2:45 PM - 3:00 PM
[SSS05-05] Fault plane modeling for dynamic rupture simulation targeting the 2011 Mj7.0 Fukushima Hamadori, Japan, earthquake
Keywords:2011 Mj7.0 Fukushima Hamadori earthquake, Fault plane modeling, Clustering
The 2011 Mj7.0 Fukushima Hamadori, Japan, earthquake (Hamadori earthquake) is thought to have occurred through a complex process involving the sequential rupture of multiple faults, not just a single fault (e.g., Kobayashi et al., 2012). However, it remains unclear why and how the sequential rupture of multiple faults occurred. Besides, attempts have been made to infer the spatiotemporal distribution of fault slip during the Hamadori earthquake using surface displacement obtained from InSAR (Kobayashi et al., 2012; Fukushima et al., 2013) and strong motion waveforms (e.g., Hikima, 2012), but it has not been done yet to investigate the rupture process of the seismic source using dynamic rupture simulations. In this study, as a preparation to investigate the seismic source rupture process during the Hamadori earthquake using dynamic rupture simulations, we conduct modeling of the ruptured fault planes during the Hamadori earthquake. Keeping in mind that the rupture process of the seismic source can be strongly influenced by the geometry of fault planes (e.g., Ando and Kaneko, 2018), we aim to construct detailed fault plane models including non-planar features such as branching and bending.
Data and Methods
We attempt fault plane modeling through analysis of aftershock distribution following the Hamadori earthquake. As the earthquake catalog, we use the catalog for the Hamadori earthquake's source area constructed by Kato et al. (2013) using the double-difference tomography method (Zhang and Thurber, 2003), along with its updated version (extended until December 31, 2015).
In the aftershock distribution analysis, firstly, clustering is conducted using HDBSCAN (Campello et al., 2013). HDBSCAN is an algorithm that performs clustering and noise detection based on the density of points, capable of detecting clusters of arbitrary shapes such as not only spherical but also planar clusters. Secondly, multiple linear regression analysis is performed for each cluster of aftershocks, selecting only clusters with coefficient of determination exceeding a threshold as clusters in which aftershocks are distributed in a planar manner. Through these procedures, if the aftershock distribution is determined with sufficient positional accuracy, detailed planar structures can be extracted with high resolution from the aftershock distribution. Thirdly, clustering by HDBSCAN is again applied to the obtained small planar clusters, grouping them into several groups, obtaining relatively large representative planes for each group, and integrating them to construct sophisticated fault plane models for dynamic rupture simulation. These algorithms are implemented using libraries such as scikit-learn, NumPy, and hdbscan (McInnes et al., 2017).
Preliminary Results
The combination of clustering with HDBSCAN and determination of planarity based on the coefficient of determination in multiple linear regression analysis enabled the extraction of numerous fine planar clusters from the aftershock distribution. Furthermore, by subjecting the fine planar clusters to further clustering with HDBSCAN, it was confirmed that groups could be formed for sets of planar clusters that are aligned in a planar manner. This presentation will report on these results and discuss the methodology for constructing the final fault plane model to be used in dynamic rupture simulations of the Hamadori earthquake.
Data and Methods
We attempt fault plane modeling through analysis of aftershock distribution following the Hamadori earthquake. As the earthquake catalog, we use the catalog for the Hamadori earthquake's source area constructed by Kato et al. (2013) using the double-difference tomography method (Zhang and Thurber, 2003), along with its updated version (extended until December 31, 2015).
In the aftershock distribution analysis, firstly, clustering is conducted using HDBSCAN (Campello et al., 2013). HDBSCAN is an algorithm that performs clustering and noise detection based on the density of points, capable of detecting clusters of arbitrary shapes such as not only spherical but also planar clusters. Secondly, multiple linear regression analysis is performed for each cluster of aftershocks, selecting only clusters with coefficient of determination exceeding a threshold as clusters in which aftershocks are distributed in a planar manner. Through these procedures, if the aftershock distribution is determined with sufficient positional accuracy, detailed planar structures can be extracted with high resolution from the aftershock distribution. Thirdly, clustering by HDBSCAN is again applied to the obtained small planar clusters, grouping them into several groups, obtaining relatively large representative planes for each group, and integrating them to construct sophisticated fault plane models for dynamic rupture simulation. These algorithms are implemented using libraries such as scikit-learn, NumPy, and hdbscan (McInnes et al., 2017).
Preliminary Results
The combination of clustering with HDBSCAN and determination of planarity based on the coefficient of determination in multiple linear regression analysis enabled the extraction of numerous fine planar clusters from the aftershock distribution. Furthermore, by subjecting the fine planar clusters to further clustering with HDBSCAN, it was confirmed that groups could be formed for sets of planar clusters that are aligned in a planar manner. This presentation will report on these results and discuss the methodology for constructing the final fault plane model to be used in dynamic rupture simulations of the Hamadori earthquake.