Improved Combined Scatterers Interferometry With Optimized Point Scatterers (ICOPS) for Interferometric Synthetic Aperture Radar (InSAR) Time-Series Analysis

Abstract

Despite the development of point-based interferometric synthetic aperture radar (InSAR) time-series methods, such as persistent scatterer interferometry (PSI), it remains difficult to identify reliable measurement points (MPs) in areas under vegetated cover. In this study, we developed a new algorithm, improved combined scatterers interferometry with optimized point scatterers (ICOPS), to monitor changes in surface deformation by combining persistent scatterers (PSs) and distributed scatterers (DSs). The algorithm was subsequently improved through the application of a machine learning process. A generalized likelihood ratio test (GLRT) was applied to identify statistically homogeneous pixels suitable for a small SAR dataset. A machine learning-based optimization process was used to overcome the increase in MPs after the combined scatterers interferometry (CSI) process. The MP optimization process used the support vector regression (SVR) algorithm to find the optimal point from all MPs in the dataset. To improve the reliability of the MP surface deformation mapping after optimization, the optimized hot-spot analysis (OHSA) method was applied to obtain spatially clustered MPs; this was possible despite a low deformation identification rate. To demonstrate the effectiveness of the ICOPS method, we applied this approach to Yellowstone Lake in the USA using 95 SAR images from the Sentinel-1 satellite that was taken during the period of 2017–2020. The CSI method produced an increase in MP density, especially in areas not covered by PSI, which indicated its ability to detect deformations in mountainous areas. Comparisons with GPS data and traditional methods produced promising results with an accuracy of 1 cm/year in terms of the root mean square error (RMSE). The optimization process in CSI also had the advantage of retrieving helpful information in the MP dataset and increased the accuracy of CSI by 12%. Deformation mapping using the optimized results provided new insights regarding the spatial clustering of surface deformation in MPs. This could provide the foundation for developing a posttime-series optimization process based on point scatter as a surface change detection tool.