Separation of Gravity Anomaly Data Considering Statistical Independence of Source Signals

Abstract

Gravity anomaly is one of efficient methods to evaluate underground structure, which is essential for estimation of ground motion due to earthquakes. Data observation is, however, costly since it requires expensive devices. In order to overcome this problem, Morikawa et al. have been working to develop a mobile gravimeter that uses force-balance (FB) accelerometer. In comparison to the conventional spring type gravimeters, it is less costly, compact and can be carried by relatively small carriers. However, it raised problems that the observed data are severely contaminated by various kinds of disturbances such as engine vibration and carrier motion. Therefore an appropriate data processing method for extracting gravity anomaly signal from such observed data is required. For that purpose, we propose to use the statistical independence property of gravity anomaly and other noisy data. The gravity anomaly and other noises are generated from different sources and it can be safely assumed that they are independent. As a scheme of considering independence of signals, blind source separation techniques are used. Second Order Blind Identification method (SOBI) separates the target sources by assuming that source and noises are un-correlated at various time-lags. Similarly, Independent Component Analysis (ICA) separates the sources by maximizing the independence of linearly transformed observed signals. An ICA algorithm namely ThinICA is proposed that implements the maximization of independence among source signals at various time-lags and thus incorporates the advantages of both SOBI and ICA. The proposed method is applied to the data observed at Toyama Bay, Japan. It is observed that the motion of carrier (ship) influences the performance of de-noising algorithm. Under certain favorable data acquisition environment, the proposed method was able to salvage the gravity anomaly data from the noise-contaminated data with the accuracy sufficient for the purpose of identification of gravity anomaly distribution.