Magnetotelluric data processing: a case study

Magnetotelluric (MT) data collected simultaneously at one or more sites may be processed by a number of different methods. Such methods attempt to remove or suppress the effect of noise on the data channels. The desired results are accurate, unbiased and repeatable estimates of the impedance tensor as a function of frequency and location. In this study we perform an investigation of the analysis of an MT data set sampled at 5 s. Both single-site (SS) and remote-reference (RR) techniques are employed to estimate the impedance tensor Z. Two biased SS estimates of Z are used to compare the performance of five coherence­ based acceptance criteria. It is demonstrated that the RR predicted coherence between local fields can be used for selecting data windows, and provides a necessary assessment of the reliability of a given RR estimate. It is demonstrated that the variance of an RR estimate depends strongly on the local signal-to-noise ratios (as monitored by the local predicted coherence) and depends weakly on the number of data windows, as long as coherences are above a moderate threshold. Although, for our data, an estimate of Z obtained using a remote electric field is grossly inaccurate, its associated predicted coherence is as efficient in selecting low-noise-level data windows as its counterpart obtained using a remote magnetic field. The relation between SS and RR predicted coherences, the latter estimated using both electric and magnetic fields, is investigated. A hybrid selection technique that uses a remote electric field is suggested.