Inversion of TEM data affected by fast-decaying induced polarization: Numerical simulation experiment with homogeneous half-space

• Published in: Journal of applied geophysics 2008-10, Vol.66 (1), p.31-43
• Main Authors: Kozhevnikov, N.O., Antonov, E.Yu
• Format: Article
• Language: English
• Subjects: Inversion; Permafrost; TEM method
• ISSN: 0926-9851; 1879-1859
• DOI: 10.1016/j.jappgeo.2008.08.001

A numerical simulation experiment was applied to explore the potential and limitations of the inversion of the TEM response to a uniform half-space with a complex frequency-dependent conductivity described by the Cole–Cole model. The experiment was made by two stages. First we assessed the quality of inversion of 100 m × 100 m coincident-loop transient response to a uniform polarizable half-space and then did comparison studies of both separate and joint inversion as applied to the synthetic data for two loop configurations of different sizes. Both of these stages included (1) choice of half-space models and loop configurations; (2) computing synthetic transients and generating pseudo-empirical responses by adding random noise; (3) inversion of pseudo-empirical transients; (4) discussion. During each stage inversion was performed twice: first one of the authors did inversion without any knowledge of the “true” models and then the same author repeated the procedure after he got some prior information from the other author. Inversion of 100 m × 100 m coincident-loop transients showed that at low chargeability m and/or exponent c the TEM response can be fitted by the 1D conductive non-polarizable models. Increasing m and/or c resulted in unambiguous IP manifestations even though no prior information was known. More than a half of the recovered models, though some of them differ strongly from the underlying ones, provide a good idea of the “true” ones. The Cole–Cole parameters found by inversion with prior information turned out to approach those of the true models even at low m and c. Joint inversion of data from two loop configurations of different sizes biased effectively the models obtained separately for each configuration toward the true ones. With some prior information, joint inversion provides correct estimates of the Cole–Cole parameters even if the true chargeability is as low as only 0.02. In the case that joint inversion was applied, square error misfit between pseudo-experimental and predicted data was controlled predominately by the fitting error for the very rapidly changing non-monotonous smaller-loop TEM response.