3D Inversion-Based Interpretation of Marine CSEM Data

摘要

The marine controlled-source electromagnetic (CSEM) method has been evolving into a geophysical imaging tool for increasingly complex geological settings in which multiple resistive bodies can be resolved. An advanced subsurface imaging workflow for 3D CSEM surveys is presented, which reproduces the resistivity distribution to within a spatial resolution determined by the frequencies included. The performance of our advanced-processing workflow is demonstrated using a case study from the Gulf of Mexico (GOM), where a dense 3D grid was acquired over an area of high-quality seismic data and well log control. At the core of our 3D workflow is an inversion methodology with approximate Hessian-based optimization and a fast finite-difference time-domain forward operator. The optimization matches the synthetic to the measured field within 100-200 iterations and is sufficiently robust in three dimensions to avoid expensive regu-larization schemes. The sensitivity of the gradient-based inversion to the starting model is addressed by investing considerable effort in building ID inversion-based starting models. At the same time, 3D inversion algorithms for survey layouts, including azimuthal data, demand high-quality data conditioning for which we present a processing sequence from time-domain electromagnetic data acquired by seabed receivers to frequency-domain data and weights for inversion. Detection and delineation of reservoirs in the presence of salt is recognized as a major challenge to CSEM methods. In order to interpret 3D data accurately in such a complex environment, the true-resistivity cube is built from a sequence of constrained inversion-based interpretation steps. Using a data set acquired in 2008 in the GOM, we demonstrate the ability of our 3D technology to resolve small (2x2 km), low-resistivity pay (Ap<5£lm) targets within the vicinity (< 1 km) of large salt bodies. In this case study, the 3D method converged within 1 week, running on 150 parallel nodes.