Source mechanism of impulsive seafloor events that track submarine lava flows

Abstract

Most of Earth’s volcanic eruptions are hidden beneath the ocean in complete darkness. Recent studies suggested that a type of impulsive event can track submarine lava flows, but their source mechanism remains uncertain. We analyze >20,000 impulsive events from the 2015 Axial Seamount eruption and find that their seismo-acoustic waveform characteristics suggest an implosive source mechanism. Integrating constraints from their spatiotemporal evolution with heat transfer estimates and geological observations, we propose that while the largest events might be related to volatiles degassed from magma, most events are generated by the implosion of bubbles formed from the vaporization of entrapped seawater by hot erupted lava. Similar events have been detected at other seamounts and slow to fast-spreading mid-ocean ridges, although eruptions at >3000-meter depth have proportionately fewer events because seawater vaporization is inhibited. Therefore, these impulsive seafloor events can be leveraged to remotely characterize eruption dynamics in most submarine volcanic settings.

Hot lava can vaporize seawater to form bubbles that implode and generate sounds that can be used to track submarine eruptions.

INTRODUCTION

Submarine volcanoes, primarily located at island arcs, hot spots, and mid-ocean ridges, account for more than 70% of volcanic activities on Earth (1) (fig. S1). However, their inaccessibility presents formidable obstacles for effective monitoring of eruptions on the seafloor. For instance, of the 9944 confirmed eruptions from 9950 BCE to 26 April 2024 in the Smithsonian Global Volcanism database (2), only ~3% are submarine eruptions. To address this observational gap, various monitoring methodologies have been developed to adapt to the unique challenges of tracking submarine volcanism. Among them, satellite-based observations can monitor for signs such as discoloration or pumice rafts related to recent shallow submarine eruptions (3, 4) but cannot resolve processes at depth. In addition, while this method technically could allow continuous monitoring, studies so far have relied on prior eruption knowledge or chance discovery at specific volcanoes (5), and the typical satellite revisit period of days/weeks precludes real-time tracking of eruption evolution (6). In comparison, direct observations through remotely operated vehicles (ROVs), autonomous underwater vehicles, and ships allow for detailed documentation of changes in bathymetry along with various video and photographic evidence of fresh lava flows (7). However, these methods are constrained by high costs and only offer a snapshot of the volcanic processes (8). On the other hand, geochemical dating and analyses of collected rock samples provide more historical context but suffer from poorer spatial and temporal resolution (9, 10).

Data recorded by ocean bottom seismometers and hydrophone arrays have been used to infer possible submarine volcanic activity based on the detection of earthquake swarms (5). However, earthquake swarms do not uniquely indicate eruptions because they are also often related to noneruptive processes such as earthquake sequences or diking (11). Recently, a type of impulsive seafloor event has been identified and suggested to be uniquely related to submarine eruptions. These impulsive signals were first identified near Hawaii and the Gakkel Ridge (12, 13) where their association with submarine eruptions was hypothesized. Later studies (14, 15) demonstrated that similar signals coincided with mapped lava flows from the 2006 eruption at the East Pacific Rise 9°50′N and the 2015 eruption at Axial Seamount. Similarly, a recent study at Fani Maoré (near Mayotte) found these impulsive events to closely align with mapped lava flow boundaries from a 2020 eruption (16). Several hypotheses have been proposed to explain these signals. One possibility is bubble dynamics due to gas exsolution (16–18), where bubble collapse and rebound generate strong pressure pulses (19). Another widely discussed hypothesis involves interactions between seawater and lava (13–15), either from seawater infiltrating cracks in cooling magma and causing explosive fragmentation (20, 21) or from seawater trapped by lava and heated to steam or supercritical states, potentially forming collapse pits (22–24). In addition, some studies suggest that the signals may result from the combustion of H₂ produced by thermal dissociation of water (25). Despite these hypotheses, the exact source mechanism of these signals has yet to be determined.

Axial Seamount, located on the Juan de Fuca Ridge in the northeastern Pacific Ocean, serves as an exceptional case study since the Ocean Observatories Initiative’s (OOI) cabled array (Fig. 1A) recorded the entire 2015 eruption, which lasted from 24 April to 21 May and was marked by notable changes in earthquake rate and seafloor deformation (15). A total of 22,383 impulsive seafloor events were cataloged spanning the entire eruption period [(26) and Materials and Methods], with the lava flows covering a total area of $11.11\times {10}^6{\mathrm{m}}^2$ with an estimated volume of $1.48\times {10}^8{\mathrm{m}}^3$ (27). In this study, we integrate seismic data analyses with constraints from heat transfer estimates, supported by published geochemical parameters and geological observations, to rule out explosion, tensile crack, and double-couple source mechanisms for these impulsive events. Instead, we conclude that while some impulsive events were related to degassed volatiles from the magma, most were generated by the implosion of bubbles formed from the vaporization of entrapped seawater by flash heating by the erupted hot lava.

Fig. 1. Impulsive events.

(A) Axial Seamount with the caldera rim (bold black line), eruptive fissures (black lines) (29), and footprint of the magma chamber (dashed line) (63). Dive tracks of ROV Jason (purple lines) and locations of collapsed pits (purple pentagons) mapped in August 2015 are also shown (27). Orange dots are earthquake epicenters (64), while red dots are impulsive events (this study). (B) World map displaying confirmed submarine eruptions with impulsive events detected (triangles) and Axial Seamount (star), all color-coded according to their water depth (12, 14, 16, 20). (C) Inferred travel path of the impulsive event arrivals. (D) Vertical-channel seismograms of the same event recorded at seven different stations, with the event source marked as a white cross in (A). The seismograms are filtered with a 4-Hz high pass filter and labeled with the station name. The event has two distinct arrivals: The first arrival has reflected once off the sea surface (B1), while the second arrival has reflected twice off the sea surface (B2).

RESULTS

Source mechanism of impulsive events

We start with the impulsive event catalog from Bohnenstiehl and Mann (26) and group the events into clusters based on their waveform similarity (Materials and Methods). This yielded 30 clusters, with each cluster having more than 40 events. The locations of 2361 impulsive events in the 30 clusters, based on the catalog of Bohnenstiehl and Mann (26), are concentrated in the rift zone north of the caldera (figs. S2 to S4). This is consistent with the rift zone accounting for more than 90% of the erupted volume with larger and thicker (67 to 128 m) lava flows compared to those near the caldera, which were only 5- to 19-m thick (27). Subsequently, we stacked all the vertical seismic waveforms within the same cluster and found that all the stacked waveforms exhibit downward first motion polarities (fig. S5). Since the events in the clusters are concentrated at the north rift zone (fig. S2), we further examined the individual waveforms of all impulsive events near the caldera, which include a total of 582 events south of latitude 46.06°N (Fig. 1A). Of these, only 95 events have waveforms with sufficient signal-to-noise ratio for confident determination of their polarity, and they all also exhibited downward polarity at all stations. Given that the first arrival of the impulsive events approached the seafloor seismic stations from above and experienced a polarity reversal during its reflection at the sea surface (Fig. 1C), a downward first motion polarity indicates the station’s presence in a dilatational zone (fig. S6).

Based on the azimuth and take-off angle between each event and station, the stations can be projected onto the event’s focal sphere. In Fig. 2A, we present data from three individual impulsive events located near the north rift zone, between the north rift zone and the caldera, and directly within the caldera, respectively. The focal mechanisms are plotted using the lower hemisphere projection, as adopted throughout this study. For impulsive events concentrated in the north rift zone, the station sampling points lie very close to the edge of the focal sphere. This poor focal sphere coverage hinders our ability to directly invert for individual focal mechanisms. We have considered the inclusion of secondary arrivals, but this did not substantially expand the focal sphere coverage (fig. S7). Therefore, we assume that all detected events have similar focal mechanisms and amalgamate all the polarities to deduce the most plausible source mechanism.

Fig. 2. Source mechanism of impulsive events.

(A) Vertical-channel seismograms (4 to 40-Hz band-pass filter) of impulsive events at three representative locations (Fig. 1A, white diamonds) recorded at various sampling points on the focal sphere. (B) Three hypothetical source mechanisms compared with observed first motion polarities. Open circles indicate downward first motions, which represent the dilatational zones. The sampling points are based on the impulsive event located at the coordinates (45°58′12.00″N, 129°59′24.00″W). Compressional quadrants are colored. All focal mechanisms are shown using the lower hemisphere projection.

The observed polarities related to a dilatational quadrant signature rule out the possibility of an explosive source mechanism (13, 18) for the impulsive events (Fig. 2B). Next, we investigate whether the impulsive events could have a tensile crack source mechanism. First, we constructed three hypothetical pure tensile crack radiation patterns (fig. S8) based on the geometry and location of mapped fissures (27) extending from the northeastern caldera to the north rift zone. However, the observed polarities do not support nor rule out the tensile crack hypothesis because they only cover the dilatational areas within the focal spheres (Fig. 2B and fig. S1). Nevertheless, we find that the impulsive signals are distinct from earthquakes on the ring fault and other recorded crustal earthquakes nearby (15, 28). First, the impulsive signals exhibit consistent polarities between hydrophones and vertical seismic channels (fig. S9), a pattern that has also been noted in a previous study (22). Second, they do not produce observable direct crustal P- or S-wave arrivals even for events within the caldera and close to the stations. These absences suggest that the source is strongly coupled with the water column but not with the solid Earth, which is inconsistent with both tensile crack and double-couple mechanisms. Last, the impulsive event locations do not align with previously mapped large fissures that formed during the eruption (Fig. 1A) (29), and those fissures did not generate detectable seismic signals. Therefore, it is unlikely that similar or smaller-scale cracks would generate the observed seismo-acoustic signals. We conclude that an implosive source mechanism is most consistent with all the existing observations (Figs. 1 and 2A).

DISCUSSION

Physical processes underlying implosive events

Impulsive events have been documented at depths of ~1600 m at Axial Seamount (20), ~2500 m on the East Pacific Rise (14), ~3300 m near Fani Maoré (30), and ~4000 m on the Gakkel Ridge (18) (Fig. 1B), where high hydrostatic pressure can explain the implosive source mechanisms. Based on existing literature, we evaluate three main hypotheses for the physical process underlying the implosive source. Below, we summarize and assess each hypothesis in turn. The first is cooling-contraction granulation (17, 31). This occurs when hot lava contacts cold seawater, quickly forming a rigid crust on its surface. As the interior cools and contracts, the outer brittle layer fractures and could collapse under high hydrostatic pressure, leading to the formation of limu o Pele particles or glass fragments (32, 33) often found near submarine volcanoes. However, collapsed pits documented at Axial Seamount possibly associated with this process produce minimal glassy ash or bubble-wall shards and are localized (27), unlike the widespread distribution of impulsive events (Fig. 1). Furthermore, repeating impulsive events with highly similar waveforms (fig. S5) suggest a nondestructive, recurring source (22), unlike one-time pit collapses. Therefore, we rule out cooling-contraction granulation as the main process generating the implosion signals at Axial Seamount.

A second hypothesis relates the observed implosive events to volatiles degassed from the underlying magma/erupted lava (17). Direct video and hydroacoustic observations have shown that magmatic gases can be a source of acoustic signals (7, 34). At Axial Seamount, which produces slightly enriched transitional mid-ocean ridge basalt (35), CO₂ is the primary volatile released during the exsolution process (36). At depth, the collapse of foam layers and the coalescence of bubbles can form CO₂-filled slugs that rise buoyantly through the magma (17). The high hydrostatic pressure at a depth of around 1600 m causes CO₂ to become a supercritical fluid with high solubility in water (31, 37, 38). Therefore, when these slugs breach the magma surface and the CO₂ inside cools, it undergoes a substantial volume decrease. This reduction in volume can be as much as 86 to 92% at pressures of 250 to 450 bars (17). This rapid contraction can lead to a pressure imbalance between the internal pressure of the CO₂ within the slug and the external hydrostatic pressure, ultimately causing an implosion. However, prior study of dissolved CO₂ and H₂O concentrations in melt inclusions and host glasses of erupted lava at Axial Seamount indicates that syn-eruptive degassing is limited (39). In addition, for gases that accumulate at the top of the magma chamber before being discharged during eruptions (17, 32), once a dike opens a pathway to the seafloor, CO₂ that degassed within the shallow magma reservoir (39) or from the deeper magmatic system should escape swiftly due to buoyancy. This is consistent with geochemical observations that erupted lavas at Axial Seamount are typically already highly degassed before eruption (40). A small number of impulsive events detected during the 2015 Axial Seamount eruption have markedly larger recorded seismic displacements, after correcting for transmission loss, compared with other events (fig. S10). These notably larger impulsive events are concentrated along the eruptive fissures (fig. S8) and primarily occur during the first ~12 hours of the eruption (fig. S10). This is similar to observations from the 2006 East Pacific Rise 9°50′N eruption, where notably larger impulsive events were recorded during the first ~5 hours and concentrated along the mid-ocean ridge axis (14). Therefore, these large-amplitude impulsive events are possibly related to the implosion of gas-rich bubbles released from magma. However, 97% of all impulsive events occurred in the north rift zone, which is ~200-m deeper than and ~9 km away from the caldera (Fig. 1A), over 30 days with similar amplitudes (fig. S10). At the greater depth of the north rift zone (Fig. 1A), the solubility of volatiles is higher, which would prevent additional volatile exsolution from the magma. Therefore, the implosion of volatiles released from magma is unlikely to be the dominant process generating these impulsive events at Axial Seamount.

A third possible mechanism is hydrovolcanic implosion (20, 22, 38), in which interactions between lava and seawater generate implosive events. Observations of active submarine lava flows in Hawaii (25) suggest that simple lava-water contact does not generate widespread impulsive events due to the initial steam layer’s insulating effect (movie S1), consistent with temperature measurements near active flows showing minimal heating of surrounding seawater (41). While observations suggest that surface-initiated cracks (42) are unlikely to penetrate deep enough for seawater to contact the molten interior of the flow (43), the common presence of pipe vesicles and spiracles in submarine lava flows provides clear evidence that water can enter lava flows from beneath (23). Additional support comes from observations offshore Kilauea, Hawaii, where frequent underwater explosions, audible concussions, and large volumes of hot water escaping through cracks in active submarine lava streams suggest dynamic seawater interaction with lava interiors (25). Therefore, the hydrovolcanic implosion process likely occurs when lava flows, with temperatures around 1200°C (40), trap seawater or water-saturated sediments as they advance (Fig. 3B). The trapped seawater, initially at a temperature around 2°C (27, 44), experiences rapid heating, which causes it to vaporize into steam (Fig. 3B). As the lava continues to cool while vapor bubbles coalesce beneath, heat conduction causes the steam to lose energy, leading to some condensation and a subsequent drop in internal pressure within the bubbles. This decrease in pressure creates a substantial pressure differential relative to the hydrostatic pressure due to the overlying ocean. Eventually, this imbalance can lead to an implosion of the steam bubbles (Fig. 3B), which generate the recorded impulsive events.

Fig. 3. Schematic diagram of the inferred processes underlying impulsive seafloor events.

(A) At shallow depths, rapid expansion of bubbles formed from vaporized seawater (blue) by hot erupted lava and volatiles exsolved from magma (yellow) can generate explosive signals. (B) At intermediate depths, implosions of predominantly vaporized seawater and some exsolved volatiles due to high hydrostatic pressure generate implosive signals. The depth boundary for transition (wavy line) from explosion-dominated to implosion-dominated regime remains to be determined though implosions were observed as shallow as ~1,600 m in this study. (C) At >~3,000 meters depths, vaporization of seawater by hot erupted lava is inhibited so proportionately fewer and smaller bubbles reach the lava surface, resulting in fewer detectable implosive events.

To further test the plausibility of the third hypothesis, we need to understand the possible bubble morphologies that generated the impulsive signals. We assume a spherical bubble where the interior contains vapor/volatile gases (Fig. 4A). Following the general form derived by Brennen (45) and consistent with the simplified version presented by Roche et al. (46), we define the natural frequency assuming isothermal gas behavior and disregard the tension term on the premise that the bubble size is adequately large (exceeding several nanometers). This approximation yields a simplified formula for estimating the bubble radius

$$R=\frac{1}{2\mathrm{\pi }f}·\sqrt{\frac{3p}{\mathrm{\rho }}}$$ (1)

where $f$ represents the natural frequency, $p$ denotes the hydrostatic pressure, and $\mathrm{\rho }$ is the water density. We set $\mathrm{\rho }=1025\text{kg}/{\mathrm{m}}^3$ and hydrostatic pressure $p=\mathrm{\rho }\mathit{gh}$ , where the acceleration due to gravity $g=9.8\mathrm{m}/{\mathrm{s}}^2$ and the water depth $h=1600\mathrm{m}$ . By conducting spectral analysis of impulsive event waveforms from all stations and averaging the normalized values, we determined the dominant frequency of the impulsive events to have a mean of 26.5 Hz with a standard deviation of 6.2 Hz (fig. S11). This frequency matches the bubble’s Minnaert frequency since the water depth is considerably greater than the bubble radius and hence is indicative of the bubble’s natural frequency. Using Eq. 1, we can estimate the bubble radius for each event (Fig. 4B).

Fig. 4. Analysis of bubble size.

(A) Schematic diagram of a spherical bubble. (B) Frequency distribution of inferred bubble radius of impulsive events. Interactions within bubble clouds may result in the apparent observation of larger bubble sizes, as groups of bubbles can oscillate collectively at lower frequencies.

From the results, we observed that 15,760 bubble radii (~70% of all impulsive events) are concentrated between 1 and 1.5 m (Fig. 4B). This inferred bubble size is consistent with observations of similarly sized vapor bubbles (~0.2 to 1 m in diameter) at the West Mata submarine volcano (~1 km below sea level) (34, 47). In comparison, at atmospheric pressure, the diameter of littoral limu bubbles can expand to several tens of meters within a few seconds (38). Given the hydrostatic pressure, the calculated bubble radii seem plausible, although we lack direct observation to fully validate these bubble sizes. However, an additional factor to consider is the interaction within bubble clouds, where closely packed bubbles oscillate in unison, often at a frequency lower than a single bubble’s natural frequency. This synchrony can cause clusters of smaller bubbles to mimic the acoustic properties of larger individual bubbles (46), so the individual bubble sizes might be smaller than our estimates. A similar bias might also result from the fact that the bubbles in our study are confined within lava instead of water. The much higher density of lava compared to water is expected to lower the natural frequency for a bubble of a given size. As a result, using the Minnaert equation with water properties to infer bubble radius from the observed frequency likely leads to an overestimation. Because of the lack of detailed models for bubble oscillations in lava, our calculated bubble radii and the resulting total bubble volume ( $V_{\text{bubbles}}=5.0\times {10}^5{\mathrm{m}}^3$ ) should be treated as an upper bound estimate.

Subsequently, we estimate the lower bound of vaporization given the volume and surface area of the erupted lava (text S1) for comparison with $V_{\text{bubbles}}$ . Under the hydrostatic pressure at a depth of $1600\mathrm{m},$ seawater would vaporize when heated to 348°C (48). Assuming an initial lava temperature of ∼1200°C (27, 40) and seawater temperature of ∼2°C (44) and considering only the heat contribution from the thin (1 mm to 1 cm) glassy zone (49–51) of the lava’s surface, we estimated the volume of water vapor that could be generated to be between $2.2\times {10}^5$ and $2.9\times {10}^7{\mathrm{m}}^3$ depending on the assumed thickness of individual lava flows (text S1). The lower bound estimate $(2.2\times {10}^5{\mathrm{m}}^3)$ , considering only the surface layer of lava flows, is of the same order of magnitude as $V_{\text{bubbles}}$ , the upper bound volume estimate of bubbles generating the detected impulsive events. At Axial Seamount, the thickest deposits where most impulsive events occurred exceed 128 m in total thickness (27). These deposits were previously interpreted as the result of multiple overlapping flows, rather than a single lava flow (20). Field and laboratory studies also suggested that individual submarine lava flows are typically 0.2 to 2 m (52–55). Therefore, the total volume of vapor that can be generated is likely closer to the upper bound estimate $(2.9\times {10}^7{\mathrm{m}}^3)$ , which is substantially greater than the bubble volume, indicating that the available heat from the erupted lava is sufficient to generate the observed impulsive events during the 2015 eruption.

Our analyses only used seismic and hydroacoustic data from the Axial Seamount. The varying depths of observed impulsive seafloor events (20) may influence the dominant underlying physical processes. At depths exceeding 3000 m, seawater heated to critical temperature would not vaporize but instead transition into a supercritical state (56, 57). Although the heated water in its supercritical fluid form would still expand to more than twice its original volume (38), they exhibit far less volumetric change compared to vapor. This reduced expansion leads to smaller pressure differentials during cooling, making it less likely to generate strong detectable seismo-acoustic signals. Therefore, hydrovolcanic implosions due to vaporization of entrapped seawater by hot erupted lava, which we propose as the dominant process generating impulsive seafloor events at Axial Seamount, might be inhibited at depths exceeding 3000 m (Fig. 3C), where both reduced bubble size and suppressed bubble formation under higher hydrostatic pressures likely contribute to fewer and smaller bubbles reaching the lava surface. We find that at Fani Maoré (~3300-m depth), the ratio of detected impulsive seafloor events to erupted lava volume (text S2) is $0.83\times {10}^{-7}\text{events}/{\mathrm{m}}^3$ (16), which is approximately three orders of magnitude lower than the $0.89\times {10}^{-4}\text{events}/{\mathrm{m}}^3$ at the East Pacific Rise (~2500-m depth) (14) and $1.51\times {10}^{-4}\text{events}/{\mathrm{m}}^3$ at Axial Seamount (~1600-m depth) (table S1). The detection capability at Fani Maoré, with three ocean bottom seismometers at ~5- to 15-km distances from the lava flow (16), was likely comparable to that at Axial Seamount and the East Pacific Rise where seven and three ocean bottom seismometers, respectively, were at distances often exceeding 10 km (14). Thus, the markedly lower number of detected events at Fani Maoré is unlikely to be due to differences in detection sensitivity. Instead, it likely reflects the inhibition of vaporization of entrapped seawater by erupted lava at Fani Maoré. This scenario should also apply to Gakkel Ridge (~4000-m depth) where only ~200 impulsive seafloor events were detected, although the lack of erupted lava volume data prevents further verification (12). Therefore, for the ~37% of all known seamounts (58) and the ~51% of mid-ocean ridges with depths greater than 3000 m (text S3), there should be proportionately fewer impulsive seafloor events detected during eruptions compared to Axial Seamount (15) and the East Pacific Rise (14), although the variability in lava flow thickness and emplacement style further introduces uncertainty to the comparison between different volcanic settings. In contrast, for submarine eruptions at very shallow depths, the reduced hydrostatic pressure may not be sufficient to induce implosions. Instead, rapid expansion of bubbles formed from the vaporization of seawater entrapped by hot erupted lava or volatiles exsolved from magma would instead generate explosive signals (Fig. 3A). This has been observed at the coastline of Kīlauea (movie S2) and is likely the case for the impulsive events related to lava ocean entry and offshore lava emplacement during the 2018 Kilauea eruption (59, 60), although it remains uncertain whether the hydroacoustic signals originate from bubble expansion or collapse (47). Regardless, impulsive seafloor events coinciding with submarine lava flows have been detected across a variety of volcanic settings, including seamounts and slow to fast-spreading mid-ocean ridges over a wide range of water depths (Fig. 1B). Combined with our analyses of the underlying source processes, we propose that similar impulsive seafloor events should be generated during most submarine eruptions, albeit with differences in the underlying dominant source mechanism, hence the number of events per unit erupted lava volume.

The detection range of these seismo-acoustic events is likely influenced by multiple factors, including network geometry, bathymetry, background noise, and event size. At Axial Seamount, some events with good signal-to-noise ratio were observed at station AXBA1 located ~28 and ~37 km from the caldera and north rift zone, respectively (figs. S12 and S13). At the East Pacific Rise, ocean-bottom seismometers near the ridge axis recorded events at up to ~12 km away (14). At Fani Maoré, autonomous hydrophones suspended in the sound fixing and ranging channel at ~1300 m depth (~2200 m above the seafloor) detected events at up to ~58 km away (30). At the Gakkel Ridge, a seismic array on an ice floe detected events at up to ~81 km away (18). These observations suggest that the seismo-acoustic events can be detected at many tens of kilometers away given the typical seismic/hydroacoustic monitoring network configurations. However, since the stations furthest away from the lava flows all detected seismo-acoustic events in these studies, the maximum detectable distance of these events remains to be constrained. Nevertheless, given current observations, their use in remote monitoring of submarine eruptions at different tectonic environments for at least the near-regional distances is feasible. This opens the door to improved monitoring and characterization of submarine eruptions using ocean bottom seismic and hydroacoustic data.

MATERIALS AND METHODS

Impulsive event catalog

We use the catalog from Bohnenstiehl and Mann (26), who used a three-step approach to derive a catalog of impulsive events. Initially, template matching with three-component waveforms (HDH, HHZ, and HHN) from AXEC2 was used, generating more than 37,000 preliminary detections from a suite of 817 waveform templates with high signal-to-noise ratios. Subsequently, they cross-correlated vertical channel waveforms from AXEC2 with data from six other stations to obtain the precise timings of the first and second arrivals (Fig. 1D) across the OOI seismic network. These events were then located using a grid search approach, leading to a catalog of 22,383 impulsive events with a mean lateral location uncertainty of ~140 m as determined by chi-squared misfit at a 68% confidence interval. Although hydrophone (HDH) data were used in the initial template matching, our subsequent analyses focus on seismic data (HHZ) since all seven stations recorded seismic waveforms, whereas only two were equipped with hydrophones, limiting the spatial coverage of HDH data.

Clustering impulsive events

We use data from the OOI Cabled Array, which includes seven seismometers (Fig. 1A) that recorded continuous data at a 200-Hz sampling rate. We adopted preprocessing measures, which include removing linear trends, mean values, and instrument responses to restore the true ground motion velocity. We further applied a 4-Hz high-pass filter to remove microseism noise (61). We found that a 0.3-s-long window centered at Bohnenstiehl and Mann’s (26) catalog arrival times gives waveforms that best preserve the primary impulsive signals without redundancy. Subsequently, we undertook spectral analysis on the waveforms to quantify the frequency content of the individual impulsive events. We found that the waveforms contain energy in the 20- to 30-Hz frequency range, which is the dominant frequency for most events (see fig. S14). However, we found that many waveforms also contained additional high-frequency energy (80 to 100 Hz). The high-frequency content is not consistently recorded by all stations even for the same event, suggesting instrumental or site effects rather than source effects. Therefore, we apply a Butterworth band-pass filter with a range of 4 to 40 Hz and two corners before conducting subsequent analyses. Because many events display similar waveforms, we group events into clusters based on waveform similarity measured using cross-correlation of the filtered, vertical channel waveforms. Clustering is performed primarily to facilitate stacking to improve the signal-to-noise ratio of the waveforms for robust first-motion determination at all stations. Only clusters with more than 40 events (i.e., the top 30 clusters) were analyzed to ensure that the stacked waveforms had sufficiently improved signal-to-noise ratio for robust analysis. We specifically use the vertical seismic component (HHZ) for clustering, as it provides consistent data across all stations, whereas hydrophone (HDH) data were only available at two locations. Our primary strategy dictates that if the cross-correlation coefficient (CCC) for two event pairs (A and B) and (B and C) separately exceeds the defined threshold, then events A, B, and C are all classified into the same cluster (62). While there is a potential issue with this algorithm—when the cross-correlation coefficients are low, the CCC between events A and C might fall below the defined threshold—our CCC threshold of 0.98 is sufficiently high to ensure waveform similarity within the clusters. This yielded 30 clusters, with each cluster having more than 40 events. Information on all clusters is provided in data S1.

Supplementary Materials

The PDF file includes:

Supplementary Text

Figs. S1 to S15

Table S1

Legends for movies S1 and S2

Legend for data S1

References

Other Supplementary Material for this manuscript includes the following:

Movies S1 and S2

Data S1