ALBERT

Description: Large-scale 3D modeling of realistic earth models is being increasingly undertaken in industry and academia. These models have proven useful for various activities such as geologic scenario testing through seismic finite-difference (FD) modeling, investigating new acquisition geometries, and validating novel seismic imaging, inversion, and interpretation methods. We have evaluated the results of the Western Australia (WA) Modeling (WAMo) project, involving the development of a large-scale 3D geomodel representative of geology of the Carnarvon Basin, located offshore of WA’s North West Shelf (NWS). Constrained by a variety of geologic, petrophysical, and field seismic data sets, the viscoelastic WAMo 3D geomodel was used in seismic FD modeling and imaging tests to “validate” model realizations. Calibrating the near-surface model proved to be challenging due to the limited amount of well data available for the top 500 m below the mudline. We addressed this issue by incorporating additional information (e.g., geotechnical data, analog studies) as well as by using soft constraints to match the overall character of nearby NWS seismic data with the modeled shot gathers. This process required undertaking several “linear” iterations to apply near-surface model conditioning, as well as “nonlinear” iterations to update the underlying petrophysical relationships. Overall, the resulting final WAMo 3D geomodel and accompanying modeled shot gathers and imaging results are able to reproduce the complex full-wavefield character of NWS marine seismic data. Thus, the WAMo model is well-calibrated for use in geologic and geophysical scenario testing to address common NWS seismic imaging, inversion, and interpretation challenges.

Description: A key goal in industry and academic seismic research is overcoming long-standing imaging, inversion, and interpretation challenges. One way to address these challenges is to develop a realistic 3D geomodel constrained by local-to-regional geologic, petrophysical, and seismic data. Such a geomodel can serve as a benchmark for numerical experiments that help users to better understand the key factors underlying — and devise novel solutions to — these exploration and development challenges. We have developed a two-part case study on the Western Australia (WA) Modeling (WAMo) project, which discusses the development and validation of a detailed large-scale geomodel of part of the Northern Carnarvon Basin (NCB) located on WA’s North West Shelf. Based on the existing regional geologic, petrophysical, and 3D seismic data, we (1) develop the 3D geomodel’s tectonostratigraphic surfaces, (2) populate the intervening volumes with representative geologic facies, lithologies, and layering as well as complex modular 3D geobodies, and (3) generate petrophysical realizations that are well-matched to borehole observations point-wise and in terms of vertical and lateral trends. The resulting 3D WAMo geomodel is geologically and petrophysically realistic, representative of short- and long-wavefield features commonly observed in the NCB, and leads to an upscaled viscoelastic model well-suited for high-resolution 3D seismic modeling studies. In the companion paper, we study WAMo seismic modeling results that demonstrate the quality of the WAMo geomodel for generating shot gathers and migration images that are highly realistic and directly comparable with those observed in NCB field data.

Description: We present an inversion method for the recovery of the geometry of an arbitrary number of geological units using a regularized least-squares framework. The method addresses cases where each geological unit can be modeled using a constant physical property. Each geological unit or group assigned with the same physical property value is modeled using the signed-distance to its interface with other units. We invert for this quantity and recover the location of interfaces between units using the level-set method. We formulate and solve the inverse problem in a least-squares sense by inverting for such signed-distances. The sensitivity matrix to perturbations of the interfaces is obtained using the chain rule and model mapping from the signed-distance is used to recover physical properties. Exploiting the flexibility of the framework we develop allows any number of rocks units to be considered. In addition, it allows the design and use of regularization incorporating prior information to encourage specific features in the inverted model. We apply this general inversion approach to gravity data favoring minimum adjustments of the interfaces between rock units to fit the data. The method is first tested using noisy synthetic data generated for a model comprised of six distinct units and several scenarios are investigated. It is then applied to field data from the Yerrida Basin (Australia) where we investigate the geometry of a prospective greenstone belt. The synthetic example demonstrates the proof-of-concept of the proposed methodology, while the field application provides insights into, and potential re-interpretation of, the tectonic setting of the area.

Description: To reduce uncertainties in reconstructed images, geological information must be introduced in a numerically robust and stable way during the geophysical data inversion procedure. In the context of potential (gravity) data inversion, it is important to bound the physical properties by providing probabilistic information on the number of lithologies and ranges of values of possibly existing related rock properties (densities). For this purpose, we introduce a generalization of bounding constraints for geophysical inversion based on the alternating direction method of multipliers (ADMM). The flexibility of the proposed technique enables us to take into account petrophysical information as well as probabilistic geological modeling, when it is available. The algorithm introduces a priori knowledge in terms of physically acceptable bounds of model parameters based on the nature of the modeled lithofacies in the region under study. Instead of introducing only one interval of geologically acceptable values for each parameter representing a set of rock properties, we define sets of disjoint intervals using the available geological information. Different sets of intervals are tested, such as quasi-discrete (or narrow) intervals as well as wider intervals provided by geological information obtained from probabilistic geological modeling. Narrower intervals can be used as soft constraints encouraging quasi-discrete inversions. The algorithm is first applied to a synthetic 2D case for proof-of-concept validation and then to the 3D inversion of gravity data collected in the Yerrida basin (Western Australia). Numerical convergence tests show the robustness and stability of the bound constraints we apply, which is not always trivial for constrained inversions. This technique can be a more reliable uncertainty reduction method as well as an alternative to other petrophysically or geologically constrained inversions based on more classical “clustering” or Gaussian-mixture approaches.