Recent Advances in 2D Frequency-Domain Marine Controlled-Source EM Inversion Using Adjoint Approximate Sensitivities

Abstract

Marine controlled-source electromagnetic (CSEM) inversion is a widely used technique for imaging subsurface structures beneath the seafloor and exploring hydrocarbon or mineral resources. Although essential for data interpretation, the inversion process is computationally intensive, particularly due to the high memory demands and time consumption associated with sensitivity calculations, even in 2D cases. This paper presents recent advances in frequency-domain marine CSEM inversion algorithms, including a novel method for approximating sensitivities that significantly accelerates 2D inversion without relying on exact sensitivity matrices. A 2.5D frequency-domain CSEM forward modeling solver is employed as a testbed, where the secondary-field formulation is adopted to mitigate source singularities and enhance numerical accuracy. In the developed 2D inversion routine, the sensitivity matrix is computed via the adjoint-state method, integrating the inner product of the adjoint and forward electric fields over each cell during each iteration. The inversion is performed using the Gauss–Newton optimization scheme. To improve computational efficiency, adjoint fields are evaluated over a simplified layered model rather than the full multidimensional true model. Numerical experiments demonstrate that the proposed sensitivity approximation achieves nearly identical inversion performance to that of the exact sensitivities. Results from both synthetic and experimental datasets confirm that the approximate sensitivities are sufficiently accurate to ensure convergence toward geologically meaningful solutions.