Polarisation filtering of magnetotelluric data - Using an advanced wavelet processing scheme to discriminate between contribution of signal and noise to the data

The magnetotelluric (MT) method investigates the structure of the Earth by studying its vertical and lateral electric conductivity distribution. For that purpose natural electromagnetic (EM) fields are measured at Earth’s surface, and thus derive a spatially and frequency dependent impedance response function that can be modelled in terms of Earth structure. Long period natural EM fields (>1 s) are generated by the interaction of electrical charged particles radiated from the Sun with the Earth’s magnetosphere and ionosphere. In phases of low solar activity the source signal for MT is weak, especially at longer periods (>1,000 s) and the effects of noise can result in poor response function estimates. A significant contribution to noise can be cultural sources fixed in space, such as mining areas, electric fences and television transmitters. Electromagnetic waves generated by such sources exhibit a preferential polarisation ellipticity and direction that differs from the natural signals generated at the Earth’s outer magnetosphere and ionosphere. In MT, the ellipticity and direction of the polarisation can be determined because the magnetic component of the electromagnetic field is measured in orthogonal directions. In addition, the continuous wavelet transform (CWT) analysis is an efficient way to localize segments of chosen polarisation in the recorded dataset in both time and frequency. We have developed an algorithm that selects data segments according to their polarisation properties allowing us to improve the signal-to-noise ratio of MT responses. After rejecting segments of certain polarisation direction and therefore low signal-to-noise ratio for the signals we wish to record, the remaining data can be used for subsequent conventional MT processing. Using synthetic data and a MT dataset collected during the PICASSO fieldwork campaign in Spain in 2007, we test our pre-processing algorithm. In this paper we present a comparative analysis and results from this study.