Denoising Application of Magnetotelluric Low-Frequency Signal Processing
Jin Li, Fanhong Ma, Jingtian Tang, Yecheng Liu, Jin Cai
Abstract
As magnetotelluric (MT) is an important method for exploring the geoelectrical structure of the underground, it has motivated in-depth research and application by many geophysicists. Nevertheless, due to the influence of the environment, the collected data are interfered with strong humanistic noise, which might result in a loss of their authenticity. To solve the above problems, many time-frequency domain methods have emerged. Based on the advantages of singular value decomposition (SVD) denoising, we propose a method of magnetotelluric noisy data processing based on multiresolution singular value decomposition (MSVD) and improved complete ensemble empirical mode decomposition with adaptive noise (ICEEMDAN), which overcomes the lack of flexibility in the construction of the matrix in SVD data processing. First, we introduce a new signal processing method by generalizing SVD to MSVD to obtain more accurate signal characteristics. Due to the difficulty of matrix selection, we suggest the singular value contribution rate as the standard to determine the suitable Hankel matrix and use MSVD to perform effective decomposition. Second, we propose the ICEEMDAN algorithm for removing impulse noise, which efficiently processes each modal component through adaptively decomposition of different thresholds. Experiments on synthetic and realistic data demonstrate that our proposed method can separate the large-scale contours of the magnetotelluric noisy data and improve the time-domain waveform quality of low-frequency signal. The apparent resistivity-phase curves, coherence and SNR are all obviously promoted.