Objective Tectonic Regionalization of CVM-S4.26 Using the k-means Clustering Algorithm

William K. Eymold, & Thomas H. Jordan

Submitted August 9, 2017, SCEC Contribution #7468, 2017 SCEC Annual Meeting Poster #229 (PDF)

The SCEC Community Velocity Model CVM-S4.26 is a representation of the P and S velocities, α and β, as a function of geographic position x and depth z on a 500-m mesh over a 496 km x 768 km area of Southern California. The model was derived by full-3D tomography using an extensive set of earthquake waveforms and ambient-field correlagrams (Lee et al., 2014a); subsequent testing has shown that it produces synthetic seismograms up to 0.2 Hz that well match observed seismograms from earthquakes not used in the inversion (Lee et al., 2014b; Taborda et al., 2016). In this study, we analyze the N = 1.52 million velocity profiles α_x (z) and β_x (z) from CVM-S4.26 using the k-means clustering algorithm (MacQueen, 1967; Arthur & Vassilvitskii, 2007) to map tectonic regions with similar profile characteristics. The k-means algorithm partitions the N profiles into k sets that minimize the inter-cluster variance via a randomly seeded iterative process. When applied to CVM-S4.26, the results are surprisingly stable for k ≤ 8; e.g., similar results are obtained when applied individually to α_x (z) and β_x (z) or to their combination α_x (z)∪β_x (z). Moreover, the clusters comprise one or more large, contiguous areas that in most cases correspond to regions with known tectonic affinities. Our preferred clustering, k = 7, distinguishes the (1) Continental Borderland, (2) Proterozoic terrains of southeastern California and western Arizona, (3) Great Valley, (4) eastern Sierra Nevada, (5) Peninsular Range batholith, (6) Mojave block, and (7) Salton Trough. Smaller areas grouped within these clusters include the Santa Barbara and Santa Maria basins (cluster 3), the eastern Transverse Ranges (cluster 4), western Sierra Nevada (cluster 5), southern Salinian block (cluster 6), and Los Angeles basins (cluster 7). These clusters are distinguished by structural commonalities in near-surface and lower-crustal velocities, Moho depth, and the amplitude of the mid-crustal low-velocity zone (MCLVZ). In particular, the MCLVZ amplitude is largest beneath the Mojave block. These strong inter-cluster correlations provide new information about how the present-day crustal structure of Southern California has evolved since Cretaceous times.

Key Words
K-means clustering, CVM, F3DT

Eymold, W. K., & Jordan, T. H. (2017, 08). Objective Tectonic Regionalization of CVM-S4.26 Using the k-means Clustering Algorithm. Poster Presentation at 2017 SCEC Annual Meeting.

Related Projects & Working Groups
SCEC Community Models (CXM)