| dc.description.abstract | Offshore site characterization for geotechnical construction projects relies on geological and geophysical survey data because geotechnical sampling techniques have limitations for offshore environments. However, the seismic images, the observed data from the most popular geophysical surveys for offshore exploration, are only reflected signals in time-domain and still difficult to apply for the geohazards evaluation in space-domain. Thus, offshore geo-site characterization requires the seismic inversion methods, which can convert from observed wave signals into the soil and rock properties along with the depth. However, this spatial inverse problem contains the mathematical non-uniqueness problem, so we need a probabilistic approach to quantify the uncertainty. The probabilistic site characterization in this study is based on the stochastic process in the Bayesian framework. The spatial Gaussian process and reversible jump Markov chain Monte Carlo (rj-MCMC) methods are developed to utilize the offshore geological drilled borehole data and geophysical seismic survey data. Bayesian inference, which integrates the observed data, model predictions, and expert’s beliefs, supports the geospatial analysis to provide probabilistic descriptions of the model parameters. It is required for the inverse problem to understand the quantified uncertainty from all model parameters and widely used for the nonlinear geophysical seismic inverse problems. This dissertation focuses on the probabilistic inversions with geophysical seismic data and starts from a synthetic offshore shallow case study to integrate geological and geophysical surveys for better ground model estimation. Geological borehole data provides the initial condition of the stratigraphy information, and geophysical seismic inversion finds out the model parameters, bulk densities, acoustic P-wave velocities, and the depth of layer interfaces. The spatial random field from Gaussian process supports the probabilistic seismic inversion, and this study introduces a new approach to reconstruct the subsurface ground model information in high-resolution. Practical applications are developed from this case study by applying the rjMCMC method, which recently showed a great advantage in geophysical studies. The main idea of the rj-MCMC method is to define the unknown subsurface geologic layers as another random variable, and design hierarchical Bayesian priors to support the convergence of the dimensions during the stochastic process. Three case studies in varying dimensions are discussed for the geological offshore stratigraphic modeling, geophysical stochastic post-stack seismic inversion, and geomechanical Bayesian full-waveform inversion (FWI). The rj-MCMC methods in those case studies lead the probabilistic site characterization to find the correct target modeling parameters, which is challenging for the unknown subsurface estimation. One of those studies is also applied to field data near Sigsbee Escarpment, a steep marine slope in the Gulf of Mexico. The results show the two-dimensional subsurface image of the ground model under the seafloor, which can help us to avoid geohazard prone areas. | en |
