SCEC Award Number 25310 View PDF
Proposal Category Collaborative Research Project (Multiple Investigators / Institutions)
Proposal Title Assessing the Community Fault Model Using Improved Earthquake Catalogs
Investigator(s)
Name Organization
Gregory Beroza Stanford University
SCEC Milestones A3-1 SCEC Groups CEM, Seismology, EFP
Report Due Date 03/15/2026 Date Report Submitted 10/04/2026
Project Abstract
 With this funding we developed a deep-learning-based method that we call "FaultNet" to discern fault surfaces within earthquake catalogs using technology that was originally developed for discerning objects within point clouds. We have applied it to existing earthquake catalogs in California and beyond to prototype this capability for use in evolving and improving the CFM. It has many other applications for fault modeling, such as describing the 3D geometry of faults for dynamic modeling, exploring the spatial variations of b-values, and for aftershock forecasting. We submitted a publication on the topic to JGR - Machine Learning.
SCEC Community Models Used Community Fault Model (CFM)
Usage Description A primary goal of the proposed research is to provide more effective methods for adding to the Community Fault Model.
Intellectual Merit This project is an example of how AI methods are being used more broadly within SCEC to develop an improved understanding of earthquake processes. Deep learning has been used previously and extensively to develop improved earthquake catalogs, it has been used to interpret surface expressions to map fault complexity, but this is the first time it has been used to develop representations of fault systems from earthquake catalogs. The project will also contribute to the development of the CFM.
Broader Impacts This project helped support graduate student Yanlan Hu, who is now a 3rd year graduate student working towards her PhD. She is currently serving as a TA for Introductory Seismology. Once published, FaultNet will be available as an open source resource through GitHub. Several research groups around the US have already expressed interest in using it for different applications that require accurate fault geometry.
Project Participants At Stanford University, Yanlan Hu (graduate student) and Greg Beroza (PI) have been involved in the project. We received helpful input from Travis Alongi of the USGS, who has developed analogous methods and has experience applying them to complex fault structures.
Exemplary Figure Figure 1. Schematic overview of the inference workflow in FaultNet. (a) Three-dimensional earthquake hypocenter catalog used as the input. (b) The denoising model distinguishes fault-related earthquakes from background seismicity. (c) The feature-learning model learns feature embeddings for the fault-related earthquakes. (d) Feature embeddings predicted from overlapping sliding windows are aggregated into a global distance matrix. (e) Agglomerative clustering groups earthquakes into individual fault clusters according to the aggregated distance matrix. (f) Clusters are reconstructed into continuous fault surfaces using radial basis function (RBF) interpolation (after Hu and Beroza, 2026).
Linked Publications

Add missing publication or edit citation shown. Enter the SCEC project ID to link publication.