Robust processing for removing train signals from magnetotelluric data in central Italy.

Smooth magnetotelluric transfer functions are used to represent the frequency relationship between electric and magnetic data. This facilitates the identification and correction of individual outliers in the electric and magnetic data, and the construction of frequency and time weights needed for robust estimates of the transfer functions. Errors in the transfer function are found by jackknife estimates of the solution covariance. Band averages provide alternate estimates of the transfer function and the damping for stabilizing the smooth estimates. The method is tested for: artificial data with specified noise, electric and magnetic outliers, periodic variations, and transfer function; and data from the Larderello geothermal region in central Italy contaminated by electric train signals