Comparison of DEM Generation and Combination Methods Using High Resolution Optical Stereo Imagery and Interferometric SAR Data

Abstract

Digital elevation models (DEM) from satellite data are generated mainly from two types of datasets using completely different methods: photogrammetry for optical stereo images (e.g. SPOT5, IKONOS) and interferometry for Synthetic Aperture Radar data (InSAR, e.g. ERS-Tandem, SRTM). Both generation methods show advantages and disadvantages but have similar accuracy values in comparison to a reference DEM. The paper aims at showing the potential for combined usage of several DSM (derived with different sensors and methods) to improve the overall accuracy. Some results are given for DEM fusion utilizing height error maps for each DSM and for DEM integration, where single point information from another DSM is inserted during the triangulation process. The quality of the DSM derived from one source and of the combined DSM depends on the steepness of the terrain and on the land cover type. For flat terrain or moderate hilly landscapes, a height accuracy in the order of 5 meters or better can be achieved for the mentioned sensors. Two test areas are chosen, where many different data sets are available and much knowledge exists from previous studies. The first test area is a region in the south-eastern part of Bavaria comprising a mostly hilly, post-glacial landscape including lakes and also mountains of the German Alps. The second test area is located in Catalonia, Spain, and includes the city of Barcelona as well as a mostly hilly terrain with some steep slopes and additionally the Mediterranean coast. The received DSM are compared qualitatively and quantitatively to the reference DEM with superior quality by looking at profiles and sub-area statistics. The results show that an improvement of the fused DSM and the integrated DSM can be quantitatively measured. Although the overall statistics for a larger region does show only a slight improvement, local errors can be reduced significantly so that the overall accuracy of the combined DSM is higher.