Design, implement, and evaluate a 3D seismic velocity model Inversion

Thesis Abstract

This study addresses the persistent challenge of accurately characterizing subsurface velocity structures essential for reliable seismic imaging, by designing, implementing, and evaluating a three-dimensional (3D) seismic velocity model inversion framework that integrates heterogeneous data sources and robust inversion strategies to reduce velocity-model uncertainty and enhance depth migrations. The aim is to develop a reproducible workflow that yields high-resolution velocity tomograms with quantified uncertainty, suitable for industrial-scale exploration and academic research. Specific objectives include (1) constructing a flexible 3D forward-modeling engine that couples ray-based travel-time modeling with full-waveform inversion for selected survey geometries, (2) implementing regularization and prior-information strategies to stabilize inversion in sparsely sampled regions, (3) incorporating Petrophysical constraints via rock-physics relations to improve velocity-density consistency, and (4) validating the framework against both synthetic benchmarks and field data to assess accuracy, robustness, and transferability. The methodology adopts a mixed-methods design grounded in geophysical modeling and quantitative analysis. The population comprises synthetic models, field-seismic datasets from a posteriori known geology, and publicly available laboratory rock-physics measurements. A two-stage sampling approach is employed first, synthetic test-ceds with varying heterogeneity levels (low, moderate, high) are generated to probe inversion fidelity; second, field data from a operated onshore/offshore survey consisting of 120 000 travel-time picks and 2 500 relevant full-waveform records are curated, with a stratified sub-sample of 30% for detailed inversion experiments. Data collection instruments include travel-time pickers, waveform simulators, and rock-physics compilation pipelines drawing on published AVO/AVP relations and porosity-velocity correlations. The analysis workflow integrates (i) a 3D forward-modeling engine combining ray-based eikonal solvers and finite-difference time-domain simulations, (ii) a Bayesian inversion framework employing Markov chain Monte Carlo to quantify parameter uncertainty, and (iii) a physics-informed regularization scheme using priors derived from lithology and porosity logs. The study also applies cross-validation with independent borehole data where available. Validity and reliability are addressed through synthetic-to-field transfer tests, sensitivity analyses, and uncertainty propagation, with performance metrics including root-mean-square error (RMSE) of velocity, structural similarity index (SSIM) for tomographic slices, and credible interval coverage probabilities. Key expected findings include demonstrable improvements in 3D velocity resolution in structurally complex zones, reduced misfit between observed and modeled travel times by at least 25–40% relative to baseline inversions, and coherent velocity-density trends aligned with established rock-physics models. The Bayesian framework is anticipated to yield credible intervals that accurately reflect uncertainties arising from data gaps and model nonlinearity, while the integration of priors is expected to stabilize inversion in regions with sparse ray coverage. Comparative analysis of synthetic and field results should reveal the importance of combining travel-time and full-waveform information to resolve velocity anisotropy and small-scale heterogeneity. The study contributes to knowledge by delivering a transparent, reproducible 3D velocity inversion workflow that explicitly handles uncertainty quantification and incorporates lithology-constrained priors. It advances practical understanding of how multi-source data integration improves seismic velocity models and their reliability for depth migration and reservoir characterization, providing transferable methods for both academic research and industry practice. The main conclusion is that a physics-informed Bayesian 3D inversion framework, when backed by rock-physics priors and combined with hybrid forward modeling, delivers superior velocity models with quantified uncertainty, enabling more accurate seismic imaging and better risk assessment in subsurface exploration. Recommendations include implementing the workflow in open-source software to foster replication, extending the approach to anisotropic inverse problems, and integrating additional data types such as gravimetric or electromagnetic measurements to further constrain velocity models.