Seismic Reservoir Characterization With Implicit Neural Representations

Abstract

AbstractSeismic reservoir characterization refers to a set of techniques aimed at estimating static and/or dynamic physical properties of subsurface geological formations from seismic data, with applications ranging from exploration and development of energy resources, geothermal production, carbon capture and storage, and the assessment of geological hazards. Extracting quantitative information from seismic recordings, such as an acoustic impedance model, is however a highly ill‐posed inverse problem due to the band‐limited and noisy nature of the seismic data. As such, seismic inversion techniques strongly rely on additional prior information that penalizes (or promotes) some features in the recovered model. This paper introduces IntraSeismic, a novel hybrid seismic inversion method that combines implicit neural representations as an effective way to parameterize a subsurface model with the physics of the seismic modeling operator. We demonstrate its effectiveness in both pre‐stack and 3D/4D post‐stack seismic inversion using synthetic and field data sets. Key features of IntraSeismic are (a) unparalleled performance in 2D and 3D pre‐stack/post‐stack dynamic and static seismic inversion, (b) rapid convergence rates, (c) ability to seamlessly include hard constraints (i.e., well data) and perform uncertainty quantification, and (d) potential data compression and fast randomized access to specific portions of the inverted model.