Automated Geo-Referencing of JERS-1 SAR Mosaics Using SRTM DEM Data

Mosaics produced from SAR images serve as base maps for many vast, remote areas such as the Amazon basin and Siberia. However, these mosaics are usually not precisely geo-referenced due to the distortion caused by terrain relief and SAR geometry, therefore their scientific value is limited. Since high-resolution digital elevation models (DEMs) were not widely available, terrain distortion was commonly not removed during the mosaic generation process. Recently, the SRTM (Shuttle Radar Topography Mission) mission has provided geolocation-accurate DEM data at high resolution (roughly 100 m) for nearly global coverage for the first time, and makes it possible to produce precisely located SAR mosaics. Rather than beginning with each individual SAR scene and rectifying it, this paper introduces an automated method for directly removing terrain distortion from SAR mosaics and precisely geo-referencing them. The proposed procedures include SAR image simulation from SRTM DEM, image matching between SAR mosaics and the simulated SAR image, automated ground control point (GCP) selection and screening, TIN-based (Triangular Irregular Network) registration using the GCPs for localized adjustment, and producing rectified SAR mosaics by transforming from the simulated SAR coordinate system to the SRTM system. The proposed method was tested in the Amazon basin using both the low water and high water GRFM (Global Rain Forest Mapping) mosaics. These mosaics, produced by NASA JPL from hundreds of JERS-1 SAR images, cover an area of 8 by 18 degrees in latitude and longitude with a resolution of ~100 m. The offset between these original mosaics and the DEM-simulated SAR image is up to 22 pixels in mountainous areas. Implemented using IDLr on a 3.06G Hz Windows platform, it cost about 3 hours to rectify both mosaics. 1-pixel geo-referencing accuracy was achieved using about 100,000 automatically selected GCPs in both cases. The rectified mosaics serve as base maps in our Amazon basin GIS database, and work great with other layers in the database.