Search

In the present study, we propose a new approach for determining earthquake hypocentral parameters. This approach integrates computed theoretical seismograms and deep machine learning. The theoretical seismograms are generated through a realistic three-dimensional Earth model, and are then used to create spatial images of seismic wave propagation at the Earth’s surface. These snapshots are subsequently utilized as a training dataset for a convolutional neural network. Neural networks for determining hypocentral parameters such as the epicenter, depth, occurrence time, and magnitude are established using the temporal evolution of the snapshots. These networks are applied to seismograms from the seismic observation network in the Hakone volcanic region in Japan to demonstrate the suitability of the proposed approach for locating earthquakes. We demonstrate that the determination accuracy of hypocentral parameters can be improved by including theoretical seismograms for different earthquake locations and sizes, in the learning dataset for the deep machine learning. Using the proposed method, the hypocentral parameters are automatically determined within seconds after detecting an event. This method can potentially serve in monitoring earthquake activity in active volcanic areas such as the Hakone region.

Since the beginning of the 21st century, volcanic unrest has occurred every 2–5 years at Hakone volcano. After the 2015 eruption, unrest activity changed significantly in terms of seismicity and geochemistry. Like the pre- and co-eruptive unrest, each post-eruptive unrest episode was detected by deep inflation below the volcano (~ 10 km) and deep low frequency events, which can be interpreted as reflecting supply of magma or magmatic fluid from depth. The seismic activity during the post-eruptive unrest episodes also increased; however, seismic activity beneath the eruption center during the unrest episodes was significantly lower, especially in the shallow region (~2 km), while sporadic seismic swarms were observed beneath the caldera rim, ~3 km away from the center. This observation and a recent InSAR analysis imply that the hydrothermal system of the volcano could be composed of multiple sub-systems, each of which can host earthquake swarm and show independent volume change. The 2015 eruption established routes for steam from the hydrothermal sub-system beneath the eruption center (≥ 150 m deep) to the surface through the cap-rock, allowing emission of super-heated steam (~ 160 ºC). This steam showed an increase in magmatic/hydrothermal gas ratios (SO2/H2S and HCl/H2S) in the 2019 unrest episode; however, no magma supply was indicated by seismic and geodetic observations. Net SO2 emission during the post-eruptive unrest episodes, which remained within the usual range of the post-eruptive period, is also inconsistent with shallow intrusion. We consider that the post-eruptive unrest episodes were also triggered by newly derived magma or magmatic fluid from depth; however, the breached cap-rock was unable to allow subsequent pressurization and intensive seismic activity within the hydrothermal sub-system beneath the eruption center. The heat released from the newly derived magma or fluid dried the vapor-dominated portion of the hydrothermal system and inhibited scrubbing of SO2 and HCl to allow a higher magmatic/hydrothermal gas ratio. The 2015 eruption could have also breached the sealing zone near the brittle–plastic transition and the subsequent self-sealing process seems not to have completed based on the observations during the post-eruptive unrest episodes.

The 2015 unrest of the Hakone volcano in Japan, which began on April 26, generated earthquake swarms accompanied by long-term deformation. The earthquake swarm activity reached its maximum in mid-May and gradually calmed down; however, it increased again on the morning of June 29, 2015. Simultaneously with the earthquake increase, rapid tilt changes started 10 s before 07:33 (JST) and they lasted for approximately 2 min. The rapid tilt changes likely reflected opening of a shallow crack that was formed near the eruption center prior to the phreatic eruption on that day. In this study, we modeled the pressure source beneath the eruption center based on static tilt changes determined using both tilt meters and broadband seismometers. In the best-fit model, the source depth was 854 m above sea level, and its orientation (N316°E) agreed with the direction of maximum compression estimated based on focal mechanism and S-wave splitting data. The extent of the crack opening was estimated to be 4.6 cm, while the volume change was approximately 1.6 × 105 m3. The top of the crack reached to approximately 150 m below the eruption center. Because the crack was too thin to be penetrated by magma, the crack opening was attributed to the intrusion of hydrothermal water. This intrusion of hydrothermal water may have triggered the phreatic eruption. Reverse polarity motion with respect to that expected from crack opening was recognized in 1 Hz tilt records during the first 20 s of the intrusion of hydrothermal water. This motion, not the subsidence of volcanic edifice, was responsible for the observed displacement.

On 29 June 2015, a small phreatic eruption occurred at Hakone volcano, Central Japan, forming several vents in the Owakudani geothermal area on the northern slope of the central cones. Intense earthquake swarm activity and geodetic signals corresponding to the 2015 eruption were also observed within the Hakone caldera. To complement these observations and to characterise the shallow resistivity structure of Hakone caldera, we carried out a three-dimensional inversion of magnetotelluric measurement data acquired at 64 sites across the region. We utilised an unstructured tetrahedral mesh for the inversion code of the edge-based finite element method to account for the steep topography of the region during the inversion process. The main features of the best-fit three-dimensional model are a bell-shaped conductor, the bottom of which shows good agreement with the upper limit of seismicity, beneath the central cones and the Owakudani geothermal area, and several buried bowl-shaped conductive zones beneath the Gora and Kojiri areas. We infer that the main bell-shaped conductor represents a hydrothermally altered zone that acts as a cap or seal to resist the upwelling of volcanic fluids. Enhanced volcanic activity may cause volcanic fluids to pass through the resistive body surrounded by the altered zone and thus promote brittle failure within the resistive body. The overlapping locations of the bowl-shaped conductors, the buried caldera structures and the presence of sodium-chloride-rich hot springs indicate that the conductors represent porous media saturated by high-salinity hot spring waters. The linear clusters of earthquake swarms beneath the Kojiri area may indicate several weak zones that formed due to these structural contrasts.

The 2015 eruption of Hakone volcano was a very small phreatic eruption, with total erupted ash estimated to be in the order of only 102 m3 and ballistic blocks reaching less than 30 m from the vent. Precursors, however, had been recognized at least 2 months before the eruption and mitigation measures were taken by the local governments well in advance. In this paper, the course of precursors, the eruption and the post-eruptive volcanic activity are reviewed, and a preliminary model for the magma-hydrothermal process that caused the unrest and eruption is proposed. Also, mitigation measures taken during the unrest and eruption are summarized and discussed. The first precursors observed were an inflation of the deep source and deep low-frequency earthquakes in early April 2015; an earthquake swarm then started in late April. On May 3, steam wells in Owakudani, the largest fumarolic area on the volcano, started to blowout. Seismicity reached its maximum in mid-May and gradually decreased; however, at 7:32 local time on June 29, a shallow open crack was formed just beneath Owakudani as inferred from sudden tilt change and InSAR analysis. The same day mud flows and/or debris flows likely started before 11:00 and ash emission began at about 12:30. The volcanic unrest and the eruption of 2015 can be interpreted as a pressure increase in the hydrothermal system, which was triggered by magma replenishment to a deep magma chamber. Such a pressure increase was also inferred from the 2001 unrest and other minor unrests of Hakone volcano during the twenty-first century. In fact, monitoring of repeated periods of unrest enabled alerting prior to the 2015 eruption. However, since open crack formation seems to occur haphazardly, eruption prediction remains impossible and evacuation in the early phase of volcanic unrest is the only way to mitigate volcanic hazard.

To understand the initial process of the phreatic eruption of the Hakone volcano from June 29 to July 01, 2015, we analyzed infrasound data using the cross-correlation between infrasound and vertical ground velocity and compared the results of our analysis to the crustal deformation detected by tiltmeters and broadband seismometers. An infrasound signal and vertical ground motion due to an infrasound wave coupled to the ground were detected simultaneously with the opening of a crack source beneath the Owakudani geothermal region during the 2-min time period after 07:32 JST on June 29, 2015 (JST = UTC + 8 h). Given that the upper end of the open crack was approximately 150 m beneath the surface, the time for the direct emission of highly pressurized fluid from the upper end of the open crack to the surface should have exceeded the duration of the inflation owing to the hydraulic diffusivity in the porous media. Therefore, the infrasound signal coincident with the opening of the crack may reflect a sudden emission of volcanic gas resulting from the rapid vaporization of pre-existing groundwater beneath Owakudani because of the transfer of the volumetric strain change from the deformation source. We also noticed a correlation pattern corresponding to discrete impulsive infrasound signals during vent formation, which occurred several hours to 2 days after the opening of the crack. In particular, we noted that the sudden emission of vapor coincided with the inflation of the shallow pressure source, whereas the eruptive burst events accompanied by the largest vent formation were delayed by approximately 2 days. Furthermore, we demonstrated that the correlation method is a useful tool in detecting small infrasound signals and provides important information regarding the initial processes of the eruption.

In the present study, we analyze the seismic signals from a continuous volcanic tremor that occurred during a small phreatic eruption of the Hakone volcano, in the Owakudani geothermal region of central Japan, on June 29, 2015. The signals were detected for 2 days, from June 29 to July 1, at stations near the vents. The frequency component of the volcanic tremors showed a broad peak within 1–6 Hz. The characteristics of the frequency component did not vary with time and were independent of the amplitude of the tremor. The largest amplitude was observed at the end of the tremor activity, 2 days after the onset of the eruption. We estimated the location of the source using a cross-correlation analysis of waveform envelopes. The locations of volcanic tremors are determined near the vents of eruption and the surface, with the area of the upper extent of an open crack estimated using changes in the tilt. The duration-amplitude distribution of the volcanic tremor was consistent with the exponential scaling law rather than the power law, suggesting a scale-bound source process. This result suggests that the volcanic tremor originated from a similar physical process occurring practically in the same place. The increment of the tremor amplitude was coincident with the occurrence of impulsive infrasonic waves and vent formations. High-amplitude seismic phases were observed prior to the infrasonic onsets. The time difference between the seismic and infrasonic onsets can be explained assuming a common source located at the vent. This result suggests that both seismic and infrasonic waves are generated when a gas slug bursts at that location. The frequency components of the seismic phases observed just before the infrasonic onset were generally consistent with those of the tremor signals without infrasonic waves. The burst of a gas slug at the surface vent may be a reasonable model for the generation mechanism of the volcanic tremor and the occurrence of impulsive infrasonic signals.

Hakone volcano, located at the northern tip of the Izu-Mariana volcanic arc, Japan, has a large caldera structure containing numerous volcanic hot springs. Earthquake swarms have occurred repeatedly within the caldera. The largest seismic swarm since the commencement of modern seismic observations (in 1968) occurred in 2001. We investigated the anisotropic structure of Hakone volcano based on S wave splitting analysis and found spatiotemporal changes in the splitting parameters accompanying the seismic swarm activity. Depth-dependent anisotropic structures are clearly observed. A highly anisotropic layer with a thickness of ~1.5 km is located beneath the Koziri (KZR) and Kozukayama (KZY) stations. The anisotropic intensity in the region reaches a maximum of 6–7% at a depth of 1 km and decreases markedly to less than 1% at a depth of 2 km. The anisotropic intensity beneath Komagatake station (KOM) decreases gradually from a maximum of 6% at the surface to 0% at a depth of 5 km but is still greater than 2.5% at a depth of 3 km. At KZY, the anisotropic intensity along a travel path of which the back azimuth was the south decreased noticeably after the 2001 seismic swarm activity. During the swarm activity, tilt meters and GPS recorded the crustal deformation. The observed decrease in anisotropic intensity is presumed to be caused by the closing of microcracks by stress changes accompanying crustal deformation near the travel path.

Immediately after the March 11, 2011, M9.0 Tohoku-Oki earthquake, seismic activity increased remarkably beneath Hakone volcano, central Japan, at an epicentral distance of 450 km. The heightened seismicity was initiated during the passage of the large-amplitude surface waves from the main shock and continued over the subsequent 2 months. We obtained hypocenters and focal mechanisms of the seismic sequence, with the aim of clarifying the physical mechanism responsible for the remotely triggered seismicity. We used data from a dense seismic network containing 56 online permanent and offline temporary stations in and around the Hakone volcano. We found that the earthquakes that occurred during the passage of the surface waves are located at the lower depth limit of ordinary seismicity in the caldera. These earthquakes have larger magnitudes than both the ordinary seismicity prior to the Tohoku-Oki earthquake and the seismicity triggered after the passage of the surface waves. The focal mechanism that we determined is a strike–slip fault type with the P-axis in the NW–SE direction, which is consistent with the focal mechanisms of earthquakes that occurred after the passage of the surface waves and the tectonic stress field in the region. We also tried to detect missing events that occurred immediately after the passage of the surface waves, by using a waveform correlation technique. The detected events are distributed near the hypocenters of the earthquakes that occurred during the passage of the surface waves. The origin times of the first four events after the arrival of surface waves are consistent with the phases of the decrease in normal stress generated by the surface waves. The results suggest that the changes in dynamic stress due to the surface waves from the 2011 Tohoku-Oki earthquake contributed significantly to the initiation of the sequence of triggered seismic activity. Assuming that normal stress changes on the faults did play an important role in the triggering of earthquakes, we propose that fluid flow induced by the oscillation of permeability on the faults is the main mechanism for the initiation of post-Tohoku-Oki earthquakes beneath the Hakone volcano.

Autocorrelation functions (ACFs) for ambient seismic noise are considered to be useful tools for estimating temporal changes in the subsurface structure. Velocity changes at Hakone volcano in central Japan, where remarkable swarm activity has often been observed, were investigated in this study. Significant velocity changes were detected during two seismic activities in 2011 and 2013. The 2011 activity began immediately after the 2011 Tohoku-oki earthquake, suggesting remote triggering by the dynamic stress changes resulting from the earthquake. During the 2013 activity, which exhibited swarm-like features, crustal deformations were detected by Global Navigation Satellite System (GNSS) stations and tiltmeters, suggesting a pressure increment of a Mogi point source at a depth of 7 km and two shallow open cracks. Waveforms that were bandpass-filtered between 1 and 3 Hz were used to calculate ACFs using a one-bit correlation technique. Fluctuations in the velocity structure were obtained using the stretching method. A gradual decrease in the velocity structure was observed prior to the 2013 activity at the KOM station near the central cone of the caldera, which started after the onset of crustal expansion observed by the GNSS stations. Additionally, a sudden significant velocity decrease was observed at the OWD station near a fumarolic area just after the onset of the 2013 activity and the tilt changes. The changes in the stress and strain caused by the deformation sources were likely the main contributors to these decreases in velocity. The precursory velocity reduction at the KOM station likely resulted from the inflation of the deep Mogi source, whereas the sudden velocity decrease at the OWD station may reflect changes in the strain caused by the shallow open-crack source. Rapid velocity decreases were also detected at many stations in and around the volcano after the 2011 Tohoku-oki earthquake. The velocity changes may reflect the redistribution of hydrothermal fluid in response to the large stress perturbation caused by the 2011 Tohoku-oki earthquake.

We investigated the spatial distribution, temporal change and some statistical features of the swarm activity in Hakone volcano after the 2011 Off the Pacific Coast of Tohoku earthquake (hereafter, the 2011 Tohoku earthquake). Though overall spatial distribution of the activity was not much different from that observed at swarm activities in recent years, its temporal change was quite different: Contrary to recent activities in which burst-like earthquake occurrence was observed repeatedly, the activity after the 2011 Tohoku earthquake declined rather rapidly and monotonously according to an inverse power law of the elapsed time from the 2011 Tohoku earthquake. This feature was most clearly seen in the change of daily number of earthquake clusters. Another notable feature of the activity was that the b value was significantly smaller than the values of recent swarm activities. These characteristics suggest that the swarm activity was induced by the sudden increase of static stress caused by the 2011 Tohoku earthquake on March 11.

Seismic activity in the Hakone volcano at an epicentral distance of 450 km was remarkably activated just after the 2011 off the Pacific coast of Tohoku Earthquake. More than 1600 events were observed in the caldera of the volcano, from 15:00 on March 11 to 12:00 on April 2. To clarify the relationship between the occurrence of the main shock and the induced activity in the Hakone volcano, we investigated the spatial distribution of hypocenters and temporal changes of the seismicity, and we examined seismographs of the main shock to identify small local events during the passage of the surface waves. Hypocenters determined with the double-difference method are mostly distributed in the N-S direction, showing several clusters of seismicity. Focal mechanisms of major earthquakes are predominantly strike-slip having the P axis in the NNW-SSE direction. These features of the hypocenter distribution and the focal mechanisms are consistent with those of earthquakes that occur ordinarily in the Hakone volcano. The onset of the local event was initiated during the Love and Rayleigh waves from the main shock, suggesting that large dynamic stress changes of 0.6 MPa dominantly contributed to initiate the sequence of seismic activity.

[1] A swarm earthquake sequence is often assumed to be triggered by fluid flow within a brittle fault damage zone, which is assumed to be highly permeable. However, there is little seismological evidence of the relation between the fluid flow within the fault damage zone and the occurrence of swarm earthquakes. Here, we precisely determine the hypocenters and focal mechanisms of swarm earthquakes that occurred in the caldera of Hakone volcano, central Japan, using data from a dense seismic network. We demonstrate that the swarm earthquakes are concentrated on four thin plane-like zones, each of which has a thickness of approximately 100 m. One of the nodal planes of the focal mechanisms agrees with the planar hypocenter distribution. The swarm earthquakes that occurred during the initial stage of the activity exhibited a migration of hypocenters that appears to be represented by the diffusion equation. Based on the spatiotemporal distribution of the earthquakes, the hydraulic diffusivity is estimated to be approximately 0.5–1.0 m2/s. The observations imply that swarm earthquakes were triggered by the diffusion of highly pressured fluid within the fault damage zone. A burst-like occurrence of the swarm earthquakes is also observed in the later stage. These swarm earthquakes are thought to have been triggered primarily by local stress changes caused by the preceding activity. The complicated spatiotemporal pattern is thought to have been caused by the effect of the fluid flow within the high-permeability damage zones as well as the stress perturbations generated by the swarm earthquakes themselves.

We have performed shear wave splitting analyses for seismograms recorded at stations located just above, and outside, the focal area of the earthquake swarm at the Hakone volcano, Japan, in August 2009. Average values of the direction of faster split shear wave polarization (Φ) at two stations above the focal area correspond to each focal alignment of the earthquake swarm. In contrast, average values of Φ at three stations outside the focal area correspond to the direction of the maximum horizontal compressional stress. We found that the average values of the time lag between the two split shear waves inside the focal area are relatively high compared with those outside the focal area. These facts suggest that cracks with a high density aligned parallel to the faults of the earthquake swarm in the focal area. Crustal fluid was selectively injected into this pre-existing cracked media accompanied by effective normal stress reduction in the cracks, resulting in the earthquake swarm.