Robust smooth magnetotelluric transfer functions

Abstract

Robust estimates of the magnetotelluric (MT) transfer function are found using an iterative reweighted method on time series data corrected for outliers and gaps. The MT transfer function, composed of several analytic functions smoothly varying in frequency, is used to represent the frequency-domain relationship between electric and magnetic time series. The smoothly varying transfer function facilitates identification and removal of electric and magnetic outliers (spikes), construction of the frequency-and time-domain weights used for obtaining robust smooth and band-averaged estimates, and separation of the time series into MT and correlated noise signals if a remote site exists that is free of the correlated noise. Errors in the transfer function are calculated using jackknife estimates of the solution covariance. The method is tested on: time series from a relatively clean MT site in central California; a test time series based on Tucson magnetic time series plus synthetic noise for a given transfer function; and time series from the Larderello geothermal region in central Italy where there are strong signals from d.c. electrified railways.