TGRS-2010-00737, EUSAR 2010 Special Issue 1 Abstract—We have studied the geographic position of several high- resolution spotlight TSX images by investigating the location of deployed radar corner reflectors. Results show that the geolocation accuracy is better than the resolution cell in both azimuth and range directions. The same corner reflectors as well as distinct points on buildings are used to estimate the absolute height from stereo viewing spotlight TSX images to within a few decimeters accuracy. Index Terms—satellite, synthetic aperture radar, geographic position accuracy, stereo height, corner reflector I. INTRODUCTION he German TerraSAR-X (TSX) satellite platform was launched in 2007. The major sensor onboard is an X-band synthetic aperture radar (SAR) system capable of acquiring data using several imaging modes [1]. The best spatial resolution of 1 meter is obtained with the spotlight mode. The TSX spotlight mode is an advanced SAR imaging mode where the raw data have been focused to single-look complex images in zero-Doppler coordinates [2]. The TSX satellite has a relatively short revisit time of only 11 days. This, together with its radar antenna steering capability, makes it possible to acquire high-resolution SAR data over the same geographic area on the Earth surface within a few days interval. The geolocation accuracy of pixels within a satellite SAR scene is strongly linked to the attitude and orbit accuracy of the satellite, as well as the SAR processing itself. The initial calibration campaign of TerraSAR-X gave excellent results in this respect, with orbit accuracies approaching 5 cm [3]. Therefore, it is interesting to investigate further if it is possible to achieve SAR image geolocation accuracies well below the resolution cell: i.e. at the centimeter scale. An overview of TerraSAR-X processing and accompanying products can be found in [4]. There are several levels of satellite orbit accuracies. The best accuracy is achieved using the science orbits. In [5] the influence of atmospheric path delay on the geolocation accuracy was thoroughly Manuscript received March 31, 2011. (This work was supported by Norwegian Defence Research Establishment). Knut Eldhuset is with Norwegian Defence Research Establishment, Land and Airsystems Division, PO Box 25, 2027 Kjeller, Norway (phone: +47-63- 807478; fax: +47-63-807212; e-mail: knut.eldhuset@ffi.no ). Dan Johan Weydahl is with Norwegian Defence Research Establishment, Land and Airsystems Division, PO Box 25, 2027 Kjeller, Norway (e-mail: dan-johan.weydahl@ffi.no ). investigated. Recently, image pixel location accuracies in the centimeter scale in range have been reported after fine-tuning some of the correction algorithms [6, 7]. Despite these facts, there will still be challenges regarding geolocation accuracy of orthorectified SAR images: the geolocation accuracy will depend strongly on the quality of the digital elevation model used in the orthorectification process [8]. In general, if there are well-defined structures that are clearly visible in a pair of TSX images acquired with different radar beam incidence angles, we may use the stereo SAR technique to estimate elevation heights at these points within the SAR image [9]. Investigations have been done with respect to generating digital elevation models (DEMs) from TSX data [10]. Results gave an overall DEM accuracy (LE 90) of around 9 meters when using 10 meters grid spacing. Another study compared the geolocation of certain well defined ground points with the TSX image data using both same-side and opposite side stereo spotlight acquisitions [11]. Here, they achieved a geolocation accuracy of retrieved 3D points in a range between 1 and 1.5 meters. Deployed corner reflectors can be seen from different incidence angle positions. We will therefore use the measured positions of such corner reflectors to validate the geographic position accuracy of the TSX spotlight image products. At the same time, we will also use our own software to estimate the absolute elevation height at these corner reflector points. First, we describe our TSX data set in Section II. Then we explain how we use the deployed corner reflectors and how they are positioned using a differential GPS system in Section III.A. In Section III.B we give a short description of our pixel location algorithm and stereo height estimator, and how we use these algorithms together with the TSX data set and corner reflectors to estimate geographical positions in both plane and height from image pixel locations (azimuth and range) within the SAR scene. In Section IV.A we use the ENVI/SARscape software to verify the position accuracy of corner reflectors. In Section IV.B we minimize the difference of GPS positions and positions estimated with our pixel location algorithm to find azimuth and range corrections. The range corrections will validate the ionospheric and atmospheric path delay given in the accompanying TSX product header. In Section IV.C we validate the TerraSAR-X product header geographic positions. In Section IV.D we improve the pixel location accuracy by finding an azimuth correction from each scene individually. We also minimize the difference of GPS positions and Geolocation and Stereo Height Estimation Using TerraSAR-X Spotlight Image Data K. Eldhuset, and D. J. Weydahl T Dette er en postprint-versjon / This is a postprint version. DOI til publisert versjon / DOI to published version: 10.1109/TGRS.2011.2160951