2D Parameterized Inversions of Marine CSEM Data

As an alternative to the computational and algorithmic complexity of regularized inversion, we investigate the use of lightly parameterized 2D inversions of marine controlled source electromagnetic (CSEM) data. Since a priori information from seismics is often available, this overdetermined scheme offers a way to make use of additional information in an inversion that will run faster than a regularized, underdetermined one. The challenge for these simple inversions comes in finding parameterizations of conductivity models that are soluble, stable, and representative of realistic geologic structures. However, an advantage of such algorithms is that they can solve for sharp boundaries. We explore the use of a Levenberg-Marquardt type optimization to find a sharp boundary model with the smallest L2 norm misfit to the data. The advent of 2D forward modeling algorithms that use adaptively refined finite element meshes [Li and Key, 2007] enables automatic re-meshing of the inverted model structure between inversion iterations, ensuring numerical accuracy and providing great flexibility in the evolution of model structure. We test strategies for parameterization of several different scenarios using synthetic and real data, and study the stability of these inversion methods.