Probabilistic Estimation of InSAR Displacement Phase Guided by Contextual Information and Artificial Intelligence

Abstract

Phase unwrapping, also known as ambiguity resolu-1 tion, is an underdetermined problem in which assumptions must 2 be made to obtain a result in SAR interferometry (InSAR) time 3 series analysis. This problem is particularly acute for distributed 4 scatterer InSAR, in which noise levels can be so large that 5 they are comparable in magnitude to the signal of investigation. 6 In addition, deformation rates can be highly nonlinear and orders 7 of magnitude larger than neighboring point scatterers, which 8 may be part of a more stable object. The combination of these 9 factors has often proven too challenging for the conventional 10 InSAR processing methods to successfully monitor these regions. 11 We present a methodology which allows for additional environ-12 mental information to be integrated into the phase unwrapping 13 procedure, thereby alleviating the problems described above. 14 We show how problematic epochs that cause errors in the tempo-15 ral phase unwrapping process can be anticipated by the machine 16 learning algorithms which can create categorical predictions 17 about the relative ambiguity level based on the readily available 18 meteorological data. These predictions significantly assist in the 19 interpretation of large changes in the wrapped interferometric 20 phase and enable the monitoring of environments not previously 21 possible using standard minimum gradient phase unwrapping 22 techniques. 23

Disentangling Shallow Subsidence Sources by Data Assimilation in a Reclaimed Urbanized Coastal Plain, South Flevoland Polder, the Netherlands

“…Temporal decorrelation is caused by atmospheric variability and changes in the physical and geometric properties of the scatter points, for example, due to seasonal changes in vegetation which result in land cover changes (Ferretti et al., 2007; Hanssen, 2001). As a result, vegetation‐rich areas are suboptimal for the analysis of subsidence by satellite imaging (Conroy et al., 2022). Therefore, the point‐wise time series of the provided InSAR data are largely located on man‐made structures because these scatter points face less decorrelation issues.…”

A Treatise on InSAR Geometry and 3-D Displacement Estimation

“…Finally, SAR acquisitions from different viewing geometries are never taken at the same moment, and since deformation phenomena, by definition, change over time this will result in incomparable displacement parameters. Especially for rapidly changing deformation phenomena such as landslides [20] or highly dynamic soils [21], it may be impossible to assume that observations from different epochs are comparable.…”

Bridging Loss-of-Lock in InSAR Time Series of Distributed Scatterers

“…The identified segments are initially treated as independent time-series. Temporal phase unwrapping (or ambiguity res-olution) is performed independently on each segment using a method aided by a machine learning model, as described by [22]. The ground surface level of peatlands is extremely unstable and prone to rapid fluctuations depending on temperature and precipitation levels, so we use a recurrent neural network (RNN) to aid in making predictions about which ambiguity level is correct.…”

“…This RNN model uses temperature, precipitation, and day of year as inputs, which is publicly available daily weather data. Detailed information about the implementation and testing of the methodology is provided in [22].…”