Development of noise reduction method based on MSSA (Multi-channel Singular Spectrum Analysis) -Application to near field noise observed in Boso Peninsula, Japan-

The Boso Peninsula is one of the most tectonically active regions in Japan due to its location on a plate boundary. Focal zones of the past enormous earthquakes (1703 Genroku Kanto Earthquake (M8.2), the 1923 Taisho Kanto Earthquake (M7.9)) and slow slip events (SSE) region are located southwest and southeast of the peninsula, respectively. Therefore, it is important to survey the subsurface resistivity structure of these regions from a geophysical point of view. The magnetotelluric (MT) survey was conducted to clarify the resistivity structure from 2014 to 2016. However, observed MT data was contaminated by artificial noise sources (e.g., leak current from DC-driven trains and power lines). Conventional noise reduction methods using remote reference (Gamble et al., 1979) and robust statistics such as BIRRP (Chave and Thomson. 2004) are inadequate to deal with the noise. The reason is that the noise included in Boso MT data is originated from a near field source and is coherent between the magnetic field and the electric field.

Therefore, we propose a NEW method using MSSA(Multi-channel Singular Spectrum Analysis) to reduce the influence of the noise. MSSA can decompose multiple time series to several principal components (PCs). In our new method, choosing PCs based on the correlation between each component, they are discriminated into trend components, quasi-periodic components (=interested MT signal), and noise components. We applied MSSA to 7ch (5ch for observation site data (horizontal magnetic field, vertical magnetic field, and horizontal electric field) and 2ch for reference site data (horizontal magnetic field)) to extract 'clean' MT data from noisy Boso MT data. In this presentation, the results of time series and MT analysis applying this method will be presented.