Introduction

Phreatic eruptions are the result of perturbations of the volcanic hydrothermal system (e.g., Barberi et al. 1992). Recent phreatic eruptions in the Japanese Islands have caused significant human casualties, with 61 fatalities during the September 27, 2014, Mt. Ontake eruption and one fatality during the January 23, 2018, Mt. Kusatsu Shirane eruption. Monitoring and forecasting are often difficult because the signals, such as excitation processes, are often subtle, which have severe implications for human safety (e.g., Kato et al. 2015).

Hakone volcano, located in the central part of Honshu Island, central Japan, recently generated a small phreatic eruption on June 29, 2015, at the Owakudani fumarole area, which is the largest fumarolic area on the volcano (Fig. 1). Although there were no fatalities, the volcanic activity caused serious human safety concerns since the area is an internationally famous sightseeing location in Japan, with many visitors to the fumarole area year-round. Furthermore, the elevated fumarolic activities caused serious damage to some of the steam wells that are used to make hot springs for the surrounding hotels, and these hotels were forced to temporarily stop providing hot springs to their visitors, resulting in significant economic losses to the area. Therefore, although the phreatic eruption was not a large event, its impact raised an important question regarding how intensively the volcano should be monitored to evaluate the risk of an eruption and mitigate the disasters related to the eruption.

Fig. 1
figure 1

Vectors indicating the GNSS movements associated with the 2015 unrest of Hakone volcano (black), and those calculated from the deep inflation source (red). Estimated deep inflation source is shown in blue circle. Two other shallow sources are shown in small squares. The gray dots represent seismicity from March 1 to October 20, 2015. The star shows the location of Owakudani

Full size image

The Hot Springs Research Institute of Kanagawa Prefecture (HSRI) has monitored Hakone volcano since 1989 using a seismometer array to detect seismic activities, GNSS and tiltmeters to detect crustal deformations, and gas sensors to observe fumarolic activities. Here, we use the GNSS data to examine the temporal changes in ground deformation and then attempt to elucidate how volcanic processes at depth, which we model as hydraulic processes, influenced the observed volcanic activity from April to September 2015.

Hakone volcano has a long eruption history, extending more than 400 kyr and including caldera-forming eruptions from 250 to 60 kyr ago (Takada et al. 2007). The most recent magmatic activity took place along a N–S trending line that traverses the center of the volcano, forming several small stratovolcanoes and lava domes. The seismic activity along the line appears to be related to a hydrothermal activity (Yukutake et al. 2011). The seismic activity of the volcano is relatively high, even during its inactive period, and several earthquakes, most of which are M ≤ 1, are observed in a given month.

We have observed major earthquake swarm activities accompanied by the inflation of the volcano since the installation of the GNSS monitoring network in the 1990s (Harada et al. 2010). As shown in Fig. 2, such volcanic unrest has been observed in 2001, 2006, 2008–2009, 2013, and 2015. With the exception of the volcanic unrest in 2001 and 2015, these observed volcanic episodes were only accompanied by earthquake swarms and inflation of the volcano. However, the 2001 unrest was accompanied by intensified steaming activity, and the 2015 event, which is discussed here, culminated in a small phreatic eruption.

Fig. 2
figure 2

Seismic activity and inflation of Hakone volcano from April 1995 to December 2017. The hypocenters were determined by HSRI. a Hypocentral distribution of Hakone volcano. b Spatiotemporal seismicity change. The vertical axis represents latitude. c Magnitude–time diagram and cumulative number of earthquakes. d Baseline length change between the Odawara and Susono-2 GEONET stations. The major volcanic earthquake swarm periods were observed in 2001, 2006, 2008–2009, 2013, and 2015. The elevated seismic activity in 2011 was triggered by the 2011 Tohoku Earthquake (M9.0)

Full size image

The 2015 activity was first recognized by the GNSS network prior to mid-April as an inflation of the volcanic edifice. The earthquake swarm activity began in late April, and the steaming activity intensified in early May in Owakudani (Harada et al. 2015). Although the seismic activity declined since early June, there was a sudden and significant increase on the morning of June 29 that culminated in a small phreatic eruption (Yukutake et al. 2017; Mannen et al. 2018). Here, we estimate the locations of the inflation sources and their inflation rates during the 2015 volcanic activity based on GNSS data analysis.

GNSS data and location of inflation sources

Here, we used GNSS data from 20 observation stations in and around the volcano (Fig. 1). Ten stations are maintained by HSRI, with the other ten GEONET stations installed by the Geospatial Information Authority of Japan (GSI). One of the HSRI stations, located 2 km west of Owakudani, was excluded from the analysis because unrelated landslide movement appears to have contaminated the data from this station. Here, GNSS data from March 1, 2015 to June 30, 2016 were analyzed. The daily coordinates of the HSRI stations and one GEONET station were determined by our routine analysis using Bernese ver. 5.0 software (Dach et al. 2007) with IGS (International GNSS Service) precise ephemerides (Doke et al. 2018b). Furthermore, we used the F3 solution released by GSI to obtain the daily coordinates of the GEONET stations. We needed to combine the two network solutions since the coordinate estimations were conducted independently between HSRI and GSI. We therefore first calculated the offset of the three-dimensional (3D) coordinates at the commonly occupied observation station at Odawara (GEONET station code: 950230) and then subtracted the daily offsets from each of the HSRI sites, relying on the GSI F3 solution. We derived the coordinated time series at each of the sites from the obtained 3D coordinates, with velocity vectors estimated from the difference between the average coordinates for March 1–10, 2015 and the average coordinates for October 11–20, which are shown in Fig. 1.

We assumed three inflation sources in the shallow crust to estimate the potential sources that may have caused the observed crustal deformation associated with the volcanic activity (see Fig. 1). Our preliminary evaluation suggested a single inflation source at ~ 5.4 km depth that reasonably reconstructed the surface dislocation (root-mean-square (RMS) error = 1.2 × 10−3). However, an interferometric synthetic aperture radar (InSAR) analysis implied that a crack opened at shallow depths in the volcano during the phreatic eruption (Doke et al. 2018a). They also implied the existence of a deflation sill at ~ 250 m above sea level, which appears to deflate during the eruption and is assumed to be the source of the hydrothermal fluid that erupted. In this study, we employed the three sources obtained by the following procedure (Table 1; Fig. 3). In our model, we used locations of the open crack and deflation sill deduced by InSAR analysis (Doke et al. 2018a) and then estimated the location of the deep source based on the MaGCAP-V software (Fukui et al. 2013). We used the inverse analysis function of this software by first defining the initial values and deviations of the models for GNSS data at each observation site. We then obtained the optimal solution using a least squares method and weighting factors, where we assigned a weight of 1.0 to the two horizontal components and 0.2 to the vertical component.

Table 1 Parameters of the assumed sources
Full size table
Fig. 3
figure 3

Conceptual diagram of the 2015 unrest of Hakone volcano. (A), (B), and (C) are the inflation sources listed in Table 1. These sources are depicted as larger than the actual sizes for visualization

Full size image

The optimal solution, which yields the best fit location and inflation volume during the 2015 volcanic unrest, is shown in Table 1. We estimated a deep spherical inflation source beneath the central cone at ~ 6.5 km below sea level and a volume change of 7.1 × 106 m3 (RMS error = 2.0 × 10−3). This deep source corresponds to the low seismic Vp/Vs region detected via tomographic analysis (Yukutake et al. 2015). Yukutake et al. (2015) suggested that this deep region contained a developed hydrothermal system.

Calculation of the inflated volume during unrest

We estimated the temporal changes in the inferred volume of each source position during the 2015 volcanic unrest and eruption. The locations of the three sources listed in Table 1 were fixed such that only the volume changes of the three sources were calculated via the MaGCAP-V inversion analyses, with the same weights given to the horizontal and vertical components as in the previous analyses. The initial position of each GNSS station was fixed as the origin, which was determined as the average of the coordinates over the March 1–10 period, about one and a half months prior to the onset of the volcanic unrest. The 10-day average positions of the GNSS stations were calculated for each 3-day shift (e.g., March 4–13 and March 7–16), with the initial position of the station origin subtracted from its 10-day averaged station positions to obtain the displacements from the origin. The resultant displacements at the GNSS stations were then used to estimate the volume changes of the three inflation sources. The best fit volume changes of the inflation sources are shown in Fig. 4. Note that the plots are at the last day of the selected 10-day intervals for monitoring purposes (which means that we cannot take the average using the “future” data).

Fig. 4
figure 4

a Volume changes of the three sources before and after the 2015 volcanic unrest at Hakone volcano (from March 2015 to June 2016). The blue, orange, and green symbols are the total volume changes of the deep spherical inflation, shallow open crack, and sill sources, respectively. Gray plots mean that an estimated error is large (RMS error ≥ 3.0 × 10−3). b Daily number of earthquakes at Hakone volcano during the same period as above

Full size image

Results

The inflation of the deep spherical source, labeled (A) in Fig. 3, appears to have started in late March 2015 (Fig. 4). The inflation rate does not appear to have been stable during the initial phase, with the volume change rates estimated via linear regressions of four intervals during this initial phase. First, there was a rapid increase of 7.9 × 104 m3/day from late March to early April, which then slowed to 2.4 × 104 m3/day in early to mid-April. The deep source underwent another rapid inflation at ~ 9.3 × 104 m3/day from late April to mid-June. The deep source appeared to be relatively stable after mid-June, with another inflation after the eruption, which occurred from June 29 to July 1. This inflation continued until the end of August at ~ 2.1 × 104 m3/day.

The inflation of the shallow open crack source, labeled (B) in Fig. 3, appears to have started gradually in mid-May and continued until early August at an inflation rate of 1.7 × 103 m3/day. It is noteworthy that there was no rate change for this source that was related to the eruption. This shallow open crack source appears to have deflated slowly after mid-September at 0.4 × 103 m3/day. Although fluctuations are still evident in the last stage, after April 2016, these may be due to the unstable solution of the inversion since the RMS error for this period (3.0 × 10−3) is much larger than the other periods. The deflation pattern continued until June 2016, but the volcanic crustal deformation subsequently became obscured by the tectonic movements of the Philippine Sea Plate.

The inflation of the shallow sill source, labeled (C) in Fig. 3, showed no significant contribution to the crustal deformation detected by this GNSS analysis. We consider that the sparse GNSS observation network was unable to detect such a local and small sill in the shallow crust in the Hakone volcano region.

Discussion

Our GNSS analysis detected the inflation of a deep spherical source in late March 2015, well before the onset of the seismic swarm activity in late April. It is interesting to note that a deep low-frequency event (DLF) began at approximately the same time as the inflation (Yukutake and Abe 2018; Mannen et al. 2018). However, the DLF occurred at 15–30 km depth, much deeper than the deep spherical source at 5–6 km depth (see Fig. 3). Yukutake and Abe (2018) hypothesized that the DLF event may be due to an increase in magmatic fluid pressure. This suggests that such fluid may have migrated to shallower depths, potentially generating inflation of the deep spherical source discussed here.

The inflation of the deep source initially ceased by mid-April but was later reactivated. The second inflation in late April appears to coincide with the April 26 onset of the earthquake swarm. This sequence is potentially related to the inflation of a hydrothermal system beneath the volcano due to the injection of hydrothermal fluids separated from the underlying magma chamber. It is expected that this injection increased the pore fluid pressure within the fracture zone and enhanced the shallow seismic activity (Yukutake et al. 2011), coinciding with the shallow inflation that initiated in late May. The continued shallow inflation during the post-eruption period notably persists after the eruption, indicating that the eruption did not ease the pore pressure to stop inflation beneath the volcano.

Here, an open crack, which is indicated by a set of InSAR images before and after the eruption (June 18–July 2 and June 7–July 5 pairs; Doke et al. 2018a), represents the shallow inflation source. As shown above, the shallow inflation began in mid-May. However, the inflation of the open crack prior to June 18 was unclear in the InSAR images. We therefore consider the shallow inflation as an inflation of the fracture system beneath the volcano instead of an open crack inflation.

The inflations in the deep spherical and shallow open crack sources appear to have ceased in late August. Late August may be considered the end of the hydrothermal fluid injection from the deep magma source since there was significantly reduced seismic activity prior to August. The shallow open crack then began to deflate from mid-September. We interpret the crack deflation as the withdrawal of hydrothermal fluid due to closure of the crack or fracture system. The hydrothermal fluid may have been released to the surface gradually from the newly formed fumarole area in Owakudani just above the open crack.

However, there is no evidence of deep source deflation, which is evident from the unchanged baseline length (Fig. 2d). The lack of deflation at depth indicates that the deep spherical source remained inflated. Magma intrusion within the crust (~ 6.5 km below sea level) could also be possible, but there are no other observations to support this hypothesis. We surmise that the lack of deflation implies the accumulation of magma and/or hydrothermal fluid at depth. Since accumulation of hydrothermal fluid and magma could be a preparatory process of larger eruptions in the future (e.g., Murase et al. 2016), volume change monitoring and hydrothermal system modeling in this depth range are critically important in the eruption forecasting of Hakone volcano.

Conclusion

We considered potential inflation sources beneath Hakone volcano at crustal depth and monitored their volume changes over the course of the precursory, eruptive, and post-eruptive phases of the 2015 eruption. The rapid inflation from late March implies the transportation of magmatic fluid into a mid-crustal magma chamber, as inferred from the contemporaneous deep low-frequency seismic events. After a short intermission in April, the deep source resumed to inflate, accompanied by a seismic activity, and continued until August. The inflation that also initiated from mid-May in the shallow source region may be due to the injection of hydrothermal fluid into the hydrothermal system above the magma chamber. This injection continued until August, and the late June phreatic eruption was not enough to mitigate the pore pressure of the hydrothermal system. The fluid injection appeared to cease in August, and the shallow source then began to deflate. However, no significant change was observed in the deep source volume. Therefore, an accumulation of magma and/or fluid may have continued to persist in the deep source area. This accumulation may be a preparatory process for a future eruption, such that intense monitoring and modeling of the magma-hydrothermal system are critically important for forecasting and disaster mitigation measures to better prepare for future eruptions.