New insights from seismic tomography on the complex geodynamic evolution of two adjacent domains: Gulf of Cadiz and Alboran Sea

Abstract

[1] In this study, we present a three-dimensional P wave upper-mantle tomography model of the southwest Iberian margin and Alboran Sea based on teleseismic arrival times recorded by Iberian and Moroccan land stations and by a seafloor network deployed for 1 year in the Gulf of Cadiz area during the European Commission Integrated observations from NEAR shore sourcES of Tsunamis: towards an early warning system (EC NEAREST) project. The three-dimensional model was computed down to 600 km depth. The tomographic images exhibit significant velocity contrasts, as large as 3%, confirming the complex evolution of this plate boundary region. Prominent high-velocity anomalies are found beneath Betics-Alboran Sea, off-shore southwest Portugal, and north Portugal, at sublithospheric depths. The transition zones between high- and low-velocity anomalies in southwest and south Iberia are associated to the contact of oceanic and continental lithosphere. The fast structure below the Alboran Sea-Granada area depicts an L-shaped body steeply dipping from the uppermost mantle to the transition zone where it becomes less curved. This anomaly is consistent with the results of previous tomographic investigations and recent geophysical data such as stress distribution, GPS measurements of plate motion, and anisotropy patterns. In the Atlantic domain, under the Horseshoe Abyssal Plain, the main feature is a high-velocity zone found at uppermost mantle depths. This feature appears laterally separated from the positive anomaly recovered in the Alboran domain by the interposition of low-velocity zones which characterize the lithosphere beneath the southwest Iberian peninsula margin, suggesting that there is no continuity between the high-velocity anomalies of the two domains west and east of the Gibraltar Strait.

1 Introduction

[2] The geodynamic evolution of the southwestern Iberian margin and the Alboran Sea is the result of the complex interaction between the African and the Eurasian plates (Figure 1). At present, there is convergence between the two plates at a rate of ~5 mm/year [Stich et al., 2006]. The plate boundary is clearly defined from the Gloria fault to the Gorringe Bank [Zitellini et al., 2009]. From the Gorringe Bank proceeding to the east, across the Strait of Gibraltar, the boundary is diffuse [Sartori et al., 1994] with different locations having been proposed for it [e.g., Rovere et al., 2004]. A narrow band of deformation (SWIM Fault Zone, SFZ in Figure 1) is considered as a precursor to the formation of a new transcurrent plate boundary between Iberia and Africa [Zitellini et al., 2009].

Figure 1

Map of the Gulf of Cadiz and surrounding regions showing seismicity related to the main tectonic features and the NEAREST land stations and marine network (green and purple triangles, respectively). The marine network includes 24 OBS plus GEOSTAR seafloor observatory. Epicentral locations are from the bulletin of the Institute of Meteorology Lisbon (http://www.meteo.pt/en/publicacoes/tecnico-cientif/noIM/boletins/). CSV = Cape St. Vincent; S = Spain; P = Portugal; MO = Morocco; SFZ = Swim Fault Zone. In inset, GF = Gloria Fault. Adapted from Geissler et al. [2010].

[3] Two main geodynamic evolution models have been proposed so far in literature for the Alboran Basin System:

1. [4]

   The collision of Europe and Africa led to lithospheric thickening during the Paleogene. The thickened continental lithosphere was later (~20 Ma) detached by convective removal [Platt and Vissers, 1989] or by delamination [Seber et al., 1996; Calvert et al., 2000]. The collapse of this lithosphere caused extension of the Alboran basin and uplift around the margin.

2. [5]

   The subduction of oceanic lithosphere caused extension within the Alboran Basin in the Miocene by slab rollback [Royden, 1993; Lonergan and White, 1997; Bijwaard and Spakman, 2000; Gutscher et al., 2002] or by slab detachment [Zeck, 1996]. Starting from their geometrical and structural similarity, a common origin has been proposed for the Calabrian and Gibraltar Arcs. In this reconstruction, the Arcs formed from the fragmentation of a single subduction zone during the tectonic evolution of the Central-Western Mediterranean region [Faccenna et al., 2004 and references therein].

[6] Different geophysical observations have been interpreted within the proposed geodynamic models. Magma evolution in time and space shows a volcanism which could be explained by a subduction scenario [Duggen et al., 2004], but it has also been invoked in support of delamination [Zeck, 1996] and of the convective removal models [Platt and Vissers, 1989; Platt, 1998; Turner et al., 1999]. Tomographic studies show a high-velocity body under the Alboran Sea, which has been interpreted as a continuous subducting slab [Gutscher et al., 2002; Piromallo and Morelli, 2003], as a broken-off slab [Blanco and Spakman, 1993], and as lithosphere which has undergone delamination [Calvert et al., 2000]. The pattern obtained from seismic anisotropy studies supports models involving a westward retreating slab [Buontempo et al., 2008; Bokelmann et al., 2010; Diaz et al., 2010]. Measures on dispersion of P waves are compatible with subducted oceanic lithosphere [Bokelmann and Maufroy, 2007].

[7] The two domains, Alboran Sea and Atlantic domain (Gulf of Cadiz), east and west of the Gibraltar Strait, show differences in their geophysical and seismological characters. In the Alboran domain, Bouguer anomaly modeling shows that the base of the lithosphere ranges from 140 km depth in the Gibraltar Strait to less than 45 km depth in the easternmost Alboran Sea [Tornè et al. 2000; Zeyen et al., 2005]. The average heat flow in the western Alboran basin is 69 ± 6 mW/m² with a generally increasing trend toward the center of the basin and to the east. In contrast, the heat flow pattern in the eastern Alboran Basin shows an average value of 124 ± 8 mW/m², which remains rather constant over the whole area [Polyak et al., 1996]. Moderate superficial seismicity and a seismic gap between 20 and 60 km depth has been observed in the Alboran Sea (Figure 2). There is intermediate-depth seismicity, located near the Strait of Gibraltar within 120 km depth and distributed in a very narrow north-south oriented vertical band [Seber et al., 1996]. Deep earthquakes are rare and occur at the bottom of the transition zone (~630 km depth) under southern Spain in the Granada area [Buforn et al., 2004], and no Benioff zone is visible. Under the Alboran Sea and southern Spain, mostly extensional and strike-slip focal mechanisms are reported and a combination of seismic and GPS data points to a transtensional regime [Stich et al., 2003, 2006].

Figure 2

Map of seismic stations used for the tomographic inversion. Stars represent marine stations from the NEAREST array (24 OBS/H, GEOSTAR) and two broadband land stations (MESJ, PDRG) installed in southern Portugal during the 2007–2008 experiment. The integrated array consists of 111 receivers. Colored dots represent subcrustal seismicity at different depth intervals. Land stations are part of the permanent networks of Spain (blue squares), Portugal (green circles), and Morocco (red triangles). Hypocentral locations are from http://www.02.ign.es/ign/layoutIn/sismoFormularioCatalogo.do (Instituto Geográfico Nacional Spain). GR = Granada area; CSV = Cape St. Vincent; GC = Gulf of Cadiz; SAP = Seine Abyssal Plain; HAP = Horseshoe Abyssal Plain; TAP = Tagus Abyssal Plain. AA′, BB′, CC′, DD′, and EE′ show the traces of the profiles shown in Figure 9. The inset shows the four regions used for the crustal correction of the traveltime residuals.

[8] In the Atlantic domain, heat flow is high above the Horseshoe Abyssal Plain (approximately 52–59 mW/m²), and it decreases eastward across the Gulf of Cadiz prism, attaining minimum values (~40 mW/m²) toward the Gibraltar Strait [Grevemeyer et al., 2009]. The low values and trend of the heat flow are typical for active thrusting and could be explained by an east-dipping subduction model, which may or may not be active [Grevemeyer et al., 2009 and references therein]. There is evidence in the Gulf of Cadiz for subcrustal earthquakes down to 60 km depth. A recent study based on the NEAREST marine array data [Geissler et al., 2010] shows that the observed seismicity is contained in a seismogenic layer of ~140 Ma old oceanic lithosphere and aligned along north-northeast–south-southwest and west-northwest–east-southeast striking structures. In the Gulf of Cadiz, repeated GPS measurements and focal mechanisms, which are mainly reverse and strike slip, point to a transpressive plate boundary [Stich et al., 2005; Geissler et al., 2010]. These studies show that there is a change of stress regime going across the Gibraltar Strait from transtensional in the Alboran domain to transpressive in the Atlantic domain.

[9] Tomographic models proposed so far do not give much information on the upper-mantle structure west of Gibraltar Strait due to the lack of long-term data series at the sea bottom. In particular, a clear image of the transition between the African and the Eurasian continental domain and the Atlantic oceanic domain is still lacking. Knowledge of the Atlantic domain and of the passage to the Alboran domain can also put some constraints on the models describing the geodynamic evolution of the Alboran domain. An open question is the present relation of the Atlantic and Alboran domains.

[10] We present a three-dimensional (3-D) tomography model imaging the upper-mantle structure below the Gulf of Cadiz-Alboran region. The inclusion, for the first time, of the Atlantic domain was possible, thanks to data recorded during a long-term (1 year) marine experiment in the Gulf of Cadiz in which 24 Ocean Bottom Seismometer (OBS) and GEOSTAR seafloor observatory were deployed, in the frame of the activities of the EC NEAREST project.

2 Data Set

[11] During the NEAREST experiment [Carrara and NEAREST Team, 2008], a seafloor seismic network was deployed for 1 year (August 2007–August 2008) in the Gulf of Cadiz and off-shore Cape St. Vincent, with average spacing of ~50 km (Figure 1). This network comprised 24 stations (OBS/H) from the German DEPAS pool, each equipped with a three-component Guralp CMG-40 T broadband seismometer with 60 s-50 Hz response (OBS) and a High Tech HTI-04-PCA/ULF hydrophone (OBH). The same type of seismometer was installed on board the GEOSTAR deep-sea multiparameter observatory deployed in the same period. Furthermore, two broadband land stations were installed in southern Portugal at the beginning of the experiment (Table S1). The OBS/H internal clocks were synchronized with GPS signal before the deployment, and the same operation was repeated on ship after the instruments were recovered. A linear drift correction was applied a posteriori, which was, on the average ~0.53 s/year. For nine OBS/H, onboard synchronization was not possible, and we applied the drift corrections calculated during other experiments where the same instruments were deployed. Figure 2 shows the seismic stations used in the inversion for the velocity structure.

[12] Direct P phases from 67 teleseismic events were identified on waveforms recorded by 25 seafloor stations, the two temporary land stations, and by some other land stations when digital seismograms were available from the European Integrated Data Archive (EIDA; http://eida.rm.ingv.it). This data set was selected from more than 200 M_w ≥ 5.5 events, which occurred during the experiment, with epicentral distance between 25° and 95°. To obtain better ray coverage, we also added 89 teleseisms recorded by the Moroccan, Portuguese, and Spanish seismic networks (Figure 2) in the period January 2007–June 2009. The corresponding arrival times were extracted from the ISC bulletin (http://www.isc.ac.uk), which also includes readings of later arrivals as pP, sP, PcP, and core phases as PKPdf. Our teleseismic data set consists of 152 events (130 P, 16 pP, 3 sP, and 3 PcP) with at least 10 P wave recordings, for a total of 6238 P wave arrival times for the tomography. Figure 3 shows the azimuthal distribution of the teleseismic sources with the epicentral locations provided by the National Earthquake Information Center (NEIC) earthquake catalogs (http://earthquake.usgs.gov/research).

Figure 3

Azimuthal equidistant map of teleseismic sources (projection center at 37.0°N–7.0°W) selected for the analysis. The events are distributed over a wide range of back azimuths. Epicentral distance ranges from 30° to 100°. Red circles represent the events used for the residual maps in Figure 5.

[13] To estimate cross-correlation delays of the first P-arrival on the waveform data we employed the semi-automated multichannel cross-correlation (MCCC) technique (VanDecar and Crosson, 1990) on the vertical channel of the unfiltered data (Figure 4). MCCC calculates relative phase arrivals by maximizing a cross-correlation matrix. After a first visual inspection, there is a preliminary selection of waveforms, and then MCCC is applied. We estimated an average error for MCCC data of 0.17 s. This estimation is 1.5–2 times the error computed by the MCCC routine. We took this more conservative approach as MCCC is known to underestimate the error (down to a fraction of the sampling interval) [Tilmann et al., 2001]. For bulletin data, mostly from impulsive P phases, we estimated an average arrival time error of 0.45 s. The entire data set of the 6238 P wave arrival times has an average error of 0.43 s, corresponding to a data variance (noise) of 0.18 s². OBS data with very low signal-to-noise ratio and time picks yielding absolute residuals of more than 10 s or relative residuals (see below) exceeding ±2.0 s were excluded from the data set. These thresholds were included to mitigate the effect of large data errors (outliers) often associated to incorrect phase identification or mispicking. Data inversion with a 2 s cutoff applied to the traveltime residuals can still take into account significant differences between the real Earth and the standard model.

Figure 4

Example of waveforms used for MCCC analysis for (a) M6.4 event from China and (b) M6.3 from Chile. Vertical bars identify P wave first break. Other examples of MCCC are shown if Figure S1 in the supporting information.

3 Relative Arrival Time Residual Analysis

[14] Relative residuals are widely used in regional arrival time teleseismic tomography to constrain the 3-D seismic structure of a local model volume beneath a seismic array. The main reason for using this methodology is that relative residuals are not significantly affected by source mislocations (errors in origin time and hypocentral location), differences between the actual and the reference source-receiver raypath below the investigated Earth volume nor by large-scale variations in mantle structure [Aki et al., 1977; Evans and Achauer, 1993]. Their inversion, however, does not produce absolute velocity, but only relative perturbations calculated starting from an initial velocity model. In other words, the resulting velocity structure only reflects deviations about some unknown average earth model, as the layer-average velocities are not constrained by the inversion and act as free parameters [Koch, 1985].

[15] In this study, we computed relative residuals on event per event basis by subtracting a weighted average residual from each absolute traveltime residual. The absolute residual, for a given seismic phase, is the difference between the observed and the predicted arrival time. The weighted average event residual was obtained using data weights 1/σ²_pick, where σ_pick is the error associated to each arrival time. For those events composed by both MCCC and ISC data, this computation was carried out separately for each subset of teleseismic arrivals. This separation takes into account the likely possibility that the data subsets may have different baselines. We found that this correction improves the data fit.

[16] Before the residual computation, the observed arrival times were corrected for Earth's ellipticity [Dziewonski and Gilbert, 1976] and station topography/bathymetry (negative for the seafloor instruments), while the predicted traveltimes, based on the ak135 global velocity model [Kennett et al., 1995], were corrected for sedimentary and crustal structure. Four velocity models were applied for the crustal correction (Table 1). These models were extracted from recent studies and from the European Moho depth map by Grad et al. [2009]. The oceanic model, in particular, takes into account the NEAREST wide-angle seismic survey [Sallarès et al., 2011]. We consider these models as representative of the different average structure of the continental and oceanic domains within the study area (see inset in Figure 2). For a given teleseismic ray, the crustal correction was computed as the difference between the traveltime in the local structure and the one in the ak135 velocity model, backtracking the ray from the Earth's surface down to the reference depth of 50 km. Values of the crustal terms range from −0.3 s for the land stations to 0.20 s for the rays ending at the seafloor receivers.

Table 1. Crustal Velocity Models Used for the Different Sectors Shown in the Inset of Figure 2a^a I, Gulf of Cadiz (Atlantic Ocean Domain) from Gutsher et al. [2009], Fullea et al. [2010], and Sallares et al. [2011]; II, Iberian Peninsula from Diaz and Gallart [2009] and references therein; III, Betics and Rif from Serrano et al. [2003], Fullea et al. [2007], Diaz and Gallart [2009], and Fullea et al. [2010]; IV, Atlas from Zeyen et al. [2005].

Depth (km)	P wave velocity (km/s)	Layer
I
0	3.50	Sediment
6	5.80	Upper crust
12	6.50	Lower crust
16	8.00	Upper mantle
II
0	6.00	Upper crust
10	6.25	Middle crust
24	6.75	Lower crust
31	8.00	Upper mantle
III
0	6.00	Upper crust
12	6.40	Middle crust
25	6.80	Lower crust
33	8.00	Upper mantle
IV
0	6.00	Upper crust
12	6.30	Middle crust
24	6.60	Lower crust
38	8.00	Upper mantle

• ^a I, Gulf of Cadiz (Atlantic Ocean Domain) from Gutsher et al. [2009], Fullea et al. [2010], and Sallares et al. [2011]; II, Iberian Peninsula from Diaz and Gallart [2009] and references therein; III, Betics and Rif from Serrano et al. [2003], Fullea et al. [2007], Diaz and Gallart [2009], and Fullea et al. [2010]; IV, Atlas from Zeyen et al. [2005].

[17] The pattern of relative arrival time residuals provides a first indication of the heterogeneity in the Earth's velocity structure inside the study volume. Negative residuals imply the presence of high-velocity anomalies, whereas positive values indicate low-velocity structures. In Figure 5, we show some significant examples of traveltime residual maps calculated for four events approaching the network from different azimuthal sectors. The residual analysis shows features that we can associate to velocity contrasts within the target volume, giving us an idea of the average properties of the studied volume and their azimuthal dependence. These results give a first indication of the 3-D velocity anomalies that exist below the study volume. The main features observed from the residual maps are as follows:

Figure 5

Pattern of relative arrival time residuals. (a) Eastern Sichuan (China) M7.9 back azimuth 53°, distance 86°. (b) Central mid-Atlantic ridge M5.9, back azimuth 196°, distance 37°. (c) Central mid-Atlantic ridge M6.5, back azimuth 226°, distance 37°. (d) Alaska Peninsula (USA) M6.3. back azimuth 345°, distance 87°. Arrows show back azimuth for each event. Epicenters of the events used for the residual maps are the red circles in Figure 3.

[18] Earlier arrivals are measured for waves approaching the Alboran Sea and the Southern Iberian margin from northeast, south, and north (Figures 5a, 5b, and 5d). This pattern suggests the presence of a clear high-velocity body below the Alboran Sea and Southern Iberian margin.

[19] Later arrivals are observed along the western Iberian margin for waves approaching from northeast, south, and southwest (Figures 5a, 5b, and 5c). This observation can be explained by the presence of a low-velocity body underlying the western Iberian margin, from northern Portugal to the Gulf of Cadiz.

[20] We observe a passage, from the Atlantic domain below the OBS network to southwestern Iberia, from earlier to later arrival times (Figures 5a, 5b, and 5c). This passage suggests a separation of the velocity structure in the two domains, as clearly displayed in the cross sections of Figure 9a.

4 Inversion Method and Model Parameterization

4.1 Inversion Method

[21] The iterative nonlinear teleseismic tomography procedure developed by Rawlinson et al. [2006] has been applied to map relative arrival time residuals as 3-D P wave velocity anomalies. As mentioned earlier, inversion of relative traveltime residuals produces a 3-D velocity structure starting from an average (usually 1-D) earth model. Although this known background/reference model is needed for the forward calculation of theoretical traveltimes, the resulting velocity perturbations cannot be considered relative to it, as relative residuals poorly constrain vertical variations in wave speed. However, if the target volume is large enough, we may assume that the horizontally averaged velocities are close to the layer velocity of the initial model [Leveque and Masson, 1999], which should be representative of the correct regional structure. Under this assumption, which is normally made in this kind of upper-mantle studies [Shomali et al., 2011, and reference therein], nonlinear iterative teleseismic tomography can improve the reconstruction of the velocity structure. This improvement is also demonstrated by numerical computations [Koch, 1985].

[22] The ak135 spherical earth model [Kennett et al., 1995], modified for the crustal structure described in the previous section, was used as initial reference model for the inversion. The tomography scheme of Rawlinson et al. [2006] uses cubic B spline functions to define a continuous smooth velocity field from a 3-D grid of velocity nodes in spherical coordinates. The fast marching method (FMM), which is a robust, computational, and efficient grid-based eikonal solver [Sethian and Popovici, 1999], is used to compute the evolution of teleseismic wavefronts and the traveltimes from the base of the model volume to the array of receivers on the surface. Outside this volume, the Earth is assumed to be spherically symmetric, which allows the use of a 1-D global reference model to rapidly compute the traveltimes from the distant sources to all grid nodes at the bottom.

[23] In the following, we give the basic formulation of the tomographic method applied, referring the reader to the Rawlinson et al. [2006] study and references therein for a detailed description. In particular, their Figure 8 provides a schematic diagram showing the FMM approach to calculate traveltimes.

[24] The inverse problem for the model parameters m (the velocity anomalies) is solved by minimizing an objective function S(m) of the form:

(1)

where g(m) are the predicted residuals, d_obs are the observed residuals, C_d is the a priori data covariance matrix, m₀ is the reference model, C_m is the a priori model covariance matrix, and D is a second derivative smoothing operator. The square root of the diagonal entries of matrix Cm indicates the uncertainty associated with the initial model parameter values. The constraints supplied by the data will result in changes to these uncertainties. ε and η are the damping and the smoothing parameter, respectively. They govern the trade-off between how well the solution satisfies the data, the proximity of the solution model to the starting model, and the smoothness of the solution model. Equation 1 is minimized through an iterative sequence of linearized inversions in which the FMM and a subspace inversion method are successively applied to solve the forward problem and to estimate the model perturbation δm_i, respectively. The subspace inversion method operates by projecting the quadratic approximation of S(m) onto a much smaller n-dimensional model space. The set of vectors that span the n-dimensional subspace are computed based on the gradient vector and Hessian matrix in model space. Finally, singular value decomposition is used to produce an orthonormal basis of linearly independent vectors removing those vectors that are redundant.

[25] For an objective function of the form of equation 1, the perturbation δm_i is given by

(2)

where A is the M × n projection matrix (M unknowns, n-dimensional subspace), G is the matrix of Frechét derivatives and γ is the gradient vector (∂S/∂m). Once matrix A has been computed and orthonormalized, vector δm_i is rapidly obtained by the inversion of the n × n matrix, then added to the current model velocity m_i to produce the new model m_i+1 = m_i + δm_i [Kennett et al., 1988; Rawlinson et al., 2006].

4.2 Model Parameterization

[26] The starting model defined for the upper-mantle structure below the study region spans 600 km in depth, 14.0° in latitude (from 30.0°N to 44.0°N), and 18.0° in longitude (from 16.0°W to 2.0°E). It comprises 8316 velocity nodes at 60 km spacing in all three dimensions (depth, latitude, and longitude). The horizontal bounds of such 3-D grid ensure that all the waveforms of our teleseismic data set pass through the base of the model, which is needed to assign traveltimes at all bottom nodes and start FMM correctly. We used the ak135 global reference model [Kennett et al., 1995] to set the initial P wave velocity below the depth spanned by the crustal correction. The input grid model incorporates the velocity profiles used for the crustal correction for each sector (Figure 2 and Table 1).

[27] The damping and the smoothing parameters were set to 5 and 2.5, respectively, by examining the trade-off between minimizing the data misfit and reducing the model complexity. This analysis was performed following the three step procedure suggested by Rawlinson et al. [2006], which is based on trade-off curves between data variance and model roughness and data variance and model variance. Finally, arrival time error estimates from the picking analysis were used to form the diagonal elements of the data covariance matrix (C_d), while the square root of the diagonal elements of the model covariance matrix (C_m) were set to 0.30 km/s.

5 Tomography Results

5.1 Resolution Tests

[28] To investigate the extent to which the P wave model is constrained by the data, we performed synthetic resolution tests using the same raypath geometry of the observational teleseismic data set to compute synthetic traveltime residuals. We choose to apply the so-called “checkerboard test,” in which the input velocity model consists of alternating regions of fast and slow anomalies with a length scale equal (or greater) to the smallest wavelength structure recovered in the solution model. The inversion of the synthetic data set will attempt to recover the checkerboard pattern, showing the regions of the model that can be considered well resolved. In spite of some limitations [Leveque et al., 1993; Husen et al., 2003], checkerboard tests have become a standard way of addressing model resolution as their results can be promptly visualized. In particular, they give a good estimate of the amount of smearing present in the tomographic images. The spatial resolution of our tomographic images was performed using different checkerboard patterns to explore a variety of wavelengths [Nolet, 2008]. In the test, we varied the cell size from 60 to 120 km by including nodes with zero perturbation, with a maximum velocity perturbation of ±0.50 km/s (~±6%). Gaussian noise was added to the synthetic data, with standard deviations of 0.5 and 0.2 s sets to simulate the noise content of the ISC and OBS data, respectively. These values are slightly greater than the error estimated for the two subsets of arrival times (see section 2). The error assigned to ISC data is in line with the results of Gudmundsson et al. [1990] for teleseismic P phases (0.39–0.9 s). Comparing all the results obtained by using the different grids, we found a resolution width (the minimum cell size that could be distinguished well) of about 100 km. Figures 6 and 7 show the results of the test for horizontal (see also Figure S2) and vertical slices, respectively. Note that the checkerboard pattern varies in latitude, longitude, and depth. Reconstruction is partial in the south due to poor raypath coverage in northern Morocco. Although smearing of anomalies is present in some regions of the model, especially along the northwest-southeast direction (Figure 6) and along raypaths at the border of the model (Figure 7), the overall recovery of the checkerboard pattern is good. The velocity anomalies imaged with real data inversion that we interpret (see the following sections) fall in the well resolved part of the model and represent structures with wavelength comparable or larger than the scale length of our synthetic model.

Figure 6

Input checkerboard velocity models and recovered patterns obtained by inverting a synthetic data set, for a sample of slices. The other slices are given in Figure S2. The test gives an indication of the regions of the model well constrained by the data.

Figure 7

Input checkerboard velocity models and recovered patterns along the vertical slices crossing the grid model in (a) the east-west direction (AA′, BB′, and CC′) and (b) in the north-south direction (DD′ and EE′). The tracking latitude/longitude of the profiles (Figure 2) is the one used for the velocity cross sections.

5.2 Three-Dimensional Velocity model

[29] The tomographic inversion was carried out through five iterations, using a subspace dimension equal to 18 and the selected damping and smoothing values. In addition to velocity parameters, station terms were also included as unknowns in the inversion. These terms, one for each receiver, were computed mainly to absorb shallow velocity perturbations unconstrained by the teleseismic ray geometry and/or not completely accounted by the applied crustal correction (Figure S3 in supporting information).

[30] The final 3-D velocity model reduces the data variance by 26% from 0.53 to 0.39 s², which corresponds to an RMS data residual reduction from 727 to 628 m s. Although the variance improvement is not very high, we find that the computed velocity models show small differences (in size and magnitude of the anomalies) with respect to realistic variations of the inversion parameters (damping, smoothing, and inversion grid size). We found that the data fit improved with the application of the crustal and baseline residual corrections. The level of RMS reduction depends particularly on the sparse array geometry and the data noise level, which is increased by the large number of bulletin data. Figure S4 shows the initial and solution traveltime residual histograms.

[31] In the following paragraphs, we describe the velocity field computed at the third iteration as only minor improvements (<1%) in the data fit were obtained through the successive inversion steps. A comparison with the residual analysis shows that the main anomalies in the 3-D model are consistent with the features shown in the residual maps (Figure 5). Figure 8 shows a series of horizontal slices extracted from the continuous P wave solution model at the labelled depths. At the shallowest depth (60 km), high-velocity anomalies are found in the Atlantic ocean beneath the area covered by the NEAREST array (HVA1), in central Iberia (HVA2), and the Alboran Sea (HVA3). A low-velocity zone appears in northern Portugal (LVA1). A series of low-velocity anomalies (LVA2) are also visible going from the Gibraltar Strait to the Atlantic domain along the southwest Iberian Margin. At 120 km depth, HVA3 becomes larger extending from the Alboran Sea to the Granada region where it is stronger. At 180–300 km depth, HVA3 becomes more pronounced and takes an L-shaped form. LVA1 now extends along the Portugal-Spain border toward the southwest Iberian Margin. Starting from 180 km depth down to 300 km, a clear high-velocity anomaly (HVA4) is imaged beneath northern Portugal. This feature becomes stronger at greater depths, extending to the Atlantic domain and to south Portugal. Below the NEAREST array, in the Gulf of Cadiz, LVA2 appears as a continuous east-west oriented band. HVA1 becomes stronger in this depth range.

Figure 8

Horizontal slices through the 3-D solution model showing the lateral heterogeneities in the investigated upper-mantle structure. Labels show the depth at which the slices are taken. The P wave velocity model is represented as perturbations from the composite starting model described in subsection 3.2. Dots represent the local seismicity at the slice depth ±30 km. The green contour includes the resolved part of the model. This curve was drawn on the basis of the checkerboard test. Hypocenter locations of local seismicity comes from the Instituto Geográfico Nacional Spain catalog 1980–2010 (812 events with M > 3, and depth >35 km).

[32] In the deeper part of the model, across the mantle transition zone at 360–600 km depth, HVA3 loses its curvature and is oriented northeast-southwest with its southern tip very close to the Gibraltar Strait. HVA4 extends north-south below Portugal and is bordered to the southwest (Atlantic domain) and to the east (central Iberia) by wide low-velocity anomalies. The bottom of the model at 600 km depth shows HVA3 below the Alboran Sea. Below HVA3, there is the deep seismicity corresponding to the Granada area.

[33] The west-east vertical profiles in Figure 9a, which cross the southern tip of the Iberian peninsula (AA′ and BB′) and the Gibraltar Strait (CC′), show the transition from the Atlantic to the Alboran domain. The three sections shows that HVA3 extends from ~60 km depth to the base of our model, and its maximum width is ~300 km in the EW direction below the Granada region (AA′). West of HVA3, beneath the Gulf of Cadiz, LVA2 extends from the top of the model down to ~250 km depth in AA′ and BB′, and it reaches the transition zone below the Gibraltar Strait in CC′. Going toward the Atlantic, we find (below the OBS array) HVA1 surrounded by LVA2. HVA1 is at least 200 km wide and is visible down to ~240 km depth. Section CC′ runs from the Atlantic ocean to the Alboran Sea across the Gibraltar Strait. It displays the part of HVA3 underlying the Alboran Sea and shows the width of the slab (100–200 km).

Figure 9

Cross sections through the 3-D solution model, computed along the great circle paths shown in Figure 2. (a) Cross sections AA′, BB′, and CC′ display the seismic structure from west to east. (b) Cross sections DD′ and EE′ cut the modeled volume from south to north. In DD′, SAP indicates the Seine Abyssal Plain; HAP, Horseshoe Abyssal Plain; TAP, Tagus Abyssal Plain. Blue triangles and red dots indicate the seismic stations and the subcrustal seismicity located within a 100 km wide zone centered on the profile. Topography/bathymetric profiles are from the Global-Integrated-Topo/Bathymetry Grid (GINA) [Lindquist et al., 2004]. Moho topography is from the European Moho depth map [Grad et al., 2009]. The green contour includes the resolved part of the model. Hypocenters deeper than 600 km show the deep events below the Granada region (AA′).

[34] The south-north cross sections in Figure 9b go through the Atlantic domain (DD′) and the Alboran Sea and Spain (EE′). Moving from south to north, section DD′ shows the north-south extension of HVA1, approximately 250 km, underlain by the deeper part of LVA2. Section EE′ crosses the Alboran Sea and shows the north-south extension of HVA3 (approximately 300 km). Subcrustal seismicity is distributed along an arc-shaped belt with the deeper events contained in the uppermost part of HVA3. To the north, below central Iberia, we find HVA2, which is visible down to ~200 km depth.

6 Discussion

[35] The investigated area, which includes the Alboran Sea and the Gulf of Cadiz, has proven to be quite complex, as reflected by its numerous and sometimes conflicting interpretations found in literature. Despite several geophysical measurements, a more comprehensive understanding of the area has been hindered by the lack of long-term instrumental coverage at sea. The NEAREST experiment, thanks to seafloor seismological recordings, allows us to better resolve the upper-mantle structure of this area.

[36] In the following discussion, we will concentrate on the Atlantic (southwest Iberian margin and Cadiz Gulf) and Alboran Sea domains. We will discuss our results with reference to the two main geodynamic models mentioned in the introduction, i.e., subduction and delamination below the Alboran Sea.

6.1 Implications for Subduction With Rollback

[37] Consistently with previous tomography studies [Calvert et al., 2000; Piromallo and Morelli, 2003; Spakman and Wortel, 2004], our model shows a prominent high-velocity body below the Alboran Sea area (HVA3; Figures 8 and 9), suggesting the presence of cold, fast lithosphere visible down to the bottom of our model at 600 km. Thanks to our improved data set, with respect to previous studies, we are able to have a more reliable and defined image of this high-velocity body. HVA3 extends continuously from the southern Iberian Margin, below the Granada area where the scarce deep seismicity occurs, to under the Alboran Sea where there is intermediate-depth seismicity. Its geometry is L-shaped at shallower depths, and below 300 km it shows an elongated shape extending toward the Strait of Gibraltar and northern Morocco. Recent geophysical observations point to the effects of subduction (active or extinct) with westward rollback of oceanic lithosphere below the Alboran Sea. Anisotropy patterns are more consistent with a slab rollback model than with delamination or convective removal of lithosphere models [Buontempo et al., 2008; Diaz et al., 2010]. Seismic wave dispersion measurements point to the oceanic nature of the sinking lithosphere [Bokelmann and Maufroy, 2007]. All these observations suggest the existence of subduction with rollback, either active or extinct, below the Alboran Sea. Some authors have proposed the existence of a continuous lithospheric slab going from the Atlantic domain, across the Gibraltar Strait, to below the Alboran Sea [Gutscher et al., 2002, and references therein]. However, subduction, if it exists, must be either very slow or finished, as implied by GPS data, which show small to none differential motion across the Gibraltar Strait [Stich et al., 2006; Serpelloni et al., 2007]. An important element imaged in our model is a strong discontinuity between the seismic structure of the lithospheres in the Alboran and Atlantic domains (cross sections AA′, BB′, and CC′; Figure 9a), which is not resolved by previous tomographic studies. Although the part just south of the Gibraltar Strait is poorly resolved, especially in the shallower layers (Figure 8), the synthetic test shows that if there were a continuous slab coming from the Atlantic domain subducting below the Alboran Sea, we should be able to detect it.

6.2 Implications for Delamination

[38] Not all data can be explained by a subduction rollback mechanism. In fact, although deep seismicity is usually associated with a subduction process, the intermediate-depth seismicity below the Alboran Sea has a distribution which is not typical of subduction zones. In contrast to the nearby Calabrian Arc [Central Mediterranean, see for example Montuori et al., 2007], which also has a small curvature radius, there is no well defined Benioff plane. Intermediate-depth seismicity is found only within the top of the high-velocity body represented by HVA3 (Figures 9a and 9b). There seems to be a gap of seismicity between 30 and 50 km [Buforn et al., 2004], whereas a great majority of hypocenters are found above 150 km depth. Similarly, in the Vrancea area there is a gap between 40 and 70 km depth, while intermediate-depth seismicity is contained above 170 km depth. The seismic gap for Vrancea has been explained as being possibly due to a decoupled seismogenic zone from the overriding plate [Sperner et al., 2001]. Furthermore, studies of igneous rocks also imply a complex geodynamic time-space evolution of the Western Mediterranean region. Thermal models for metamorphic units from the floor of the Alboran Basin are consistent with postcollisional rapid exhumation and associated heating. Geochemical and geochronological data show a transition from postcollisional subduction related to intraplate-type magmatism which occurred between 6.3 and 4.8 Ma [Duggen et al., 2005]. These processes imply a role for lithospheric delamination [Platt, 1998].

[39] Now we will try to determine a geodynamic evolution that is compatible with the above constraints and our tomography. The high-velocity body under the Alboran Sea area appears as isolated, adjacent to a pronounced low-velocity anomaly, and possibly separated from the overlying lithosphere. The recovered geometry suggests its evolution in time: the older section of the slab, corresponding to a planar deeper part (Figure 8), was subjected to rollback and not (or minimally) deformed by the Africa-Eurasia compression, whereas the younger, shallower part, found above ~300 km depth, was subjected both to rollback and compression and still is subjected to the African-Eurasia compression, which caused its arcuate shape. The separation of the high-velocity oceanic lithospheric structure in the Atlantic and Alboran domains in our model is consistent with a subduction with westward rollback process that has come to a stop east of the Gibraltar Strait, in agreement with the reconstruction proposed by Lonergan and White [1997]. This evolution of the slab geometry also agrees with the sequence of tectonic events reconstructed by Iribarren et al. [2007], who favor a westward slab retreat process active from Middle to Late Miocene time. Successively, oceanic subduction with rollback triggered continental-edge delamination under northwestern Africa and southern Iberia, as proposed by Duggen et al., 2004 on the basis of petrological studies.

[40] Now we will comment other interesting parts of the model. A clear high-velocity anomaly, HVA1, is imaged for the first time under the NEAREST array, in an area roughly underlying the Horseshoe Abyssal Plain. This area is part of a diffuse convergent margin where old oceanic lithosphere is hypothesized [Zitellini et al., 2009; Geissler et al., 2010]. The thickness of HVA1 (~80–150 km in cross sections AA′–BB′; Figure 9a) agrees with values proposed in literature for old (~140 Ma) oceanic lithosphere [McKenzie et al., 2005; Conrad and Lithgow-Bertelloni, 2006]. Low-velocity anomalies are also visible, LVA1 and LVA2. LVA2 together with HVA1, define what we interpret as the passage from the Atlantic oceanic to the Iberian continental lithosphere (Figures 8 and 9a). The passage from HVA1 to LVA2, in the southwest-northeast direction, is in good agreement with the position of the south Portuguese stress regime anomaly observed by Stich et al. (2003) and marks the transition from oceanic to continental crust. HVA1 is visible from the crust down to sublithospheric depths. The east-west shape of HVA1 suggests incipient subduction (Figure 9a). In fact, the Gorringe bank (Figure 1) is considered an example of young incipient margin that could develop into a subduction zone [Gurnis et al., 2004].

[41] At sublithospheric depths, below the Atlantic domain, we image wide areas of lower than standard P wave seismic velocities (LVA2), down to the bottom of our model. At these depths, it is commonly agreed that temperature plays a first order role in determining lateral seismic velocity heterogeneity [e.g., Ranalli, 1996], so we interpret such low values (down to approximately −0.16 km/s, about −2% of the reference value) as being due to the presence of a hot upper mantle. Isotopic and geochemical studies show that the magmatism of western Portugal and of the adjacent Atlantic domain that occurred during the Mesozoic can be explained by the presence of a common sublithospheric regional mantle melting anomaly [Merle et al., 2006; Miranda et al., 2009, and references therein]. Recent data show that this thermal anomaly is likely the source of tertiary and quaternary alkaline magmatism in the eastern North Atlantic region [Merle et al., 2006; Grange et al., 2010]. The space-time evolution of magmatism in this region has been explained by assuming that the Iberian plate has rotated (approximately 30° anticlockwise) above a fixed deep-rooted thermal anomaly (mantle plume) [Sibuet et al., 2004; Grange et al., 2010].

[42] Although it is outside the main area of interest, we briefly comment on the persistent feature HVA4, imaged beneath Portugal from 180 km depth down to the bottom of the model. Generally, high-velocity bodies found at these depths are associated to colder and denser sinking lithosphere commonly considered of oceanic origin.

7 Conclusions

[43] Teleseismic tomography has been used to investigate the 3-D seismic structure of the upper mantle beneath the Alboran and Atlantic domains using seafloor recordings from the Gulf of Cadiz integrated with land-based seismic data. Thanks to the seafloor data we obtained a first image of the upper-mantle west of the Gibraltar Strait and an improved image of the Alboran Sea high-velocity anomaly. Information extracted from our model, crossed with independent geophysical and geological data, can be explained by a now extinct oceanic subduction below the Alboran domain, with westward rollback. Subduction stopped east of the Gibraltar Strait and was superseded by continental-edge delamination.

[44] The recovered geometry of the slab suggests its evolution in time. The older part, 360 km depth and below, is planar and could represent a time when rollback was the main force acting on the slab. The more superficial part, 300 km and above, has a curvature with a small radius which could be the result of the African-Eurasian compression acting on the slab in a clamp-like effect. This effect possibly contributed to the end of rollback and influenced the transition toward the new regime.

[45] The plate boundaries between oceanic lithosphere and continental domains are imaged in our 3-D model as lateral passages between high and low seismic velocity at lithospheric depths. The separation of two high-velocity bodies in the Atlantic and the Alboran domains suggests that the two lithospheres could have been subjected to independent geodynamic evolution, although in both the still active Eurasian-African compression played an important role.

[46] The results coming from this study are promising and show that long-term marine measurements are needed to define the upper-mantle structure west of the Gibraltar Strait and to clarify the passage between the uppermost mantle of the Atlantic and Alboran domains. In the future, by extending the area covered by the NEAREST array, especially off the Moroccan coast, we may get a more complete picture of crust and upper-mantle in the two domains and better understand their relation.

Acknowledgments

[47] The NEAREST project was funded by EC (GOCE, contract no. 037110). The authors thank Captain E. Gentile, the crew, G. Carrara, all participants of the R/V URANIA expeditions in 2007 and 2008, and N. Zitellini for coordinating the project. We are grateful to all people and institutions involved in the NEAREST project. We thank the “German Instrument Pool for Amphibian Seismology (DEPAS),” hosted by the Alfred Wegener Institute Bremerhaven, for providing the ocean-bottom seismometers. We thank John Platt and two anonymous reviewers for their useful comments which helped us improve the manuscript. We also thank T. Dahm and F. Frugoni for the useful discussions and D.-T. Ton-That for helping with the english revision. We thank N. Rawlinson for providing the original form of the inversion code, J. C. VanDecar and R. S. Crosson for making the MCCC code available, and F. Tilmann for useful suggestions. Most figures were produced using GMT software [Wessel and Smith, 1991].