Search

We propose a Bayesian approach for jointly determining the locations of earthquake hypocentres and velocity structures in a region divided by a structural boundary, such as a fault zone. To accommodate the Markov chain Monte Carlo (MCMC) method in a region containing lateral velocity contrast, we incorporated a data-driven station clustering scheme when estimating earthquake hypocentre locations, a 1-D velocity model assigned to each cluster and station corrections for traveltimes. The proposed method realizes the clustering of stations reflecting velocity structures beneath the station network in the framework of the MCMC method. We verified the performances of the proposed method using numerical experiments with different distributions of earthquake hypocentres relative to the structural boundary. The results demonstrated that the proposed method reconstructed earthquake hypocentre locations and velocity structures involving lateral contrast robustly to earthquake distributions. Additionally, the number of 1-D velocity models (the number of station clusters) was optimized by comparing resultant posterior probability proportional to distances between the observed and theoretical traveltimes. We also applied the proposed method to estimate the earthquake hypocentre locations and velocity structures around the source area of the 2004 Mid-Niigata prefecture earthquake, Japan, to examine the applicability of the proposed method to real observation data. The stations were divided into east and west clusters, coinciding with the lateral velocity variation related to the geological setting. Therefore, the proposed method accurately determines earthquake hypocentres in a region divided by a structural boundary without any knowledge regarding the geological structure. [ABSTRACT FROM AUTHOR]

Crustal stress is a crucial parameter for understanding tectonics and assessing the occurrence of large earthquakes. Stress at seismogenic depths over a wide area has been inferred from focal mechanism solutions. The existence of focal mechanism solutions determines the area that can be potentially studied. Therefore, we studied the crustal stress orientations in Japan from focal mechanism solutions of small earthquakes. First, we obtained about 220,000 focal mechanisms using the P‐wave first‐motion polarities picked by a deep learning model. Next, we performed a linearized inversion on the nationwide stress orientation model at a grid of 0.2‐degree intervals. The estimation errors reflect the quality and quantity of focal mechanism solutions. Overall, the results indicate that Japanese islands are under an east–west compressional stress, whereas some areas along the Pacific coast of northeast Japan are under east–west extensional stress. A closer look shows the spatial trends of stress orientation may vary near the geological borders, including the Hinagu and Futagawa faults in Kumamoto Prefecture (Kyushu Island) and the Median Tectonic Line in Shikoku Island. Our stress model also revealed the existence of stress anomalies around Cape Shionomisaki in the Kii peninsula. Most of the active faults are at approximately optimal orientation for the derived stress orientation and stress ratio, although uncertainties in the fault geometry, particularly the dip angle, substantially affect this evaluation. The stress map helps understand the overall vision of the crustal stress in Japan and further study the local stress distribution. Plain Language Summary: Stress, an internal force of the Earth, causes earthquakes and crustal deformation and forms characteristic landforms. Therefore, stress is a key to understanding the dynamics of the Earth. This study uses one of the features of earthquakes caused by crustal stress as an indicator. This feature is the "focal mechanism" corresponding to the orientation of the fault plane and slip direction on the fault. First, we derived numerous focal mechanism solutions for small earthquakes in Japan. The massive seismic waveform data processing was supported by a deep learning model we previously developed. The stress orientations consistent with the focal mechanism solutions were then estimated. The high‐resolution (about 20 km) nationwide stress map shows a general compression trend in the east–west direction, but in some spots around the Pacific coastal area of northeastern Japan, the compression trend is in the north‐south direction. In addition, it was found that the stress orientation changes along geological boundaries such as active faults and major tectonic lines. Furthermore, spots where the stress orientation differs from the surrounding area were identified. This stress map will be indispensable for future earthquake assessments and understanding of the formation mechanism of the Japanese islands. Key Points: We generated a nationwide crustal stress map with 0.2‐degree grids from focal mechanism solutions of about 220,000 microearthquakesWe observed a stress anomaly spot and stress orientation changes along geological borders, such as faults and tectonic linesMost of the active faults are almost optimally oriented to the estimated stress field [ABSTRACT FROM AUTHOR]

We present a deep‐learning approach for earthquake detection using waveforms from a seismic array consisting of multiple seismographs. Although automated, deep‐learning earthquake detection techniques have recently been developed at the single‐station level, they have potential difficulty in reducing false detections owing to the presence of local noise inherent to each station. Here, we propose a deep‐learning‐based approach to efficiently analyze the waveforms observed by a seismic array, whereby we employ convolutional neural networks in conjunction with graph partitioning to group the waveforms from seismic stations within the array. We then apply the proposed method to waveform data recorded by a dense, local seismic array in the regional seismograph network around the Tokyo metropolitan area, Japan. Our method detects more than 97% of the local seismicity catalog, with less than 4% false positive rate, based on an optimal threshold value of the output earthquake probability of 0.61. A comparison with conventional deep‐learning‐based detectors demonstrates that our method yields fewer false detections for a given true earthquake detection rate. Furthermore, the current method exhibits the robustness to poor‐quality data and/or data that are missing at several stations within the array. Numerical experiments using subsampled data demonstrate that the present method has the potential to detect earthquakes even when half of the normally available seismic data are missing. We apply the proposed method to analyze 1‐h‐long continuous waveforms and identify new seismic events with extremely low signal‐to‐noise ratios that are not listed in existing catalogs. We also show the potential portability of the proposed method by applying it to seismic array data not used for the training. Key Points: We develop a convolutional neural network (CNN) for automated earthquake detection using waveforms from a nonequispaced seismic arrayStations with similar waveforms are clustered via graph partitioning, substantially improving the CNN‐based earthquake detection accuracyThe proposed CNN detects small‐magnitude earthquakes with very low signal‐to‐noise ratios that were not listed in existing catalogs [ABSTRACT FROM AUTHOR]

We estimate the velocity structures of P (Vp) and S waves (Vs) in southern Hokkaido, Japan, where a large number of crustal low‐frequency earthquakes (LFEs) have occurred. To investigate the relationship between structure and crustal seismicity, we compare the obtained velocity heterogeneity between the crustal LFE regions, which are characterized by active volcanoes, Quaternary volcanoes, and seismic swarms. The tomographic results show that low‐velocity zones (LVZs) with high Vp/Vs are located around the Moho discontinuity beneath active volcanoes, and the LVZs extend horizontally from beneath the active volcanoes to other LFE regions. In southern Hokkaido, crustal LFEs also appear inside and below the upper crust as well as around the Moho discontinuity. These intermediate‐depth and deep LFEs are located around the LVZs with a high Vp/Vs ratio, indicating the contribution of partial melt to the LFEs genesis, even at a distance from the active volcanoes. At intermediate depths of the crust, LFEs are distributed even in the high‐velocity zones, adjacent to the LVZs. This indicates that the partial melt migrations within the crust affect the activity of LFE at the intermediate depths. Below the seismic swarms in the upper crust, LVZs with crustal LFEs are distributed, probably owing to upward migrations of fluid originating from the solidification of the partial melt; this indicates a link between the shallow and deep crustal seismicity in southern Hokkaido. Key Points: Constructing detailed P and S wave velocity structures in southern Hokkaido, JapanPartial melt and their movement are related to the genesis of crustal low‐frequency earthquakesShallow and deep crustal seismicity are linked through fluid and melt migrations [ABSTRACT FROM AUTHOR]

We investigate P-wave attenuation, Q p − 1 , in the Pacific slab beneath northeastern (NE) Japan, adopting for the first time the spectral ratio technique for intraslab earthquakes. When seismograms of two earthquakes are recorded at a station and their ray paths to the station are largely overlapped, station-dependent amplification and structural effects on the overlapped rays can be canceled out from the ratio of the spectral amplitudes of the seismograms. Therefore, adopting the spectral ratio technique for intraslab earthquakes has a great advantage for the precise evaluation of Q p − 1 in the slab because the structural effects above the slab, including the high-attenuation mantle wedge, are removed. For estimating the intraslab Q p − 1 , we determined corner frequency of the intraslab earthquakes using the S-coda wave spectral ratio as the first step. Then, we evaluated the inter-event path attenuation, Δ t ⁎ , from the ratio of the spectral amplitudes of P waves. The obtained result shows that P-wave attenuation in the Pacific slab marks Q p − 1 of 0.0015 ( Q p of ∼670) at depths of 50–250 km. This indicates that the P-wave attenuation in the Pacific slab is weaker than that in the mantle wedge. The relatively high- Q p − 1 is correlated with the distributions of intraslab earthquakes, suggesting that the P-wave amplitude is more attenuated around active seismicity zones in the slab. Therefore, our observations likely indicate the presence of fractures, hydrous minerals, and dehydrated fluid around seismogenic zones in the slab at intermediate depths. [ABSTRACT FROM AUTHOR]

The 1906 Colombia-Ecuador earthquake induced both strong seismic motions and a tsunami, the most destructive earthquake in the history of the Colombia-Ecuador subduction zone. The tsunami propagated across the Pacific Ocean, and its waveforms were observed at tide gauge stations in countries including Panama, Japan, and the USA. This study conducted slip inverse analysis for the 1906 earthquake using these waveforms. A digital dataset of observed tsunami waveforms at the Naos Island (Panama) and Honolulu (USA) tide gauge stations, where the tsunami was clearly observed, was first produced by consulting documents. Next, the two waveforms were applied in an inverse analysis as the target waveform. The results of this analysis indicated that the moment magnitude of the 1906 earthquake ranged from 8.3 to 8.6. Moreover, the dominant slip occurred in the northern part of the assumed source region near the coast of Colombia, where little significant seismicity has occurred, rather than in the southern part. The results also indicated that the source area, with significant slip, covered a long distance, including the southern, central, and northern parts of the region. [ABSTRACT FROM AUTHOR]