Hybrid Dem Generation and Evaluation from Spaceborne Radargrammetric and Optical Stereoscopic Dems

With the launch of the recent high-resolution Synthetic Aperture Radar (SAR) satellite missions, e.g., TerraSAR-X, RadarSAT-2, and CosmoSkymed, the technique of radargrammetry motivates interest again to potentially improve upon the level of detail and accuracy of Digital Elevation Models (DEM) from a global coverage perspective as an alternative to data derived from Interferometric SAR (IFSAR), Light Detection and Ranging (LiDAR), or the Shuttle Radar Topography Mission (SRTM). A stereoradargrammetric DEM extraction capability has been developed in-house with large area operational capacity. A number of DEMs, posted at 10 meters, have been generated from the TerraSAR-X strip-map mode stereo images, and evaluated relative to Intermap’s NEXTMap airborne DEM (vertical accuracy specification is 1 meter RMSE for a 5 meter grid). Vertical accuracies of 3~5m (RMSE) were observed in the low-relief, bare terrain areas, which outperformed, in these cases, a 90 m SRTM DEM that had been re-sampled to a 10 meter grid. However, the implementation of the stereo radargrammetry concept has some limitations. For instance, the resulting DEM is severely strongly affected by speckle in the images. Therefore the terrain needs to be overly smoothed and this results in a loss of spatial resolution and accuracy. Although the stereo DEM can be improved with the application of a speckle reducing filter and through development of more efficient image matching algorithms, another approach is through merging with a complementary source, such as a photogrammetrically-derived DEM. The medium-resolution optical stereo images from ALOS/PRISM, are candidates for this approach. However, photogrammetric DEMs also have their intrinsic limitations, for instance, cloud coverage, and in the areas of interest the potential absence of Ground Control Points (GCPs). Without vendor-provided Rational Polynomial Coefficients (RPC), the performance of a 5m posting PRISM stereoscopic DEM from the nadir and backward view combination dramatically depends on the number and quality of input GCPs. Therefore, the fusion of the radar and opticallyderived DEMs offers complementary advantages for the creation of a hybrid product. Our preliminary evaluation results show that such a hybrid DEM after fusion does demonstrate superior characteristics in terms of accuracy and level of spatial detail, compared to each individual input. In this work we will outline the process and demonstrate the results visually and quantitatively. * Corresponding author.