S1 - TOPS InSAR

SAR images coregistration is of fundamental importance for the generation of SAR interferograms. The high azimuth coregistration requirements imposed by the TOPS acquisition mode converts the coregistration of stacked time series images in a demanding task due to temporal decorrelation effects. A joint coregistration approach appears to be the optimal solution for coregistering long stacks. The problem of a joint coregistration has been already addressed in the literature for the case of coherently correlating speckle signals. A similar procedure can be applied when employing the established Enhanced Spectral Diversity (ESD) technique for azimuth coregistration of TOPS images.

The methodology for the retrieval of the (rigid) azimuth shift of a slave image with respect to a master image, due to orbital or timing error, will be in first place introduced. A Joint-coregistration approach employing the retrieved shifts of all possible pair combinations will be detailed. A Weighted Least Squares (WLS) estimation is applied to retrieve the azimuth shifts of each image with respect to a selected master image.

A stack of Sentinel-1 images acquired in IW mode over Mexico City covering approximately 2 years of acquisitions and eventually other sites is used to evaluate this procedure, comparing its performance to the standard single-master approach. Moreover a validation of the method will be provided highlighting the phase biases that might appear when only a geometric approach or single-master approach are used.

The contribution focuses on the current status of the ESA study entitled “INSARAP Sentinel-1 Constellation Study”, and investigates the interferometric performance of the S1A/S1B units. General aspects like the interferometric compatibility in terms of common range and Doppler bandwidth and the burst synchronization will be addressed. Besides the first interferometric results with both units, time series results over the pilot sites combining both satellites will be also shown. In addition, advanced investigations concerning the exploitation of the overlap areas, the estimation and correction of ionospheric artefacts, or processing approaches for fast moving scenarios (e.g., glaciers) will be presented.

Please see the attached PDF for further details and results.

This contribution analyses the geolocation accuracy of Sentinel-1 in the IW mode. Considering the geolocation accuracy has been investigated in the frame of the corresponding commissioning phases for the S1A and S1B units, this contribution is focused on certain aspects not addressed before. On the first place, the 2D (range / azimuth) geolocation accuracy will be addressed by taking into account certain processing aspects that are currently not being corrected by the operational Sentinel-1 processor. On the second place, the geolocation accuracy is also evaluated in elevation by exploiting a stack of images and the corresponding spread of the perpendicular baselines in order to perform imaging in the direction perpendicular to the line of sight, i.e., tomographic imaging. The following two paragraphs are dedicated to describe both investigations. In order to perform the analysis, an ascending stack of Sentienl-1A images acquired over DLR’s calibration test site close to Munich has been used, which is composed of 18 images acquired between November 25, 2015 and July 22, 2016.

2D geolocation accuracy: the S1 processed data does not properly include the correction of two geometric effects, namely, the bistatic correction and the intra-pulse motion. The former introduces a range-dependent azimuth shift of about ±1.8 m for the complete IW swath, being 0 m at the reference range at the middle of the swath. The later introduces an azimuth-dependent range shift, which reaches values of ±[35,48] cm at burst edges. Both effects bias the geolocation accuracy, hence limiting the potential of what otherwise could be a very accurate ranging instrument, as already demonstrated with TerraSAR-X. Another aspect related to the processing is the azimuth variance of the geometry, i.e., the fact that targets at different azimuth positions require different matched filters due to the space variance of the spaceborne geometry, coupled with the fact that the target is observed under a given Doppler [1]. In this contribution, these three effects are first analyzed separately and then the final results after applying all corrections are presented. We shall see that such proper consideration of these geometric effects does improve the geolocation accuracy of Sentinel-1 in both range and azimuth direction. The contribution will describe in detail the different effects and how to overcome them. Figure 1 shows the final result with the measured error over several CRs before and after the corrections. The original standard deviations without the corrections are 0.27 m and 0.43 m in range and azimuth, respectively, while the standard deviations after the corrections are 0.10 m and 0.22 m, respectively, hence showing a relevant improvement in the geolocation accuracy.

3D geolocation accuracy: as mentioned, along with the accuracy in the range-azimuth geolocation within a SAR image, if a stack is available, the vertical position of the phase center of a corner reflector can be determined by means of SAR tomography via a backprojection approaches. For this analysis, also the Capon beamformer has been employed to enhance the resolution of the tomographic retrieval consequence of the reduced total perpendicular baseline spanned, as a direct consequence of the reduced orbital tube. For a good match between the data and the nominal geometry, a data calibration procedure has to be foreseen. In particular, once a reference point is chosen and its phase removed, it is necessary to filter out the atmospheric and the deformation contributions. Therefore, a PSI preprocessing step is usually required. In the context of this analysis, our reference point will be a corner reflector that will be called CRref. After calibration, the tomographic processing will allow one to estimate the offsets in height w.r.t. the height of this reference corner (that will have height equal to zero being the reference). In particular, to evaluate the vertical accuracy, the height of another corner reflector CRi (relatively close to CRref) will be estimated and compared with the ground measurements performed by the calibration team of the DLR-HR. The residual discrepancy will then be compared with the theoretical expected accuracy that is defined as the Cramér-Rao lower bound for high-SNRs [2]. We shall see that, independently on the focusing method (Fourier/Capon), the achieved vertical accuracy is within the expected values.

Please see attached pdf file.

In the last two years the European Space Agency (ESA) SAR sensors, Sentinel 1A and B (S1), have acquired a huge amount of data over the whole Earth surface. Clearly, different SAR applications will benefit by employing these large stacks of S1 data. In particular, applications like crop monitoring, ground surface deformations survey, land cover mapping, and all the others based on the spatial-temporal analysis of a large stack of data can take great advantage of the features of the sensor: short revisiting time, wide area processing, high slant-range resolution, improved radiometric accuracy, availability of dual pol. data, different values of incidence angle and, by now, more than two years of archived data.

However, due to the TOPS acquisition mode, some important aspects have to be considered. In particular, applications that exploit the coherent signal need a very high accuracy during the image co-registration. Typically, it is proven that images acquired by standard modes (or stripmap) have to be aligned with an accuracy of less than 0.1 pixel to have no impact on the phase quality. This parameter is stronger (less than few thousandths of pixel) for the background acquisition mode of S1, which is based on TOPS mode.

In literature, there is a set of different coregistration approaches proposed.  Depending on the processing chain, different steps are performed in different ways achieving different performances. The objective of this paper is to present the adaptations made on our processing chain in order to accurately coregistrate large stack of data and at the same time and share some problems and discussions found in the coregistration process of S1.

We are going to focus our attention on the Coherent Pixel Technique (CPT) of the UPC, and what are the main methodological modifications we have made in order to adapt it to S1. The whole process for correctly coregistrate one SLC, the slave, over another one, the master, can be resumed into 5 main steps: de-ramping of the slave image, geometric offset estimation and application, constant offset estimation through amplitude correlation, Enhanced Spectral Diversity (ESD) to correct a residual offset and re-ramping.

For most InSAR application, if all these steps are properly applied, the two images will be perfectly aligned. We present an overview of the different problems that can affect the correct coregistration process. In first instance, the orbits and time synchronization inaccuracies produce a miss-matching between consecutive bursts. We have detected in 15% of the SLCs processed an offset between consecutive burst up to 0.006 lines that produce a phase ramp of almost one azimuth fringe in a single burst. A same azimuth time reference has to be set in order to overcome this problem as we are working in a burst per burst basis. In this way the miss-matching offset will be corrected through the geometric matrices.

Second, the temporal decorrelation in vegetated areas, the presence of sea/lakes or areas affected by layover effect is another issue that can affect the cross-correlation step. In case of two SLC images separated in time it could be possible to incur in a false detection in the estimation of the constant offsets. This wrong offset would produce a differential azimuth shift between the different bursts that ESD cannot recover. As presented in the literature, ESD finds the residual azimuth offset exploiting the spectral gap between two consecutive bursts. This step works well if the previous steps have been successfully completed, this means that the residual azimuth offset is below 0.07 lines.

Therefore, in this paper we will present different strategies to overcome those problems like approaching the coregistration with different image subsets, masking incoherent areas, or the impact of playing with different burst and sub-swaths during the ESD for achieving a proper estimation of the fine offsets required.

In September 2016, the space segment of the Sentinel-1 (S1) mission was declared complete and operational. Due to the short duration of the commissioning phase, a rigorous assessment of the S1 constellation InSAR time series performance could not be performed. However, the initial analysis of basic InSAR parameters - based on short stacks - showed that “all lights are green”. Therefore, the work presented here will not address the interferometric compatibility of the two units, but  rather focus on on the implications of the 6-day revisit time and on the challenges and opportunities connected with large-scale interferometric processing.

The objective of the InSARap project is essentially to answer the following questions:

What can we do already - “out-of-the-box”?

What can we do after algorithmic consolidation and/or improvement?

What can we still not do?

Our approach was to answer them by studying the following topics:

InSAR geodetic validation: quantitative analysis of InSAR performance measures and other measures that can indirectly influence the interferometric phase. Farms of corner reflectors are exploited for this purpose.

Algorithms: it is evident that some areas of the processing workflow need further algorithmic development for the large-scale case. We will explore new algorithmic opportunities in light of the 6-day sampling provided by the constellation.

Geophysical interpretation: Better understanding of unexplored signal components, e.g., the along-track deformation in burst overlap zones and potentially correlated components with 6-day sampling.

Considering the broad nature of this contribution, different test sites and deformation phenomena will be used as examples, from corner reflector farms in Poland to ice sheet flow in Antarctica. Importantly, the inherent limitations of the S1 mission will also be discussed, focusing on the  temporal and spatial sampling. 

To conclude, in this contribution we will summarize our initial assessment of the S1 constellation performance, including proposed necessary updates to the processing workflows currently regarded as community consensus.