Temporal changes in electrical resistivity at Sakurajima volcano from continuous magnetotelluric observations

Journal of Volcanology and Geothermal Research 199 (2011) 165–175

Contents lists available at ScienceDirect

Journal of Volcanology and Geothermal Research j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / j vo l g e o r e s

Temporal changes in electrical resistivity at Sakurajima volcano from continuous magnetotelluric observations Koki Aizawa a,d,⁎, Wataru Kanda a,b, Yasuo Ogawa b, Masato Iguchi a, Akihiko Yokoo a,e, Hiroshi Yakiwara c, Takayuki Sugano d a

Sakurajima Volcano Research Center, Kyoto University, Yokoyama 1722-19, Sakurajima, Kagoshima 891-1419, Japan Volcanic Fluid Research Center, Tokyo Institute of Technology, Ookayama 2-12-2, Meguro-ku, Tokyo 152-8551, Japan Nansei-Toko Observatory for Earthquakes and Volcanoes, Faculty of Science, Kagoshima University, 10861 Yoshino-cho, Kagoshima 892-0871, Japan d Earthquake Research Institute, University of Tokyo, Yayoi 1-1-1, Bunkyo-ku, Tokyo 113-0032, Japan e Department of Geophysics, Graduate School of Science, Tohoku University, Aramaki-Aza Aoba 6-3, Aoba-ku, Sendai, Miyagi 980-8578, Japan b c

a r t i c l e

i n f o

Article history: Received 21 June 2010 Accepted 1 November 2010 Available online 10 November 2010 Keywords: magnetotellurics resistivity volatile degassing hydrothermal system

a b s t r a c t Continuous magnetotelluric (MT) measurements were conducted from May 2008 to July 2009 at Sakurajima, one of the most active volcanoes in Japan. Two observation sites were established at locations 3.3 km east and 3 km west–northwest of the summit crater. At both observation sites, the high-quality component of the impedance tensor (Zyx) showed variations in apparent resistivity of approximately ± 20% and phase change of ± 2°, which continued for 20–180 days in the frequency range between 320 and 4 Hz. The start of the period of changes in apparent resistivity approximately coincided with the start of uplift in the direction of the summit crater, as observed by a tiltmeter, which is one of the most reliable pieces of equipment with which to detect magma intrusion beneath a volcano. A 2D inversion of MT impedance suggests that the resistivity change occurred at a depth around sea level. One of the possible implications of the present finding is that the degassed volatiles migrated not only vertically through the conduit but also laterally through a fracture network, mixing with shallow groundwater beneath sea level and thereby causing the observed resistivity change. © 2010 Elsevier B.V. All rights reserved.

1. Introduction The monitoring of subsurface magma is an essential approach in terms of predicting volcanic eruptions and contributing to hazard mitigation. Daily imaging of the location, volume, and physical properties (e.g., pressure and gas fraction) of subsurface magma enables predictions not only of eruption timing, but also its location, duration, and degree of explosivity. Geodetic measurements (strain, tilting, and GPS) are currently the most practical methods with which to investigate changes in subsurface magma, because such data are sensitive to subtle pressure changes and have high temporal resolution. For example, at Sakurajima volcano, Japan, Vulcanian-type eruptions are routinely predicted in advance by up to 1 day based on data from a strainmeter and tiltmeter installed in a tunnel at the volcano (e.g., Ishihara, 1990; Iguchi et al., 2008a,b). However, it is generally difficult to predict an

⁎ Corresponding author. Earthquake Research Institute, University of Tokyo, Yayoi 1-1-1, Bunkyo-ku, Tokyo 113-0032, Japan. Tel.: +81 3 5841 5746; fax: +81 3 3812 6979. E-mail addresses: [email protected] (K. Aizawa), [email protected] (W. Kanda), [email protected] (Y. Ogawa), [email protected] (M. Iguchi), [email protected] (A. Yokoo), [email protected] (H. Yakiwara), [email protected] (T. Sugano). 0377-0273/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.jvolgeores.2010.11.003

eruption over the coming weeks or months. In addition, some eruptions occur without significant ground deformation. It is a promising procedure to monitor changes in subsurface structure as an indicator of changes in subsurface magma. Previous studies have used seismic methods to investigate changes in structure beneath active volcanoes and geothermal areas (e.g., Foulger et al., 1997; Nishimura et al., 2000; Miller and Savage, 2001; Foulger et al., 2003; Gerst and Savage, 2004; Yamawaki et al., 2004; Nishimura et al., 2006). A recent study of temporal change in seismic structure (4D tomography) beneath Etna volcano, Italy, clearly imaged change in the structure of Vp/Vs ratio (Patanè et al., 2006). The authors attributed the change in Vp/Vs to subsurface magma movement and corresponding degassing of volatiles. Another approach involves using seismic noise records to monitor seismic structure, as reported by Brenguier et al. (2007, 2008). Based on the premise that long-term averaging of seismic noise produces a random source field, the authors imaged changes in seismic velocity at Piton de la Fournaise volcano, Reunion Island. These recently developed seismic methods have given rise to the possibility of monitoring magma movement and predicting eruptions, although such approaches require a dense seismometer network. The monitoring of electrical resistivity structure also shows promise in terms of imaging subsurface magma, because magma

166

K. Aizawa et al. / Journal of Volcanology and Geothermal Research 199 (2011) 165–175

(melt) and degassed volatiles (if mixed with groundwater) are highly conductive. Indeed, repeated DC electrical measurements revealed a significant resistivity change coincident with the 1986 eruption of Izu-Oshima volcano, Japan, from which the movement of subsurface magma was inferred (Yukutake et al., 1990; Utada, 2003). At Miyakejima volcano, Japan, significant resistivity change was observed before the formation of a crater in 2000, interpreted to represent disturbance of the hydrothermal system (Zlotnicki et al., 2003). Because these pioneering studies used DC electrical measurements with a few receivers, the temporal resolution (several weeks to months) or the spatial resolution (hundreds of meter) beneath the observation line are relatively coarse. In the present study, we provide the first results of long-term (over 1 year), continuous magnetotellurics (MT) observations of an active volcano using the natural electromagnetic source field. Because MT impedance is stable and has high temporal resolution (Eisel and Egbert, 2001; Hanekop and Simpson, 2006; Kappler et al., 2010), the MT technique is suitable for monitoring subsurface resistivity structure. 2. Observations Continuous MT observations were carried out at Sakurajima volcano, which is one of the most active volcanoes in Japan. Sakurajima is located in southern Kyushu, and its summit is ~ 1100 m above sea level. The volcano is located in the southern part of the 20 × 20 km Aira caldera (Fig. 1), which emitted 100 km3 of eruption products during an event 22,000 years ago (Aramaki, 1984). Since 1955, eruptions have occurred mainly at the summit crater (Minami-dake). Volcanic activity is characterized by Vulcanian

eruptions (~ 8000 explosions in the past 50 years), effusive eruptions, and continuous ash emissions. Showa crater, which is located 500 m east of the Minami-dake summit crater, started to erupt in June 2006 after lying dormant for 58 years. Activity at Showa crater gradually replaced that at Minami-dake, with eruptions becoming increasingly vigorous: 548 explosions were recorded in 2009. The magma volume presently emitted from Sakurajima is in the order of 0.001 km3/year (Iguchi et al., 2008a,b). In 2008, we established two continuous MT observation sites at locations 3.3 km east of Showa crater (KURMT) and 3 km west– northwest of the crater (HARMT) (Fig. 1). This paper is the first report of long-term continuous MT measurements performed with the aim of investigating the relationship between volcanic activity and resistivity change, although MT observations were previously carried out for 6 weeks at Ruapehu volcano, New Zealand, to investigate the relationship between earthquakes and resistivity change (Hanekop and Simpson, 2006). Naturally occurring geomagnetic fields and induced earth electric currents in the frequency range of 320– 0.0005 Hz were measured using Phoenix MTU5 systems. Sampling frequencies were 2400 Hz (1 s every 4 min), 150 Hz (16 s every 4 min), and 15 Hz (continuous), from which high-frequency variations were removed using an anti-aliasing filter. The electrodes were buried at 30 cm depth at spacings of ~ 20 m; the contact resistance between electrode and soils was maintained below 3000 Ω. The observation sites were visited approximately weekly for maintenance of the apparatus (e.g., changing batteries, sweeping away volcanic ash, and checking that cables had not been disconnected as a result of animal activity) and for collecting the time series data stored on a 4 GB CompactFlash card. The MT time series during eruptions contain

Fig. 1. Locations of sites of continuous MT measurements at Sakurajima volcano (squares) and sites of other geophysical/geochemical observations (crosses; see Section 5).

K. Aizawa et al. / Journal of Volcanology and Geothermal Research 199 (2011) 165–175

impedance at a higher time resolution than that of the error bars. Next, we employed remote-reference processing (Gamble et al., 1979) using geomagnetic data recorded at other sites. For KURMT, we employed geomagnetic data recorded at HARMT. For HARMT, we employed geomagnetic data recorded at the ESA observatory of the Geographical Survey Institute of Japan (located 1500 km northeast of Sakurajima volcano), because the data quality at KURMT is always low relative to HARMT due to the artificial noise. In both cases, noisy hours were carefully removed by visual inspection of the data. Note that single-site processing produced similar results (sounding curves) to that of remote-reference processing for frequencies between 320 and 4 Hz. For frequencies below 4 Hz, remote-reference processing resulted in a significant improvement in the quality of impedance, but the errors were still too large to enable an analysis of temporal change. Later, we discuss the impedance between 320 and 4 Hz (audio-frequency band) as obtained by remote-reference processing. Fig. 2 shows the example of sounding curves (days of good and poor data quality) by remote-reference processing. In general, the data quality of HARMT is better than that of KURMT. In both sites, the quality of Zyx is better than Zxy. The data quality of the estimated impedance was also checked by E-predicted coherency, which is the coherency between the predicted E-field (calculated from impedance

frequent pulses generated by volcanic lightning, and the relationship of these pulses to the style of eruption has been reported previously (Aizawa et al., 2010). 3. Analysis and data quality At Sakurajima volcano, the occurrence of electrical current leakage from railways around Kagoshima City means that noise removal is important in detecting resistivity change. We first calculated the daily impedance tensor by single-site processing. The impedance tensor Zij is a complex number defined in the frequency domain as 

Ex ð f Þ Ey ð f Þ



 =

Zxx ð f Þ Zyx ð f Þ

Zxy ð f Þ Zyy ð f Þ



 Bx ð f Þ ; By ð f Þ

ð1Þ

where Ex, Bx and Ey, By are the electric and geomagnetic fields in the north–south and east–west directions, respectively. The Fourier transform of the electric and magnetic fields was obtained using a cascade decimation algorithm (Wight and Bostick, 1980), for which the time interval between estimates was twice that used in the nexthigher octave. Time series data were processed daily because processing every 2 h does not produce the temporal variation of

Phase (Degree)

Ap. Res. (Log ohm-m)

HARMT

KURMT

4

4

3

3

2

2

1

1

0

0

90 75 60 45 30 15 0

90 75 60 45 30 15 0

3

2

1

0

-1

-2

-3

4 May 2008 (Day of excellent quality)

3

2

Log frequency (Hz)

Ap. Res. (Log ohm-m)

167

1

0

-1

-2

-3

Zxy Zyx

Log frequency (Hz)

4

4

3

3

2

2

1

1

5 May 2008

Phase (Degree)

0 90 75 60 45 30 15 0

3

2

1

0

-1

Log frequency (Hz)

-2

-3

90 75 60 45 30 15 0

(Day of poor quality)

3

2

1

0

-1

-2

-3

Log frequency (Hz)

Fig. 2. Examples of MT sounding curves. Open and solid circles indicate Zxy and Zyx, respectively. Error bars represent ± two standard deviations. Note that this study considers temporal variations in impedance between 320 and 4 Hz.

168

K. Aizawa et al. / Journal of Volcanology and Geothermal Research 199 (2011) 165–175

and the measured geomagnetic field) and the measured E-field. For example, Ey-predicted coherency is calculated as CohEypred =

〈j〈Ey Eypred j〉 b N : crossspectrum 〈Ey Ey 〉〈Eypred Eypred 〉

where Ey pred = Zyx ⋅Bx + Zyy ⋅By

ð2Þ ð3Þ

Fig. 3 shows temporal change in E-predicted coherency. Eypredicted coherency exceeds 0.9 at both sites, while Ex-predicted coherency is 0.6–0.9 at KURMT, suggesting the low quality of Zxx and Zxy. This low quality probably reflects artificial noise produced by power lines located near KURMT. Indeed, noise spikes are regularly seen in the Ex data at KURMT. Below, we use Zyx data from both MT sites to compare impedance at the two sites. The difference between Zyx and the other three components at HARMT is briefly discussed in Section 4.4. Because of the large errors, the tipper (transfer function between the vertical and horizontal geomagnetic fields) was not used in this study. 4. Features of the data 4.1. Temporal change in the high-quality component (Zyx) of the impedance tensor Fig. 4 shows temporal variations in apparent resistivity and phase calculated from Zyx, using data for the following frequencies: 320, 229, 159, 97, 40, 22.5, 13.7, 8.1, and 4.1 Hz. Apparent resistivity and phase in Zyx are expressed as follows, ρa ð f Þ =

1 2 jZ ð f Þj 2πf yx

  ϕð f Þ = arg Zyx ð f Þ

ð4Þ ð5Þ

where f is frequency, ρa(f) is the apparent resistivity, and ϕ(f) is the phase. The lowest frequency (4.1 Hz) corresponds to 0.8 and 2.5 km of skin depth, assuming uniform 10 and 100 Ω m half spaces, respectively. Fig. 4 shows that the changes in apparent resistivity and phase are relatively small at both sites, suggesting that the overall resistivity structure remained stationary during the observation period. For both sites, high resistivity gradually changes to low resistivity with

decreasing frequency, indicating that the deeper part is more conductive than the shallower part. This finding is typical of the impedance in volcanic areas, and can be interpreted as an upper, water-poor resistive surface layer (e.g., lava and unsaturated zone) and a lower water-rich region. However, it is generally difficult to deduce the hydraulic structure based solely on resistivity structure. In particular, the interpretation of a conductive region is speculative because water and hydrothermally altered clay minerals (e.g., smectite) are both conductive (e.g., see the discussion in Aizawa et al., 2009). At Sakurajima, boreholes have been drilled within 300 m of the MT observation sites. We confirmed that the vertical iron casing in the boreholes does not significantly distort the MT impedance at both site by using the 3D forward code (Mackie et al., 1994). Fig. 5 shows the temperature–depth profile obtained by logging these boreholes (Sumiko Consultants Co., 1985, 1990), and the observed location of water table. The logs show that the water table occurs at around the depth of sea level, below which lies hot spring water (~30–38 °C). At the KURMT site, a much colder area (~20 °C) occurs at elevations of −300 to −100 m, interpreted to represent invading seawater, given the proximity of the shoreline. The low apparent resistivities of ~0.5 (log Ω m) at KURMT (i.e., the lowest frequency in Fig. 4) are interpreted to represent the seepage of highly conductive seawater. Because the maximum temperature is relatively low, it is unlikely that the deeper conductive zone represents the presence of melt. Fig. 6 presents the relative temporal changes in Zyx, showing the deviations from arithmetic mean values for each frequency in Fig. 4, throughout the observation period. There exist variations in apparent resistivity of approximately ±20% and variations in phase of ± 2°; the changes continue for 20–180 days. Temporal changes in apparent resistivity show a negative correlation between the two sites. Continuous MT observations (Ex, Ey, Hx, Hy, and Hz) were also performed by Nittetsu Mining Corporation at a site located 30 km north of Sakurajima, as a remote-reference site for many MT surveys conducted in Japan. We analyzed these data by single-site processing, revealing a lack of significant changes in apparent resistivity. Therefore, we reasonably assume that significant impedance (apparent resistivity and phase) change is localized at Sakurajima. 4.2. Time lag between the two sites To enable a comparison of temporal changes at the two sites, we extracted the apparent resistivities at frequencies of 22.5 Hz at

Fig. 3. Temporal changes in E-predicted coherency. Ex and Ey correspond to the electric field in the N–S and E–W directions, respectively. Small ‘plus’ symbols indicate the frequency used in Figs. 6–8. White contour indicate the coherency of 0.9.

K. Aizawa et al. / Journal of Volcanology and Geothermal Research 199 (2011) 165–175

169

Fig. 4. Temporal changes in apparent resistivity and phase (Zyx).

HARMT and 97 Hz at KURMT (Fig. 7a). There occur smooth temporal changes and a negative correlation between the two sites. The reason we extract the different frequency is that temporal change is relatively insignificant in the lower frequencies at KURMT (see Fig. 6). Although the temporal changes recorded at the two sites appear to occur simultaneously, there exists a slight time lag, especially in the case of a negative correlation (days 0–190 and 330–430). Roughly speaking, the temporal change at KURMT appears to slightly precede that at HARMT (see around days 20 and 100). In contrast, at around day 280, temporal changes at KURMT appear to lag behind those at HARMT. To quantify these time lags, cross-correlation between the two time series RXY(τ) was calculated as a function of the time lag, using the following equations: CXY ðτÞ =

N 1   ∑ ðXiτ − XÞ⋅ðYi − YÞ; N i=τ + 1

ð6Þ

RXY ðτÞ =

CXY ðτÞ ; CXX ð0Þ⋅CYY ð0Þ

ð7Þ

¯ , and τ represent the time series, its average value, and the whereXi, X time lag, respectively. We calculated the cross-correlations not only

HARMT (Elevation = 407 m)

for the frequencies shown in Fig. 7a, but also for all combination of frequencies. Fig. 7b shows the cross-correlations during the period except around day 280 (days 0–190, and 330–430), showing that the negative correlation attains a maximum, with an approximately 7-day delay of HARMT behind KURMT. In contrast, around day 280, temporal change at HARMT appears to precede that at KURMT. The implications of these time lags are briefly discussed in Section 6. 4.3. 80-day variation in the HARMT record The HARMT record appeared to show an 80-day variation. The timings of the lowest resistivity correspond approximately to 20 June, 10 September, and 1 November 2008, and 24 February 2009, being similar to the solstice and equinox periods. This 80-day variation became vague in the summer of 2009. 4.4. Difference between Zxy and the other three components In this study, we used only the Zyx component because of its high quality. Fig. 8 shows the apparent resistivities derived from the four components of the impedance tensor for HARMT in the same frequency. The temporal variations in the four components are basically similar, supporting the validity of using Zyx as being representative of variations in resistivity. However, there is a notable difference between Zxy and Zyx in the period between days 0 and 90.

400

Elevation (m)

5. Comparison of resistivity data with tides, rain, ground deformation, eruptions, and earthquakes 200

KURMT (Elevation = 66 m)

0

0

-200

-200

20

40

20

40

Temperature (°C) Fig. 5. Temperature data recorded at boreholes located close to the MT sites. Dashed horizontal lines indicate the observed water table.

To investigate the mechanism of resistivity change, we compared the time series of apparent resistivity with various geophysical data. Fig. 9 shows precipitation, oceanic tides, apparent resistivity (Zyx), ground deformation, eruptions with major infrasound, and seismicity around the volcano. Given that tides and rainfall are able to induce subsurface resistivity change (e.g., Utada et al., 1998), these effects on resistivity should be evaluated first to investigate the mechanism of resistivity change. Fig. 9 shows no significant relationship between temporal change in apparent resistivity and tides. On the other hand, intense rain appears to induce a short-term temporal change in apparent resistivity at HARMT (see the periods around days 45 and 150). However, the long-term temporal change in apparent resistivity cannot be explained solely by rain (at HARMT, rain apparently causes a resistivity decrease between days 25 and 60, but causes a resistivity

170

K. Aizawa et al. / Journal of Volcanology and Geothermal Research 199 (2011) 165–175

Fig. 6. Temporal changes in apparent resistivity and phase (Zyx). Relative change is normalized by the average values for each frequency.

increase between days 150 and 180), thereby indicating that an alternative mechanism of resistivity change is required. The relationship between apparent resistivity and ground deformation is significant. Although the time lag noted in Section 4.2 is apparent in Fig. 9, apparent resistivity starts to change at approximately the time when the tunnel tiltmeter starts to show uplift in the direction of the summit area (marked by the vertical dashed line in Fig. 9). Ground tilt causes the induction coils to tilt, resulting in a change in the impedance. However, the degree of impedance change expected from tilting of the induction coils (10− 6 rad) is smaller than that observed by approximately 10− 5. Deep low-frequency earthquakes (DLFs) occur mainly beneath the southern part of the volcano within the lower crust (Yakiwara and Goto, 1999). Fig. 9 shows that the relatively large DLFs (M N 1.0) appear to occur at convex and concave peaks in resistivity change. Intense shallow seismicity may have induced the observed resistivity change because fracture development can lead to change in the connectivity of fluid in the rock. At Sakurajima, BL-type earthquakes, whose dominant frequency is around 1 Hz (Iguchi, 1994), occur in or around the vertically elongated cylindrical zone

Log ohm-m Log ohm-m

2.50 2.45

It is possible that changes in groundwater level are related to changes in apparent resistivity. However, a numerical forward calculation using a

(b)

Days since 1 May 2008 0

2.55

6. Interpretation of resistivity change

30

60

90

120

150

180

210

240

270

300

330

360

390

420

10 July 2009 HARMT 22.5 Hz

2.40 2.35 1.40 1.35 1.30 1.25 1.20

KURMT 97 Hz

0.2

Cross – correlation

(a)

(which is probably related to the magma conduit) beneath the summit crater at a depth of 0–2 km below sea level. A-type (volcanotectonic) earthquakes (Iguchi, 1994) occur around the conduit at a depth of 0–3 km below sea level; however, these shallow earthquakes show no clear relationship with changes in apparent resistivity. Fig. 9 also shows no clear relationship between eruption events and apparent resistivity at a higher-frequency band (320–4 Hz) (i.e., eruptions cannot be predicted based solely on change in apparent resistivity). It should be noted that the forecasting of volcanic activity in upcoming weeks or months remains difficult in Sakurajima because there exists no relationship between mid-term ground deformation and eruption events. It may be possible to detect the temporal change in resistivity directly associated with magma movement, provided the MT site is established near the erupting crater at Sakurajima or at volcanoes that are not severely contaminated by EM noise.

0.0

-0.2

-0.4 -15 -10

-5

0

5

10

15

Lag (days) HARMT delay Fig. 7. (a) Temporal variations in apparent resistivity (Zyx) at a frequency of 22.5 Hz at HARMT and a frequency of 97 Hz at KURMT. The employed frequencies are marked in Figs. 3, 4, and 6. Error bars represent ± two standard deviations. (b) Cross-correlation of two time series without a period indicated by arrow in (a). Thick dashed line: cross-correlation between the two time series shown in (a). Thin dashed lines: cross-correlation calculated using all the other combinations of data at different frequencies. Thick solid line: the average of all cross-correlation shown in (b).

Zxx

Zyy

Log ohm-m Log ohm-m Log ohm-m

Zyx

Log ohm-m

Zxy

KURMT (97 Hz) 1.6 1.5

2.1 1.4 1.3 1.40 1.35 1.30 1.25

2.0 2.5 2.4

1.20 1.0 0.5 0.0 -0.5

2.0 1.9 1.8 1.7 1.6 1.5 1.0 0.9 0.8 0.7 0.6 0.5

-1.0 0.0 -0.5 -1.0 -1.5 -2.0 0

30

60

90

120

150

180

210

240

270

Days since 1 May 2008

300

330

360

390

420

0

30

60

90

120

150

180

210

240

270

300

330

360

390

420

Days since 1 May 2008

Fig. 8. Temporal variations in apparent resistivity for the four components of the impedance tensor (Zxy, Zyx, Zxx, and Zyy), as recorded at HARMT and KURMT. To enable a comparison, only the data in the frequencies shown in Figs. 3, 4 and 6 are shown.

K. Aizawa et al. / Journal of Volcanology and Geothermal Research 199 (2011) 165–175

HARMT (22.5 Hz)

171

172

K. Aizawa et al. / Journal of Volcanology and Geothermal Research 199 (2011) 165–175

Fig. 9. Comparison of temporal variations in apparent resistivity (Zyx) with rain, tides, ground deformation, eruption periods with major infrasound strength, and earthquakes around Sakurajima volcano. Vertical dashed lines indicate the possible start time of uplift in the summit direction. (a) Precipitation data recorded at a site located 150 m from HARMT. (b) Oceanic tides recorded at Furusato (Fig. 1). The raw data are shown along with a 1-day moving average. (c) and (d) Apparent resistivity data, as shown in Fig. 7. (e) Ground deformation (tilt record) observed in a tunnel at Arimura (Fig. 1). The radial component, which represents the direction of Minami-dake crater, was smoothed using a 1-day moving average. Higher values indicate uplift in the summit direction. (f) Infrasound data recorded at Arimura by low-frequency microphone. Gray hatched area represents a period of continuous ash emissions. (g) Deep low-frequency earthquakes (DLFs) occur in the southern part of Sakurajima at depths of the lower crust. Note that the vertical axis indicates the depth (km) of DLFs. Large and small stars represent DLFs with magnitudes of N1.0 and b 1.0, respectively. (e) BL-type earthquakes. (f) A-type earthquakes (volcano-tectonic earthquakes).

mesh and resistivity model (see Section 7) revealed that a 3-m change in the level of the water table results in only a 1% apparent resistivity change at KURMT and no significant change at HARMT. The groundwater level at

the borehole near KURMT has shown little variation (±1 m) since the start of observations in September 2008. These results indicate that the resistivity change is not explained by a change in groundwater level.

K. Aizawa et al. / Journal of Volcanology and Geothermal Research 199 (2011) 165–175

The observed correlation between resistivity change and ground deformation or DLF suggests that a component of the change in apparent resistivity was caused by volcanic activity. Uplift of the summit is believed to reflect the accumulation of magma at a depth of several kilometers (Ishihara, 1990; Hidayati et al., 2007; Tateo and Iguchi, 2009); therefore, the time correspondence between apparent resistivity and uplift suggests that resistivity change is related to the intrusion of magma beneath the summit. It should be noted that the ground tilt does not reflect the magma intrusion itself, but the balance between input and output of the subsurface magma. The output of magma, in the form of erupted material or drain-back of subsurface magma, is difficult to quantify. Consequently, it is not surprising that we do not observe a one-to-one correlation between inflation and changes in apparent resistivity. The relationship between resistivity and DLF is unexpected because the apparent resistivity at 22.5 and 97 Hz reflects shallow structures at depths of ~200 m to 2 km, whereas DLFs occur at deeper levels in the crust. Because DLFs beneath volcanoes are sometimes interpreted to be related to the migration of magma or supercritical fluid (e.g., Foulger et al., 2003; Nakamichi et al., 2007), the relationship might indicate that the resistivity change is related to fluid movement in the lower crust. The 80-day variation in apparent resistivity recorded at HARMT may reflect an unknown seasonal effect. Given that volcanoes may show cyclic behavior (e.g., Denlinger and Hoblitt, 1999; Voight et al., 1999; Peltier et al., 2008), the apparent 80-day variation recorded at HARMT may be explained in terms of volcanic activity. One possible origin of resistivity change is the behavior of volatiles (i.e., the gas phase at the temperature of magma). During magma ascent, volatiles are exsolved from the magma (referred to as “opensystem degassing”). Volatiles are generally considered to migrate vertically through the volcanic conduit, and are discharged from craters or fissures around the summit. In addition to this vertical degassing, we propose that volatiles can escape laterally through the conduit wall into the interior of the volcano. This mechanism is sometimes called “lateral degassing” or “lateral gas loss,” and has been studied by observations, experiments, and modeling (e.g., Baubron et al., 1990; Woods and Koyaguchi, 1994; Ida, 2007). The laterally degassed volatiles are finally absorbed by groundwater (via dissolution) at shallow levels beneath the volcano. As a result, the amount of dissolved ions changes, resulting in turn in a change in the resistivity of groundwater. As a result of continuous monitoring of hot spring water at Furusato spa (Fig. 1), Ohta (1986) found that the resistivity and temperature of groundwater show periods of synchronous change for several tens of minutes preceding explosive eruptions at the summit crater. The resistivity of hot spring water became resistive when the temperature increased. To explain this phenomenon, Ohta proposed that the pressure and amount of emanations (H2O, NaCl, HF, HCl, SO2, H2S, CO2, H2, N2, etc.) from deep magma show temporal increases. According to Ohta (1986), the hot spring water with large contribution of seawater is conductive relative to magmatic fluid (main component is considered to be H2O). Therefore, the resistivity of hot spring water increases even when the contribution of magmatic fluid increases. Although the time scale of resistivity change differs between that described by Ohta (1986) and the present study, our interpretation is basically consistent with Ohta's explanation. Because KURMT is located close to the shoreline, a large amount of cold and highly conductive seawater is mixed into the shallow groundwater system (Figs. 4 and 5). This situation is similar to that described by Ohta (1986), and it is considered that the resistivity of the groundwater increases with increasing proportion of volatiles. The CO2 dissolution (Fleury and Deschamps, 2008) or saturation (Kiessling et al., 2010) to the groundwater may also contribute to the resistivity increase of KURMT. On the other hand, the groundwater beneath HARMT is

173

expected to be rich in dilute meteoric water that is more resistive than volatiles, and the groundwater resistivity decreases with increasing contribution of volatiles. This explains the negative correlation between apparent resistivity at the two sites considered in the present study. By assuming that summit uplift is caused by a pressure source at a depth of approximately 3 km (Ishihara, 1990; Hidayati et al., 2007; Tateo and Iguchi, 2009), we account for the observation that the onset of summit uplift is approximately synchronous with the onset of change in apparent resistivity. We assume that the degassed volatiles travel rapidly to areas of shallow groundwater, thereby explaining the approximate synchronicity between the onset of summit uplift and the onset of resistivity change. On the other hand, previous numerical simulations have demonstrated that changes in the flow of groundwater occur slowly at locations far from the magma intrusion, in the case of an aquifer with moderate permeability (Hayba and Ingebritsen, 1997; Hurwitz et al., 2002, 2003). Geochemical monitoring of the hot spring water at Izu-Oshima volcano suggests that subsurface volatiles migrate at a speed of several to tens of meters per day (Sano et al., 1988), whereas our hypothesis assumes a speed of hundreds to thousands of meters per day. However, given that Sakurajima is located within Aira caldera (Fig. 1), a fracture network may have developed beneath the volcano, which would act as a migration path for volatiles (e.g., Ohta, 1986). Indeed, during the 1779 and 1914 eruptions at Sakurajima (in both cases, the erupted product was approximately 2 km3), groundwater in wells around the shoreline boiled at 20–30 h after the increase in seismicity (Imura, 1988), suggesting that the volatiles which degassed from the magma migrated rapidly to the surface through fractures, shortly after intrusion of the magma to shallow levels. Furthermore, at Masaya volcano, Nicaragua, fumarolic activity at a site located 3 km from the possible zone of magma intrusion showed an instantaneous change in response to increased volcanic activity at the summit crater (Pearson et al., 2008). Therefore, it seems reasonable that lateral degassing would occur at Sakurajima and that volatiles would travel rapidly through fractures toward the surface. The time lag in resistivity change between the two sites is explained by the difference in travel times between the degassing zone and the MT sites. If an MT site is located relatively far from the degassing zone, it takes much longer for the volatiles to migrate, resulting in a delay in the onset of resistivity change. Because the time lag between the two MT sites is variable through the observation period (Fig. 7), it is suggested that the degassing zone is mobile. The magma supply system at Sakurajima may consist of multiple routes, with the dominant route varying over time (Hidayati et al., 2007). Therefore, short-term changes in the magma supply system may have occurred during the observation period. 7. Change in resistivity structure modeled by 2D inversion To assess the feasibility of the interpretation, we modeled the change in resistivity structure using a two-dimensional (2D) model under certain assumptions. First, we assume that the resistivity of layers above the aquifer does not change. This assumption is necessary to obtain a stable result because the available frequency was limited during the observation period. The highest measured frequency (320 Hz) of MT impedance corresponds to a skin depth of 300 m for a 100 Ω m uniform half space, which is too coarse to investigate the resistivity structure between the surface and the aquifer. To investigate the validity of this assumption, it is necessary to obtain continuous MT data at a higher-frequency band (10,000– 320 Hz); however, such measurements were performed on only one day in 2007 (Kanda et al., 2008). Based on this first assumption, we fixed the structure above the aquifer using the data provided by Kanda et al. (2008). The AMT impedance (apparent resistivity and phase of Zyx), which was taken within 100 m of the present MT sites, is first

174

K. Aizawa et al. / Journal of Volcanology and Geothermal Research 199 (2011) 165–175

inverted using the code proposed by Ogawa and Uchida (1996) with a 2D mesh including the sea and topography (Fig. 10a). In the 2D calculation, we fixed the resistivity value from the surface to a depth of 40 m at KURMT (21 m above the water table) and from the surface to a depth of 300 m at HARMT (87 m above the water table) by using the results of the inversion of AMT data. The second assumption is that the 2D strike is N–S. Because Zxy data at KURMT are of poor quality (Figs. 2 and 3), the validity of this assumption cannot be rigorously assessed. After fixing the 2D strike to N–S, we inverted the impedance estimated from continuous MT measurements (Fig. 4). The apparent resistivity and phase of Zyx are inverted daily, for which the best-fit model for a given day is used as the first model for the following day. In the inversion, an error floor of 5% is given for apparent resistivity and for phase. For a calculation producing 435 days of data, the RMS misfits for the best-fit models for each day are less than 0.5, indicating that the temporal variations in impedance tensors are reasonably well reproduced by the model. Fig. 10b shows the temporal change in structure immediately beneath the MT sites. The obtained structure is interpreted in terms of the drillhole data (Fig. 5). The surface resistive region corresponds to a water-poor region such as lava or an unsaturated zone; the underlying conductive layer corresponds to a water-rich region. The occurrence of a highly conductive layer beneath KURMT indicates the large contribution of invading seawater to the groundwater system. Fig. 10c shows temporal variations in resistivity at a depth of 50 m below sea level, as extracted from Fig. 10b. The results are similar to those obtained for temporal change in apparent resistivity (Figs. 6, 7, and 10d). However, the ±20% variation in apparent resistivity is explained by the approximately ±150% variation in groundwater resistivity (in decimal scale rather than log scale). Although this result is based on many assumptions and involves a high degree of uncertainty, it may indicate that a large amount of volatiles was degassed laterally from the magma body.

8. Conclusion We conducted continuous magnetotelluric (MT) measurements at Sakurajima volcano, Japan, revealing variations in apparent resistivity of approximately ±20% and variations in phase of ±2° at two observation sites. Some of the changes in apparent resistivity correlate with ground deformation, suggesting an origin related to volcanic activity. We interpret these changes to represent mixing between volatiles and groundwater at around sea level beneath the volcano. We speculate that the volatiles exsolved from rising magma migrated laterally through a fracture network. Measurements of volcanic gas provide an indication of the volume of gas emitted vertically from the crater. However, it is generally difficult to measure the amount of laterally degassed volatiles. The measurement of CO2 flux from soil (e.g., Baubron et al., 1990; Chiodini et al., 2001; Hernández et al., 2001) or of gravity change (Gottsmann et al., 2006) is an alternative approach in this regard. The present results suggest that monitoring of electric resistivity by MT at a higher-frequency band (320–4 Hz) provides new information regarding lateral degassing beneath a volcano. The MT method has the advantage of high temporal resolution. Acknowledgements We thank the staff of Sakurajima Volcano Observatory for help during field surveys. The geomagnetic data used for remotereference processing were provided by the Esashi Geomagnetic Observatory, Geographical Survey Institute. Tiltmeter data were provided by the Osumi Office of the River and National Highway, Ministry of Land, Infrastructure, Transport and Tourism. Critical reviews by two anonymous referees greatly contribute to make this paper concise. A.Y. acknowledges the assistance of a JSPS research fellowship.

Fig. 10. Results of 2D inversion using apparent resistivity and the phase of Zyx. (a) Mesh used in the 2D inversion. (b) Temporal change in the resistivity structure immediately beneath the MT sites. See the text for details. (c) Temporal change in resistivity at 50 m beneath sea level, as extracted from (b). (d) Apparent resistivity data (Fig. 7c–d) were also shown for comparison.

K. Aizawa et al. / Journal of Volcanology and Geothermal Research 199 (2011) 165–175

References Aizawa, K., Ogawa, Y., Ishido, T., 2009. Groundwater flow and hydrothermal systems within volcanic edifices: delineation by electric self-potential and magnetotellurics. J. Geophys. Res. 114, B01208. doi:10.1029/2008JB005910. Aizawa, K., Yokoo, A., Kanda, W., Ogawa, Y., Iguchi, M., 2010. Magnetotelluric pulses generated by volcanic lightning at Sakurajima volcano, Japan. Geophys. Res. Lett. 37, L17301. doi:10.1029/2010GL044208. Aramaki, S., 1984. Formation of the Aira caldera, southern Kyushu, 22,000 years ago. J. Geophys. Res. 89, 8485–8501. Baubron, J.C., Allard, P., Toutain, J.P., 1990. Diffuse volcanic emissions of carbon dioxide from Vulcano Island, Italy. Nature 344, 51–53. Brenguier, F., Shapiro, N.M., Campillo, M., Nercessian, A., Ferrazzini, V., 2007. 3-D surface wave tomography of the Piton de la Fournaise volcano using seismic noise correlations. Geophys. Res. Lett. 34, L02305. doi:10.1029/2006GL028586. Brenguier, F., Shapiro, N.M., Capillo, M., Ferrazzini, V., Duputel, Z., Coutant, O., Nercessian, A., 2008. Towards forecasting volcanic eruptions using seismic noise. Nat. Geosci. 1, 126–130. doi:10.1038/ngeo104. Chiodini, G., Frondini, F., Cardellini, C., Granieri, D., Marini, L., Ventura, G., 2001. CO2 degassing and energy release at Solfatara volcano, Campi Flegrei, Italy. J. Geophys. Res. 106, 16213–16221. Denlinger, R.P., Hoblitt, R.P., 1999. Cyclic eruptive behavior of silicic volcanoes. Geology 27, 459–462. Eisel, M., Egbert, G.D., 2001. On the stability of magnetotelluric transfer function estimates and the reliability of their variances. Geophys. J. Int. 144, 65–82. Fleury, M., Deschamps, H., 2008. Electrical conductivity and viscosity of aqueous NaCl solutions with dissolved CO2. J. Chem. Eng. Data 53, 2505–2509. doi:10.1021/ je8002628. Foulger, G.R., Grant, C.C., Ross, A., Julian, B.R., 1997. Industrially induced changes in Earth structure at the Geysers geothermal area, California. Geophys. Res. Lett. 24, 135–137. Foulger, G.R., Julian, B.R., Pitt, A.M., Hill, D.P., Malin, P.E., Shalev, E., 2003. Threedimensional crustal structure of Long Valley caldera, California, and evidence for the migration of CO2 under Mammoth Mountain. J. Geophys. Res. 108 (B3), 2147. doi:10.1029/2000JB000041. Gamble, T.D., Goubau, W.M., Clarke, J., 1979. Magnetotellurics with a remote magnetic reference. Geophysics 44, 53–68. Gerst, A., Savage, M.K., 2004. Seismic anisotropy beneath Ruapehu volcano: a possible eruption forecasting tool. Science 306, 1543–1547. Gottsmann, J., Wooller, L., Martí, J., Fernández, J., Camacho, A.G., Gonzalez, P.J., Garcia, A., Rymer, H., 2006. New evidence for the reawakening of Teide volcano. Geophys. Res. Lett. 33, L20311. doi:10.1029/2006GL027523. Hanekop, O., Simpson, F., 2006. Error propagation in electromagnetic transfer functions: what role for the magnetotelluric method in detecting earthquake precursors? Geophys. J. Int. 165, 763–774. doi:10.1111/j.1365-246X.2006.02948.x. Hayba, D.O., Ingebritsen, S.E., 1997. Multiphase groundwater flow near cooling plutons. J. Geophys. Res. 102, 12235–12252. Hernández, P.A., Notsu, K., Salazar, J.M., Mori, T., Natale, G., Okada, H., Virgili, G., Shimoike, Y., Sato, M., Pérez, N.M., 2001. Carbon dioxide degassing by advective flow from Usu Volcano, Japan. Science 292, 83–86. Hidayati, S., Ishihara, K., Iguchi, M., 2007. Volcano-tectonic earthquakes during the stage of magma accumulation at the Aira caldera, southern Kyushu, Japan. Bull. Volcanol. Soc. Jpn. 52, 289–309. Hurwitz, S., Ingebritsen, S.E., Sorey, M.L., 2002. Episodic thermal perturbations associated with groundwater flow: an example from Kilauea Volcano, Hawaii. J. Geophys. Res. 107 (B11), 2297. doi:10.1029/2001JB001654. Hurwitz, S., Kipp, K.L., Reid, M.E., Ingebritsen, S.E., 2003. Groundwater flow, heat transport, and water-table position within volcanic edifices: implications for volcanic processes in the Cascade Range. J. Geophys. Res. 108, 2557. doi:10.1029/ 2003JB002565. Ida, Y., 2007. Driving force of lateral permeable gas flow in magma and the criterion of explosive and effusive eruptions. J. Volcanol. Geotherm. Res. 162, 172–184. Iguchi, M., 1994. A vertical expansion source model for the mechanisms of earthquakes originated in the magma conduit of an andesitic volcano: Sakurajima, Japan. Bull. Volcanol. Soc. Jpn. 54, 161–186. Iguchi, M., Yakiwara, H., Tameguri, T., Hendrasto, M., Hirabayashi, J., 2008a. Mechanism of explosive eruption revealed by geophysical observations at the Sakurajima, Suwanosejima and Semeru volcanoes. J. Volcanol. Geotherm. Res. 178, 1–9. doi:10.1016/j.jvolgeores.2007.10.010. Iguchi, et al., 2008b. GPS Observation Around Sakurajima and Aira Caldera. In: Ishihara, K. (Ed.), 10th Joint observation of Sakurajima volcano, pp. 53–62. Imura, R., 1988. Reconstruction of sequence of the An-ei eruption of Sakurajima volcano (A.D. 1779–1782) using the historical records. Bull. Volcanol. Soc. Jpn. 43, 373–383. Ishihara, K., 1990. Pressure sources and induced ground deformation associated with explosive eruptions at an andesitic volcano: Sakurajima volcano, Japan. In: Ryan, M. (Ed.), Magma Transport and Storage. Wiley, New York, pp. 335–356.

175

Kanda, et al., 2008. AMT measurement on Sakurajima volcano. In: Ishihara, K. (Ed.), 10th Joint Observation of Sakurajima Volcano, pp. 89–104. Kappler, K.N., Morrison, H.F., Egbert, G.D., 2010. Long-term monitoring of ULF electromagnetic fields at Parkfield, California. J. Geophys. Res. 115, B04406. doi:10.1029/2009JB006421. Kiessling, D., Hattenberger, C.S., Schuett, H., Schilling, F., Krueger, K., Schoebel, B., Danckwardt, E., Kummerow, J., the CO2SINK Group, 2010. Geoelectrical methods for monitoring geological CO2 storage: first results from cross-hole and surfacedownhole measurements from the CO2SINK test site at Ketzin (Germany). Int. J. Greenhouse Gas Control 4, 816–826. doi:10.1016/j.ijggc.2010.05.001. Mackie, R.L., Smith, J.T., Madden, T.R., 1994. Three-dimensional electromagnetic modeling using finite difference equations: the magnetotelluric example. Radio Sci. 29, 923–935. Miller, V., Savage, M., 2001. Changes in seismic anisotropy after volcanic eruptions: evidence from mount Ruapehu. Science 293, 2231–2233. Nakamichi, H., Watanabe, H., Ohminato, T., 2007. Three-dimensional velocity structures of Mount Fuji and the South Fossa Magna, central Japan. J. Geophys. Res. 112, B03310. doi:10.1029/2005JB004161. Nishimura, T., Uchida, N., Sato, H., Ohtake, M., Tanaka, S., Hamaguchi, H., 2000. Temporal changes of the crustal structure associated with the M6.1 earthquake on September 3, 1998, and the volcanic activity of Mount Iwate, Japan. Geophys. Res. Lett. 27, 269–272. Nishimura, T., Tanaka, S., Yamawaki, T., Yamasato, H., Sano, T., Sato, M., Nakahara, H., Uchida, N., Hori, S., Sato, H., 2006. Temporal changes in seismic velocity of the crust around Iwate volcano, Japan, as inferred from analyses of repeated active seismic experiment data from 1998 to 2003. Earth Planet. Space 57, 491–505. Ogawa, Y., Uchida, T., 1996. A two-dimensional magnetotelluric inversion assuming Gaussian static shift. Geophys. J. Int. 126, 69–76. Ohta, K., 1986. The behavior of hot spring water in relation with the volcanic activity of Sakurajima volcano. In: Kamo, K. (Ed.), 5th Joint observation of Sakurajima volcano, pp. 103–114. Patanè, D., Barberi, G., Cocina, O., Gori, P.De., Chiarabba, C., 2006. Time-resolved seismic tomography detects magma intrusions at Mount Etna. Science 313, 821–823. doi:10.1126/science.1127724. Pearson, S.C.P., Connor, C.B., Sanford, W.E., 2008. Rapid response of hydrologic system to volcanic activity: Masaya volcano, Nicaragua. Geology 36, 951–954. doi:10.1130/ G25210A.1. Peltier, A., Famin, V., Bachèlery, P., Cayol, V., Fukushima, Y., Staudacher, T., 2008. Cyclic magma storages and transfers at Piton de La Fournaise volcano (La Réunion hotspot) inferred from deformation and geochemical data. Earth Planet. Sci. Lett. 270, 180–188. Sano, Y., Nakamura, Y., Notsu, K., Wakita, H., 1988. Influence of volcanic eruptions on helium isotope rations in hydrothermal system. Geochim. Cosmochim. Acta 52, 1305–1308. Sumiko Consultants Co. LTD., 1985. Report of the Drilling for Geological Survey at Haruta-Yama. Sumiko Consultants Co. LTD., 1990. Report of the Drilling at Kurokami. Tateo, Y., Iguchi, M., 2009. Ground deformation associated with BL-type earthquake swarms at Sakurajima volcano. Bull. Volcanol. Soc. Jpn. 54, 175–186. Utada, H., 2003. Interpretation of time changes in the apparent resistivity observed prior to the 1986 eruption of Izu-Oshima volcano, Japan. J. Volcanol. Geotherm. Res. 126, 97–107. Utada, H., Yoshino, T., Okubo, T., Yukutake, T., 1998. Seismic resistivity changes observed at Aburatsubo, central Japan, revisited. Tectonophysics 299, 317–331. Voight, B., et al., 1999. Magma flow instability and cyclic activity at Soufriere Hills Volcano, Montserrat, British West Indies. Science 283, 1138–1142. Wight, D.E., Bostick, F.X., 1980. Cascade decimation—a technique for real time estimation of power spectra. Proceedings of the IEEE International Conference on Acoustic, Speech Signal Processing, pp. 626–629. Woods, A.W., Koyaguchi, T., 1994. Transitions between explosive and effusive eruptions of silicic magmas. Nature 370, 641–644. Yakiwara, H., Goto, K., 1999. Activity of deep low-frequency earthquakes beneath Sakurajima volcano in 1998. Japan Geoscience Meetings, Yoyogi, Japan, June 1999. Yamawaki, T., Nishimura, T., Hamaguchi, H., 2004. Temporal changes of seismic structure around Iwate volcano inferred from waveform correlation analysis of similar earthquakes. Geophys. Res. Lett. 31, L24616. doi:10.1029/2004GL021103. Yukutake, T., Yoshino, T., Utada, H., Watanabe, H., Hamano, Y., Shimomura, T., 1990. Changes in the electrical resistivity of the central cone, Mihara-yama, of Oshima volcano observed by a direct current method. J. Geomagn. Geoelec. 42, 151–168. Zlotnicki, J., Sasai, Y., Yvetot, P., Nishida, Y., Uyeshima, M., Fauquet, F., Utada, H., Takahashi, H., Donnadieu, G., 2003. Resistivity and self-potential changes associated with volcanic activity: the July 8, 2000 Miyake-jima eruption (Japan). Earth Planet. Sci. Lett. 205, 139–154.