SCEC Award Number 21103 View PDF
Proposal Category Individual Proposal (Integration and Theory)
Proposal Title Sequencing the CVM: Looking for Lithotectonic Blocks in Southern California in Seismic Models Using Machine Learning
Name Organization
Laurent Montesi University of Maryland Vedran Lekic University of Maryland
Other Participants Graduate Research Assistant (1 person)
SCEC Priorities 1b, 3b, 3e SCEC Groups CXM, Seismology, SDOT
Report Due Date 03/15/2022 Date Report Submitted 06/16/2022
Project Abstract
Defining lithotectonic blocks over Southern California is important to understand how the faults that separate these blocks are loaded and how they interact with each other to generate earthquakes. Most block definitions are based on geological evidence, which is collected mainly at the surface. Seismic velocity models like the CVM present a complementary view of the region. Here we use a machine learning algorithm (t-SNE) to organize velocity profiles into a sequence that identifies sort profiles based on their similarity. Several geographically coherent regions appear but the results may be improved by adopting multidimensional sequencing algorithms.
Intellectual Merit The research advances knowledge of the organization of the Southern California lithosphere by providing unbiased definitions of blocks based on the similarity of their seismic velocity structure, without input from geological information. These blocks can be used as an alternative framework to understand the rheology and stress field of the region.
Broader Impacts This project introduces the broader SCEC community to machine learning algorithms and their novel application to tectonics.
Exemplary Figure Figure 1

Illustration of the way that the velocity profiles entered in an unsorted manner are organized into a sequence (bottom) by t-SNE. Color coding the profiles with their position in the sequence reveals geographically coherent and geologically sensible regions that provide a new view of the blocks in Southern California.