Error tracking of radargrammetric DEM from RADARSAT images

Abstract

In the 1960s, stereoscopic methods were first applied to radar images to derive ground elevation. Unfortunately, research uncovered contradictions between error propagation theory and practical results. These contradictions combined with the lack of stereo radar pairs led to the decline of radargrammetry. The launch in 1995 of RADARSAT, with its various operating modes and specific geometric characteristics, has turned the tide. The error propagation of the radargrammetric DEM generated from different RADARSAT stereo configurations is then quantitatively evaluated along the full processing chain (stereo model set-up with ground control points (GCPs), image matching and three-dimensional (3-D) intersection). Two matching algorithms are used: 1) automatic and 2) computer-assisted visual matching. The GCP collection method using stereoscopic plotting is a requisite to achieve the best results for the stereo model and DEM. The automatic matching gives slightly better results than the computer-assisted visual matching, except when the radiometric disparities in the stereo images are too large. Since the geometric advantage (not involved in the automatic matching) can compensate for the radiometric disadvantage, visual matching, which combines both aspects, is a better method in these conditions. Consequently, these two algorithms can be used in a two-step method to generate the best DEM whatever the stereo configuration. Since the relief is an important parameter in the final accuracy, geometric versus radiometric disparity tradeoffs and general guidelines are suggested for selecting RADARSAT stereo pairs for DEM generation as a function of terrain relief.