HR: 1340h
AN: S43A-1061    [Abstracts]
TI: Advances in Non-linear Dynamic Inversion of Strong-Motion Data
AU: * Corish, S M
EM: corish@lanl.gov
AF: Los Alamos National Laboratory, P.O. Box 1663, MS D443, Los Alamos, NM 87545 United States
AU: Bradley, C R
EM: cbradley@lanl.gov
AF: Los Alamos National Laboratory, P. O. Box 1663, Los Alamos, NM 87545 United States
AU: Olsen, K B
EM: kbolsen@sciences.sdsu.edu
AF: San Diego State University, 550 Campanile Dr., San Diego, CA 92182
AB: Dynamic modeling of earthquakes, in which rupture is controlled by the yield stress and distributions of initial stress and friction on a fault, explicitly solves the mechanical problem of rupture propagation, and therefore can provide mechanically consistent insight about the natural processes involved in rupture propagation and arrest. As this information can be accessed most readily by a nonlinear inversion procedure, it is crucial to have an efficient, physically plausible inversion scheme. Many inversion methods divide the fault spatially into patches or rectangular subfaults, and find optimal values of the parameter values for each subfault. However, inverting for the number of subfaults required to capture the complexity of a large fault can force the problem to be computationally prohibitive. Here we introduce an alternative method of modal decomposition of the pre-stress and slip-weakening distance describing the initial conditions on the fault plane. Inversion of the strong-motion data using a direct search method known as the neighborhood algorithm (Sambridge, 1999) then seeks to optimize linear combinations of the modes that produce ruptures with a low misfit to the data. Our results indicate that parameterization of the fault plane in terms of component waves requires substantially fewer parameters to achieve a satisfactory fit to the data than parameterization by subfaults. The reduced dimensionality of the parameter space allows for more thorough exploration of the parameter space and accelerated convergence, resulting in a more efficient and robust inversion. In addition, a linear combination of orthogonal modes necessarily imposes a smoothness on the parameter distributions, which is an improvement over previous unphysical patch subdivision of the fault.
DE: 7209 Earthquake dynamics (1242)
DE: 7250 Transform faults
DE: 7290 Computational seismology
SC: Seismology [S]
MN: Fall Meeting 2005