Teleseismic receiver functions from a seismic experiment in a small area of the northern Apennines, Italy, reveal strong crustal structure variations across the mountain chain. Receiver functions imaging and full waveform inversion technique are used to constrain the S-wave velocity profile in the crust and to reconstruct the geometry of the main seismic discontinuities at depth. We highlight the presence of the main mode-converting discontinuities in the study area. Most importantly, we identify the crust-mantle transition which is represented, almost everywhere in the study area, by a sharp S-wave velocity increase (over 4 km/s) at depth between 35 and 40 km. However, farther west, the S-wave velocity reaches values typical for the sub-crustal mantle at about 54 km depth, possibly marking the locus where the subducting Adriatic plate starts to dip into the mantle. Here the presence of a shallower discontinuity at about 36 km depth, with S-velocity values around 4 km/s, can be interpreted as the Moho signature of the overriding Tyrrhenian plate.
 The Apennines are a relatively young (Neogene) orogen produced by the emersion of the accretionary prism related to the subduction zone running along the Italian peninsula. The northern part of the chain (northern Apennines) consists of a NE verging thrust-fold belt formed as result of the collision (in the Oligo-Miocene) between the Euroasiatic plate and the Adriatic microplate, once the W-ward subduction of the Tethyan oceanic lithosphere was completed [Alvarez, 1972; Argnani, 2002]. Major geophysical and geological differences across the northern Apennines chain allow to distinguish two different domains: an internal domain, to the W, characterized by high heat flow, volcanic activity, thinned crust, low topography [Jolivet et al., 1994]; an external domain, to the E, with relatively higher topography, thicker crust, lower heat flow [Collettini and Barchi, 2002]. At present, local seismicity reveals a coherent pattern of compression along the external part of the northern Apennines, while extension is found in the internal region [Frepoli and Amato, 1997]. The transition between these two domains takes place in a narrow belt running along the northern Apennines, and possibly represents the surface expression of the colliding plates boundary at depth. The study area is a small sector (about 0,5°Lon × 0,5°Lat) of the northern Apennines, located at 43°N, just at the transition between the internal and the external domains. The chemical, physical and kinematic complexities related to the transition from one domain to the other are reflected in an equally complex crustal structure in this narrow area of the northern Apennines. For this reason, while there is a general agreement on the first order trend of the crust-mantle boundary across the whole northern Apennines, from the Tyrrhenian to the Adriatic coasts, the Moho depth and geometry in this transitional area are debated. Results from previous studies, based both on seismic exploration experiments (the refraction DSS experiments [Ponziani et al., 1995] and the CROP03 deep reflection survey [Pialli et al., 1998]) and on teleseismic Receiver Function interpretations [Piana Agostinetti et al., 2002; Mele and Sandvol, 2003] are inconclusive and somewhat controversial on the crustal attributes of this area. In this study we make use of a dense configuration network (Figure 1), to obtain a fine representation of the crustal structure and of the Moho topography in this complex plate boundary region from Receiver Function analysis.
2. Receiver Functions Data Set and Analysis
 We use data recorded at 14 seismic stations of a local temporary array between October 2000 and May 2001. Stations were equipped with three component broad-band sensors, Lennartz 5-s or Guralp (CMG-40), coupled with Reftek 72A digitizers [Piccinini et al., 2003]. Inter-station distance ranges between 5 and 10 km, allowing for a very dense coverage of teleseismic rays in the study area. We select 100 teleseisms with good S/N ratio with magnitude M ≥ 5.5 and epicentral distance (Δ) between 30° and 105°. We apply the Receiver Function (RF) method, a technique commonly employed to obtain depth estimates of crustal and/or mantle interfaces under three-component seismic stations [Langston, 1979; Steckler et al., 2008]. We compute the RF through the frequency-domain deconvolution technique developed by Di Bona  which allows us to estimate single RF variance and has been proven useful for the analysis of RFs from temporary seismic stations [Piana Agostinetti et al., 2002; Lucente et al., 2005]. RFs are estimated from the deconvolution of the P-wave coda recorded on the vertical component from the horizontal ones. RFs are low-passed with a Gaussian filter to exclude frequencies above ∼1.0 Hz. We select only RFs with a high S/N ratio, displaying a clear first direct-P pulse and a limited variance. Our final selection consists of 55 RFs at 14 stations (between 1 and 5 RFs at each station). The limited operating period of the network does not allow for a complete back-azimuth coverage at the recording sites, therefore we concentrate our analysis on the radial components of the RFs. In fact, the analysis of the tangential components of the RFs, which is informative about the presence of 3-D and/or anisotropic structures beneath the seismic station, requires an almost complete azimuthal coverage to be effective. In order to highlight the structures related to the main structural element of the area, i.e., the Apennines orogen, we project the radial RFs at each stations on a direction normal to the mountain local strike direction (∼N60°E; cf. Figure 1), between stations C001 and B005 (AA′ in Figure 1). Following the scheme proposed by Ferris et al. , we stack all the RFs at single station, obtaining, for each station, a “summary” RF. In the stacked RFs, the single RFs are weighted with the inverse of their standard deviation. Then, summary RFs are binned in 50% overlapping bins as function of the point projection distances along the AA′ profile (Figure 1). The summary RFs are binned in 5 km overlapping intervals at 2.5 km spacing, in the way that each trace influences two adjacent bins only. An advantage of this technique is that combining data from multiple stations into single binned RFs, improve the S/N-ratio by stacking down the signal-generated noise related to topographic scattering [Clouser and Langston, 1995]. By plotting the traces for each bin we obtain high resolution RF image along the entire profile of 40 km length (Figure 2). The continuity, the amplitude variation, and the time shift of RF pulses through adjacent bins allow the construction of a picture of the subsurface structure in terms of the S-wave velocity discontinuities across the Apennines (Figure 2). RF image of the subsurface structure reveals a high complexity along the considered profile (Figure 2). The first and last bins appear substantially different, suggesting a transition in the bulk crustal properties from side to side of our profile. The main pulses show coherent variations between adjacent bins and give us some insight on this transition. We focus our analysis on the evolution of four principal Ps phases from A to A′ (Figure 2). The first is a positive phase (PS1, hereinafter), which arrives between 1.0 and 1.5 s. This phase is clearly visible on some of the bins, while on other bins it interacts with the direct P-pulse, resulting in a broader pulse between 0.0 and 1.0 s. Its time-delay seems to decrease from SW to NE. The second is a negative phase at about 2–3 s (PS2) that becomes more evident starting from 10 km on the profile, and broadens between 25 and 40 km. A positive phase, at about 4 s, is seen in the first 10 km along the profile (PS3); this phase seems to double between ∼10 and ∼25 km into two different pulse which arrive between 4 and 6 s. Last, we observe a clear positive conversion at about 5.5 s in correspondence of the NE termination of the profile (PS4), between 35 and 40 km. While the two phases PS3 and PS4 are sharp and well defined on the SW, and NE sides, respectively, these are not clearly distinguishable under the central part of the profile, between 10 and 35 km, where they appear to interact with each other. On the SW side, a positive pulse is observed at about 7 s.
3. Receiver Functions Inversion
 To translate the S-wave velocity discontinuities recognized on the binned RF profile into tectonic elements present in the study area we performed the 1-D inversions of the five highest quality bins (Figure 2), devoted to reconstruct detailed S-velocity models across the mountain chain. Solving the RF inverse problem for crustal structure is crucial to correctly interpret the RF image and associate the highlighted pulses to buried seismic boundaries. We apply a Monte Carlo inversion scheme called neighbourhood algorithm (see Sambridge [1999a, 1999b] for details), which has been widely used for RF studies [Piana Agostinetti et al., 2002; Steckler et al., 2008]. We model the subsurface structure at various distances along the profile (Figure 2), as a stack of 6 horizontal isotropic layers with constant densities, whose thickness, Vp/Vs, and S wave velocity values vary between a-priori fixed minimum and maximum boundaries. Each binned RF has been inverted using a set of six different inversion schemes (see the auxiliary material for details), in order to find common, hence more stable and robust, features among the best-fitting model families obtained in each inversion. Best-fit S-velocity models for the selected bins are shown in Figure 3a, together with the best-fit model families and parameters space boundaries. A common feature to all the S-waves velocity profiles is a very sharp shallow interface, varying between 5 and 8 km depth, marking a S-velocity increase of about 1 km/s. Based on the main trend of the S-waves velocity with depth (Figure 3a), the retrieved models can be divided in two classes: the S-wave velocity profile P1, on the SW, shows a gradual increase of the S-velocity with depth, while the velocity profiles P2-5 are characterized by the presence of strong velocity inversions within the crust. By the similarities found at a greater level of detail and complexity, models P2-5 can be further separated in two pairs (Figure 3a): on the internal domain (profiles P2,3) the upper crust is characterized by the succession of 2 thin high and low velocity layers from 6 down to 20 km depth, followed, in the lower crust, by a more standard velocity-increasing-with-depth profile [Christensen and Mooney, 1995]. The two easternmost models, P4 and P5, on the external domain, are characterized by a thicker high velocity layer in the upper crust, down to about 20 km depth, and by a broad low-velocity zone downward, in the lower crust, with S-wave velocities as low as 2.6 km/s at P4 and extending down to the Moho at P5 (Figure 3a). In spite of the strong differences in the crustal structure, these Vs models (P2,3 and P4,5) reveal a coherent Moho signal along the profile from 10 to 45 km. In this area, the crust-mantle transition is characterized by a strong positive velocity jump (ΔVs ∼ 1 km/s) of the S-wave velocities above 4.0 km/s at depth between 36–39 km (Figure 3a). On the westernmost profile, P1, the interface marking the transition to S-wave velocity values typical for the sub-crustal mantle (i.e. > 4.0 km/s) [see Christensen and Mooney, 1995, and references therein] is found at about 54 km depth, but with a weaker velocity jump (ΔVs ∼ 0.4 km/s). However, on this S-wave velocity model, an interface is present at about 36 km depth, where Vs reaches 3.9 km/s (with a ΔVs ∼ 0.4 km/s). We observe that the crustal structure transition (between profiles P2,3 and P4,5) lies beneath the Gubbio basin, where the topography starts to raise toward the local Apennines height, while the level land on the SW, the Tiber valley, corresponds on the surface to the striking change in the Moho topography at depth (Figure 3a, 3b).
 The coherence -or incoherence- between nearby S-wave velocity models gives us the basis for geological and geodynamical interpretations of the RF image (Figure 2) and the 1D velocity inversion results (Figure 3a). In the following we summarize the relations between the topography, the peculiarities in the Vs models, and the depths of the main discontinuities found along the profile, and report them on Figure 3b. We associate PS1 pulse to the first interface (red dashed line in Figure 3b) which likely represents the transition between the sedimentary cover and the crystalline basement. Despite the lithological variability [Mirabella et al., 2008], the S-wave velocity in the sedimentary cover and the velocity contrast at its bottom are almost constant through the study area (Vs ∼ 2.5 km/s and ΔVs ∼ +1 km/s, respectively). Its thickness varies between 5 and 8 km, in good agreement with recent interpretation of reprocessed seismic reflection profiles crossing the area just few tens of kilometers to the S, which constrain the top of the crystalline basement at depth between 5 and 10 km [Mirabella et al., 2008]. The general low Vs value in the sedimentary cover and the constant velocity contrast at the underlying basement, can be explained by the internal structure of the sedimentary cover, which has been extensively involved in the last tectonic phases, causing rocks fracturing and mixing of the different lithologies. The crystalline basement is modeled in the inversions as a high velocity layer (Vs > 3.2 km/s; profiles P2-5 in Figure 3a). The nature and composition of the basement at are largely unknown because of the lack of samples of the deep Palaeozoic rocks in the study area [Speranza and Chiappini, 2002]. In the adjacent areas, only the shallower portion of this structure was encountered by deep wells, revealing at its top the presence of the Verrucano Formation and of Palaeozoic phyllites [see Pauselli et al., 2006, and references therein]. The thickness of this high velocity layer (light-blue area in Figure 3b) is about 5 km at P2-P3, while it doubles under the eastern part of the profile (P4-P5). A negative pulse PS2 is clearly distinguishable starting from 10 km along the profile (Figure 2), and marks, in our interpretation, the bottom of the basement. This negative pulse PS2 represents the downward transition to a lower-velocity layer; it is sharp under the W portion of our profile and progressively broadens going toward E. One possible interpretation for the inversion of the S-wave velocity is in the mechanism of formation of the northern Apennines by crustal eastward accretion of major thrust sheets imbricated through fault ramps cutting the basement. This mechanism of crustal accretion produces crustal slices doubling and inverted stratigraphic sequences. This hypothesis is consistent with a thick-skinned style of deformation, where the basement is involved in the major thrust sheets [Speranza and Chiappini, 2002; Pauselli et al., 2006]. In this hypothesis the lithologies atop of the basement would have acted as decollement level, favoring the formation of the fault ramps in the basement. However, due to the limited knowledge of the deep crust nature and composition, we cannot exclude other factors as responsible for the presence of the Vs inversions in the crust, as the presence of weaker lithologies or fluid accumulation at depth. The positive phase PS4 (Figure 2) is generated by the lowermost interface in the five constrained models (Figure 3a), and is interpreted as the crust-mantle boundary (blue dashed lines in Figure 3b). Owing to the transition to similar S-wave velocity underneath, we hypothesize the continuity of this seismic discontinuity across the whole profile (Figure 3a, 3b). Therefore, the deeper Moho found at P1 (Figure 3a, 3b), would mark the locus where the Adriatic plate passes from an almost flat standing, in the external Apennines, to the steep standing in the subduction zone. In this hypothesis, the positive pulse PS3, corresponding to the interface present in profile P1 at about 36 km (Figure 3a), is generated by the overriding Tyrrhenian Moho. The volume included between the Tyrrhenian and Adriatic Moho, is not sufficiently resolved by our data: it has been modeled as an unique layer with thickness of about 20 km and Vs velocity of about 3.8 km/s (see profile P1 in Figure 3a, 3b). We speculate that this volume could be composed by a mix of upper mantle and subducted crustal materials, where mantle Vs is lowered by melting, high heat-flux and/or partial hydration due to fluid release from the subducting plate, while the partial eclogization of the subducted lower crust produces an increase of the crustal Vs, and both these phenomena result into an indistinct velocity signature of the volume between the two Moho. We produced a detailed picture of the subsurface structure in the transition zone between the internal (SW) and the external (NE) domains of the northern Apennines, by projecting 1-D information from RFs computed at closely spaced sites on a 2-D profile crossing the Apennines orogen. The RF analysis and imaging produce evidences for the presence of two overlapping Moho in the study area, a Tyrrhenian Moho, to the SW, and an Adriatic Moho, to the NE. The hypothesis of the existence of a Moho doubling in this area, already proposed by some authors [Ponziani et al., 1995; Mele and Sandvol, 2003], find here a further, independent, support. Therefore, this transitional area of the northern Apennines can be viewed as the surface signature of the boundary between the two colliding plates (the Euroasiatic plate and the Adriatic microplate) at depth. Moreover, our analysis provides a clear indication of the locus where the Adriatic Moho starts to dip into the mantle, i.e., beneath the Tiber valley. The geographical correspondence between the change of the dip angle in the subducting plate and the lowest topography could suggest the existence of a relationship between these two tectonic elements and deserves to be further investigated.
 We are grateful to Massimo Di Bona for his RFs computing code and C. Chiarabba for useful discussion. We thank the Città di Castello (2000/2001) working-group. N. Piana Agostinetti was supported by project Airplane (funded by the Italian Ministry of Research, Project RBPR05B2ZJ_003). Comments from two anonymous reviewers improved the quality of the manuscript.