Singular Value Decomposition denoising of airborne Time Domain ElectroMagnetic data in anthropized area
Résumé
In anthropized areas, airborne Time-Domain ElectroMagnetic (TDEM) data may be affected by spikes, oscillations, shifts and/or bumps. Since EM noise spectrum is complex, thresholding and stacking standards techniques are hardly efficient in such environment. Time-consuming and subjective manual cleaning is therefore required. We propose an alternative fast and efficient user-assisted filtering approach. We adapted the Singular Value Decomposition (SVD) to denoise TDEM data. The SVD uses principal component analysis to separate noise and geological components of the signal; the latters are used to reconstruct the EM decays with the geological signal only. The SVD procedure was applied on an anthropized area, in the center of France. The comparison between each reconstructed decay and its corresponding measured decay allowed rejecting efficiently mainly spikes and oscillations. In a second phase, an ad hoc analysis of the map of weights of the "noisy components" showed high correlation with man-made installations and provided a tool to reject mainly most likely soundings biased by shifts. Bumped decays were also localized based on the analysis of specific SVD components. The established denoising procedure provides accurate denoising tools and makes, at least, the manual cleaning much less time consuming and less subjective.