Principal component analysis of InSAR data

English Abstract:

Geodetic data, the spatial and temporal surface expression of complex geophysical processes in the earth, is being acquired today at unprecedented rates and accuracies. Differential Interferometric Synthetic Aperture Radar (DInSAR) is a satellite remote sensing technique that is used extensively for mapping ground deformation with a high spatial resolution and sub-centimeter precision over a large area. Spatial resolution of the modern SAR sensors ranges from 1 to 20 m over areas from 10x10 km to 200x200 km. Images produced with advanced DInSAR techniques can contain millions of data points. Previous research has demonstrated that an eigenpattern decomposition technique known as principal component analysis (PCA) can be used to identify a unique, finite set of correlated deformation patterns for a given regional network of GPS stations. Similar in nature to the empirical orthogonal functions (EOF) historically employed in the analysis of atmospheric and oceanographic phenomena, the method derives the eigenvalues and eigenstates from the diagonalization of the covariance matrix. After decomposing large data sets into their orthonormal eigenvectors and associated time series, based upon the spatiotemporal relationships that exist in the data, the decomposition can be used to study those modes most responsible for the various signals and their sources or to remove those uninteresting modes in the system. One additional benefit is that these modes are ordered naturally by the predominant spatial wavelength of the signal and can be used to identify the potential anthropogenic, volcanic and tectonic sources responsible for those signals. Here we demonstrate its application to DInSAR data from several different regions and its ability to isolate the handful of important signals in the thousands of data points in remote sensing images.