17:15 〜 18:30
[SCG39-P29] Development of microseismicity catalog near the Japan Trench shallow slow earthquake region using a deep learning module
キーワード：スロー地震、低周波微動、深層学習、地震計アレイ、日本海溝、海底地震計
The Japan Trench subduction zone is a cradle for wide spectrum of seismic and aseismic activities, such as the Mw 9.0 Tohoku-Oki earthquake, slow slip event, very low frequency earthquakes, and tectonic tremors. In a past few years, the S-Net, the cabled network of ocean bottom seismometers (OBSs) and dense networks established by free-fall type OBSs reveal tectonic tremor in a shallow portion of the subduction zone [e.g. Nishikawa et al., 2019; Tanaka et al., 2019; Ohta et al. 2019]. However, abundant OBS data sets have not fully utilized to understand micro-seismicity of the same region in contrast to tectonic tremor activity. The OBS networks have often used to relocate hypocenters of earthquakes cataloged in the JMA unified catalog [e.g. Shinohara et al., 2012]. However, the JMA unified catalog is developed only using onshore seismic network, which has limited detectability and location capability on earthquakes occur offshore. Meanwhile, it is uncommon attempt to detect micro-earthquakes, which are exclusively visible from offshore seismic network despite of its necessity to fully-expose interplay between fast, standard seismicity and slow seismicity at the shallow trench; the attempt is definitely important to understand the wide spectrum of rupture processes in the subduction zone.
Here, we perform detection and phase picking of microearthquakes to a data set of an OBS network, which was deployed in offshore Fukushima near the trench to develop a high-quality offshore seismicity catalog, with applied the Earthquake Transformer (EQT) [Mousavi et al., 2020]. The EQT has already been trained with a set of onshore seismic and ambient noise signals [Mousavi et al., 2019]. We attempt to apply the trained model directly to the OBS data to examine its performance on offshore environment. Our OBS data set is composed of 3 dense arrays with a small aperture and 3 single station OBSs. First, we select one OBS from each array in the network as well as the single station OBSs. Seismograms of the selected stations are then filtered between 2 Hz and 30 Hz to suppress low frequency noise and increase signal-to-noise ratio. The filtered continuous waveforms are then used to detect and pick P- and S-wave arrivals from microearthquakes. The picked arrival times are finally used to estimate event hypocenters with hypomh [Hirata and Matsu’ura, 1987].
As a result of the experiment, the EQT detects more than 5000 microearthquakes, which is six times number of events than the earthquakes located by the JMA in the same region during the observation period. Although most of the detected events are located outside of the region where tectonic tremors have been observed [Nishikawa et al., 2019; Ohta et al., 2019], there are about 60 events located in close proximity of the tectonic tremors.
To evaluate closeness of the tectonic tremors and the microearthquakes, we perform array beamforming in the 3 arrays mentioned above to compare two different types of phenomena in slowness domain. It is generally unclear whether those microearthquakes occur within the tremorgenic region or completely separated from each other, because the tectonic tremors have relatively large error on their hypocenter compares to the microearthquakes because they are located with the envelope correlation method [e.g., Ide, 2010; Obara, 2002]. After filtering their waveform in 2 Hz and 8 Hz, the tectonic tremors are beamformed with the duration based on Ohta et al. (2019), and the microearthquakes are beamformed for 2 seconds S-wave arrival time. Comparison in the slowness domain suggests at least 3 ordinal, or fast earthquakes occur within the source region of tectonic tremors.
Here, we perform detection and phase picking of microearthquakes to a data set of an OBS network, which was deployed in offshore Fukushima near the trench to develop a high-quality offshore seismicity catalog, with applied the Earthquake Transformer (EQT) [Mousavi et al., 2020]. The EQT has already been trained with a set of onshore seismic and ambient noise signals [Mousavi et al., 2019]. We attempt to apply the trained model directly to the OBS data to examine its performance on offshore environment. Our OBS data set is composed of 3 dense arrays with a small aperture and 3 single station OBSs. First, we select one OBS from each array in the network as well as the single station OBSs. Seismograms of the selected stations are then filtered between 2 Hz and 30 Hz to suppress low frequency noise and increase signal-to-noise ratio. The filtered continuous waveforms are then used to detect and pick P- and S-wave arrivals from microearthquakes. The picked arrival times are finally used to estimate event hypocenters with hypomh [Hirata and Matsu’ura, 1987].
As a result of the experiment, the EQT detects more than 5000 microearthquakes, which is six times number of events than the earthquakes located by the JMA in the same region during the observation period. Although most of the detected events are located outside of the region where tectonic tremors have been observed [Nishikawa et al., 2019; Ohta et al., 2019], there are about 60 events located in close proximity of the tectonic tremors.
To evaluate closeness of the tectonic tremors and the microearthquakes, we perform array beamforming in the 3 arrays mentioned above to compare two different types of phenomena in slowness domain. It is generally unclear whether those microearthquakes occur within the tremorgenic region or completely separated from each other, because the tectonic tremors have relatively large error on their hypocenter compares to the microearthquakes because they are located with the envelope correlation method [e.g., Ide, 2010; Obara, 2002]. After filtering their waveform in 2 Hz and 8 Hz, the tectonic tremors are beamformed with the duration based on Ohta et al. (2019), and the microearthquakes are beamformed for 2 seconds S-wave arrival time. Comparison in the slowness domain suggests at least 3 ordinal, or fast earthquakes occur within the source region of tectonic tremors.