10:45 AM - 12:15 PM
[STT44-P09] A seismograph network design for hypocenter determination using sensor selection algorithm: Numerical examinations
Keywords:Earthquake hypocenter, Hypocenter determination, Station selection
The locations of earthquake hypocenters are fundamental information in seismology. In particular, accurate determinations of hypocenter distribution for aftershocks of a large inland earthquake and seismic swarm are important to understand their spatiotemporal activity and generation mechanisms. Meanwhile, the spatial scale of such swam-like seismicity is often less than the intervals of permanent seismograph station networks. A campaign seismograph network is temporally installed for the accurate hypocenter determination of earthquakes. Thereby, an optimum design of such the campaign seismograph network will contribute to improving the determination accuracy [e.g., Kraft et al., 2013].
In this study, we introduce the station selection algorithm, which was proposed by Nakai et al. [arXiv], for seismograph network design in the hypocenter determination problem. This algorithm bases the D-optimum design of experiments and the greedy method for the station selection. In this approach, we define the parameter sensitivity matrix for candidate stations, of which elements reflect the sensitivity of stations to locations and origin times of earthquakes as the travel time. Using this sensitivity matrix, we sequentially add a station that minimizes the volume of an error ellipsoid of estimated hypocenters [e.g., Saito et al., 2021; Nakai et al., arXiv]. This strategy of the station selection corresponds to adding observations at an arbitral point which maximize the determinant of the Fisher information matrix.
We numerically verify the validity of the proposed method. Here, we consider a single earthquake located at a depth of 10 km. The candidate stations are scattered within 30 km of the epicentral distance. To simplify the numerical examination, we consider only the P wave. The P-wave velocity is homogeneous and assumed to be 5.8 km/s. Because at least four observations should require determining an earthquake hypocenter, we focus on the results in which four or more stations are selected. We note that the proposed method constructs the triangular quadripartite station layout when four stations are selected. The station layouts selected by the proposed method expect that volumes of the error ellipsoid are greater than the average of those evaluated those obtained in the random selections.
Acknowledgment:
This study was supported by the JST CREST [grant number JPMJCR1763].
In this study, we introduce the station selection algorithm, which was proposed by Nakai et al. [arXiv], for seismograph network design in the hypocenter determination problem. This algorithm bases the D-optimum design of experiments and the greedy method for the station selection. In this approach, we define the parameter sensitivity matrix for candidate stations, of which elements reflect the sensitivity of stations to locations and origin times of earthquakes as the travel time. Using this sensitivity matrix, we sequentially add a station that minimizes the volume of an error ellipsoid of estimated hypocenters [e.g., Saito et al., 2021; Nakai et al., arXiv]. This strategy of the station selection corresponds to adding observations at an arbitral point which maximize the determinant of the Fisher information matrix.
We numerically verify the validity of the proposed method. Here, we consider a single earthquake located at a depth of 10 km. The candidate stations are scattered within 30 km of the epicentral distance. To simplify the numerical examination, we consider only the P wave. The P-wave velocity is homogeneous and assumed to be 5.8 km/s. Because at least four observations should require determining an earthquake hypocenter, we focus on the results in which four or more stations are selected. We note that the proposed method constructs the triangular quadripartite station layout when four stations are selected. The station layouts selected by the proposed method expect that volumes of the error ellipsoid are greater than the average of those evaluated those obtained in the random selections.
Acknowledgment:
This study was supported by the JST CREST [grant number JPMJCR1763].