Adaptive hybrid MSVD-VMD method for robust denoising of magnetotelluric signals

The magnetotelluric (MT) method utilizes naturally occurring electromagnetic fields at the Earth’s surface to investigate subsurface electrical structures at depth. However, low-frequency MT signals are often severely contaminated by complex noise, compromising both data quality and interpretational accuracy. To address this challenge, we propose a novel two-stage denoising framework that integrates Multiresolution Singular Value Decomposition (MSVD) with Variational Mode Decomposition (VMD). In the first stage, MSVD is employed to suppress high-amplitude and broadband noise while preserving the essential structural features of the original signal. The preliminarily denoised signal is then fed into the second stage, where an adaptive VMD algorithm, guided by the Thomson formula and equipped with a dynamic decomposition threshold, is applied to effectively eliminate residual impulsive noise and mitigate modal aliasing. Experimental evaluations on both synthetic and field MT data demonstrate that the proposed method achieves superior multiscale noise separation and significantly enhances the time-domain fidelity of MT waveforms. Furthermore, validation using apparent resistivity and phase responses reveals that the proposed method consistently outperforms conventional robust estimation techniques in terms of interpretational reliability and accuracy.