Novel algorithms for pair and pixel selection and atmospheric error correction in multitemporal InSAR

Abstract

Entering the SAR's golden era began with the launch of Sentinel-1A/B satellites in 2014 and 2016 with 6–12 day revisit time, much larger stacks of high-resolution SAR images are available over a given area to perform time series analysis. Algorithms that deal with large stack sizes face several challenges, including interferometric phase quality degradation due to signal decorrelations, phase closure error caused by applied multilooking, and tropospheric phase delay. Here, we present an improved SBAS-type algorithm suitable for processing a large stack of SAR images at an arbitrary resolution. We develop a new pair selection strategy that applies dyadic downsampling combined with widely used Delaunay Triangulation to identify an optimal set of interferometric pairs that minimize systematic errors due to short-lived signals and closure errors. We develop and apply a novel statistical framework that selects elite pixels accounting for distributed and permanent scatterers. Also, we implement a new tropospheric error correction that takes advantage of smooth 2D splines to identify and remove error components with fractal-like structures. We demonstrate the effectiveness of the algorithms by applying them to 3 large datasets of Sentinel-1 SAR images measuring non-linear surface deformation over various terrains. Compared with independent GNSS observations, we find that over the rural/natural terrains adjacent to San Andreas fault in southern California, our approach yields a standard deviation of 0.48 cm for time series differences in both ascending and descending tracks. While in urban areas, such as Los Angeles, standard deviation difference with GNSS time series is 0.30 cm.