Hydroacoustic Signals Originating from Marine Volcanic Activity at Kadovar Island, Papua New Guinea, Recorded by the Comprehensive Nuclear-Test-Ban Treaty International Monitoring System

Hydroacoustic signals originating from marine volcanic activity at Kadovar Island (Papua New Guinea), recorded by the Comprehensive Nuclear-Test-Ban Treaty (CTBT) International Monitoring System (IMS) hydroacoustic (HA) station HA11 Wake Island (USA), are examined herein. Episodes of high volcanic activity were identified on two occasions, separated by a period of 1 month. The events studied pertain to an initial eruption series during a period between January and February 2018. Based on local visual observations, the Kadovar volcano began to erupt at the summit and then created a new vent spot near the coast. This series of events also included the collapse of a lava dome. Direction-of-arrival estimates for the hydroacoustic signals detected at HA11 were computed using a cross-correlation technique, which allowed for the discrimination between hydroacoustic signals originating from the Kadovar volcanic activity and numerous other hydroacoustic signals attributed to seismic activity in the Pacific Ocean. The Kadovar-related seismic signals could not be identified by regional IMS seismic stations, suggesting a submarine origin of these events. On the other hand, hydroacoustic signals originating from the Kadovar volcanic activity were identified by the seismometer at Manus Island, which is located between Kadovar and HA11. The study suggests that a series of explosive bursts followed by an unusual rumble and a broadband signal plus rumble may constrain the time of the lava dome collapse event at Kadovar Island to 00:30 UTC, 00:33 UTC, and 00:46 UTC on 09 February 2018. Given the compatibility of this observation with the tsunami generation reported by eyewitnesses on the nearby island of Blup Blup, the authors interpret this particular hydroacoustic signal as being a remote observation of this tsunamigenic event. The objective of this study was to assess the potential added value of IMS hydroacoustic data for remote surveillance of geohazards in otherwise sparsely monitored areas.


Introduction
The history of remote monitoring for submarine volcanoes using global seismo-acoustic networks is described in the recently published literature (Tepp & Dziak, 2021), which includes hydrophone arrays of the Sound Surveillance System (SOSUS), the Polynesian Seismic Network (RSP) for T wave (i.e., an acoustic wave in the ocean which has been converted to/from a seismic wave at the seafloor-water interface) detection, and the hydroacoustic network of the Comprehensive Nuclear-Test-Ban Treaty (CTBT) International Monitoring System (IMS). The hydrophone arrays at Wake Island (US territory) of the CTBT IMS and the former SOSUS have proven to be sensitive in detecting submarine volcanic eruptions from the Pacific basin because of their location in the central Pacific. For example, the 2010 South Sarigan seamount eruption activity was investigated using hydroacoustic, infrasound, and seismic data by Green et al. (2013), and the energy release during the largest explosion was studied by Talandier et al. (2020). Tepp et al. (2019) compared the hydroacoustic data with the seismic data of volcanic activity of the 2014 Ahyi seamount eruptions in the Northern Mariana The views expressed herein are those of the authors and do not necessarily reflect the views of the CTBTO Preparatory Commission. The use of particular designations of countries or territories does not imply any judgement by the Commission as to the legal status of such countries or territories, of their authorities and institutions or of the delimitation of their boundaries.
Islands. Detailed information on CTBT IMS hydroacoustic stations is provided in the Appendix.
A volcanic island can collapse and sometimes cause a devastating tsunami. For example, the Anak Krakatau Island in Indonesia collapsed during eruptions in 2018 and caused a tsunami that devastated the nearby coastlines (Grilli et al., 2019;Muhari et al., 2019;Ye et al., 2020;Hunt et al., 2021). Such risks exist along all of the world's marine volcanic chains. Therefore, the establishment of remote monitoring for volcanic eruptions of uninhabited remote islands or submerged seamounts is a topic of high interest, especially for areas that are not accessible, or where in situ observations are lacking. Additionally, volcanic eruptive intervals vary, i.e., from a few years to thousands of years, depending on individual volcanoes (Deligne et al., 2010;Croweller et al. 2012;Papale, 2018) or on the duration of the volcanic life, which is sometimes very short (less than a few weeks) (Plank et al., 2020). Therefore, volcanic activity originating from such events is likely to be missed and the associated risks to human activity are difficult to evaluate unless one can resort to remote monitoring techniques.
Implementation of instrumental and/or visual observations for volcanic activity in the ocean is challenging for areas where volcanoes are located on uninhabited islands or are completely submerged. Therefore, interpretation of far-field observations is essential for establishing an effective method for remote monitoring of volcanic eruptions in the ocean. In this work, we associate locally reported volcanic events with hydroacoustic signals recorded in the far field. A series of volcanic events at Kadovar, one of the small volcanic Schouten Islands in Papua New Guinea (Fig. 1), has provided an opportunity to study how hydroacoustic signals originating from such volcanic activity can be identified in the far field. In this study, we examine the hydroacoustic data acquired by the water-column-suspended hydrophones of the CTBT IMS station HA11 Wake Island. We analyze the hydroacoustic signals with reference to the observations and the documented events for a series of volcanic activity at Kadovar.

Documented Eruptions of the Kadovar Volcano
The volcanic history of the last known eruptive activity at Kadovar can be traced back to the year 1700 (Simkin & Siebert, 1994), but comprehensive documentation is lacking. The earliest existing detailed report of volcanic activity at Kadovar is a thermal anomaly which was confirmed in 1976; however, this was not associated with an eruptive event . It has been reported that past volcanic activity in the region where Kadovar Island is situated in the western part of the Bismarck volcanic arc, such as the volcanic flank collapse of Ritter Island, is thought to have generated a tsunami, based on submarine surveys conducted in the area (Johnson, 1987;Silver et al., 2009). As for Kadovar Island itself, however, accounts of documented volcanic activity are scarce, except for a short period of seismic unrest in 2015, which was reported by the Rabaul Volcanological Observatory (RVO) (Global Volcanism Program, 2018).
More recently, eruptions were confirmed around midday local time (UTC ? 10 h) on 05 January 2018 initially by eyewitnesses; the eruption was also captured by Himawari-8 satellite imagery at 02:20 UTC on the same day (Global Volcanism Program, 2018). In situ seismometers are effective instruments for monitoring volcanic activity over time, i.e., volcanic seismicity or eruptions. However, there are no in situ instruments to monitor the volcanic activity on Kadovar Island, and therefore remote monitoring techniques and/or visual observations must be relied upon. Although Kadovar Island is inhabited with a population of approximately 600 persons, all residents were evacuated to Blup Blup Island, at a distance of 12 km (Fig. 1b), during the period when the volcano was active, according to the United Nations Office for the Coordination of Human Affairs (UN OCHA) online document (UN OCHA, 2018). Residents evacuated to Blup Blup Island witnessed the lava dome collapse and the resulting tsunami. This episodic event was confirmed by satellite imagery analysis (Plank et al., 2019). Although these observations were conducted at different points in time due to satellite passage tracks, they still provide 1354 H. Matsumoto et al. Pure Appl. Geophys. valuable ground truth for this series of volcanic activities.

Key Episodes of the 2018 Kadovar Eruptions
Because of a lack of in situ monitoring, little ground truth was available for this series of Kadovar volcanic eruptions; however, some key episodes of this series of volcanic activity, with their dates and times, were identified using local reports including flyover photos. Volcanic seismicity increased a few weeks before the first eruption on 05 January 2018, according to the Asia-Pacific Economic Cooperation (APEC) conference reports submitted by Papua New Guinea. A short report by UN OCHA also contains a description of a new vent spot near the coast, which was created between 01:30 UTC and 04:00 UTC on 11 January 2018 (UN OCHA, 2018). This ground truth can also be confirmed by comparing two local flyover photos taken at each of the two times mentioned above. Furthermore, the Global Volcanism Program (GVP) stated in a report that a fracture of the flank had been confirmed on 06 January 2018 and that this fracture had been increasing in width, and a new vent spot near the coast exhibited ''vigorous'' activity on 12 January 2018. On 13 January 2018, the RVO confirmed with a local flyover observation five vent spots in total, three at the summit and two at the coast, and a lava dome visible on the coast (Global Volcanism Program, 2018). Imagery from satellite synthetic aperture radar (SAR), Sentinel-1 (S1), also led to the identification of a lava dome on the coast on 14 January 2018, which was not observed on 10 January 2018 when a flyover by Fine-mode Advanced Land Observing Satellite-2 (ALOS-2) could access the island (Plank et al., 2019). During the period 18-22 January 2018, the RVO estimated that the horizontal extent of the Maps of the western Pacific Ocean area covered by the present study. a The locations of Kadovar Island (red star) and the IMS HA11 hydroacoustic station at Wake Island (orange triangles). The distance between Kadovar Island and HA11 is about 3500 km. HA11 is composed of the north and the south triplets denoted by H11N and H11S (orange triangles). Dark and bright red triangles represent abovesurface and submarine volcanoes registered by the Smithsonian Global Volcanism Program. Radial marks are back azimuth (BAZ) from H11N in increases of 15°. b An enlarged map around Schouten Islands, Papua New Guinea. Two straight lines indicate geodesic paths bearing from Kadovar Island to H11N and H11S Vol. 180, (2023) Hydroacoustic Signals Originating from Marine Volcanic Activity at Kadovar Island lava dome ranged between 150 and 200 m out to sea (Global Volcanism Program, 2018). Additionally, the satellite imagery suggested that the speed of the lava dome growth slowed after 23 January 2018 (Plank et al., 2019). Overall, three key episodes during the initial 1-month series of the Kadovar eruptions can be identified: (1) on 11 January 2018, when a new vent spot near the coast was first confirmed, (2) a period from 14 to 23 January 2018, when the lava dome growth was rapid and extended offshore, and (3) the period after 23 January 2018, when the growth of the lava dome slowed. The first episode on 11 January 2018, during which a new vent spot emerged near the coast, appears to have produced more hydroacoustic signals than the other two episodes; however, the volcanic activity lasted during the entire period between 11 and 23 January 2018 and then ceased gradually after 23 January 2018. The first episode therefore suggests a phase of intense underwater activity in addition to what was visible on the surface.
Another notable episode occurred on 09 February 2018, when part of the flank along with the lava dome on the coast collapsed, causing a tsunami (Plank et al., 2019). Blup Blup Island experienced 5-6 tsunami waves all less than 1 m in height. The activity of the nearshore coastal vent spot continued after this episodic event and, a few days later, lava flows connected to the island. A lava dome grew again, albeit with a slightly smaller footprint (Plank et al., 2019). According to the satellite imagery, the activity of the coastal vent spot lasted until August 2018 (Plank et al., 2019). The new lava dome did not collapse and, to the best of the authors' knowledge, remains at the time of writing of this manuscript.

Characteristics of Hydroacoustic Recordings Associated with the Kadovar Eruption
This study examined hydroacoustic recordings originating from the initial stage of the 2018 Kadovar volcanic activity, a period from January to February 2018. The hydroacoustic data were recorded by the water-column-suspended hydrophones of the IMS hydroacoustic station HA11 at Wake Island. The analysis period includes not only volcanic eruptions but also flank and lava dome collapses into the sea, allowing in some cases the formulation of hypotheses for associating some of these volcanic phenomena with far-field hydroacoustic signals recorded at a distance of approximately 3500 km from the source (Fig. 1a).
The first step of the analysis consists of separating the Kadovar volcanic activity hydroacoustic signals recorded at HA11 from those originating from other hydroacoustic sources. Figure 2 shows examples of our processing of the HA11 hydroacoustic dataset. The waveforms shown in Fig. 2a were recorded between 09:40 UTC and 09:50 UTC on 11 January 2018 when the Kadovar volcanic activity was relatively high. A frequency pass-band filter between 10 and 60 Hz was applied in order to suppress lowfrequency signals associated with the ocean microseismic background and high frequencies, where the signal-to-noise ratio (SNR) of the Kadovar arrivals is low. Based on the geodesic path from HA11 to Kadovar, arrivals with a back azimuth of 225°were associated with Kadovar Island (Fig. 1a). The signal arrival group shown in Fig. 2 is one of the loudest received during the period from January to February 2018. The group arrives first at the south triplet, followed by the north triplet with about a 1-min time lag; the north triplet (H11N) is distanced approximately 100 km further from Kadovar Island than the south triplet (H11S) (Figs. 1a and 3a). The amplitude of the arrivals is different between the two triplets, which is attributable to the fact that the path between Kadovar and H11S is partially obstructed by the Orluk Atoll, whereas the path to H11N is unobstructed (Fig. 4). H11N consistently showed a better SNR than H11S (see Fig. 4 and its caption for details), and therefore the presentation in this paper focuses on H11N.
Spectrograms of time series including the arrival group make it possible to visualize the timefrequency characteristics of the Kadovar volcanic activity. Figure 2b shows a spectrogram for the period from 08:00 UTC to 12:00 UTC on 11 January 2018 at H11N1 (the first hydrophone sensor of the H11N triplet). The hydroacoustic signal associated with the volcanic eruption can be seen at 09:45 UTC.
A signal having frequency content lower than 30 Hz is also seen at 09:25 UTC. This signal is associated with an earthquake near Torishima Island, south of Japan, with a magnitude of 3.9, which occurred at 08:50:33 UTC (International Seismological Centre, 2018).
The data presented in the spectrograms were calibrated by taking into account the frequencyamplitude-phase (FAP) response of the hydrophone provided by the CTBT's IMS (Fig. 3b). This example clearly shows the broadband nature of the arrivals, which are interpreted as originating from the Kadovar volcanic eruptions. The power spectral density (PSD) of the arrival group recorded at H11N1 is compared in Fig. 2c to the ocean background noise of the timeseries preceding the arrival. The PSD of ocean background noise was derived from the 1-h period before the arrival of the group. The PSD of the volcanic event shown in this figure exceeds the ocean background noise by 30 dB in the frequency range An example of hydroacoustic signals associated with the Kadovar volcanic activity acquired at the HA11 station at Wake Island on 11 January 2018 when volcanic activity, e.g., a new vent spot near the coast, was first confirmed. a Waveforms recorded by six hydrophones of the north and the south triplets labeled H11N1 through H11N3 and H11S1 through H11S3, respectively, for the period from 09:40 UTC to 09:50 UTC. Band-pass filter with 10 Hz to 60 Hz is applied to the original time-series dataset to emphasize hydroacoustic signals for the south triplet particularly. b Spectrogram of H11N1 for the period from 08:00 UTC to 12:00 UTC on the same day, including signal induced by an earthquake. Note that the waveform raw and de-meaned time series shown in the top box is not filtered. c Power spectral density (PSD) of the signal of H11N1 at 09:45 UTC compared with the ocean ambient noise background. PSD associated with the Kadovar volcanic eruptions is broadband. d Spectrogram focusing on a period from 09:44 UTC to 09:48 UTC. A long-lasting signal with relatively narrow frequency content was recorded between two prominent signals. All of these signals are originated from the Kadovar volcanic activity as suggested by the back azimuth presented in Fig. 6 Vol. 180, (2023) Hydroacoustic Signals Originating from Marine Volcanic Activity at Kadovar Island a Cross-sectional bathymetric profiles between Kadovar Island and the IMS hydroacoustic station, HA11 at Wake Island, and b detailed bathymetry around Mortlock Islands with the geodesic paths from Kadovar to H11N and H11S. The red and blue lines shown in panel a represent the bathymetry along the geodesic paths from Kadovar to H11N and H11S, respectively. The figures show that an atoll, namely, Orluk Atoll, is located along the direct path from Kadovar to H11S at a distance of approximately 1700 km, partially obstructing the SOFAR channel. The direct path to H11N passes a shallow seamount area near the Mortlock Islands; however, it is deep enough relative to the SOFAR channel axis to not significantly obstruct the signal propagation 1358 H. Matsumoto et al. Pure Appl. Geophys. between 10 and 30 Hz. At higher frequencies, the PSD decreases gradually. Below 5 Hz, these arrivals are indistinguishable from the ocean background noise.

Direction of Arrival Derived by Crosscorrelation Analysis
Cross-correlation analysis was applied to the HA11 hydroacoustic data for the period from January to February 2018 in order to quantify the volcanic activity at Kadovar over time from the point of view of hydroacoustic signals generated by the volcano. Details concerning this analytical procedure can be found in the previous study pertaining to volcanic activity at Ioto Island in Japan (Matsumoto et al., 2019). Since the SNR of the arrivals originating from the Kadovar volcanic activity was relatively low, a sufficient time-window overlap had to be employed in the cross-correlation analysis in order to reduce the likelihood of missing arrivals; therefore, a 25 s overlap with a 30-s sliding time-window was used. The frequency range of the analysis was set to be between 10 and 60 Hz, as this was the band in which the signals originating from the volcanic activity exhibited a high SNR, as demonstrated in the PSD.
The triplet data cross-correlation analysis makes it possible to determine the direction of arrival (DOA) of each incident hydroacoustic signal. DOAs around the true back azimuth to Kadovar Island and their detection counts estimated from H11N for the entire period of January and February 2018 are shown in Fig. 5. Similar to the analysis conducted in the earlier study on the volcanic activity at Ioto Island (Matsumoto et al., 2019), the criteria for labeling an arrival as originating from the Kadovar volcanic activity were consistency (i.e., summation of the cross-correlation time lag between each sensor pair in the triplet, generally referred to as cross-correlation residual, which should be zero in the case of a theoretically perfect cross-correlation), incident signal apparent horizontal wave speed (phase velocity) between hydrophones representative of an H-phase (i.e., an acoustic wave originating in the water column) propagating in the Sound Fixing and Ranging (SOFAR) channel, and back azimuth pointing in the direction of the source. The SOFAR channel is a natural acoustic waveguide which develops around the sound-speed minimum in the oceans (Jensen et al., 2011). It acts as a natural waveguide for hydroacoustic waves and minimizes energy loss from interaction with boundaries, thus enabling efficient propagation of underwater sound. The thresholds for the present analysis were set as consistency of less than 0.1 s, phase velocity between 1.33 km/s and 1.63 km/s, and back azimuth between 224°and 227°. It should be noted that a larger consistency number is associated with a less accurate estimation of the DOA. In order to take into account the uncertainty in the relative distance between the hydrophones suspended within the triplet, we reject arrivals which have a consistency larger than 0.1 s or which deviate by more than 10% from the ideal range of phase velocity for a horizontally propagating hydroacoustic wave. Since it is impossible to follow the actual location of three hydrophones in the water column, and because the currents and the sound speed may change with time, we assume the parameters for DOA estimation remain fixed during the observation. It should also be noted that a part of the HA11 dataset was not available for some time between 08:00 UTC on 05 January 2018 and 00:00 UTC on 10 January 2018, as a result of a temporary equipment outage at the station.
With this caveat in mind, the cross-correlation analysis indicates that no significant hydroacoustic signals from Kadovar were detected at HA11 on 10 January 2018; the first identified hydroacoustic signal occurred around 09:00 UTC on 11 January 2018, although the continuous data were available since 00:00 UTC on 10 January 2018. It should also be noted that two detections were made at 11:21 UTC on 01 January 2018, before the temporary interruption of the HA11 data stream at 08:00 UTC on 05 January 2018. We confirmed that these two detections were identified as a double-count due to the cross-correlation time-window overlap, because the back azimuths of these two detections are estimated to be 224.3°and 225.7°with a time lag of 10 s between them, and therefore this was counted as a single arrival. Additionally, the phase velocity and the back azimuth of these detections were respectively slightly slower (* 1.41 km/s) and offset (* 0.8°) than those of most other detections associated with the eruptive Vol. 180, (2023) Hydroacoustic Signals Originating from Marine Volcanic Activity at Kadovar Island 1359 series of Kadovar volcanic activity starting from 11 January 2018 (Fig. 5, top panel). However, it is noted that this signal cannot be clearly recognized in the spectrogram (Supplementary Fig. 1). The accuracy of event location obtained from cross-correlation analysis of IMS hydrophone triplet data reported in the past for other studies, e.g., Nielsen et al. (2021), suggests that the same method applied to the HA11 recordings of the volcanic activity at Kadovar should lead to a location accuracy of better than a few tens of km, considering the distance of propagation. Plank et al. (2019) identified 30 seismic events in the Kadovar region, including undetected weaker events during the period between the beginning of January and the end of June 2018, so there is a high degree of confidence that the volcanic seismicity started at that time. Therefore, we assessed that this early signal might be associated with the early undersea volcanic seismic activity of Kadovar Island before the beginning of the eruption sequence.
No other likely seismic precursors of the volcanic activity could be detected by HA11 in early January.
The first documented above-surface eruption of this series at Kadovar Island was captured at 02:20 UTC on 05 January 2018 by satellite imagery (Plank et al., 2019), a time during which HA11 was transmitting data. However, no hydroacoustic signal from Kadovar could be detected around this time on 05 January, suggesting that this first part of the eruption series was occurring above water. Alternatively, it is also possible that a hydroacoustic signal was produced but was too weak to be detected at HA11. The cross-correlation analysis shows that the volcanic activity picked up momentum on 11 January 2018 and was observed without significant interruption during the entire day on 12 January 2018. The maximum activity recorded gave as many as 78 detections per hour around 19:00 UTC on 14 January 2018, followed by a gradual decrease in volcanic activity (Fig. 5). The activity increased again sharply on 19 January 2018, albeit with fewer detections than Following a quiet period of more than 10 days, during which no hydroacoustic arrivals associated with volcanic activity at Kadovar were detected, an abrupt increase in activity was again observed on 09 February 2018. This lasted only 1 day. Further occasional detections continued beyond the end of February 2018 (Global Volcanism Program, 2018). Overall, two episodes of high activity, separated by a quiet period, were identified in the hydroacoustic data: the first period started on 11 January 2018, and the second period started on 09 February 2018. In the following section, the analysis focuses on these two periods and the detailed features of the detected signals.

First Episodic Activity from 11 January 2018
The cross-correlation analysis shows that the first hydroacoustic detections of the major series of Kadovar volcanic activity started at 09:45 UTC on 11 January 2018 (Fig. 5). The spectrogram focusing on the first detected hydroacoustic signal associated with the Kadovar volcanic activity at H11N1 on 11 January 2018 is shown in Fig. 2d. Corresponding DOAs by cross-correlation analysis and the rootmean-square (RMS) amplitude of the raw data over time are provided in Fig. 6. The cross-correlation analysis suggests that the group of arrivals originating from the Kadovar volcanic activity lasted at least 2 min. The group can be divided into three independent signals centered at 09:45:22 UTC, 09:46:20 UTC, and 09:46:40 UTC, respectively. The highfrequency content of the first, second, and third signals is up to 100 Hz, 40 Hz, and 80 Hz, and their durations are 30 s, 40 s, and 10 s, respectively (Fig. 2d). Both the first and the third signals have typical characteristics of hydroacoustic arrivals from a submarine volcanic explosive eruption, i.e., a short duration with broadband frequency content Chadwick et al., 2008;Dziak et al., 2012;Heaney et al., 2013;Tepp et al., 2019).
The second signal displays some unique features, and is a longer-duration event. The frequency content of this signal is mostly below 40 Hz and its duration is 40 s, suggesting that the source mechanism is different from that of the first and third events. Data from the nearest IMS seismic station, auxiliary station AS75 PMG at Port Moresby, Papua New Guinea, approximately 700 km from Kadovar Island, were therefore considered in the search for further information. The seismometer recording included a 40-min-long signal train with peak frequency between 10 and 15 Hz identified from 08:40 UTC, followed by some impulsive bursts (Fig. 7b). Although one cannot exclude a priori that those seismic signals could be associated with the Kadovar volcanic activity, comparing the seismic recording at PMG to the HA11 data the recording patterns did not make it possible to confirm such an association because the number of arrivals and the timefrequency signal features between the two stations did not coincide. Additionally, the seismic data from the Manus Island seismometer, which is located between Kadovar and HA11 and is part of the Australian National Seismograph Network (ANSN, network code AU) is examined (Fig. 7c). The distance from Kadovar to the station at Manus Island (MANU) is 354 km. A dense chain of T wave arrivals was identified in the MANU data during the period between 08:20 UTC and 09:10 UTC. Given the travel time of the hydroacoustic signal from Kadovar to MANU, the origin time is estimated to be 4 min prior to the recording time. On the other hand, signals recorded at PMG are dominated by seismic waves, because most of the propagation path between Kadovar and PMG is on land, and therefore the time lag between a source at Kadovar and a receiver at PMG is shorter than the lag time between Kadovar and MANU. Therefore, the interpretation is proposed that the 40-min-long train of seismic signals recorded at PMG around 09:00 UTC could be associated with a series of the Kadovar volcanic eruptions. One of the loudest signals was recorded by the MANU seismometer at 09:05 UTC, and it is worth noting that the same signal might also be identified at HA11 at 09:45 UTC shown in Fig. 2d. However, the other loud signals recorded at 10:41 UTC, 11:12 UTC, and 11:28 UTC at MANU could not be obviously Vol. 180, (2023) Hydroacoustic Signals Originating from Marine Volcanic Activity at Kadovar Island 1361 identified in the recording at HA11 compared with the signal at 09:05 UTC. Similar situations can be found in the literature, for example, a comparison of seismometer and hydroacoustic observations associated with the Ahyi seamount submarine eruptions in the Northern Mariana Islands showed that a near-field seismometer sometimes recorded the volcanic eruptions as relatively long-duration signals (Tepp et al., 2019) whereas the hydrophone signals were shorter in duration. Steep slopes in particular, such as those that define Kadovar Island, are more efficient at transferring seismic energy into hydroacoustic energy (Talandier & Okal, 1998); therefore, volcanic earthquakes from Kadovar can produce shorter and sharper T waves. Although some of the detected signals at PMG can correspond to those at MANU, further discussions of the correlation of each signal between PMG and MANU may be pursued in a future study. It should also be noted that the PMG IMS seismic station is an auxiliary three-component seismic station, consisting of a single broadband seismometer. It may be possible to estimate DOA from a three-component seismometer, but possibly with high uncertainty. Obviously, this uncertainty may be reduced if more than one three-component seismometer or seismometer array is available. Nevertheless, a single three-component seismic station such as MANU should provide sufficient information to confirm that the arriving signals could be associated with the Kadovar volcanic activity. This analysis was however not carried out within the scope of this work.
According to the bulletin report by the GVP (Global Volcanism Program, 2018), the fracture of the southeast flank of Kadovar Island was confirmed on 06 January 2018. It extended from the summit to the coast and was seen to have become wider on 12 January 2018, indicating that a part of the flank collapsed and flowed into the sea. It is not possible to constrain the time of the fracture collapse from local reporting; however, it is plausible that the first loud undersea explosion shown in Fig. 2d could have contributed to triggering the fracture collapse on the flank. It could also possibly be the other way around-the flank collapse formation could have released the pressure on the magma system or provided a conduit, leading to the explosive activity.

Second Episodic Activity, With a Peak on 09 February 2018
According to the GVP (Global Volcanism Program, 2018), a partial collapse of the lava flow was observed at 08:30 UTC on 01 February 2018. However, there was no detection of hydroacoustic signals at that time at HA11 (Fig. 5). MANU detected a T wave arrival at 08:33 UTC followed by the discrete signals, suggesting that this collapse could be associated with explosions because of the frequency content and the duration time ( Supplementary Fig. 2). This episode may have produced the hydroacoustic signals, but these may have been too weak to be detectable in the far field.
Another significant episode occurred on 09 February 2018: the newly grown lava dome collapsed and, following a series of strong explosions, disappeared completely under the sea (Plank et al., 2019). Similar phenomena were observed during the 2016-2017 eruptions of Bogoslof Island, Alaska, USA; two of three lava domes created during this eruptive series were destroyed by explosions within 5-10 days after their formation (Waythomas et al., 2020). Plank et al. (2019) examined the SAR satellite data between 24 January and 21 February 2018 and found evidence indicating that the collapse comprised not only the lava dome but also a considerable part of Kadovar Island after this episode. Uncertainties regarding the timing of this collapse series remain because no in situ ground truth observations are available.
HA11 data for the period between 23:00 UTC on 08 February 2018 and 02:00 UTC on the following day were reviewed because several hydroacoustic signals associated with the Kadovar volcanic activity were detected since the last detection on 05 February  (Fig. 5). Spectrograms of HA11 signals are shown in Fig. 8. A 3-h spectrogram with back azimuths is shown in Fig. 8a. Three prominent signals at 23:23 UTC, 00:03 UTC, and 00:41 UTC were associated with earthquakes not originating from Kadovar, as indicated by the back azimuths of 357°, 217°, and 1°, respectively. Hydroacoustic signals originating from an earthquake can be discriminated easily from other sources by their relatively long duration and low-frequency content (\ 30 Hz) characteristics. Shorter and higher-frequency bursts (i.e., spike-like signals) are also recognized in the spectrogram. These are the signals which we consider more closely in the analysis below, as they are more typical of submarine volcanic activity.
Focusing on the period from 00:20 UTC to 01:40 UTC on 09 February 2018 (Fig. 8b) and a narrower band of azimuths around the DOA from Kadovar, it is seen that more detections become visible. It should be noted that the cross-correlation time-window overlap leads to some double-counts. Considering the propagation distance of 3500 km and the fact that the DOAs cover a very small range of angles (within 0.5°) for the various 1-min arrival clusters, the signals can be interpreted as coming from the same source without reflections along the propagation path. The back azimuths of these signals are within a few tenths of a degree from the true back azimuth to Kadovar Island. This small offset in arrival angle is consistently negative. This offset coincides with the reported location of the newly emerged vent, which was located slightly offshore east of Kadovar (see, e.g., Figure 7 of Plank et al., 2019). Interpreting these shorter hydroacoustic arrivals as explosive eruptions suggests that at least four explosive eruptions repeated at 5-min intervals between 00:20 UTC and 01:00 UTC. A similar pattern is repeated between 01:20 UTC and 01:40 UTC. In contrast, some characteristics of the signals recorded by HA11 between 01:00 UTC and 01:20 UTC are different from those of the earlier series. The back azimuths of these arrivals are closer to the island than those of the earlier arrivals, suggesting that the hydroacoustic source(s) of these events may have been located in a shallower area. Three hydroacoustic signals with noteworthy characteristics were observed within a short period of time between 01:10 UTC and 01:15 UTC (Fig. 8c). The first and the third signals of this group are of slightly longer duration than the second signal, i.e., 0.5-1 min for the first and the third signals and 10 s for the second signals, and longer than other signals associated with Kadovar explosive volcanic eruptions. Each of these two signals is also followed by a 20-30-s ''rumble,'' which can be recognized in the spectrogram. This suggests a different source mechanism compared with the other Kadovar submarine volcanic explosions. For better clarity, these three signals are also compared side by side in Supplementary Fig. 3.
Previous near-field and far-field observations of hydroacoustic signals originating from submarine landslides are available in the literature and provide helpful information for interpreting the signals in Fig. 8c. Firstly, it is noted that the frequency content of the signals shown in Fig. 8c coincides with the frequency content of hydroacoustic signals that have been reported as originating from submarine landslides at other locations (Caplan-Auerbach et al., 2001Chadwick et al., 2012;Okal, 2003). More precisely, a low-frequency arrival followed by a long-lasting rumble is reported in Caplan-Auerbach et al. (2001) for a hydroacoustic signal associated with a submarine landslide triggered by a lava bench collapse of the Kilauea volcano in Hawaii, USA.
The in situ hydrophone at the NW Rota-1 submarine volcano in the Mariana detected longlasting hydroacoustic signals with broadband background noise, which were interpreted as being associated with submarine landslides triggered by the volcanic eruptions . There were also two potential landslide signals identified within a half-day originating from NW Rota-1, each of which had a duration of 2-4 h. This duration is quite long for a landslide and the distinction between landslide and eruption signal characteristics in this dataset is still a topic of debate in the community (Tepp, personal communication). In this case it was seen that the more broadband frequency signal also lasted longer.
Hydroacoustic signals associated with a submarine landslide after the Papua New Guinea earthquake in 1998 were also characterized by a long duration, which was comparable to that of the T wave generated by the mainshock of the earthquake (Okal, 2003). The duration time was 47 s, and the frequency range between 7 and 12 Hz was predominant for this possible submarine landslide event in that case.
A series of lava bench collapses into the sea associated with the volcanic activity of Kilauea, Hawaii, USA, was identified by near-field observations recorded at the Hawaiian Undersea Geo-Observatory (HUGO) (Caplan-Auerbach et al., 2001). Hydroacoustic signals originating from that source were also recorded by the far-field hydrophones of the Pacific Marine Environmental Laboratory (PMEL). Comparing HUGO's relatively near (50 km distant) observations and the PMEL's far-field (5000-7000 km distant) observations, shows that the rumbles remained, while high-frequency content of the hiss were attenuated more in the farfield observation. The far-field recordings from the lava bench collapse reported in Caplan-Auerbach et al. (2001) are similar to the first and third signals seen between 01:10 UTC and 01:15 UTC in Fig. 8c, where a relatively short broadband arrival is followed by a clearly identifiable rumble. A long-lasting hydroacoustic signal like a rumble was also identified Vol. 180, (2023) Hydroacoustic Signals Originating from Marine Volcanic Activity at Kadovar Island at 01:26 UTC, which was preceded by a broadband signal slightly longer in duration than the other explosive bursts recorded from this event series (Fig. 8d). Further proceeding in the comparison with hydroacoustic signals from landslides and lava bench collapses reported in the literature, it is noted that a hiss signal generated by the lava bench collapse of Kilauea lasted a few hundred seconds and had a broadband frequency content in the near field. However, it could not be clearly identified in the far field, as the hiss may have been attenuated in that case. Similar to the hydroacoustic observations during the Kilauea lava bench collapses, the fact that no hiss signals were identified at HA11 during the Kadovar lava dome collapses may have been a result of attenuation in the far field.
It is also possible that the rumble may be relatively weak because of the smaller lava dome collapse volume compared with the other events reported in the literature. The volume of lava dome collapse for Kadovar was estimated to be 726,000 m 3 (Plank et al., 2019), while those for the NW Rota-1 submarine landslide and the 1998 Papua New Guinea tsunami event are on the order of 10 7 m 3  and 10 9 m 3 (Okal, 2003), respectively. Therefore, the events reported in the literature are much larger than Kadovar. On the other hand, the Kilauea bench collapse whose volume was estimated to be 1-10 9 10 5 m 3 (Caplan-Auerbach et al., 2001) is very similar to the Kadovar case, supporting our observations that the far-field rumble signal appears to be relatively weak. Furthermore, other previous instances of seismo-acoustic observations for surficial mass movement at subaerial volcanoes suggest that the same frequency content up to 20 Hz are predominant (e.g., Allstadt et al., 2018).
The above comparison between the Kadovar data reported here and the previous literature suggests that the significant difference in size between the Kadovar lava dome collapse and the other landslide events can result in different time-frequency characteristics for the rumble signals. Overall, the characteristics regarding frequency content are similar to the previous hydroacoustic observations for submarine landslides, though the duration of the present rumble signal is relatively short, likely because of a combination of far-field attenuation effects and relatively small lava dome collapse volume.
Based on the above analysis and comparison with results reported in the literature, we therefore interpret the two explosive burst-and-rumble signals and the broadband signal plus rumble identified at 01:10 UTC, 01:13 UTC, and 01:26 UTC, respectively, as being associated with the lava dome collapse of Kadovar. Frequency content differs between the broadband burst and the rumble; i.e., the rumble signal is characterized by lower frequencies than the burst ( Fig. 8c and Supplementary Fig. 3), suggesting that this source mechanism is different from the one that generated the broadband burst, or that the burst and rumble signals describe different parts of the flank/lava dome collapse. This interpretation is supported by visual reports of these events: The GVP bulletin (Global Volcanism Program, 2018) reported a collapse of the lava dome, followed by the tsunami. Five to six tsunamis less than 1 m high were observed by witnesses at Blup Blup Island at 00:50 UTC on 09 February 2018, before the main collapse of the lava dome. Given that the tsunamis were observed before the reported main collapse of the lava dome, it is possible that a series of collapse events took place. Kadovar Island is 365 m high and is located at a water depth of approximately 1000 m (Fig. 1b), suggesting that the slope of the seafloor is relatively steep. Since tsunamis are long waves (i.e., gravity waves), it is conceivable that the lava dome collapse might also have disturbed the water mass in the deeper layer of the ocean.
The distance between Kadovar Island and Blup Blup Island is 12 km, and consequently the tsunami travel time is expected to be less than 3 min (Fig. 9). Therefore, the visual report of the tsunami suggests that the triggering event occurred a few minutes earlier than 00:50 UTC. On the other hand, the travel time of the hydroacoustic sound from Kadovar to HA11 Wake Island is approximately 40 min, assuming a typical phase velocity of hydroacoustic propagation of 1.48 km/s. The three hydroacoustic rumble signals recorded at 01:10 UTC, 01:13 UTC, and 01:26 UTC at HA11 are therefore expected to have origin times at Kadovar of 00:30 UTC, 00:33 UTC, and 00:46 UTC, respectively. Therefore, the hydroacoustic observation is consistent with the visual observation of the tsunamis at Blup Blup Island, and it is therefore proposed that the three signals recorded by HA11 at 01:10 UTC, 01:13 UTC, and 01:26 UTC (Figs. 8c and 8d) can be associated as having originated from a collapse event which triggered the tsunami. An illustration of this source mechanism is shown in Fig. 9 of Plank et al. (2019). An explosive burst following these hydroacoustic rumble signals was identified at 01:31 UTC, which might be associated with the explosions triggering the main collapse of the lava dome of Kadovar, as it is suggested that this hydroacoustic signal was generated at 00:51 UTC, i.e., within a few minutes after 00:50 UTC as documented in the GVP bulletin. Given that there was neither a tsunami reported after 00:50 UTC nor significant rumble signals recorded at HA11 at the corresponding theoretical arrival times, the effect of the main collapse on the seawater was likely small, suggesting that the lava dome was destroyed by the abovewater explosion.

Figure 9
Tsunami travel times assuming the source is located in Kadovar. Numbers along the contours are in minutes Vol. 180, (2023) Hydroacoustic Signals Originating from Marine Volcanic Activity at Kadovar Island 1367

Summary of the Evolution of the Volcanic Activity
Observations from in situ sensors were not available at the early stage of a series of volcanic eruptions of Kadovar in 2018. The present study made it possible to further investigate the Kadovar volcanic activity using remote hydroacoustic recordings from the CTBT IMS hydrophone station HA11 Wake Island at a distance of 3500 km from the source. HA11 complemented data which were otherwise available from remote seismic monitoring and visual and satellite observations (Plank et al., 2019). The emergence of a new vent spot near the coast of Kadovar Island was confirmed on 11 January 2018 for the first time by a local flyover observation 6 days after the first eruption of the summit. The hydroacoustic signals originating from the Kadovar volcanic activity recorded on 11 January 2018 by HA11 coincided with the first observations of this new vent spot near the coast. Since there was no hydroacoustic detection at HA11 on 10 January 2018 when the summit erupted, hydroacoustic detections at HA11 are interpreted as being associated with events that took place under the sea surface, which might have led to the growth of the new vent spot at the coast. Remote hydroacoustic recordings and volcanic seismicity reported by in situ observations have previously shown a good correlation, as demonstrated in an earlier study of submarine eruptions of Ioto Island (Matsumoto et al., 2019).
The growth rate of the lava dome projection measured in the initial 10 days (i.e., a period from 14 January 2018 to 24 January 2018) was more than 10 times as fast as that during the following 14 days (i.e., the period from 24 January 2018 to 07 February 2018) (Plank et al., 2019). Thus, most of the growth of the lava dome took place before 24 January 2018, which correlates with the period during which numerous hydroacoustic arrivals from Kadovar were observed at HA11 (Fig. 5). Additionally, the regrowth of the lava dome after the first collapse on 09 February 2018 showed a slightly faster rate (i.e., 1.5 times as fast) than the previous measurement period, suggesting that the vent spot became active again. Indeed, the number of hydroacoustic detections during the regrowth period (i.e., after 09 February) showed an increase. Therefore, the variation in the lava dome growth rate over time appears to correlate well with the hydroacoustic observations at HA11 for the entire period (Fig. 5).
The first episode associated with the hydroacoustic signals started on 11 January 2018. Episodes of high volcanic activity with as many as 80 detections per hour were identified on two occasions (i.e., two peaks could be identified at 19:00 UTC on 14 January 2018 and 01:00 UTC on 09 February 2018.), separated by a 10-day period of relative quiet (Fig. 5). The second episode was associated with the lava dome collapse on 09 February 2018. Two explosive bursts followed by a rumble and a broadband signal plus rumble were identified in the HA11 data and characterized in time-frequency space (Figs. 8c and 8d). Based on similarities with landslide/collapse events reported in the literature (Allstadt et al., 2018;Caplan-Auerbach et al., 2001Okal, 2003;Tepp et al., 2020), we suggest that these hydroacoustic signals are associated with the lava dome collapse of Kadovar Island, implying that the collapse times are 00:30 UTC, 00:33 UTC, and 00:46 UTC, followed by an explosion at 00:51 UTC on 09 February 2018. We examined the seismic data of the IMS PMG and ANSN MANU seismic stations around 00:00 UTC (Fig. 10), and concluded that seismic signals associated with the two episodes triggered at 00:30 UTC and 00:33 UTC might have been recorded as longlasting signals. A weak signal was identified at 00:31 UTC in the MANU data; however, there was no such signal before or after this time (Fig. 10b). The events at 00:46 UTC and 00:51 UTC could not be distinguished from the seismic background noise in the PMG data but were recognizable as a seismic event followed by a weak signal at 00:46 UTC, at least in the MANU data, suggesting an indication of the likely submarine origin of these episodes. One of the largest signals at MANU is identified at 00:57 UTC, suggesting that the origin time is 00:53 UTC, assuming this is a T wave, which is about 2 min later than the expected origin time from the hydroacoustic analysis of HA11. A nominal sound speed of 1.48 km/s was used to estimate the in-water travel times. Given the longer distance to HA11 (i.e., * 3500 km), it is more likely that sound-speed 1368 H. Matsumoto et al. Pure Appl. Geophys. variations along the propagation path may have a stronger effect on the arrival time at HA11 compared with MANU (i.e., * 350 km), to which the propagation path is shorter. Tepp et al. (2020) investigated T waves (i.e., seismo-acoustic signal) originating from the Bogoslof eruption and recorded by a seismometer at a distance of 700 km and compared with the near-field hydrophone data and regional seismic data. Further examinations may be needed to find a correspondence for each signal acquired by an intermediate-field seismic station such as MANU and the far-field hydroacoustic station HA11.

Conclusions
Following an eruption series at Kadovar Island, Papua New Guinea, the hydroacoustic dataset acquired by the CTBT IMS hydroacoustic station HA11, Wake Island, located approximately 3500 km northeast of Kadovar, was examined in order to associate the episodes reported by local and satellite observations with the far-field hydroacoustic signals. The volcanic activity of Kadovar restarted in 2018 after three centuries of rest and comprised not only an increase in vent spots but also the collapse of a flank and a newly created lava dome, which caused small tsunamis before the main collapse of the lava dome. The detection of hydroacoustic signals at Wake Island correlates well with the local observations and made it possible to constrain the timing of the lava dome collapse based on the characterized signals, suggesting at least three collapses taking place within a period of 16 min.
The direction of arrival of the hydroacoustic signals detected at HA11 were computed using a crosscorrelation technique, which allowed the discrimination between hydroacoustic signals originating from the Kadovar volcanic activity and numerous other hydroacoustic signals attributed to seismic activity in the Pacific Ocean. The series of the Kadovar eruptions demonstrated how challenging it can be to interpret signals in the absence of other corroborating information.
The detection of hydroacoustic signals in the far field over time correlates with the volcanic episodes confirmed near the coast rather than above the surface, suggesting that most of the hydroacoustic signals detected at HA11 originated under the sea surface. The present analysis suggests that a series of explosive bursts followed by an unusual rumble and a broadband signal plus rumble may constrain the time of lava dome collapse events at Kadovar Island to 00:30 UTC, 00:33 UTC, and 00:46 UTC, followed by an explosion at 00:51 UTC on 09 February 2018.
In conclusion, this analysis highlights the potential for remote hydroacoustic observations, such as those made possible by the CTBT IMS hydroacoustic stations, for the identification and monitoring of geophysical hazards in areas where in situ or local data are not available.