Geometric evaluation, automated DEM and orthoimage generation from along-track stereo ASTER images

A cloud-free ASTER scene combination covering 61.5km /spl times/ 63km Zonguldak testfield in the north-west Turkey has been analysed. It comprises the nadir and backward views with a base-to-height ratio of 0.6. The pixel size on the ground is 15m. The bundle orientation was executed with the related module of PCI Geomatica V9.1.4 software package and resulted the 3D geo-positioning to an accuracy of about 14m in planimetry and 13m in height. This level of accuracy can be provided using the number of GCPs up to 14 which are distributed over the scene uniformly. Based on the scene orientation, a DEM of the area has been determined by an automatic image matching and PCI system yielded a DEM with 30m cell size. For the validation of extracted DEM, different groups of GCPs selected over the testfield were utilized. In this analysis, GCPs were located in the raster DEM in according to their planimetric coordinates, then the heights are estimated by the bilinear interpolation of the neighboured grid cells. This was done by the program DEMINT and mean square differences was obtained in the range of 12 to 14m. Moreover, matched DEM was checked against reference DEM based on digitised contour lines from the 1:25000 scale topographic maps using program DEMANAL. The discrepancies between the two DEMs were determined as reference DEM minus matched DEM. Then positive biases resulted which show that matched DEM occurred under the reference DEM. These biases appeared also in the superimposition of contours from two DEMs. Image of DZ discrepancies is displayed as a function of grey values as well and highest residuals occurred mainly at the ridges. In the meantime, RMSE-Z from the comparison of both DEMs obtained very close in the range of /spl plusmn/21-22m. Finally, orthoimage was generated using matched DEM and nadir image component of ASTER stereopair without problem. Planimetric accuracy check of this product was realized using the GCPs and shows no systematic error pattern overall.