Very Local Subsidence Near the Hot Spring Region in Hakone Volcano, Japan, Inferred from InSAR Time Series Analysis of ALOS / PALSAR Data

: Monitoring of surface displacement by satellite-based interferometric synthetic aperture radar (InSAR) analysis is an e ﬀ ective method for detecting land subsidence in areas where routes of leveling measurements are undeveloped, such as mountainous areas. In particular, InSAR-based monitoring around well-developed hot spring resorts, such as those in Japan, is useful for conserving hot spring resources. Hakone Volcano is one of the major hot spring resorts in Japan, and many hot spring wells have been developed in the Owakudani fumarole area, where a small phreatic eruption occurred in 2015. In this study, we performed an InSAR time series analysis using the small baseline subset (SBAS) method and ALOS / PALSAR scenes of the Hakone Volcano to monitor surface displacements around the volcano. The results of the SBAS-InSAR time series analysis show highly localized subsidence to the west of Owakudani from 2006–2011 when the ALOS / PALSAR satellite was operated. The area of subsidence was approximately 500 m in diameter, and the peak rate of subsidence was approximately 30 mm / year. Modeling using a point pressure source suggested that the subsidence was caused by a contraction at approximately 700 m above sea level (about 300 m below the ground surface). The rate of this contraction was estimated to be 1.04 × 10 4 m 3 / year. Hot spring water is collected from a nearby well at almost the same depth as the contraction source, and its main dissolved ion component is chloride ions, suggesting that the hydrothermal ﬂuids are supplied from deep within the volcano. The land subsidence suggests that the fumarole activity is attenuating due to a decrease in the supply of hydrothermal ﬂuids from deeper areas.


Introduction
Land subsidence due to the over-extraction of groundwater needs to be carefully monitored as it can lead to the destruction of buildings and infrastructure in urban areas. In Japan, land subsidence is monitored by leveling [1]. Although leveling measurements can be used to precisely determine vertical displacements, they can only be used to monitor displacements if benchmarks have been installed; if local subsidence occurs in areas away from benchmarks, it may be overlooked. Consequently, conventional leveling measurements are not very effective for detecting land subsidence in mountainous areas where leveling routes are not well developed. There are numerous volcanoes in Japan, many of which have hot springs and are popular tourist destinations. Large-scale resort development has occurred in these mountainous areas, which has meant that extensive drilling for new sources of hot spring water has occurred. Therefore, it is important to monitor land subsidence in mountainous areas with many large spa resorts, not only to maintain the infrastructure, but also to conserve hot spring resources.
Interferometric synthetic aperture radar (InSAR) analysis, using satellite-based sensors, is one of the most effective tools for detecting unknown land subsidence [2]. While InSAR is less precise than conventional leveling measurements, the technique does not require a network of ground-based observations. In addition, InSAR analysis is well suited for application in mountainous areas, where there is no existing observation network. SAR satellites were launched by governments and private companies, primarily to monitor the natural and physical characteristics of the Earth, including vegetation, natural resources, and crustal deformation for disaster monitoring. Various wavelengths are used for land surface observations, ranging from the L-band to the X-band, depending on specific requirements. InSAR analysis using short-wavelength sensors (e.g., X-band and C-band) generally does not provide good coherence in forested areas, which makes it challenging to monitor ground movement. Conversely, InSAR analyses using long-wavelength (L-band) sensors can achieve good coherence, even in forested areas, enabling users to monitor ground deformation. It is therefore considered that monitoring by L-band sensors will play an important role in clarifying ground deformation in mountainous areas in places such as forested area of Japan.
The Hakone Volcano (Figure 1), located in the plate collision zone of the Izu Peninsula with Japan mainland [3], has been active for approximately 400,000 years and is one of several active volcanoes in Japan. Hakone is a caldera volcano with a complex of stratovolcanoes of basaltic to andesitic composition forming the caldera rim (≥230 ka) ( Figure 2) [4]. Within the last 37,000 years, andesitic eruptive activity has formed a central cone of the volcano [4]. The latest magmatic eruption occurred about 3000 years ago on the northern slope of the central cone and caused a sector collapse that formed an amphitheater [4]. Owakudani, the largest fumarole area of the volcano, is located at the eastern margin of the amphitheater (Figure 3). In the Owakudani, hot spring water is artificially produced from the self-discharged volcanic steam that comes up from borehole (steam wells), which are several hundred meters in depth. Total heat flux of natural fumaroles and steam wells added together was estimated to be approximately 15 to 20 MW from 1983 to 2012, and it dropped in recent years [5]. The recent activity of the volcano started around the year 2000, and a small-scale phreatic eruption occurred in Owakudani in 2015 [5]. Hakone Volcano is located approximately 100 km west of Tokyo (Figure 1), the capital of Japan. The region surrounding the volcano has a well-developed infrastructure, comprising railways, roads, hotels, and other recreational amenities. It is a major hot-spring tourism destination in Japan that is visited by approximately 20 million tourists a year.   Based on seismic activity and InSAR analyses, the occurrence of the phreatic eruptions of Hakone Volcano and its associated seismic swarm activities are thought to be related to hydrothermal activity in the shallow parts of the volcano [7,8]. It is considered that this activity is closely related to the formation of hot spring water [9]. It is therefore considered important to clarify the reasons for ground deformation in this area, not only for preventing a volcanic disaster, but also to better understand the formation mechanism of hot springs in Hakone Volcano and the conservation of hot spring resources.
In this study, we performed an InSAR time series analysis using the small baseline subset (SBAS) method [10], and data obtained from the Phased Array type L-band Synthetic Aperture Radar (PALSAR) aboard the Advanced Land Observing Satellite (ALOS). We detected highly localized subsidences in the fumarolic area, which is the principal hot spring area of Hakone.

Materials and Methods
We used data from the ALOS/PALSAR satellite launched by the Japan Aerospace Exploration Agency (JAXA), which has a wavelength of 23.6 cm (L-band). The satellite always observes the right side of the azimuth direction, and its revisit interval is 46 days [11]. ALOS/PALSAR was operated from 2006 to 2011, and its successor, ALOS-2/PALSAR-2, has been in operation since 2014. In this study, we obtained 24 scenes from Path 407-Frame 690 observed from the ascending orbit (observation from the western sky) and 22 scenes from Path 59-Frame 2910 observed from the descending orbit (observation from the eastern sky) which covers Hakone Volcano ( Figure 1 and Table 1). The off-nadir angles for the scenes of both the ascending and descending orbits were 34.3 • . The ALOS-2/PALSAR-2 data were not used in this study because the data acquisition interval was long, and the data available was insufficient for the analysis. The ENVI+SARscape software program (Sarmap Sa, Caslano, Switzerland) was used for the SBAS-InSAR time series analysis in this study. Given that coherence is generally worse when the perpendicular baselines (Bperp) between ALOS/PALSAR orbits are more than 1000-1500 m, the analysis was conducted by extracting the interference pairs whose Bperp was less than approximately 1000 m. The pairs with weak interference were then removed visually, and 21 scenes-64 pairs in the ascending orbit and 21 scenes-57 pairs in the descending orbit were used for the analysis ( Figure 4). Three of the scenes in the ascending orbit were excluded because of the large Bperp between the other scenes ( Table 1). One of the scenes in the descending orbit was excluded because it was observed after the Tohoku-Oki Earthquake (M9.0) of 11 March 2011, and was affected by the earthquake displacement (Table 1).
To improve the signal-to-noise ratio, we averaged over approximately 15 × 15 m for each pixel in the azimuth and range directions. To remove the influence of topography in the initial interferogram, we used 10 m mesh elevation data obtained from the Geospatial Information Authority of Japan (GSI) and the geoid model of the Earth Gravitational Model 2008 [12]. The adaptive interferogram filtering algorithm [13] was applied to reduce the noise, and interferograms were unwrapped using the minimum cost flow approach [14]. During the time series inversion analysis, we used the global navigation satellite system (GNSS) of the GSI for establishing ground control points (GCPs) (Figure 1). The InSAR displacements were corrected with the displacements of the GCPs. For the GCPs, we used daily coordinates of GNSS sites published by the GSI and filtered these coordinate values in a 10-day time window. The velocity distributions obtained from the InSAR time series analysis were converted into a geographic coordinate system with an approximately 25 × 25 m mesh. Then, the velocity maps were obtained by masking pixels with a height estimation error of more than 5 m or a velocity estimation error of more than 8 mm/year. For the descending orbit, the data include Mount Fuji, which is the highest mountain in Japan (3776 m) and forms an independent peak with no continuity with the surrounding mountains. Including Mount Fuji in the scene causes a large discrepancy between the displacements of GCPs and pixels in the InSAR time series analysis. However, when the western part of the scene, including Mount Fuji, was cropped, the displacements estimated by InSAR and the GCPs almost agreed. Therefore, we employed the result that excluded Mount Fuji.

2.5-D Analysis
The LOS velocities obtained for ascending and descending orbits were transformed to the velocities of the quasi-eastward and quasi-upward components by applying 2.5-D analysis [15]. Quasi-eastward and quasi-upward velocities (V QE , V QU ) are estimated from LOS velocity in ascending and descending orbits (V A , V D ), and these unit vectors ( A line vector where the LOS plane and horizontal plane intersect is represented as → L (y, −x, 0), and its magnitude is as follows: Angles (θ A , θ D ) between → L and LOS vectors ( are estimated as follows: Quasi-eastward and quasi-upward velocities (V QE V QU ) are estimated as follows: The strike and dip of the LOS plane varied slightly from point to point, but these were approximately 271 • and 85 • , respectively.
The quasi-eastward velocity shows that almost the entire area was displaced westward (Figure 7a). This westward displacement is considered to be a regional tectonic displacement caused by the Pacific Plate, which is subducting beneath the northeastern Japan [16,17]. On the other hand, the areas near Mount Fuji and Hakone Volcano showed eastward velocities (Figure 7a,b). Since expansions of volcanic edifices were observed by GNSS around Mount Fuji and Hakone Volcano during the study period [18,19], the eastward velocities might be consistent with the GNSS displacements. The quasi-upward velocity findings showed that the areas around Mount Fuji, Hakone Volcano, and the Izu Peninsula were uplifted, and the Ashigara Plain, located to the east of Hakone Volcano, subsided slightly (Figure 8a). These displacements are thought to be caused by tectonic deformations by the Izu Peninsula colliding with the main part of the Japanese Islands [3] and volcanic activities [18,19]. To the west of Owakudani, where volcanic rock debris-avalanche deposits are distributed on the surface (Figure 2), a maximum subsidence rate of about 30 mm/year was observed within an area with a diameter of about 500 m (Figure 8b). The displacement appears to have progressed at a constant rate over the observation period ( Figure 9). Moreover, regarding the quasi-eastward component, both sides of the center of the subsidence area were displaced inwards; that is, the western side was displaced eastward, and the eastern side was displaced westward (Figure 7b). This result indicates that contractional displacement was dominant to the west of Owakudani.   Figures 5b and 6b) to remove the displacement around the Hakone Volcano.

Modeling
Since the contractional displacements are dominant, we used a model involving a point pressure source in an elastic half-space body [20] to explain the subsidence to the west of Owakudani. A modeling tool in ENVI+SARscape, which implements inversion of the Levenberg-Marquardt least-squares approach [21], was used to estimate the volume variation and 3-D position of the pressure source. During the inversion analysis, the influence of topography was compensated for by the elevation data. Moreover, the offset parameters of the InSAR images were also estimated to consider the LOS velocity around the Hakone Volcano. After the optimum values of the model parameters were obtained by the inversion, the variations in the residual root mean square (RMS) values were obtained by forward modeling with varying parameters (i.e., altitude and volume variation) to validate the sensitivities of each parameter.
The modeling results show that the subsidence can be explained by a point pressure source contracting at a rate of 1.04 × 10 4 m 3 /year at the height of about 700 m above sea level (about 300 m below the ground surface) just under the area of subsidence (Table 2 and Figure 10). Based on the residual RMS values (Figure 11) estimated by the forward modeling, the parameters of the model are considered well-constrained, especially with respect to altitude. For example, a residual RMS value of 4 mm or less would be achieved at an altitude of 620-750 m and a volume variation of −1.4 × 10 4 to −6.0 × 10 3 m 3 /year. However, it may be inappropriate to apply a point pressure source model, because the estimated model has a very shallow underground distribution, and it may not be deformed as an elastic body. Modeling that takes into account the physical properties of the ground is a future issue. At a minimum, the modeling results suggest that the cause of subsidence may be due to very shallow contraction, possibly several hundred meters below the ground surface.  Table 2. Estimated offsets for the ascending and descending orbits are +0.15 and +2.43 mm/year, respectively. The color scales used for the velocity are the same as those in Figures 5 and 6.   (Table 2).

Discussion
Hot spring wells have been developed in the Owakudani region for decades, and many of these wells are located to the east of the subsidence area. Most of the hot spring wells in Owakudani are steam wells; hot water for bathing is artificially produced by adding local groundwater to the self-discharged volcanic steam that comes up from boreholes, which are several hundred meters in depth. The hot water produced by steam wells in Owakudani is acidic with high concentrations of chloride ions, which are derived from a magma activity of about 10 km in depth [22,23].
On the other hand, in the western region of Owakudani, most of the hot spring water is drawn from the wells. These hot spring waters contain primarily sulfate and bicarbonate, which are naturally produced by volcanic gas added to groundwater distributed near the surface, except for well MO39 (Figures 12 and 13). The hot spring well MO39, which is closest to the area of subsidence, is a steam well that produces acidic spring water from self-discharged volcanic steam with chloride ions as its main dissolved chemical component ( Figure 13).   Figure 12 shows the depth and characteristics of the hot spring wells to the west of Owakudani. To prevent cold water from mixing with the hot spring water, casing pipes are typically installed from the top of the well to a certain depth in the wells. The casing pipes allow for the collection of the hot spring water or steam directly from the depth of the uncased part of the borehole, where the hot spring aquifers are exposed. Based on the structure of the wells to the west of Owakudani, hot spring water, or steam is collected from around 600-800 m above sea level ( Figure 12). These data indicate that the hot spring water or steam in this region is distributed at a depth of several hundred meters below the ground surface. Interestingly, the contraction source estimated by this study is also located at a depth of several hundred meters, which is almost coincident with the hot spring source. Therefore, the subsidence observed in the area is considered to be due to the contraction of the reservoir from which the hydrothermal fluids are being supplied from below.
So, what has caused this reservoir contraction? The Owakudani fumaroles, which is located a few hundred meters east of the area of subsidence, are more active than the fumaroles to the west of Owakudani ( Figure 14). In Owakudani, an area 200 m in diameter was locally uplifted during volcanic unrest in 2015, and a phreatic eruption occurred in the vicinity of the uplifted area [7,24]. This local uplifting was explained by expansion sources approximately 80-150 m below the ground surface [7,24]. This expansion suggests the existence of a reservoir that supplies hydrothermal fluids and causes the fumarolic activity on the surface of Owakudani. The hydrothermal fluids with highly concentrated chloride ions are considered to be directly supplied to this reservoir through cracks from the deeper parts of the volcano [7]. Conversely, there are no such active fumaroles to the west of Owakudani, where the subsidence was observed (Figure 14), although there are small fumarolic activities that make a discharge. However, in 1872, the French engineer Jean-Paul Ishidor Vidal is on record as having measured the temperature of the steam in the western area of Owakudani as 103 • C [5,25,26]. This implies that the area of subsidence was once a fumarole zone with heated steam. Moreover, chloride ion that is contained as a major anion in hot spring water of MO39 (Figure 13), which is the closest to the area of subsidence, is implying the participation of the hydrothermal fluids supplied directly from the deeper parts of the volcano [22]. This suggests that the observed subsidence is associated with attenuation of fumarole activity in response to a decrease in the supply of hydrothermal fluids from below. It is also possible that this attenuation may have arisen due to closure of the fluid path from deeper regions, such as a crack. Another possible cause of subsidence may be excessive pumping of hot spring water. However, MO39, the well closest to the subsidence area, is a steam well and uses self-discharged steam, not artificially pumping hot spring water. It is not considered that forced dissipation of pore water pressure, such as land subsidence resulting from excessive groundwater pumping, has occurred. To accurately clarify these factors, it will be necessary to continuously monitor the discharge volume and temperature of the hot spring water.
In this study, we used SAR data acquired by the ALOS/PALSAR satellite from 2006 to 2011. After the termination of ALOS/PALSAR in 2011, the ALOS-2/PALSAR-2 satellite was launched. We consider it essential to continue monitoring the surface displacement in the area to the west of Owakudani to clarify whether the subsidence is continuing or not; this will form part of future studies.

Conclusions
Based on an SBAS-InSAR time series analysis, we observed highly localized subsidence in an area to the west of Owakudani in the Hakone Volcano, Japan, from 2006 to 2011 using ALOS/PALSAR data. The area of subsidence was approximately 500 m in diameter, and the peak rate of subsidence was approximately 30 mm/year. The results of point pressure source modeling suggested that the cause of the observed subsidence was due to reservoir contraction at approximately 700 m above sea level (about 300 m below the ground surface). The rate of contraction was estimated to be 1.04 × 10 4 m 3 /year.
Based on the structure of the hot spring wells and their chemical components, the contraction source is considered to be a reservoir containing hydrothermal fluids. Moreover, the subsidence suggests that the attenuation of fumarole activity is occurring due to a decrease in hydrothermal fluid supply.
In addition, the findings of this study demonstrated the application of InSAR surface displacement measurements to monitor hydrothermal activity in shallow parts of volcanic areas. Monitoring surface displacement is considered useful, not only for evaluating the potential occurrence of phreatic eruptions but also for clarifying the production mechanism of hot spring water in volcanic regions.