Seismic travel time tomography has been proven as an effective tool in reconstruction of Earth's p-wave velocity model in offshore projects. Geological and geotechnical data are collected in 1D through direct sampling with high vertical resolution and seismic travel time tomography data are acquired with lower vertical resolution, yet higher aerial coverage, to help us detect the structural events of the study area and establish the statistical relationship between seismic inversion output, i.e. velocity model, and geo-mechanical parameters of the soil. Data of different resolution and scale are to be integrated to infer the underlying velocity field which may be further correlated to other geotechnical and geo-mechanical parameters. Constructing an initial model of velocity is an essential step that requires interpolating p-wave velocity in spatial domain utilizing velocity logs data. In deterministic methods, the initial model is generated through Kriging interpolation and is perturbed until it converges to a model that regenerate the recorded travel time data with an acceptable error band. In this study, we propose a probabilistic method to invert the seismic travel time tomography data which utilizes geostatistical priors, i.e. models that shows the spatial pattern and continuity of the underlying velocity field inferred from limited amount of data in the boreholes. The method starts with a Bayesian analysis over the Kriging mean and covariance model parameters, and posterior samples from this step is input to a geostatistical simulator that generates realistic prior models as an input to probabilistic inversion process. The proposed method is applied on a synthetic ground model of p-wave velocity. The final output of the proposed method is the mean and standard deviation of the p-wave velocity field.
Moslem Moradi and Zenon Medina-Cetina
Keywords: Bayesian Inversion, Geostatitsics, Seismic Inversion, Travel Time Tomography
Publisher: DFI