The NNW-SSE-trending Red Sea Rift developed over the last 23Ma16 as a result of the ~ 0.39°/Ma counter-clockwise rotation of the Arabia Plate relative to Nubia about a pole located in the southeastern Mediterranean (the pole used in this study is 24.22°E, 31.61°N, yielding opening rates that increase from ~ 7 mm/yr in the NNW to 16 mm/yr in the SSE)17,18,19 (Fig. 1a). Towards the SSW, the Arabia/Nubia plate boundary splays into two branches that isolate the Danakil microplate18,19,20 before linking to the Gulf of Aden and Ethiopian rifts to form a rift-rift-rift triple junction (Fig. 1a). This triple junction is closely associated with the Afar mantle plume21, whose initiation preceded rifting and relative to which the Nubia plate moves NW at ~ 15 mm/yr22.
From the stable ANS to the plate boundary, the Red Sea rift (Fig. 1a) is composed of 1) a coastal plain bounded by the external rift escarpments and underlain by tilted blocks and their accompanying syn-rift sediments16, 2) a smooth offshore shelf marked by a gradually increasing elevation toward the SSE18, and whose basement is blanketed by more than 1 km of Miocene evaporites23,24,25, and 3) a discontinuous 15-70-km-wide axial trough, essentially restricted to the Southern and Central Red Sea, where the oceanic crust and its rugged volcano-tectonic sea-floor are exposed except at the so-called “inter-trough zones” occupied by allochtonous salt flows26,27,28,29.
The weak penetration of geophysical surveys across the Miocene evaporites led to ambiguities about the nature of the offshore shelves18. In particular, forward modeling of gravity anomalies and the absence of a clear magnetic signature have led some authors to conclude that the axial trough approximately captures the whole extent of the oceanic lithosphere (proximal OCT end-member: oceanic spreading < 5 Ma and diachronous along axis)23,29,30. Conversely, seismic reflection data and segmentation trails inferred from vertical gravity gradient patterns have been used by others to place the ocean-continent transition (OCT) at the outer edge of the shelves (distal OCT end-member: oceanic spreading ~ 13-14-Ma-old and synchronous)18,31.
Whereas most debate about the recent evolution of the Red Sea Rift has focused on the position of the OCT, little attention has comparatively been paid to the along-axis variability of oceanic spreading mode27. Here, we combine GEBCO 2021 (GEBCO Compilation Group, 2021) and high-resolution bathymetry with satellite free air gravity anomalies (see Methods) to characterize the topographic, gravity, and volcano-tectonic signatures of the axial trough North of its intersection with the Danakil microplate (Figs. 2 and 3) and identify long- and short-wavelength along-axis gradients. These gradients are then compared to primary characteristics of the continental margin (lithospheric thickness and distribution of volcanism, Fig. 1b), and integrated into a kinematically-constrained numerical model of plume-assisted rifting.
Along-axis topography, gravity, and volcano-tectonic patterns
Long-wavelength along-axis topography of the Red Sea Ridge features a relatively stable axial depth of ~ 2000 m.b.s.l. in the Northern and Central Red Sea (N-RS and C-RS labelled on top of Fig. 3; reference to specific location of Fig. 3 will be hereafter given in kilometers from the Arabia/Nubia pole, and denoted as K##), followed by a gradual shallowing up to ~ 1100 m.b.s.l. when approaching the Afar plume in the Southern Red Sea (Fig. 3a). To a first order, variations of the axial Mantle Bouguer Anomaly (MBA; see Methods and Supplementary Fig. 1 for calculation procedure) occur symmetrically to this topographic trend (Fig. 3a). Superimposed on the long-wavelength topographic signal are short-wavelength undulations characterized by peak-to-peak distances that apparently decrease from ~ 250 km South of K2000 to 60–100 km farther North (Fig. 3b). The positive anomalies show no clear relation with the inter-trough zones, but instead correlate with large axial volcanoes. The negative anomalies generally correspond to the Red Sea “Deeps”26 and exhibit a bi-modal distribution with vertical residuals scattering around − 250 m and − 600 m. In general, these anomalies leave no or very little MBA signature (Fig. 3a and 3b).
The slope distribution of the flanks of the axial trough also displays large lateral variability (Fig. 3c). Nevertheless, the similarity between NE flank and SW flank distributions clearly indicates that the axial trough is highly symmetric in cross-section. The 5% threshold used in constructing Fig. 3c is intended to capture the extent of the axial trough (see Methods). Its outer (and so vertical) extension is thus visualized by the shallowest terrains plotted in Fig. 3c, and shows that the top of the axial trough records a long wavelength topographic signal similar to that of the axial ridge. Antithetic slopes (i.e., opposite to the average flank slope) are locally observed in association with large axial volcanoes NW of K2000 and become widespread farther South (Fig. 3c). Two ridge segments (Nereus around K1570 and Suakin around K1960, also located in Fig. 1) display unusually steep flanks relative to other sections (Fig. 3c), and correlate with the two most negative anomalies (~-600m) of residual topography evidenced in Fig. 3b.
Both flanks of the axial trough are also characterized by a similar pattern of normal faulting (Fig. 3d), which again attests to the symmetric nature of the spreading centers. In general, fault spacing and fault offset apparently increase away from the axial volcanic ridge, which might reflect a resolution and/or preservation bias (see Methods). As for the topographic analysis, the Nereus and Suakin segments exhibit a distinctive pattern, with wider fault spacing and larger fault offsets (Fig. 3d and 4). The maximum fault offset along most of the axial trough is around 600–700 m, but increases to ~ 1300 m at the Nereus and Suakin segments. The magmatic contribution to plate opening (M: sense of (ref. 6)) determined from this faulting pattern (Methods) displays a very subtle decrease from SSE to NNW (~ 0.92 to ~ 0.85), which locally becomes much stronger at the Nereus and Suakin segments (Fig. 3d).
The analysis of the sea-floor morphology is hampered by both the variable width of the stripe covered by high-resolution bathymetry, and the presence of large portions of terrains occupied by allochtonous evaporites26 (Fig. 2). Since the maximum possible proportion of volcanic terrains includes areas that are unclassified in Fig. 3f, it is visualized from this Figure as the sum of the red-tone envelopes (see Methods). Accordingly, the Nereus and Suakin segments appear as the most magma-poor ridge segments, with less than 20% of volcanic sea-floor.
Although topographic, tectonic, and volcanic metrics vary continuously along the Red Sea axial trough, the Nereus and Suakin segments appear to share prominent anomalies, characterized by a deeper less volcanic spreading ridge flanked by more spaced larger displacement normal faults. These “anomalous segments” occur along the less oblique and less segmented sections of the Red Sea Ridge (Fig. 3e). On the other hand, they seem to align with major geophysical and geological structures of the Arabian margin. In particular, Nereus and Suakin occur next to the Lunayyir and Tufail onshore volcanic fields (located in Fig. 1b), which have the largest seaward extension at the scale of the entire Red Sea (Fig. 3g), and are markedly younger (last ~ 1 Ma and last ~ 3.5 Ma32,33 compared to last ~ 12 Ma). They also occur next to the Proterozoic Hijaz and Asir terranes of the Arabian margin13, whose bounding suture zones are intimately associated with strong density variations at Moho depth inferred from the joint inversion of seismic and gravity data15, and strong gradients in lithospheric thickness as mapped from Sp-receiver functions14 (Fig. 1b). In between, the Jeddah terrane exhibits a denser crust15 and thicker lithosphere (Fig. 1b).
Variability of oceanic spreading mode
Thermo-mechanical models of spreading ridges predict that the fraction of plate separation accommodated magmatically (M) exerts a primary control on the pattern of normal faulting, such that increasing M will be recorded on the sea-floor by smaller and less spaced normal fault scarps6,7. It is worth noting that an accurate estimation of M from bathymetry data requires tectonic displacements to be readily reflected in the sea-floor topography. It is known that this assumption may become unverified when M becomes so little that spreading switches to a detachment mode34. In that case, the progressive flexure of detachment faults will lead tectonic displacements to be poorly reflected in the bathymetry35,36. The sea-floor topography of the Red Sea axial trough documented above and by (ref. 4 and 5) however lacks typical characteristics of a detachment mode of spreading, such as domal corrugated surfaces, across-ridge topographic asymmetry, or gently dipping fault surfaces1,2,6 (Figs. 3 and 4). Instead, a negative relation between the slope of the axial trough flanks and the proportion of volcanic terrains seems to be the general rule along the Red Sea axis (Fig. 3c and 3f). This is at odds with the gentler scarps that would be otherwise expected along the less volcanic segments if spreading switched to a detachment mode. More generally, there is no avolcanic segment along the Red Sea Ridge (at least where the basement is exposed) despite the slow- to ultra-slow opening. These observations support the estimation of M from observed normal fault scarps, as well as its relevance for tracking lateral variations of the magmatic contribution along the Red Sea Ridge.
Lateral variations of the faulting pattern, as encapsulated into M, display two superposed signals. A subtle long-wavelength decrease of M (from ~ 0.92 to ~ 0.85) is observed towards the NNW, except around K1700 where it increases along with the apparent magma focusing signed by the large Hatiba Mons volcano27 (see the large positive anomaly around K1700 in Fig. 3b). Superposed onto this signal are larger drops near the Nereus and Suakin segments (M ~ 0.3–0.4) (Fig. 3d). These latter are also accompanied by a local doubling of the maximum fault offset (passing from an average of 600–700 m to ~ 1300 m), that goes along with a decrease of the number of overlapping faults (Fig. 3d). These two signals point to a slight regional northward decrease of the proportion of plate separation accommodated magmatically, and a more pronounced local decrease at the Suakin and Nereus segments. This interpretation is consistent with the locally lower proportion of exposed volcanic sea-floor (Fig. 3f).
On the other hand, the few high-resolution bathymetric data available SE of K2000 (Fig. 2) show that the axial volcanic ridges are generally wider and more continuous than farther NNW26,27. Flank antithetic-slopes, interpreted as large tectonically-splitted volcanoes, also become more abundant along this section (Fig. 3c), which might be indicative of a magmatically more robust character of the spreading ridge. The data therefore show that plate opening is less and less accommodated magmatically when moving towards the NNW, and that this trend is locally amplified at the Suakin and Nereus segments.
Whereas the regional trend could easily be explained by a combination of decreasing spreading-rate and increasing distance from the Afar plume towards the NW (Fig. 1a and 3), the anomalous character of the Suakin and Nereus segments appears more enigmatic. In particular, there is no evidence that the local decrease of M is controlled by the segmentation of the ridge (Fig. 3), as observed along other slow-spreading ridges9. Instead, the Suakin and Nereus segments are located along the less oblique and less segmented sections of the Red Sea Ridge (Fig. 3e).
Moreover, along-axis variations of the MBA surprisingly show an absence of typical bull-eye patterns centered on segment centers35,38. The MBA also gradually increases from the Suakin segment towards the SSE (Fig. 3a), which is contrary to what would be expected from an increasing crustal thickness. In general, MBA varies symmetrically with the mean axial depth at long wavelengths (Fig. 3a), which suggests that the axial ridge is not isostatically compensated. These observations converge to suggest that the expected gravity signature is hidden by a dynamic topographic support, as proposed for example by (ref. 37) along the Reykjanes Ridge.
Differential focusing of plume-related heat by the thickness of the marginal lithosphere
Lateral spreading of the hot and buoyant Afar plume into the laterally-confined Red Sea Rift has been already documented from the 4He/3He signature of axial basalts39 and the density structure of the upper-mantle beneath the Red Sea Ridge15, and thus constitutes a natural candidate for this dynamic support (Fig. 1a). We think it constitutes the primary cause of the shallow axial depth of the Southern Red Sea, but also of the western termination of the Gulf of Aden (Fig. 1a). Although it does not exclude the hypothesis of a flexural lifting induced by enhanced erosion of the Arabian margin as recently advocated by (ref. 40), this latter should translate into an ENE-WSW cross-sectional bathymetric asymmetry, which is not specifically observed (Fig. 1a).
We suggest that the narrowness of the Red Sea basin is the key condition for focusing buoyant plume material beneath the spreading ridge, and that this condition exerts primary thermal and gravitational controls on the regional trend observed in Fig. 3. Accordingly, we argue that the anomalous character of the Nereus and Suakin segments most likely records a locally decreased focusing. Specifically, we propose that the spatial correspondence between these two segments, the onshore Lunayyir and Tufail volcanic fields, and the shallower LAB of the Hijaz and Asir terranes (Fig. 1b and 3) is best explained by an outboard deflection of the heat source where it becomes less laterally confined due to shallowing of the marginal lithosphere.
This is demonstrated using a modified version of the numerical model of (ref. 41), which is appropriate to track the effect of LAB topography variations on the lateral spreading of low-density ponded material (see Methods). Our simulations are kinematically constrained to fit with both the rotational opening and lithospheric thinning of the Red Sea and Gulf of Aden, and the coeval motion of the Afar hotspot relative to the lithospheric plates, and account for the shallower pre-rift LAB of the Hijaz and Asir terranes in the form of two rift-oblique sub-lithospheric channels (Fig. 5) (see Methods). These simulations show that channeling of plume material beneath the tectonically thinned rift is accompanied at the intersections with these sub-lithospheric channels by the development of on-axis plume thickness anomalies and outboard viscous fingering during the continental rifting phase (Fig. 5a and S3).
Following the onset of oceanic spreading, the on-axis anomalies are dissipated through along-axis diverging flow of the buoyant material and continuing viscous fingering towards the continental margins (Fig. 5b). These primary features account for the coeval development of comparatively less magmatic segments at Nereus and Suakin and onshore volcanism of Lunayyir and Tufail. We also emphasize that the only uplifted Arabian marine terraces are precisely located at the suspected onshore outlet of the Nereus/Lunayyir deflected flow along a 20-km-long coast-line section (Supplementary Fig. 2), which might consistently indicate dynamic uplift.
The simulation shown in Fig. 5 prescribes a synchronous onset of oceanic spreading (see Methods) consistent with the distal OCT end-member18,31, and for which the on-axis divergence of plume material lasts ~ 1 Ma. Sensitivity tests show that models prescribing a northward propagating onset of spreading (in line with the proximal OCT end-member42) differ only in terms of the response time, with the dissipation of the on-axis anomaly taking up to ~ 8 Ma (Supplementary Fig. 3). We therefore suggest that additional chronologic constraints on the proximal onshore volcanism might provide indirect insights about the rhythm of propagation of sea-floor spreading in the Red Sea.
On the other hand, onshore volcanism is highly asymmetric at rift scale (Fig. 1). In the context of our model, this could be explained if syn-rift lithospheric thinning was also asymmetric as previously proposed by (ref. 43), and/or if the pre-rift lithospheric structure of the Hijaz and Asir terranes on the Nubian margin (for now poorly constrained14,15) lacked the shallow LAB features seen on the Arabian conjugate.
The mechanism advocated here is similar to that proposed by (ref. 12) to account for the distribution of syn-rift volcanism in Iceland and the North Atlantic region. Our study shows from in-situ observations that this mechanism may persist beyond the stage of continental break-up, and points to the decisive role played by pre-rift lithospheric inheritance on the mode of early sea-floor spreading in plume-assisted rift settings. By differentially focusing thermal lithospheric thinning, this mechanism might also determine the final geometry of the associated oceanic basins.