Three-dimensional controlled-source electromagnetic and magnetotelluric joint inversion

Abstract

S U M M A R YThe growing use of the controlled-source electromagnetic method (CSEM) and magnetotellurics (MT) for exploration applications has been driving the development of data acquisition technologies, and three-dimensional (3-D) modelling and imaging techniques. However, targeting increasingly complex geological environments also further enhances the problems inherent in large-scale inversion, such as non-uniqueness and resolution issues. In this paper, we report on two techniques to mitigate these problems. We use 3-D joint CSEM and MT inversion to improve the model resolution. To avoid the suppression of the resolution capacities of one data type, and thus to balance the use of inherent, and ideally complementary information content, different data reweighting schemes are proposed. Further, a hybrid model parametrization approach is presented, where traditional cell-based model parameters are used simultaneously within a parametric inversion. The idea is to limit the non-uniqueness problem, typical for 3-D imaging problems, in order to allow for a more focusing inversion. The methods are demonstrated using synthetic data generated from models with a strong practical relevance.Large-scale inverse problems are usually underdetermined, meaning that there are more unknowns, typically in the form of highly digitized model meshes, than data. This adds to the problem that errors are associated with every geophysical datum. The resulting issue is referred to as the problem of non-uniqueness of inverse solutions. To mitigate this problem and to improve the resolution in an inversion, it is common to take advantage of complementary natures of different geophysical data sets. In electromagnetic problems, magnetotelluric (MT) data usually provides the conductivity information on a more gross scale, while controlled-source electromagnetic methods (CSEM) have a better ability of illuminating rather subtle targets, particularly thin resistors. MT data have been successfully combined with time-domain CSEM data to invert for one-dimensionally layered models (Hui-Ping et al. 1996;Meju 1996;Rovetta et al. 2008). With CSEM data responding stronger to thin resistive targets, the combination with MT data has a strong relevance for providing a less ambiguous interpretation of data measured over hydrocarbon prospects. Mackie et al. (2007) report a proof of concept showing the improved resolution by combining marine CSEM and MT synthetic data for mapping thin resistors.Even with improved resolution capabilities, the solutions of 3-D large-scale cell-based (or pixel-based) inversions with finely sampled models usually remain non-unique. Several strategies have been reported to limit the ambiguities for reconstructed targets and its conductivities. For cell-based problems, model-smoothing constraints are commonly applied, limiting the solutions to a class of geologically more meaningful ones, that is, avoiding conductivity variations that are unphysical. A different approach is to actually address the underdeterminacy by casting the...

Efficient Inversion of Multi-frequency and Multi-source Electromagnetic Data: Final report

“…Furthermore, experience inverting multiple data types (e.g. Commer & Newman 2009) demonstrates that multiple tradeoff parameters may be required to allow for differential weighting of disparate data types. And, one approach to joint inversion is to enforce structural similarity between two or more distinct physical parameters (e.g.…”

Three-Dimensional Magnetotelluric Inversion: An Introductory Guide for Developers and Users

“…Because of its minimal memory requirement, NLCG for MT inversion has been gaining in popularity in the past decade as shown in many publications (Rodi and Mackie 2001;Newman and Alumbaugh 2000;Newman and Boggs 2004;Commer and Newman 2008;Kelbert et al 2008;Commer and Newman 2009). Most of these algorithms use the Polak-Ribiere formula.…”

“…This is closer to a least-squares inverse problem and often produces a rougher model. The second functional is used in most algorithms, for example, in NLCG (e.g., Newman and Alumbaugh 2000;Rodi and Mackie 2001;Mackie and Watts 2007;Commer and Newman 2009;Zhanxiang et al 2010), in GN (e.g., Haber et al 2000, Sasaki 2001, 2004, Gunther et al 2006, in QN (e.g., Haber 2005; Avdeev and Avdeeva 2009), and in GN-CG (e.g., Mackie and Madden 1993;Newman and Alumbaugh 1997;Gunther et al 2006;Siripunvaraporn and Egbert 2007). In these algorithms, λ must be pre-selected and its value remains fixed.…”

Integrating Electromagnetic Data with Other Geophysical Observations for Enhanced Imaging of the Earth: A Tutorial and Review

“…Here U data m ð Þ and U reg m ð Þ are the data misfit and regularization terms as used by virtually all geophysical inversion approaches (e.g., Pedersen 1977;Constable et al 1987;Oldenburg 1990;Commer and Newman 2009;Fichtner and Trampert 2011;Menke 2012) and k is the Lagrange parameter for the regularization. The additional term U coupling m ð Þ mathematically defines the relationship between different subsets of model parameters.…”

“…Naturally, the discussion applies also to combining electromagnetic and gravity data (e.g., Maier et al 2009), but to a certain degree even when inverting different types of electromagnetic data together (Commer and Newman 2009;Haroon et al 2015) or electromagnetic with DC resistivity data (e.g., Candansayar and Tezkan 2008;Yogeshwar et al 2012;Hoversten et al 2016). Even though in the latter cases the physical parameter under consideration is electrical conductivity, magnetotellurics is sensitive to horizontal conductivity, whereas controlled-source methods are sensitive to horizontal and vertical conductivity which leads to effective anisotropy in finely layered sedimentary environments .…”