PP and PS joint inversion with a posterior constraint and with particle filtering

The Bayesian framework works well in amplitude versus offset (AVO) inversion, which merges multi-information together to generate posterior distributions of P-wave velocity, S-wave velocity and density. Most existing AVO inversion methods utilize PP reflection seismic data to predict the three elastic parameters. These methods are not usually sensitive to S-wave velocity and density, which make the inversion methods inaccurate and unstable. One way of solving these problems is to perform PP and PS joint inversion by incorporating PS seismic data. Another way is to provide a relatively accurate prior model. In this paper, we apply a particle filtering technique to produce a prior model for the PP and PS joint inversion. In the Bayesian inversion setting, the prior model works as the regularization term. Particle filtering is a Bayesian recursive method that combines prior information with observed data to provide a posterior constraint to reduce the joint inversion's uncertainty. We generate synthetic models with different signal-to-noise ratios to validate our new method. Comparisons are provided with the traditional joint inversion, which adopts the Gaussian prior model. The inversion results show that the three elastic parameters are retrieved well when the signal-to-noise ratios are high. As the signal-to-noise ratio reduces, our new method can depict more detailed changes than the traditional inversion method, and improves the inversion accuracy apparent in the target layers. (paper)