Path Processing and block adjustment with RADARSAT-1 SAR images

Abstract

The objectives of this research study was to determine the conditions of experimentation and application of path processing and block adjustment with synthetic aperture radar (SAR) images when few controls are available. The path and block processing enabled the simultaneous adjustment of all images together to reduce the control point requirement. The method is based on the three-dimensional physical model developed for multisensor images at the Canada Centre for Remote Sensing, Natural Resources Canada. These processes were applied to 15 RADARSAT-1 SAR fine mode images (five paths by three rows) acquired over the Rocky Mountains, Canada from different look angles (F1 and F4) with a weak intersection geometry (6/spl deg/ angle). The first results of the least squares block adjustment showed that the same errors were obtained with image paths or block (20 m for three-image paths and five-path block) as with a single image (18 m). In addition to ground control points (GCPs), elevation tie points (ETPs), with a known elevation value, were used in the overlaps because the 6/spl deg/ look-angle difference of overlapping paths was small. However when using only GCPs in the outer paths for block adjustments, the error results deteriorated from 25 m in both directions for the three-path block to 270 m in X direction for the five-path block. This deterioration was a combination of the image pointing and cartographic errors of GCPs (25-30 m) and the weak 6/spl deg/ intersection angle. Consequently, GCP distribution every two paths was the solution with this dataset, and better results (35 m) was then achieved using a reduced number of GCPs in the outer paths (25 or even 10 GCPs) and middle path (six GCPs) and 20 ETPs in each overlap. However, the combined image pointing and cartographic errors of GCPs (25-30 m) are included in these 35-m error results, and the internal accuracy of the block should thus be better (around one resolution). Finally, the same minimum requirement of GCPs, as a function of their accuracy and the overlap intersection geometry, can be applied for an image, a path or a block.