## Publication 156

##
Robust magnetotelluric inversion

###
Matsuno, T., A.D. Chave, A.G. Jones, M.R. Muller, and R.L. Evans

### Abstract

A robust magnetotelluric (MT) inversion algorithm has been developed on the basis of
quantile-quantile (q-q) plotting with a confidence band and statistical modelling of inversion
residuals for the MT response function (apparent resistivity and phase). Once outliers in the
inversion residuals are detected in q-q plots with confidence bands and the statistical
modelling with the Akaike Information Criterion, they are excluded from the inversion data
set and a subsequent inversion is implemented with the culled data set. The exclusion of
outliers and the subsequent inversion is repeated until the q-q plots are substantially linear
within confidence bands, outliers predicted by the statistical modelling are unchanged from
the prior inversion, and the misfit statistic is unchanged at a target level. The robust inversion
algorithm was applied to synthetic data generated from a simple 2-D model and observational
data from a 2-D transect in southern Africa. Outliers in the synthetic data, which come from
extreme values added to the synthetic responses, produced spurious features in inversion
models, but were detected by the robust algorithm and excluded to retrieve the true model.
An application of the robust inversion algorithm to the field data demonstrates that the
method is useful for data clean-up of outliers, which could include model as well as data
inconsistency (for example, inability to fit a 2-D model to a 3-D dataset), during inversion
and for objectively obtaining a robust and optimal model. The present statistical method is
available irrespective of the dimensionality of target structures (hence 2-D and 3-D
structures) and of isotropy or anisotropy, and can operate as an external process to any
inversion algorithm without modifications to the inversion program.

### Source

*Geophysical Journal International*, **196**, 1365–1374, doi:10.1093/gji/ggt484.
[PDF]

MTNet Home Page
Alan's Home Page

Alan G Jones / 14 February 2014 /
alan-at-cp.dias.ie