Exciting news! We're transitioning to the Statewide California Earthquake Center. Our new website is under construction, but we'll continue using this website for SCEC business in the meantime. We're also archiving the Southern Center site to preserve its rich history. A new and improved platform is coming soon!

Cluster analysis of the long-period ground-motion simulation data – application of the Nankai Trough megathrust earthquakes scenarios

Takahiro Maeda, Hiroyuki Fujiwara, Toshihiko Hayakawa, Satsuki Shimono, & Sho Akagi

Published August 15, 2017, SCEC Contribution #7697, 2017 SCEC Annual Meeting Poster #268

Extracting important hazard information and its trends from the big data leads to better risk assessment and quick respond in case of emergency when megathrust earthquake happens. A cluster analysis was applied to the big data of the long-period ground-motion simulation data. These big data of simulation data were calculated by a three dimensional finite difference method for megathrust earthquakes in the Nankai Trough using various source models (earthquake scenarios) with different source area, rupture starting point, asperity, rupture velocity, source time function and fmax. These source models were constructed according to the source modeling method based on theoretical and empirical rules. The cluster analysis based on a principal component analysis (PCA) and the k-means method classifies earthquake scenarios with similar ground-motion distribution into clusters. The ground-motion distribution was represented by ground motion index value at Q meshes. By applying PCA to Q-dimensional N-scenario data, P principal components were extracted. The ground-motion distribution of each scenario was represented by the coefficients for P principal components. Then the P-dimensional N-scenario data were classified into k-scenario clusters by the k-means method. For our simulation data, the number of input scenarios (N) was 369 and the number of data output points (Q) was 77609. The ground-motion index values were the peak ground velocity and velocity response (period of 3, 5, 7, 10, 20 seconds). When we used a velocity response of 5 second and assume P=5, the cumulative contribution ratio up to the fifth principal component was about 75%, and the ground-motion distribution by the 369 scenario was considered to be roughly expressed using these five principal components. We classified the scenarios into 30 clusters by the k-means method and examined the correspondence between the scenario and the scenario parameters. It was found that the clusters were composed by the common scenario of the source area, the rupture starting point and the shallow-asperity location. By clarifying the relationship between these clusters and scenario parameters, there was a possibility that candidates of scenarios to be added next can be extracted.

This research was supported by CREST, JST.

Key Words
Clustering, Long-period ground motion, Finite difference simulation

Citation
Maeda, T., Fujiwara, H., Hayakawa, T., Shimono, S., & Akagi, S. (2017, 08). Cluster analysis of the long-period ground-motion simulation data – application of the Nankai Trough megathrust earthquakes scenarios. Poster Presentation at 2017 SCEC Annual Meeting.


Related Projects & Working Groups
Ground Motions