[S-EM16] Electromagnetic Induction in the Earth and Planetary Interiors, and Tectono-Electromagnetism

[SEM16-04] An attempt to reduce near field effect for Boso MT data using MSSA (Multi-channel Singular Spectrum Analysis)

*Shu Kaneko¹, Toru Mogi², Chie Yoshino², Katsumi Hattori^2,3,4 (1.Graduate school of Science and Engineering, Chiba University Japan, 2.Graduate School of Science, Chiba University Japan, 3.Center for Environmental Remote Sensing, Chiba University, Japan, 4.Disaster Medicine Research Institute, Chiba University, Japan)

Keywords:MT method, Multi-channel Singular Spectrum Analysis, Noise reduction, Boso peninsula

In 2014 ~ 2016, the MT (magnetotelluric) survey was conducted to study the subsurface electrical resistivity structure in Boso Peninsula, Japan, such as electric property of a part of the focal zone of the Genroku earthquake and the Taisho Kanto earthquake and a part of the slow slip region. However, due to the location of the area, the obtained data is affected by manmade noise such as leak current from DC-driven trains, factories, and power lines. The noise breaks the assumption of the MT method due to the distance from the noise source to the observation site being too nearby. It is called the near-field effect. One of the specific properties of the noise is that the noise correlates between measured magnetic and electric fields. The conventional noise reduction method, BIRRP, was applied to the data to process the noise, but due to the influence that originated from the noise, the MT response function was not sufficiently estimated. Therefore, the new noise reduction method is needed to obtain a reasonable MT result. Thus, Multi-channel Singular Spectrum Analysis (MSSA) is introduced. MSSA can decompose C channels time series into (C×window length (WL)) principal components (PCs) by performing singular value decomposition of the trajectory matrix, which is created by shifting the C-channel time series data by one data which column is window length L. Therefore, MSSA is unique because it does not require basis functions and is helpful for natural electromagnetic field signals, which are non-stationary processes.

The following is an overview of this method for short-period signals (period of about 40 to 300 seconds).
1. Decompose 7ch data (Hx, Hy, Hz, Ex, Ey, Rx, and Ry, H, E, and R are measured magnetic field, electric fields, and reference magnetic field, respectively, and x, y, and z are south to north, west to east, and vertical downward component, respectively) with application MSSA to the raw data into (7 × WL) PCs.
2. Create the detrended time series by summing PCs that have short-period signals.
3. Decompose7ch detrended time series with application MSSA into (7 × WL) PCs.
4. Create MT signal by summation selecting high correlation PCs that have large amplitude.

The effectiveness of the method is confirmed by applying this method to real Boso MT data (12:00-14:00 (JST)) using the magnetic field of the JMA memambetsu observatory as the reference. It is found that this method can remove about 1 nT of artificial noise in the magnetic field and obtain clean magnetic field data. As for the electric field, it is found that this method can remove about 1/2 of the original amplitude of the noise. After cleaning MT data with MSSA, the effectiveness in MT analysis is checked by Robust Remote Reference. As a result, the MT sounding curve became smoother than those results by the BIRRP. It indicates the effectiveness of the proposed noise reduction.