A high-resolution bathymetry and backscatter survey was carried out in February 2022 from the OSS Handin Tide using the 6000 m depth-rated Hugin Superior AUV (autonomous underwater vehicle) provided by FUGRO. The AUV was equipped with a Kongsberg EM2040 Mk2, 200 kHz multibeam echosounder, EdgeTech Subbottom profiler, a CTD (conductivity, temperature, and depth), and a USBL (ultra-short baseline) system for navigation. The AUV flew at a constant altitude of 90 m in bottom-following mode. The area mapped covers 42.7 km² of the eastern AVR of Hadarba Deep and parts of its sub-circular axial high. The navigation data were post-processed in the software NavLab before being merged with the bathymetry data and post-processed using the software Qimera from QPS. The resulting computed grid has a cell size of 2 m ( Figure 2A ). Multibeam backscatter was mosaiced in QPS FMGT from the raw data and cleaned in Generic Sensor Format (.GSF) exports ( Supplementary Figure 1 ). The final mosaic has a pixel size of 0.8 m.

Sub-bottom profiles, with a vertical resolution of 10 cm (given by EdgeTech) were acquired along the AUV track and imported in SonarWiz 7 software from Chesapeake Technology to determine sediment thickness. Sediment thickness extracted from sub-bottom profiles is used to infer the age of the volcanic events. To avoid underestimation on steep slopes, or overestimation linked to sediment pounding, the sediment thickness was extracted from regions with relatively low variations in relief, with no significant change in sediment thickness. Sedimentation rates reported from the central and northern Red Sea are highly variable, as it experiences both pelagic sedimentation, dust input from the continents, and local turbidity currents. In the study area, sedimentation rates are not known, and in the absence of sediment cores, the sediment rate used in this study is estimated from the very limited reference data of the overall Red Sea. Here we apply an average sedimentation rate of 14 ± 3 cm/ka based on data from Stoffers and Ross (1974) and Kuptsov and Palkina (1986). The resolution limit of 10 cm of the sub-bottom profiler is used to infer the maximum age of eruptive units not displaying sediments in the profiles. Where no sediment thickness was determined, relative ages between eruptive units are determined based on the stratigraphic relationships between lava flows, patterns of fracture distribution, and geometry at the flow front as well as backscatter intensity (BI).

Tectonic features are manually digitized based on the bathymetry, derive slope map and backscatter mosaic. Faults are digitized following the fault crest, and fissures are digitized following their centerlines. Tectonic density is calculated using the “Kernel Density estimation” feature of SAGA Next Gen in QGIS using a search radius of 500 m and a pixel size of 10 m.

The tectonic strain was estimated along axis-perpendicular profiles, spaced roughly 500 m apart ( Supplementary Figure 2 ), based on the measurements of fissure width (W), vertical fault throw (D), and dip (α). Fissure width was measured on the backscatter mosaic. Indeed, small fissures (< 4 m wide) are not clearly evidenced in the bathymetric data, and the fissure width is generally overestimated due to the effect of data resolution and interpolation. Fault throw (D) is defined as the depth difference between the top and bottom of the scarp, determined at the slope break with the surrounding seafloor. Fault dip α is defined as the maximum slope value along the scarp to avoid underestimation associated with the presence of mass wasting and/or post-faulting lava flows, along the fault scarp (see method Figure 2 in Le Saout et al., 2022). Horizontal fault heave (H) associated with individual fault is calculated by the equation:

The tectonic strain (Ts) along profiles is then calculated from the equation:

where H_tot is the cumulative horizontal heave (H and W) along a profile, and L is the length of profiles.

Here eruptive units are defined as one flow or several lava flows with similar morphology, emplaced at similar times, therefore corresponding to a given eruptive sequence (Chen et al., 2021). Eruptive units are manually digitized based on abrupt changes in seafloor morphology, flow direction, backscatter intensity, and/or tectonic pattern using the bathymetric map, its derived products (e.g., slope, ruggedness index), and the backscatter mosaic.

Each individual eruptive unit is given a dominant morphology determined from the quantile classification of the ruggedness index (RI) calculated using the “Terrain Ruggedness Index” feature of SAGA Next Gen in QGIS with a search radius of 4 ( Figure 2C ) and a relative sediment cover based on backscatter intensity (BI, Figure 2B ). The seafloor is subdivided into three RI categories: smooth (RI< 0.87), moderately rough (RI: 0.87 to 1.5), and rough (RI > 1.5). Based on BI, the seafloor is classified as: not to slightly sedimented (BI > -20.9 dB), moderately sedimented (BI: -20.9 to -24.85 dB), and fully sedimented (BI< -24.85 dB). Using RI and to a lower-level BI, the eruptive units have been classified into 4 categories: Smooth terrains are dominated by low RI and medium to low sediment cover, i.e., high backscatter ( Figure 2D , yellow-light green). The often-flat surface displays lava channels, collapses, and tumuli, while their lava front generally consists of rough seafloor. Sedimented terrains are also dominated by low RI but are characterized by low BI ( Figure 2D , dark green). Hummocky terrains are dominated by a high RI, and often form elongate ridges ( Figure 2D , light to dark blue). Smooth hummocky terrains are used to describe terrains with a moderate RI. Both hummocky and smooth-hummocky terrain are independent of BI. Flat-topped seamounts constitute a fifth class. They are subcircular, monogenetic edifices (Clague et al., 2000), with smooth-flat tops and rough flanks, that have been classified based on these specific characteristics, independent from BI and RI. Bias in the classification method can come from the proportion of faults and fissures that will increase the calculated RI, and thus the proportion of hummocky terrain, but also increase BI and provide lower apparent sediment cover. However, deformation is relatively low and appears to affect mainly rough terrains.

The volume of eruptive units is calculated assuming a pre-existing horizontal base whose depth is determined by the average depth of the lava flow fronts or by the depth of collapses for enclosed lava flows. The volume is calculated between this surface and the mapped seafloor depth. Due to the assumption of the horizontal base, volumes of lava flow burying pre-existing edifices tend to be overestimated, while eruptive units filling pre-existing depressions are likely to be underestimated. This method also does not account for hollow features in the lava flows (lava tunnels). Nevertheless, a simple positive or negative difference of 1 m would result in volume variations of only 0.3 to 2.5x 10⁶ m³, thus <4% of the estimated volume for all except two of the studied lava flows. These two thin lava flows (S3 and S4) have an uncertainty of up to 25% of the estimated volume.

Lava fragments of 5 locations on different young lava flows were collected by wax corer. Major elements analyses were carried out with a Jeol JXA-8200 ‘‘Superprobe’’ electron microprobe at GEOMAR, Kiel using an acceleration voltage of 15 kV. Major elements of the glasses were measured with a defocused spot of 5 µm and a beam current of 10 nA. Counting times were 20/10 s (peak/background) for Si, Al, Mg, Ca, Na and P, 30/15 s for Ti, Fe, K, S and Cl and 40/20 s for Mn and F. The average was taken of 12 analyses per sample with 3 analyses for 4 glass chips to assure homogeneity. Mineral point analyses of plagioclase (Pl), clinopyroxene (Cpx) and olivine (Ol) were measured with a beam current of 20 nA and a focused beam spot for Pl, and 1 µm spot for Cpx and Ol. For all elements counting times of 20/10 s were used, apart for Ti and K in Pl and Ti and Cr in Cpx and Ol (30/15 s). For calibration and monitoring of data quality, we used natural reference samples from the Smithsonian Institute (Jarosewich et al., 1980). Relative analytical precision is generally <2.5%, but for the glasses up to 20% for MnO, P₂O₅ and Cl and up to 30% for SO₃ and F; for Pl up to 10% for FeO, MgO and K₂O and 25% for TiO₂ and for Cpx and Ol up to 25% for MnO and Cr₂O₃ and up to 75% for K₂O and NiO.

The 2 m AUV bathymetry reveals numerous faults (n=498) and fissures (n=1488) ( Figures 3 ; Table 1 ) not detectable on the ship-based bathymetry ( Figure 1 ). Faults are 25 to 3033 m long, averaging 231 m, and the fault scarps have vertical throws ranging from 1 to 130 m, averaging of 20.7 m ( Figure 3B ), with maximum apparent fault dips ranging from 7° to 85° (mean value: 52°, median value: 54° and maximum dip frequency: 70°; Figure 3C ). The major vertical offsets are concentrated on three faults (up to 130 m high) that delineate three long, parallel, major bathymetric steps that extend beyond the AUV data coverage (from west to east: W1, W2 (east dipping), and E1 (west dipping; Figure 3A ) and can be delineated on the ship-based bathymetry up to 7.5 km north of the mapped area ( Figure 1 ). This distribution of the deformation resulted in the development of a 3.6-4.5 km wide graben between W1 and E1 (here referred as axial valley), with a 1.5-1.8 km wide inner graben between W2 and E1. Most of the remaining faults are located within this axial valley between W1 and E1, and distributed along those bathymetric steps forming elongated fault systems, as shown by the fracture density map ( Figure 3A ). Fault scarps are dominantly east-facing (304 east-facing vs. 196 west-facing), but west-facing faults have on average larger vertical throw (27.0 vs 18.0 m respectively; Table 1 ). In contrast, the fissures (10–780 m long, 1.7 to 15 m wide, and up to 7.7 m deep) show no significant pattern and have generally no correlation with these fault systems. Faults and fissures are dominated by a 142.3° ± 27.36°N orientation, almost orthogonal to the spreading direction [42.9 ± 2.9°N at 22.5°N/37.75°E (Argus et al., 2011); Figure 3A ]. While some faults and fissures are oblique to the general trend (i.e., at the northwest corner or on the flat-topped seamount SM4; Figure 4A ), their low number does not affect the mean azimuth.

The tectonic strain, estimated along thirteen axis-perpendicular profiles (P4-P8 and P10-P17 from north to south in Figure S2 ) is shown in Table 1 . Cumulative fault throws and, thus tectonic extension were measured within the axial valley, between the top of E1 and W1 fault scarps ( Figure 3B ). No values were determined along profiles P1-P3 and P9, as they do not capture W1 in their profiles. The calculated apparent extension between E1 and W1 ranges between 2.5% and 11.1%, depending on the profile and the value used for the dip (measured, mean (54°) or most common (70°); Table 1 and Figure 3C ). In the absence of a well-defined spreading axis, as can be seen on fast- and some intermediate-spreading segments (e.g., Chadwick and Embley, 1998), the ridge axis is estimated at the center of the graben delineated by W2 and E1, corresponding to the axis of the youngest lava flows and fitting to a change in direction of the major faults (see section 4.3 and Figure 4 ) to evaluate the extension of the western and eastern side. Figures 3D, E show a clear asymmetry in accommodation of the extension between East and West sides of the ridge axis. Indeed, a similar amount of extension is accommodated over a shorter distance on the east side of the axis, e.g., 100 m of horizontal extension is accommodated over a length of 0.7-1 km on the east versus 1-3.2 km on the west side of the axis ( Figures 3D, E ). This is clearly evidenced by the overall higher tectonic strain side of the ridge ( Figure 3E ). The asymmetry is also reflected in the style of the deformation with a larger amount of fissure on the west side of the ridge, which is also shown by the difference between cumulative fault throw and cumulative extension.

In total, 90 eruptive units have been identified based on their distinct morphology, tectonic pattern, and/or sediment thickness. Individual flow units vary from 4.6x10² to 6.4x10⁶ m², of which 41 have a surface exceeding 2x10⁵ m² ( Figure 4 ; Table 2 ). Eruptive units consisting of smooth terrains are the most common type (27 eruptive units covering a total surface of 14.8 km² or ~34.7% of the mapped area). Smooth terrains are observed all over the mapped area; however, they are dominant in the east between W2 and E1. 37% of the smooth eruptive units display lava flow collapses (15 to 600 m in diameter) or inflated structures such as tumuli. Hummocky terrains are the second most common morphology, with 41 eruptive units over a surface of 13.7 km² (32.2%). They are mainly located in the southwest section of the mapped area forming either elongated units up to 4.2 km long or small individual cones. Smooth-hummocky terrains, with a total of 14 units over 11.0 km² (25.8% of the mapped area), are mostly observed in the northwest. This category includes SH1, which covers most of the surface of a large dome volcano (3.5 km in diameter; Figure 2A ). Flat-topped seamounts are rare, only 4 seamounts covering a surface of 2.2 km² (5.2% of the mapped area) were identified. They occur along W1 and E1 and west of the axial valley. A small portion of the mapped seafloor (< 1 km² or<2.1%) has both high RI and low BI (< -26 dB outside of tectonic structures) with no evidence of volcanic structures (i.e., collapses and tumuli). They are considered as sedimented terrains and are only observed south and west of W1, thus outside of the axial valley.

The eruptive volumes have been determined for sixteen eruptive units characterized by either well-defined flow front or deep collapses enabling estimates of flow thickness. These selected eruptive units are labeled in Figure 4A . Eruptive volumes of 4.2x10⁶ to 146.4x10⁶ m³ have been calculated for smooth eruptive units, 35.45x10⁶ to 38.22x10⁶ m³ for smooth-hummocky units, and 37.6x10⁶ to 206.6x10⁶ m³ for hummocky eruptive units. The four flat-topped seamounts have volumes estimated at 21x10⁶ to 99x10⁶ m³ ( Table 2 ). Several of the measured flows are only partially mapped (i.e., S1, S4, S5, and SH3); thus, their calculated volume is most likely underestimated.

Rock fragments were collected from units S2 (2 locations: from the central cone (S2-c) and the rim (S2-r) of the lava field), S4, S5, and SH2. The glass and mineral chemistry is presented in Figure 5 and Supplementary Table 1 . Glasses are of typical mid-ocean ridge basaltic composition and span a narrow range, falling within the fields defined by literature samples of surrounding Red Sea Deeps (Thetis, Hadarba, and Hatiba Deep; Ligi et al., 2012; van der Zwan et al., 2015; van der Zwan et al., 2023a); Figure 5 and Supplementary Table 1 ). While most of the glasses are homogeneous, sample SH1 shows two distinct compositions for 2 glass chips each (SH2a and SH2b), indicating local variations (within the 30 cm of wax core sampling). These variations within one sample location, as well as the variation within unit S2 (S2-c and S2-r), is larger than the variation between the different lava flows ( Figure 5 ). The samples lay on a fractional crystallization trend of decreasing SiO₂ (50.44-49.99 wt%), Al₂O₃ (14.00-13.24 wt%) and CaO (11.51-10.47 wt%), and increasing TiO₂ (1.75-2.32 wt%), Na₂O (2.67-2.82), K₂O (0.14-0.20 wt%), P₂O₅ (0.13-0.21 wt%) SO₃ (0.32-0.38 wt%) and Cl (0.01-0.04 wt%) with decreasing MgO (6.15-6.82 wt%). Proxies for melt degree (Na₈), depth of melting (Fe₈), mantle source variation (K/Ti) and hydrothermal influence (Cl/K), are all very homogenous between the samples and fall within the Red Sea fields defined by literature samples ( Figures 5B, C ; Supplementary Table 1 ).

Small plagioclase minerals (0.02-1 mm in size) are present in all samples and have a bytownite composition, with limited anorthite (An) content variations between An₇₅-An₈₂. This variation can be observed within a sample, and plagioclase compositions between samples are indistinguishable ( Figure 5D ). Pyroxenes and olivines are observed in samples S2-c, S4, and S5 as microlites of 0.1-0.2 mm. Olivines have forsterite (Fo) contents of Fo₇₆ – Fo₈₁, while the clinopyroxene have an augite composition ( Figures 5E, F ). Similar as the plagioclases, the clinopyroxene, and olivine compositions overlap between samples.

Due to the rough terrain, sediment thickness could only be extracted from 40 positions with low relief and constant sediment thickness along the AUV sub-bottom profiles ( Supplementary Figure 3 ), characterizing 21 of the eruptive units. Measured sediment thickness ranges from 0.4 ± 0.1 to 4.2 ± 0.1 m ( Figure 6A ). Three of the eruptive units have sediment thicknesses below the limit of detection of the sub-bottom profiles, thus< 10 cm. Completely sedimented regions are found to be covered by more than 1.4 ± 0.1 m of sediments. Hummocky terrains have sediment thicknesses ranging from 0.8 ± 0.1 to 2.1 ± 0.1 m but are generally more than 1.6 ± 0.1 m, while smooth terrains cover the full ranges of data from <10 cm (below resolution limit) to 4.2 ± 0.1 m of sediments. The large volcanic dome has measured sediment thickness ranging from 1.2 ± 0.1 to 3.7 ± 0.1 m, with relatively constant sediment thickness on its summit (1.2-1.6 ± 0.1 m) and a general increase observed in its flanks without clear evidence of change in flow units. Sediment thickness estimates are in correlation with variations in backscatter intensity.

Flow morphology is commonly used as a proxy to infer the rheological and physical properties of an eruption. Based on the comparison of the determined morphologies with those of similar resolution studies with ground truth data (e.g., McClinton et al., 2013; Chen et al., 2021), smooth terrains seem to be associated with the emplacement of sheet and lobate flows, hummocky terrains with the emplacement of pillow mounds or pillow ridges, while smooth hummocky terrains mark a transition between lobate and pillow flows. Laboratory experiments have shown that the morphology of deep (>1500 m) submarine lava flows is controlled by the lava viscosity, underlying slope, and local flow rates (Bonatti and Harrison, 1988; Gregg and Fink, 1995; Gregg and Smith, 2003). Sheet flows transition toward lobate and pillow flows by increasing viscosity and decreasing extrusion rate and underlying slope. Constraining the underlying slope is challenging as it is locally controlled by volcanic edifices buried by more recent eruptions. Nevertheless, in Hadarba Deep, the regional bathymetry deepens towards the south at an angle of 1.5 to 3.5° (excluding the dome volcano) without clear variations between seafloor morphology, indicating this does not have a major effect. Within our mapped area, only 4 eruptive units were sampled ( Figure 4A ), and none of them are associated with hummocky terrains. The homogeneous compositions of the basaltic glasses and minerals and the chemistry of the lava flows that fall along fractional crystallization trends point to a role of crystallization to explain the limited spread. Calculation of crystallization pathways at seafloor depth (0.2 kbar), mid-crustal or lower crustal depth (0.8 and 1.6 kbar; assuming a 5 km crust) with Petrolog3 (Danyushevsky and Plechov, 2011) indicate that the limited spread in lava chemistry can be explained by variations in degrees of crystallization of around 20% (depending on the exact depth; Figure 5A ). Melting degree and depths (as indicated by Na₈ and Fe₈, cf. Langmuir et al., 1992) are also consistent between the samples, and similar to other Red Sea samples, overlapping with ‘normal’ mid-ocean ridges not influenced by mantle plumes ( Figure 5B ). Low K/Ti and little elevated Cl/K, which is on the low end of the Red Sea field, indicate that all magmas have a relatively refractory source, and underwent only little assimilation of hydrothermal crust ( Figure 5C ; cf. van der Zwan et al., 2015). Therefore, all magmas seem to have had a similar source, underwent similar magmatic processes, and potentially are related to the same event. The overall smooth terrain of these lavas is consistent with their primitive low-viscosity basaltic composition (glass MgO > 6; Pl >An₇₅; Ol >Fo₇₆). In addition, the low crystallinity and lack of phenocrysts (<1 mm crystals) in the samples also imply little crystallization in stagnation levels, in agreement with the high-extrusion rates of low-viscosity lava. However, analyses of the glass samples and their minerals show no significant compositional variations between the lava flows that would explain changes in morphologies between smooth and smooth-hummocky. Similar observations have been made along the Mid-Atlantic Ridge (Cann and Smith, 2005) and Galapagos Spreading Center (McClinton et al., 2013), and on a larger scale for different volcano types in the Red Sea (Augustin et al., 2016), suggesting that extrusion rates influenced flow morphology to a greater extent than the chemistry. Thus, the differences in morphology observed in Hadarba Deep could be used as a proxy for extrusion rates. The transition between smooth to smooth-hummocky and hummocky terrain is likely associated with a decrease in extrusion rates, potentially marking the end of eruptive sequences.

The maximum measured sediment thicknesses (4.2 ± 0.1 m, Figure 6A ) and an average sedimentation rate of 14 ± 3 cm/ka (Stoffers and Ross, 1974; Kuptsov and Palkina, 1986), gives a maximum age of 30 ± 9 ka for the mapped area and 15 ± 5 ka within the axial valley (i.e., between W1 and E1). If the volcanic activity was limited at the ridge axis and given the spreading rate in this region (11-12 mm/yr; Argus et al., 2011; Viltres et al., 2022) and the width of the axial valley (~3.5-4.5 km), the maximum age of the crust within the axial trough should reach ~400 ka; or ~800 ka, if we consider that the mapped region corresponds to the eastern tip of an overlapping spreading center and that the total extension could be distributed between the two axes. Such variation between this maximum crustal age and the age of the upper flows is similar to observations made along the Mohns Ridge, where ~50% of the valley floor have ages younger than ~25 ka (Stubseid et al., 2023) and reflect the wide region of crustal generation along slow- and ultraslow spreading ridges (Perfit and Chadwick, 1998; Chen et al., 2021).

Within this timeframe, different eruptive units can be identified based on their morphology ( Figures 2 , 4 ), and relative age ( Figure 6 ), as well as the tectonic pattern ( Figure 3 ). They are used to determine eruptive phases and investigate the evolution of the eruptive activity of the study area in Hadarba Deep. Three volcanic phases (V1-V3) have been identified within the axial valley based on recurrent patterns between successive eruptive units and their sediment thickness. Phase V1 is associated with the emplacement of elongated hummocky terrain separated by smooth and smooth-hummocky terrain. It is also associated with the formation of the four flat-topped seamounts. The abrupt variation in fault and fissure distribution between eruptive units ( Figure 6B ) within this V1 indicates alternation between hummocky eruptions and smooth to smooth-hummocky eruptions, with at least three such cycles evidenced. This would suggest cyclic variations in extrusion rate and possibly magma supply. However, whether the episodes of deformation occur in between individual eruptive events or are limited to a period of lower extrusion rate associated with the waning of magma supply at the end of each phase remains unclear. The sediment thickness varying from 1.5 ± 0.1 to 2.1 ± 0.1 m indicates ages between ~10 ± 4.5 and 15 ± 5 ka. Eruptive units emplaced during this period are primarily at the surface on the southwest section of the study area, but observed all across the axial valley (from E1 to W1), and thus potentially more covered on the eastern side. This, therefore, indicates a period associated with a wide zone of dike injections (exceeding 4.5 km) over the whole axial valley.

Phase V2 appears to mark the last building phase of the dome volcano but can also be found east of E1. The relatively constant sediment thickness at the dome summit (1.2 to 1.4 m) indicates ages between ~8.5± 3 and 10 ± 3.5 ka. While the increase of sediment thickness on its flank (up to 3.7 ± 0.1 m; Figure 6A ), could result from the redistribution of sediment down the slope, an initial construction of the edifice anterior to V1 cannot be excluded. V2 is mainly associated with the emplacement of smooth and hummocky smooth terrains with lava channels exceeding 1.5 km but is topped by hummocky terrain, which attests to an overall decrease of volcanic activity toward the end of phase V2. No eruptive units with similar ages have been found south of the volcanic dome, and V2 could mark a regression phase of the volcanic activity with magma focusing towards the segment center (axial high). The eruptive units of V2 have experienced significant deformation, especially along W1, W2, and E1 ( Figures 3A , 4A ). However, there is no significant variation in the tectonic pattern between the different units of V2, indicating that the deformation of the V2 lava flows is posterior to the last eruption of this volcanic phase.

Finally, phase V3 marks the youngest volcanic phase, indicated by their strongest backscatter signals. Three eruptive units (S1, S2 and S4) have sediment thicknesses of<10 cm, confirmed by limited sediments in the wax cores, and thus are younger than 700 yrs. No ages could be determined for most of the earlier stages of V3, but most eruptive units (70%) appear to be of similar age or younger than S5, which has a thickness of 40 cm, equal to 2.9 ± 1.6 ka. Phase V3 is dominated by the emplacement of smooth terrains (i.e., sheet flows and lobate flows), pointing to elevated extrusion rates. Not dissimilar to V1, V3 seems to have experienced changes in morphology with the occurrence of smooth terrains by high extrusion rates, alternating with periods of slightly decreased extrusion rates forming smooth-hummocky terrains, indicating episodic activity associated with waxing and waning of the magma supply. However, the lower amounts of hummocky terrain point towards overall higher magmatic activity. This is also indicated by the Na₈ ( ± 2.4) and Fe₈ ( ± 10.5) on the lower end of the mid-ocean ridges spectrum, indicating higher degrees of melt at shallower average depths, more similar to faster-spreading ridges ( Figure 5B ). High eruption rates are consistent with the absence of extensive magma storage as indicated by their low crystal content and lack of phenocrysts and similar compositions. The narrow distribution of the lava flows across the axial valley, indicates that dikes injections are centered within a much narrower region of the valley (<1.5 km), and lava flow emplacement is mostly constrained by W2 and E1.

Overall, the high proportion of smooth flows ( Figure 4 ), and the low tectonic strain (<10%, Figure 3 ), are similar to those of intermediate spreading segments (e.g., at 86°W and 92°W along the Galapagos spreading center; McClinton et al., 2013) and those of SWIR 50°28’E (Chen et al., 2021). Together those parameters indicate that the mapped region is part of a magmatically robust segment in concert with it being part of an axial high, and this for at least the last 15 ± 5 ka. However, the above interpretation also shows that the volcanic activity over this period is not steady. The variability between the different volcanic phases indicates periods of (1) lower extrusion rates over the whole axial valley (V1), (2) regression of the volcanic activity toward the axial high, but with local higher extrusion rates (transition V1 to V2), (3) increase of tectonic extension (between V2 and V3), and (3) renewed volcanic activity with higher extrusion rates (V3). This could indicate a decrease of magma supply from V1 to V2, followed by an increase during V3. These variations could also explain changes in the width of the injection zone. Previous studies have shown that periods of robust magma supply dominated by high extrusion rate eruptions develop a narrower well-defined axis, while during periods of lower magma supply, the activity appears unfocused, occurring within the entire width of the axial valley (50°28′E, Chen et al., 2021). In addition to the variations between each volcanic phases, there are also significant changes in flow morphology (i.e., extrusion rates) between eruptive events. Based on the number of observed cycles of morphological changes and the ages of V1 and V2, variations in extrusion rate within the phases show a periodicity of 1-2 ka.

Accurately dating volcanic events along a MOR is challenging, and an eruption history over relevant periods of time is rare. Up to date, only a few repeated eruptions have been directly observed (9°50′N EPR, Axial Seamount (JdFR), and the Coaxial segment of the JdFR), giving intervals of recurrence of 4 to 20 years (Rubin et al., 2012; Clague et al., 2017), while most other information is from indirect methods. Eruption frequency is often estimated by either (1) dating eruptions sequences using radiometry, paleointensity, or estimation of the sediment thickness and sedimentation rate (Bergmanis et al., 2007), (2) calculating the number of meter-wide diking events needed to accommodate the magmatic extension (e.g., Hooft et al., 1996; Curewitz and Karson, 1998; Colman et al., 2012), or (3) based on the time-averaged rate production of the extrusive layer and average flow volume (Perfit and Chadwick, 1998; Sinton et al., 2002; Rubin et al., 2012). This latter method infers that the frequency of eruption decreases with the spreading rate, and recurrence intervals >1000 yr should be expected for a slow- and ultraslow- spreading segment (Rubin et al., 2012).

Assuming a steady state of the magmatic system, the number of distinguishable eruptive units per volcanic phase and sediment ages can be used to calculate an interval of recurrence of ~106 ± 135 yr (in V1), ~115 ± 600 yr (in V2) and ~250 ± 116 yr (last 2.9 ± 1.6 ka of V3). Note that these estimated eruption rates have large errors resulting from the high uncertainty in the sedimentation rate. The three youngest flows (Flow S1, S2, and S4), whose ages are better constrained (<700 ± 200 yr), also indicate an interval of recurrence of<240 ± 50 yr, thus a much higher frequency than expected for ultra-slow spreading segments. Such short intervals could be related to the age uncertainties and the method used to delineate the eruptive units. Indeed, previous studies have shown that disconnected flows could be part of the same eruptive event (e.g., Colman et al., 2012; Yeo, 2014; Yeo et al., 2016; Clague et al., 2017), leading to an underestimation of the frequency. Phase V1 and V2 have a high proportion of disconnected cones or mounds. These smaller edifices might be part of larger eruptive events, and contribute to an overestimation of the eruptive frequency. But in the absence of geochemical information, integrating those smaller edifices into larger events remains challenging. However, in phase V3, while geochemical analyses indicate a possible relation between the lava flows, variations in BI indicate age variations with recurrence interval consistent over 3 ka and 700 yr periods.

The high eruption frequency over the last 15 ka is likely to be related to the long-term magmatic state of the segment. Indeed, the magmatic budget is known to fluctuate not only at the time scale of individual eruptions (Bowles et al., 2014; Clague et al., 2017) but also over a larger time scale. Such variations are known to influence flow morphology and eruption frequency, which become more frequent as the magmatic budget increases (Colman et al., 2012). Faults located ~6.5 km east and west from the current ridge axis may mark the end of the last high-magmatic phase and would indicate a long-term magmatic cycle of ~590 ka, similar to those found along slow- and ultraslow-spreading segments (150-500 ka; Cordier et al., 2010; Rioux et al., 2016; Klischies et al., 2019; Chen et al., 2021). Faults located ~6.5 km east and west from the current ridge axis may mark the end of the last high-magmatic phase and would indicate a long-term magmatic cycle of ~590 ka, similar to those found along slow- and ultraslow-spreading segments (150-500 ka; Cordier et al., 2010; Rioux et al., 2016; Klischies et al., 2019; Chen et al., 2021). Variability in long-term magmatic supply can also be observed off-axis in vertical gravity gradient (VGG) data (Augustin et al., 2021): in contrast to Thetis Dome and Hatiba Mons that form rift-perpendicular ridges indicating long-term magma focusing, the Hadarba segment center shows a patchy VGG pattern pointing to larger variations in magma supply. Nevertheless, considering (1) the consistency of eruption intervals between the phases, (2) the generally very low tectonic strain (<10%, Figure 3E , see 5.4), and (3) the high proportion of smooth terrains even during V1, a relatively large magma supply has been located under this segment, for at least 15 ka, without clear sign of decline. In such magmatically robust segment, a recurrence interval of a few hundred years seems realistic. This would indicate that ultraslow-spreading ridges could sustain robust magmatic activity with high eruption frequency over several kyrs.

The eastern axis of Hadarba Deep has a well-defined, ~3.6-4.5 km wide axial valley offset from the rift flanks by 1.5 to 104 m high fault scarps (W1 and E1, Figures 2A and 3A ). The shallow axial valley (<105 m deep) and the low tectonic strain between W1 and E1 (<10.9%), require that 90% of the spreading is accommodated by magmatic intrusions, and are thus indicative of a magmatically robust segment (e.g., Sempéré et al., 1993; Hooft et al., 2000; Olive and Dublanchet, 2020), at least in geologically recent time. This is in agreement with the low seismicity ( Figure 1B ) that has been proposed to indicate an increase in magma extension (Metz et al., 2013; Augustin et al., 2016). The differences in fault patterns and tectonic strain with respect to the current ridge axis ( Figure 3 ), indicate an asymmetry for the accommodation of the extension between W1 and E1. An asymmetry in the extension and fault pattern is not uncommon at a segment end (e.g., Escartín et al., 1999 and references therein). As the study area is the eastern branch of an overlapping spreading center ( Figure 1 ), the asymmetry could indeed be a result of a change in the rheology of the lithosphere, a difference in the spreading rate between the eastern and western flanks, or a difference in the width of the active deformation zone. Potential differences in spreading rate cannot be constrained by the available data, and although there is a lack of large vertical offset outside of the mapped region ( Figure 1 ), it cannot be excluded that the zone of active deformation extends further east and west. Indeed, the active deformation zone on slow- and ultra-slow spreading segments can exceed 15 km (Escartín et al., 1999; Chen et al., 2021), over three times the width of the map region. However, based on the prolongation of W1, W2 and E1, the asymmetry appears to continue north, well outside the overlapping region, into the axial high ( Figure 1 ). Thus, it is unlikely merely related to it being at a segment end. The relationship between the fault pattern and the youngest eruptive unit seems to indicate that at least part of the asymmetry is related to the recent change in volcanic activity. Deformation along E1 and W2 seems to have primarily occurred between the end of V2 and the early phase of V3 when the eruptive activity became more focused on the eastern section of the axial valley. Such relationship, may indicate that the localization of the deformation is controlled by the magmatic activity.