著者
平野 史朗 川方 裕則 土井 一生
出版者
日本地球惑星科学連合
雑誌
日本地球惑星科学連合2019年大会
巻号頁・発行日
2019-03-14

Revisiting earthquake catalogs have revealed that 40% or more of major earthquakes are accompanied by foreshock activities, at least in California [Abercrombie & Mori 1996 Nature] and Japan [Tamaribuchi et al. 2018 EPS].To investigate whether the foreshocks are magnification and activation of background seismicity, we have to compare waveforms due to the foreshocks and background events that might be sometimes uncataloged because of their small sizes.We can mine even small seismic signals similar to some template waveforms from continuous waveform records by using a matched-filter analysis based on cross-correlation coefficients (CC) between the template waveforms and continuous records.However, in the conventional analysis, we have to define a threshold of CC to detect similar seismic waveforms, which have been chosen subjectively and empirically.Then, we propose a threshold-free method to detect outliers from the empirical distribution of CC among seismic waveforms.In our framework, empirical distributions of the coefficients are modeled by the theory of extreme value statistics, and the detectability is automatically determined from Akaike's Information Criterion (AIC), depending on data.We applied the method of seismic signal detection to 2-years-continuous records before an M5.4 earthquake in Nagano, Japan (June 30, 2011) that followed 27 foreshocks cataloged by JMA.First, we found that the empirical frequency distribution of CC between the continuous records and foreshocks did not follow a normal distribution, which means that we cannot estimate the possibility of a false positive by assuming the normal distribution as a model.Instead, we also found that the maximum value of CC in every few seconds follow the Gumbel distribution after elimination of some outliers.The elimination can be achieved by comparing AICs of data including and excluding the outlier candidates.Given this method, we found a similar event ~2 months before the mainshock and 3 similar events 3-4 days before the mainshock.This result implies that the foreshocks were not similar to background events, and hence, not magnification or activation of them.
著者
駒形 亮太 平野 史朗 川方 裕則 直井 誠
雑誌
JpGU-AGU Joint Meeting 2020
巻号頁・発行日
2020-03-13

地震波形には、特定地域の震源特性や地下構造などの情報が豊富に含まれている。地震カタログに載らない微小地震による地震波を捉えることができれば、もちろん前震などの地震活動の特徴を捉えることにも有効であるが、地下構造をより詳細に知ることにも有効である。地震カタログに載らない微小地震による地震波を捉える方法はいくつか存在する。多く用いられている方法は、既知のテンプレート波形を用いた、相関係数によるテンプレートマッチングである(例えば、Doi and Kawakata, 2012; Kato et al., 2012)。しかし、この方法では既知の波形の類似波形を見つけることに他ならず、既知の波形を用いない解析を行えば、さらに多くの微小地震による地震波が捉えられる可能性がある。そこで、本研究ではテンプレート波形を用いずに高速な類似波形検出を行えるハッシュ法に注目した。Yoon, et al. (2015)はLocality Sensitive Hashing(LSH)を用いたハッシュ法の一種である、Fingerprint and Similarity Thresholding (FAST)という類似地震波形検出手法を提案した。FASTはBaluja&Covell.(2008)により音声検索において有効性が認められているWaveprintという手法を元に開発された。FASTは主に2つの構成要素から成り、1つ目はLSH特性を持つハッシュ関数による波形の特徴抽出、2つ目は波形ペアのハッシュ値の類似性であるJaccard係数を近似的に評価する類似検索である。そして、FASTの LSHとして開発されたFingerprintingを用いることで、地震カタログに載らない小さな類似波形のペアも見つけられることが確認されている。しかし、FASTはspectrogram 生成やWavelet変換という複雑な処理を行う必要がある。本研究では、より少ない手順で計算可能な新たなハッシュ関数を2つ提案し、それらとFASTの3手法による類似波形ペアの検出を行った。そして、検出された類似波形ペアの類似度と相関係数の関係、計算実行に要する時間の2点を中心にそれらの性能比較を行った。 連続波形記録のspectrogramを見ると、一般に地震波形は一時的な高エネルギーイベントとして現れる。FASTはこの特性を利用しており、spectrogram生成やHaarWavelet変換を必要とする。一方、時系列で連続波形記録を見ると、一般に地震波形は周囲の常時微動に比べ、振幅が大きくなることが期待される。そこで、本研究ではこの特性に着目し、FASTとは異なり、連続波形記録の時系列情報そのままで計算できるようなハッシュ関数を設計した。1つ目は、Fei et al. (2015)によって画像検出のために提案されたaHashを地震波検知のために改造し、常時微動と地震波の識別に強くした2bit-aHash、2つ目は連続波形記録から切り出した波形ウィンドウ振幅の絶対値の順位でハッシュ値を定める全く新しい手法のkHashである。2bit-aHashは時系列の振幅の後続N個の平均値と標準偏差を用いる。平均値±標準偏差から逸脱している正の振幅には10、負の振幅には01を対応させ、それ以外をノイズとして00に対応させる。kHashは、波形ウィンドウの時系列の振幅の絶対値上位k%の正の振幅には10、負の振幅には01を対応させ、それ以外をノイズとして00に対応させる。解析データとして、長野県中部で発生した Mj5.4の地震発生を含む2011年6月29日19:00~2011年6月30日18:59(JST)のHi-net松本和田観測点で記録された連続速度波形記録を用いた。結果として、2bit-aHash、kHashは地震カタログに載っていない、地震波のような類似波形ペアを検出することに成功した。そして2bit-aHash、kHashは検出波形波形ペア間の相関係数が高く、類似度の高さと相関係数の高さに相関がみられた。一方、FASTは複雑な処理を行う割に、検出される類似波形ペア間の相関係数が他の2手法に比べてばらつきが大きいことが分かった。加えて全体の実行時間は2bit-aHash、kHash共にFASTよりも約4~5倍高速であった。理由の一つとして、特徴抽出の際2bit-aHash、kHashがFASTよりも簡単に計算できることが挙げられるが、それだけでなく、類似検索の実行時間も数十倍高速になっていることが判明した。Yoon, et al. (2015)によると、解析に用いるデータのサンプル数nが非常に長くなれば類似検索の実行時間がO(n2)に近づいていくが、本研究の結果は2bit-aHashやkHashが類似検索時の実行時間の大きな改善にも貢献することを意味する。類似検索アルゴリズムではノイズ同士のペア、ノイズとイベント波形のペアなど、極端にJaccard係数が低い波形ペアはアルゴリズム内での類似度が定義できず、データベースから削除される。プロセスを精査したところFASTではこのような無駄なペアが削除されずに多数存在してしまうことが明らかになり、これを類似波形の候補として全て保持・検索しなければならないことが速度低下を招いていたものと考えられる。逆に2bit-aHash、kHashで出力された波形ペアのハッシュ値のJaccard係数はFASTのものよりも全体的に低く、無駄なペアがかなり削除されたため、速度が向上した。以上より、今回新たに提案した2bit-aHash、kHashの2手法はFASTよりも実行時間が特徴抽出だけでなく類似検索においても高速になり、また検出波形ペア間の相関係数が高くなりそのばらつきが少なくなることが示された。
著者
小笠原 宏 川方 裕則 石井 紘 中谷 正生 矢部 康男 飯尾 能久 南アフリカ金鉱山における半制御地震発生実験国際共同研究グループ
出版者
公益社団法人 日本地震学会
雑誌
地震 第2輯 (ISSN:00371114)
巻号頁・発行日
vol.61, no.Supplement, pp.563-573, 2009-07-31 (Released:2013-11-21)
参考文献数
54
被引用文献数
2 3

Experimental sites with potential earthquakes up to M ∼ 3 in coming few years are known beforehand from mining schedule at 2-3 km depths in South African gold mines, which allows us to deploy various borehole instruments including Ishii strainmeters, geophones, accelerometers and AE sensors. Deployment of these wide-dynamic-range and high-resolution observations in the past 15 years has led to many findings about the earthquake rupture and its preparation stage. High-sampling seismograms obtained at close proximity of M > 1 earthquakes have demonstrated similarities of these earthquakes to natural, greater earthquakes in many aspects, including stress drop, energy efficiency, and complexity of rupture propagation. Some of larger mine earthquakes are preceded by perceivable abnormal seismicity. However, no immediate precursors for earthquakes with M ∼ 2 were observed by our high-resolution strain and AE sensors installed within the dimension of mainshock rupture. In contrast, aseismic strain-step events that we had recently discovered were sometimes preceded by further slower forerunners. Ongoing projects bring in novel technologies such as field-scale AE monitoring and fast-response strainmeters, and novel targets including mines being flooded for closing operation.
著者
福山 英一 山下 太 徐 世慶 溝口 一生 滝沢 茂 川方 裕則
雑誌
JpGU-AGU Joint Meeting 2020
巻号頁・発行日
2020-03-13

We have been conducting meter-scale rock friction experiments using the large-scale shaking table at NIED since 2012. We have completed 5 series of experiments, each of which included about 20 experiments. One of the purposes of these experiments was to investigate the spatial scaling of the friction since the friction laws we use today were derived from centimeter-scale experiments. Another purpose was to monitor rupture evolution and local stress field using near-fault high-resolution measurements. In this talk, we will showcase some key results derived from our rock friction experiments.Regarding the spatial scaling of friction, we recognized that the local frictional strength was not uniform on the fault and its spatial variation had a significant impact to the macroscopic frictional strength (Yamashita et al., 2015). In addition, the scaling behavior seems different between rock-on rock friction and that with a gouge layer. In the rock-on-rock case, gouge generation changes the strength in space. But if the gouge layer already exists, strength depends on the rearrangements of the gouge particles (Yamashita et al., 2018).Regarding rupture evolution on laboratory fault, we pointed out a previously overlooked difficulty in direct measuring the two-dimensional (2D) evolution of the rupture front. Under very special condition, we could overcome this difficulty by installing 2D strain gauge arrays inside the rock sample. We found that the free surface effects at both edges of the fault had a significant effect on rupture nucleation (Fukuyama et al., 2018). In addition, the strain behavior close to the fault edge might not be the same as that on the fault, even if the sensors were installed within 10 mm away from the fault. Using numerical simulations, we could reproduce the observed strain data by extrapolating a simple friction behavior on the fault surface, suggesting that the way of deriving the friction law needs to be revised (Xu et al., 2019).We also discovered some interesting fault behaviors during our experiments. By changing loading rate or fault surface condition, we could frequently reproduce super shear rupture events in the laboratory, which were thought to be rare in nature. By investigating the cohesive zone length of the rupture front in the supershear regime, we showed that the experimental results could reach a good match with one of the theoretical predictions Fukuyama et al. (2017). Moreover, we observed slow slip events with supersonic propagation velocity during some experiments (Fukuyama et al., 2019), whose interpretation is still underway.The above results bridge the gap between the traditional small-scale lab experiments and the field observations, and can be useful for improving our understandings of fault rheology and earthquake physics.
著者
舩木 伸江 河田 惠昭 矢守 克也 川方 裕則 三柳 健一
出版者
自然災害科学会
雑誌
自然災害科学 (ISSN:02866021)
巻号頁・発行日
vol.24, no.4, pp.447-471, 2006
参考文献数
58

In Japan, there is concern that great earthquake disasters, in Tokai, Tonankai, Nankai and the Tokyo Metropolitan area, could occur within the next few decades. Once one of these disasters happens, a larger number of deaths than in the 1995 Great Hanshin-Awaji Earthquake Disaster, which killed more than 6,000 people, could possibly occur. Therefore, it is necessary to find an early solution to the problem of mortuary care and cremation of deceased people after large-scale disasters. However, there has not yet been enough discussion about how to deal with, bury, and cremate dead bodies. This study first sorts out several problems related to mortuary care and cremation by examining 34 documents of the Great Hanshin-Awaji Earthquake Disaster. Next, it identifies remaining problems after the Great Hanshin-Awaji Earthquake Disaster. Third, it analyzes new issues related to the mortuary care and cremation when largescale disasters occur. Finally, several important findings are provided for improving present problems in the Japanese system of mortuary care and cremation.
著者
辻村 優志 川方 裕則 福山 英一 山下 太 徐 世慶 溝口 一生 滝沢 茂 平野 史朗
出版者
日本地球惑星科学連合
雑誌
日本地球惑星科学連合2016年大会
巻号頁・発行日
2016-03-10

For inland earthquakes such as the 2007 Noto Hanto earthquake (Doi and Kawakata, 2013) and the 2008 Iwate-Miyagi earthquake (Doi and Kawakata, 2012), foreshocks were reported to occur in the vicinity of main shock hypocenter. Moreover, for interplate earthquakes such as the 2011 off the Pacific coast of Tohoku earthquake (Kato, et al., 2012) and 2014 Iquique earthquake in Chile (Yagi et al., 2014), migration of foreshocks toward the main shock hypocenter was detected in one month before the main shock. In order to understand the generation mechanism of foreshocks, it is important to investigate under what environments foreshocks occur.Since 2012, stick-slip experiments have been carried out using a large-scale biaxial friction apparatus at NIED (e.g., Fukuyama et al., 2014). Based on the experimental result that foreshocks were detected only in the later period of each run, Kawakata et al. (2014) suggested that the foreshocks occur only after the generation of gouge. In this study, we carried out a series of stick-slip experiments with and without pre-existing gouge along a fault plane to confirm if fault gouge affects the foreshock activity. When foreshocks are detected, we estimate the hypocenter locations of foreshocks.We used two rectangular metagabbro blocks to make the simulated fault plane, whose dimension was 1500 mm long and 500 mm wide. The experiments were conducted under normal stress of 1.33 MPa and loading speed of 0.01 mm/s up to approximate slip amount of 8 mm. During each experiment, we continuously measured elastic waves to detect foreshocks. The sensor distribution is shown in the figure below. Gouge materials were prepared naturally during preceding experiments whose sliding speed was as high as 1 mm/s.To roughly detect foreshock activity, we calculated cumulative amplitude of continuous waveform data every 0.01 seconds. During an experiment without pre-existing gouge materials (LB13-004), a few foreshocks were detected. On the other hand, during an experiment with pre-existing gouge materials (LB13-007), much more foreshocks were detected. Then we estimated hypocenters of foreshocks for a stick-slip event (event 44) in LB13-007. Although the initial phases of the main shock were contaminated due to the coda wave signals of preceding foreshocks, the hypocenter of the main shock was roughly estimated near the right end of the fault plane. Foreshocks began to occur in the left half of the fault plane, but most of later foreshocks occurred near the right end.Therefore, we confirmed that foreshock activity was high when gouge materials were present along a fault plane, and found a similar hypocenter migration of foreshocks toward the main shock hypocenter, which was reported for interplate earthquakes.In the future, we shall examine the data obtained from other experiments to confirm if the aforementioned features are common.Acknowledgments: This work was supported by NIED research project “Development of monitoring and forecasting technology for crustal activity” and JSPS KAKENHI Grant Number 23340131.
著者
豊本 大 川方 裕則 平野 史朗 土井 一生
出版者
日本地球惑星科学連合
雑誌
日本地球惑星科学連合2016年大会
巻号頁・発行日
2016-03-10

Recently, small foreshocks have been frequently detected using a cross-correlation technique (e.g., Bouchon et al., 2011, Science). For inland earthquakes, foreshocks whose hypocenters were estimated to be adjacent to the mainshock hypocenter were detected from several tens of minutes before the main shock occurrence (Doi and Kawakata, 2012, GRL; 2013, EPS). Toyomoto et al. (2015, SSJ) tried to detect foreshocks of an M 5.4 earthquake in central Nagano prefecture on June 30, 2011, in a similar manner to Doi and Kawakata (2013). Using the continuous waveforms of the vertical component at N.MWDH (Hi-net) station (the epicentral distance of the mainshock is 4.5 km), they newly detected 14 foreshocks with a cross-correlation criterion of 0.6, in addition to 27 foreshocks listed in the JMA (Japan Meteorological Agency) unified hypocenter catalogs. To efficiently detect small foreshocks for other inland earthquakes, it is necessary to design how to set the cross-correlation detection criterion for foreshock detection.In this study, we carried out foreshocks detection of the same earthquake in the same method as Toyomoto et al. (2015, SSJ) using the waveform record of N.MNYH (Hi-net) station (epicentral distance of main shock is 5.3km). In this case, the maximum correlation coefficients during one minute tended to be higher than those for records at N.MWDH station, and the result of detection strongly depends on a threshold employed in a cross-correlation method. This indicates that we should not use a universal threshold independent of data. One of alternative way is to use the standard deviation of cross-correlation coefficients. Then, we made a histogram of the cross-correlation coefficients of 1-day data. The histogram of N.MWDH data is Gaussian and the cross-correlation coefficients obey a normal distribution with the average of zero. Although the histogram of N.MNYH data is not Gaussian, so the cross-correlation coefficients have a large-deviation. In such a case, a criterion depending on the standard deviation is inadequate.Acknowledgments:We used continuous waveform records of NIED high-sensitivity seismograph network in Japan (Hi-net) and the JMA unified hypocenter catalogs.
著者
米田 直明 川方 裕則 平野 史朗
出版者
日本地球惑星科学連合
雑誌
日本地球惑星科学連合2016年大会
巻号頁・発行日
2016-03-10

It has been reported that the b-value decreases prior to large earthquakes in nature (e.g., Imoto, 1991) and failure of a rok sample in laboratories (e.g., Scholz, 1968) . To discuss a temporal variation of the b-value, a sufficient number of earthquakes is required. In general, calculation of b-value prior to large earthquakes requires long-term data because seismic activity is not always high at that term. In other words, the temporal resolution of b-value variation before a large earthquake is usually low. Therefore, sufficiently high seismic activity before the large earthquake is required to evaluate the b-value variation precisely.For example, two major earthquakes occurred in northern Tochigi Prefecture: Mj6.3 in 2013 and Mj5.1 in 2014. The two events followed the increase of seismic events. One possible cause of this increase is the Mw9.0 Tohoku earthquake in 2011 (e.g., Aketagawa, 2011).In this study, we try to detect the temporal variation of the b-value in northern Tochigi Prefecture where a large number of earthquakes could be observed in a short period prior to the two major events. First, to increase the temporal resolution, we calculate the b-value for a circular region with 20km radius from the epicenter of the Mj6.3 event; the result is shown in Figure A. While the b-value was greater than 1.0 and stable before March 2011, it dramatically decreased to ~0.6 after the occurrence of the Tohoku earthquake in 2011 and recovered to around 1.0 almost within one year. After that, it decreased to ~0.7 again following the Mj6.3 event in 2013 and recovered to ~1.0 within a small period. Although it decreased to ~0.75 again following the Mj5.1 event in 2014, it did not recover but continued, at least, one year. Regarding these different variations in each sequence, we considered the seismic activity in northern Tochigi precisely. We consider regions 1, 2, and 3. The region 1 is located south of the source region of the Mj6.3 event and includes an active fault. The regions 2 and 3 include the source areas of the Mj6.3 and Mj5.1 events, respectively. The temporal variation of b-value for each region is shown in Figure B, C, and D. In region 1, constant seismic activity has continued for the whole term and the b-value was stable and greater than 1.0. The b-values are also stable but ~1.0 in region 2 and ~0.75 in region 3. On the basis of these results, we found that the temporal variation of the b-value of the entire region is affected by the temporarily activated one of the three regions. However, in regions 2 and 3, the numbers of events to calculate the b-value precisely are insufficient despite their activation. So we found that we cannot detect temporal variation of the b-value prior to the major events. This finding tells us that we need to consider the target region carefully when we research the temporal variation of the b-value.AcknowledgmentsIn this study, we used the JMA unified hypocenter catalogue.
著者
植村 美優 川方 裕則 平野 史朗
出版者
日本地球惑星科学連合
雑誌
日本地球惑星科学連合2016年大会
巻号頁・発行日
2016-03-10

On the basis of experimental studies (e.g. Yoshimitsu et al., 2009 and Lockner et al., 1977), it has been expected that seismic velocity decreases prior to earthquakes. To detect temporal variation in the velocity, stable monitoring of the velocity for a long time is required. Seismic interferometry using micro-tremors is one of the potential techniques which enable us to detect such variation if seismic stations are densely located. With a seismic interferometry technique, some researchers have tried to detect the velocity variation before and after an earthquake using seismograms of a station pair whose interval was longer than ~20 km, but remarkable variation preceding target earthquakes have never been reported. If we can use seismograms of a station pair with a shorter interval, we might be able to detect the variation. In this study, we chose the 2014 Nagano Kamishiro Fault Earthquake (Mj 6.7) as a target, whose source fault (Kamishiro fault) is located between two NIED Hi-net seismic stations (N.HBAH and N.HKKH). The interval of these stations is about 7.3km.At first, we investigated how frequency contents of micro-tremors depend on time, such as day or night, weekday or weekend. After checking, we confirmed that seismograms on Saturday night are the best for our analysis. After applying one-bit normalization, we divided continuous seismograms into one-minute seismograms. Then, we calculated the cross-correlation function of each one-minute seismograms pair of two stations, and stacked all cross-correlation functions for a period of six hours, on Saturday night. Finally, we obtained stacked cross-correlation from 2011 to 2015.We found obvious and pulse-like phases around -2s, from which we estimate apparent seismic velocity ~3.5km/s. Further, we found the increase and decrease in velocity during two years before the earthquake. However, the variation of average velocity is as large as 10%, and we cannot find any corresponding phase in positive time. Moreover, we could not find any coseismic variation. It is suggested that distribution of the micro-tremor sources is anisotropic and asymmetric in space and unstable in time even though we focused only on November and December for every year. Consequently, if we try to detect the structure variation around a seismic source fault, we should confirm that the spatio-temporal distribution of the micro-tremors source does not change.Acknowledgments: We used continuous waveform records of NIED high-sensitivity seismograph network in Japan (Hi-net).
著者
若松 修平 川方 裕則 平野 史朗
出版者
日本地球惑星科学連合
雑誌
日本地球惑星科学連合2016年大会
巻号頁・発行日
2016-03-10

火山性地震には多くの種類が存在する。しかしながらその分類方法は統一されておらず、研究者や火山によって分類方法や基準は異なっている。火山性地震のどの部分に着目して分類するかは筆者によって様々であるし、また同じ名称で分類していても、分類の基準となる値が違うこともある(西村・井口,2006)。以上のように、文献によって火山性地震の分類基準・名称が異なると、混乱を招く。西村・井口(2006)は、「このような基準のずれや分類項目の複雑さは、火山性地震や微動の研究を分かりづらくしている理由の一つである」と述べている。したがって、火山性地震の分類方法を統一することは重要であると言える。そのためには、火山性地震の連続波形データから、十分な量の火山性地震を抽出する必要がある。 しかしながら、火山に設置された地震計には、火山性地震による揺れ以外にも、人の生活に起因する揺れが観測される可能性がある。そのため、火山性地震とそれ以外の揺れを分け、火山性地震のみを抽出するためには、データ毎に火山性地震かそれ以外のことに起因する揺れかということを判断し、火山性地震以外の揺れをノイズとして取り除いていく、という作業が必要となる。これらの作業を行うためには、火山性地震が検出できるよう、火山活動が活発となっている時期がある火山を研究対象とする必要がある。また人の生活に起因するノイズについて解析することで、それらのノイズの強さや周波数を推定でき、火山活動が活発な時期でも人の生活に起因するノイズを発見・除去しやすくなると考えられるため、火山活動が活発ではない時期もある火山が望ましい。 以上の点から、本研究では、2015年4月~9月に火山活動が活発化した箱根火山を研究対象とし、気象庁が公開している二ノ平観測点上下成分の連続波形記録を用いて、火山性地震ではないと思われる揺れを検出・除去できるよう試みた。 二ノ平観測点のデータには、観測点の近くにある彫刻の森駅を発着している電車の波形が記録されていた。以上のことは、彫刻の森駅に電車が発着していた時間と、揺れが発生していた時間で同じであったことから推定できた。 これら電車の発着による揺れの検出について、電車の波形のテンプレートを複数選出し、これらテンプレートと観測波形で似ている部分を検出することを試みた。具体的にはまず、2015年3月29日午前5時~午前9時までの間に発着した20回分の電車による波形を54個に分けたものをテンプレートとした。これらのテンプレートのエンベロープを計算して位相情報をなくした後、時間窓1秒、ずれ0.2秒で移動平均を計算することでスムージングを行なった。これらの処理を観測波形全体にも施し、テンプレートと観測エンベロープ相関をとった。この相関を用いて、電車波形を検出する方法について検討した。 以上の処理を2015年3月29日のデータに適用したところ、合計で116回あった電車による揺れのうち、112回は検出できた。また、24時間のうち、約300秒は電車が来ていない時間帯にも関わらず電車による揺れとして検出された。これらの処理を火山活動が活発化していた時期に適用することで、火山性地震を検出しやすくなることが期待される。また、以上の処理を2015年6月29日のデータに適用したところ、電車が彫刻の森駅に発着している時間にも関わらず、電車波形として検出されなかった部分があった。これらの部分には、火山性の地震波と思われるシグナルが卓越していた。これは、火山性地震の検出という本研究の目的とてらして、成功したといえる。謝辞:本研究にあたり、気象庁火山観測網のデータを使用させていただいた。