Optimizing Seismic Inversion Results Through Fully SeismicDriven Background Model Generation

The optimization of inversion algorithms, coupled with increasing high-performance computing capability, favour the rapid availability of project deliverables for prospect delineation, stratigraphic interpretation, or investigation of the applicability of machine learning-based solutions such as lithofacies or rock type prediction. Quantitative interpretation experts face a series of challenges when running model-based seismic inversion. One of them is the construction of the low frequency model. Because seismic data is bandwidth limited, the low frequency trend is usually obtained from well data; one option is to propagate that trend through a geocellular model and use it as the initial guess in a deterministic seismic inversion workflow. However, project deadlines, technology, and even asset choices may have an impact on the quality of the geologic model, which can result in an over-simplified structural and stratigraphic representation of the reservoir. The improvements in imaging algorithms enable the development of new strategies for increasing the use of seismic data in the construction of a more geologically consistent background model. The level of detail from seismic data clearly invalidates the application of a simple geostatistical approach when representing the micro-layering within a reservoir, especially if we want to consider the presence of discontinuities such as faults, intrusive bodies, or any unconformable event in a single formation. High-resolution models, enabled by geo-cellular modelling workflows, create inversion results that demonstrate greater consistency with input seismic data. This paper discusses two different approaches to the conventional workflow used to create the low-frequency model: 1. A structural seismic guided interpolation algorithm that maintains the structural trends observed in the seismic data. 2. An advanced reservoir modelling approach, paired with a volumetric reservoir-scale interpretation workflow. The use of a Convolutional Neural Network (CNN) to perform an automatic interpretation of internal horizons is introduced which can further accelerate the geologic model building.