Multi-temporal SAR interferometry (MTInSAR) techniques are able to derive displacement maps and displacement time series over coherent objects on the Earth, used for monitoring either geophysical ground deformation or structural instabilities. Nowadays, several datasets are available at different wavelengths, spatial resolutions, and revisit time, collectively covering long time periods (even more than 20 years). In particular, short revisit times, by improving the temporal sampling, make theoretically possible to catch high-rate and non-linear kinematics, which typically characterize warning signals like, for instance, those related to pre-failure of artificial infrastructures.
However, MTInSAR algorithms generally fit the displacement signal by using a linear model, which is computationally convenient, and also, more robust than higher-order models for what concerns the errors affecting the InSAR phase. Moreover, the analysis of the MTInSAR products is often performed by only considering the mean displacement rate, computed over the monitoring period, which is the information typically displayed on the displacement maps. However, this approach suffer for two main limitations: 1) by analysing time series covering long time periods (as more and more common with the datasets available nowadays), it is more probable that nonlinear displacements occur; 2) the mean displacement rate, though useful for distinguishing between stable and unstable areas on the ground, often mask interesting nonlinear signals, useful for early warning. Therefore, in order to fully exploit the content of MTInSAR products, methods are needed for automatically identifying relevant changes along displacement time series, and, consequently, classifying the targets on the ground according to their kinematic regime. This also allows performing a more reliable ground deformation spatial analysis, by distinguishing among spatial patterns of different kinematics.
Recently, approaches have been proposed to tackle this problem, which use different strategies. This work presents some general considerations and a possible new procedure based on statistical characterization of displacement time series, which allows, under certain constraints, recognising automatically different kinematic classes, and estimating the relevant parameters useful for target characterization in time.
The theoretical formulation of the proposed method is introduced, then the algorithm performance is evaluated under simulated scenarios, by varying simulation parameters such as type of kinematic model, temporal sampling rate, number of observations, and signal to noise ratio. Finally, some results of the analysis of displacement time series derived by MTInSAR processing of real datasets are presented.