Journal Article

The advantages of logarithmically scaled data for electromagnetic inversion

Brent Wheelock, Steven Constable and Kerry Key

in Geophysical Journal International

Volume 201, issue 3, pages 1765-1780
ISSN: 0956-540X
Published online April 2015 | e-ISSN: 1365-246X | DOI: https://dx.doi.org/10.1093/gji/ggv107
The advantages of logarithmically scaled data for electromagnetic inversion

More Like This

Show all results sharing these subjects:

  • Geophysics
  • Oceanography and Hydrology

GO

Show Summary Details

Preview

Non-linear inversion algorithms traverse a data misfit space over multiple iterations of trial models in search of either a global minimum or some target misfit contour. The success of the algorithm in reaching that objective depends upon the smoothness and predictability of the misfit space. For any given observation, there is no absolute form a datum must take, and therefore no absolute definition for the misfit space; in fact, there are many alternatives. However, not all misfit spaces are equal in terms of promoting the success of inversion. In this work, we appraise three common forms that complex data take in electromagnetic geophysical methods: real and imaginary components, a power of amplitude and phase, and logarithmic amplitude and phase. We find that the optimal form is logarithmic amplitude and phase. Single-parameter misfit curves of log-amplitude and phase data for both magnetotelluric and controlled-source electromagnetic methods are the smoothest of the three data forms and do not exhibit flattening at low model resistivities. Synthetic, multiparameter, 2-D inversions illustrate that log-amplitude and phase is the most robust data form, converging to the target misfit contour in the fewest steps regardless of starting model and the amount of noise added to the data; inversions using the other two data forms run slower or fail under various starting models and proportions of noise. It is observed that inversion with log-amplitude and phase data is nearly two times faster in converging to a solution than with other data types. We also assess the statistical consequences of transforming data in the ways discussed in this paper. With the exception of real and imaginary components, which are assumed to be Gaussian, all other data types do not produce an expected mean-squared misfit value of 1.00 at the true model (a common assumption) as the errors in the complex data become large. We recommend that real and imaginary data with errors larger than 10 per cent of the complex amplitude be withheld from a log-amplitude and phase inversion rather than retaining them with large error-bars.

Keywords: Inverse theory; Magnetotellurics; Marine electromagnetic

Journal Article.  12826 words.  Illustrated.

Subjects: Geophysics ; Oceanography and Hydrology

Full text: subscription required

How to subscribe Recommend to my Librarian

Users without a subscription are not able to see the full content. Please, subscribe or login to access all content. subscribe or login to access all content.