SCEC Award Number 25341 View PDF
Proposal Category Collaborative Research Project (Multiple Investigators / Institutions)
Proposal Title Combining hierarchical MCMC and dynamic rupture simulations in a supercomputing framework to infer off-fault damage parameters
Investigator(s)
Name Organization
Alice-Agnes Gabriel University of California, San Diego Zihua Niu Ludwig Maximilian University of Munich (Germany) Yehuda Ben-Zion University of Southern California
SCEC Milestones C1,2,3-1, C1-2, C2-2, C3-1 SCEC Groups RC, FARM, CEM
Report Due Date 03/15/2026 Date Report Submitted 05/08/2026
Project Abstract
 Theoretical models predict high-stress concentrations at an earthquake’s rupture front, which need to be accommodated by off-fault anelastic processes. While decades of laboratory experiments have reasonably constrained parameters governing fault friction, those governing off-fault brittle damage remain poorly understood. In this project, we will use a novel framework linking SeisSol 3D dynamic rupture simulations to modern uncertainty quantification (UQ) software frameworks to perform - for the first time - a Bayesian inversion for off-fault damage parameters, efficiently utilizing computationally intensive 3D dynamic rupture simulations and seismic and geodetic observations. With HPC-empowered forward simulations and hierarchical uncertainty quantification methods like Multilevel Delayed Acceptance MCMC (MLDA) drastically reducing compute cost, sampling Bayesian posteriors for computationally intensive models now becomes feasible. As a pilot study, we will use 10 near-fault GNSS recordings, seismic data at 8 strong motion seismometers and fault-parallel offsets of the Mw7.1 Ridgecrest mainshock to quantify uncertainties in a reference dynamic rupture model with lab-constrained frictional parameters and continuum off-fault brittle rheology parameterization. We will validate our results with kinematically inverted seismic moment release rate and compare the modeled off-fault damage patterns to observed aftershock distribution, measured from 450 near fault sensors in dense 1D and 2D arrays around the rupture zone. This study represents the first attempt to perform a Bayesian inversion for off-fault damage parameters using this computationally efficient approach. The findings will enhance our understanding of the mechanics of off-fault damage and its influence on earthquake rupture dynamics.
Intellectual Merit This award produced the first Bayesian inversion for off-fault damage rheology parameters in 3D dynamic rupture models of a real earthquake. Methodologically, we showed that prefetching MLDA, a Gaussian-process surrogate plus polynomial-order-2 and order-3 SeisSol forward models, makes Bayesian inference tractable for 3D dynamic rupture; the level-3 model alone is ~48× cheaper than the published reference. Scientifically, simultaneously constraining on-fault rate-and-state friction and off-fault Drucker–Prager cohesion exposed parameter trade-offs invisible to deterministic best-fit searches and recovered different effective strengths on the NW and SE segments of the Ridgecrest fault. The open-source SeisSol/UM-BRIDGE workflow is portable across supercomputers.
Broader Impacts This project supported PhD student Zihua Niu, visiting at SIO/UCSD, who led the inversion design, forward modeling, and analysis as and presented the work at the 2025 SCEC Annual Meeting. The award also enabled close collaboration between SIO/UCSD, USC, the Karlsruhe Institute of Technology, and Ludwig-Maximilians-Universität München, fostering exchange between the dynamic-rupture and Bayesian-inference communities. The HPC-optimized workflow that resulted from the project lowers the barrier for the broader earthquake science community to perform uncertainty-quantified dynamic rupture inversions: SeisSol, UM-BRIDGE, and the prefetching MLDA implementation of Kruse et al. are open source, documented, and reusable.
Project Participants Postdoctoral researcher Zihua Niu (Scripps Institution of Oceanography, UC San Diego) led the project as Early-Career Researcher, with scientific supervision from Alice-Agnes Gabriel (SIO/UCSD) and Yehuda Ben-Zion (University of Southern California). External collaborators contributed the methodological framework underpinning the inversion: Linus Seelinger and Mario Kruse (Karlsruhe Institute of Technology, Germany) developed and supported the prefetching MLDA implementation in UM-BRIDGE, and Heiner Igel and Nico Schliwa (Ludwig-Maximilians-Universität München, Germany) collaborated on the SeisSol forward-model setup and on integration with the Ridgecrest reference dynamic rupture simulations of Taufiqurrahman et al. (2023).
Exemplary Figure Figure 2. Bayesian posterior ρ(m∣dobs) from MLDA inversion. Diagonal panels (top to bottom): 1D marginal distributions of off-fault plastic cohesion scaling factors γ0, γ1 and on-fault direct-effect scaling factors α0, α1. Super-diagonal panels: 2D marginal posteriors for each parameter pair, with effective samples from the eight Markov chains shown as blue dots and the posterior density estimated by Gaussian kernel density. Panel (k): fault-parallel surface offsets predicted by the posterior models (blue) compared to the optical-image correlation data of Antoine et al. (2021, dashed black) and the reference model of Taufiqurrahman et al. (2023, solid black). Adapted from the EPSL paper (Niu et al., in press).
Linked Publications

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