10:45 〜 12:15
[SSS07-P12] 東北地方太平洋沖で発生した地震からのPコーダ波中に見られる低周波変換波の特徴，起源，および応用の可能性
キーワード：変換波、低周波、東北地方太平洋沖、時系列クラスタリング、波形シミュレーション
Arrival times and amplitudes of converted phases that appear on seismograms are useful to relocate earthquake hypocenters and to estimate Earth’s structure. We found a phase in the P coda from the earthquakes off the Pacific coast of the Tohoku district, particularly off Iwate and Miyagi prefectures, and reported a preliminary analysis in the previous JpGU meeting. The phase that appears about 8 s after the P wave onset has not been mentioned or analyzed so far due to its low-frequency nature that obscures viewing on the seismograms of the short-period seismometers. We visualized the phase by correcting for the response of seismometers and by applying waveform stacking to improve the signal-to-noise ratio. However, we have not searched the phase in the wide spatial range of the Tohoku district. Here we re-examined the phase using a comprehensive spatial range dataset and applying waveform clustering with machine learning and wave propagation simulation with a 3D Earth model.
We prepared seismograms observed by Hi-net stations for earthquakes along the plate boundary between the subducting Pacific plate and the overriding plate in an area from Hokkaido to Ibaraki prefecture. The period and the magnitude range are from 2004 to 2022, equal to or greater than 4.5, respectively. After the correction of instrumental response and the alignment with P-wave, we stacked the seismograms in some epicentral distance ranges and azimuth, reducing the number of seismograms to be analyzed. Finally, we applied the machine learning tools tslearn package to the dataset for waveform clustering.
When we applied the technique to data on a broad spatial extent, various seismograms were classified as the same cluster. Therefore, we applied the clustering for spatially restricted data sets that contain events within a limited epicenter location bounded by the iso-depth contours of the Pacific plate. This restriction significantly improved the result of clustering. The most significant clusters are for the earthquakes off Iwate and Miyagi prefectures, which we have previously reported. We assigned the number of clusters as six by using the elbow method. The clusters include the low-frequency converted phase with variable shape, frequency, and the number of phase peaks. The quite interesting feature is their separation in space. For example, in the area off the coast of Iwate prefecture, there are four clusters within an area of E-W and N-S extent with 50 and 150 km. The two large clusters are separated to the east and the west, in the N-S direction parallel to the Japan Trench. Two additional small clusters are located westernmost part of the large cluster and between the two large clusters forming separated patches.
We examined the phase from the synthetic seismograms by a 3D waveform propagation simulation using the OpenSWPC package. We used the Earth structure based on the JIVSM velocity model. The synthetic seismograms also contain some phases in the P coda, with consistent lapse time from the P-wave, frequency, and the number of the wave packet. This observation indicates that the known structure can explain the origin of the phase. We found that the shape of the phase is variable among the clusters. The difference in the phase waveform is not due to focal mechanisms because the mechanisms are primarily thrust faulting. By calculating the seismograms with different focal depths, we found that the peak-to-peak interval of the phase varies with the depth. Cross-correlation (CC) between the observed and the simulated phase shows a single peak on the depth-CC plot. We obtained a depth with maximum cross-correlation by applying polynomial approximation to the plot. This method is potentially helpful in estimating the focal depth of offshore earthquakes, though we need to investigate many earthquakes with different locations thoroughly.
We prepared seismograms observed by Hi-net stations for earthquakes along the plate boundary between the subducting Pacific plate and the overriding plate in an area from Hokkaido to Ibaraki prefecture. The period and the magnitude range are from 2004 to 2022, equal to or greater than 4.5, respectively. After the correction of instrumental response and the alignment with P-wave, we stacked the seismograms in some epicentral distance ranges and azimuth, reducing the number of seismograms to be analyzed. Finally, we applied the machine learning tools tslearn package to the dataset for waveform clustering.
When we applied the technique to data on a broad spatial extent, various seismograms were classified as the same cluster. Therefore, we applied the clustering for spatially restricted data sets that contain events within a limited epicenter location bounded by the iso-depth contours of the Pacific plate. This restriction significantly improved the result of clustering. The most significant clusters are for the earthquakes off Iwate and Miyagi prefectures, which we have previously reported. We assigned the number of clusters as six by using the elbow method. The clusters include the low-frequency converted phase with variable shape, frequency, and the number of phase peaks. The quite interesting feature is their separation in space. For example, in the area off the coast of Iwate prefecture, there are four clusters within an area of E-W and N-S extent with 50 and 150 km. The two large clusters are separated to the east and the west, in the N-S direction parallel to the Japan Trench. Two additional small clusters are located westernmost part of the large cluster and between the two large clusters forming separated patches.
We examined the phase from the synthetic seismograms by a 3D waveform propagation simulation using the OpenSWPC package. We used the Earth structure based on the JIVSM velocity model. The synthetic seismograms also contain some phases in the P coda, with consistent lapse time from the P-wave, frequency, and the number of the wave packet. This observation indicates that the known structure can explain the origin of the phase. We found that the shape of the phase is variable among the clusters. The difference in the phase waveform is not due to focal mechanisms because the mechanisms are primarily thrust faulting. By calculating the seismograms with different focal depths, we found that the peak-to-peak interval of the phase varies with the depth. Cross-correlation (CC) between the observed and the simulated phase shows a single peak on the depth-CC plot. We obtained a depth with maximum cross-correlation by applying polynomial approximation to the plot. This method is potentially helpful in estimating the focal depth of offshore earthquakes, though we need to investigate many earthquakes with different locations thoroughly.