| Abstract | Synthetic aperture radar (SAR) imagery exhibits significant radiometric and geometric sensitivity to local terrain relief. These effects are most pronounced in imagery acquired at steep-viewing
geometries such as those of ERS-1 and the near-range modes of Radarsat. In order for such imagery to be accurately registered with map-based information and with images from other satellite sensors, these geometric distortions must be corrected in a
geocoding process with the aid of a digital elevation model (DEM). In addition, many monitoring applications will require geographic coverage in excess of nominal image swath widths. As a result, efficient production of seamless mosaics is essential.
The Radarsat Topographic Correction System (RTCS) has been designed and built to meet these requirements, providing high processing throughput in a user-friendly, desktop environment. This stand-alone, low repeat-cost system is a synergy of CCRS R&D
efforts in geocoding and mosaicking, Intermap's developmental and operational expertise in radar production software (through its STARMAP and STARPRO) systems, and elements of PCI's EASI/PACE functionality. The project was sponsored by the Canadian
government, with Intermap as prime contractor and PCI as its subcontractor. The design philosophy was such as to emphasize geometric accuracy and production performance. Considerable attention was paid to mathematical rigour in the terrain
geocoding portion in order to take advantage of more precise DEMs as they become available. Production throughput was addressed not only through algorithm optimization, but also through design of an interface which simplifies the user interaction,
while retaining operator choice for selection of modes, parameters and manual or automated operation. Simultaneous operations (for instance loading data, collecting ground control and geocoding) on different scenes may be performed simultaneously in
order to increase efficiency. The RTCS is able to process georeferenced Radarsat images from a broad range of mode, as well as standard ERS-1 scenes. Since elevation information is an integral data source, a flexible DEM data handling capability has
been developed. It includes database management facilities, as well as the assembly of scene-specific coverage from a diverse set of DEM sources. RTCS provides three methods to acquire ground control points (GCPs): (1) interactively, using a
modified form of PCI's GCPWorks for GCP database management, (2) automatically, using a simulated SAR image or (3) automatically using an image chip database. In interactive GCP collection, the operator selects matching points between the uncorrected
SAR image and the corresponding point on paper or digital map sheets, or on simulated SAR imagery, or a combination of the two. Automatic GCP collection operates by correlating image chips from the uncorrected image with image chips taken from either
a predefined image chip database, or by searching for a matching point in a corresponding simulated SAR image. The SAR image simulation models the radiometric variations due to topography in a SAR image, based on the local angle of incidence of
the microwave radiation. In this manner, edges and points of topographic features produced by regions of shadow, layover and foreshortening may be used as GCPs. This technique is well suited for automated implementation in regions of moderate to
rugged terrain and where the pixel resolution of the image is matched with the spatial resolution of the DEM. In the absence of pronounced topographic features (for example, in the prairie regions of Canada), conventional control point marking based
on the identification of cultural features (e.g., bridges, roads, railways) may be invoked. This technique also has the advantage that the simulated imagery may be used to correct for the radiometric influence of the terrain in the SAR image. This is
possible because the simulated SAR image is derived from the DEM only, using a universal backscatter model in which the radiometric response is modulated by theamount of range foreshortening. The techniques used to create large area mosaics build
on the STARPRO system developed by Intermap. Image seams may be defined either automatically or manually. For manual operation, tools are provided to define and edit the seams, to specify the image laydown order and seam topology, and to specify
image enhancements and zooming. Image pairs are interleaved on the display at a user-specified zoom level to show the appropriate level of detail. The automated seaming algorithm is fast and robust. Where terrain conditions permit, it will be the
preferred option. However it can be supplemented or replaced by a fast operator-driven seam operation treated either as an editing stage or a stand-alone activity. Outputs from the system include precision-geocoded SAR imagery (also referred to as
ortho-rectified or terrain-geocoded imagery), as single or mosaicked scenes, in a user-specified map projection and pixel spacing. The user may also produce various co-registered value-added image products, including shadow/layover masks; incidence
angle masks; simulated, geo-coded SAR images; and elevation colour-coded images. The output products are CEOS-formatted in 8- or 16-bit operator-selectable mode. Examples of these products of the RTCS will be presented at the Conference. The
software is currently hosted on a SPARCStation 20 platform with 96 Mbyte memory, 8.4 Gbyte disc drive, Exabyte tape drive and CD drive. It will soon be available on other UNIX platforms and eventually a de-rated version will be available on PC
platforms. It is intended to offer commercial geocoding/mosaicking services with this system (Intermap) as well as to licence the software (Intermap and PCI) to external users. |