(This is a sample cover image for this issue. The actual cover is not yet available at this time.)

Transcription

1 (This is a sample cover image for this issue. The actual cover is not yet available at this time.) This article appeared in a journal published by Elsevier. The attached copy is furnished to the author for internal non-commercial research and education use, including for instruction at the authors institution and sharing with colleagues. Other uses, including reproduction and distribution, or selling or licensing copies, or posting to personal, institutional or third party websites are prohibited. In most cases authors are permitted to post their version of the article (e.g. in Word or Tex form) to their personal website or institutional repository. Authors requiring further information regarding Elsevier s archiving and manuscript policies are encouraged to visit:

2 Earth and Planetary Science Letters (2012) Contents lists available at SciVerse ScienceDirect Earth and Planetary Science Letters journal homepage: Upper mantle structures beneath the Carpathian Pannonian region: Implications for the geodynamics of continental collision Y. Ren a,n, G.W. Stuart a, G.A. Houseman a, B. Dando b, C. Ionescu c, E. Hegedüs d, S. Radovanović e, Y. Shen f, South Carpathian Project Working Group a School of Earth and Environment, University of Leeds, LS2 9JT Leeds, UK b RockTalk Imaging Ltd., Chacewater, Cornwall TR4 8PN, UK c National Institute of Earth Physics, Bucharest, Romania d Eötvös Loránd Geophysical Institute, Budapest, Hungary e Seismological Survey of Serbia, Belgrade, Serbia f Graduate School of Oceanography, University of Rhode Island, USA article info Article history: Received 7 March 2012 Received in revised form 14 June 2012 Accepted 19 June 2012 Editor: P. Shearer Keywords: finite-frequency tomography upper mantle structure Carpathian Pannonian system continental collision abstract The Carpathian Pannonian system of Eastern and Central Europe represents a unique opportunity to study the interaction between surface tectonic processes involving convergence, extension and convective overturn in the upper mantle. Here, we present high-resolution images of upper mantle structure beneath the region from P-wave finite-frequency teleseismic tomography to help constrain such geodynamical interactions. We have selected earthquakes with magnitude greater than 5.5 in the distance range , which occurred between 2006 and The data were recorded on 54 temporary stations deployed by the South Carpathian Project ( ), 56 temporary stations deployed by the Carpathian Basins Project ( ), and 131 national network broadband stations. The P-wave relative arrival times are measured in two frequency bands ( Hz and Hz), and are inverted for Vp perturbation maps in the upper mantle. Our images show a sub-vertical slab of fast material beneath the eastern Alps which extends eastward across the Pannonian basin at depths below 300 km. The fast material extends down into the mantle transition zone (MTZ), where it spreads out beneath the entire basin. Above 300 km, the upper mantle below the Pannonian basin is dominated by relatively slow velocities, the largest of which extends down to 200 km. We suggest that cold mantle lithospheric downwelling occurred below the Pannonian Basin before detaching in the mid-miocene. In the Vrancea Zone of SE Romania, intermediate-depth ( km) seismicity occurs at the NE end of an upper mantle high velocity structure that extends SW under the Moesian Platform, oblique to the southern edge of the South Carpathians. At greater depths ( km), a sub-circular high velocity anomaly is found directly beneath the seismicity. This sub-vertical high-velocity body is bounded by slow anomalies to the NW and SE, which extend down to the top of the MTZ. No clear evidence of a residual slab is observed above the MTZ beneath the Eastern Carpathians. These observations suggest that intermediate-depth seismicity in Vrancea Zone is unlikely to be due to slab tearing, but rather could be explained by either gravitational instability or delamination of mantle lithosphere. & 2012 Elsevier B.V. All rights reserved. 1. Introduction The Carpathian Pannonian system includes the Pannonian and Transylvanian Basins and the surrounding Eastern Alpine, Carpathian and Dinaric mountain belts (Fig. 1). The structure, formation and tectonic evolution of the Carpathian Pannonian region are well documented, with extensive studies by several n Corresponding author. address: earyr@leeds.ac.uk (Y. Ren). authors in the past three decades (Royden et al., 1982; Horváth, 1993; Horváth et al., 2006; Kovács et al., 2007; Ustaszewski et al., 2008; Schmid et al., 2008; Mat enco et al., 2010; Ismail-Zadeh et al., 2012). The complex closure of the Tethyan Ocean during the collision of the African and European plates in Jurassic to Tertiary times produced a series of subduction zones and collisional belts (Schmid et al., 2008; Handy et al., 2010). The Carpathian Pannonian system was formed after the continental collision between the Adriatic microplate and the European continent in the Cretaceous (Royden et al., 1982; Horváth et al., 2006). During the Neogene, the inner Carpathian region was composed of two X/$ - see front matter & 2012 Elsevier B.V. All rights reserved.

3 140 Fig. 1. Topographic map of the Carpathian Pannonian system with surface structural features: red filled patterns indicate outcrops of calc-alkaline and silicic volcanic rocks after Harangi and Lenkey (2007); CSVF represents Central Slovakian volcanic field. Abbreviations on the map: PAL, Peri-Adriatic Line; MHL, Mid-Hungarian Line; BL, Balaton Line; PKB, Pieniny Klippen Belt; Cam. F., Camena Fault; IMF, Intra-Moesian fault; V. Z., Vrancea Zone. Black stars represent all earthquake hypocenters occurred since 1900 with depth 460 km and magnitude 42:0 from the NIEP catalog ( (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) independent units: the Alcapa block in the NW and Tisza Dacia block in the SE (Csontos, 1995; Tischler et al., 2008). The Alcapa block, bounded by the Pieniny Klippen Belt Zone to the north and the Peri-Adriatic and Balaton lines to the south, occupies the northern Pannonian Basin and the inner western Carpathians south of the Pieniny Klippen Belt (Kovács and Haas, 2011). The Tisza Dacia block is composed of two sub-units: the Tisza block underlying most of the Great Hungarian Plain and the Dacia block comprising the Transylvanian basin, bounded by the Eastern Carpathian and South Carpathian mountain ranges (Csontos and Vörös, 2004) (Fig. 1). As the Adriatic microplate pushed northward, the Alcapa block was extruded to the ENE with a counterclockwise rotation of about 601 (Márton and Fodor, 1995), whereas the Tisza Dacia block moved ESE with a clockwise rotation of about 901 (Pâtras-cu et al., 1994). The movements were accompanied by convergence and uplift in the Carpathians, and possibly subduction of oceanic lithosphere along an Eastern Carpathian margin (Royden et al., 1982; Horváth, 1993). In Mid Miocene, with the Alcapa and Tisza Dacia blocks juxtaposed, lithospheric extension formed the Pannonian Basin (Csontos et al., 1992). Ustaszewski et al. (2008) estimated a 290 km SW NE extension in the Alcapa unit and up to 180 km of extension in the Tisza Dacia unit during its emplacement into the Carpathian embayment. A commonly accepted driving mechanism for rifting and extension of the Pannonian Basin is the southwestward subduction of oceanic lithosphere along a retreating Carpathian arc, with slab roll-back associated with the diapiric upwelling of asthenospheric mantle beneath the basin (Horváth, 1993; Horváth et al., 2006). Regional tomography using P-wave travel times shows a broad fast anomaly lying horizontally beneath the Pannonian region in the mantle transition zone (Wortel and Spakman, 2000; Piromallo and Morelli, 2003). These authors interpreted this fast material as remnant subducted slabs from the Tethyan margin beneath a generally slow Pannonian. Although structural and petrological evidence indicates Eastern Carpathian subduction prior to 10 Ma, alternative ideas have emerged in recent years. Analyzing the Mid-Miocene volcanic rocks in the western Carpathian region, Kovács and Szabó (2008) pointed out that subduction may not have developed along the Western Carpathians and proposed that mantle flow, associated with the eastward extrusion from the Alpine convergent zone, could initiate the extension (Kovács et al., 2012). Houseman and Gemmer (2007) suggested that an unstable thickened crust, produced by Adriatic convergence, could trigger gravitational instability of continental lithosphere that would drive downwelling beneath the Carpathians and extension of the Pannonian Basin. Contrary to this idea, however, Dando et al. (2011) obtained tomographic images which showed a high-velocity body at depths greater than 300 km crossing the Pannonian Basin; they inferred that break-off of this feature from the overlying lithosphere may have initiated extension. The Vrancea Zone is another puzzling part of the Carpathian Pannonian system due to its unique intra-continental situation. Located in the bend zone of the southeastern Carpathian arc (Fig. 1), the Vrancea region is characterized by the occurrence of intermediate-depth earthquakes in a narrow, nearly vertical column (Roman, 1970; Fuchs et al., 1979; Oncescu et al., 1984; Wenzel et al., 1998). Several seismic tomographic studies show a very localized, quasi-vertical, high-velocity body beneath the Vrancea zone extending from 60 to 400 km depth and co-located with the seismogenic zone (Lorenz et al., 1997; Fan et al., 1998; Wenzel et al., 1998; Wortel and Spakman, 2000; Martin et al., 2005, 2006; Tondi et al., 2009; Koulakov et al., 2010). Some of these studies have also pointed out the presence of a low-velocity anomaly located NW of the Vrancea seismogenic zone beneath Transylvania (Fan et al., 1998; Wortel and Spakman, 2000; Martin et al., 2006). This low-velocity body has been attributed to asthenospheric upwelling induced by subduction or lithospheric delamination (Gîrbacea and Frisch, 1998; Knapp et al., 2005). Measurements of seismic wave attenuation using data from

4 141 Vrancea earthquakes (Popa et al., 2005) provide further evidence of this low-velocity, high attenuation structure which acts to filter out the high-frequency components of the seismic waves propagating to the NW. This low-velocity body also correlates well with anomalously high surface heat flow (Demetrescu and Andreescu, 1994). The physical and rheological nature and the cause of the seismicity within the high-velocity body beneath Vrancea continue to be debated. In the pioneering work of McKenzie (1970) and Fuchs et al. (1979), the intermediate-depth earthquakes are attributed to a remnant oceanic slab sinking vertically into the mantle. Based on their tomographic studies, Wortel and Spakman (2000) and Sperner et al. (2001) proposed that the subduction of an oceanic slab beneath the Eastern Carpathians was followed by slab break-off, with the break-off point migrating SE towards Vrancea since 10 Ma (Nemcok et al., 1998). In this model, the seismogenic zone beneath Vrancea corresponds to the final stage of slab break-off and explains the history of time-progressive Neogene volcanism in the Eastern Carpathian hinterland Fig. 2. (a) Distribution of broadband stations used in this study. Blue filled squares represent stations of the temporary network deployed in South Carpathian Project ( ). Black circles mark stations from the temporary network deployed in the Carpathian Basins Project ( ). White inverted triangles depict additional permanent broadband stations used in this study. Yellow triangles represent seismic stations from Romanian Regional Seismic Network. (b) Azimuthal projection of the 1180 events from 2006 to 2011, with magnitude 45:5, used in the tomographic inversion of P waves. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

5 142 (Seghedi et al., 2004). Knapp et al. (2005), however, suggested that delamination of continental lithosphere also could explain the seismogenic, high-velocity body beneath Vrancea. Using numerical simulations, Schott and Schmeling (1998) and Morency and Doin (2004) demonstrated that mantle delamination and detachment likely occur beneath a hot low-viscosity lower crust if the lithosphere is substantially thickened. Fillerup et al. (2010) speculate that an eclogitic root below the lower crust in the Eastern Carpathians could have initiated the delamination process and therefore would form the core of the seismogenic zone, although direct evidence (e.g., xenoliths) for such an eclogitic root is lacking. Lorinczi and Houseman (2009) showed that mantle downwelling produced by gravitational (Rayleigh Taylor) instability of continental mantle lithosphere can predict the depth distribution of present-day seismic moment release within the seismic zone. This drip model also explains lithospheric thinning beneath the Transylvanian Basin and crustal thickness variations from Transylvania to Vrancea, but does not necessarily exclude an earlier phase of subduction and retreat occurring around the Eastern Carpathians. In order to improve the seismological constraints on the different geodynamical models used to explain the formation of the Carpathian Pannonian system, we set up the South Carpathian Project (SCP), a passive seismic experiment in which 54 broadband seismographs were deployed between June 2009 and June 2011 (Fig. 2a). Combining this dataset with that from the previous Carpathian Basins Project (CBP; Dando et al., 2011) and data from a further 131 national network stations forms a uniquely dense seismic network across the Carpathian Pannonian system (Fig. 2a). In this paper, we present high-resolution images of P-wave velocity structures in the upper mantle beneath the Carpathian Pannonian system derived from inversion of relative P-wave teleseismic travel time residuals using 3D finitefrequency kernel tomography (Hung et al., 2004). The resulting images provide important constraints on the geodynamic processes that have formed the Carpathian Pannonian system, in particular on the formation and evolution of the Pannonian Basin, the question of subduction beneath the Eastern Carpathians and the cause of intermediate-depth seismicity beneath Vrancea. 2. Data and inversion 2.1. Data sets The data used in this study are assembled from: the South Carpathian Project (SCP, ), the Carpathian Basins Project (CBP, ), the Romanian Seismic Network and permanent stations in the region sourced from IRIS (Incorporated Research Institutions for Seismology), Orfeus (Observatories and Research Facilities for European Seismology) and GFZ Seismological data archives. Fig. 2a shows the distribution of seismographs used in this study. We established the South Carpathian Project (SCP), a major temporary deployment (June 2009 June 2011) of seismic broadband systems extending across the eastern Pannonian Basin, Transylvanian Basin and the South Carpathian Mountains in Hungary, Romania and Serbia. The temporary network consisted of 17 Guralp CMG-40T (broadband to 30 s), 13 Guralp CMG-3T (broadband to 120 s) and 24 Guralp CMG-6TD (broadband to 30 s). The Carpathian Basins Project (CBP; Dando et al., 2011) consisted of 56 seismographs in Austria, Western Hungary and Serbia, operating between 2005 and In addition, seismograms from 131 national network broadband stations in the region (Fig. 2a) in the period were assembled. We have considered all teleseismic events with magnitude greater than 5.5 from the Harvard CMT catalog ( that occurred between January 2006 and April Only teleseismic data with average epicentral distance between 301 and 951 were used in this study. Our final P-wave dataset includes relative arrival times from 1180 earthquakes (Fig. 2b) Measurements Using multichannel waveform cross-correlation (VanDecar and Crosson, 1990), relative P arrival times between stations were measured in two frequency bands (high: Hz and mid: Hz), from vertical component seismograms. To minimize and remove problematic data, all traces were visually inspected after applying the cross-correlation derived time-shifts. To detect remaining sources of error such as cycle skippings or temporary timing problems, the relative arrival-time residuals were also analyzed as a function of back-azimuth for each station; residuals within a 21 back-azimuth bin that differed by more than 2 standard deviations from the median of the bin were removed. Furthermore only relative arrival times from event records with a signal-to-noise-ratio (SNR) greater than 6.0 and an inter-trace correlation coefficient (CC) higher than 0.85 were inverted. The SNR was computed as the ratio of the maximum peak-to-trough amplitude in a window (15 s for high-frequency, and 30 s for midfrequency) around the predicted arrival time to the root mean square (rms) amplitude of a 60 s noise window taken 50 s before the arrival; the correlation coefficients were obtained directly from the multichannel cross-correlation method (VanDecar and Crosson, 1990). Finally, we only selected events with at least 10 P relative arrival times. After each step of selection, the MCCC procedure was rerun to obtain new relative arrival-time residuals. Fig. 3 shows examples of high-frequency ( Hz) and midfrequency ( Hz) P-wave relative arrival-time residuals before and after the selection procedure for the station ARSA. With an initial data set of 92,844 and 69,846 P-wave relative arrival times at high and medium frequencies, respectively, the selection procedures reduce the numbers of measurements to 85,886 and 54,144 respectively. The averages of estimated measurement uncertainties (using the methodology of VanDecar and Crosson, 1990) were 0:015 s and 0:046 s, respectively, for the high and medium frequency P-waves. However, we consider that the measurement uncertainties derived from the MCCC technique are most likely underestimated, as has been reported in several previous studies (e.g. Tilmann et al., 2001; Bastow et al., 2005; Dando et al., 2011) Inversion The inversions of relative arrival time residuals for P-wave velocity perturbation models were performed using the finitefrequency methodology developed by Hung et al. (2004). This technique has been applied successfully to study mantle structures in several regions with dense distributions of seismographs (Hung et al., 2004; Ren and Shen, 2008; Liang et al., 2011). The details of the inversion technique are described by Hung et al. (2004). Briefly, the finite-frequency theory is based on the idea that the travel time of a seismic wave is sensitive to threedimensional structure around the ray path (Dahlen et al., 2000) as defined by Z Z Z dt ¼ KðxÞdcðxÞ=cðxÞ d 3 x ð1þ where K is the 3D Fréchet sensitivity kernel for the time shift dt produced by anomalous velocity dc at each point x throughout the region. In the evaluation of (1) the velocity anomaly is

6 143 Fig. 3. Example of measured P-wave relative arrival-time residuals in (a) high-frequency band ( Hz) and (b) mid-frequency band ( Hz) for the station ARSA (47.251N, E) before (left) and after (right) the selection process. Fig. 4. (a) Trade-off curve for P-wave inversions using simple damped least squares scheme (DLS) with 2ryr500 (Eq. (3)). The red square represents the preferred model from the trade-off analysis. (b) The preferred DLS P-wave velocity model at 75 km depth using y ¼ 25. (c) Trade-off curve for P-wave inversions using convolution quelling scheme (CQS with a smoothing distance of 40 km, equivalent to about one grid interval in each direction). The red circle represents the preferred model from the tradeoff analysis. (d) The preferred CQS P-wave velocity model at 75 km depth using y ¼ 25 and smoothing width 40 km (Eq. (4)). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) assumed constant in each cell of the discretized model. Previous studies showed that seismic tomography using the finite-frequency formulation produces better resolved structures compared to classical ray-based technique, in terms of amplitude, location and shape of anomalies (Hung et al., 2004; Montelli et al., 2004). In the present study, the volume to be inverted has a latitude range from 411 to 521, a longitude range from 111 to 311 (Fig. 1), and a depth range from the surface to 1200 km. We use IASP91 as the reference model; inversions based on PREM and AK135 background models show only minor differences at shallow

7 144 Fig. 5. Resolution tests using synthetic P-wave checkerboard models. (a) Horizontal slices of the input P-wave synthetic velocity models at 75 km and the retrieved P-wave synthetic velocity models at 75 km, 150 km, 225 km, 338 km and 525 km. The Gaussian type input anomalies are 150 km wide in each direction and have peak magnitude of 72.0%. The section at 150 km (between layers of anomalous material) should have peak magnitude of 0%. (b) Profiles N S and W E (as shown in a) of input and retrieved synthetic models. levels. The model is parameterized using a set of constant velocity cells centered on a 3D regular grid consisting of nodes with a grid spacing of 46.8 km in longitude, 43.7 km in latitude and 37.5 km in depth. The basic discretized inversion problem can then be formulated as d i ¼ G il m l ð2þ where d i is the ith relative arrival time, G il represents the contribution to the differential sensitivity kernels of the lth node for the ith arrival time relative to a reference station, and m l is the velocity anomaly at the lth node. In order to prevent variations in crustal structure below stations being mapped into the mantle model, an additional station term is incorporated into the inversion to represent arrival time shifts caused by crustal and uppermost mantle heterogeneity, after correction for station elevation. Like all tomographic studies, our inverse problem is highly underdetermined and regularization is required to find an optimal and reliable solution. To test different regularization approaches, we

8 Author's personal copy conducted inversions using two schemes: simple damped least squares (DLS) and convolution quelling solutions (CQS) (Hung et al., 2004). In the case of DLS method, the minimum-norm criterion is applied in order to suppress unconstrained variables (Menke, 1984) ^ ¼ ðgt G þ y2 IÞ 1 GT d m ð3þ 145 where I is the identity matrix, all data are equally weighted, and y is the damping parameter, which is determined empirically from the trade-off curve between variance reduction and model norm. In the case of CQS method, the inversion formulation is similar to DLS, except that the model parameters are also subject to a smoothing operator corresponding to a prescribed Gaussian function with an a Fig. 6. Sections from model SCP_P1 showing P-wave velocity perturbations relative to the reference model IASP91 at depths: 75 km, 150 km, 225 km, 338 km, 412 km and 525 km. White filled patterns on the 75 km velocity map indicate outcrops of calc-alkaline and silicic volcanic rocks. Regularization parameters indicated in Fig. 4c and d have been used in the inversion.

9 146 priori imposed correlation length ^m ¼ Q ðq T G T GQ þy 2 IÞ 1 Q T G T d where Q is the convolution operator. Specific details about this procedure and test cases can be found in the studies by Meyerholtz et al. (1989), Chiao et al. (2006) and Hung et al. (2004, 2011). Eq.(4) shows that CQS regularization amounts to a special type of weighted damped least squares, in which Q is comparable to classical roughness regularizations which minimize the first spatial derivative or the Laplacian operator. However, in the CQS formulation, the roughness parameters are now explicit; their values correspond to the widths of the smoothing function. Fig. 4 shows a comparison between the two regularization schemes in terms of trade-off between model norm and variance reduction, and the effect on the tomographic images at 75 km depth. Our tests using y between 2 and 500 show that inversions using DLS and CQS schemes (with a smoothing width of 40 km) yield similar results (Fig. 4b and d), with a slightly higher data variance reduction for DLS if y isthesame(fig. 4a andc). However, the CQS procedure emphasizes longer wavelength features (Fig. 4b and d). The solutions discussed in this paper (Fig. 6) were obtained using a damping parameter (y ¼ 25) and convolution quelling with a smoothing width of 40 km (as used to obtain Fig. 4d). The variance reduction obtained from this regularization is 82% (Fig. 4c), from an initial variance of s 2 to a final variance of s Checkerboard resolution tests Fig. 5 shows an example of checkerboard resolution tests we performed to evaluate the data coverage and inversion technique. The checkerboard input models are set up as follows: in each direction, the input velocity anomaly is of Gaussian type over three grid intervals with a maximum amplitude of þ2% at the center, is zero for the next two grid intervals and then repeats ð4þ with the sign reversed. The width of each anomalous region is 150 km in each direction. The synthetic travel times are calculated by multiplying the sensitivity kernel matrix evaluated for the real data set and the input model. For the inversion of the synthetic travel times, we used the same damping parameters, sources and station locations as used in the inversion of the real data set. The synthetic checkerboard anomalies are recovered beneath much of the Carpathian Pannonian region throughout the upper mantle as indicated in Fig. 5. Distortion of the regular anomalies increases gradually with depth and we find that the recovered peak velocity anomaly decreases systematically from 85% at 75 km to 55% at 525 km. At depths less than about 45 km, the model is poorly resolved because ray paths arrive almost vertically below the stations. In the real data inversion variations in this near surface layer are accounted for by the station terms. At depths below the mantle transition zone (700 km), our models are not well resolved everywhere and smearing is significant, so we do not interpret structures in the lower mantle. 3. P-wave velocity model: SCP_P1 Fig. 6 shows depth slices through our preferred P-wave velocity model for the Carpathian Pannonian region. While our tomographic model agree in general with the results from previous tomographic studies (e.g. Fan et al., 1998; Wortel and Spakman, 2000; Martin et al., 2006; Koulakov et al., 2010; Dando et al., 2011), the addition of the SCP data set enables much better resolution, especially beneath eastern Hungary and Romania. Here we focus on the observed structures at different depths beneath two important areas: the Eastern Alpine/Pannonian region and the Transylvania/Vrancea zone region. Above 250 km depth, P-wave model shows predominantly low velocity zones beneath the Pannonian regions and the Transylvanian Fig. 7. Vertical sections through the P-wave tomographic model of SCP_P1, for the Eastern Carpathians (1, 2 and 3), the Vrancea Zone (4) and the Eastern Alps Southern Apuseni Mountains (5).

10 147 region (see Fig. 6a and b). The most prominent low velocity anomalies are those beneath the Western Carpathians and the eastern Pannonian basin (Békés and Makó sub-basins); the latter were also observed in the tomographic models of Dando et al. (2011), using a sub-set of our data. The Western Carpathian low-velocity anomaly clearly extends beneath the Vienna basin at depths of km (see Fig. 6a). Recent seismic tomographic studies by Chang et al. (2010) and Dando et al. (2011) also reported generally slow velocities down to at least 200 km beneath the Pannonian basin. These lowvelocity anomalies coincide with high heat flow associated with rifting depocentres (Tari et al., 1999). The slow anomaly beneath the Western Carpathians is also associated with the Neogene Central Slovakian calc-alkaline volcanic domains (Kovács and Szabó, 2008). Another broad slow anomaly is observed clearly beneath the Transylvanian basin and the eastern half of the Eastern Carpathian arc, where it extends down to 300 km (Fig. 6a c and cross-sections 1 3 in Fig. 7). The Eastern Carpathians are also characterized by magmatic activity since the Miocene, as summarized by Seghedi et al. (2004). Geochronological measurements suggest migration of the locus of volcanism from NW to SE along the Eastern Carpathian arc in Romania since 10 Ma (Szabó et al., 1992; Csontos, 1995; Pécskay et al., 1995a, 1995b, 2006; Szakács and Seghedi, 1995; Mason et al., 1998; Seghedi et al., 1998). Our inversions also reveal an intense localized slow anomaly to the East of the Vrancea region in the upper 200 km (Fig. 6a c). This feature may be associated with the deep trough of sediments in the Foc-sani Basin. At depths below about 100 km, a broad fast anomaly is evident beneath the eastern Alps in the P-wave velocity models (Fig. 6b d) as found by Mitterbauer et al. (2011). Below about 300 km this fast anomaly extends across the Pannonian region (Fig. 6e and cross-section 5 in Fig. 7). This structure was also observed by Dando et al. (2011), who noted its along-strike continuity with the downwelling slab beneath the Eastern Alps. At 412 km, this alignment is strongly developed in our P-wave model due to a concentration of fast material now resolved beneath the Apuseni Mountains, though its continuity is interrupted beneath the Great Hungarian Plain. Beneath the Vrancea zone, the P-wave velocity models show a localized fast anomaly approximately 300 km 200 km in area extending from 100 km in depth to about 400 km, and trending SW oblique to the southern edge of the South Carpathians (Fig. 6b d, and crosssection 4 in Fig. 7). A localized fast anomaly beneath Vrancea was also observed in previous seismic tomographic studies (Wortel and Spakman, 2000; Martin et al., 2006; Koulakov et al., 2010) and coincides with the zone of intermediate depth seismicity (e.g Tondi et al., 2009). This high-velocity structure has been interpreted as a remnant of subducted oceanic lithosphere (Sperner et al., 2001; Martin et al., 2006) or as delaminated mantle lithosphere (Koulakov et al., 2010; Knapp et al., 2005). Our images show that the intermediate-depth seismicity is aligned with the NE end of this fast anomaly at 100 km depth, but lies directly above a more axi-symmetric fast anomaly at km depth (see Fig. 8). In the mantle transition zone, we observe a fast P-wave anomaly beneath the entire Pannonian region with a ring-like appearance (Fig. 6f). Regional tomographic studies (e.g. Wortel and Spakman, 2000; Piromallo and Morelli, 2003) have also reported this broad fast anomaly in the mantle transition zone and associated it with remnant oceanic slabs which have subducted along the Inner-Carpathian arc. Our results, based on many more data, have provided significantly improved resolution of this anomaly. 4. Synthetic block models Fig. 8. Horizontal slices through the P-wave tomographic model in the Vrancea region at depths: 112 km, 150 km 188 km, 225 km, 262 km, 338 km, 412 km and 525 km. Green dots in different maps show the earthquakes occurring at corresponding depths, and the green ellipse represents a downward projected contour showing the horizontal distribution of all intermediate-depth seismicity. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) In order to further assess the validity of our tomographic models, we have also conducted tests using simple synthetic block models (Fig. 9) which approximate the structures observed in the real data inversion (Figs. 6 and 7). The inversion uses the same damping parameters, sources and stations as used for the observed data. Our synthetic block model consists of the following: (1) a tabular fast anomaly between 75 km and 400 km depth beneath Vrancea region; (2) an elongated slow anomaly between 75 km and 300 km depth beneath the Eastern Carpathians; (3) a fast anomaly between 75 km and 400 km beneath the Eastern Alps; (4) a broad fast anomaly in the mantle transition zone beneath the Pannonian region; and (5) a thin slow velocity layer directly below the center of the MTZ fast anomaly. Actually, the configuration of this synthetic block model is similar to the test

11 148 Fig. 9. Resolution tests using a simplified synthetic block model inspired by observed structures in the P-wave inversions of Fig. 6. (a) Horizontal slices of the input (left) and retrieved (right) P-wave synthetic models at 112 km, 225 km, 338 km and 600 km. (b) Vertical sections through input and retrieved P-wave synthetic models corresponding to the five sections shown in Fig. 7. presented by Dando et al. (2011). The amplitudes of input anomalies are þ2% and 2% for fast and slow structures, respectively. In general, all the synthetic anomalies are well reproduced by the inversion in terms of locations and shapes (Fig. 9). The amplitude of recovered anomalies is typically between 50% and 80% of the input values. Minor smearing along ray paths is evident, but the recovery of the input model provides us with confidence in the model obtained from the real data inversion. With increasing depth, both lateral and vertical smearing increases, especially beneath the Eastern Carpathians and the central Pannonian Basin (see maps at 338 and 600 km in Fig. 9a, and sections 1 3 in Fig. 9b). 5. Discussion Our P-wave tomographic images show well-resolved structures in the upper mantle beneath the Carpathian Pannonian system and provide important clues to the geodynamic processes that have shaped this region. In this section, we discuss the implications of these results for geodynamic models of the region. In particular, we consider: (1) evolution of the Eastern Alps and genesis of the Pannonian Basin, (2) the evidence for subduction having occurred beneath the Eastern Carpathians, and (3) development of the high-velocity structure associated with the intermediate-depth seismicity beneath the Vrancea zone Structures beneath the Eastern Alps and Pannonian basin Lippitsch et al. (2003) interpreted a northeastward dipping fast anomaly beneath the eastern Alps as Adriatic lithosphere subducted beneath the European plate. Recent studies by Mitterbauer et al. (2011) and Dando et al. (2011) show that the eastern Alpine fast anomaly in the mantle is sub-vertical and persists down to about 400 km. Our tomographic images confirm these results, but the increased aperture of our array enables us to see that the eastern Alpine fast anomaly extends right down into the transition zone (Fig. 6f). Mitterbauer et al. (2011) explained this structure in terms of delamination, then subduction of the continental European lower lithosphere during ongoing convergence between Adria and Europe. At least in the mantle, this structure does not have a clear N or S vergence and could be explained by a symmetrical downwelling of both European and Adriatic mantle lithospheres. The eastern Alpine orogenic zone is topographically high and is characterized by both thickened crust and lithosphere. Lillie et al. (1994) modeled the gravity signature of the lithosphere in the Eastern Alps and pointed out that the present topography is not sufficiently high to compensate isostatically for the crustal root. They proposed that an excess mass provided by the protrusion of relatively dense lithosphere into the asthenosphere is necessary to balance the compensation. Thus both gravity signature and seismic structure are consistent with the presence of a sub-vertical, relatively dense and seismically fast mantle body extending downward beneath the Eastern Alps.

12 149 Dando et al. (2011) used data from the Carpathian Basins Project to show that the Eastern Alpine fast anomaly extends eastward across the Pannonian Basin at depths below about 300 km, and merges downward into the fast transition zone beneath the Pannonian region. Our tomographic images using more data and different tomographic techniques confirm the existence and depth extent of this structure (Fig. 6d e, Fig. 7; Section 5) and furthermore link it to fast material beneath the Apuseni Mountains region of western Romania. At 412 km the along-strike continuity of this band of fast material which stretches from the Alps to the Apuseni Mountains and then southward beneath the Balkans is interrupted only briefly by a loss of amplitude beneath the Great Hungarian Plain. We interpret this extended high velocity structure between 300 and 400 km depth as the remnant of mantle downwelling beneath a pre-pannonian convergent belt which resembled the Eastern Alps but extended right across the Pannonian region before the basin extended. This structure evidently records the most recent phase of mantle downwelling beneath the Pannonian Basin, and its detachment from the lithospheric convergent zone may have been the trigger for mid-miocene extension of the Pannonian Basin (Dando et al., 2011). Below 400 km, the volume of fast material occupying the MTZ beneath the Pannonian region, probably signifies a sustained period of subduction (e.g., Handy et al., 2010) followed by Alpine-style convergent downwelling. This fast material appears to have spread outward from beneath the center of the basin, displacing slower material outward to create a front aligned approximately with the Carpathian Arc separating the fast Pannonian MTZ from slower material beneath the Bohemian Massif, the West Carpathians and the Moesian Platform. Our synthetic block model (Fig. 9) supports the interpretation of Dando et al. (2011) that the relatively slower central part of this fast Pannonian anomaly in the MTZ is an artifact arising from depression of the 660 km discontinuity beneath the Pannonian Basin, consistent with the interpretation of receiver functions (Hetényi et al., 2009) The mode of basin extension and evidence for Carpathian subduction In general, the Pannonian basin is characterized by both thinned crust and lithosphere ( km for crustal thickness and km for the thinnest part of the lithosphere; Babuška et al., 1987; Horváth, 1993; Lenkey et al., 2002; Horváth et al., 2006), and anomalously high heat flow (4100 mw m 2 ; Tari et al., 1999). The driving force in the extension of the Pannonian Basin is commonly attributed to slab roll-back with northeastward retreat of the Eastern Carpathians until about 10 Ma (Royden et al., 1983a, 1983b; Horváth, 1993; Linzer, 1996; Horváth et al., 2006), followed eventually by progressive slab detachment beneath the Eastern Carpathians (Wortel and Spakman, 1992, 2000). Huismans et al. (2001) summarized the Neogene subsidence and sedimentation history and pointed out that the extension occurred in two phases: a first phase where the crust and lithosphere have been thinned equally with a stretching factor of (Royden et al., 1983b; Lankreijer et al., 1995); and a second phase where the stretching factor of the lithosphere is locally as high as 4 8 without additional thinning of the crust (Sclater et al., 1980; Royden et al., 1983b; Lankreijer et al., 1995; Horváth et al., 2006). Although the first phase of extension could be attributed to a slab roll-back mechanism, the later phase, in which the mantle lithosphere is strongly extended, more likely is driven by asthenospheric upwelling (Huismans et al., 2001). Houseman and Gemmer (2007) proposed indeed that the lithospheric extension might have been driven entirely by gravitational instability involving a diffuse upwelling beneath the Pannonian Basin and focussed downwelling beneath the Carpathian arc. The generally slow seismic velocities of the Pannonian upper mantle support the idea that mantle upwelling has occurred beneath the basin, but the above mentioned models clearly do not capture the complexity of the process that is evident in our tomographic images. Our images (Fig. 6) show that, at 75 km, slow material is localized in several regions, notably beneath the Western Carpathians, the Eastern Carpathians, and the deep basins of the Great Hungarian Plain. At these levels, mantle upwelling is evidently localized in regions that either are associated with the basin depocentres and extreme lithospheric thinning (Tari et al., 1999), or are associated with post-extensional volcanic centers where fluids or melts might explain the low velocities (Green et al., 2010). Above about 250 km (Fig. 6b and c) the Pannonian region is generally slow, but the slowest velocities are actually found beneath the Carpathian arc, and indeed just outside the arc, associated with the Foc-sani Basin of Romania. If roll-back of a subducted slab has occurred beneath the Eastern Carpathians, there remains no clear evidence of that slab in the upper mantle (above 410 km) from the tomographic images presented here. Recent numerical experiments by van Hunen and Allen (2011) imply that a cold mantle signature will persist where downwelling previously occurred, even if the major mass anomaly has detached from the lithosphere. Our images actually show slow velocities in the upper mantle beneath the Eastern Carpathians (Fig. 6b and c and Fig. 7); evidently any subducted slab (e.g. Wortel and Spakman, 2000) or fast material developed from gravitational instability of the continental mantle lithosphere (Houseman and Gemmer, 2007) has fallen rapidly and completely into the MTZ, contrary to the progressive slab detachment proposed by Wortel and Spakman (2000) or Sperner et al. (2001). The evidence for surface convergence across the Eastern Carpathians is of course compelling, as summarized recently by Mat enco et al. (2010), which leaves the essential question of why the expected fast signature of downward mantle flow is generally not present beneath the Carpathians? Two explanations seem possible: the Carpathian convergence may have occurred above the relatively fast trans-pannonian anomaly (described in the preceding section), before detachment occurred and the Carpathian orogen was displaced north-eastward. Alternatively, one might argue that any shallow fast anomaly beneath the Carpathians has been recently obliterated, possibly by upward migration of hot fluids released on the periphery of the major MTZ anomaly (Fig. 6f) Intermediate-depth seismicity in Vrancea zone Three distinct geodynamical mechanisms have been advanced to explain the intermediate-depth seismicity in the Vrancea zone: (1) slab tearing or break-off (Wortel and Spakman, 1992, 2000; Nemcok et al., 1998; Sperner et al., 2001); (2) gravitational instability (Lorenczi and Houseman, 2009); and (3) delamination of mantle lithosphere (Gîrbacea and Frisch, 1998; Gvirtzman, 2002; Knapp et al., 2005; Fillerup et al., 2010). Slab break-off as described by Wortel and Spakman (2000) is the final stage of a subduction process beneath the Eastern Carpathian mountain range. After the oceanic slab has been consumed by subduction, the dense slab progressively separates from the surface, from north to south beneath the Eastern Carpathian arc. The observation that calc-alkaline and alkaline volcanism occurred in the Eastern Carpathians from about 14 Ma until 2 Ma, with a gradual southward migration of the eruptions along the chain (Pécskay et al., 1995a; Mason et al., 1998) has also been interpreted in support of progressive slab detachment. A northwest to southeast shift from normal subduction related calc-alkaline affinity to more diverse compositions at 3 Ma has been attributed by

13 150 Seghedi et al. (2011) to steepening of the slab and opening of a slab window. If the break-off had occurred progressively, however, we should see the signature of a remnant slab in the upper mantle north of the Vrancea zone, where break-off would have occurred most recently. While our P-wave velocity models show clearly the fast anomaly beneath and extending to the SW of the Vrancea zone, there is no evidence for a remnant subducted slab beneath the adjacent Eastern Carpathians at any depth above the transition zone. In fact the inner-carpathian region, north of Vrancea, is dominated by slow anomalies (see cross-sections 1 3 of Vp models in Fig. 7). Our resolution tests using a slab-like structure along the inner Eastern Carpathian arc show that the region is well resolved in our tomographic models (Fig. 9). An alternative explanation of the Vrancea tectonic activity suggests that the mantle lithosphere is in the process of delamination (Gîrbacea and Frisch, 1998; Gvirtzman, 2002; Knapp et al., 2005; Fillerup et al., 2010). Under continental collision conditions, delamination could be initiated by an anomalously dense and over-thickened lithospheric mantle (Houseman and Molnar, 1997; Schott and Schmeling, 1998; Morency and Doin, 2004; Göǧüs- and Pysklywec, 2008). An alternative trigger for delamination can arise from metamorphic transformation of mafic lower crust into eclogite (Rudnick and Fountain, 1995) at depths of around km. An eclogitic lower crustal root is gravitationally unstable and can drive downward flow which removes the mantle lithosphere (Elkins-Tanton, 2007; Fillerup et al., 2010). Our NW-SE section across the Vrancea zone (Fig. 10) shows clearly the P-wave fast anomaly, which is nearly vertical from 80 km to 400 km. Wortel and Spakman (2000) showed a similar image of this Vrancea fast anomaly apparently connected to high wave speed lithosphere to the SE which encouraged the interpretation of a subducting or delaminating lithospheric slab. However, such a connection is not evident in our model; instead this fast anomaly is bounded by two slow regions dipping to the NW and SE (Figs. 8 and 10). While the observed slow anomaly on the NW side could be associated with hot asthenosphere that has flowed upward to fill the gap induced by the delaminated lithosphere, the slow anomaly on the SE side is difficult to explain using the delamination model. The slow anomalies, however, are more easily explained with the mechanism proposed by Lorinczi and Houseman (2009). In that model the fast anomaly corresponds to a mass of continental mantle lithosphere which has formed in a drip-like gravitational instability, with slow anomalies on the NW and SE sides caused by adjacent asthenospheric upwellings. This drip model has the advantage that it provides a quantitative explanation of the distribution of seismic moment release rate in the deep seismic zone (Lorinczi and Houseman, 2009). Our images show that the seismicity at depths of around km occurs at the NE end of a high velocity region which extends at shallow depths to the southwest (see Fig. 8); below about 250 km the high velocity material lies almost directly beneath the seismogenic zone. This dense material presumably provides the driving force for the high rates of seismic moment release. Our images show no connection between the Vrancea anomaly and the high velocity material in the mantle transition zone (Fig. 10). To our mind, this implies that the Vrancea structure is a recent feature and post-dates the events that produced the fast-anomaly in the mantle transition zone and the extension of the Pannonian Basin. 6. Conclusions Our high-resolution P-wave velocity model of the upper mantle beneath the Carpathian-Pannonian region, obtained using finite-frequency tomography of teleseismic P relative arrival time residuals results from an unusually large, dense, broadband array: 54 stations deployed in the South Carpathian Project (SCP, ), 56 stations deployed in the Carpathian Basins Project (CBP, ) and 131 national network stations. Our tomographic images therefore are better resolved than previous studies in the eastern Pannonian and Transylvanian regions. Images from our P- wave model show that fast material beneath the eastern Alps, extends down into the transition zone and eastward across the Pannonian Basin below 300 km depth. This fast material extends downward into the mantle transition zone (MTZ) and appears to spread outward beneath the entire Pannonian basin. The fast region in the MTZ is clearly bounded by slower material beneath the Western Carpathians and the Moesian platform. Above 300 km, the upper mantle below the Pannonian basin is dominated by relatively slow velocities, the largest of which extends down to nearly 200 km and underlies the 47 km thick sediments of the Makó Békés rift basins. The Alpine anomaly supports a model of mantle downwelling initiated by continental collision but probably driven by gravitational instability beneath the present-day eastern Alps. We interpret that similar downwelling occurred beneath the Pannonian Basin, before slab detachment triggered the lithospheric extension process that produced the Pannonian Basin. In the Vrancea region, the seismicity occurs at the NE end of a high velocity structure that extends SW in the upper 200 km, oblique to the southern edge of the South Carpathians; this structure dips to the NE and becomes more axi-symmetric to depths of 400 km. The Vrancea structure is broadly consistent with models based on either delamination of mantle lithosphere or lithospheric gravitational instability occurring beneath the SE corner of the Carpathians. We find no clear evidence of residual slabs beneath the Eastern Carpathians in the upper 400 km, contrary to models for the evolution of the Pannonian Basin that depend on the idea of slab roll-back and progressive slab detachment beneath the Eastern Carpathians. Acknowledgments Fig. 10. NW SE profile across the Vrancea Zone through our P-wave tomographic model (same as section 4 in Fig. 7 with geodynamic interpretation). Green dots show the projected locations of intermediate-depth earthquakes in the Vrancea region. The South Carpathian Project was supported by NERC standard Grant NE/G005931/1. The seismological equipment used in the SCP network was provided by SEIS-UK as the NERC Geophysical