HR: 1340h
AN: S43C-1023 [Abstracts]
TI: Two-Step MCMC Seismic Tomography
AU: * Liu, J
EM: liu34@purdue.edu
AF: EAS Department, Purdue University, 550 Stadium Mall Drive, West Lafayette, IN 479072051
United States
AU: Braile, L
EM: braile@purdue.edu
AF: EAS Department, Purdue University, 550 Stadium Mall Drive, West Lafayette, IN 479072051
United States
AB:
Damping is commonly used to stabilize the tomography result in conventional linear travel time inversion methods. Concern
exists with the subjective damping factor selection and unreliability of solution uncertainty derived from linear inversion.
In this study, a nonlinear approach, the Two-Step Markov Chain Monte Carlo method (MCMC) is applied to the travel time
tomography problem. In the first step of the MCMC, no damping or smoothing is applied, so as to find the solution and its
uncertainty objectively from the observed data. As the estimated uncertainty is not only related to the data error but also
the investigation geometry (ray coverage) and the parameterization, the appropriate cell size is found first. The grid is
gradually changed from coarser to finer, and the RMS error between the observed data and the predicted data from the mean
solution will become smaller, and the estimated uncertainty increases. Once the RMS error won_t change significantly, the
grid size is considered to be appropriate. The estimated uncertainty (variance) from the first step is used as the damping
factor in the second step MCMC when smoothing is applied. In this way, the reliable model information (good ray coverage)
extracted from the observed data will be less affected by the smoothing process. This new approach is applied to a complex
2-D data set recorded over a fault zone, including refraction, wide-angle reflection and CMP reflection data. Simultaneous
inversion of these data improves the model information and hence the resolution. The MCMC method also provides reasonable
estimates of model uncertainty.
DE: 3200 MATHEMATICAL GEOPHYSICS (New field)
DE: 3260 Inverse theory
SC: Seismology [S]
MN: 2004 AGU Fall Meeting