MainTwo-thirds of Earth’s surface was created along the 65,000-km-long MOR system, a network of narrow tectonic boundaries in which new lithospheric plates form and drift apart at a scale of centimetres per year. Under MORs, mantle rises and partially melts on decompressing. Buoyant magma ascends towards the ridge axis, where it crystallizes as new oceanic crust within spreading segments bounded by TFs or non-transform offsets. Magmatic emplacement, however, rarely accounts for all of the plate divergence and part of the extension is accommodated by slip on normal faults of all sizes1. Over hundreds of thousands of years, this interplay of magmatic and tectonic deformation shapes abyssal hills, which cover most of the ocean floor10,11. Little is known, however, on how these processes unfold on short timescales. Seafloor spreading is thought to proceed as a succession of ‘quantum’ events of extension, involving earthquakes and dyke intrusions12. These events can last several hours to several months and recur on decadal timescales, depending on the local spreading rate and magma supply. Many questions remain about how stresses accumulate between these events13,14, the degree to which activity on spreading segments and nearby TFs is synchronized15 and why MORs release so little seismic energy given their rapid time-averaged deformation rates8,9. These knowledge gaps originate from the challenge of simultaneously and continuously monitoring, for several years, the low-level seismic activity and the horizontal and vertical displacements of the seafloor along submarine plate boundaries16,17. We took on this challenge by installing an autonomous seismogeodetic observatory on an active MOR segment and its abutting TF and were lucky enough to capture a notable seafloor spreading event two months into our experiment.Setting of the OHA-GEODAMS experimentThe OHA-GEODAMS (Observatory with Hydro-Acoustics and Geodesy near Amsterdam Island) observatory was deployed late February 2024 across the axis of the SEIR at 37° 07′ S (segment I1 (refs. 18,19)) and across the nearby Amsterdam TF (Fig. 1a). This region marks the northernmost extent of the Saint Paul–Amsterdam volcanic plateau, which the SEIR is now splitting apart18,19,20. The local spreading rate is 61 mm year−1 based on magnetic anomalies21,22 and 63 mm year−1 according to space geodesy23. The seafloor produced at segment I1 is typical of intermediate-spreading MORs, with an approximately 2,000-m-deep axial valley flanked by fault-bounded abyssal hills, hundreds of metres tall and spaced roughly every 2 km (Fig. 1b). This area experiences one to two Mw ≥ 5 earthquakes each year either on the Amsterdam TF or the Boomerang TF that bound segment I1 and has the advantage of being regularly visited by R/V Marion Dufresne on her yearly voyages to maintain the French scientific research stations in the Southern Indian Ocean islands.Fig. 1: Tectonic setting and seismicity of the captured seafloor spreading and transform faulting event.a, This event occurred along the SEIR segment I1 (in the black rectangle), located on the Saint Paul–Amsterdam plateau in the Southern Indian Ocean (oblique Mercator projection). The seismogeodetic experiment included a wide array of autonomous hydrophones (stars), two arrays of acoustic ranging beacons and one BPR (in the purple rectangles) and repeated swath-bathymetry surveys of the axial valley and its intersection with the two abutting TFs. b, Earthquake crisis of 26 and 27 April 2024. GCMT focal mechanisms (red with Mw magnitudes in black) and ISC events (orange stars with mb magnitudes in blue) from land-based catalogues are numbered sequentially (italicized numbers; Extended Data Table 1) and were all relocated from the recordings of the hydrophone array, which also recorded many lower-magnitude hydroacoustic events (grey dots). Horizontal displacements were measured along direct-path ranging baselines (orange lines) between acoustic beacons (white numbers in blue circles) distributed in two arrays. Vertical displacement in the axial valley floor was measured with a self-calibrating BPR (cyan hexagon).Our observatory comprises an array of five autonomous hydrophones, encompassing the whole Saint Paul–Amsterdam plateau (Fig. 1a), to monitor the regional seismic activity. Passive acoustics (Methods) have proved to be very effective in locating low-magnitude seismic activity in MOR contexts, with a location accuracy of 1–2 km or better and completeness on the order of mb = 3.9 (refs. 15,24,25,26). They have been particularly successful at detecting seafloor spreading events involving dyke injection and eruption27,28,29,30. To monitor horizontal displacements at the ridge–transform intersection, we deployed 15 acoustic-ranging beacons or transponders (TP (refs. 3,7,31)) on the seafloor (Fig. 1b), 11 straddling the axial valley (TP1–TP11) and four across the Amsterdam TF (TP12–TP15). To make baselines as long as possible, the beacons were mounted on high tripods and placed on ridge crests and cliff edges. Across the axial valley, beacons were set along two parallel lines, 1.3 km apart with common end-stations, to extend the lateral coverage of the array and to ensure redundancy in case of instrument failure. Overall, the network spans 11 km, with baselines ranging in length from 1,100 to 4,300 m. Owing to the rugged and elevated terrain, not all beacons can communicate with all of their neighbours; still, a total of 32 baselines can be monitored by direct-path ranging. The TF array forms a diamond spanning the Amsterdam TF, with four 1,940–4,940-m-long baselines. Ranging measurements were repeated every 4 h in the ridge array and every 2 h in the TF array to measure the relative displacements between transponders (Methods). Finally, we deployed a drift-controlled bottom-pressure recorder (BPR) between transponders TP6 and TP7 (Fig. 1b) to monitor vertical displacements of the axial valley floor (Methods). Even though all of these approaches have been used previously to study ground deformation associated with seafloor spreading2,3,4,5,6,7, this is the first time that they have all been implemented simultaneously on a MOR segment and abutting TF exhibiting substantial tectonic activity.Seismicity and seafloor displacementsOn 26 April 2024, at 19:56 UTC, a notable swarm of extensional seismicity began in the middle of segment I1 with five small events detected by the hydrophones, followed by a Global Centroid Moment Tensor (GCMT)32 Mw = 4.9 normal-faulting earthquake at 20:09, located under the axial valley (number 1 in Fig. 1b). This initial seismic activity, with a subsequent Mw = 5.1 and 16 lower-magnitude earthquakes, migrated more than 8 km towards the southeastern end of the axial valley until 20:35 (Fig. 2a). It was followed minutes later by a series of seven Mw ≥ 5 normal-faulting (GCMT), eight 4.8 > mb > 4.4 (International Seismological Centre (ISC))33 and four lower-magnitude earthquakes, which migrated more than 9 km from the segment centre to its northwestern end, in the direction opposite to that of the initial migration phase. Such migrating patterns are typical signatures of dyke propagation events along extensional plate boundaries34,35,36, although the migration rates involved in this case (2–3 m s−1, from the regressions shown in Fig. 2a) are an order of magnitude faster than those documented in analogous, albeit longer, sequences (Table 1 in ref. 29). Remarkably, the axial valley floor is 200 m shallower and narrower at the southeastern end of the segment compared with its northwestern end (Fig. 1b). A southeastward dyke intrusion from the segment centre may thus have been hindered by topographic stresses37, giving way to a more mechanically favourable northwestward intrusion.Fig. 2: Initial sequence of seismic events and vertical displacements of the valley floor.a, Sequence of events in the axial valley interpreted as the result of dyke propagations, initially to the southeast (N139) and then to the northwest (N319), at an average speed of about 3 m s−1. Along-axis distances are relative to the BPR location. GCMT focal mechanisms (green), ISC (orange stars) and hydroacoustic events (blue dots) are shown with their Mw (black) and mb (blue) magnitudes, respectively. The lower part shows the BPR recordings of the coseismic subsidence of the valley floor as the dyke propagated; episodes of subsidence exceeding 1.7 cm min−1 are shaded. Map view of the events and dyke propagations is shown in Extended Data Fig. 9. b, Longer-term subsidence (blue curve, 1 dbar ≈ 1 m) and temperature changes (orange curve) recorded by the BPR in the axial valley floor. The framed area corresponds to the time window in panel a. The brown curve is the cumulative number (N) of seismic and hydroacoustic events reported at the top of the panel and distributed between the axial valley and the two adjacent TFs.In the meantime, the BPR in the axial valley floor recorded a series of discrete subsidence events, that is, sudden pressure increases amounting to between 1.7 and 3.3 cm of vertical motion coincident with each of the first large normal-faulting events (Fig. 2a). Using the rupture size inferred from the GCMT magnitude38 and fault dips given by the focal mechanisms, we find that these earthquakes, modelled as circular ruptures in an elastic half-space (Methods), could not have occurred any deeper than about 5 km below the seafloor to cause the observed coseismic subsidence (Extended Data Fig. 1). These markedly shallow depths are in strong contrast to those estimated by the GCMT catalogue (12–16 km below the seafloor), probably because of the poor teleseismic coverage of the area.Subsidence of the axial valley floor accelerated after a Mw = 5.1 event at 21:03 (number 5 in Fig. 1b), in a stepwise manner, averaging about 5 cm min−1 while the putative dyke propagated to the northwest. When the propagation stopped at about 21:40, cumulative seafloor subsidence had reached 1.2 m (Fig. 2a). Subsidence then slowed abruptly and continued at a lesser rate of about 0.6 cm min−1 until 01:30 on 27 April (Fig. 2b). It slowed further to 1.2 cm day−1 until 00:00 on 3 May and beyond (Extended Data Fig. 2). The total subsidence amounted to 4.2 m over a period of 6 days, of which 83% occurred in the first 16 h. This is almost twice the total subsidence of 2.45 m measured over 12 days during the 2015 eruption of Axial Seamount, of which 91% occurred during the first 24 h (ref. 17). Like at Axial Seamount, we interpret this vertical displacement of the valley floor as resulting from the drainage of an underlying magma reservoir (for example, axial melt lens39). The rapid subsidence between 21:03 and 21:40 may specifically represent the drainage from the magma reservoir into a northwest-propagating dyke. Rising near-bottom water temperature recorded by the BPR suggests that lava may have reached the seafloor as early as 22:00 (Fig. 2 and Extended Data Fig. 2). Past that time, slower seafloor subsidence may reflect magma drainage through open fractures connecting the reservoir to the seafloor.Nearly concurrently, the direct-path acoustic ranging array straddling the axial valley recorded substantial horizontal displacements (Fig. 3). During the February 2025 maintenance cruise, given the challenge to reposition the beacons on the seafloor, only the central station TP7 was recovered to access its full dataset. Baselines remained fairly constant until the ranging session at 21:12 on 26 April, which nearly coincided with the Mw = 5.2 normal-faulting event at 21:09 (number 8 in Fig. 1b), at the beginning of the rapid subsidence phase of the BPR. Afterwards, baselines markedly changed but fluctuated about their new lengths, probably because of rapid changes in sound speed induced by fluctuating ambient temperatures (Fig. 2b). The temperature field was then certainly not uniform near the axial valley floor (that is, around TP7) and affected all acoustic travel times to the elevated beacons on the valley walls. Nevertheless, by averaging the ranging measurements before the ambient temperature increased by more than about 0.1 °C, baselines from station TP7 to the two stations located on top of the southwest valley wall (TP8 and TP9) clearly lengthened by up to 1.30 m, whereas, on the opposite side of the valley, baselines shortened by 0.53 m (TP4 and TP5) and 0.86 m (TP3; averaging windows shown in Fig. 3). These displacements are one order of magnitude larger than those measured across the Juan de Fuca Ridge at 46° N during the 25 January 1998 dyke intrusion at Axial Seamount4,31. However, our baselines do not show any obvious extension related to some inflation before the 26 April events, as was observed at Axial Seamount. All tripods tilted by several degrees during the swarm of nearby Mw ≥ 5 quakes. Because we cannot tell in which direction they tilted, the uncertainties in the baseline changes include the worst-case scenario (that is, acoustic heads fully tilting apart or closer). Extension between TP8–TP9 and TP7 continued until June 2024 and reached up to +1.80 m, whereas TP3–TP4–TP5 ended up closer to TP7 by −0.64 to −0.86 m (Extended Data Fig. 3). Notably, baseline TP6–TP7 also diminished. This could be because of the subsidence of the two transponders, as recorded by the built-in pressure gauge in TP7 (Extended Data Fig. 2c), possibly suggesting that the deflation of the magma reservoir was centred below the BPR and TP7 (Fig. 3), which both subsided by about 4 m.Fig. 3: Initial baseline changes in the axial valley, and extent and thickness of the new lava flows.Displacements in metres are measured from transponder TP7 along the coloured baselines to the other transponders (numbered circles), which are either lengthening (orange) or shortening (blue). Greyed patches with 10-m contours in light blue show the extent of the new lava flows, which exceed a thickness of 90 m in places (see greyscale in the upper left; volumes of patches 1 to 10 are given in Extended Data Fig. 6). Insets show the initial changes in the baselines with up to 1.30 m extension to the southwest (TP8, TP9) and 0.86 m contraction to the northeast (TP3, TP4, TP5). Coloured rectangles show averaging windows and 1σ uncertainty. The vertical grey line corresponds to the time of the rapid subsidence of the BPR (between 21:00 and 21:40 on 26 April; Fig. 2b). The baseline changes are interpreted as the combined result of a propagating dyke and a deflating magma lens at depth below TP7 (purple rectangle) and normal slip along a fault bordering the axial valley and dipping 60° inwards towards the dyke (Fig. 4). Longer-term changes in the baselines are shown in Extended Data Fig. 3.The initial displacements (Fig. 3) along a cross-axis line joining TP9 to TP3, through TP7 and TP5, provide critical constraints to evaluate the interplay of different tectonomagmatic deformation sources during the early stages of the spreading event, that is, fault slip, deflation of a magma reservoir and intrusion of a dyke. Station TP7 subsided by 3.75 m ± 0.55 m, whereas TP9 moved away from TP7 and TP5 and TP3 moved closer to TP7. Such massive subsidence is readily explained by a deflating magmatic sill40 that would also induce horizontal contraction bringing all transponders closer to TP7 (Extended Data Fig. 4). Lengthening of the baselines thus requires sources of extension between TP9 and TP7, such as the opening of a dyke and/or slip on a normal fault.Using 2D elastic dislocation models (Methods), we randomly sampled 10 million combinations of sill, dyke and fault geometries to assess how well they could account for the observed displacements (Extended Data Fig. 5). Of these, 2,203 yielded a satisfactory root mean square (RMS) misfit (<20 cm). They have in common: (1) a sill at least 3,500 m deep collapsing by 12–18 m; (2) a dyke rooted at the sill, extending to sub-seafloor depths of 83–377 m and opening by 1.0–3.3 m; (3) 0.6–3.1 m of slip on a normal fault down to a few kilometres; and (4) a total horizontal extension of 2.1–4.0 m, distributed between the dyke and the fault. Most models favour the dyke taking up more extension than the fault. As an illustration, Fig. 4 presents a plausible model with a sill, 2,500 m wide and 3,600 m deep, deflated by 9.8 m; a dyke rising to 57 m depth beneath the geodetic network and opening by 1.4 m; a northeast-dipping axial-valley bounding fault slipping by 1.9 m down to 1,500 m depth. Such a large amount of slip, one order of magnitude larger than expected from any Mw ≈ 5 earthquakes of the swarm (Methods), implies that the fault motion largely occurred aseismically. In this particular scenario, the total horizontal extension amounts to 2.4 m (60% dyke, 40% fault), which is equivalent to 38 years of spreading at 63 mm year−1.Fig. 4: Example of a 2D deformation model fitting the observed vertical displacements and baseline changes relative to TP7 along a profile perpendicular to the axial valley.a, Overall baseline changes (ΔL; Methods; continuous line) and vertical subsidence (dashed line). The overall horizontal displacement is 2.4 m (60% dyke, 40% fault). b, Geometry of the model components below a bathymetric cross-section: a 3,577-m-deep sill deflating by 9.8 m, a 1.4-m-wide dyke rising 3,520 m from the sill up to 57 m below seafloor, between TP9 and TP7 (310 m southwest of TP7) and 1.9 m of slip down to 1,526 m depth on a northeast-dipping normal fault (total RMS misfit = 17 cm). A schematic map view of the sill, dyke and fault is shown in Fig. 3. The individual contribution of each component—sill, dyke and fault—to the displacements is shown in Extended Data Fig. 4. This is one of the 2,203 solutions best fitting the data (RMS misfit < 20 cm) out of 10 million randomly sampled models. The only fixed parameters were the fault dip set at 60°, based on focal mechanisms, the fault location along a southwest valley wall between TP7 and TP9 and the sill width across axis set at 2,500 m and centred on TP7 (Methods). Coloured arrows show the expected type of motion on the fault, dyke and sill.New lava flows on the seafloorDuring the deployment of the observatory (2024) and first maintenance cruise (2025), we repeated a ship-based swath-bathymetry mapping of the axial valley, its flanks and its connection with the Amsterdam TF. To compensate for the lack of high-resolution mapping near the seabed, we collected several overlapping swaths at reduced speed during each leg. Subtracting the digital terrain models (DTMs)41,42 from the 2024 and 2025 surveys revealed marked changes in valley-floor topography (Fig. 3), despite the low resolution (20 m) of the models with depth-sounding uncertainties up to 5 m. Relative to 2024, the 2025 survey mapped large areas with positive depth changes, locally exceeding 90 m. We interpreted these areas as the outlines of substantial new lava flows along two distinct parallel lines. The thickest one, forming a 4-km-long patch (number 7 in Fig. 3), outpoured at the foot of the southwestern valley wall, which experienced several large normal-faulting events. The second line of massive lava flows forms a series of distinct or connected patches (numbered from 1 to 4) aligned with the axial volcanic high that was mapped along the ridge axis in 2024. A 40-m-tall circular patch (number 8) formed only 600 m away from beacon TP6. The northwest-propagating dyke probably fed the central line of lava flows but also supplied magma that possibly extruded through the western fault plane. Two other patches (numbers 9 and 10, with thicknesses up to 20 m) are located at the southeastern end of the segment and were probably fed by the early southeast-propagating dyke. The total volume of new lavas ranges from 148 to 160 million m3 (Extended Data Fig. 6), in the range of lava emplacements reported in comparable MOR contexts (for example, the Galápagos, Gorda, Juan de Fuca ridges; Fig. 12 in ref. 43).The timing of lava effusion can be constrained through indirect evidence: (1) the change of slope in the rapid subsidence phase of the BPR, which coincides with a gradual increase of the sea-bottom temperature starting on 26 April at 22:00 (Fig. 2b and Extended Data Fig. 2) and (2) the occurrence of numerous short (<10 s) and energetic hydroacoustic events, called H-waves and generated in the water column by lava–seawater interactions16,44,45,46,47 (Extended Data Fig. 2). These events begin around 06:00 on 27 April. There may have been earlier ones, masked by the intense seismic activity recorded by the hydrophones (Extended Data Fig. 7). H-wave events greatly increased after 27 April. More than 2,150 were recorded by at least three hydrophones by 12 May, after which they stopped, suggesting that the eruption lasted only 16 days, thus with an average rate of 9–10 million m3 of lava extrusion per day (effusive rate of 116 m3 s−1) and a final short outburst on 15 and 16 June 2024. The three main pulses of H-wave events match increases in the sea-bottom temperature at the BPR (Extended Data Fig. 2). Their sources cluster around the northern lava flows, which are also the most voluminous. At Axial Seamount and the East Pacific Rise, the occurrence of these impulsive events were similarly interpreted as marking the onset and duration of lava extrusion16,44,45.Transform activation and strain budgetShortly after the dyke stopped propagating northward (21:40 on 26 April), seismic activity started in the area of the Boomerang TF with unexpected off-axis normal-faulting events (numbers 21 and 23 in Fig. 1b), with fault planes oblique relative to the strike of the axial valley, as well as strike-slip events (number 24 in Fig. 1b). Twenty minutes later (at 22:23), the Amsterdam TF produced several strike-slip earthquakes, including a notable Mw = 5.9 event (number 29 in Fig. 1b). This intense activity resulted in substantial tilts among the transponder array straddling the TF. Together with large cyclical changes in the sound speed, this tilting hindered our ability to detect any resolvable change in the length of baselines (Extended Data Fig. 3b). Overall, the seismicity timeline (Fig. 2b) strongly suggests that the rupture of the Boomerang TF, followed by that of the Amsterdam TF, was triggered by the dyke propagation along the adjacent ridge axis. A similar causal relationship was inferred following the 2006 eruption at the East Pacific Rise at 9° 50′ N (ref. 15). Estimation of Coulomb stress changes (Methods) imparted by a dyke on strike-slip receiver faults representative of the Boomerang and Amsterdam TFs supports this hypothesis (Extended Data Fig. 8). Altogether, the transform seismicity triggered by the 2024 seafloor spreading event amounts to about 12 PJ km−1 of seismic moment release on the Amsterdam TF and about 3 PJ km−1 along the Boomerang TF (1 PJ = 1015 J). These values are not unusual compared with earlier bursts of moment release, which have been occurring regularly on these two TFs for several decades, as recorded by the GCMT catalogue32 (Fig. 5).Fig. 5: Cumulative moment budget along the Amsterdam and Boomerang TFs and spreading segment in between (37° 05′ S, segment I1) based on the GCMT catalogue32 and horizontal potency48 of the 2024 spreading event.a, Seismic moments (M0), expressed in 1015 J or PJ, have been normalized by the length (L) of the TFs and ridge segment: 115, 54 and 25 km, respectively. The dashed line indicates the expected horizontal potency rate (and its equivalent moment rate for a fault dipping at 60°) for a ridge segment continuously spreading at 6.3 cm year−1. b, Probability density function (PDF) of the equivalent horizontal potency48 released by fault slip—seismic and aseismic (grey histogram; modelled in Extended Data Fig. 5c); relative to the median normal-faulting moment (black symbol with range corresponding to the 16th and 84th percentiles), it shows that the 2024 event was about 24% seismic and about 76% aseismic. The same quantity can be estimated for the dyke as the product of its opening by its height (Extended Data Fig. 5d,e). These distributions can be combined to estimate the total horizontal potency of the 2024 spreading event (tectonic + magmatic; red histogram); the median estimate of 12.3 × 103 m2 (red symbol ≈ 850 PJ km−1) would correspond to the geometric moment accumulated over about 39 years of divergence of the SEIR at 6.3 cm year−1.By contrast, the approximately 25-km-long I1 spreading segment released no detectable normal-faulting moment before 2017. Between 2017 and 2023, it released about 14 PJ km−1 and the 2024 migrating swarm events (Extended Data Fig. 9) accounts for an extra approximately 22.6 PJ km−1. Notably, about 98% of the elastic models that fit our geodetic observations involve a normal-faulting moment larger than 22.6 PJ km−1 (brown histogram in Extended Data Fig. 5c). The median normal-faulting moment of these models is 92.3 PJ km−1 (range 29.1–212.9 PJ km−1, corresponding to the 16th and 84th percentiles, capturing 68% of the probability mass). By definition, this ‘geodetic’ moment includes both a seismic and an aseismic component, which, given our median estimate, implies that fault slip during the 2024 event was about 24% seismic and about 76% aseismic. This demonstrates the ability of MOR normal faults to experience aseismic slip events, which partly explains the well-documented deficit of extensional earthquakes in these settings8,9.Another main contributor to this deficit is the fact that MOR normal faults only accommodate a fraction (1 − M)11 of the total plate divergence over geological time. Our results now show how this phenomenon manifests itself on a decadal timescale. Extending a 5-km-thick lithosphere at 63 mm year−1 amounts to a horizontal geometric moment rate of 315 m3 year−1 per metre (that is, m2 year−1) along axis. The geometric moment, also known as potency48, is a kinematic quantity that, unlike the scalar moment, allows for the direct inclusion of the contribution from dykes and faults. Our median estimate for the horizontal potency released by fault slip (seismic + aseismic) during the 2024 event, assuming a dip of 60°, is 1.3 × 103 m2. The same quantity can be estimated for the dyke, as the product of its opening by its height (Extended Data Fig. 5d,e), yielding a median estimate of 10 × 103 m2. Combining these distributions yields the total horizontal potency of the 2024 spreading event (tectonic + magmatic), with a median estimate of 12.3 × 103 m2 (range 6.4–17.9 × 103 m2; red histogram in Fig. 5). This amounts to the geometric moment accumulated during 39 years (range 20–57 years) of divergence at the SEIR.We conclude that spreading segments undergo decades of quiescent stretching as abutting transforms accrue displacements by frequent slip events. The stresses that build up in the MOR segment are released during quantum events of seafloor spreading that involve both dyking and fault slip, including notable aseismic slip on axis-bounding normal faults, akin to a caldera collapse event, such as the Kīlauea’s 2018 collapse49. This may be the primary mechanism by which MOR normal faults accumulate displacement, potentially explaining their seismic moment deficit in global catalogues8.MethodsSeismicityLarge-magnitude events were extracted from the GCMT catalogue32 and from the ISC bulletin33,50. All events were relocated from our hydroacoustic records (Extended Data Table 1).HydroacousticsThe hydroacoustic array consisted of five autonomous hydrophones (Fig. 1a) moored in the axis of the SOFAR (sound fixing and ranging) channel and anchored to the seafloor. They continuously record sounds in the 1–125-Hz frequency band (250 Hz sampling rate), which include acoustic waves generated by seismic events near their epicentres24,27 (tertiary waves or T-waves) or by hot lava–seawater interactions (short impulsive events called H-waves)16,44,45,46,47, among other sounds (icequakes, large baleen whale calls or big vessels25). Source events can be precisely located (within 1–2 km)15,24,25 by trilaterating the T-wave or H-wave arrival times at each hydrophone, all synchronized with the GPS time. Following the approach of Fox et al.24, with at least four arrival times, we can derive the uncertainties in latitude, longitude and origin time from the covariance matrix of the nonlinear least-squares inversion. In that paper, they do not exceed 2 km (major axis) versus 1 km (minor axis) at 1σ (68% confidence level) inside the array; also, for each event, we systematically performed 10,000 random Monte Carlo draws within 1 s and 1 m s−1 intervals from the arrival times and sound speeds, respectively, yielding the same figures. Automatic event detection, arrival time association and trilateration are based on newly developed methods51,52. Unfortunately, the hydrophone located northwest of Amsterdam Island (top-left in Fig. 1a) failed in early April 2024. All reported events were consequently located from the acoustic records of the four remaining hydrophones. Hydroacoustic recordings were crucial in accurately repositioning the GCMT focal mechanisms and ISC events, provided in land-based catalogues with location uncertainties of 20 km or more owing to the remoteness of the area. Between the start of seismic activity on 26 April until 2 May 2024, nearly 500 T-wave events were identified and located (versus 22 GCMT and 52 ISC events) and more than 2,000 H-wave events (Extended Data Fig. 2).Direct-path rangingHorizontal displacements of the seafloor were measured by acoustic direct-path ranging between Exail-manufactured Canopus beacons deployed on the seafloor. All acoustic transponders were mounted on a 3.50-m-high tripod and equipped with pressure, temperature and, for some, conductivity sensors to infer the sound speed at ranging times. During the maintenance cruise in 2025, limited datasets were downloaded with an acoustic modem to ensure that all beacons were running properly (inclination, battery level, sensors, some ranging measurements). The direct-path ranging schedule was designed to ensure that the batteries of the beacons would last 4 years. Accordingly, each station in the ridge array ranged to the neighbouring stations in acoustic line of sight every 4 h. In the TF array, ranging sessions were set 2 h apart. Hence, during a ranging session, all baselines are measured twice, that is, both ways, with a precision of 1.5 microseconds for the two-way travel times (<3 mm at 1,500 m s−1). Travel times were then converted into distances based on the sound speed inferred from the ancillary sensors, using Del Grosso’s formula53. The converted ranging uncertainties increase with the length of the baseline and with the uncertainties in the sound speed, at best derived from the harmonic mean of the sound speeds measured at both ends of a baseline. However, because, in 2025, we only recovered (and redeployed) stations TP7 and TP13 to access their whole datasets, all baselines shown are solely based on the sound speed at station TP7 or at station TP13 (Fig. 3 and Extended Data Fig. 3).Two perpendicular inclinometers also monitor the beacon attitude. All tripods tilted during the seismic crisis, up to 12° for some stations (1° tilt of the 3.50-m-high tripod will cause 6 cm of displacement of the acoustic head). Because tripods are not oriented on the seafloor, we considered all possible orientations relative to the baseline direction. Given the size of the measured displacements from TP7 (Fig. 3 and Extended Data Fig. 3), the resulting uncertainties remained nearly two orders of magnitude smaller. However, for baselines from TP13, they obscured any signal (Extended Data Fig. 3b).All tripods were blindly lowered with a cable to the seafloor from R/V Marion Dufresne. It was therefore very challenging to position them on bathymetric escarpments or cliff edges and ensure that they were stable and roughly vertical, and that each beacon could communicate with its neighbours, especially on a rugged basaltic seabed and with maps that had a pixel resolution of 20 m × 20 m at best. Deployments of the 15 beacons took 6 h per tripod on average (ranging from 3 h to 11 h).Bottom-pressure recorderVertical displacements of the seafloor were estimated with pressure sensors. A self-calibrating (that is, drift-controlled) BPR was set on the floor of the axial valley to precisely monitor its vertical deformation (cyan hexagon in Fig. 1b). For redundancy, the BPR, manufactured by RBR Ltd., has two Paroscientific Digiquartz sensors (model 46K-313) measuring the ambient ocean pressure through a single oil capillary. They also regularly measure the pressure inside the instrument casing as a calibrating ‘zero’. These ‘zero’ measurements, corrected for air-pressure variations inside the instrument with a Paroscientific barometer (model 216B-102), are used to estimate the sensor drift, with the assumption that the drift of Paroscientific pressure sensors at low pressure (that is, atmospheric pressure) is the same as that at seafloor depth, which amounts to neglecting the scale factor. Lower-resolution pressure sensors (without drift control) were also fitted to the acoustic beacons.The BPR was lowered from the ship with a cable and dropped 30 m above the seafloor. Its recordings covered an 11-month period, from 26 February 2024 to 23 January 2025, with a sampling rate of 1 s. Calibration measurements were initially set every 2 days until 3 March, then every 7 days until 7 April, every two weeks until 6 June and finally every month. Owing to the battery running low, the last successful calibration took place in August 2024 but sampling continued. Switching the valve from ambient pressure to zero and back again consumes the most power.Following previous work54, the Paroscientific raw pressure measurements were corrected for the instrumental drift derived from fitting the ‘zero’ measurements to an exponential and then a linear decay model. The contribution of tides to pressure variations was then removed using harmonic analysis (UTide software55) to obtain a time series of pressure reflecting, in the first order, the contribution of vertical seafloor displacement and oceanic variations unrelated to tides (Fig. 2b and Extended Data Fig. 2).Swath bathymetryDuring the deployment cruise in February 2024 and the maintenance cruise in February 2025, we performed swath-bathymetry surveys from the hull-mounted EM 122 Kongsberg echosounder of R/V Marion Dufresne. To increase the sounding density, several overlapping swaths were collected at a speed of 5 knots. After editing and merging all tracks, DTMs with a resolution of 20 m could be achieved with a vertical uncertainty up to 5 m (about 0.2% of the water depth according to Kongsberg). These DTMs were the only information available to deploy the acoustic beacons and BPR. They are also the basis for the depth changes we documented between the two surveys in Fig. 3 and Extended Data Fig. 6.Elastic dislocation modelsThe first three large-magnitude GCMT earthquakes in the normal-faulting swarm (Mw = 4.9, 5.1 and 5.0; event numbers 1, 2 and 3 in Fig. 1b and Extended Data Table 1) coincide with sudden increases in the BPR-recorded seafloor pressure, which we interpret as coseismic seafloor subsidence by 1.7, 1.8 and 3.3 cm, respectively. This implies that these earthquakes cannot have occurred at too great a depth, otherwise they would have caused no resolvable seafloor subsidence. To quantify this, we modelled the events as uniform slip S on circular shear cracks of radius R (ref. 38), with a typical stress drop Δσ = 3 MPa. In this framework, the average fault slip is \(S=\frac{16}{7{\rm{\pi }}}\frac{\Delta \sigma }{G}R\) (in which G = 30 GPa denotes the shear modulus) and its radius can be estimated from the moment of each event M0 = GπR2S. We assign to each modelled rupture patch the dip of the steepest focal plane from the event focal mechanisms, 59°, 54° and 54°, respectively. This yields rupture-patch radii of 1.6, 2.0 and 1.8 km and downdip slips of 12, 15 and 13 cm. We then mesh each circular patch with triangular dislocations and use the analytical solutions of Nikkhoo and Walter56 to evaluate the maximum amount of seafloor subsidence that could be caused by each of these earthquakes, assuming that they occurred at depth D within an elastic half-space of Poisson’s ratio equal to 0.25. This amount scales as D−2 and our results (Extended Data Fig. 1) indicate that these earthquakes could not have caused seafloor subsidence greater than 1.7 cm if they had occurred any deeper than about 5 km. The Mw = 4.9 and 5.1 earthquakes, in particular, cannot have occurred at depths greater than about 3.5 and about 2.9 km, respectively. The observed coseismic subsidence thus implies that the earliest earthquakes in the swarm are confined to markedly shallow depths <5 km below the seafloor.To model and interpret the observed seafloor subsidence and changes in acoustic baselines in terms of simple tectonomagmatic deformation sources, we used analytical models of shear and tensile dislocations in a 2D elastic half-space. We specifically considered: (1) uniform normal slip on a 60°-dipping fault, from the seafloor down to a depth Hf (ref. 57); (2) uniform opening of a vertical dyke between depth Hd1 and Hd2, located at a distance xd to the southwest of TP7; and (3) uniform closing of a horizontal sill of cross-axis extent W, located at depth Hd2 (ref. 58). The only imposed geometrical constraints were to align the fault with a southwest wall of the axial valley between TP7 and TP9, to place the dyke between TP7 and TP9 and to centre the sill at the location of TP7 with a width of 2.5 km, comparable with axial melt lenses imaged at intermediate-spreading MORs59. The corresponding 2D fields of (small) displacements were first recast as horizontal and vertical displacements (respectively noted as Δux and Δuz) relative to transponder TP7, set at position x = 0 and z = 0. We then converted these displacements to relative changes in slope distance between TP7 and any point located at position x and elevation z relative to TP7 (with the approximation that acoustic ray paths are straight lines).Introducing \(L=\sqrt{{x}^{2}+{z}^{2}}\), the initial slope distance between a point and TP7, the relative baseline change is:$$\Delta L=(\sqrt{{L}^{2}+C}-L)\,{\rm{with}}\,C=\Delta {u}_{x}\times (2x+\Delta {u}_{x})+\Delta {u}_{z}\times (2z+\Delta {u}_{z}).$$We ran 10 million random models that uniformly sample xd, Hd1, Hd2, Hf (<Hd2) and the amount of fault slip, dyke opening and sill compaction within the bounds shown in Extended Data Fig. 5 (prior distributions as blue histograms). For each model, we calculated the RMS misfit relative to the baseline changes between transponders TP7 and TP9 (+1.13 m ± 0.26 m), TP5 (−0.51 m ± 0.19 m) and TP3 (−0.86 m ± 0.48 m), as well as the seafloor subsidence recorded at TP7 (−3.75 ± 0.55 m). We retained 2,203 models that had a RMS misfit better than 20 cm and plotted the distributions of their input parameters in Extended Data Fig. 5 (brown histograms). Bracketed values given in the text for the different parameters correspond to the 16th and 84th percentiles of these distributions, equivalent to ±1σ in a purely Gaussian distribution.Coulomb stress changesWe calculated Coulomb static stress changes caused by a segment-scale dyke intrusion event, approximated as a vertical ellipse with a vertical minor axis of 2 km and a horizontal major axis of 18 km, with an azimuth of N319. The centre of the ellipse is located 1 km below the free surface of an elastic half-space with Poisson’s ratio and shear modulus equal to 0.25 and 10 GPa, respectively. The ellipse is meshed with triangular dislocations56, each assigned a uniform opening of 1 m. Coulomb stress changes60 are computed on two types of receiver fault. The first is a strike-slip fault oriented N024, which corresponds to the focal mechanism of the Mw = 5.3 (at 22:07 on 26 April 2024) event that occurred within the Boomerang transform boundary. The second is a vertical, left-lateral strike-slip fault oriented N049 representing the Amsterdam TF orientation. Extended Data Fig. 8 shows the corresponding stress changes along horizontal cross-sections 5 km below the seafloor. We chose this depth to be below the reach of the modelled dyke to account for the fact that the seismogenic lithosphere is probably thicker around the Amsterdam and Boomerang TFs than along the SEIR segment I1.