[S-EM12] Electric, magnetic and electromagnetic survey technologies and scientific achievements

[SEM12-P06] Attempt of noise reduction in magnetic field data based on MC-NMF

*Rei Amano¹, Tada-nori Goto¹ (1.University of Hyogo)

Keywords:Magnetic filed, Signal separation, Noise reduction, Nonnegative matrix factorization : NMF

The magnetotelluric (MT) is one of the electromagnetic survey methods for the estimation of deep subsurface structures by observing natural fluctuations of geomagnetic and electric fields. It is known to be difficult to obtain high quality data when there is strong artificial noise in the survey area. Remote reference (RR) has been conventionally applied to reduce the effects of regional noise. The method conducts simultaneous measurements at two or more points located far from each other (mainly magnetic field fluctuations). However, it is difficult to remove noise when noise is continuous mixed into observation data. In this study, we use a method called Multi-Channel Nonnegative Matrix Factorization (MC-NMF), which can extract characteristic spectral patterns. The noise separation and reduction were attempted using MC-NMF.
we conducted application experiments on synthetic data obtained by adding noise to the magnetic field data from the Kakioka Magnetic Observatory. The performance of signal separation was evaluated by decomposing the synthetic data. The usefulness of MC-NMF was also confirmed by comparison with the results of the application of singular value decomposition. Furthermore the result of reconstructing the data by deleting the basis containing a lot of noise shows it was possible to remove the added noise. It can be concluded the MC-NMF is effective for noise reduction.