InSAR Theory and Techniques (2)

Most recent SAR sensors (including TerraSAR-X or Cosmo-Skymed) use wide band signals (typically 150 MHz) to achieve metric range resolution. One can also take advantage of wide band to split it in sub-bands and generate several lower-resolution images from a single acquisition, each being centered on a slightly different frequency. This process, named Multi Chromatic Analysis (MCA), corresponds to performing a spectral analysis or a spectral decomposition of SAR images. Spectral decomposition is performed by extracting sub-bands in range frequency domain, each extracted sub-band leading to a range sub-look of the observed scene, here named a chromatic view, centered on its own carrier frequency. Performing an inverse Fourier transform of each sub-bands to go back in the image space, leads to a stack of lower resolution SAR images.

Split-Band SAR interferometry (SBInSAR) is based on such spectral decomposition performed on both images of an InSAR pair, yielding a stack of sub-band interferograms, each one having its own central carier frequency or wavelength and corresponding fringe rate. Scatterers keeping a coherent behavior in each sub-band interferogram show a phase that varies linearly with the carrier frequency, the slope being proportional to the absolute optical path difference. This potentially solves the problems of phase unwrapping on a pixel-per-pixel basis provided the coherence level is high enough to measure this linear behavior with high accuracy.

A first approach, which is highly time consuming, consist in splitting each image of a given interferometric pair leading to N pairs to be processed. This implies performing as much registrations. A second and more efficient approach consist in splitting each image after coregistration. But in that case, the optical phase path corresponding to the applied coregistration must be taken into account and re-added to the phase measurement derived by split band interferometry. In other words, when performing a SBInSAR processing on already coregistered images, we obtain a correction to this coregistration phase to get the absolute phase measurement. This correction phase can also be viewed as a coregistration improvement.

In this paper, we present a SBInSAR processor and its application on TanDEM-X data over the Nyiragongo volcano. The processor uses already coregistered interferometric pairs. From the stack of interferograms, linear regression is conducted at each point to derive the phase slope, which is proportional to the absolute phase after addition of the coregistration phase. The processor also computes the corresponding local height and several figures of merit, including the RMS error on the regression, the correlation coefficient R2, the RMS (one sigma) error on the intercepts and the spectral coherence. The RMS on the intercept is of prime importance because it gives the effective accuracy of the derived absolute phase. It has to be better than one phase: if two distant points have an RMS error on the intercepts better than one fringe, then absolute phase difference can be computed and the two points can be connected even if separated by non-coherent regions. In practice, SBInSAR allows getting the fringe number while Classical interferometry gives the fractional part of the phase.

We developed and applied the SBInSAR technique in the frame of the Vi-X project. Vi-X derives from « Study and monitoring of Virunga volcanoes using Tandem-X ». It is in line with continuous efforts of the Royal Museum for Central Africa (RMCA), the European Centre for Geodynamics & Seismology (ECGS) and the National Museum of Natural History of Luxembourg (NMNH) to monitor and understand the geodynamical processes of the Virunga Volcanic Province (VVP, east of the Democratic Republic of Congo (RDC), in the frame of a global risk management/evaluation of the area. The project aimed at assessing the use of TanDEM-X SAR data to help in studying an active volcano zone.

We applied SBInSAR to the peculiar case of the Nyiragongo volcano crater, located near Goma in the VVP.

One of the platforms inside the crater (P3), the deepest, is varying of several tens of meters a year due to the presence of a very active lava lake. The flanks of the crater are so steep that classical phase unwrapping always fails connecting P3 to the surrounding, preventing a correct height measurement of the platform to be made. We thus applied SBInSAR processing to perform a lava lake leveling or a lava deposit monitoring, having in mind the development of a tool for monitoring the volcano activity. Using three TanDEM-X data sets, we show that SBInSAR processing allows getting a first height estimation of the lava lake level. However, the spectral diversity resulting from the bandwidth (100MHz) of TanDEM-X does not allow getting a phase accuracy better than about 3 fringes in the most favorable cases. However, as can be seen on the figure below, the derived height map shows clearly the height difference between the inner part of the crater and the volcano flanks, the inner part being impossible to connect to the flanks using classical unwrapping methods.

This work was carried out under Belgian Science Policy Contract NR SR/00/150

The ESA campaign AlpTomoSAR is aimed at investigating the capabilities of L-Band SAR tomography concerning the study of the internal structure of glaciers. The campaign took place from the end of February to the beginning of March 2014 at the Mittelbergferner glacier, on the Austrian Alps.

SAR data were acquired using a fully-polarimetric airborne L-Band SAR system, developed and operated by Metasensing. The data were acquired by flying repeatedly along an oval-like racetrack, resulting in multiple passes from two opposite viewpoints.

The campaign also includes the collection of in-situ measurements of snow and glacier ice properties, carried out by ENVEO at the Mittelbergferner glacier in the same period as the airborne surveys.

The SAR sensor has been flown on a CASA C-212 aircraft operated by BLOM CGR. Aircraft altitude was about 4500 m, that is about 1300 m over the illuminated scene  - surface elevation at the Mittelbergferner glacier ranging from 2800 m to 3400 m. As expected, the aircraft not being equipped with automatic piloting systems, the flights were perturbed by turbulent phenomena in proximity of the peaks. The standard deviation of the flight trajectories was assessed in about 10 m, with peaks of over 50 m.

Such acquisition geometry determined a number of challenges concerning data processing.

In the first place, along track focusing needed to be carried out precisely accounting for navigational data, in order to cope with trajectory deviations, as well as surface topography.

A subsequent InSAR step was required to finely estimate airborne trajectory down to a fraction of the wavelength, in order to remove phase screens due to platform positions.

Tomographic focusing was complicated by the presence of large range migration differences among different passes, which required image coregistration to vary as a function of focusing depth.

Preliminary results clearly indicate the presence of in-depth scattering contributions from tenths of meters beneath the surface.

The aim of this paper is to illustrate and discuss all processing procedures that were implemented to achieve accurate 3D imaging at the illuminated site. We will also consider the repeatability of such measurements using spaceborne platforms.

3D SAR Tomography (TomoSAR) exploiting multibaseline data [1-4] is an important evolution of SAR Interferometry, to sense complex scenarios with multiple scatterers mapped in the SAR cell. Beyond layover solution in garbled urban/infrastructure areas [2,4], and recent ice investigations [5], emerging tomographic remote sensing applications regard forest scenarios [3,6,7], e.g. tree height and biomass estimation, sub-canopy topographic mapping, and search-and-rescue/surveillance. However, these scenarios are characterized by temporal decorrelating (canopy) scatterers, and an open problem exists of possible height blurring and accuracy losses for TomoSAR techniques - and Pol-InSAR - to be applied [6], in particular to biomass inventory and monitoring. Notably, the partial coherence can significantly impact not only spaceborne implementations of forest TomoSAR based on monostatic SAR systems, like the BIOMASS and NISAR projects, which can be affected by both long-term and short-term temporal decorrelation; also companion (or tandem) satellites systems, such as the investigated SAOCOM-CS , even adopting 3D correlative (track-pair) imaging processing [8], can still be sensibly influenced by short-term decorrelation depending on their formation-flying configuration design. It is thus important to extend characterizations of forest temporal decorrelation, also accounting for the height-varying nature of the decorrelation phenomena [6,9,10].

Regarding stratified analyses of forest long-term temporal decorrelation, large area characterizations can be obtained by advanced Differential Tomography (Diff-Tomo) 4D/4D+ processing [11,12] of classical airborne multipass data with sparse baseline-time sampling. In fact Diff-Tomo, furnishing “space (height)-time” signatures of multiple dynamic scatterers [11], while initially usefully applied to subsidence monitoring in garbled urban areas (see e.g. [13]), has actually the potential  to extract dynamic information in layered media, including separating in the height dimension different temporal coherence levels that are mixed (undiscriminated) in the classical (overall) coherence analysis [6,9,14].

This separation is based on the concept of space-time signatures of long-term temporal decorrelation [6], that has been confirmed with non-parametric Diff-Tomo analysis of BioSAR-1 ESA campaign P-band forest data with 2 months time span, acquired by the DLR’s E-SAR airborne platform, over a boreal region in Sweden [9]. A large-scale P-band separation of height-varying temporal coherences [14], over 500 hectares of the boreal area airborne data, has been obtained with a semi-parametric [9] (AR model) 4D+ Diff-Tomo processor, for overall coherence down to about 0.4. Diff-Tomo large-scale separation of temporal decorrelation mechanisms of canopy and ground has been also extended to multiple polarizations (HV, HH, VV). Corresponding results are reported, average HV pol. canopy and ground coherence time estimates  having been quantified in about 6 and 13 months, respectively, and average separated ground coherence time having resulted to be about three-fold rising in HH pol. with respect to HV pol., the phenomenological interpretation being consistent with the typical scattering mechanisms. The coverage of such 4D+ Diff-Tomo airborne analyses goes beyond what achievable with dedicated tower-radar based measurements of the TropiScat campaign, and the 4D+ Diff-Tomo method may be applied to  incoming  AfriSAR data to extend  AfriScat analyses.

Concerning forest short-term temporal decorrelation [6,10,12,14], in this work first samples are presented of an innovative short-term coherence profiling along the height dimension, in particular aiming to characterize the short-term coherence time magnitudes and their expected variability along the tree structures [6,10]. Extending the setup in [14], a first small scale experiment has been carried out using a ground-based miniradar, currently at X-band, HH pol., capable of extremely quick repeated acquisitions that are exploited to produce a “quick Diff-Tomo” characterization of short-term decorrelation processes. The test scene is located in Pisa, Italy and comprehends a few poplar and elm trees, the wind condition being light breeze. After data processing, unique space-time signatures of short-term temporal decorrelation are produced; from these, short-term coherence times resulted to be measurable, and their values appear reasonable. Also, a first indication of height-varying behaviour of the short-term coherence time parameter of forest decorrelation processes has been obtained through direct radar measurements.

These long- and short-term phenomenological investigation methodologies can be employed to better characterize perturbation effects of forest temporal decorrelation on Pol-InSAR [10] and TomoSAR [1,3,6] methods, and to tune decorrelation-robust TomoSAR algorithms [12,14] or satellite system designs, respectively. In particular, they may be useful in the framework of developments of the spaceborne forest Pol-InSAR and TomoSAR remote sensing missions BIOMASS and SAOCOM-CS, and related supporting measurements campaigns may be oriented  accordingly to these findings.

[1] A. Reigber, A. Moreira, “First Demonstration of Airborne SAR Tomography using Multibaseline L-band Data,” IEEE TGARS, 38(5), pp.2142-2152, 2000.

[2] G. Fornaro, F. Serafino, F. Soldovieri, “Three Dimensional Focusing With Multipass SAR Data,” IEEE TGARS, 41(3), pp. 507-517, 2003.

[3] M. Nannini, R. Scheiber, R. Horn, “Imaging of Targets Beneath Foliage with SAR Tomography,” EUSAR’08.

[4] F. Lombardini, F. Cai, D. Pasculli, “Spaceborne 3-D SAR Tomography for Analyzing Garbled Urban Scenarios: Single-look Superresolution Advances and Experiments," IEEE JSTARS, 6(2), pp.960-968, 2013.

[5] L. Ferro-Famil, C. Leconte, F. Boutet, X. Phan, M. Gay, Y. Durand, “PoSAR: A VHR Tomographic GB-SAR System Application to Snow Cover 3-D Imaging at X and Ku Bands,” EuRAD’12.

[6] F. Lombardini, F. Cai, “3D Tomographic and Differential Tomographic Response to Partially Coherent Scenes,” IGARSS’08.

[7] M. Pardini, K. Papathanassiou, “Robust Estimation of the Vertical Structure of Forest with Coherence Tomography,” ESA PolInSAR ’11 Workshop.

[8] F. Lombardini, M. Pardini, "Sector Matched Interpolation for 3D SAR Correlative Interferometric Imaging," Tech. Report, University of Pisa, Edizioni ETS, Pisa, July 2008. 

[9] F. Lombardini, F. Cai, “Evolutions of Diff-Tomo for Sensing Subcanopy Deformations and Height-varying Temporal Coherence,” ESA Fringe’11 Workshop.

[10] M. Lavalle, S. Hensley, “Demonstration of Repeat-Pass POLINSAR Using UAVSAR: the RMOG Model,” IGARSS’12.

[11] F. Lombardini, “Differential Tomography: A New Framework for SAR Interferometry”, IEEE TGARS, 43(1), pp.37-44, 2005.

[12] F. Lombardini, F. Cai, "Temporal Decorrelation-Robust SAR Tomography," IEEE TGARS, 52(9), pp.5412-5421, 2014.

[13] F. Lombardini, M. Pardini, “Superresolution Differential Tomography: Experiments on Identification of Multiple Scatterers in Spaceborne SAR Data,” IEEE TGARS, 50(4), pp.1117-1129, 2012. 

[14] F. Lombardini, F. Viviani, F. Cai, F. Dini, “Forest Temporal Decorrelation: 3D Analyses and Processing in the Diff-Tomo Framework,” IGARSS’13.

Despite over two decades of efforts towards mitigation, atmospheric contamination due to spatio-temporal changes in tropospheric temperature, pressure and humidity remains the largest source of error for the measurement of ground motions with InSAR. This is particularly problematic when considering automated processing and analysis of large SAR datasets from Sentinel-1.

Whilst many different methods have been proposed to mitigate atmospheric errors, those that rely on external data can rarely be applied due to data limitations, and those that use the InSAR phase observations themselves can erroneously remove ground deformation signals as well. Numerical weather models, on the other hand, present an opportunity to correct InSAR data systematically using independent data and a standardized approach that should, in principle, work in all conditions and locations, and at all times.

One of the most widely used numerical weather model datasets is ERA-Interim (ERA-I), produced by the European Centre for Medium-range Weather Forecasts. Corrections based on these data have seen rapid uptake by the InSAR community and are now widely used, but attempts to validate the correction method on small test sites show that it works well in some regions (e.g. Jolivet et al., 2014) but not in others (e.g. Garthwaite et al., 2013, Walters et al., 2013).

Here we systematically test wet delays derived from ERA-I for the first time by comparing them with global satellite spectrometer measurements of water vapour acquired during 2003-2010 from the MERIS instrument on board ENVISAT. Comparing the two datasets in 10×10 degree regions, we find significant geographical variation in the quality of ERA-I predicted wet delays, as measured by the RMS misfit between the two datasets.

We find that ERA-I retrieves wet delays better at mid-high latitudes (> 20-30 degrees), with mean RMS misfit values of ~1-2 cm line of sight delay (approaching the level of uncertainty on MERIS), than at low latitudes, where mean RMS misfit values are around 3-4 cm.

Globally, there are strong negative correlations between the quality of the ERA-I wet delays and both temperature and humidity in the tropospheric boundary layer, which enable us to predict the efficacy of the ERA-I correction method for any given SAR acquisition.

We suggest that for automated atmospheric correction of SAR data for Sentinel-1, ERA-I is currently a viable option, but uncertainties should also be estimated for these corrections on the basis of temperature and humidity. We propose a new method to estimate and incorporate these uncertainties in time-series analysis of deformation.

References:

Garthwaite, M. C., H. Wang, and T. J. Wright (2013), Broadscale interseismic deformation and fault slip rates in the central Tibetan Plateau observed using InSAR, J. Geophys. Res. Solid Earth, 118, 5071–5083, doi:10.1002/jgrb.50348.

Jolivet, R., P. S. Agram, N. Y. Lin, M. Simons, M.-P. Doin, G. Peltzer, and Z. Li (2014), Improving InSAR geodesy using Global Atmospheric Models, J. Geophys. Res. Solid Earth, 119, 2324–2341, doi:10.1002/2013JB010588.

Walters, R. J., J. R. Elliott, Z. Li, and B. Parsons (2013), Rapid strain accumulation on the Ashkabad fault (Turkmenistan) from atmosphere-corrected InSAR, J. Geophys. Res. Solid Earth, 118, 3674–3690, doi:10.1002/jgrb.50236.

The current synthetic aperture radar (SAR) theory makes a quite restrictive assumption – linearity – in the SAR imaging model, for the convenience of mathematical derivation. That is to say the imaged area is considered as an ensemble of individual point scatterers whose scattered fields and, hence, their responses in the SAR image superimpose linearly (Bamler & Hartl, 1998). This is the so called the first Born approximation. However, the reality is, for sure, more complicated than such approximation.

This work presents a step towards a better understanding of the scattering mechanism of different objects, and the occurrence of single scatterer, as well as multiple scatterers within a resolution cell. We back trace individual SAR scatterer to high resolution optical images where we can analyze the geometry, material, and other properties of the imaged object.

Please refer to the PDF file for the full abstract.

During the round table, seed questions proposed by the chairs will be discussed with the audience.