Spatiotemporal characteristics of the displacement field revealed with principal component analysis and the mode-rotation technique

Abstract

Principal component analysis (PCA) was applied to the displacement field recorded at 147 GPS stations in the Kanto-Tokai district, Japan, for the period from January 1999 to February 2003. Some prominent geophysical events occurred during this period. Using PCA, we attempted to decompose the displacement field into some representative spatial modes and their corresponding temporal modes to quantitatively extract the primary geophysical events that have caused rapid crustal deformation in this region. The displacement field was roughly decomposed into modes corresponding to the following three major events: (1) The Miyake-Kozu volcanic event, which occurred in the region south of Tokyo in 2000. It included large-scale dyke intrusion and intensive earthquake swarm between the Miyakejima and Kozushima islands. (2) The Tokai slow-slip event, which is considered to have begun just after the Miyake-Kozu volcanic event and is ongoing as of November 2005. It occurred on the boundary between the subducting Philippine Sea Plate and the overriding Eurasian (Amurian) Plate in the Tokai district. (3) The Boso slow-slip event, which lasted for approximately 10 days in October 2002. It occurred on the boundary between the subducting Pacific Plate and the overriding North American Plate near the Boso Peninsula. In the decomposition process, we combined the mode-rotation technique with ordinary PCA. This is because PCA causes contamination between modes, i.e., a mode obtained with PCA is contaminated by other modes. For example, the first principal mode corresponding to the Miyake-Kozu volcanic event, which is by far the largest in magnitude, was clearly seen in many other modes. In order to remove such contamination, we applied the mode-rotation technique, in which the principal axes derived from PCA are rotated so as to minimize the correlation coefficients between different temporal modes for a time period in which some prominent change occurs. Combining the mode-rotation technique with PCA exhibited a better performance in separating the displacement field.