DEM generation and accuracy assessment from stereo ASTER imagery

Abstract

In this study, Digital Elevation Models (DEMs) were generated from ASTER stereo images for a study area enclosing the city of Ankara and covering an area of 3600 km/sup 2/. Majority of the area was covered by large scale orthophotos, which were used as the main source to collect the Ground Control Points (GCPs). For those areas that were not covered by the orthophotos the GCPs were collected through a differential GPS. The bundle adjustment process and the generation of DEMs were carried out using PCI Geomatica image analysis software. In order to find the best GCP combination, DEMs were generated using different number of GCPs (8, 16, 24 and 32). DEM accuracies were assessed based on different parameters including the check points, slopes and land cover types. The accuracies of the resulting DEMs revealed that 16 GCPs were good compromise to produce the most accurate DEM. It was found that the post processing and blunder removal were crucial stages that affect the overall DEM accuracy up to 38%. There was a strong linear relationship between the accuracy of DEMs and the terrain slopes. For urban and forested areas, the accuracy in elevation was found to be 8.03 m and 12.69 m, respectively. The overall accuracy computed for 2,171,664 points produced 10.92 m elevation accuracy.