Multibeam bathymetry data of the sampling areas were collected during expeditions P408 of the German R/V Poseidon (Schmidt et al., 2011) as well as 64PE-350/351 and 64PE-445 of the Dutch R/V Pelagia (Schmidt et al., 2013; Augustin et al., 2019). R/V Pelagia operated a 30 kHz Kongsberg Maritime AS EM302 multibeam system for water depths up to 7,000 m. R/V Poseidon was equipped with a hull-mounted ELAC Nautic Seabeam 3,050 echo sounder with a 50 kHz frequency and a maximum working depth of 3,000 m. The digital terrain model (DTM) has a spatial resolution of 30 m, and the grids are publicly available from the PANGAEA data repository (Augustin et al., 2016b; Augustin et al., 2020). Backscatter data from this area were only usable from R/V Pelagia 64PE-445 due to poor weather conditions on the earlier expeditions. The backscatter mosaic has a resolution of 30 m (Supplementary Material 1). The nomenclature of the Red Sea Deeps used in the text and figures follows the compilation of Augustin et al. (2014) and references therein.

This study presents one seismic line (64PE-445-18SPS, segment 03) crossing the Discovery Deep derived from a larger dataset collected during expedition 64PE-445 (Augustin et al., 2019). As a seismic source, we used a Delta Sparker system with a dominant frequency of approximately 300 Hz. Seismic energy was recorded using a Microeel solid-state streamer with 24 channels and a length of 100 m resulting in an in-line common-midpoint bin spacing of 8 m. Processing of the seismic data was carried out using VISTA seismic processing software and comprised trace-editing, simple frequency filtering (50–2000 Hz), predictive deconvolution for bubble suppression, followed by NMO correction (1,500 m/s), CMP stacking, finite-difference post-stack migration, as well as top-muting and white noise removal. Interpretation of the seismic data was carried out using the KingdomSuite software by IHS. For a detailed overview of the method, see Hübscher and Gohl (2014).

Gabbro samples were recovered with a gravity corer from the Discovery Deep at 21°16.9′N/38°02.7′E (Figure 1) in the Red Sea Rift during expedition 64PE-445 SaltAx (August–September 2018) at a water depth of 2,240 m (Augustin et al., 2020). No sediment core was recovered from this station, but an assemblage of hard fragments was caught in the core catcher, consisting of mm-cm sized gabbroic rock and carbonate fragments, shells, iron hydroxides, and aragonite crystals. Some of the carbonate fragments have a manganese coating. From the 13 fragments, the largest pieces were selected for X-ray fluorescence (XRF; 2 fragments), thin section preparation, and electron probe microanalyzer (EPMA) analyses (3 fragments).

Major- and main trace-element concentrations of two gabbro fragments were determined by X-ray fluorescence at the University of Kiel, Germany (Table 1). The samples were prepared as two glass tablets, adding 2.4 g LiBO₄ to 0.6 g of the pulverized and homogenized sample. Measurements have been performed with a Panalytical Zetium XRF, and the data were automatically internally corrected for spectral overlaps as well as for matrix effects using the De Jongh Alpha calibration model (De Jongh, 1973). Loss on ignition has been determined at 1,100°C and an ignition time of 24 h.

Major-element concentrations of the main mineral phases olivine (21 crystals), plagioclase (26 crystals), clinopyroxene (28 crystals), and ilmenite (49 crystals) on the three thin section fragments have been measured using a JEOL JXA-8200 electron microprobe at the GEOMAR Helmholtz-Centre for Ocean Research, Kiel, Germany. All minerals have been measured with a beam acceleration voltage of 15 kV. Beam currents of 10 nA for olivine, 20 nA for plagioclase and clinopyroxene, and 50 nA for ilmenite have been used. We used a focused beam for olivine, plagioclase, and oxides and spot sizes of 1 µm for clinopyroxenes. Analytical precision (2σ standard deviation) of the main elements in each silicate phase is <2% for SiO₂, CaO, MgO, and Al₂O₃, <5% for Na₂O and FeO, and <10% for K₂O, TiO₂, and Cr₂O₃. For data quality monitoring and calibration, reference samples from the Smithsonian Institute have been used (Jarosewich et al., 1980) (see supplementary data for accuracy of the standard measurements). All measurements have been performed on fresh, non-poikilitic crystals (Figure 2).

The bathymetric maps of the Discovery Deep and surrounding areas reveal numerous volcanic structures, faults, and depressions (Figures 1B,C). The Discovery Deep is a semi-circular depression SE of the Atlantis II Deep with a low rugose seafloor, deepening from north to south with a gentle slope from about 2,200 m to 2,800 m. It is bounded to the west by a significant, east-facing normal fault with an exposed fault scarp of about 680 m and a talus fan that partly extends into the Discovery Deep. The deep has a smooth topography, which is interrupted by large, oval-shaped, and crater-like depressions. These craters are up to 300 m long, 250 m wide, and 50 m deep (Figure 1B). East of the Discovery Deep, two more groups of crater-like depressions are visible, they are aligned N-S and follow the general trend of the Red Sea Rift (chains A and B; Figures 1B,C). These craters are up to 500 m wide and 100 m deep, and the crater chains are 1,100 m and 1700 m in length. The Hummocky volcanic terrain is present around the craters as well as a very flat, cratered volcanic edifice further to the east (Figures 1B,C). Volcanic edifices and volcanic terrain are abundant around Discovery Deep and the southern part of the Atlantis II Deep, with volcano sizes ranging from 100 m to >3 km in diameter. While most structures are flat-topped volcanoes, we found some volcanoes with craters, in addition to smaller volcanic cones and hummocky volcanic terrain (Figures 1B,C). Multibeam backscatter mosaics reveal a relatively even, low backscattering seafloor in the observed area, indicating a sedimented area. Only a few high backscattering patches that may be less sedimented are visible in the data within the Discovery Deep (not related to apparent bathymetric features). Further visible are the main fault and the talus fan, and other fault scarps are visible in the data (see Supplementary Appendix 1).

The internal architecture of the Discovery Deep is revealed by the seismic profile illustrated in Figure 3A. This line crosses the distinct east-facing escarpment and traverses the Discovery Deep slightly north of the central craters before crossing the craters observed east of the Deep (Figure 1B). The seismic line is generally characterized by a thin sequence of layered reflections above the acoustic basement. This upper sequence has higher amplitudes in Discovery Deep compared to the low-amplitude sequence west of the Deep. Towards the east, we observed that this upper unit becomes more chaotic above a hummocky acoustic basement. Above the Discovery Deep, we identify a subtle horizontal reflection at 100–200 ms two-way travel time (TWT) above the seafloor and corresponds to the high-salinity brine. In addition to that, we detected a prominent concave reflection on the hanging wall of the fault that loses expression west of the projected location of the craters Figure 4).

Thirteen gabbro fragments were collected, ranging in size from 0.5 to 2.5 cm (Figure 2A). All fragments look in hand specimen very similar, and none of the gabbro samples had any type of coating or encrustations, such as basalt, manganese, or an alteration layer.

Thin sections of the studied gabbro fragments consist of approximately 95% fresh minerals (Figures 2A,B) and show the ophitic texture of the rock. Plagioclase has a modal proportion of 45 vol% with large, prismatic, subhedral crystals, ranging in size from 0.5–2.5 mm and showing albite twinning and concentric zoning. Some plagioclase crystals show a poikilitic texture with a chadacryst phase that is too fine-grained to be identifiable. A few plagioclases are rich in fluid inclusions. Clinopyroxene crystals (35 vol%) range from 0.4–1.3 mm in size and are subhedral to euhedral, intergrown with plagioclase. Orthopyroxene and transparent spinel each constitute a minor proportion of 5 vol%. Olivine crystals (5 vol%) mostly show sizes in the range from 0.2 to 0.5 mm, with few crystals that reach 1 mm. Hence, the olivine crystals are smaller than both plagioclases and pyroxenes and either fill the intergranular space between these major phases or occur as inclusions in clinopyroxenes. Olivines are anhedral to subhedral and are free of serpentinization. However, most olivine crystals show a corona of dissolution, possibly altered to iddingsite. Opaque oxide minerals are accessory and randomly distributed but mainly associated with brownish, fine-grained patches of mixed unidentified alteration minerals with a combined proportion of 5 vol%.

The rock fragments have LOIs of 0.64 and 1.75 wt% and a bulk rock Mg# (Mg# = Mg/(Mg + Fe)) of 0.45 and 0.47. The content of alkalis (K₂O+ Na₂O) is 2.13 wt% (Figure 5A) for both measured fragments, and thus, the samples fall within the geochemical field defined for gabbros in the total alkali-silica diagram (Middlemost, 1994). Al₂O₃ shows variations of 18.5 and 17.1 wt% for the two fragments. FeO concentrations in the newly described rocks are within the range of other Red Sea gabbros (<9 wt%) and lower than reported samples from the Tihama Asir complex 10–15 wt%) (Figure 5B). All rift-related gabbro data included in this study show Al₂O₃ concentrations in the range of 13–23 wt%. The bulk rock chemistry of the two gabbro samples shows higher Al₂O₃ and lower SiO₂ concentrations than basaltic glasses from the rift axis but mostly overlap for other major elements (Figure 5C).

Trace element concentrations of the gabbros are generally low and in the same range as the samples from the northern Red Sea (Zabargad Island (Bonatti and Seyler, 1987; Piccardo et al., 1988), Brothers and QUSEIR gabbro (Ligi et al., 2018), and the Tihama Asir gabbros (Ghent et al., 1979). Compatible elements incorporated in the main silicate phases (Ni, Cr) show higher concentrations, and Sr is elevated due to the compatibility in plagioclase. The Zr is with 49 and 48 ppm in the two fragments and thus enriched relative to the gabbro analyses from the Mid-Atlantic Ridge (Miyashiro et al., 1970; Bonatti et al., 1971; Dostal and Muecke, 1978; Tiezzi and Scott, 1980; Drouin et al., 2009; Chen et al., 2013) or SW Indian Ridge (Meyer et al., 1989; Coogan et al., 2001; Holm, 2002; Ray et al., 2011).

Olivine crystals show a normal zonation with cores of on average Fo₈₀ and rims of on average Fo₇₃. The Ni and Cr contents show a positive correlation with the Fo content of the crystals (Figures 6A,B). Hence, both elements are more concentrated in the cores (774–1,465 ppm NiO, 18–376 ppm Cr₂O₃) than in the rims (629–1,266 ppm NiO, 19–216 ppm Cr₂O₃). In contrast, Mn concentrations increase towards the rims up to 6,130 ppm. The plagioclase also show a zonation, with An-rich cores (An₇₆) and Na-richer compositions (An₄₄) towards the rims. From the core to rim, FeO and TiO₂ increase with decreasing An content followed by a decrease in FeO, TiO₂, and An in the rim measurements (Figures 6G,H). Measurements have been performed on fresh, non-poikilitic crystals. Representative measurements for the main mineral phases are documented in Table 2.

All measured clinopyroxenes have been classified as augites by stoichiometry. Mg numbers (Mg# = Mg/(Mg + Fe)) ranging from 78.7 to 29.2, indicating a normal zonation. Crystal cores are FeO-poor (av. 7.3 wt%) with increasing FeO-concentrations towards the rims (av. 12.7 wt%) along with an enrichment of MnO and TiO₂ and a decrease in Al₂O₃, Cr₂O₃, NiO, and CaO (Figures 6A–E). The Cr₂O₃ concentrations in clinopyroxenes decrease exponentially with decreasing Mg#, from 0.82 wt% to below detection limit (Figure 6D). Aluminum in clinopyroxene generally shows higher concentrations in the cores (Figure 6D). Two distinct groups of clinopyroxene core compositions are visible in Al₂O₃, TiO₂, and Cr₂O₃, one with higher and one with lower concentrations of these oxides. This grouping is mainly visible in the core measurements and is less marked in the rim compositions (Figure 6E). The two distinct groups appear in all three fragments and are not related to mineral size, particular neighbor minerals, areas on the section, or analytical issues but occur throughout consistently throughout the thin sections and are larger than the analytical error (see error bars in Figure 6).

The distribution of Al^VI/Al^IV (Figure 6F) in the clinopyroxenes, calculated from the stoichiometric formulas shows that our measured clinopyroxenes are clearly distinct from the Al values of the primary clinopyroxene in the Zabargad metagabbros but fit very well with the oceanic gabbro field from Hodges and Papike (1976).

The measured intergrown opaque minerals are Fe–Ti oxides, with a TiO₂ concentration of 10–25 wt% and therefore classified as ilmenites.

The gabbro fragments from the Discovery Deep are the first ones that have been recovered from the Red Sea rift axis. No other deeper crustal rocks have been found so close to the rift axis before, despite its slow-spreading character, which was earlier attributed to higher volcanic activity related to the Afar plume (Augustin et al., 2016b). The samples studied in this publication were not recovered from a core complex but instead from a crater structure with a limited sediment cover. The sampled craters also show a stronger backscatter signal (Supplementary Appendix 1) than the surrounding deeper parts of the surveyed area, possibly as a result of coarser talus or erupted material with little or no sediment cover (e.g., Hewitt et al., 2010). Even though most parts of the deep are filled with metalliferous sediments (Gurvich, 2006), no sediments were recovered during the sampling (Augustin et al., 2020). Hence, the crater appears to be sediment-free, even though located in a largely sedimented, brine-filled depression. Eruption craters have been observed in the Red Sea Rift in several areas usually covered by evaporites and sediments (e.g., between the Erba and Poseidon Deeps). Based on their morphology and location within the neovolcanic zone, these eruptive craters have been classified as volcanic craters (see Augustin et al., 2016a for a detailed discussion). Although the craters in Discovery Deep have a slightly higher depth/width ratio compared to the other Red Sea craters, their morphology, the roundness of the crater, the lack of sedimentary muds, and also their location in the neovolcanic zone (supported by the mid-ocean ridge geomorphology in the study area) point to a similar origin (Figures 1B,C). The possibility of (explosive) eruptions in the deeper marine settings, as is necessary to create such craters, is rather new and poorly understood but has also been introduced for the Charles Darwin Volcanic Field, Cape Verdes (Kwasnitschka et al., 2013), and for a monogenetic volcano field on the Southwest Indian Ridge (Dick, 2021).

Figures 3B–D illustrate the information obtained from the seismic line crossing the Discovery Deep, slightly north of the crater. This interpretation is based on the described seismic facies and on the geological map shown in Figure 1C. We interpret the stratified sequence on top of the acoustic basement as pelagic sediments. The higher-amplitude apron within the deep is likely metalliferous sediments, as was indicated by extensive ground-truthing (Gurvich 2006). Furthermore, the eastward dipping normal fault can be traced to depth with a talus fan extending from the fault, potentially up to the location of the craters. The visible concave reflection underneath the approximate location of the craters fulfills the criteria defined by Planke et al. (2005) for the interpretation of seismic events as sill reflections: it is 1) of high amplitude, has 2) a saucer-shape, and 3) terminates abruptly. We approximate the depth of this potential sill to be 400 m below the sea floor level (assuming a seismic velocity of approximately 3 km/s). This sill could potentially have provided a magma and/or heat source for the eruptive craters, but based on the seismic line, we do not see a reflector that could indicate a feeder dike from the sill to the craters. The absence of such a steep dipped feeder dike in the seismic image could be result of the limited vertical resolution, caused by the short streamer length of only 100 m. On the other hand, concerning the complex three-dimensional shape of a sill intrusion and the position of the seismic line that is slightly north of the crater structures, the displayed profile could be out of plane to provide evidence for a feeder from the potential intrusion to the crater.

The interpretation of the crater as a volcanic eruption crater could indicate a volcanic emplacement of the gabbro fragments by the eruption that formed the crater. If the sill was the driving mechanism to form the crater structures on the seafloor, we would expect it to form in an early phase of the intrusion due to the excess of heat (and melt) rather than in a later stage where progressive cooling of the intruded material takes place. So, unless the sill continues to the surface at the crater, it is unlikely that the gabbro directly represents the cooled sill itself. More likely, deeper crustal rocks could be transported to the surface as xenoliths that are typically found in volcanic craters. The absence of host basalt attached to the gabbros, which may seem inconsistent with this hypothesis, could be explained by the fact that the collected fragments are derived from a reasonably large xenolith (consistent with the homogenous appearance and chemistry of the fragments) and/or the occurrence of a more phreatic explosion rather than a phreatomagmatic eruption. Due to our unusual sampling method with a gravity core of only 12.5 cm diameter, we cannot confirm or exclude the presence of basalt.

Alternatively, the gabbros could represent expelled material already in place from the talus fan potentially uplifted by the normal fault or a combination of these two mechanisms. We have no indications of when the exposure took place. However, the lack of hydrothermal alteration despite exposure to the 45°C temperature, high-salinity brine, as well as the lack of sediment coverage in an area with high sedimentation rates (up to 10 cm/kyr; Stoffers and Ross, 1974) point to a rather recent nature of the craters. Therefore, both the uplift of the gabbroic material by a fault and the eruption as a xenolith from larger depth or even a combination of the two are possible scenarios for the emplacement of the gabbro fragments. More detailed investigations of the crater, the talus, and the fault would be needed to distinguish between these scenarios.

The rock fragments overlap in major- and trace-element chemistry with gabbros found at the Mid-Atlantic Ridge and the Southwest Indian Ridge (Dostal and Muecke, 1978; Meyer et al., 1989; Coogan et al., 2001; Holm, 2002; Drouin et, al., 2009; Godard et al., 2009). Considering the grain size of 0.5–2.5 mm, the samples can be classified as a typical mid-ocean ridge gabbro, indicating a slower cooling and crystallization of the fragments at greater depths before exposure of the fragments at the surface where they were sampled. Significant chemical or petrographic differences between the studied fragments have not been observed. The fragments show a chemical overlap with measurements from axial MORB basalts in the Red Sea, and we do not see evidence for continental crustal contamination. The elevated content of Zr, which is incompatible in most mafic-ultramafic systems, compared to gabbros from other slow-spreading ridges along with the low Mg# suggests the Red Sea Rift gabbros to be relatively evolved. Thus, we suggest that the gabbros originate from an asthenospheric melt, which ascends beneath the Red Sea Rift but undergone significant crystallization, which formed the more evolved composition.

The petrography (e.g., olivine inclusions in clinopyroxene and plagioclase) suggests an early crystallization of olivine. The fact that these minerals are smaller than the Pls and Cpx even though they crystallized earlier indicates a termination of olivine crystallization. Recrystallization of olivine, which would also lead to smaller crystal sizes, would equilibrate the element budget of the olivines instead of the typical fractional crystallization of Ni and Cr vs #Mg patterns that we observe here (Figures 6A,B), as experimental (Grove and Bryan, 1983) and empirical studies of phenocrysts (Stakes et al., 1984) have shown. Some olivines are entirely altered, as is made apparent by the subhedral to euhedral relict olivine pseudomorph shapes that are filled with alteration products. The intergrowth and similar crystal size range of the two main mineral phases plagioclase and clinopyroxene suggest that they co-formed. The enrichment of FeO and TiO₂ during the crystallization of the plagioclase cores and the trend inflection at around An₅₂ indicates the onset of the intergranular ilmenite crystallization to have taken place simultaneously to the formation of plagioclase rims (Figures 6G,H).

The grouping of the core chemistries of clinopyroxenes (observed in 24 crystals), visible in Cr₂O₃, Al₂O₃, and TiO₂, indicates the crystallization of clinopyroxene from different magma batches. The later rim crystallization appears to have happened in one environment, as the rim measurements are scattered and do not follow the core chemistry grouping. Neither the cores nor the rims of the clinopyroxene are in equilibrium with the whole rock chemistry, based on the Fe–Mg partitioning that has K_d (Fe–Mg)^cpx-liq values (equilibrium between clinopyroxene and a nominal coexisting liquid) of 0.26–1.22 for cores and 0.24–1.99 for rims, which is significantly higher than the equilibrium values of 0.27 ± 0.03 (Putirka, 1996). The clinopyroxenes of different geochemical groups do not translate into different fragments or noticeable heterogeneities in the recovered gabbros. Therefore, a formation in a single, heterogenous magma reservoir with either local or temporal changes, implying significant small-scale differences in the reservoir’s element budget, seems unlikely. Hence, crystallization thus occurred in at least two different environments and under disequilibrium conditions, which implies a history with multiple magma batches and/or melt reactions rather than simple crystallization in one magma reservoir.

A trend towards lower bulk rock Mg# along with a decrease of An in plagioclase, Mg# in clinopyroxene, and Fo in olivine has also been described for shallow gabbros from Hole 735B, interpreted as a progressive upwards differentiation within the crust (Boulanger et al., 2020). Furthermore, a similar trend in FeO vs An content in plagioclase has been reported from Hole 735B at the SW Indian Ridge (SWIR; Dick et al., 2002). Our plagioclase compositions also show an enrichment in FeO with a decrease in An content and with progressive crystallization from core to rim, followed by a decrease in FeO at An contents of An₅₅ (Dick et al., 2002). The similarity of the trends in SWIR and the Red Sea rift gabbros may imply that, equally as the SWIR gabbros, our samples were likely formed by crystallization in a mush with a change to an iron–titanium-rich environment in the last stages, as inferred from ilmenite formation. Dick et al. (2002) interpreted such an evolution as a migration of late iron-titanium-rich liquids into fractures in the upper part of the sequence. This is not observed in our samples. As we also see the joining of the two different groups of clinopyroxene at this stage, we interpret that processes such as the mobilization of the crystal mush and mush or magma mixing could have played a role.

For the off-axis gabbroic samples in the Red Sea, crystallization pressures have been determined using the methods of Nimis et al. (1999), which gives values of 2.8 ± 0.12 kbar and 3.0 ± 0.26 kbar (Ligi et al., 2018). Using the same methods, typical values for temperature input (Soesoo, 1997), comparable petrographic assemblages from the SWIR (Ozava et al., 1991) and the calibration for tholeiitic series. Pressure estimations after Putirka et al. (1996) calculations fall in the same range with pressures from 1.0 to 5.0 kbar, though most data points fall between 2.0 and 4.1 kbar. In both cases, the calculations only work for the core compositions, not for the rims that are far out of equilibrium with a nominal coexisting fluid. However, both barometer models are calibrated based on extrusive samples rather than gabbroic rocks and require equilibrium conditions, which is not entirely the case. Therefore, these methods provide only first-order information on the depth. Thus, while the crystallization pressures are likely lower than for the off-axis samples, the absolute values may be off. Assuming an oceanic crustal thickness of 5–10 km (Rihm et al., 1991), these crystallization pressures would indicate mid-lower crustal to mantle depths. However, the comparably evolved nature of samples from the Atlantis Bank, SWIR occura in depths of >400 m below the seafloor (Boulanger, 2020) and thus imply a crystallization of the Red Sea gabbro fragments in a shallow magma reservoir—consistent with the estimated depth of the sill intrusion. Considering the constraints from the geochemistry and the comparisons with gabbros from other slow-spreading ridges, the barometer calculations’ uncertainty, and the models’ limitations for these types of disequilibrium plutonic rocks, it is likely that the pressure calculations are overestimated. We therefore conclude that these gabbros likely occurred within the upper part of the oceanic crust. The coordination of aluminum in the clinopyroxene (Figure 6F), a pressure-sensitive proxy, supports the samples to be ‘regular’ oceanic gabbros, forming in a low-pressure environment.

Based on their distance from the rift axis, spreading velocities, and geobarometric estimations, Zabargad (8–9 kbar) gabbros were interpreted as being formed by underplating and gabbros from Brothers Islands (2.5–3.5 kbar) as shallow intrusions into a progressively thinned continental crust (Bonatti and Seyler, 1987; Ligi et al., 2018). The off-axis samples from Brothers Island and the QUSEIR drill hole show a wide range of similarities with our samples in chemistry, all lacking evidence for contamination with continental lithosphere. Pressure calculations for Brothers Island match with our numerical clinopyroxene estimations (Bonatti and Seyler, 1987), but these are affected by the same limitations as we faced. This could be a similar oceanic crustal origin for those samples, which would support studies that argue for a more comprehensive extension of oceanic crust in the central and northern Red Sea (Sultan et al., 1992; Augustin et al., 2021). On the other hand, the Red Sea Rift gabbros show a significantly lower Al^VI proportion than primary clinopyroxene from Zabargad Island. This suggests lower crystallization pressures for the rift axis gabbros and could suggest a transitional stage between oceanic crust and continental rifting.