A Caltech Library Service

A Bayesian Approach to Earthquake Source Studies


Minson, Sarah Ellen (2010) A Bayesian Approach to Earthquake Source Studies. Dissertation (Ph.D.), California Institute of Technology. doi:10.7907/3RT9-3215.


Bayesian sampling has several advantages over conventional optimization approaches to solving inverse problems. It produces the distribution of all possible models sampled proportionally to how much each model is consistent with the data and the specified prior information, and thus images the entire solution space, revealing the uncertainties and trade-offs in the model. Bayesian sampling is applicable to both linear and non-linear modeling, and the values of the model parameters being sampled can be constrained based on the physics of the process being studied and do not have to be regularized. However, these methods are computationally challenging for high-dimensional problems.

Until now the computational expense of Bayesian sampling has been too great for it to be practicable for most geophysical problems. I present a new parallel sampling algorithm called CATMIP for Cascading Adaptive Tempered Metropolis In Parallel. This technique, based on Transitional Markov chain Monte Carlo, makes it possible to sample distributions in many hundreds of dimensions, if the forward model is fast, or to sample computationally expensive forward models in smaller numbers of dimensions. The design of the algorithm is independent of the model being sampled, so CATMIP can be applied to many areas of research.

I use CATMIP to produce a finite fault source model for the 2007 Mw 7.7 Tocopilla, Chile earthquake. Surface displacements from the earthquake were recorded by six interferograms and twelve local high-rate GPS stations. Because of the wealth of near-fault data, the source process is well-constrained. I find that the near-field high-rate GPS data have significant resolving power above and beyond the slip distribution determined from static displacements. The location and magnitude of the maximum displacement are resolved. The rupture almost certainly propagated at sub-shear velocities. The full posterior distribution can be used not only to calculate source parameters but also to determine their uncertainties. So while kinematic source modeling and the estimation of source parameters is not new, with CATMIP I am able to use Bayesian sampling to determine which parts of the source process are well-constrained and which are not.

Item Type:Thesis (Dissertation (Ph.D.))
Subject Keywords:Bayesian analysis; earthquake source; finite fault model; Tocopilla earthquake; CATMIP
Degree Grantor:California Institute of Technology
Division:Geological and Planetary Sciences
Major Option:Geophysics
Thesis Availability:Public (worldwide access)
Research Advisor(s):
  • Simons, Mark
Thesis Committee:
  • Heaton, Thomas H. (chair)
  • Beck, James L.
  • Helmberger, Donald V.
  • Kanamori, Hiroo
  • Lapusta, Nadia
  • Simons, Mark
Defense Date:4 June 2010
Record Number:CaltechTHESIS:06062010-235315977
Persistent URL:
Minson, Sarah Ellen0000-0001-5869-3477
Default Usage Policy:No commercial reproduction, distribution, display or performance rights in this work are provided.
ID Code:5918
Deposited By: Sarah Minson
Deposited On:29 Jul 2010 15:47
Last Modified:07 Jan 2020 22:16

Thesis Files

PDF (Full thesis) - Final Version
Creative Commons Attribution Non-commercial Share Alike.

[img] Video (AVI) (Animated version of Figure 2.2: CATMIP algorithm example) - Supplemental Material
Creative Commons Attribution Non-commercial Share Alike.


Repository Staff Only: item control page