Methodology and Techniques - PSI

InSAR has a great capability for retrieving the deformation time series of huge amounts of ground objects on a bi-/weekly basis. Such time series unveil the evolution of geophysical processes, anthropogenic hazards and the structural health situation of public infrastructure. It is important to understand these processes in order to reduce the impact of natural and anthropogenic hazards and infrastructure degradation.

Unfortunately, time series modelling is not straightforward. Especially in urban infrastructure monitoring, every single radar scatterer may have its own dynamic behavior, irrespective of its neighbors, which implies that the applied functional models may differ for each point. Consequently, the problem is ill-posed unless additional constraints are introduced.

We have demonstrated that a probabilistic approach [1] can be used to determine the most probable time-series model of every InSAR measurement point. We use multiple hypothesis testing, given a library of potential physically realistic deformation models and a complete stochastic model of the measurements. We use the Gauss-Markov model to describe the functional and stochastic model, and we implicitly assume that the unknown parameters are deterministic and uncorrelated with each other. We consider the noise of the measurements due to atmospheric influence, sensor noise, and data processing errors, to determine the stochastic model.

However, in practice, adjacent InSAR measurement points may exhibit a homogeneous or smooth behavior, in either space or time. This implies that the parameters of interest may be correlated and considered to be stochastic variates.  Without considering this parameter signal correlation information, parameter estimation would be biased and therefore unreliable. Yet, simply applying global smoothing/multi-looking in space or time to filter the signal, is too harsh and will invoke more biases. Therefore, in the current study we propose to apply regularization in the parameter estimation, per cluster of points, based on available a priori signal information. As the signal information cannot be derived directly from the InSAR measurements, we obtain this information from other external sources (expert elicitation) and use them as constraints. This approach improves the accuracy, precision and reliability of the InSAR results. We demonstrate this approach both via simulations and on real data.

[1] Chang, L., Hanssen, R.F., 2015. A probabilistic approach for InSAR time series postprocessing. IEEE transactions on Geoscience and Remote Sensing 54, 421–430.

Using persistent scatterer (PS) time-series InSAR, deformation of objects can be measured in order of millimeters. However, the exact physical nature and location of each scatterer is poorly known. Unlike conventional geodesic methods, PS scatterers are generally not pre-defined receivers or benchmarks. The occurrence of PS is strongly dependent on the specific orientation, geometry, and other characteristics of objects on the earth’s surface, in relation to the parameters of the transmitted radar signals (e.g. direction, wavelength, polarization). Thus, though high-precision deformation estimates can be achieved, these uncertainties are a limitation to the use of this technique.

One solution to solve this problem is to estimate the 3D coordinates of scatterers by multi-baseline datasets, like persistent scatterer Interferometry[1, 2], Stereo-SAR[3], or SAR tomography[4]. However, the estimated positions, which are in order of several meters in cross-range direction for PS-InSAR, are still insufficient to detailed interpretation. Stereo-SAR requires the identification of (physically) identical scatterers, visible in both imaging geometries, which is not always possible for data stacks from different orbital tracks. SAR tomography only distinguish scatterers if the distance between scatterers is longer than the Rayleigh resolution in elevation[5]. Another way is to extract physical information of scatterers (size, material and temperature etc.) by building the time series amplitude function [6, 7], which also requires to solve the phase ambiguities of the scatterers. Consequently, the most important problem still is the understanding the origin and nature of PS, and the accurate estimation of its position. 

Here, we attempt to improve our understanding of scattering mechanisms in an urban context in a new way, by simulating urban landscapes with varying level-of-detail (LOD),see Fig.1. We use a 3D SAR simulator based on Ray-tracing[8] to predict the radar scattering by illuminating a 3D scene by a known SAR sensor. The ’rays’ can follow multiple reflections within the object scene, yielding some ’points’ to behave as PS point scatterers. These potential scatterers will be predicted and localized. As the detected scatterers change with various level of detail (LOD) 3D models[9], we will explore the LOD effect on the identified scatterers. This yields useful information to improve the interpretation of actual PSI results, since it can be assessed whether specific elements of, e.g., a building will behave as PS or not. We report on the differences observed by illumination from various direction, as well as the differences due to different radar sensors. The simulated signals with their 3D coordinates may further support the connection between radar scatterers and real objects. 

REFERENCES

[1]  A. Ferretti, C. Prati, and F. Rocca. Permanent scatterers in SAR interferometry. IEEE Transactions on Geoscience and Remote Sensing, 39(1):8–20, January 2001.

[2]  S. Gernhardt, S. Auer, and K. Eder. Persistent scatterers at building facades–evaluation of appearance and localization accuracy. ISPRS Journal of Photogrammetry and Remote Sensing, 100:92–105, 2015.

[3]  C. Gisinger, U. Balss, R. Pail, X. X. Zhu, S. Montazeri, S. Gernhardt, and M. Eineder. Precise three-dimensional stereo localization of corner reflectors and persistent scatterers with terrasar-x. IEEE Transactions on Geoscience and Remote Sensing, 53(4):1782–1802, April 2015.

[4]  X. X. Zhu, S. Montazeri, C. Gisinger, R. F. Hanssen, and R. Bamler. Geodetic sar tomography. IEEE Transactions on Geoscience and Remote Sensing, 54(1):18–35, Jan 2016.

[5]  X.X.Zhu, R.Bamler. Demonstration of super-resolution for tomographic sar imaging in urban environment. IEEE Transactions on Geoscience and Remote Sensing, 50(8):3150–3157, 2012.

[6]  D. Perissin. SAR super-resolution and characterization of urban targets. PhD thesis, Politecnico di Milano, Italy, 2006.

[7]  P. Dheenathayalan and R.F. Hanssen. Radar target type classification and validation. In Geoscience and Remote Sensing Symposium (IGARSS), 2013 IEEE International, pages 923–926. IEEE, 2013.

[8]  S. Auer. 3D synthetic aperture radar simulation for interpreting complex urban reflection scenarios. PhD thesis, Technische Universita ̈t Mu ̈nchen, 2011.

[9]  F. Biljecki, H. Ledoux, and J. Stoter. An improved LOD specification for 3d building models. Computers, Environment and Urban Systems, 59:25–37, 2016. 

Due to the short revisit time and high data acquisition capacity of current satellites, much emphasis has recently been placed on the development of deformation monitoring and rapid disaster response capability using InSAR. This requires efficient, fast data processing, due to the need for timely updates on movements in the case of, for example, earthquakes and volcanic activity, and also to limit the computing resources required to process the vast quantities of data being acquired. High resolution is typically not a critical requirement in the case of volcanic or tectonic applications. In urban monitoring, however, differentiating between the movements of different buildings, or between buildings and the surrounding land, can be crucial, requiring processing of the data at the highest resolution possible. Here we present Rapid time series InSAR (RapidSAR), a method that can efficiently update high resolution time series of interferograms, and demonstrate its effectiveness over urban areas.

The RapidSAR method uses ensembles of neighbouring pixels with similar amplitude behaviour through time to estimate the coherence of pixels on an interferogram-by-interferogram basis. Newly acquired images can be rapidly ingested due to the individual coherence estimate, as the remainder of the time series does not have to be reprocessed. The coherence estimate does not suffer from smearing, as is the case with the conventional boxcar method. The timely, high quality coherence estimate makes the RapidSAR method suitable for urban monitoring. The individual point selection maximizes the amount of information extracted from the time series. The downside of this is that the selection of points for each individual interferogram varies, making the time series analysis more challenging. We overcome this by connecting points in both time and space.

We demonstrate the effectiveness of the method over urbanized areas. We show how the algorithm is able to successfully extract a high density of points in full Sentinel-1 resolution, and is able to distinguish coherent points on buildings from incoherent points surrounding them. We further examine the effectiveness of the time series estimation using the dense time series available from Sentinel-1. Finally, we show that the method is able to manage the high data volumes, both in space and time, generated by the mission. 

The German TerraSAR-X and TanDEM-X satellites are characterized by unique features such as providing images with high spatial resolution and an unprecedented geometric accuracy. The latter has been significantly improved in the recent years by quantifying and removing the most prominent error sources which affect radar range and azimuth time measurements, a method called imaging geodesy [1]. Moreover, if corrected time observations of a specific target are available from SAR acquisitions with different viewing geometries, it has been demonstrated that the stereo SAR method is capable of delivering 3-D absolute coordinates of the target with accuracies in the decimeter to centimeter regime, depending on the target being a corner reflector or an opportunistic persistent scatterer (PS) [2].

As a first step towards the inclusion of such accurately localized point targets into phase-based stacking methods, in [3] the concepts of imaging geodesy and stereo SAR were used to transform the relative estimates of SAR tomography (TomoSAR) into absolute 3-D point clouds by absolutely localizing the reference point. The improvement in the localization accuracy of the resulting point cloud has encouraged us to continue expanding the mentioned framework by automatic detection and absolute localization of useful PS candidates which are visible from SAR images acquired from different viewing geometries, either same-heading or cross-heading orbit tracks. This will generate multiple Ground Control Points (GCPs) which can be used as a reference network in multi-pass InSAR techniques for reliable estimation and removal of atmospheric phase screen and for support in phase-unwrapping. The availability of such points can also be relevant for non-InSAR applications such as detection of large magnitude motions which are invisible from InSAR time-series approaches or as tie points for improving the registration of remotely sensed optical images.

This contribution is dedicated to introducing an automated processing chain for generating GCPs from SAR data. The procedures start with the identification of high quality PS candidates, in some strategies with the aid of external data, to precisely extract PS timings in a stack of non-corregistered SAR images. The subsequent steps are the correction of PS timings and the absolute localization with the Stereo SAR method. It will also be demonstrated how these GCPs can be used to improve the geocoding accuracy of InSAR, in particular PSI, point clouds leading to absolute positions. Finally, preliminary results based on TerraSAR-X high resolution spotlight images over the city of Oulu, Finland are reported.

Bibliography:
[1]          M. Eineder, C. Minet, P. Steigenberger, X. Y. Cong, and T. Fritz, “Imaging Geodesy - Toward Centimeter-Level Ranging Accuracy With TerraSAR-X,” IEEE Transactions on Geoscience and Remote Sensing, vol. 49, no. 2, pp. 661–671, Feb. 2011.
[2]          C. Gisinger et al., “Precise Three-Dimensional Stereo Localization of Corner Reflectors and Persistent Scatterers With TerraSAR-X,” IEEE Transactions on Geoscience and Remote Sensing, vol. 53, no. 4, pp. 1782–1802, Apr. 2015.
[3]          X. X. Zhu, S. Montazeri, C. Gisinger, R. F. Hanssen, and R. Bamler, “Geodetic SAR Tomography,” IEEE Transactions on Geoscience and Remote Sensing, vol. 54, no. 1, pp. 18–35, Jan. 2016.

PS-InSAR interferometry is a well-established technique to estimate linear and non-linear ground displacements as well as the atmospheric phase screen (APS) in an InSAR data. It achieves the highest accuracy in measuring deformation. To reach this accuracy a high density of high quality points (PS) is required for model fitting. In rural regions, where the availability of PS points is limited, the density of high quality points is often too low to guarantee accurate results. To mitigate this, a common technique is to exploit distributed scatterers (DS) that have enough phase quality for it to be used in the PS-InSAR analysis. To increase the phase quality of DS and preserve PS points in the image scene, adaptive filtering of interferograms should be implemented prior to PS and DS candidate selection and PS-InSAR implementation.

This paper addresses the scattering property based adaptive filtering of dual polarized Sentinel-1 interferograms for application to permanent scatterer interferometry in a rural region. We first demonstrate the derivation of scattering mechanisms from 14 dual polarized Sentinel-1 data acquired between August 2015 and August 2016. We implemented an adaptive filtering procedure to estimate the complex coherences for different interferogram pairs to preserve PS and filter DS located in the image scene. We further implemented phase quality optimization to achieve high accuracy in differential phases for both PS and DS candidates. Finally, the selected candidates are processed jointly in a PS-InSAR processing work-flow to estimate ground deformation .

Preliminary results indicated that PS points were well preserved and that the signal to noise ratio of DS was increased by applying scattering property based adaptive filtering. Adaptive filtering and polarimetric optimization increased the number of pixels available for PS-InSAR analysis. A robust model fitting and a more reliable PS-InSAR analysis result is anticipated from the proposed filtering.