Variational Bayesian Independent Component Analysis for InSAR Displacement Time-series with Application to Central California, USA

The exploitation of ever increasing Interferometric Synthetic Aperture Radar (InSAR)16 datasets to monitor the Earth’s surface deformation is an important goal of today’s geodesy.17 In this study our observations consist of deformations along the Line-Of-Sight (LOS) di18 rection of the satellite. Our observations are the result of the combination of a multi19 tude of sources (either volcano-tectonic or non-tectonic deformation). In most cases, we20 are facing a Blind Source Separation (BSS) problem. Natural approaches to tackle BSS21 problems are multivariate statistical techniques that attempt to decompose the dataset22 into a limited number of statistically independent sources, under the assumption that23 the different physical mechanisms contributing to the observations have independent foot24 prints in space and/or time. We show the capabilities of a variational Bayesian Indepen25 dent Component Analysis (vbICA) algorithm in dealing with synthetic InSAR time se26 ries and compare it to the commonly used FastICA algorithm. We explore the effective27 ness of the spatial and temporal mode decompositions. We apply vbICA on data rela28 tive to the San Joaquin Valley (SJV) and the Central San Andreas Fault (CSAF), Cal29 ifornia, spanning the time range 2015/03/01-2019/07/14. The proposed approach likely30 isolates the contribution of shallow and deep aquifers to the surface deformation as well31 as the elastic and inelastic deformation. We present a 1-dimensional compaction esti32 mation of the elastic and inelastic storage coefficients adopting a formalism that takes33 into account the last century water level history. Concerning the CSAF, the algorithm34 helps separating tectonic loading from seasonal behavior concentrated in the Quaternary35 sediments of the Salinas Valley.