Introduction

The presence of partial melt within the crust of the southern Tibetan and central Andean plateaus is inferred from low seismic velocities and low electrical resistivity1,2,3,4,5,6,7,8,9. Despite similarity in geophysical observations, the origin and dynamics of the crustal partial melt are attributed to different processes in the two plateaus. Models include partial melting due to crustal thickening in the Tibetan Plateau2 and by intrusion of asthenospheric material in the central Andes10. The timing and cause of plateau uplift and crustal thickening remain controversial, and non-uniqueness of interpretations of crustal structure has contributed to competing geodynamic models. The rise of the Tibetan Plateau has been attributed to underthrusting of Indian lower crust11, oblique subduction of lithospheric blocks within the plateau12, and lateral crustal flow13,14. In the central Andes, the removal of dense mantle lithosphere and the addition of magma to the lower and middle crust are proposed to promote plateau uplift10,15.

The uplift of the Gangdese Arc in the Lhasa Block of the Tibetan Plateau began >60 Ma16,17, predating continent-continent collision18, and the central Andean Plateau has been uplifting for 23 Myr19. The long uplift history of these plateaus introduces challenges in inferring their deep structure at the onset of uplift. Therefore, processes leading to the initiation of plateau uplift may better be studied at young collisional plateaus. The uplift of the East Anatolian Plateau (EAP) in Turkey with a topography of ~1.5–2.0 km began around 12 Ma in relation to the collision between the Arabian Plate and Eurasia20 with a current convergence rate of 20–30 mm/yr21.

The EAP has very low Vs anomaly (around −6 to −9% with respect to regional average) in the uppermost mantle22,23, low Pn velocity (generally Vp < 8.0 km/s and some zones have Vp < 7.8 km/s)24,25, and high lithosphere temperature (800–900 °C at Moho)26. It is broadly covered by Neogene to Quaternary volcanic rocks27,28. Dynamic support of the plateau by upwelling asthenosphere has been proposed29 based on the assumption of a 38–46 km thick crust, as indicated by low-frequency (f ≈ 0.72 Hz) Receiver Function (RF) studies, which interpret a strong P-wave converter as the Moho under the assumption of constant average crustal Vp/Vs = 1.73 throughout the region30. This assumption of constant Vp/Vs ratio contradicts seismic tomography models which include significant low-velocity bodies, especially in the active volcanic provinces22,23, suggesting the presence of intra-crustal, partially molten zones. These zones would require high crustal Vp/Vs and should also produce strong P-RF converters at their base, which easily would be mistaken for the seismic Moho.

Here we apply a refined method31 for imaging the crustal structure of the EAP by seismic RF. We combine data from a temporary deployment of seismic stations at ~50 km interval32 and permanent stations from the Kandilli Observatory across the EAP (Fig. 1 and Supplementary Table 1). Our 450 km long RF profile extends from N to S across the EAP in its central part with abundant volcanic fields, the Bitlis Suture, and the northern part of the Arabian Plate. We image strong and weak P-RF converters, and we apply a new gravity-based interpolation method to identify the Moho (Methods)31. We demonstrate that the base of the zone that includes partially molten bodies was earlier interpreted as the Moho30. To reduce the bias in our interpretation of the RF results, we constrain the upper 80 km of the lithosphere by integrating the RF results with regional seismic tomography22,23,33,34, magnetotelluric35, geothermal26 and geochemical models27.

Fig. 1: Topographic map and tectonic setting of the East Anatolian Plateau (EAP).
figure 1

Main map: regional topography; grey transparent shading—Neogene to Quaternary volcanics; inverted triangles—seismic stations used in this study, red inverted triangles—stations used for illustrated receiver functions with synthetic tests (Fig. 2b); thick black line—location of receiver function profile; white lines—collisional structures with subduction direction marked by white triangles, except for zones with controversal subduction direction29,67,68; red lines with arrows—large strike-slip faults. Insert: geographical location of the study area with blue stars showing epicentres of teleseismic events used for the calculation of RFs.

Full size image

Results

Crustal receiver function profile

Our P-wave RF model (Fig. 2a) shows the following:

  1. 1.

    The Arabian Plate has a crustal thickness of 35–40 km, as identified by a strong converter in an otherwise non-converting environment.

  2. 2.

    Shallow positive converters close to the margin of the Arabian Plate mark a main detachment which we interpret as the Main Bitlis Thrust (MBT) at a depth of 10–15 km. This blind thrust36 shares strong similarity to the Main Himalaya Thrust in Nepal11,37,38.

  3. 3.

    The Bitlis Massif to the south of the Muş Suture has a crustal thickness of ~40 km, with a gradually N-dipping Moho converter (M2) to ~50 km depth in the EAP.

  4. 4.

    The EAP includes two positive converters: M1 at ~30 km and M2 at ~50 km depth (Fig. 2a and Supplementary Table 1). Typically, M1 has larger amplitude than M2 (Supplementary Fig. 1). Additionally, we observe a strong negative converter TL which is located ~10 km shallower than M1 within the EAP. These sub-horizontal converters are observed on all stations within the EAP and are not identified outside the EAP.

Fig. 2: Receiver Function (RF) profile across the East Anatolian Plateau (EAP).
figure 2

a Depth-migrated RF profile (red and blue colors represent increase and decrease of seismic impedance with depth, respectively). Black line atop RFs shows topography along the profile and its variation within a ± 40 km-wide corridor (grey shading). Color boxes at the top mark four tectonic units (BM – Bitlis Massif, NWIF – NW Iranian Fragment) separated by suture zones (BS – Bitlis Suture, MS – Muş Suture). Inverted triangles show the locations of seismic stations. Grey wiggly lines are the stacked RFs at each station migrated to depth. TL and M1 converters surround a ~ 10 km thick intra-crustal low-velocity zone, M1 is the previously interpreted RF Moho30,45, which we interpret as the top of a mafic layer above our interpreted Moho M2. b Crustal velocity models for two stations (red triangles in a) based on modelling with synthetic RFs (Supplementary Fig. 2). The ~10 km thick low-velocity intra-crustal zone (red color) has an electric resistivity of 1 Ωm35 and a density of 2.60 g/cm3; and the ~20 km thick lower crust has an electric resistivity of 10–30 Ωm35 and a density of 3.05 g/cm3. c Sketch of our interpretation. Numbers are density in g/cm3 constrained by gravity modeling (Supplementary Fig. 5b). Solid line M2 – the interpreted Moho, dashed line M1 – top of mafic underplate.

Full size image

The crustal layer between TL and M1 has a very high Vp/Vs ratio of ~2.0 and a low Vs velocity ranging from 2.9 to 3.1 km/s based on calculations of synthetic RFs (Fig. 2b and Supplementary Fig. 2). Results from inverting the RFs for velocities also suggest a low Vs of 2.96-3.33 km/s except for station CMCY in northern EAP (Fig. 3a, Supplementary Figs. 3 and 4). The lower layer between M1 and M2 has a high P-wave velocity (Vp~7.2 km/s) and a high Vp/Vs ratio (~2.0) in its lower part (Fig. 2b).

Fig. 3: Seismic and electrical structure along the RF profile.
figure 3

Plot on the top – topographic profile within a ± 40 km-wide corridor along the profile, seismic stations are shown by inverted triangles with code names on the top. a Vertical Vs profiles at each station from non-linear inversion63 of RF (cf. Methods and Supplementary Figs. 3, 4). Probablity scale originates from the inversion algorithm, cf. Supplementary Fig. 3. Dashed lines contour the intra-crustal low-velocity zone (LVZ) at the EAP, interpreted to include bodies of partial melt. b Migrated RF at individual stations displayed in the depth domain to show the data coverage of the profile. c, e P- and S-wave models obtained by interpolation of our velocity inversion results in (a). d, f Seismic tomographic S- and P-wave models33. Superimposed vertical profiles show the individual velocity profiles from our velocity inversion. g Electrical resistivity structure35 with Vs from our RF inversion superimposed. Notice the correlation between low-resistivity bodies and low-velocity zones.

Full size image

Other geophysical observations

We use the long-wavelength filtered Bouguer anomaly for resolving the density contrast across the Moho and for regional scale isostatic calculations, whereas the resolution of the gravity data (~9 km laterally39) may be too low to model the dimensions of individual melt pockets. A body wave travel time tomographic velocity model33 shows that the crustal layer between M1 and M2 (30–50 km depth) in the EAP has a Vp velocity between 6.6 and 7.2 km/s (Fig. 3f). Gravity modeling (Supplementary Fig. 5b) suggests that the layer between M1 and M2 has a high density of 3.05 g/cm3, assuming a density of 3.20 g/cm3 in the hot uppermost mantle at 50-70 km depth (Fig. 2c). This layer also includes several bodies with low electrical resistivity of 10–30 Ωm35 (Fig. 3g).

The low-velocity anomaly between TL and M1 (20–30 km depth) cannot be clearly resolved by body wave tomography due to vertical resolution limitations33. It is, however, observed in Rayleigh wave data at a period of 30 s22. Strong low-density (2.60 g/cm3, Fig. 2 and Supplementary Fig. 5b) and low-resistivity (~1 Ωm) zones35 (Fig. 3g) are unique for the EAP in the study region. The converters TL, M1, and M2 along with magnetotelluric anomalies are only observed in the volcanic province of the EAP, which has 0.5–1.0 km higher topography than neighboring blocks (Fig. 3).

Crustal converters and the presence of magma

The parameters of the layer between TL and M1 ( ~ 1 Ωm electric resistivity35, Vs~3.0 km/s22, and Vp/Vs ratio ~2.0 at 20–30 km depth) are similar to a layer containing crustal silica melt observed in the central Andes, where joint inversion of surface wave dispersion and RF constrains Vs = 1.9-3.1 km/s40, with average Vs~3.1 km/s from RF synthetic modelling3, and electric resistivity lower than 1 Ωm7,41. The Vp/Vs value of ~2.0 between TL and M1 indicates a distinct Vs decrease in this layer which may be caused by the presence of either fluids or partial melt. Considering the young magmatic activity in the EAP (Fig. 4a) and the temperature close to the felsic solidus at 30 km depth (Fig. 4b), the presence of pockets with felsic partial melt is the most likely explanation for this depth interval at 20–30 km. Heat flow data are absent along our profile, but the 12–18 km deep Curie temperature depth indicates high heat flow26. For a dihedral angle of 20–40° in the equilibrium geometry model, with an aspect ratio of 0.10–0.1542, we estimate the melt volume fraction by a forward model for partial melting43 to be between 3 and 13% for a Vs value between 2.96 and 3.33 km/s (Supplementary Fig. S6a).

Fig. 4: Evolution of magmatic activity along the profile.
figure 4

a Profile on the top – topography marked with 4 tectonic units (abbreviations as in Fig. 2). Lower diagram - Age of volcanism28 within a ± 100 km-wide corridor along the RF profile. b Regional geotherm (thick red line)26 and experimental solidi. Line G: for wet felsic rock49 which suggests partial melting of felsic crust at a depth of ~30 km (pink layer). Line B: for wet mafic rocks48 which implies partial melting of the crust below 30 km depth (blue layer). Line P: for peridotite with small amount of C-H-O69,70 which implies mantle melting below 70 km depth (purple layer). c Temporal change of subduction slab dip (thick grey lines) from 20° to 26° due to roll-back27,34. Blue arrows illustrate migration of fluids and mafic melts extracted from the mantle wedge into the lower crust. Upward migration of fluids and felsic magma together with partial melting due to heat flux (red arrows) create partially molten material in mid-crustal layer (pink color) in the EAP. d Geodynamic sketch of the tectonic evolution (from right to left). Colored circles at surface show ages of volcanism. Magma composition27 is illustrated at depth by colors. Converters M1 (black dotted line) and M2 (black solid line) bound the 20 km thick underplated mafic layer, the red layer above M1 is the zone containing pockets with felsic partial melt.

Full size image

The MT model (Fig. 3g) indicates the presence of three major, individual melt bodies in the EAP separated by ~70 km, but the resolution of the available data is insufficient to provide measures of their dimension due to leakage of the low-resistivity in the model. The depth to the top of these bodies (~20 km) corresponds to the top (TL) of the upper low-velocity zone (LVZ), which extends continuously across the EAP (Fig. 3c), and the LVZ thickness may indicate the vertical extent of the melt pockets. The receiver functions and the seismic tomographic models have insufficient resolution to outline the individual partially molten bodies, but the recurrent low-velocity zone at all stations in the EAP indicates that the material between TL and M1 generally is close to the solidus44, also outside the bodies containing partial melt.

Previous RF interpretations selected M1 as the Moho30,45 because it is the strongest converter in the RF section. By the new method31 based on long-wavelength gravity-interpolation of converters from the Arabian Plate we, instead, find that the strong converter represents the strong contrast between the low-velocity partially molten layer and the high-velocity (Vp~7.2 km/s) and high-density (~3.05 g/cm3) lower crust, most likely with mafic composition (Supplementary Fig. 5b). The lower part of this layer has a Vp/Vs of ~2.0 and an electric resistivity of 10–30 Ωm35 which indicates the presence of a, not yet fully solidified, mafic magmatic material. High Vp of ~7.2 km/s is typical for mafic rocks at 1.4 GPa and 900 °C ( ~ 50 km depth in the EAP)43. This Vp value is lower than in the underplated lower crust with Vp = 7.4-7.6 km/s at the Baikal Rift Zone46, but it is similar to other underplated layers like in the Kuril Arc, the Norwegian-Danish Basin and the Donbas Basin47. It is unlikely that this layer may represent the underthrusted Neo-Tethys oceanic slab, since tomographic images and geochemical results indicate that the oceanic slab has broken off27,34.

Geodynamic evolution and magmatism

Most of the recent magmatism28 along the profile is concentrated in the EAP with variable magma composition ranging from felsic to mafic and a clear pattern of north to south younging of the onset of magmatic activity since 12 Ma (Fig. 4a and Supplementary Fig. 7). The occurrence of very young magmatism, with age less than 1 Myr, suggests the presence of magma chambers and partially molten bodies within the crust that can still be detected using geophysical data. The predicted geotherms26 intersect the experimental wet mafic rock solidus48 at depths of ~30–50 km and cross the wet felsic solidus49 at a depth of ~30 km (Fig. 4b). This indicates that the mafic lower crust between M1 and M2 at 30 to 50 km depth, especially its lower part with high temperature, may still be partially molten which is consistent with the Vp/Vs of ~2.0. Further, the low velocity zone between TL and M1 at 20–30 km depth probably includes partial melt of felsic composition, considering that the estimated temperature of ~600 °C at a depth of 30 km26 is below the wet mafic melting solidus, the high Vp/Vs of ~2.0 and the low resistivity of ~1 Ωm.

We explain the lateral change of magmatism from 12 Ma to 3 Ma by spatial changes in subduction dynamics and geometry. Assuming that the subducting Neo-Tethys slab induces melting of the mantle wedge at a depth of ~70 km (Fig. 4c), and observing the magmatism of various ages at distances of 345 ± 15 km (~12 Ma), 295 ± 36 km (~5 Ma), 230 ± 21 km (~3 Ma), the subduction angle would have steepened from 20 ± 1.5° (~12 Ma) to 26 ± 6° (~5 Ma). Alternatively, the magmatism might have been caused by delamination of the lithospheric mantle and lower crust28,50, but this mechanism does not explain the clear southward younging of magmatism initiation. We propose that underplating47 (addition of mafic magma to the lower crust or below the original Moho) led to magma fractionation and that the accompanying release of fluid and heat initiated melting of preexisting continental crust, which may explain the presence of the mid-crustal partially molten bodies.

Slab roll-back and break-off in the EAP have earlier been proposed based on geochronology, geochemistry and seismic tomography27,34. We extend this interpretation by our geophysical model (Fig. 4d), which suggests that the magmatism moved laterally from north to south due to slab roll-back and steepening followed by slab break-off, and that this evolution caused intensive addition of mafic magma to the lower crust and intra-crustal partial melting.

Discussion

The free air gravity anomaly difference between the Arabian plate and EAP, of about 30-40 mGal, implies that the topography is mainly controlled by isostatic buoyancy (Supplementary Fig. 8). Compared to the Arabian plate, the EAP has an additional elevation of ~0.75 km. Selecting the 1 km topography in the Arabian plate as reference, our new seismic-density model (Fig. 2c) explains the actual topography of the EAP by isostasy for a density of 3.3 g/cm3 in the mantle lithosphere and 3.2 g/cm3 in the asthenosphere (Fig. 5a, Supplementary Fig. 8a).

Fig. 5: Plateau uplift by underplating and crustal partial melting.
figure 5

a East Anatolian Plateau, (b) central Andes Plateau, (c) Tibetan Plateau. Left panels: schematic cross-sections through the three plateaus constrained by regional models with topographic profiles on top. Middle panels in (b), c profile locations. Right panels – 1D crustal density (in g/cm3) models used for calculating the predicted topography (dotted lines with values on the right) from the free air gravity, assuming isostasy with a compensation depth of 80 km and with the Arabian Plate as reference model (middle panel in (a) with 1.0 km topography) (Methods and Supplementary Fig. 8). The predicted elevation for the EAP lithosphere model in Fig. 2c is 1.7 km as observed (a). The elevation of 4.5 km in the Altiplano/Puna Plateau (b) is explained by a lithosphere model with the same densities as for the EAP, similar thickness of the high-velocity lower crust3, and the presence of a felsic partially molten crustal body71. The high topography of the Lhasa block (c) reflects a balance between the degree of lower crust eclogitization11,52,72 and the thickness of the melt-hosting layer inconclusively imaged seismically with two end-member models. Both models explain the observed 5.6 km high topography. Model I2 has a 40 km thick layer (with bulk 2.60 g/cm3) hosting partially molten bodies and a 20 km thick eclogitic lower crust (3.30 g/cm3, with 39–55% eclogitization). Model II11 has a 10 km thick layer with partially molten bodies, above a lower crustal layer with density of 3.15 g/cm3 (15–22% eclogitization).

Full size image

The presence of a 20 km thick layer containing high-velocity, mafic material in the EAP is not unique. A similar size (~200 km long and ~20 km thick) high-velocity lower crust has been also mapped by RF below the central Andean plateau3,10 (Fig. 5b) and in parts of the Lhasa Block of the Tibetan plateau11 (Fig. 5c). The free air gravity anomalies in the central Andean plateau and the Lhasa block are up to ~90 mGal and ~40 mGal, respectively. We calculate the isostatic topography by adding the necessary small corrections for the isostatic equivalent of the free air anomalies (Fig. 5bc, Methods and Supplementary Fig. 8). The results indicate that isostasy is the primary cause of the topography.

In the central Andean Plateau, our model predicts the observed topography with the same densities of 2.60 g/cm3 for the layer with partially molten bodies and 3.05 g/cm3 for the underplated lower crust as in the EAP. For the southern Tibetan Plateau, two end-member models for the thickness of the upper to middle crustal layer with partially molten bodies have been proposed: either (1) it is thick and the continuation of a seismic “bright spot” downward to the top of mafic lower crust2 or (2) it is limited to only 10 km thickness11. Although model 1 now appears unrealistic, we consider both models for the Tibetan Plateau, where the observed topography may be fit by densities of 3.30 g/cm3 or 3.15 g/cm3 in the lower crust for the two end-member models, respectively (Fig. 5c). This indicates that the mafic lower crust may be partially metamorphosed into eclogite: Eclogitization may be caused by the very deep level of the lower crust in Tibet (70–80 km)51,52,53.

Our observations of the crustal structure of three plateaus in collisional settings with different topography and uplift duration (12 Ma in EAP, 23 Ma in Altiplano, and 60 Ma in Tibet) indicate that (Fig. 6):

  1. 1.

    the high-density, high-velocity mafic lower crust has comparable thicknesses of ~20 km in all three plateaus;

  2. 2.

    the thickness of a mid-crustal layer containing bodies of felsic melt has key influence on plateau elevation, and this thickness appears positively correlated with the duration of uplift and the present plateau elevation.

  3. 3.

    crustal thickness increases through the growth of the mid-crustal layer.

Fig. 6: Relation between plateau uplift duration, thickness of the middle crustal layer and crustal thickness in EAP, Andes and Tibet.
figure 6

The middle crustal layer is defined as the layer between the brittle upper crust and the mafic lower crust. Thickness of the mid-crustal layer and topography versus (a) age of plateau uplift and (b) crustal thickness.

Full size image

The case of the central Andes shows that continent-continent collision is not required for plateau uplift, and the uplift of the Altiplano plateau is consistent with the recent observation of pre-collision uplift of the Gangdese Arc in southern Tibet18. For a thin overriding plate, like the Japanese Arc with a ~15 km thick upper crust, neither plateau uplift nor intra-crustal partial melting is observed, but a ~15 km thick mafic lower crust may be present54. We speculate that underplating may occur, but the thickness of the continental crust may be insufficient to generate large volumes of felsic magma by crustal melting. However, we find that plateau uplift, where the overriding plate is continental, generally can be explained by mafic underplating in association with intra-crustal partial melting. The final high plateau is built by inflating the weakened, low viscosity continental crust containing bodies of partial melt, while the uplift-induced regional climate change may affect the high topography and vice-versa55,56.

Our observations of the physical state of the lithosphere during the initial stage of plateau uplift provide a quantitative framework for the factors that lead to the formation of bodies of a partially molten material within the crust, and their buoyancy effects on continental plateaus are enhanced by mafic underplating of new lower crust17,57. Our revised geodynamic understanding of continental uplift explains topographic change at pre-collisional stage18. Our results indicate that subduction below continental lithosphere is a necessary condition for collisional thickening by underplating, intra-crustal melting and related plateau uplift, and the process is independent of the nature of the subducting plate.

Methods

Receiver functions (RF)

The (radial) Ps RF is calculated by time domain iterative deconvolution by ITERDECON58 with a Gaussian filter of 2.5, which corresponds to a cut off frequency of ~1.24 Hz. RFs were processed for epicentral distances of 30–90° and each RF from each station has been manually inspected (Supplementary Fig. 9). RFs with a large difference from the overall characteristics of other RFs were rejected in the final RF results. We manually picked the P-wave arrival time of each RF (Supplementary Fig. 9c). For some seismic phases with weak amplitude, we also used the multiple-taper correlation (MTC)59 to estimate the RFs (Supplementary Fig. 9d); a comparison of the RF results from different deconvolution methods ensures recognition that the weak seismic phases are not noise artifacts. Synthetic RF tests were calculated by the IRFFM (Interactive Receiver Function Forward Modeller) software60.

Assuming a single-layer medium, the crustal thickness H and the crustal average κ = Vp/Vs are determined by H-κ stacking61 of the converted Ps wave and the reflected multiples from the Moho interface. We use weight parameters of 0.6, 0.3, 0.1 for the Ps, PpPs, PsPs+PpSs phases to stack the peak amplitudes (Supplementary Fig. 1b). The sensitivity of the results to the assumed constant Vp=6.3 km/s is insignificant within uncertainty (Supplementary Fig. 10). We record all possible H-κ pairs, and use the gravity field data to determine the final H-κ values according to a new method31 (see details in the section on Gravity interpolation of RF results).

RFs were migrated into the depth domain (Figs. 2a, 3b) based on the assumed crustal P-wave velocity of 6.3 km/s, and the Vp/Vs ratio (Supplementary Table 1) retrieved from the H-κ stacking with application of gravity interpolation of the RFs (Supplementary Fig. 11). The final RF and velocity profiles are obtained by applying a moving average filter with a radius of 30 × 2 km (horizontal × vertical) to the migrated RF rays and the inverted Vp and Vs profiles.

Velocity inversion of RFs

We estimate vertical P- and S-wave velocity profiles from the RFs by applying three different methods: linear RF inversion62, non-linear RF inversion by the transdimensional Markov-chain Monte Carlo (MCMC) algorithm63 and the Neighborhood Algorithm (NA)64,65. The results of the non-linear algorithms are similar, including the identification of the low-velocity zones in the middle crust. The first algorithm assumes a constant Vp/Vs = 1.73. We first check that this method leads to similar results as an earlier application for the EAP30, when constrained to the upper 40 km of the crust. Application of the same method to more realistic depths of 50 and 56 km for detecting the Moho, produces results that include the pronounced low-velocity zone between 20 and 30 km depth as well as a Moho at around 50–60 km depth (Supplementary Fig. 3a). By extending the covered depth interval, the visual fit between calculated and observed RF is substantially improved (Supplementary Fig. 3a).

The non-linear transdimensional MCMC inversion algorithm63 estimates both Vp and Vs profiles (and thereby Vp/Vs). Given the strong variation in Vp/Vs along the EAP profile, this method provides better resolution of depth and velocity variation, and it leads to a better fit between calculated and observed RF (Supplementary Fig. 3b). The results identify a clear Moho transition in addition to a strong velocity contrast at the base of the mid-crustal layer with partially molten bodies. The determined vertical velocity profiles for individual stations correlate very well between each other along the profile as documented by the velocity profiles (Fig. 3c, e). The resulting vertical Vs profiles also correlate with the occurrences of high-conductivity bodies in the MT model (Fig. 3g). The NA inversion64,65 identify the same low-velocity zones beneath stations within EAP although the results are less detailed as the NA is restricted to 6 layers in the models (Supplementary Fig. 12). Each inversion included 1000 iterations with 20 models generated at each iteration and 10 Voronoi cells re-sampled at each iteration. The results are consistent with results from the MCMC inversion which confirms that the inverted Vs profiles are robust.

As the inversion was carried out independently for each station, the strong lateral continuity of the resulting vertical velocity profiles testifies to the presence of anomalous bodies (melt pockets) in the same depth interval along the profile, and it justifies the horizontal interpolation of the results between the stations.

Gravity calculations

The horizontal resolution of ~9 km and the amplitude accuracy of ~9 mGal of the EIGEN-6C4 gravity data39 may be insufficient for modelling the individual molten bodies in the crust with unknown size. Instead, we use the long wavelength gravity signal for calculation of the isostatic topography and for resolving the density contrast across the Moho. The 2-D gravity forward calculations are made with the FastGrav software. The Bouguer gravity anomaly is calculated for the EIGEN-6C4 gravity data39 and the ETOPO1 topography data66, assuming a crustal density of 2.67 g/cm3. To model the gravity signal from a deep Moho, the short-wavelength Bouguer anomaly fluctuations, originating from the shallow crust, are removed with a Gaussian filter of 500 km31.

Model 1 (Supplementary Fig. 5b) explains the observed gravity anomaly and is consistent with the RF profile. It includes the crustal layer with partially molten bodies (with bulk layer density of 2.60 g/cm3, Vs~3.0 km/s, Vp~6.0 km/s) and the mafic underplated layer (density of 3.05 g/cm3, Vs of 3.6-4.1 km/s, Vp~7.2 km/s) for an asthenosphere density D1 of 3.2 g/cm3. Model 2 without the mid-crustal low-velocity layer and the underplated lower crust (Supplementary Fig. 5c) cannot fit the gravity data unless the density of the asthenosphere beneath the EAP22 is unrealistically high (>3.3 g/cm3, cf. green line) or the density of the lower crust is high (≥2.95 g/cm3, cf. black line), corresponding to an underplated lower crust, which supports the presence of an underplated lower crustal layer.

Gravity interpolation of RF results

We apply a new method31 for regional determination of Moho depth variation by minimizing the difference between the long period gravity signal and potential Moho depths along the profile, selected from the results of H-κ stacking31. The method selects the optimum Moho depth at each seismic station from several possible candidates derived from the H-κ stacking by testing the crustal models against gravity. We adopt an empirical correlation between regional long-wavelength (>248 km) Bouguer gravity anomaly (ΔgB) and Moho depth (D), so that a lower value of the Bouguer gravity anomaly corresponds to a deeper Moho: ΔgB = a + bD, where a and b are empirical regional parameters. For a known ΔgB, we minimize the difference between the Moho depth obtained as (ΔgB–a)/b and multiple Moho candidates from the regional RFs by scanning the a, b parameter space. This procedure31 determines the approximate Moho depth for any given ΔgB value, and allows us to select the closest RF Moho depth (Supplementary Fig. 11)31.

Predicted topography

To calculate the predicted topography in the three plateaus, we assume regional isostasy with pressure equilibrium at a depth of 80 km below the different tectonic units. The Arabian Plate is close to isostatic equilibrium with a free air (FA) gravity anomaly of ~20 mGal, and we adopt it as reference model. The lithostatic pressure is Pref = toρo+Σhiρi, where to and ρo are thickness and density of the material above sea level, hi and ρi are thickness and density of different layers from sea level to 80 km depth. Isostatic topography for other tectonic units is therefore tiso = (Pref - Σhiρi)/ρo. The predicted free air gravity anomaly is FA = 2πG(toρo+Σhiρi). At full isostatic equilibrium (and neglecting lithosphere flexure) FA should be near-zero, but the central Andes and Tibet have FA anomalies of ~90 mGal and ~40 mGal, respectively. We assume that the extra ΔFA anomaly originates from a non-isostastic topography component of Δt = ΔFA/2πG/ρo in addition to the isostatically predicted topography, tiso,with the Arabian Plate adopted as reference. This non-isostatic correction allows for including the observed FA into the isostatic calculation.

Quantitative comparison with tomography and MT models

Age data of magmatic rocks in Eastern Anatolia28 within a ± 100 km wide corridor is projected onto the profile (Fig. 4; Supplementary Fig. 7). The absolute P-wave and S-wave velocity profiles33 are located along the RF profile. The MT model35 is smoothed by parameters of alpha = 1, tau = 335, to compare to the RF result, as these parameters provide the smallest misfit between the magnetotelluric data and the resistivity model35.

We calculate the P-wave and S-wave velocities at different melt fraction, pressure and temperature by forward modeling43 of seismic velocity and electrical conductivity and compare with the absolute P- and S-wave velocity profiles33 and the smoothed resistivity model35 along the RF profile (Fig. 3; Supplementary Fig. 6).