SCEC Award Number 13102 View PDF
Proposal Category Individual Proposal (Integration and Theory)
Proposal Title Prediction uncertainty in kinematic source inversion
Investigator(s)
Name Organization
Jean-Paul Ampuero California Institute of Technology
Other Participants Zacharie Duputel
SCEC Priorities 2, 3, 6 SCEC Groups SIV, Seismology, FARM
Report Due Date 03/15/2014 Date Report Submitted N/A
Project Abstract
 We developed a general approach to explicitly handle inadequacies of our predictions (forward models) in finite earthquake source inversions. Using this new formalism, we can move from traditional deterministic approaches to more appropriate stochastic forward modeling approaches, enabling realistic probability distributions of predictions for a given source model. These advances lay the groundwork for a new generation of earthquake source models based on a general formalism that rigorously quantifies and incorporates the impact of uncertainties in fault slip inverse problems. In particular, this can be used to account for uncertainty in the Earth model elastic parameters.

We assessed the viability of our approach using simple synthetic test cases for static data. We found that accounting for prediction uncertainty can greatly improve earthquake source models by providing a physical basis for the relative weighting between disparate data sets, and including realistic estimates of uncertainty on inferred model parameters. Avoiding overly simplistic assumptions for the prediction error by capturing the relevant physics of the forward problem lead to more reliable images of earthquake rupture phenomena.

We also applied this new formalism to actual observations of the 2013 Mw 7.7 Balochistan earthquake. We derived the slip posterior probability density function using a Bayesian approach, including a full description of the data covariance and accounting for uncertainty in the elastic structure of the crust. Our solution indicates that this event is mainly associated with shallow strike-slip motion on a dipping fault, which is unprecedented for such large event.
Intellectual Merit Our research attempts to enable a new generation of earthquake source model that are more resistant to over-fitting of data, provide a physical basis for the relative weighting between disparate data sets, and include more realistic estimates of uncertainty on inferred model parameters. This effort exploit recent advances in Bayesian earthquake source modeling and new massively parallel computational approaches using GPUs. The benefits of this approach include improving the sharpness of our images of seismic and aseismic fault slip processes - thereby directly impacting both our models for fault mechanics and inferences of seismic hazard.

Our research addresses the following SCEC4 research priorities and requirements:
• 2a: The tools we developed support the “improvement of earthquake catalogs, including non-point-source source descriptions” by providing a general framework to have realistic error estimates on point-source and finite-fault model parameters.
• 3c: Our results support “theoretical and numerical modeling of specific fault resistance mechanisms for seismic radiation and rupture propagation”, by assessing the reliability of seismological constraints on such processes.
• 6b: By providing better constraints on the earthquake rupture, our results contribute to “modeling of ruptures that includes realistic dynamic weakening mechanisms, off-fault plastic deformation, and is constrained by source inversions”, and to “produce physically consistent rupture models for broadband ground motion simulation.”

Our research directly addresses a priority for FARM: “propose source-inversion methods with minimal assumptions, and provide robust uncertainty quantification of inferred source parameters”. It also relates to Computational Science disciplinary activities: “provide tools and algorithms for uncertainty quantification in large-scale inversion and forward-modeling studies”, and Seismology disciplinary activities: “develop strategies for robust uncertainty quantification in finite-fault rupture models”.
Broader Impacts This project provided training and research opportunities for a postdoctoral scholar, Zacharie Duputel, who is now hired as a CNRS researcher in France. He continues to lead this project, especially through regular visits to Caltech. The multidisciplinary research addressed herein gathered together geodesists, seismologists and computational scientists.

The algorithms and methods developed here are now directly applicable to other crustal deformation problems such as models of volcano deformation or to rapid source estimations problems for warning purposes. As we have done in the past, all of our tools developed will be documented and openly available to the geophysical community as open source. Open source software for 3D kinematic rupture was developed, incorporated in the spectral element code SPECFEM3D, distributed and maintained online through the Computational Infrastructure for Geophysics.
Exemplary Figure Figure 1.
From Jolivet et al. (2013). Slip distribution of the September 24th, 2013 Mw 7.7 Balochistan earthquake inferred from satellite-based observations of co-seismic displacements, accounting for uncertainties in the depth-distribution of shear modulus (shown by the red histograms on the right). The color of each subfault patch indicates the slip amplitude, brightness represents relative slip uncertainty. Arrows and their associated 2-σ error ellipse indicate the slip direction and its uncertainty. The inversion was performed using the ALTAR Bayesian algorithm, initially developed by Minson et al. (2013).