Comparison of deterministic and stochastic approaches to crosshole seismic travel-time inversions  

作　　者：YanZhe Zhao YanBin Wang 

机构地区：[1]Department of Geophysics, School of Earth and Space Sciences, Peking University

出　　处：《Earth and Planetary Physics》2019年第6期547-559,共13页地球与行星物理（英文版）

基　　金：supported by the National Natural Science Foundation of China (grant nos. 41930103 and 41674052)

摘　　要：The Bayesian inversion method is a stochastic approach based on the Bayesian theory.With the development of sampling algorithms and computer technologies,the Bayesian inversion method has been widely used in geophysical inversion problems.In this study,we conduct inversion experiments using crosshole seismic travel-time data to examine the characteristics and performance of the stochastic Bayesian inversion based on the Markov chain Monte Carlo sampling scheme and the traditional deterministic inversion with Tikhonov regularization.Velocity structures with two different spatial variations are considered,one with a chessboard pattern and the other with an interface mimicking the Mohorovicicdiscontinuity(Moho).Inversions are carried out with different scenarios of model discretization and source–receiver configurations.Results show that the Bayesian method yields more robust single-model estimations than the deterministic method,with smaller model errors.In addition,the Bayesian method provides the posterior probabilistic distribution function of the model space,which can help us evaluate the quality of the inversion result.The Bayesian inversion method is a stochastic approach based on the Bayesian theory. With the development of sampling algorithms and computer technologies, the Bayesian inversion method has been widely used in geophysical inversion problems. In this study, we conduct inversion experiments using crosshole seismic travel-time data to examine the characteristics and performance of the stochastic Bayesian inversion based on the Markov chain Monte Carlo sampling scheme and the traditional deterministic inversion with Tikhonov regularization. Velocity structures with two different spatial variations are considered, one with a chessboard pattern and the other with an interface mimicking the Mohorovi?i? discontinuity(Moho). Inversions are carried out with different scenarios of model discretization and source–receiver configurations. Results show that the Bayesian method yields more robust single-model estimations than the deterministic method, with smaller model errors. In addition, the Bayesian method provides the posterior probabilistic distribution function of the model space, which can help us evaluate the quality of the inversion result.

关 键 词：stochastic APPROACH DETERMINISTIC APPROACH Bayesian inversion MARKOV Chain MONTE Carlo Tikhonov REGULARIZATION 

分 类 号：P[天文地球]