, characteristic changes of both the P- and S-wave anisotropic parameters and their relation to distinct tectonic features represent independent arguments for locating sources of anisotropic signal in the lithospheric part of the upper mantle in stable continental regions. We may thus ask, which of the potential anisotropic sources—the lithospheric mantle, the sublithospheric mantle (asthenospheric flow), or even the subducted slab—dominates the apparent anisotropic signal observed in the province around convergent plate margins, or, what is the mutual proportionality of these sources.
Babuška et al.  showed that anisotropic signal detected in the P spheres, i.e., in distribution of the negative and positive directional terms extracted from relative residuals, can be modeled in the continental lithosphere by peridotite aggregates with plunging symmetry axes. 3D self-consistent anisotropic models of individual domains of mantle lithosphere result from joint inversion/interpretation of both the P- and S-wave anisotropic parameters [Babuška and Plomerová, 2006]. While a cumulative character dominates in the P-wave travel times, shear-wave splitting characteristics are more complex in the case that waves propagate through a more complicated medium than a single anisotropic “layer”. Inclination of symmetry axes in the lithospheric domains causes a dependence of evaluated splitting parameters on initial polarization (back-azimuth). Then, the characteristic π-modality, a priori assumed in the case of horizontal symmetry axes, disappears. Modeling variations of apparent splitting parameters (often from 2D evaluation of azimuthal anisotropy), by double-layer models with horizontal symmetry axes, leads to an additional reduction of the back-azimuth variations to π/2 periodicity [Savage and Silver, 1993]. Moreover, these models do not generate the observed bipolar P pattern.
 Subcrustal lithosphere beneath the TYR region is too thin (~30 km, see Figures 11 and 12, and, e.g., Miller and Piana Agostinetti ) to generate the distinct bipolar pattern of the P residuals and the ~1 s split delay time (see Figures 3 and 11). Moreover, fabrics of oceanic lithosphere tend to be of a flat-sandwiched structure with subhorizontal high velocities coupled with the sublithospheric flow. Thickness of this layer is comparable with wavelength of the teleseismic shear waves recorded by broad-band seismometers, thus at a limit of “visibility”. Dominating trench-parallel fast S polarizations in the TYR region and also in the western rim of the TR1 zone allow us to associate most of the anisotropic signal westward of the NA range with a slab-parallel flow in the asthenosphere. The fast shear-wave polarizations only slightly rotate to ~45° of the trench-perpendicular direction in the southern TYR region. No distinct dependence of the polarizations on back-azimuth also supports the interpretation by a subhorizontal flow. Nevertheless, the fast S polarization azimuths at the stations in the TYR for waves from the north-east become to be partly controlled also by the anisotropy of the Adriatic slab; see the SKS ray paths and greater variations of the splitting parameters in Figure 11. However, slab-parallel orientation of the fast S polarizations dominated at the TYR region, is not in agreement with a trench-perpendicular flow that would be expected for an extensional region related to a slab retreat, as was already pointed out in Plomerová et al.  and Salimbeni et al. . Further to the south of the NA, the fast S polarizations evaluated by Margheriti et al.  tend to be closer to the standard trench-perpendicular model, but this orientation might be related to a slab “window” between the NA and the Southern Apennines [Lucente et al., 2006].
 How then to explain the strong bipolar P-sphere pattern attaining up to 0.9 s difference in the extremes of the directional terms in the TYR region, if the lithosphere is thin and of oceanic type with a flat fabric and assuming the recent tomographic velocity model of Benoit et al.  is correct? Contribution from the fast isotropic heterogeneity to the P pattern appears negligible (cf. observed and synthetic P spheres, Figures 2, S4, 8, and S5). However, we have to consider that the thick continental Adriatic lithosphere exhibits clear signs of anisotropy (see Figure 5 and discussion below). Assuming orientation of anisotropy within the slab remained unchanged relative to the plate boundaries, the rays to the stations in the TYR region propagate from the north-east along the high-velocity directions relative to those from the south-west (see Figure 2). Ray paths within the short but thick subducted slab (see Figures 11 and 13) are long enough to create the observed P-pattern. This finding can be considered as an independent indication that a former oceanic subduction resulting from the Europe-Africa-plate collision was followed by a steep short subduction of the thicker continental Adriatic plate [Handy et al., 2010].
 Differences in distributions of shear-wave polarizations expressed in the rose diagrams (see Figure 5) evidence complex structure of the upper mantle. Similar to other continental regions [e.g., Babuška and Plomerová, 2006, 2012], including those of Archean ages [Plomerová et al., 2011], the observed changes in anisotropic signals of body waves allow us to delimit boundaries of domains of the upper mantle with consistent fabrics. On the basis of geographical and back-azimuth variations of shear-wave polarization, Plomerová et al.  proposed a domain-like structure of the Adriatic plate (ADR). In the FER, the NA orogen-perpendicular fast S polarizations prevail, whereas in the SD, the apparent polarizations tend to orientate orogen parallel (see Figure 4). Distributions of polarization azimuths determined for the waves propagating from the south-west differ from those characterizing propagations from the north-east (see Figure 6). This finding and the different polarizations within the two domains testify thus for different mantle structures of both regions. Diffuse structure of the rose diagrams of ADR1 and ADR2 for waves propagating from the north-east (see Figure 5; see also Figure 6 of Salimbeni et al. ) turns to clearly bimodal distributions in the circular diagrams constructed for the FER and SD separately. We associate these two contributions in the apparent polarizations with the shear-wave propagation through anisotropic sublithospheric mantle and then through different blocks of the thick Adriatic lithosphere with their own fabric. One can hardly expect sudden and distinct changes of the asthenosphere flow along the NA slab, but sudden fabric changes of individual blocks/domains of the continental mantle lithosphere have been well documented [Babuška and Plomerová, 2006].
Salimbeni et al.  prefer a double-layer model with a horizontal “fast” axis at 170° in the upper layer and 135° in the lower layer for the BARR and RAVR stations. A contribution to the apparent split delay time δt from the upper (lithosphere) layer of this model doubles that from the lower (asthenospheric) layer. Both stations belong to our FER domain. We constructed a synthetic rose diagram (see Figure 7c) for the double-layer model of Salimbeni et al.  and compared it with the rose diagram of the FER domain for the same back-azimuths (see upper right diagram in Figure 6). However, there is only a little correlation between the rose diagrams, probably due to a low number of waves coming from different back-azimuths to characterize the structure beneath single stations as Salimbeni et al.  also admit.
 Having a better distribution of rays is not only a question of a length of period of an array registration. Ray-path limitation is a reality, because waves arrive only from earthquakes, whose locations are restricted to specific regions. Therefore, interpreting jointly independent sets of body-wave anisotropic parameters broadens fan of ray paths sampling the upper-mantle volume, and it is challenging to retrieve more realistic self-consistent models of the upper mantle. In the case of modeling, the mantle lithosphere in stable continental regions, the bipolar pattern of the P spheres can be modeled by peridotite aggregates with plunging symmetry axis a in the case that azimuth of the fast S polarizations and dip direction from the P-spheres agree, or, with plunging symmetry axis b, i.e., dipping (a,c) foliation plane, if the fast S polarizations parallel the strike of the (a,c) plane [Babuška et al., 1993]. Following this rule and interpreting directly the fast S polarizations (see Figure 6), one could relate the relatively stable north-south component to the anisotropy reflecting the slab-parallel flow in the sublithospheric upper mantle on the Adriatic side. In such a case, the lithosphere fabric of the FER could be approximated by an anisotropic model with the a axis dipping at azimuth ~210° and that of SD by an anisotropic model with the b axis of symmetry and the southward dipping (a,c) foliation striking at ~300°. However, these models, approximated by two anisotropic layers with horizontal symmetry axes [Savage and Silver, 1993], did not accommodate the visualized apparent polarizations.
 In the case of a grid search through double-layer anisotropic models of the upper mantle beneath the ADR, we found an azimuth of high velocities and a delay time for each of the layers in both the FER and SD regions (see Figure 7 and Table 1) to fit the observed rose diagrams (see Figure 6). The thickness of upper layer beneath the FER corresponds to the thickness of the mantle lithosphere of ~50 km (Po Plain basin) resulting from the static terms of P-wave travel-time residuals (see Figure 11). The models show high velocities at azimuth 80° and 60° in the lower layers (sublithospheric upper mantle) of the FER and SD domains, respectively, far from the slab-parallel orientation predicted due to the suggested slab roll-back [e.g., Faccena et al., 2003]. Fast velocity orientations in the upper layers at 20° (200°) and 0° (180°) in the FER and SD lithosphere, respectively, correlate well with the south-south-west azimuths of the dipping high velocities inferred from the P-spheres (see Figures 2 and S4).
Menke and Levin  developed a waveform fitting technique that allows testing radial-horizontal and tangential-horizontal components of shear waves for splitting as predicted by one- and double-layer anisotropic models with horizontal high-velocity directions. Without the necessity of fitting the apparent parameters, the method yields consistent results and provides additional information on statistical significance of the double-layer solution. However, in the case of the NA, the double-layer solutions for the upper mantle proposed by this method for some stations (Table 1) differ from solutions by the method fitting the apparent splitting parameters [Savage and Silver, 1993]. It is surprising that for two nearby stations ZOCR and RAVR, both located in the FER domain of the Adriatic plate, the fast-velocity directions in the upper and lower layers result in mutually interchanged azimuths (Table 1). Since the waveform fitting method searches for azimuth anisotropy only, it is not suitable for regions with complex structure and especially for retrieving mantle lithosphere fabric, for which there are independent indications on fabric inclination. Menke and Levin  admit that some processes, other than those producing the double-layer flat anisotropy, affect the shear-wave splitting. Fabric inclinations, either in the continental lithosphere or due to complex nonhorizontal flow in the sublithospheric mantle, can be the sources.
 Similar regionalization of the NA province into three regions—the Tyrrhenian, Adriatic, and Overlapping domains, the last one coinciding with our TR—comes from the harmonic decomposition of migrated radial and transverse receiver functions [Bianchi et al., 2010]. The procedure takes into consideration anisotropy in the uppermost mantle. The authors associate the revealed symmetry directions of anisotropy with characteristics of the mantle material at depths of 40–70 km, i.e., with fabrics of the mantle lithosphere, particularly of the thick Adriatic plate. Prevailing north-north-east to north-east directions are in agreement with axis orientations we derive in this study, as well as the distinct regionalization according to lateral changes of the anisotropic parameters [Plomerová et al., 2006].
 Simple estimate on a depth source of anisotropy below a dense array is usually based on calculating Fresnel zones [e.g., Alsina and Snieder, 1995], assuming horizontal fast velocities in the upper mantle. Salimbeni et al.  estimated the depth interval of anisotropic layers and concluded that both the double-layer models with a horizontal axis of symmetry and the Fresnel-zone estimate suggest a lithospheric contribution [see also Plomerová et al., 2006].
4.3 Lithosphere thickness around the steep Adriatic subduction
 Differences in estimates of the lithosphere thickness beneath the northern Italy stem from independent data and methods [surface waves—Panza et al., 1980; S-wave receiver functions (SRF)—Miller and Piana Agostinetti, 2012; P-wave travel-time residuals—Babuška and Plomerová, 1990]. All of the estimates are consistent in showing a shallow LAB beneath the TYR westward of the NA and a thicker continental Adriatic lithosphere eastward of the NA (Figure 13). The modification of the method presented by Babuška and Plomerová , i.e., separation of the static terms according to the wave azimuth relative to the strike of the NA mountain range and plotting the LAB depths at piercing points of the rays and the LAB boundary, results in a smooth model of LAB relief (see Figures 11 and 12) in contrast to LAB estimates averaged for all back-azimuth ranges and plotted directly beneath each station [e.g., Miller and Piana Agostinetti, 2012]. We relate the banana shape of the deep LAB relief beneath the NA with the Adriatic slab subduction (Figure 13).
 The model width of the Adriatic subduction is of ~80–100 km (see Figures 11 and 12) can be an independent sign of the continental plate subduction [Bianchi et al., 2010]. Diffuse values of the LAB estimates located at the north-eastern edge of the slab probably reflect the wave propagation along the high-velocity directions within the anisotropic slab. Less well-defined LAB beneath the Adriatic slab along the southern profile (see Figure 12) can be an independent indication of a fragmentation of the slab, whose continuation south of 43°N is questioned by results of independent methods [Benoit et al., 2011; Spakman and Wortel, 2004].
Piccinini et al.  calculated the P- and S-wave spectra to infer attenuation in the Earth mantle and to analyze its variations beneath the NA and surroundings. Stations located south-westward of the 700 m elevation isoline of the Apennine chain above the steep subduction (Tyrrhenian side of the NA) exhibit high attenuation, while a low attenuation prevails on the Adriatic side of the NA orogen (Figure 13). Subvertical orientation of the Adriatic slab imaged by tomography [e.g., Benoit et al., 2011] excludes a presence of a standard asthenosphere wedge above a less steep subduction. Thomson et al.  state from measurements of low-temperature thermochronometers (apatite (U-Th)/He and fission track) that “no single model of wedge kinematics is likely appropriate for the NA over the long term” and that “different lithospheric geodynamic processes have acted at different times in different lateral segments of the orogen”. To explain the high attenuation evaluated above the subducting continental Adriatic lithosphere relative to its flat part leads us to refresh the idea of the slab detachment [Levin et al., 2007]. The potential detachment filled with the low-velocity asthenospheric material would have to be narrow, with width below the teleseismic shear-wave length (~40 km), and in its initial stage.
4.4 Seismic Anisotropy and Tectonics
 Seismic anisotropy measured by means of shear-wave splitting, detection of QL waves or evaluating transverse components of receiver functions shed light on the lithospheric and sublithospheric parts of the upper-mantle fabrics, their past and present deformation, and lateral/vertical changes of their structures. The main results of anisotropy studies evidence a complex structure of the region, in which none of the basic models, e.g., a slab roll-back or a slab detachment, works within the whole province [Plomerová et al., 2006; Levin et al., 2007]. Shear-wave anisotropy recognizes different mantle fabrics westward and eastward of the north-west/south-east oriented NA orogen [Salimbeni et al., 2008], with additional subregions, particularly in the Adriatic domain [Plomerová et al., 2006], while the QL waves differentiate the mantle structure north and south of the 44°N latitude [Levin et al., 2007].
 Results of this study are in agreement with previous investigations mentioned above, distinguishing regions with principal mantle tectonics: (1) the TYR with the thin oceanic lithosphere and distinct north-west/south-east oriented flow in the sublithospheric mantle, (2) two regions with relatively thicker continental lithosphere belonging to the Adriatic plate underlain by the mantle, which is deformed inconsistently with the retreat-related subduction-zone parallel flow, and (3) the approximately 80 km wide TR in between the Tyrrhenian and ADRs and above the steeply subducted part of the Adriatic plate (Figure 14 and Table 2). The steep high-velocity heterogeneity [Benoit et al., 2011] makes evaluation of particularly P-wave anisotropy difficult, and we admit that small remnant effects could be preserved in the overall anisotropy patterns. Proposed geologic models of the NA region are derived mostly from shallow observations; however, teleseismic measurements focus on deeper processes which take place in the upper mantle and drive the shallow tectonics. Evaluation of seismic anisotropy can thus help to differentiate especially between subduction with or without a slab roll-back or a slab break-off, lateral extent of these phenomena, including a slab continuity, or its fragmentation.
Table 2. Summary of the Upper-mantle Anisotropic Models in the Northern Apennine Region
Lucente et al.  map a long route of the progressive eastward retreat of a regional-scale subduction zone (trapped between two continents) through the west-central Mediterranean mantle in the last 30 My. At present, the upper-mantle fabrics resulting from seismic anisotropy do not show the simple deformation pattern in the upper mantle either due to a slab roll-back, i.e., the trench perpendicular in the supra-slab region and the orogen-parallel in the subslab region, or, due to a slab break-off resulting in the sublithospheric mantle flow towards the site of the detachment. Variability of the anisotropy as to its orientation evidences a complex structure of the NA and fragmentation of the Apennine chain [Lucente et al., 2005; Miller and Piana Agostinetti, 2012].
 Though, e.g., Pondrelli et al.  and Miller and Piana Agostinetti  assume that the Adriatic plate subduction beneath the NA is still active, results from body-wave anisotropy presented in this paper as well as those from surface-wave study of QL waves by Levin et al.  suggest that the subduction-zone retreat has ceased. Formerly eastward dominated asthenospheric flow due to the slab roll-back seems to be deflected on the Tyrrhenian side to the north-west/south-east direction along the short steep slab [Benoit et al., 2012; for the most recent velocity tomography image]. Then, the north-western movement of the Tyrrhenian crust, detected north of ~41°N from GPS measurements [Serpelloni et al., 2005], appears coherent with the upper-mantle flow. The flow slightly turns clockwise towards the NA chain. The same rotation can be observed along the NA in direction from south to the north (see Figures 4 and 5, and Table 1). The recent isotropic tomography from the RETREAT data [Benoit et al., 2011] as well as the tomography from data of permanent stations only [Lucente et al., 1999] do not show a continuous slab along the whole collision zone in the Italian peninsula [Wortel and Spakman, 2000; Piromallo and Morelli, 2003]. The authors limit the high-velocity heterogeneity in the upper mantle related to the NA subduction to latitudes north of ~42°, being separated from the Southern Apennines by a tear beneath the Central Apennines. The separation is visible also in mapping of the LAB by SRF [Miller and Piana Agostinetti, 2012] and in a distinct change of size and direction of residual geodetic velocities [Serpelloni et al., 2005]. Low Pn velocities [Mele et al., 1997] and young magmatism [Rosenbaum et al., 2008] surrounding 43°N, together with high flux of CO2 [Gambardella et al., 2004] southward of 43°N also indicate presence of hot asthenosphere in shallow mantle depths (Figure 13). Up to now, there is no convincing anisotropy measurement indicating a mantle material flow related to a tearing associated to the ending part of the NA subduction [see also Levin et al., 2007; Margheriti et al., 2003].
 Structure of the upper mantle on the Adriatic side of the NA is complex, reflecting the mantle-lithosphere structure of ~80 km thick continental Adriatic plate (see Figures 11, 12, and 13) as well as deformation in the sublithospheric mantle constricted by nearby subductions beneath the Western and Eastern Alps, and Dinarides [Babuška et al., 1990; Piromallo and Morelli, 2003; Lippitsch et al., 2003; Koulakov et al., 2009]. However, due to the lack of well distributed shear-wave polarization measurements, none of the techniques used resulted in a unique double-layer anisotropic model of the mantle beneath the FER and SD Adriatic domains (Table 1; see also Menke and Levin ; Plomerová et al. ; Salimbeni et al. ). Nevertheless, the models agree with a change of the asthenospheric flow from the north-eastern towards the east-north-eastern (60° in SD of the Adria, and 80° in the FER). Alternatively, Levin et al.  hypothesize that the mantle flow towards the site of the subducted slab detachment from the surface lithosphere north of 44°N, originally proposed by Wortel and Spakman , would explain the apparent north/south trending SKS fast polarizations and the lack of QL scattering. However, the subduction zone parallel polarizations in the Tyrrhenian domain and prevailingly null or very week split measurements in the Alps-NA transition domain (see Figure 4; see also Plomerová et al. ) most probably reflect the fact that the broad-band shear waves sample multistructures at scales of their wavelength. Nearby subductions beneath the Western and Eastern Alps, as well as beneath the Dinarides (Figure 14), might block a mantle flow expected in the case of a broader NA slab break-off, similarly as in the case of a noticeable subduction-zone retreat. Seismicity in the NA is mostly limited to the crust [Chiarabba et al., 2005]. Normal faults prevail in the crust above the slab and reverse faults in the crust north-easterly from the NA mountain range [Pondrelli et al., 2006]. This is in accord with a continuing extension along the NA and an active compression of the crust above the Adriatic lithosphere (Figure 13) and can reflect a possible pull of the sinking Adriatic lithosphere.
 In the upper mantle beneath the NA, we recognize regions of different fabrics in the mantle lithosphere and also in the sublithospheric mantle. Based on joint analysis of directional terms of relative P-wave travel-time residuals and shear-wave splitting evidencing anisotropy in the upper mantle, we model the upper-mantle fabric in the Tyrrhenian, Transition, and ADRs. The last one comprises two subregions with distinct anisotropy of the continental lithosphere. The TR between the Adriatic and TYRs is about 80 km wide. Joint analysis of the two different and independent data sets (P-wave travel-time residuals and shear-wave splitting) allows us to infer anisotropic structures of the mantle lithosphere oriented generally in 3D with inclined symmetry axes. Tests of the well-known trade-off between the heterogeneity and anisotropy showed differences between the synthetic P spheres calculated for the standard tomographic model of isotropic velocity perturbations [Benoit et al., 2011] and the observed P spheres implying that we succeeded in detecting anisotropy in both types of body waves.
 Back-azimuth dependence of body-wave anisotropic parameters along with their geographical variability require double-layer models and anisotropy with inclined symmetry axes, particularly in the upper layer associated with the continental mantle lithosphere of the Adriatic plate. Slab-parallel fast shear-wave polarizations reflect predominantly sublithospheric flow beneath the Tyrrhenian plate. Anisotropic signal expressed in the P-spheres at stations located above the thin and flat oceanic-type Tyrrhenian lithosphere reflects most probably the inherited anisotropy of the subducted slab itself. Anisotropy evaluated in two Adriatic subregions indicates combined effects of fossil anisotropy with a dipping symmetry axis in the mantle lithosphere domains and anisotropy due to the north-easterly oriented flow in the asthenosphere. Fabric of the upper mantle in the TR between both major regions is complex. The region might be composed of slices of the Adriatic mantle lithosphere sheared off during the collision with the European plate.
 Static terms of relative P-wave travel-time residuals, divided according to the strike of the subduction zone (~135º), provided first detailed depth estimate of the LAB in the region of the NA. We estimate thickness of the thin Tyrrhenian lithosphere at ~50 km and of the thicker continental Adriatic lithosphere at ~80 km. We model a depth of the subduction of the Adriatic plate down to no more than 200 km.
 Generally accepted model of the subduction roll-back does not work in the current stage of the NA evolution. Slab-parallel flow in the sublithospheric mantle on the Tyrrhenian side and a north-easterly oriented flow on the Adriatic side support the idea of the ending subduction-zone retreat. Mass transfer in the sublithospheric mantle might be blocked in the upper 200 km by the frame of the surrounding subductions—the Alps to the north and the Dinarides to the east, both formed as a result of independent motions of the microplates between the colliding Europe and Africa.
 We thank all members of the Retreat Seismic Working Group, particularly to L. Margheriti, S. Pondrelli, N. Piana Agostinetti, F. P. Lucente, D. Piccinini (INGV Roma), to J. Park and V. Levin (Yale University, USA), and to B. Růžek, P. Jedlička (IG Prague) for their cooperation during the field work and data processing. To draw most of the figures, we used the GMT plotting software [Wessel and Smith, 1998]. The research was partly supported by grants No. IAA300120709 of the Grant Agency of the Academy of Sciences of the Czech Republic, No. P210/12/2381 of the Grant Agency of the Czech Republic, No. 111-10/253101 of the Grant Agency of Charles University, and No. SVV-2013-267308 of the Ministry of Education, Youth, and Sports of the Czech Republic. We greatly appreciate the thoughtful review comments provided by Francesco Pio Lucente, an anonymous reviewer and Thorsten Becker, Editor-in-Chief, which have improved substantially the manuscript.