4:15 PM - 4:30 PM
[SSS12-10] Hierarchical clustering of seismic source processes for the 1993 Hokkaido-Nansei-Oki earthquake and the 2024 Noto Peninsula earthquake
Reverse fault-type seismicity is known to be prevalent in the Japan Sea coastal region. These earthquakes occur on pre-existing normal faults that formed during the opening of the Japan Sea as well as on the faults newly created under the present compressional stress regime. Large earthquakes in this region may involve the sequential rupture of multiple old and new faults. It is essential to analyze them using source-process models that capture variations in fault geometry and the complex rupture propagation. However, an objective approach for systematically identifying fault-rupture linkages within high-degree-of-freedom source-process models has yet been established.
In this study, we analyze the source processes of the 2024 Noto Peninsula earthquake and the 1993 Hokkaido-Nansei-Oki earthquake using the Potency Density Tensor Inversion (PDTI) method, a high-degree-of-freedom seismic waveform inversion technique. To objectively extract fault-linkage information, we apply hierarchical clustering to classify the characteristics of the fault rupture.
The PDTI represents each source knot as a summation of five basis double-couple components of fault slip. The PDTI does not impose constraints on the fault slip direction within a predefined model plane, allowing for the construction of a high-degree-of-freedom source-process model that estmates fault geometry and slip direction. Hierarchical clustering is then applied to the PDTI model to classify the hierarchical structure of the fault geometry and the rupture process.
The hierarchical clustering treats each knot as an independent cluster and progressively merges them into higher-order clusters based on similarity. In this study, cosine similarity is used as the similarity metric, with standardized n-vector and v-vector components as input data. Here, the n-vector represents a vector normal to the nodal plane. The v-vector corresponds to the slip direction. For cluster merging, Ward’s method is employed to minimize variance within clusters.
Clustering is applied to the PDTI model of the 2024 Noto Peninsula Earthquake. No obvious clusters are observed during the first 10 seconds after the origin time. However, from 10 to 23 s, a northeast-southwest elongated cluster with a 70-km major axisis formed along the Suzu-Wajima coast. The P-axis azimuth of this cluster is approximately 295°. Between 23 and 35 s, another cluster is formed, extending about 50 km from Saruyama to the Wajima coast, approximately 30 km southwest of the rupture initiation point. The P-axis azimuth of this cluster is approximately 286°, indicating a counterclockwise rotation relative to the preceding Suzu-Wajima cluster. Additionally, from 25 to 35 s, a cluster with a nearly identical P-axis azimuth but a strike rotated counterclockwise by approximately 10° is also identified to the south (inland) of this cluster. During the same period, from 26 to 32 s, a cluster with a P-axis azimuth of approximately 235° is formed offshore of Suzu, about 30 km northeast of the rupture initiation point. After 35 s, no dominant cluster is observed. These results demonstrate that our cluster analysis can objectively extract fault segment characteristics discussed in the previous studies.
For the 1993 Hokkaido-Nansei-Oki Earthquake, we performed the PDTI analysis and then applied the hierarchical clustering to the PDTI result. During the first 10 s after the origin time, a reverse fault-type cluster with a strike of approximately 153° is formed, extending about 10 km westward from the epicenter. From 5 to 10 s, a cluster with a strike-slip component is distributed, centered approximately 10 km southeast of the epicenter. After 10 s, it transitions into a reverse fault-type cluster with a strike of approximately 200°, centered about 15 km northeast of the epicenter. From 25 to 43 s, this cluster migrates offshore to the west of Okushiri Island, approximately 60 km southeast of the epicenter. After 43 s, no dominant cluster is observed.
In this study, we analyze the source processes of the 2024 Noto Peninsula earthquake and the 1993 Hokkaido-Nansei-Oki earthquake using the Potency Density Tensor Inversion (PDTI) method, a high-degree-of-freedom seismic waveform inversion technique. To objectively extract fault-linkage information, we apply hierarchical clustering to classify the characteristics of the fault rupture.
The PDTI represents each source knot as a summation of five basis double-couple components of fault slip. The PDTI does not impose constraints on the fault slip direction within a predefined model plane, allowing for the construction of a high-degree-of-freedom source-process model that estmates fault geometry and slip direction. Hierarchical clustering is then applied to the PDTI model to classify the hierarchical structure of the fault geometry and the rupture process.
The hierarchical clustering treats each knot as an independent cluster and progressively merges them into higher-order clusters based on similarity. In this study, cosine similarity is used as the similarity metric, with standardized n-vector and v-vector components as input data. Here, the n-vector represents a vector normal to the nodal plane. The v-vector corresponds to the slip direction. For cluster merging, Ward’s method is employed to minimize variance within clusters.
Clustering is applied to the PDTI model of the 2024 Noto Peninsula Earthquake. No obvious clusters are observed during the first 10 seconds after the origin time. However, from 10 to 23 s, a northeast-southwest elongated cluster with a 70-km major axisis formed along the Suzu-Wajima coast. The P-axis azimuth of this cluster is approximately 295°. Between 23 and 35 s, another cluster is formed, extending about 50 km from Saruyama to the Wajima coast, approximately 30 km southwest of the rupture initiation point. The P-axis azimuth of this cluster is approximately 286°, indicating a counterclockwise rotation relative to the preceding Suzu-Wajima cluster. Additionally, from 25 to 35 s, a cluster with a nearly identical P-axis azimuth but a strike rotated counterclockwise by approximately 10° is also identified to the south (inland) of this cluster. During the same period, from 26 to 32 s, a cluster with a P-axis azimuth of approximately 235° is formed offshore of Suzu, about 30 km northeast of the rupture initiation point. After 35 s, no dominant cluster is observed. These results demonstrate that our cluster analysis can objectively extract fault segment characteristics discussed in the previous studies.
For the 1993 Hokkaido-Nansei-Oki Earthquake, we performed the PDTI analysis and then applied the hierarchical clustering to the PDTI result. During the first 10 s after the origin time, a reverse fault-type cluster with a strike of approximately 153° is formed, extending about 10 km westward from the epicenter. From 5 to 10 s, a cluster with a strike-slip component is distributed, centered approximately 10 km southeast of the epicenter. After 10 s, it transitions into a reverse fault-type cluster with a strike of approximately 200°, centered about 15 km northeast of the epicenter. From 25 to 43 s, this cluster migrates offshore to the west of Okushiri Island, approximately 60 km southeast of the epicenter. After 43 s, no dominant cluster is observed.