A Comparative Study on the Geometric Accuracy of TerraSAR-X and Sentinel-1

Kurzfassung

Our previous studies revealed the unprecedented geolocation accuracy of TerraSAR-X based on high resolution SAR images acquired in Spotlight modes HS, HS300 or ST300. In contrast, the obtained geolocation performance of imaging modes like ScanSAR with significantly lower resolution but significantly higher swath width has not been discussed at an equivalent level of detail so far. The measurement results presented here shall close this gap. Based on TerraSAR-X ScanSAR mode and Sentinel-1 Interferometric Wideswath (IW) mode, that are comparable w.r.t. their resolution and swath width, we evaluated the attainable geolocation accuracy. The applied standard method is to compute the absolute location error (ALE) between the measured positions of corner reflectors (CRs) in SAR images and their expected positions derived from the precise knowledge of the geodetic coordinates of their geometric phase center and the solution of the zero-Doppler equations. In the process, also correction values for signal propagation delays and for geodynamic effects have to be thoroughly considered. The permanent Australian CR array located at the Surat Basin (East Australia), installed and operated by Geoscience Australia (GA), is a test site very well suited for our measurements. 40 CRs with an inner edge dimension of 1.5 to 2.5 meters are distributed in an area of about 100x100 square kilometers. The precise CR coordinates have been published by GA. Thus, multiple measurements at different positions within an image become possible and may disclose potential geometric distortions of the SAR image as they might be introduced e.g. by processor approximations in SAR focusing. For the 12 TerraSAR-X ScanSAR datatakes of this testsite, the standard deviations of our ALE measurements amount to about 8 centimeters in azimuth and about 4 centimeters in range. For comparison, in former measurement series for Spotlight mode applied at other testsites using 1.5 meters CRs, we observed a standard deviation of the ALE of about 1.5 - 2 centimeters in azimuth and about 1 centimeter in range. However, the decrease in geolocation accuracy from Spotlight to ScanSAR is expected. It is largely explained by the lower signal to clutter ratio of the CRs caused by the coarser resolution in ScanSAR mode. The observed standard deviation of the ALE fits well to the expectation value that we computed on the base of theoretical considerations originally published by Stein in 1981. Comparing the measured ALE of different CRs at different positions in the image, we observed almost no correlation between the ALE and the azimuth or range location in the image. At most, there might be slight traces of a variation of the azimuth bias within a ScanSAR beam from early to late azimuth. This trend still needs to be confirmed and if true, its cause to be investigated. In the case of Sentinal-1 IW mode, 18 datatakes were investigated. Here, we found significant trends of the azimuth and range ALE on the CR position within the image. In a detailed analysis, we identified several processor approximations causing these artifacts. When computationally correcting the image coordinates for the identified processor approximations in a post-processing step, the geolocation accuracy of Sentinel-1 IW becomes similar to TerraSAR-X ScanSAR. In the future, these corrections will be distributed in an extended Sentinel-1 annotation product provided free of charge by ESA ESRIN. Plots of the detailed measurement results and analyses shall be presented in a poster at the Science Team Meeting. The research work was founded by ESA ESRIN within the framework of the FRM4SAR project (ESA Contract No. 4000119113/16/I-EF).