Singular Spectrum Analysis for Modeling Geodetic Time Series

Abstract

Geodetic time series obtained via space-geodetic techniques, e.g. site displacements from continuous gps observations and aggregated basin water storages from grace, display a seasonal behavior. Much focus has been given to separate such periodic signal from other signals buried in the geodetic time series, e.g. linear or non-linear trends. Conventionally, these seasonal signals are retrieved with constant amplitudes by the classical least squares estimation approach. Recently, singular spectrum analysis (ssa) has been successfully applied to extract the time variable seasonal signals from the gps time series. This study aims to extend the application of ssa to other geodetic time series. Through two examples, i.e. water level time series from satellite altimetry and grace-derived basin averaged equivalent water height time series, the capabilities of ssa to separate the non-linear trend from seasonal signals are demonstrated. In addition, it is shown that the so-called w-correlation analysis is beneficial in determining the optimal embedding window size, which is a key parameter during ssa analysis.