Four different multibeam echosounders (MBESs) acquired data during the four repeat surveys of the Havre volcano (Table 1): the 2002 pre-eruption survey used a Kongsberg EM 300 (30 kHz) echosounder on RV Tangaroa (Figure 1A); the post-main eruption response survey in October 2012 used a Kongsberg EM 302 (30 kHz) echosounder, again on RV Tangaroa (Figure 1B); and the 2015 targeted survey used both a Kongsberg EM 122 (12 kHz) on RV Roger Revelle (Figure 1C) and a Reson SeaBat 7125 (400 kHz) on the AUV Sentry (Figure 1D). Bathymetry data were processed following the best practices outlined in respective voyage reports or published articles, including those by Wright et al. (2006) for 2002 data, Wysoczanski et al. (2013) for 2012 data, and the MESH Voyage Report for 2015 data. Data processing includes standard sound speed and position corrections and acoustic artifact (“ping error”) cleaning.

All digital elevation models (DEMs) of the Havre volcano were projected into WGS84 UTM Zone 1S. All DEMs, excluding the high-resolution (1 m) AUV DEM, were resampled down to 25 m to fit the coarsest resolution surface, using QGIS v3.16. All DEM extents were matched and aligned along eastings and northings. The DEM pixel numbers were also matched to ensure that rasters were dimensionally divisible to allow for accurate change detection (Wheaton et al., 2009).

AUV Sentry collected vector magnetic data using an APS three-axis flux gate magnetometer at an average altitude of ∼50 m above the seafloor. The three-component magnetic data (x, y, and z) were transformed into a total-intensity magnetic field. The measured magnetic field was then corrected for the magnetic effect originating from the vehicle itself by fitting the sinusoidal variation in the magnetic field data observed as the AUV spins during its descent/ascent to/from the seafloor (Tivey et al., 2003). Following the magnetometer spiral calibration, residual noise from the AUV was less than 0.2% of the magnetic anomaly magnitude. Magnetic anomalies were obtained by subtracting the International Geomagnetic Reference Field (IGRF; Thebault et al., 2015). The resulting magnetic anomaly data were then interpolated onto a 20-m-spaced grid using a minimum-curvature algorithm. The magnetic anomaly grid was then inverted to correct for topographic effects and the uneven track of the AUV. We used a discrete set of juxtaposed prisms with 50-m horizontal resolution, with tops defined by the bathymetric level and bottoms at a constant depth of 2,000 m. The resulting crustal magnetization of each block was then determined by least-squares linear optimization.

Spatial analysis toolboxes in QGIS v3.16 and System for Automated Geoscientific Analyses (SAGA) algorithms were used to create bathymetric derivatives for all four DEMs, including the slope, standard deviation of the slope, and topographic position index (TPI) (Conrad et al., 2015) (Figure 2 shows the slope and standard deviation of the slope). The slope was used to delineate obvious changes in the slope between large-scale features and contacts or boundaries of various volcanic products and geomorphological features at all scales. Broad and fine-scale TPIs were used to note areas of discrete high and low topography to delineate fault features, ridge lines, and gullies.

The standard deviation of the slope is a robust proxy for seafloor texture that has been used to delineate various lava products, both submarine and subaerially (Whelley et al., 2014; Whelley et al., 2017; Korzeniowska et al., 2018). The textures of giant pumice (GP) are emphasized by the 1-m resolution standard deviation of the slope layer, which was used to extract GP coverage maps. Areas with >4 standard deviation of the slope were re-classified as GP and converted to a binary class, and footprints of the GP areas were extracted. Polygons >1 m and <10 m were retained, with polygons >10 m excluded from the analysis to remove stacked clasts and limit the analysis to observed maximum GP sizes (Carey et al., 2018; Ikegami et al., 2018; Mitchell et al., 2019). The resulting polygons were used as a proxy for individual clasts. Areas of known acoustic artifacts or features were also removed, e.g., extruded fresh lavas, caldera walls, and ping errors. The standard deviation of the slope also emphasizes a known artifact in the 2012 EM 302 data along the southwestern wall. This artifact was perpetuated through all analyses; any interpretation within that area is treated with due caution.

Approximately 250 h of video footage was collected over 12 dives during the 2015 survey using the ROV Jason (Figure 3). Stills were analyzed using both a forward-looking Insite SuperScorpio Sony Camera and a downward looking MISO-OSI 16 MP camera. The video was analyzed from the three Insite Mini Zeus 1080i Camera video footages. Water temperature values were recorded underway using a miniature autonomous plume recorder (MAPR); discrete seafloor and sub-seafloor temperature values were recorded using a stand-alone heat flow (SAHF) meter. Extensive event logs were recorded during the twelve 2015 ROV dives by onboard scientists and technicians.

Post-survey, voyage event logs were cleaned (removing erroneous entries, fixing spelling, and consolidating terms) and re-coded in QGIS to indicate where key features were observed. These logs helped guide video footage interrogation and geomorphic boundary refining. Using these logs, we analyzed the video footage for regional and local topographic, geomorphic, and lithic composition to verify surface slope, roughness, and textures observed in bathymetric derivatives.

Initial geomorphic change detection (GCD) calculations were conducted by Carey et al. (2018) and Ikegami et al. (2018), with volumetric calculations undertaken only on lava and dome regions between the three datasets. In this section, we expand the GCD to analyze all three lower-resolution (25 m) DEMs, including both geomorphic increase and decrease. We estimate and incorporate a uniform 10-m detection limit threshold into all GCD analyses to account for assumed sensor differences and vertical uncertainty in deep water (>1,000 m w.d.) and uncertainty in the three grids (Brasington et al., 2003; Wheaton et al., 2009; Schimel et al., 2018). Any observed change that occurs within this 10-m threshold is deemed untrustworthy and excluded from further analysis; a change above this threshold is deemed trustworthy. A detection limit threshold was chosen as the grids were provided in processed form; uncertainty estimations were based on known sensor uncertainties between systems and certainty loss with depth.

Three difference rasters were created to compare the periods 2002–2012, 2012–2015, and 2002–2015. All different rasters were then overlain with the high-resolution 1-m grid to determine whether volumetric changes correspond with observable morphological features in the high-resolution (1 m) DEM. Volume calculations were made by multiplying the height change (increase or decrease) per pixel of change. Volume change estimates in areas with high degrees of slope were treated with caution.

Geomorphic units of lower-resolution (25 m) DEMs were delineated in QGIS (v3.16) based on the slope, roughness, topographic position metrics, and the GCD analysis. Geomorphic units of the high-resolution (1 m) DEM were identified and hand-drawn in QGIS based on the slope, roughness, topographic position metrics, the large-scale GCD, and observable morphological features cross-referenced with video footage. The fine-scale map units were also based on a synthesis of previous geological analyses undertaken by Wright et al. (2006), Carey et al. (2014), Carey et al. (2018), Ikegami et al. (2018), McPhie et al. (2020), and Murch et al. (2020). Criteria for allocation to specific geomorphic units, along with statistical and geometric measures of each unit, are described in Supplementary Material Table 1. The definitions and naming conventions of geological facies from the previous literature (Carey et al., 2018; Ikegami et al., 2018; McPhie et al., 2020) predominantly remain unchanged, including the alphabetization of lava lobes and domes.

Four geomorphic groups encompassing both constructional and destructional processes are interpreted in the large- and fine-scale maps: (1) large-scale tectonic features, e.g., faults and ridges; (2) large-scale volcanic features, e.g., calderas and craters; (3) coherent volcanic products, e.g., lava lobes and domes; (4) clastic volcanic products, e.g., ash–lapilli–block and GP deposits; and (5) mass-wasting features, e.g., debris deposits, scarps, and mega-blocks. Geomorphic units are given in Table 2. The spatial characteristics of units are given in Supplementary Material Table 1.

The fine-scale map produced from the 1-m AUV bathymetry allowed us to define broad-scale unit boundaries more precisely, record geological facies, and interpret geomorphological processes obscured in the lower-resolution data. The scale of the high-resolution bathymetry means any geomorphic units that feature volcanic material <1 m in size, including ash, ash–block–lapilli, or scree deposits, were largely unresolvable. These deposits are defined for locations where video footage is available but cannot be robustly defined elsewhere. Mass-wasting evidence along steep slopes (i.e., scarps, scars, and debris deposits) could be delineated in some areas. Similarly, hydrothermal venting areas were recorded and included both diffuse and point-source fluid venting, hydrothermally altered material and sediment, and ecology suggestive of nearby venting (i.e., extensive bacterial mats) (Figures 3E, J, O, T).

The large-scale geomorphic and geologic (Figure 4A) and Havre volcano maps (Figures 4B–D) show tectonic and volcanic features, including the volcanic edifice of Havre, the edifice surrounding Havre rock to the SE, a partial edifice to the NE, the rifted back-arc terrain to the NW, arcuate balconies (also called “concentric escarpments” by Wright et al. (2006)), numerous cone-like edifices (i.e., volcanic cones and small seamounts) distributed across both the rifted zone and the Havre volcano edifice, igneous ridge crests [“fissure ridges” as described by Wright et al. (2006)], the terrace breach feature, and a mega-scarp on the SE [interpreted as a sector collapse by Wright et al. (2006)].

All large-scale features observable in the lower-resolution datasets outboard of the outer rim did not change over the time-series analysis, with no recent observable foundering of the edifice, new cone growths, or other volumetric changes (Figure 4D). This stasis is also noted for some features inboard of the outer rim, including the caldera terrace, outer-rim arms, intra-caldera ridges (“elongated fissure cone”; Wright et al. (2006)), and volcanic craters. Several features within the caldera rim show significant changes (Figures 4B–D). Descriptions of all broad-scale features outside the caldera rim reflect a synthesis of bathymetric surveys conducted in 2002, 2012, and 2015.

The fine-scale geomorphic and geologic map of the Havre volcano produced from 1-m resolution AUV bathymetry and ROV video footage has allowed us to resolve deposition products and emplacement processes to a level of detail not achievable with low-resolution data (Figure 5). Over 4,070 events recorded during video transects revealed a variety of geologic and geomorphic features, including mass-wasting features (scree, rubble, debris, and scarps) (Figures 3A, B, F, G), lava (Figures 3D,M), lava spines (Figures 3H, S), pumice (Figure 3L), GP (Figures 3G, L, P, Q, R), breccia (Figures 3C, N), lapilli (Figures 3D, I), ash (Figures 3G, K, Q, R), scarps, cliffs (Figure 3K), contacts, boundaries, hydrothermally altered rocks and sediment, and hydrothermal venting evidence (water discharge, hydrothermal vents, and microbial and bacterial mats) (Figures 3E, J, O, T). Dominant geomorphic and geologic units are described below, focusing on the constructional and destructional differences compared to the pre-eruption survey. The following sections discuss the work by Carey et al. (2018), Ikegami et al. (2018), and McPhie et al. (2020), followed by new observations.

The geomorphic change detection (GCD) recorded a total area of 24.7 km² that experienced a geomorphic height increase and an area of 5.1 km² that experienced a height decrease between 2002 and 2012 (Figure 9A). This translates to a 0.6-km³ volume estimate of material gained and a 0.1-km³ volume estimate of material lost. A much smaller area of geomorphic height increase (3.7 km²) was recorded between 2012 and 2015. This translates to an estimated volume gain of 0.1 km³. A large area of deficit was also recorded in the 2012–2015 analysis; however, this aligns with a grid artifact from initial data collection in 2012 and is not a genuine geomorphic change (Figure 9B). The geomorphic change recorded between 2002 and 2015 is a more realistic representation. Here, an 8.6 km² area of geomorphic increase and 11.47 km² of geomorphic decrease were recorded, translating into an estimated volume growth of 0.3 km³ and a volume deficit of 0.23 km³.

Lava lobes and domes in total comprise a total volume of 0.2 km³, with half (0.1 km³) attributed to the combined dome OP emplacement; these volumes accord with calculations made by Carey et al. (2018) and Ikegami et al. (2018). Most notably, an additional volume growth of 0.001 km³ was recorded on dome OP, occurring after the 2012 survey (Figures 9B, E). This change aligns with the tulip-shaped growth emplacement of dome P in the spoon-shaped amphitheater of dome O, as discussed in Section 4.2. Profiles and contours of dome OP also support this volume change between 2012 and 2015 (Figure 10A). Growth patterns between 2002 and 2012 extend radially from the boundary of dome OP downslope toward the caldera floor in an NW orientation (Figure 9A); these are reduced in the 2002–2015 analysis (Figure 9C). The smooth-flanked pre-2012 lava dome J, between K and M, shows evidence of volume deficit on its northern side: an area of 0.03 km² underwent an estimated 0.005 km³ volume deficit between 2002 and 2012 (Figures 7, 9C).

Growth patterns between 2002 and 2012 also trace the full extent of lava C laterally spread toe across the caldera floor, including the ridge-swale chaotic domain and the soft sediment deformation domain (Figure 9A). The toe, including chaotic, low-relief, and deformed sediment domains, has an extent of 0.94 km², translating into a volume of 0.014 km³. In addition to the 0.03-km³ volume of lava C proper (Carey et al., 2018), this results in a total volume estimate for lava C-transported materials of 0.044 km³. These growth patterns are less obvious in the 2002–2015 GCD (Figure 9C). Notably, no volume change was observed for the western sheet (McPhie et al., 2020) in any of the three GCDs (Figure 5; Figure 9). An additional 0.0055-km³ geomorphic volume gain was also estimated downslope and inboard of lava domes K, M, and N (Figures 9A, C), aligning with pre-2012 lobe-shaped features along the southern caldera wall. Nearby, growth was also recorded ∼400 m NE of the major scarp that bounds lava G, H, and I (Figures 9A, C).

The largest degree of geomorphic decrease was observed along the northern caldera wall, with a total area of 2.6 km² undergoing change between 2002 and 2015 (Figures 9F, 10C), translating into a volume estimate of a 0.08 km³ deficit. No equivalent volume of deposition that corresponds to this deficit was observed within the caldera bowl, although evidence of remobilization and mass wasting is noted in the 1-m bathymetry (Figure 6M). A notable area of geomorphic decrease (0.17 km²), translating into a 0.005-km³ volume deficit estimate, was also observed immediately upslope of lava A occurring prior to 2012, as discussed by McPhie et al. (2020) (Figures 9A, D). This deficit lies beneath a major scarp (see Section 4.4.2), and the mega-blocks resting on the caldera floor downslope from the scarp comprise an area of 0.065 km².

When the GCD analysis from 2002–2012 is applied to the mega-blocks, they translate into a 0.0006 km³ volume addition. The other scarp noted on the southern caldera rim, bounding lava domes G, H, and I on their northern side (Figure 9D), shows no comparable geomorphic decrease in the GCD. The observable mega-breccia has an area of 0.03 km²; these clasts are too small to compare to the GCD (25 m resolution). A small 0.004-km³ volume deficit between 2012 and 2015 was also recorded on the eastern caldera wall between several prominent ridges; evidence of remobilization and mass wasting is noted below this area of geomorphic decrease (Figures 9G, 6N). Most of this estimated erosion formed post October 2012 as there is little observable evidence of mass wasting along the eastern caldera wall between 2002 and 2012.

Our analysis of the video footage reveals that 11 of the 12 ROV dives show evidence of hydrothermalism in a variety of forms (Figure 11): fluid venting, “chimlets” (small spires of sulfur-colored deposits; Figure 3E), mass bacterial mat growth, and hydrothermally altered sediments and rocks (Figures 3J, O, T). The dive on the eastern wall showed no hydrothermal evidence. Active hydrothermalism was observed in 7 of the 12 ROV dives in 2015 as active fluid discharge or shimmer (Figure 11). Fluid flow was observed in several different environments, including the margins of lava flows, at cracks along the caldera rim, and in the center of lava domes. This highlights the diversity of heat and fluid gradients across the caldera.

A maximum temperature of ∼69°C was recorded by the temperature probe at lava dome N, where one of the more active fluid expulsions was observed in 2015 (Figure 11B). This active vent field was the location of fluid expulsion and numerous chimlets and miniature Fe-oxidizing bacterium-blanketed chimneys (Figure 3E) with a dominant colony of crabs (unknown species). These 10–20-cm tall spiers rise from a centimeter-thick yellow deposit (likely Fe-oxidizing bacteria) that lies on top of ash deposits. Active fluid expulsion, in the form of observable shimmer (due to contrasts in the refractive index, temperature, and/or salinity anomalies) or liquid flow, was observed across the apex of dome P, near the bounding scarp of both lava domes I and G, mid-slope of the western caldera wall, and on the inner rim of the northern wall. Active fluid venting was also recorded acoustically in 2012 as a 197-m tall flare in water-column backscatter data of the EM 302 echosounder (Figure 11C) (Lamarche et al., 2016). This acoustic flare was aligned with lava dome N (Figure 11B). Evidence for past hydrothermal activity was observed as altered lavas in the northern section of the caldera wall (Wright et al., 2006) and elevated ³He concentrations (de Ronde et al., 2007).

The crustal magnetization distribution shown in Figure 12 is characterized by large values (up to 10 A/m) on the north wall of the caldera. A broad circular structure with similar large values is observed extending northward from the lava lobes to the caldera floor, encompassing the areas of positive geomorphic change. Domes O and P have smaller magnetization, i.e., not exceeding 2 A/m. An area of significantly low magnetization (<0.3 A/m) is observed in proximity to the large region of geomorphic deficit located at the base of the N caldera wall. Other areas of significantly low magnetization (<0.3 A/m) are observed north and west of domes O–P.

The 1-m resolution bathymetry maps have shown smooth, bulbous mounds associated with the northern sides of lava domes K–L–M, which we interpret as shallow intrusions, or cryptodomes. These newly discovered cryptodomes are comparable to the partially intrusive mechanism described for lava dome N by Ikegami et al. (2018). The smooth texture of the bulbous mounds reflects the intrusion into and subsequent uplift of the unconsolidated water-saturated sediment pile. The interactions between high-level intrusions with water-saturated sediment could be expected to trigger explosive magma–water interactions and associated clastic products and features. One newly identified feature on the seafloor is a small circular depression around the site where lava L was visited and sampled by the ROV in 2015. It is plausible that this circular depression is a crater, although more video and sediment analysis is required to test this idea. Giant pumice clasts are observed on top of the extrusive components of domes K–L–M–N; however, the GP density on the observed sediment above the cryptodomes of these domes is lower. These observations suggest that the GP clasts were remobilized downslope as the intrusions grew, indicating a post-GP deposition timing for the cryptodome growth.

In addition to cryptodomes associated with domes K–L–M–N, we suggest the geomorphic height increase of mounds, ∼150 m downslope and inboard of lavas K, L, M, and N (Figures 7, 9C), are cryptodomes. Although volcanoclastic deposits or mass wasting from lava domes K, M, and N may contribute some of the 0.0055-km³ volume addition, the high aspect ratio of these bulbous and even flat-topped mounds matches those of K–M–N that spread laterally downslope (Figures 7, 9C). We observe a lower GP density on these interpreted cryptodomes, which also suggests that these cryptodomes grew post-GP emplacement.

The domes/cryptodomes of K–L–M–N and wholly intrusive cryptodomes 150 m downslope indicate structural controls on emplacement, with similarly oriented caldera wall faults and dikes in the upper and mid-slope region of the southern caldera wall. The intrusive component of all domes K, L, M, and N is asymmetrical and always on the downslope side, indicating that dike-related intrusion was pushing the southern caldera wall northward at a shallow level to accommodate intrusive growth. In contrast, no cryptodomes were observed on the SW side of the caldera near the vents of lavas A–E, perhaps related to the effusion rate, structure, or lithology of the caldera and walls or complex history.

These new discoveries indicate that the 2012 Havre silicic caldera eruption included more significant shallow-level intrusion than has been identified in the past. The volume of the new cryptodomes was ∼0.0055 km³, which is approximately equivalent to the extrusive volume of domes K, L, M, and N. Their emplacement likely took place prior to and after the deposition of the GP apron, inferred to be from 18 July 2012. The longer-term cryptodome growth along faults in the caldera walls identified in this paper highlights that submarine caldera eruptions can be of long durations, similar to terrestrial analogs, and that walls could be unstable during growth.

An extended timeline of intrusion and lava effusion for this eruption is also indicated by the observation of dome P growth filling the spoon-shaped cavity of dome O between the post-eruption October 2012 survey data and the survey in 2015. This late-stage emplacement of dome P extends the known volcanic emplacement timeline from 3 months (Carey et al., 2018) to a maximum of 3 years. The presence of fresh spines and the absence of ash or lapilli on dome P suggest that it was emplaced after all ash in the water column had been deposited. Such prolonged, low-volume dome growth is common after paroxysmal eruptions of terrestrial arc volcanoes (Thouret, 1999; Gómez-Vazquez et al., 2016; Burchardt et al., 2019). The absence of GP clasts on dome P suggests that any deposited clasts were either moved downslope from mass movement or buried by the subsequent lava emplacement.

The source of the newly identified background GP deposit and its dispersal direction remain unclear (Figure 8). Possibilities include (1) a non-Havre volcano source; (2) a Havre pre-2012 eruption product; (3) a later deposit from the Havre 2012 pumice raft as ribbons of the raft were observed moving back over the caldera weeks later (Jutzeler et al., 2014); and (4) a deposit derived from sedimentation of water-logged 2012 GP from the mid-water column that had a different current direction. We were unable to speculate on the most likely option for the genesis and timing of the background GP deposit without the analysis of the physical samples or the video footage. ‘

We propose the small-scale geomorphic deficit noted along the eastern caldera wall occurred after the 2012 survey and before the 2015 survey and is the result of mass-wasting remobilizing material from the caldera wall to the floor. Geomorphic change is most obvious when comparing the 2002 and 2015 datasets and closely traces the caldera wall ridges and gullies observed in the 1-m AUV bathymetry (Figure 9G). An M3.5 earthquake occurred on 29 March 2013, ∼1.5 km N of this zone of deficit (International Seismological Centre, 2022). Although the eastern deficit has no comparable geomorphic deposition zone, a large portion of this geomorphic is likely within the <10-m detection limit threshold (Figure 10B), particularly at the toe. Here, mass-wasting lobes and sediment waves in the 1-m bathymetry provide evidence of remobilization.

The mega-breccia deposit on the southern caldera wall, draped on top of lava F and the lower western slopes of the intra-caldera ridge, provides an updated timeline for the emplacement of F relative to G, H, and I (Figure 6L). The proposed mechanism that generated the mega-breccia is the intrusion of a rhyolite dike beneath G, H, and I, which destabilized early dome growth and triggered collapse across the major scarp (McPhie et al., 2020). As a portion of the mega-breccia rests on top of lava F, the emplacement events of G, H, and I occurs post-date F, separated by an interval of time. The steep and linear headwall scarp makes it likely that the collapse occurred after the emplacement of these domes.

The source for the mega-blocks on the southwestern portion of the caldera floor (excluding the original solitary block) has been proposed to be the major scarp upslope, observed as a geomorphic deficit area upslope of lava A (McPhie et al., 2020) (Fig. j). We support this hypothesis, given that the volume of deficit and volume of mega-blocks are comparable, and the direction of flow aligns with the scarp. As no mega-blocks rest on top of lava A or C and the lava shows flow morphology around the blocks, the mega-blocks are inferred to pre-date the lava lobe emplacement.

We propose two factors contributing to the volume deficit observed along the northern caldera wall between 2002 and 2015, including (1) a high degree of slope across the steep cliffs introducing elements of inaccuracy from the multibeam echosounders (depth and system dependent), particularly the initial 2002 bathymetry survey and (2) subtle mass wasting of material that comprises the caldera wall.

The EM 302 (2012), EM 122 (2015), and Reson SeaBat 7,125 (2015) bathymetry is comparable in this area (Figures 9F, 10C); however, a minimum ∼40 m deficit from the 2002 bathymetry was observed. Areas of high slope are known to yield inaccurate bathymetric values (Lurton, 2003). A 6-m bathymetric difference on the caldera floor was previously noted, with the 2002 and 2015 low-resolution grids both offset ∼6 m from the 2012 and 2015 1-m AUV bathymetry. Bottom detection algorithms vary significantly between systems, affecting their ability to locate the seafloor in the presence of multiple stacked clasts of GP. Given the comparable bathymetry from 2012 to 2015, the older multibeam echosounder used in 2002 may have been unable to accurately rectify this area of high slope.

While mass-wasting features noted in the 1-m bathymetry, such as head scarps, lateral scarps, scars, and debris deposits, reveal that extensive remobilization of material along the steep caldera wall area likely contributes to some of the observable deficit, no corresponding area of deposition is noted on the caldera floor. A portion of the erosion deposit that corresponds to these mass-wasting features may be spread across the caldera floor but obscured due to the 10-m detection limit threshold applied in GCD processing; however, this would not account for the 0.08-km³ volume deficit. However, crustal magnetization in this region (Figure 12) shows a broad area of very low values (<0.3 A/m). This could represent an expression of the chaotic redistribution of magnetized blocks following remobilization along the caldera wall. This breaks down the coherent orientation of the magnetized blocks, inducing a random distribution of the blocks and a significant reduction in the crustal magnetization (Lerner et al., 2022).

The most active field of hydrothermalism observed in the video footage aligns with the dominant WE fault features that pierce lava N (Figure 11B). This vent field lies ∼270 m from the location of the only active fluid expulsion evidence noted in the 2012 survey (from the 197-m-tall acoustic flare in the 30-kHz echosounder). This flare resembles those with a gas phase, suggestive of CO₂ gas or liquid-phase CO₂ or sulfur, as observed by Konno et al. (2006); Linke et al. (2014); Nakamura et al. (2015); and Stucker et al. (2017), which would have a high enough impedance contrast to seawater to be observable.

Evidence of hydrothermalism observed across all videos and terrain during the 2015 survey indicates that hydrothermalism is widespread across the Havre volcano. The hydrothermalism observed in 2015, however, suggests a diffuse or ephemeral system, lacking voluminous black smokers or associated endemic biota (e.g., Bathymodiolus mussels, stalked barnacles, and alvino caridina shrimps) commonly observed at other caldera volcanoes along the Kermadec arc (de Ronde et al., 2007; Wysoczanski and Clark, 2012). Furthermore, in proximity to the observed vent fields, we do not find clear evidence of magnetic “burn holes” in the magnetic anomaly data (Figure 12), i.e., localized areas of demagnetization caused by a hydrothermal alteration (Caratori Tontini et al., 2019). The broad area of observed negative magnetic anomalies shown in Figure 12 is likely too large to be caused by hydrothermal alteration effects, particularly considering the limited hydrothermal manifestations observed on the northern caldera wall. On the other hand, the area of lobes A–F, where several active hydrothermal manifestations occur, shows intense magnetization values. This may indicate that the current active hydrothermal system at the Havre volcano may be young, possibly post-2012 eruption, where the hydrothermal system may have not altered a significant volume of rocks yet as the time scale of such an effect on magnetic data is >∼100–1,000 years (Caratori Tontini et al., 2019). However, the demagnetized areas located west and north of domes O–P lie in proximity to hydrothermal manifestations. These regions were not affected by the 2012 eruption, and they could be related to the presence of a pre-existing and still ongoing hydrothermal system.

We have no pre-2015 video footage, so we were unable to speculate whether the Havre volcano hydrothermal system is persistently diffuse or whether the 2012 eruption disrupted a formerly fulsome hydrothermal system. Re-surveying the area using towed or ROV cameras would shed light on the colonization or recolonization habits of these vents or their change in the flow regime.