IntraSeismic: a coordinate-based learning approach to seismic inversion

Abstract

Seismic imaging is the numerical process of creating a volumetric representation of the subsurface geological structures from elastic waves recorded at the surface of the Earth. As such, it is widely utilized in the energy and construction sectors for applications ranging from oil and gas prospection, to geothermal production and carbon capture and storage monitoring, to geotechnical assessment of infrastructures. 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 data. This paper introduces IntraSeismic, a novel hybrid seismic inversion method that seamlessly combines coordinate-based learning with the physics of the post-stack modeling operator. Key features of IntraSeismic are i) unparalleled performance in 2D and 3D post-stack seismic inversion, ii) rapid convergence rates, iii) ability to seamlessly include hard constraints (i.e., well data) and perform uncertainty quantification, and iv) potential data compression and fast randomized access to portions of the inverted model. Synthetic and field data applications of IntraSeismic are presented to validate the effectiveness of the proposed method.

Figures (15)

• Figure 1: Schematic representation of the IntraSeismic framework: The input to IntraSeismic consists of spatial coordinates that are processed through a nonlinear mapping module. Initially, this module maps the input into a higher-dimensional space of trainable feature vectors using multiresolution hash encoding. Subsequently, these vectors are input to a MLP that outputs the impedance model. In the subsequent physics modeling module, the output from the MLP is added to the background model. A modeling operator then computes the predicted data from the impedance model, which is then incorporated into the loss function.
• Figure 2: Inversion of the synthetic 2D Marmousi dataset with a noise level of $\sigma=0.1$: a) seismic data, b) background model, c) Tikhonov regularization, d) TV-regularization using the Primal-Dual solver, e) DIP, f) DIP with additional priors, g) PnP, h) IntraSeismic, and i) ground truth.
• Figure 3: SNR evolution through iterations for IntraSeismic and benchmark methods in the inversion of the Marmousi data.
• Figure 4: Inversion of the synthetic 3D SEAM dataset: a) seismic data, b) background model, c) Tikhonov regularization, d) TV-regularization using the Primal-Dual solver, e) DIP, f) DIP with additional priors, g) IntraSeismic and h) ground truth.
• Figure 5: SNR evolution through iterations for IntraSeismic and benchmark methods in the inversion of the SEAM data.