Multichannel singular spectrum analysis of InSAR datasets: data-adaptive
interpolation and decomposition of Sentinel-1 time series at Pacaya
Volcano, Guatemala
Abstract
This paper presents the first application of multichannel singular
spectrum analysis (M-SSA) to radar satellite geodesy. We apply M-SSA to
Sentinel-1 Interferometric Synthetic Aperture Radar (InSAR) time series
processed for Pacaya Volcano in Guatemala in two steps. First, we
produce, in an iterative and data-adaptive way, estimates of missing
data points to obtain evenly sampled time series. The resulting time
series are then decomposed with M-SSA into long-periodic nonlinear
trends and oscillatory modes providing a sparse representation of the
signals present in the data. The M-SSA approach presented herein is
designed to deal with very large datasets such as collections of InSAR
time series. Combining M-SSA with power spectrum analysis show that the
dominant frequencies of the main oscillatory modes correspond to 1, 1.5,
2, 3, 5.8 and 6.8 cycle per years. These frequencies are consistent with
the seasonal variability of the regional hydrological system, as
determined from correlograms of rainfall time series and M-SSA modes
extracted from time series of regional gravity anomalies using Gravity
Recovery and Climate Experiment (GRACE) data, Global Navigation
Satellite Systems (GNSS) time series recorded in Guatemala City, and
phase delay maps derived from a global weather model. While some of the
seasonal oscillations correlate well with topography, others show
significant spatial asymmetries. The extracted nonlinear trends show
large amplitudes around the summit and within the area covered by the
2014 lava flows and, to a lesser extent, the 2010 lava flows. This
nonlinear trend correlates with interannual variability of the regional
water cycle.