[S-TT42] Advancing Earth Science through Optic Fiber Sensing Techniques and Integrated Analysis

[STT42-12] Development of GUI software for picking up seismic phases from distributed acoustic sensing records

Hiroaki Yamahana¹, *Masanao Shinohara² (1.Graduate School of Science, The University of Tokyo, 2.Earthquake Research Institute, The University of Tokyo)

Keywords:Distributed Acoustic Sensing, GUI software, Python, Phase identification

Recently, real-time seafloor seismic observations have been performed using seafloor cable systems in the regions of Kuril trench, Japan trench, and Nankai trough. However, the spatial density of stations in the marine area is generally lower than that of land seismic networks. Optical fiber sensing technology such as distributed acoustic sensing (DAS) provides a seismic observation with super high density of stations along a fiber for a long distance. The high density seismic data contribute to the determination of precise hypocenter distribution or seismic structures in the marine area. However, it is difficult to identify seismic phases for all channels (seismic stations) within DAS dataset due to several disadvantages: First, sensitivity becomes low for the direction perpendicular to the fiber. Second, picking up signals for all channels manually are costly due to the large number of channels. Finally, existing software for manual picking has difficulty in picking up seismic phases using channel coherence information. To promote seismic analyses using DAS records, it is important to enable low-cost arrival time picking using channel coherence information. One way to overcome these disadvantages, we can apply automatic pickers based on deep learning methods like PhaseNet-DAS. The distributed model of PhaseNet-DAS was trained using DAS observations on land and arrival time data by PhaseNet, which is an automatic picker for conventional seismometers. We applied the original PhaseNet-DAS to marine DAS records. As a result, S-wave and PS-wave arrivals were precisely picked up, but identifying P-wave arrivals was difficult. To increase the performance of PhaseNet-DAS on marine DAS records, training on marine DAS records is thought to be necessary, but it is currently difficult because we do not have numerous precise arrival time data on marine DAS records.
Existing GUI software to pick up seismic arrival times is not suitable for DAS records, because it does not utilize channel coherence information which is an advantage of DAS records. In addition, optimization of operation for many channels of DAS records and huge data size is needed. From these backgrounds, we decided to develop a new GUI software to identify seismic phases effectively for DAS records. The new GUI software has a graphic image indicating amplitudes of data for a spatiotemporal domain in addition to a conventional waveform display for a single channel. The image is displayed as fast as possible by using a fast graphic library. The aspect ratio and displayed range of the image can be smoothly adjusted by dragging a mouse or using a keyboard. An arrival time for a single channel is picked by clicking, and arrival times for multiple channels are easily picked by dragging a mouse. Furthermore, because name and number of phases for picking are freely specified, we can mark not only P- and S-wave arrivals but also other arrivals such as converted waves. The GUI software was developed by using Python and DAS records can be given in numpy.ndarray format which is one of the most common formats of array data. Therefore, DAS records processed with existing libraries such as DASPy or Xdas, or any other Python programs, can be directly handled by the developed GUI software.
Using the GUI software, we identified arrival times from DAS data of an M2.8 earthquake with a distance of 21 km obtained by the off-Sanriku seafloor cable system. We could identify arrival times of P-waves with small amplitudes, reflected P-waves from the sea surface and converted waves from P-wave to S-wave at the bottom of a sedimentary layer. These arrival times are consistent with the seafloor topography and the seismic reflection image along the cable. After we pick up arrival times for more events, the results will contribute to hypocenter locations, focal solution determination using DAS records, and training of the deep learning models.