Mantle structure and dynamic topography in the Mediterranean Basin



[1] We study the contribution of mantle flow to surface deformation within the Mediterranean Basin. Flow is modeled numerically based on lateral changes in mantle temperature estimated from tomography models. We find that modeling results are significantly affected by the properties of the selected tomography models. Shear-velocity models based on surface-wave observations achieve the highest resolution of upper-mantle structure, and, as a result, are most successful in predicting microplate motion and dynamic topography.

1. Introduction

[2] The observed topography of the Earth's surface is explained by the combined effects of shallow processes like erosion or sedimentation, and deep ones like isostasic compensation and the upward push/downward pull associated with convective flow in the mantle [Gurnis, 2001]. The latter term, known as dynamic topography, has been evaluated in a number of studies, where models of mantle density are defined on the basis of seismic tomography, and used to predict the associated mantle flow [Forte, 2007]. The large-scale patterns of topography in western North America [Moucha et al., 2008], southern Africa [Lithgow-Bertelloni and Silver, 1998] and Arabia [Daradich et al., 2003], for example, have been explained in terms of whole-mantle convection.

[3] In regions where the resolution of tomography models is high, it is possible to model smaller scale, shallow (top few hundred km of the upper mantle) convection processes, accordingly explaining the shorter-wavelength component of topography. One example is the Mediterranean Basin, where mantle structure has been illuminated by a number of high-resolution tomography models. Observed plate velocity, seismicity and topography suggest that the region is composed of a mosaic of blocks that move independently from the overall convergence [Serpelloni et al., 2007]: (i) the southern part of Adria, separated from the rest of the plate by the middle-Adriatic shear zone [Favali et al., 1993; D'Agostino et al., 2008], slowly moves eastward with respect to Eurasia; (ii) in the same reference frame, Anatolia moves westward by ∼2 cm/yr, while (iii) Aegea moves predominantly southward ( ∼3 cm/yr) [Reilinger et al., 2006]. Vertical motion in the area is, at least at some sites, as fast as the horizontal, as illustrated by holocene notches, marine terraces and karst caves along the southern Italian coast line [Ferranti et al., 2006], the Hellenic arc and in Spain [Casas-Sainz and de Vicente, 2009]. The Anatolia-Iran highplain is also related to large-scale uplift, with less well constrained timing.

[4] Faccenna and Becker [2010, hereafter FB10] have shown that 3-D density heterogeneity and corresponding flow in the upper mantle can at least partially account for this complex pattern of horizontal and vertical motion. In the following we explore how different modeling results for microplate motion and topography throughout the Mediterranean basin are obtained on the basis of different tomography models. We find that shear-velocity models of the region provide a better fit of geophysical observables than achieved by the compressional-velocity ones. In addition, predictions of the horizontal and vertical motion of Iberia, Adria and Anatolia based on the new high-resolution models considered here are now more consistent with observations. This is a significant step to understand, in general, the style of convection beneath converging margins, and towards the identification of a consensus seismic model of the Mediterranean upper mantle.

2. Surface Deformation in the Mediterranean Region

[5] The Mediterranean region is a diffuse plate boundary where sharp topographic features and deformation are distributed on a wide area. We show in Figure 1 the difference between observed Earth's topography and its isostatic component, computed based on crustal densities and layer thicknesses from the global crustal model Crust2.0 [Bassin et al., 2000] (Figure 1a), and from the regional crustal model Eurocrust07 of Tesauro et al. [2008] (Figure 1b), with densities again taken from Crust2.0. Both maps and model predictions are filtered by convolution with a 250 km-radius Gaussian function to eliminate artifacts related to gridding and to remove small-scale structure that is flexurally supported. This “residual” topography should approximately coincide with dynamic topography caused by mantle flow alone. In areas where Figures 1a and 1b differ (e.g., Anatolia), the more recent model Eurocrust07 (Figure 1b) should be taken as reference.

Figure 1.

(a) Residual topography after correcting for isostatic adjustment based on the crustal model Crust2.0 [Bassin et al., 2000]. (b) same as Figure 1a, isostatic adjustment computed from Eurocrust07 [Tesauro et al., 2008], with density values taken, again, from Crust2.0.

[6] Our set of geophysical observations is augmented by the plate velocities of Adria and Anatolia (GPS measurements reported by Serpelloni et al. [2007]), and Africa and Arabia (model NUVEL-1A by DeMets et al. [1990, 1994]), with respect to a fixed Eurasian plate.

3. Seismic Imaging of Mantle Structure

[7] FB10 employed a set of three tomographic models of the Earth's mantle as alternative input parameters of their mantle-flow models: the global P-velocity model of Li et al. [2008] (dubbed model MITP08 in the following); the Mediterranean upper mantle model of Piromallo and Morelli [2003] (PM01); the S-velocity upper-mantle model of Lebedev and van der Hilst [2008], merged with SMEAN [Becker and Boschi, 2002] in the lower mantle (LH08).

[8] PM01 and MITP08 are high-resolution models based on large travel-time databases, including phases that are particularly sensitive to regional-scale, upper-mantle structure; yet they were derived without making use of surface-wave observations, which provide a stronger constraint on uppermost mantle structure, but are almost only sensitive to S velocity. LH08 is based on observations of surface waves, but has by construction relatively low resolution of regional-scale structure.

[9] Here we additionally consider two S-velocity models that resolve short-scalelength features under the Mediterranean Basin: the global, multiple-resolution model LRSP30EU of Boschi et al. [2009], based on teleseismic measurements of surface-wave phase velocities, and the regional model of Schmid et al. [2008], merged with S20RTS [Ritsema et al., 1999] outside the region of interest, obtained from a combination of surface-wave and S-wave observations (Schmid in the following).

[10] Horizontal sections through all the mentioned models with the exception of PM01 are shown in Figure 2. FB10 show that results associated with PM01 are broadly consistent with what they find from MITP08. As discussed by Boschi et al. [2009], tomographic models of the upper mantle under the Mediterranean Basin share the same long-wavelength pattern. Dominant features are high velocities under the east European craton, associated with deeper-than-average continental roots; a narrower high-velocity body under (and north of) the Hellenic arc, interpreted as a northward dipping subducting slab; low velocities down to ∼200 km under the Western Mediterranean and the Aegean sea, explained by spreading activity in both regions [Wortel and Spakman, 2000]. At larger depths (∼600 km), upper mantle structure is dominated by a system of large-scale fast heterogeneities extending from southern Iberia and the western Mediterranean to the Alps, Carpathians, Aegean sea and Anatolia (e.g., Figure 2 of Wortel and Spakman [2000], Figure S1 of FB10), most likely related to subduction [Capitanio and Goes, 2006].

Figure 2.

Horizontal sections through the tomography models described in section 3, at depths of (left) 100 and (right) 300 km below the Earth's surface. (a–d) Models already employed by FB10. (e–h) Models on which new mantle flow calculations presented here are based.

[11] Important differences are apparent at shorter scales:

[12] 1. P-velocity models are characterized by shorter-wavelength signal with respect to S-wave ones, but this might reflect the nonuniformity of body-wave data coverage, while surface waves, used to derive the S-wave models of Figure 2, sample the upper mantle more uniformly.

[13] 2. Remarkable differences between the S models of Figure 2, and the P ones employed by FB10, include a fast S heterogeneity under Adria, whose sign is inverted in the P models, and an equally strong slow heterogeneity under Iberia, not seen in the P models or in LH08.

[14] 3. The upper mantle under Anatolia is slow at 100 km depth according to Schmid, and, to some extent, LRSP30EU, in agreement with the idea of a relatively thin lithosphere; fast according to MITP08.

[15] 4. LH08 is a global model whose wavelength is by construction longer than that of the other models in Figure 2: its horizontal parameterization grid is much coarser than that of, e.g., LRSP30EU (compare Figure 5 of Lebedev and van der Hilst [2008] with Figure 2b of Boschi et al. [2009]).

4. Modeling

[16] We estimate mantle density structure by multiplying tomographically mapped seismic velocities by a constant scaling coefficient. For this approach to be valid, we have to assume that the effect of lateral variations in chemical composition on seismic velocities is negligible compared to that of temperature variations (FB10). The resulting 3-D density models are then input parameters of the global, finite-element code CitcomS [Zhong et al., 2000], which we use (in the CIG version available at to model mantle flow. The other important input parameter is the assumed mantle rheology: we reproduce the preferred set-up of FB10, with (i) Newtonian viscous rheology, (ii) a 3-layer profile with viscosity equal to 1021 Pa · s between 100 and 660 km depth, and 5 × 1022 Pa · s elsewhere, and (iii) 100 km-wide, 100 km-thick surficial weak (0.01 reduction with respect to ambient viscosity) zones along main plate boundaries. FB10 verified that flow-modeled dynamic topography in this set-up is relatively insensitive to changes in rheological parameters, including lateral viscosity variations. As discussed by FB10 we prescribe the velocities of the African and Arabian plates as boundary conditions for flow computation. We do not require our flow models to fit plate velocities observed in Adria and Anatolia; these velocities are an output parameter, and we can use their similarity to corresponding GPS observations to evaluate the performance of different models.

5. Results and Discussion

[17] We show in Figure 3 the contribution of mantle flow to the surface tectonics of the Mediterranean Basin, computed on the basis of the tomography models of section 3. Comparing Figures 3 and 1, it is apparent that surface-wave based models LH08, LRSP30EU and Schmid achieve a better fit of observation-based estimates of dynamic topography than the body-wave model MITP08. The same is true for GPS measures (yellow arrows in Figures 3b, 3d, 3f, and 3h) and dynamic predictions (white) of microplate motion. These observations are confirmed by the linear correlations between Figures 1a and 1b and modeled dynamic topography, and the mean lengths of vector difference/mean dot-products between modeled and observed velocities in Figures 3b, 3d, 3f, and 3h, given in Table 1. We explain this finding in terms of the specific properties of different tomography models. Surface waves are more sensitive than body waves to relatively shallow, upper-mantle structure; as a result, models of the upper mantle based on surface-wave data are more accurate. Because dynamic topography is mostly controlled by upper-mantle structure, modeling results based on surface-wave tomography explain it best.

Figure 3.

Maps of (a, c, e, and g) modeled dynamic topography (colour scale as in Figure 1) and (b, d, f, and h) horizontal velocity (white arrows), shown in the same order as the tomography models of Figure 2 on which they are based. GPS observations of Adria and Anatolia velocity (orange arrows), and the velocity (from NUVEL-1A) of Africa and Arabia relative to Eurasia (grey arrows) are also shown in Figures 3b, 3d, 3f, and 3h.

Table 1. Regional Linear Correlation Between the Estimates of Residual Topography Based on the Crustal Models Crust2.0 and Eurocrust07, and Dynamic Topography Computed From a Suite of P and S Tomography Results; Mean Length of Vector Differences Between Predictions and GPS Observations of Plate Velocity; and Mean Normalized Dot Product of Modeled Plate Velocities and GPS
 Crust2.0 (Corr.)Eurocrust07 (Corr.)GPS Data (Diff.)aGPS Data (Prod.)b
  • a

    Larger values naturally correspond to worse fit.

  • b

    Neglecting datapoints with velocity <0.3 cm/year.


[18] Evaluating the comparative performance of the different S models is complicated by the difficulty inherent in correlating features of shorter spatial wavelength, at the limit of resolution for both tomography and crust models. Table 1 shows that the higher-resolution models LRSP30EU and Schmid are not improving, in comparison with LH08, the fit to observed dynamic topography, controlled by the longer wavelengths. Yet, LRSP30EU and Schmid are most successful at fitting measured plate velocities. They likewise effectively predict a number of local geological features:

[19] 1. Iberia: FB10 suggest that the plateau-like morphology of the mesetas and alkaline volcanism can be explained in terms of mantle flow [Casas-Sainz and de Vicente, 2009]. Our new models (Figures 3e and 3g) support this idea more convincingly.

[20] 2. Massif Central: positive dynamic topography presumably related to decompression melting [Lucente et al., 2006] is a robust feature of our new models, and not of those based upon MITP08 or LH08.

[21] 3. Adria: the dynamic uplift predicted in this area based on PM01 and MITP08 is in contrast with the observations of Figure 1. Models LRSP30EU and Schmid predict negative dynamic topography throughout Adria, in good agreement with residual topography data [Shaw and Pysklywec, 2007]. The small amplitude of Adria's horizontal motion is also predicted here more accurately than in the study of FB10, with model LRSP30EU also fitting, at least to some extent, its north-northeast direction.

[22] 4. Aegea: the southwestward motion of Aegea, associated with the retreat of the Hellenic slab, is reproduced in direction, if not in amplitude, by LRSP30EU (Figure 3f) and Schmid (Figure 3h).

[23] 5. Anatolia: expected positive dynamic topography in this region [Şengör et al., 2003], recovered only partially in Figures 3a and 3c, is a robust feature of our new models.

[24] Two important conclusions arise. First, after applying the method of FB10 to a larger set of tomography models, we further substantiate their conclusions: small-scale mantle flow explains the region's tectonics to an even larger extent than originally suggested. Second, surface-wave based mantle models fit observed dynamic topography and horizontal plate motions well: this confirms their reliability as reference maps of regional heterogeneity.


[25] All figures were generated with the Generic Mapping Tools software [Wessel and Smith, 1991]. LB is grateful to Domenico Giardini for his support and encouragement. TWB was partially supported by NSF projects MEDUSA (EAR-0633879) and PICASSO (EAR-0809023). We are thankful to all authors who shared the tomography models we discuss, to A. Morelli, S. Lebedev, F. Capitanio and anonymous reviewer for their comments and suggestions. We thank CIG and its contributors for sharing and maintaining the CitcomS software.