Application of higher-order statistics in magnetotelluric data processing

Magnetotelluric (MT) responses some times have scattering at individual frequencies with big error bars or shape distortion, which lower the effectiveness of MT curves in describing underground electrical structures. The main reason for this is that natural MT signal is usually weak, and the traditional data processing method has weakness. This paper uses the higher-order statistics method to test the Gaussianity of MT signals. The results show that the signals are non-Gaussian. On the basis that the higher-order statistics vanish for any Gaussian process, this paper uses higher-order spectra to reconstruct power spectra, then estimates MT responses using reconstructed power spectra, to effectively suppress Gaussian colored noise. Synthetic tests show that the method presented in this paper is superior to the power spectral method in suppressing Gaussian colored noise, and can extract more useful information.