An Approach to improvement of the quality of magnetotelluric data based on continuous wavelet transform and independent component analysis

Ogawa, Hiroki; Asamori, Koichi; Hama, Yuki*

It is generally difficult to improve the quality of magnetotelluric (MT) data if noise is correlated between the electromagnetic fields. Independent component analysis (ICA) is a statistical method for transforming the observed data into signals that are mutually as independent as possible. This paper proposes the way to process MT data accompanied with intentional coherent noise in the time-frequency domain by utilizing continuous wavelet analysis and frequency-domain ICA. At each frequency, two procedures are conducted that are composed of lowering the effect of noise through ICA applied to the wavelet coefficients of horizontal electromagnetic components and selecting these reconstructed coefficients in terms of S/N ratio. As a result of the remote reference processing using the calculated spectra, the apparent resistivity curve is clearly improved, while the way to handle the separated signals leaves much to be desired in the ICA.