Journal of Applied Geophysics 94 (2013) 1–14

Contents lists available at SciVerse ScienceDirect

Journal of Applied Geophysics

j ourna l homepage: www.e lsev ie r .com/ locate / jappgeo

Detection of microwave emission due to rock fracture as a new tool forgeophysics: A field test at a volcano in Miyake Island, Japan

Tadashi Takano a,⁎, Takashi Maeda b, Yoji Miki c, Sayo Akatsuka c, Katsumi Hattori d, Masahide Nishihashi d,Daishi Kaida d, Takuya Hirano d

a College of Science and Technology, Nihon University, Funabashi, 274-8501, Japanb Earth Observation Research Center (EORC), Japan Aerospace Exploration Agency (JAXA), Tsukuba, 305-8505, Japanc Institute of Space and Astronautical Science (ISAS), Japan Aerospace Exploration Agency (JAXA), Sagamihara, 229-8510, Japand Graduate School of Science, Chiba University, Chiba, 263-8522, Japan

⁎ Corresponding author. Tel.: +81 47 469 5529; fax:E-mail addresses: takano@ecs.cst.nihon-u.ac.jp (T. T

(T. Maeda).

0926-9851/$ – see front matter © 2013 Elsevier B.V. Allhttp://dx.doi.org/10.1016/j.jappgeo.2013.03.016

a b s t r a c t

a r t i c l e i n f o

Article history:Received 24 September 2012Accepted 30 March 2013Available online 12 April 2013

Keywords:Microwave emissionRock fractureVolcanic activityEarthquakeField testMiyake Island

This paper describes a field test to verify a newly discovered phenomenon of microwave emission due to rockfracture in a volcano. The field test was carried out onMiyake Island, 150 km south of Tokyo. Themain objectiveof the test was to investigate the applicability of the phenomenon to the study of geophysics, volcanology, andseismology by extending observations of this phenomenological occurrence from the laboratory to the naturalfield.We installed measuring systems for 300 MHz, 2 GHz, and 18 GHz-bands on the mountain top and mountainfoot in order to discriminate local events from regional and global events. The systems include deliberate datasubsystems that store slowly sampled data in the long term, and fast sampled data when triggered. We suc-cessfully obtained data from January to February 2008. During this period, characteristic microwave pulseswere intermittently detected at 300 MHz. Two photographs taken before and after this period revealedthat a considerably large-scale collapse occurred on the crater cliff. Moreover, seismograms obtained by near-by observatories strongly suggest that the crater subsidence occurred simultaneously with microwave signalson the same day during the observation period.For confirmation of the microwave emission caused by rock fracture, these microwave signals must be clearlydiscriminated from noise, interferences, and other disturbances. We carefully discriminated the microwave datataken at themountaintop and foot, checked the lightning strike data around the island, and consequently concludedthat thesemicrowave signals could not be attributed to lightning. Artificial interferences were discriminated by thenature of theirwaveforms. Thus,we inferred that the signals detected at 300 MHzwere due to rock fractures duringcliff collapses. This result may provide a useful new tool for geoscientists and for the mitigation of natural hazards.

© 2013 Elsevier B.V. All rights reserved.

1. Introduction

Japan has suffered several huge earthquakes in recent history. Inparticular, the Great East Japan Earthquake in March 2011 resultedin the death of about 20,000 people and caused heavy damage tohouses and social infrastructure including the Fukushima nuclearpower station (Cyranoski and Brumfiel, 2011). In light of such events,a greater demand than ever is placed on seismologists to clarify themechanism of great earthquakes and to predict their occurrence(Sagiya, 2012).

These requirements are currently not realistic (Uyeda et al., 2012).One reason may be the lack of knowledge and experience. In addition,study tools are limited to monitoring data from mechanical vibration

+81 47 467 9683.akano), maeda.takashi@jaxa.jp

rights reserved.

sensors (Fujinawa and Noda, 2007) and remote sensing measurementsof ground deformation (Massonnet et al., 1993). Therefore, new toolsare required, such as those in the field of electro-magnetic measure-ments, measurements of the Earth's atmosphere and ionosphere,and hydrological instruments. Among these approaches, however, theVAN method of ground potential measurement (Varotsos et al., 1993)lacks an understandable explanation of the causes of the signals, andit is difficult to discriminate the signal from the noise. The correlationbetween the ionospheric disturbances and earthquakes needs a physi-cal explanation, which could be provided by laboratory experiments(Molchanov and Hayakawa, 1998).

Meanwhile, Maki et al. (2006) recently found that electromagneticemissions at 300 MHz, 2 GHz, and 22 GHz aswell as at lower frequencyalready reported by former works (Cress et al., 1987; Nitsan, 1977)are caused by fracturing rock for the first time ever in laboratory exper-iments. Generally, measurements of electromagnetic emissions at mi-crowave bands are particularly difficult because the emitted signal's

(a) Location of Miyake Island. (b) Seismometer network in Miyake Island.

Fig. 1. Overview of Miyake Island. (a) Location of Miyake Island. (b) Seismometer network in Miyake Island.

2 T. Takano et al. / Journal of Applied Geophysics 94 (2013) 1–14

frequency is extremely high, and it has instantaneous properties with aconsiderably small time constant. Maki et al. exploited a novel measur-ing system that includes an antenna and a low noise amplifier for eachof three frequency bands, and a largememorywith excellent triggering.They used rock samples of quartzite, granite, gabbro, and basalt forexperiments. Electromagnetic emissions at 300 MHz and 2 GHz weredetected as for all rock samples. That at 22 GHz was detected only inthe case of quartzite due to the measurement difficulty. These emissionswere not thermally excited andwere discriminated from thermal noises.At 300 MHz and 2 GHz, the received powerwas largest for quartzite andgabbro, the second largest for granite, and the smallest for basalt. It isimportant that gabbro emitsmore power than granite though gabbro in-cludes less quartz than granite. Accordingly, this phenomenon is relatedto not only piezoelectricity but also other rock properties such as stiffnessand hardness, but has not yet been explained completely (Takano et al.,2010a,b, 2011). Therefore, the detection of microwave emission mayoffer a new tool to geosciences and seismic geophysics.

Before the above-mentioned finding, Geng et al. (1999) reportedmicrowave emission from rock fractures, but could not confirm thewaveform, spectrum, or power level. There have also been reports of

(a) Locations of our test sites at the top and foot of the mountain.

(

Fig. 2. Overview of our test sites at the top and foot of the mountain at Miyake Island. (a) Losite at the mountaintop.

electromagnetic emissions at lower frequencies (Cress et al., 1987;Nitsan, 1977). In addition, it is known that volcanic plumes are electri-cally charged because of the ejection of ions and atoms, vaporizationof water, and wind effects, and that the electromagnetic energy in vol-canic plumes can be emitted at lower frequencies (James et al., 2008).A proposed mechanism for the electromagnetic emissions at lower fre-quencies is the polarization effect due to piezoelectricity and chargemovementwhen piezoelectric rocks fracture. However, this model can-not explain the strong emission fromgabbro,which has been confirmedexperimentally (Maki et al., 2006). A model of current flow along thesurface due to electrical charging was proposed, based on friction andcrack generation during fracture (O'Keefe and Thiel, 1995). However,the time constant in this case is about 130 μs, so the high frequency can-not be explained with polarization due to piezoelectricity or the move-ment of charges, which were adopted in Yoshida and Ogawa (2004). Amodel that includes discharge across a microcrack could explain theexperimental results of the waveforms (Maki and Takano, 2004). Chargearising from bond dislocation between atoms or thermally excited elec-trons may cause relatively high voltages in some materials (Ohnishiet al., 2007). Kinetic excitation of the inner electrons and the nucleus,

b) Scenery around the test site at the mountaintop.

Crater cliff Crater center Antennas

cations of our test sites at the top and foot of the mountain. (b) Scenery around the test

Fig. 3. Configuration of the total measuring system. Rx: Receiver, DAQ PC: Data Acquisition PC, HDD: Hard Disk Drive, COM PC: Communication PC.

3T. Takano et al. / Journal of Applied Geophysics 94 (2013) 1–14

rather than the outermost electrons, could explain the experimentalemission power results and the weak dependency on piezoelectricity(Takano et al., 2010a,b).

It is certainly true that rock fracture occurs in association with vol-canic activity and earthquakes (Takano and Maeda, 2009). However,microwave detection during such natural events has not yet beenreported or described, probably because geoscientists did not knowabout the phenomenon and the relevant microwave technology.Therefore, it is important to verify the validity in a field test, sothat geoscientists can utilize the newly found phenomenon as astudy tool.

To investigate the validity of this technique, we planned a field testto detect possible emissions (Takano et al., 2008, 2009). The objectiveswere: (1) to confirm microwave emission during naturally occurringphenomena to support the experimental findings, (2) to confirm amethod for extrapolating experimental findings to large-scale events,and eventually (3) to show the applicability of the phenomenon togeophysics studies.

We considered several types of rock fracture for the field test andcompared the advantages and disadvantages of each, as follows:

(1) Cliff collapse around the crater in a volcano: A suitable volcanocan be selected according to its activity history. The emittedmicrowave propagates through the air, so we can eliminatethe ambiguity arising from heterogeneity of attenuation duringunderground propagation.

(2) Volcanic eruption: A suitable volcano can be selected, but theinterval between events may be too long, and it may not be

Table 1Estimation of S/N ratios of microwave signals due to rock fractures. In laboratory: Sample volreceiver = 2 × 103 m, noise figure = 3 = 5 dB.

Term 300 MHz

Wavelength [m] 1Bandwidth [Hz] 3 × 107

Emitted power in the experiment [W] 5.6 × 10−12

Radius of crashed rock sphere [m] 10Volume of crashed rock sphere [m3] 4.2 × 103

Extrapolated microwave emission [W] 6.9 × 10−4

Propagation loss of the ground 1Antenna gain 10 dBi = 10Received signal power [W] 1.09 × 10−11

Noise power [W] 3.5 × 10−13

Signal to noise power ratio 31

possible to predict the event well in advance. In addition, theexpelled lava and hot rocks may emit microwave noise dueto thermal excitation, which is difficult to discriminate fromemission due to rock fracture.

(3) Earthquake: The location and time of the event cannot be pre-dicted. Although attenuation has been well studied recently(Hosono et al., 2009), we still do not have sufficient knowledgeof microwave attenuation during underground propagation.

Therefore, at present, microwave signals due to rock fracture canbe detected from cliff collapsesmore easily than the other cases becauseof the known location and time, and can be considered an example ofthe canonical problem (Yasukawa et al., 2009). We thus decided tofocus on the microwave signals emitted during cliff collapses aroundthe crater of a volcano.

This paper first describes the outline of the field test, includingscientific as well as operational factors of the test site. Then, the con-struction and characteristics of the measuring system are presented.The measuring system was designed to support the acquisition of fielddata with limited funding andwork force resources. The correlation be-tween detected microwave signals and changes in the cliff shape isdescribed. Moreover, the seismogram signals from nearby observato-ries are also compared to show further evidence that cliff collapsesoccur at the same time as the microwave emissions. To demonstratethe reliability of the results, the method for discriminating microwavesignals from obstructive disturbances will be presented. The extentand depth of analysis of other data types will be restricted only to as-pects that relate to microwave emission phenomena.

ume = 3.4 × 10−5 m3, quartzite. In field test: Distance between a rock fracture and the

2 GHz 22 GHz

0.15 0.01362 × 108 5 × 108

2.7 × 10−11 1.73 × 10−12

10 104.2 × 103 4.2 × 103

3.3 × 10−3 2.1 × 10−4

1 115 dBi = 31.6 15 dBi = 31.63.8 × 10−12 1.92 × 10−15

2.3 × 10−12 5.8 × 10−12

1.7 3.3 × 10−4

Fig. 4. Accumulated 300-MHz signals at the top of the mountain during the test term from January 9 to February 4, 2008.

4 T. Takano et al. / Journal of Applied Geophysics 94 (2013) 1–14

2. Outline of the test and estimation of the received power

In selecting the field test site, the following factors were considered:extent of volcanic activity, electric power and communication infra-structure, seismometer network, transportation to the site, and theresearchers' security and accommodation. Comparing these factorsamong active volcanoes in Japan, we selected Mount Oyama volcanoon Miyake Island, which is located 150 km south of Tokyo, as shownin Fig. 1(a). This volcano has been active since a large eruption in2000, which forced all inhabitants of the island to evacuate. Several vol-canological and seismological studies have beenperformed on this largeeruption, and are helpful to our later data analysis (Fujita et al., 2001;Geshi et al., 2002; Ukawa et al., 2000).

As shown in Fig. 1(b), the Japan Meteorological Agency (JMA) hasinstalled six seismometers (MYCR, MKJA, MIG2, MYSA, MYKO, andMYYA) around Miyake Island. These seismometers, in particular,MYCR and MKJA, which are located around the crater, are useful forinvestigating tremors to determine their hypocenter locations andto clarify their relation to the detected microwaves. The raw datafrom these seismometers are not available to the public, but weobtained them through the cooperation of JMA.

Fig. 5. Relationship between the number of pulses and t

Fig. 2 shows an overview of our test sites at the top and foot of themountain. Fig. 2(a) shows the positions of our test sites relative to thecrater. The crater is about 1 km in diameter and 550 m in depth. Weinstalled measuring systems at two observation sites at the top andfoot of the mountain, denoted by square symbols. The two seismom-eters around the crater (MYCR and MKJA) are shown in black circles.

Fig. 2(b) shows the scenery around the test site at the top ofthe mountain. The antennas receive microwaves, which may beemitted by rock fracture around the crater cliff. The measuring instru-ments are located in an extreme environment very close to the crater,from which toxic gasses including H2S and SO2 are emitted almostconstantly. Furthermore, rain mixes with these gasses to form astrong acid.

As shown in Fig. 2(b), at the test site at the top of the mountain,microwave antennas were pointed toward the crater cliff. The anten-nas' field of view (FOV) covered the crater cliff, as indicated by dottedlines in Fig. 2(a). Meanwhile, the microwave antennas at the test siteat the mountain foot were pointed toward the zenith to receivemicrowave signals that cover the whole island.

Fig. 3 shows the configuration of the entire measuring system. Theobservation frequencies are 300 MHz, 2 GHz, and 18 GHz. The first

he pulse height for all signals detected at 300 MHz.

Fig. 6. Relationship between the duration and the pulse height for all signals detected at 300 MHz. The black circle for each pulse height represents the mean value of the durations.The lower and upper ends of the bar around the black circle represent the minimum and maximum values of the durations.

5T. Takano et al. / Journal of Applied Geophysics 94 (2013) 1–14

two frequencies are adopted in laboratory experiments on rock frac-tures (Maki et al., 2006), and 18 GHz is the same frequency as thatobserved by the microwave radiometer AMSR-E on the remote sens-ing satellite Aqua (Kawanishi et al., 2003). The use of high frequencieshas several advantages over the use of lower frequencies, which in-clude ease of calibration of the signal power, reduction of interferencefrom a directional antenna, and an inherent low-noise environment.Observations were carried out from November 2007 until May 2009with several interruptions resulting from measuring system failures.

Prior to the test, we calculated the relationship between receivedpower and emitted power based on the test site parameters and theantenna gain. The microwave signal power emitted in associationwith rock fractures was calculated from extrapolation of the experi-mental data from the size of small laboratory sample to that of anactual fractured rock in the field. It was assumed that the cratercliff at which the rock fractured was at a distance of 2 km from thefield antennas at most and that the volume of fractured rock was4.2 × 103 m3. The values of the emitted power in the laboratory exper-iment were calculated using the waveforms from quartzite reported in

Fig. 7. 300-MHz signals detected in the slow sampling mode on January 12. The orange (purwere strongest at 12:18 on this day.

Maki et al. (2006), because this is the only study that successfully mea-sured all frequencies of 300 MHz, 2 GHz, and 22 GHz. The comparisonbetween frequencies and the power estimation in optimal conditionscan be accomplished using these values, though Miyake Island is com-posed predominantly of basalt. The received power was estimated aslisted in Table 1.

For the noise characteristics of the receiving system, we were ableto estimate the S/N ratio of the signal power according to the thermalnoise power, as summarized in Table 1. The S/N ratio at 300 MHz is31, which is large enough to allow the signal to be discriminatedfrom the noise. The observation is marginal at 2 GHz, with an S/Nratio of 1.7, and may be impossible at 22 GHz, where the S/N ratiois less than unity.

3. Measuring system constitution

The measuring system should be versatile enough to discriminatelocal phenomena relevant to crater cliff collapses on the mountain topfrom island-wide phenomena, as well as to simultaneously support

ple) line represents the 300-MHz signals at the top (foot) of the mountain. Both signals

(a) Diurnal changes of detected signals. There are three groups of pulses (A, B, and C).

(b) Waveforms in signal B from 20:23:38.650 to 20:23:39.150.

Fig. 8. 300-MHz signals detected in the slow sampling mode on January 18, 2008. In each figure, the orange (purple) line represents 300-MHz signals at the top (foot) of the mountain.(a) Diurnal changes of detected signals. There are three groups of pulses (A, B, and C). (b) Waveforms in signal B from 20:23:38.650 to 20:23:39.150.

6 T. Takano et al. / Journal of Applied Geophysics 94 (2013) 1–14

regular observations and detailed data analyses. As shown in theblock diagram of Fig. 3, the microwave receiving systems at the topof the mountain consist of a frequency-peculiar antenna and a micro-wave receiver (Rx) including a low noise amplifier for each of threefrequency bands, a data storage device (HDD), and a data transmissionlink including HUB, Router, and Wi-Fi circuit. Each receiving system atthe three different frequency bands is limited to narrow bandwidthsthat are called sub-bands. Gathering data from all sub-bands, we canobtain the spectrum information of the whole band from 300 MHz to18 GHz. This configuration is inevitable due to the difficulty in usingan ultra-wide band receiver including antenna. Electric power wassupplied through cables from the mountain foot.

A similar receiving system was installed at the mountain foot,with the antennas pointing toward the zenith. In this configuration,the side lobe is at 90° relative to the main lobe, which points towardthe target crater cliff. Therefore, the directivity at the angle to the cliffis−13.5 dB lower than at the main lobe, and the free-space loss fromthe collapsed cliff to the mountain foot antenna is 4.3 dB larger than

that to the mountain top antenna, considering the distances inFig. 2(a). The received power at the foot site calculated from thesetwo factors is −17.8 dB, which is 0.016 times weaker than that atthe top site. This value is sufficiently small for the foot antenna todiscriminate the real signals generated in rock fractures in the craterfrom the island-wide interference noise that could be detected byboth antennas.

The detected signals were first sampled at a lower sampling rateof 1 kHz to reduce the frequency bandwidth, which enabled regulartransmission to laboratories in Tokyo. Using this configuration, wemonitored the operational status of themeasuring system and obtainedrough data within the limits imposed by our communication capabili-ties. The circuit for detecting the received microwaves operates mostlyin the peak holdmode. In thismode, the peak value of a pulse is held fora timewindow of 1 ms. This mode is equivalent to low-pass filtering toreduce the information rate. When a strong signal of a significant eventis detected, the data are sampled at a faster rate of 1 MHz. The strongsignal is a trigger to activate faster sampling. During faster sampling,

Fig. 9. Expanded waveform of the impulse signal B1 in the fast sampling mode at 20:23:38.910.

7T. Takano et al. / Journal of Applied Geophysics 94 (2013) 1–14

the detection circuit operates in the rectified mode, which suppliesthe absolute amplitude of observed microwave signals in frequencybands lower than 1 MHz. This mode is used to evaluate the waveform

(a) Photo taken from the top site on January 9, 2008November 11, 2008

(c) Aerial photo taken on March 11, 2008 (courtesa crater cliff with th

12

3

Fig. 10. Change of the shape of the crater cliff. (a) Photo taken from the top site on January 9on March 11, 2008 (courtesy of Japan Coast Guard). A red circle indicates a crater cliff with

precisely. The data saved during the faster sampling are transmittedoccasionally on a communication line or recorded to a hard disk. Notethat the sampling rate is much lower than a radio frequency or the

(b) Photo taken from the top site on

y of Japan Coast Guard). A red circle indicatese large collapse.

Top site

, 2008. (b) Photo taken from the top site on November 11, 2008. (c) Aerial photo takenthe large collapse.

8 T. Takano et al. / Journal of Applied Geophysics 94 (2013) 1–14

intermediate frequency of a heterodyne receiver due to the limitation ofthe memory and communication capabilities. This scheme, called anunder-sampling system, gives the smallest limit of the waveform.

For visual data, we initially used data from a video camera installedon the crater cliff; however, the resolutionwas too poor to allow changesin the cliff to be recognized, and the camera accidentally fell to the craterbottom. Photos taken on the occasions of our visit also provided valuablevisual data, but the temporal resolution was very limited.

4. Results of microwave detection

Many microwave signals were detected at 300 MHz during thetest term from January 9 to February 4, 2008. In contrast, the 2-GHzsignals were much weaker and were often contaminated with inter-ference signals, probably from a radar-transmitting antenna, and no18.7-GHz signals were detected at all.

Fig. 4 shows the time-series data for the 300-MHz signals detectedin the slow sampling mode at the mountain top during the test period.

(a) Diurnal change

(b) Detailed temporal change with microw

Fig. 11. Variation in the absolute ground velocity at MYCR (mountaintop) on January 18. Taccording to their amplitude, duration, and waveform by JMA. A, B, and C in the lower panground velocity coincident with the microwave signals A, B and C. The D' and E' indicate examconsidered as an error in data (See Fig. 16). (a) Diurnal change. (b) Detailed temporal change

In this figure, the datawere examined every 10 s, and the largest ampli-tude in each time bin was plotted. Obvious interference signals wereeliminated beforehand as described in the last paragraph of Section 6.Fig. 4 indicates that the signals did not appear continuously, but ratherintermittently in clusters. In particular, the signals detected on January12 and on January 18 are the most significant but are quite differentto one another in nature, having the greatest number of signals andthe strongest signal, respectively. The reason for this difference will beclarified later using several auxiliary data types.

The data in Fig. 4 were then processed to obtain the distributionaccording to the pulse amplitude, which is a basic statistical character-istic of the detected signals. The pulse amplitude indicates the initialpeak of a decaying signal in the slow sampling mode. The receivingamplifier output includes not only signal power but also noise power,so we have to discriminate the true signal from the noise. Consideringthe noise level of our receiver, we set the discrimination level of a signalto be 0.08 V. Fig. 5 shows the result. The number of pulses decreasesexponentially up to a height of 1.35 V. Almost all pulses stronger than

ave signals

he peaks represented by diamonds were classified as earthquakes or volcanic tremorsel indicate the microwave signals shown in Fig. 8(a). A', B' and C' indicate the absoluteples not accompanied by microwave signals. In particular, the E' expressed by X symbol iswith microwave signals.

Fig. 12. Variations in the absolute ground velocity at MYCR (mountaintop) and MKJA (mountain foot) on January 12. The symbols and lines are drawn similarly to Fig. 11, and graylines and circles are added as for MKJA data. After 12:18, the MYCR data were lost due to failure.

9T. Takano et al. / Journal of Applied Geophysics 94 (2013) 1–14

1.35 V occurred on January 12 and 18. The percentage of pulses stron-ger than 1.35 V was 0.3% of the total pulses.

The duration of a pulse varies significantly between detectedsignals. We defined the duration time of a pulse as the time lapse be-tween the signal rising above and falling below the discriminationlevel of 0.08 V. Under this definition, a pulse can include more thanone data sample in the duration time. We then analyzed the relation-ship between the duration and the pulse height. The result is shownin Fig. 6. The black circle for each pulse height represents the meanvalue of the durations. The lower and upper ends of the bar aroundthe black circle represent the minimum and maximum values ofthe durations. In this figure, three distinctive groups of data can beobserved.

The first group of pulses in Fig. 6, indicated by P, extends throughthe abscissa range from 0 V to 3 V with almost linear increases induration. This trend is attributed to the time response of the receiverin the peak hold mode, which is later explained using a concretewaveform shown in Fig. 8(b). Accordingly, the duration depends onthe peak amplitude. Usually, each data point in this trend is a singlepulse, but two pulses are included in the level from 2.35 V to 2.4 V,as shown in Fig. 5. The strongest signal has a 3-V pulse height and50-ms duration and was detected on January 18. This group looks

Fig. 13. All the seismograms from Miyake Island for quake A' in Fig. 11(a). The origin of the aMKJA is located at the foot of the mountain.

quite different from noise, and is believed to include pulses generatedby large cliff collapses over a short duration, as will be explained later.

The second group is in the abscissa range from 0 V to 1.25 V withdurations longer than 50 ms, and is indicated by Q. There are morepulses at the lower pulse height, as shown in the distribution of Fig. 5.Therefore, this group corresponds to noise. Each datum in this grouphas maximum and minimum duration values in addition to the meanvalue. The distribution of the duration inside each datum has a stronglybiased shape. For example, from 0.15 V to 0.2 V on the abscissa, themaximum value is 210 ms, but the mean value is near 0 ms. This indi-cates that although most pulses have short durations, a few pulses atthis level have notably long durations.

The third group of pulses is from 1.3 V to 1.45 V, from 1.6 V to1.65 V, and from 2.15 V to 2.2 V on the abscissa and is indicated byR. This group is extraordinarily high and wide, and all the pulseswere detected on January 12. It will be demonstrated later in thisdocument that these signals were generated by lightning.

For confirmation of the origin of the signals, we extended the signalon January 12, which is shown in Fig. 4, and checked its relation withthe signal obtained at the mountain foot site. The result for this day isshown in Fig. 7. In this figure, the orange and purple lines representthe signals at the mountaintop and foot, respectively. The signals at

bscissa axis in each panel is the same (18:44:57). MYCR is located at the mountaintop.

Fig. 14. All the seismograms from Miyake Island for quake B' in Fig. 11(a). The origin of the abscissa axis in each panel is the same (20:20:28). MYCR is located at the mountaintop.MKJA is located at the foot of the mountain.

10 T. Takano et al. / Journal of Applied Geophysics 94 (2013) 1–14

themountaintop are quite dense, with only small intervals between thetwo time periods, while those at the mountain foot are sparser but oc-casionally have amplitudes comparable with those at themountain top.

We then extended the signal on January 18 in Fig. 4, and show thechange on that day with orange lines in Fig. 8. We can see quite adistinct difference in the signals on January 18 and that on January12, shown in Fig. 7. In Fig. 8(a), the 300-MHz signals detected onJanuary 18 consist of completely separate events. At around 19:00,the strongest signal A was recorded, with an amplitude of 3 V atthe mountain top. This corresponds to the black circle indicated byA in Fig. 5. Close to signal A, several signals can be recognized, oneof which is 1.6 V. At around 20:30, the strong signal B of 1.2 V wasdetected with an associated signal of 1.0 V, and signal C was detectedat around 22:00. In addition, there were two weaker signals at around05:00 and 14:00. Purple lines are added to the plots to represent thesignals at the mountain foot. For example, at the same time as signalA, a weak signal was also recorded. As the amplitude of this signalis one-twentieth of the amplitude of signal A, and the decaying wave-form is confirmed to be similar to signal A, it is inferred that this signalwas emitted at the crater cliff, and detected at the mountain foot by theantenna side lobe. Signals B and C are also associatedwithmuchweakersignals. Accordingly, signals A, B, and C are thought to have originated atthe mountaintop.

Fig. 15. All the seismograms from Miyake Island for quake C' in Fig. 11(a). The origin of the aMKJA is located at the foot of the mountain.

After finding the impulse signals in Fig. 8(a), we can enlarge thewaveform even more. Fig. 8(b) shows signal B with an abscissascale of 50 ms/division. We can see the details of the signal, whichis composed of several pulses. The strongest pulse B1 lasts about30 ms from 20:23:38.910. There is no signal at the time of B1 at themountain foot. Since the signal detector was operated in the peakhold mode, the amplitude decays with a half-amplitude time constant,about 10 ms in this case. Therefore, we cannot discriminate smallerpulses that are enclosed within this decaying envelope.

We can expand the waveforms in Fig. 8(b) even more using thefast sampling mode, as shown in Fig. 9. In this mode, the detectoroutputs a rectified waveform from the microwave of both polari-ties. The signal amplitudes in these figures were not preciselycalibrated. From examining Fig. 9, it becomes apparent thateach pulse in Fig. 8(b) is actually composed of thousands of shortpulses.

5. Correlation with crater cliff collapses and quakes

A possible correlation between microwave emission and cliffcollapse is studied in this section. We photographed the shape of thecrater cliff from a fixed observation point in a fixed magnification withthe same camera each time we visited the test site. Fig. 10(a) and (b)

bscissa axis in each panel is the same (21:45:24). MYCR is located at the mountaintop.

11T. Takano et al. / Journal of Applied Geophysics 94 (2013) 1–14

shows the cliff shapes on January 9, 2008 and November 11, 2008,respectively. Comparison of the two scenes indicates that areas 1, 2,and 3, highlighted with red circles, all collapsed significantly.

In addition, we obtained an aerial photo taken by the Japan CoastGuard on March 11, 2008, shown in Fig. 10(c). In this figure, a whitesquare symbol represents our test site at the mountaintop, and thered circles correspond to areas 1, 2, and 3 in Fig. 10(a). Based onFig. 10(a) and (c), the time of the large collapse can be narroweddown to the period between January 9 and March 11, 2008.

To estimate the collapsed volume, we calculated the lateral size ofthe largest collapsed area, area 2, in Fig. 10(a) based on the viewangles and the distances to the cliff. The normal thickness of area 2was calculated from Fig. 10(c) comparing the dent with the craterdiameter. Accordingly, the collapsed volume was modestly estimatedto be 94,000 m3, which is believed to be large enough to exceed thecritical value for signal detection, as shown in Table 1. In addition,since other areas of the cliff including areas 1 and 3 collapsed, micro-wave signals were likely to be emitted several times.

To determine the time of cliff collapses more precisely, we can ob-tain and analyze seismological data from several observatories shownin Fig. 2(a). We first give an overview of the seismic signals from sta-tion MYCR, at the mountaintop. Fig. 11(a) shows the variation in theabsolute ground velocity on January 18. The time scale is determinedto be equal to Figs. 7 and 8(a) for comparison purposes. We can seemany quake signals during the analyzed term because the station islocated on an active volcano. The peaks, represented by diamonds,were classified as earthquakes or volcanic tremors by JMA, accordingto their amplitude, duration, and waveform. The peaks, representedby small black circles, were not recognized as earthquakes or volcanictremors by the criteria of JMA. Accordingly, quakes due to cliff col-lapses may be included in the latter category. Fig. 11(b) shows thedata from 18:00 to 22:00 in expanded form. The 300 MHz microwavesignals A, B, and C in Fig. 6 are included in Fig. 11(b). They coincidewith increases in the absolute ground velocity (A', B', and C'). However,there are ambiguities in determining the timing because the time bin ofmicrowave reception is 10 s, which is comparable to the duration of thequake, and because of asynchronous measurement between the micro-wave receivers and the seismometer.

In contrast, a number of strong peaks of the seismogram atMYCR were not associated with corresponding microwave signalsin Fig. 11(a), such as D' and E'. Naturally, not all of these peaks weregenerated by cliff collapses, because the seismograms also reflect earth-quakes or volcanic tremors. The emitted microwave signals may notreach the receiving antenna, either because of the collapse location or

Fig. 16. All the seismograms in Miyake Island for quake D' in Fig. 11(a). The origin of the abMKJA is located at the foot of the mountain.

because of the main lobe direction of the receiving antenna and radiowave being blocked by obstacles. Therefore, we will focus on particulartimeswhen analyzing the seismograms, especially around the time thatmicrowave signals A, B, and C occurred, in order to extract commonfeatures of the seismograms in these events.

For comparison purposes, variations in the absolute ground velocityat MYCR and MKJA (mountain foot) on January 12 are shown inFig. 12. The symbols and lines are drawn similar to those in Fig. 11, andgray lines and circles are added as MKJA data. At 11:00, a largest peaksignal was detected, and it was three times larger than the largest signalon Jan 18. In fact, it rainedheavilywith lightning on this day. After a light-ning strike at 12:18, theMYCRdatawere lost due to failure. In correspon-dence to this time, the microwave emission in Fig. 7 is so dense thatquakes happened at times that correspondedwithmicrowave emissions.However, we cannot infer the physical relation between the microwaveemission and seismogram due to the ambiguity of timing determinationas described before.

According to previous studies (Fujita et al., 2001; Ukawa et al.,2000), seismograms associated with the volcanic activity in MiyakeIsland can be classified into the three types, as shown below:

(1) Volcano-tectonic event: The P and S waves can be clearly distin-guished, as in a typical earthquake.

(2) Low-frequency (LF) earthquake: The P and S waves cannot beclearly distinguished. The dominant frequency in the waveformis lower than 1 Hz. The waveform slowly rises, lasts more thanseveral tens of seconds, and slowly decays. LF earthquakes arelikely to indicate magma migration.

(3) Swarm-like activity of LF earthquakes: These events have featuressimilar to those of type (2) events, but the dominant frequency inthe waveform is higher than that of type (2) events. This eventwas observed a few hours before tilt steps and originated atshallow depths (0–2 km) around a crater.

Fujita et al. (2001) suggested that type (3) is associated with cra-ter subsidence, which is actually similar to a cliff collapse. In addition,these seismogram features were studied in cases of landslides andglacier collapses, and presented in Dammeier et al. (2011) and Feng(2011). Therefore, let us view the details of the waveforms in someinstances in reference to the above-mentioned classification.

The seismograms at MYCR, MKJA, MYKO, MIG2, MYSA, and MYYAfor quake A' in Fig. 11 are shown in Fig. 13. In these seismograms, the or-igins of the abscissa axes are the same (18:44:57). For all components(NS, EW, and UD), MYCR, located on the mountaintop, was excitedfirst and had the strongest amplitudes. Then a few seconds later, MKJA

scissa axis in each panel is the same (00:25:42). MYCR is located at the mountaintop.

Fig. 17. Seismogram of MYCR at E' in Fig. 12(a). The origin of the abscissa axis in eachpanel is the same (04:43:28).

12 T. Takano et al. / Journal of Applied Geophysics 94 (2013) 1–14

at the mountain foot was excited. In contrast, MYKO, MIG2, MYYA, andMYSA at the coastal areaweremuchmoreweakly excitedwith a greaterdelay. Accordingly, the hypocenter of quake A' was located closer toMYCR than other seismometers. If the hypocenter was deep under-ground, the distances to all observatories would be almost equal toeach other. Therefore, it is reasonable to conclude that the hypocenteris located close to the ground surface. In addition, the dominant fre-quency in the waveform at MYCR is higher than 1 Hz, which coincideswith the features of type (3) events. Accordingly, quake A' is inferredto be generated by a cliff collapse.

Next, all seismograms for quake B' are shown in Fig. 14. The originsof the abscissa axes are the same (20:20:28). For all components, MYCRwas excited first and had the strongest amplitudes. A few seconds later,MKJA was excited. MIG2 and MYSA, located in the coastal area, wereexcited another several seconds later. MYKO and MYYA at the coastalarea were quite weakly excited. Accordingly, the hypocenter of quakeB' was located nearest to MYCR and close to the ground surface. Thewaveform at MYCR coincides with the features of type (3) events.Accordingly, quake B' is also inferred to be generated by a cliff collapse.

The seismograms for quake C' are shown in Fig. 15. Only MYCRshowed the features of type (3) events, while the other seismogramsbarely showed any peaks. Therefore, quake C' is also inferred to be gener-ated by a cliff collapse. Consequently, we conclude that the seismogramsfor A', B', and C' indicate the cliff collapses that simultaneously generatedmicrowave emissions A, B, and C, as shown in Fig. 11(b). However, theseconclusions have ambiguity, as several other seismograms also show thesame features as A', B', and C'.

In contrast, all the seismograms for event D' shown in Fig. 11(a)are shown in Fig. 16. The origins of the abscissa axes are the same(00:25:42). The P and S waves can be clearly distinguished for all com-ponents as is a feature of type (1) events. In addition, the amplitudesmeasured by MYCR, MKJA, MIG2, and MYYA were excited almostsimultaneously and with similar amplitude, though MYCR was themost strongly excited among all seismometers. Therefore, the distancesfrom the hypocenter of quake D' toMYCR,MKJA,MIG2, andMYYAwerealmost the same, and thus the hypocenter was not located near the cra-ter surface. Accordingly, around the time of D', no cliff collapse occurred,and this agrees with the fact that no 300-MHz microwave signals weredetected, as indicated in Fig. 8(a).

A strong seismometer signal was also detected at E', as shown inFig. 11(a). The relevant seismogram from MYCR is shown in Fig. 17.Only the up–down component changed significantly, in the down direc-tion,while the north–south and east–west componentswere completelyquiet. Therefore, we regard this response as an instrumental error.

Table 2Eight periods when lightning strike data were investigated.

Periods

0:00:00–23:59:59, Jan. 09, 20080:00:00–23:59:59, Jan. 12, 20080:00:00–23:59:59, Jan. 18, 20080:00:00–23:59:59, Jan. 21, 20080:00:00–23:59:59, Jan. 23, 20080:00:00–23:59:59, Feb. 02, 20080:00:00–23:59:59, Feb. 03, 20080:00:00–23:59:59, Feb. 04, 2008

6. Discrimination from obstructing factors

Several factors obstruct the detection ofmicrowave signals generatedfrom rock fractures. The first factor is emission due to lightning. Light-ning is caused by movement of electrical charge in the air. The temporalchange of this movement is represented by a rectangular pulse with awidth T and height A. Its frequency spectrum is then (A/πf) sin (πfT),where f is its frequency, which would extend through the 300 MHzband. However, the nature of lightning at microwave frequencieshas not been reported so far. Considering the high sensitivity of ourreceiver, we cannot rule out the influence of lightning on our receiverat 300 MHz. Accordingly, we thoroughly investigated the occurrenceof lightning during the eight periods listed in Table 2 when strong300 MHz microwave signals were detected on the mountain top, asshown in Fig. 4.

Lightning strike data were provided by the Japan Lightning Detec-tion Network (JLDN) (Ishii et al., 2005), which determines the locationsof lightning strikes using sensors installed throughout Japan. The light-ning sensors observe the LEMP (lightning electromagnetic impulse)in frequency ranges from VLF (3–30 kHz) to LF (30–300 kHz). The

lightning sensor closest to the test site is on Kouzu Island, 30 km north-west of Miyake Island.

According to JLDN, the only lightning strike detected near the craterin the periods shown in Table 2 was at 12:18 on January 12, 2008. Fig. 7indicates that the twomicrowave signals at 12:18were the strongest onthis day, and that several signals were received simultaneously at themountaintop and foot. The meteorological data also reported that itrained heavily with lightning most of the time, and that it did not rainin the periods from 0:00 to 3:00 and from 21:00 to 24:00 on this day.This raining time clearly corresponds to the period of microwave emis-sion in Fig. 7. Accordingly, many signals were likely to be caused bylightning. Group R in Fig. 6 is typical of data that originates from light-ning. Note that the other data in Fig. 6might have originated from light-ning but cannot be discriminated from data that originated from theother sources. On the other hand, the weather was completely clearwith no lightning observed on January 18, according to themeteorolog-ical data.

The second factor is rain, which has a strong correlation with light-ning. Rain changes the electrical characteristics of the ground. Therefore,we gathered meteorological data on the precipitation in the periodsshown in Table 2. It rained only on January 12, 2008, while the weatherwas clear without rain on the other days.

The third factor is artificial radio interference. Two kinds of artificialinterference were observed in several cases, as shown in Fig. 18. One ofthem has a regular periodicity in the waveform, and the other has anabrupt appearance with a high frequency component and an offset ofthe noise level. The interference signals were differentiated by findingthe waveforms and by comparison of signal levels at the mountaintop

(a) Periodic signal with 10-second intervals.

(b) High frequency with abrupt appearance and the noise level offset.

Fig. 18. Two kinds of artificial interference removed from the microwave data received at the mountaintop and the mountain foot. (a) Periodic signal with 10-second intervals.(b) High frequency with abrupt appearance and the noise level offset.

13T. Takano et al. / Journal of Applied Geophysics 94 (2013) 1–14

and the mountain foot. We succeeded in eliminating this interferencethrough filtering.

7. Conclusions

A field test was carried out in the extreme environment of thevolcano at Miyake Island. The novel measuring system worked wellbut with several interruptions. Microwave signals were detected inter-mittently at 300 MHz. Each signal was composed of pulses with quitenarrow widths, which were estimated to be several μs from the fast-sampling data. We performed a statistical analysis of the receivedsignals to identify peculiarities in several cases. There were extraordi-narily high-amplitude pulses in the slow-sampling data on two days:January 12, 2008 and January 18, 2008.

It was revealed by photos of the rock cliff that the volcanic cratersignificantly changed in shape between January 9 and March 11, 2008.Analysis of seismograms indicated subsidence of the crater ratherthan typical earthquakes. Local quake activity at shallow depths wasverified by the seismograms installed around the island. The correlationbetween the microwave emissions and ground quakes was significant.These facts strongly support the case that separate microwave pulseson January 18, 2008 were generated by cliff collapses.

By using meteorological data and comparing the detection datacollected at the mountaintop with that collected at the mountainfoot, we concluded that the dense group of pulses on January 12,2008 originated from lightning. Meanwhile, several artificial interfer-ences were observed, but we differentiated and eliminated themwithour filtering software.

Overall, microwave emission was detected during volcanic activityand is believed to be generated by rock fracture in cliff collapses.Microwave emission associatedwith rock fracture is a newly discoveredphenomenon, andwe can conceivemany applications for it in scientificand practical fields. This technique may offer a new scientific toolfor geophysics and rock physics research. It is already known thatmicrowave emission is closely related to microcrack occurrence andthe resultant voltage generation (Maki and Takano, 2004). Therefore,the relationship of material parameters such as brittleness, hardness,

and electric conductivity with microwave emission should be clarified.In addition, the domain structure of a material could be investigatedusing microwave diagnosis.

In practical applications, the results described in this paper offer thepossibility that this technique could be applied to volcano monitoring.The same technique could also be applied to other natural disastersassociated with rock fractures, such as earthquakes and landslides. In-deed, we have already developed a data analysis method for extractinglocal and faint microwave anomalies from the data obtained from asatellite-bornemicrowave sensor, andhave successfully detected the sig-nature of a volcanic eruption and an earthquake (Maeda and Takano,2008, 2010). However, the timescale of the hazardous event and the rel-evant rock fracture has not yet been clarified. If rock fractures around anasperity before a great earthquake, this technique could be valuable formitigating social losses by predicting earthquake occurrence.

Acknowledgments

We thank the Japan Meteorological Agency and Dr. S. Yoshida andDr. K. Nagata of the Earthquake Research Institute, University of Tokyo,for arranging the observation site, providing seismometer data, anddiscussing the obtained results. We express our gratitude to the JapanCoast Guard for providing photographs of the cliff collapse. Finally, wethank Dr. K. Tachikawa of JAXA for his continuous support for thisproject.

References

Cress, G.O., Brady, B.T., Rowell, G.A., 1987. Sources of electromagnetic radiation from fractureof rock samples in the laboratory. Geophysical Research Letters 14, 331–334.

Cyranoski, D., Brumfiel, G., 2011. Fukushima impact is still hazy: chaos and bureaucracyhamper assessment of nuclear crisis. Nature 477, 139–140.

Dammeier, F., Moore, J., Haslinger, F., Loew, S., 2011. Characterization of alpine rockslidesusing statistical analysis of seismic signals. Journal of Geophysical Research 116,F04024. http://dx.doi.org/10.1029/2011JF002037.

Feng, Z., 2011. The seismic signatures of the 2009 Shiaolin landslide in Taiwan. NaturalHazards and Earth System Sciences 11, 1559–1569. http://dx.doi.org/10.5194/nhess-11-1559-2011.

Fujinawa, Y., Noda, Y., 2007. Research and development of earthquake early warning-application systems for various users. BUTSURI-TANSA 60 (5), 375–386.

14 T. Takano et al. / Journal of Applied Geophysics 94 (2013) 1–14

Fujita, E., Ukawa, M., Yamamoto, E., Okada, Y., Kikuchi, M., 2001. Volcanic earthquakesand tremors associated with the 2000 Miyakejima Volcano eruption (in Japanese).Journal of Geography 110 (2), 191–203.

Geng, N., Deng, M., Cui, C., 1999. Simulated experimental studies on earthquake predic-tion by using infrared and microwave remote sensing. In: Hayakawa, M. (Ed.), Atmo-spheric and Ionospheric Electromagnetic Phenomena Associated With Earthquakes.Terra Scientific Publishing Company (TERRAPUB), Tokyo, pp. 751–758.

Geshi, N., Shimano, T., Chiba, T., Nakada, S., 2002. Caldera collapse during the 2000eruption of Miyakejima Volcano, Japan. Bulletin of Volcanology 64, 55–68.

Hosono, K., Osada, H., Saegusa, K., Takano, T., 2009. Microwave propagation loss in aground crack larger than a wavelength. Proc. the 2008 IEICE General Conference(B-2-3).

Ishii, M., Saito, M., Fujii, F., Hojo, J., Matsui, M., Itamoto, N., Shinjo, K., 2005. LEMP fromlightning discharges observed by JLDN. IEEJ Transactions on Power and Energy 125(8), 765–770.

James, M.R., Wilson, L., Lane, S.J., Gilbert, J.S., Mather, T.A., Harrison, R.G., Martin, R.S.,2008. Electrical charging of volcanic plumes. Space Science Reviews 137 (1–4),399–418. http://dx.doi.org/10.1007/s11214-008-9362-z.

Kawanishi, T., Sezai, T., Ito, Y., Imaoka, K., Takeshima, T., Ishido, Y., Shibata, A., Miura,M., Inahata, H., Spencer, R.W., 2003. The advanced microwave scanning radiometerfor the earth observing system (AMSR-E), NASDA's contribution to the EOS for globalenergy andwater cycle studies. IEEE Transactions on Geoscience and Remote Sensing41 (2), 184–194.

Maeda, T., Takano, T., 2008. Discrimination of local and faint changes from satellite-bornemicrowave radiometer data. IEEE Transactions on Geoscience and Remote Sensing46 (9), 2684–2691.

Maeda, T., Takano, T., 2010. Detection algorithm of earthquake-related rock failuresfrom satelliteborne microwave radiometer data. IEEE Transactions on Geoscienceand Remote Sensing 48 (4), 1768–1776.

Maki, K., Takano, T., 2004. Analysis of the microwave radiation field from the dischargedue to an impact destruction (in Japanese). National Convention of the Institute ofElectrical Engineers of Japan (IEEJ), p. 87 (1-081).

Maki, K., Takano, T., Soma, E., Ishii, K., Yoshida, S., Nakatani, M., 2006. An experimentalstudy of microwave emission from compression failure of rocks. Journal of theSeismological Society of Japan 58 (4), 375–384.

Massonnet, D., Rossi, M., Carmona, C., Adragna, F., Peltzer, G., Feigl, K., Rabaute, T., 1993.The displacement field of the Landers earthquake mapped by radar interferometry.Nature 364, 138–142.

Molchanov, O., Hayakawa, M., 1998. Subionospheric VLF signal perturbations possiblyrelated to earthquakes. Journal of Geophysical Research 103 (17), 489–504.

Nitsan, U., 1977. Electromagnetic emission accompanying fracture of quartz-bearingrocks. Geophysical Research Letters 4, 333–336.

O'Keefe, S.G., Thiel, D.V., 1995. A mechanism for the production of electromagneticradiation during fracture of brittle materials. Physics of the Earth and PlanetaryInteriors 89, 127–135.

Ohnishi, H., Chiba, S., Soma, E., Ishii, K., Maki, K., Takano, T., Hasegawa, S., 2007. Studyonmicrowave emissionmechanisms on the basis of hypervelocity impact experimentson various targat plates. Journal of Applied Physics 101 (9) (124901-1-8).

Sagiya, T., 2012. Suggestions for the future of seismology from the temporary committee(summary). Japan Geoscience Union Meeting 2012 (U06-01).

Takano, T., Maeda, T., 2009. Experiment and theoretical study of earthquake detectioncapability by means of microwave passive sensors on a satellite. IEEE Geoscienceand Remote Sensing Letters 6 (1), 107–111.

Takano, T., et al., 2008. Detection of microwave frequency signals from earthquakesand volcanic activities. AGU Fall Meeting 2008 (S13C-1837).

Takano, T., et al., 2009. Signal detection of microwave frequency due to rock fracture involcanic activities. Geophysical Research Abstracts 11 (EGU2009-9176).

Takano, T., Ikeda, H., Maeda, T., 2010a. Consideration of the mechanism of microwaveemission due to material destruction. Journal of Applied Physics 108, 1–5 (083722).

Takano, T., Ikeda, H., Sugita, S., Maeda, T., 2010b. Consideration on the mechanismof microwave emission due to rock fracture. Geophysical Research Abstracts 12(EGU2010-6398).

Takano, T., Maki, K., Soma, E., Hattori, K., Maeda, T., 2011. Fundamentals of microwavedetection in association with rock fracture and experimental data for various rocks.Geophysical Research Abstracts 13 (EGU2011-9638).

Ukawa, M., Fujita, E., Yamamoto, E., Okada, Y., Kikuchi, M., 2000. The 2000 Miyakejimaeruption: crustal deformation and earthquakes observed by the NIED Miyakejimaobservation network. Earth Planets Space 52 (8), xix–xxvi.

Uyeda, S., et al., 2012. Proposal for earthquake prediction program I: short-term prediction.Japan Geoscience Union Meeting 2012 (U06-03).

Varotsos, P., Alexopoulos, K., Lazaridou, M., 1993. Latest aspects of earthquakes predictionin Greece based on seismic electric signals, two. Tectonophysics 224, 1–37.

Yasukawa, H., Takumi, I., Hata, M., 2009. Signal detection technology on earth diastrophismusing electromagnetic wave observation. International Symposium on IntelligentSignal Processing and Communication Systems 2009, pp. 473–476.

Yoshida, S., Ogawa, T., 2004. Electromagnetic emissions from dry and wet graniteassociated with acoustic emissions. Journal of Geophysical Research 109, B09204.http://dx.doi.org/10.1029/2004JB003092.