ihich are 
with an a 
Jucted at 
tigations 
occurred 
s into the 
; 3 times 
entioned 
4] were 
the final 
neglected 
the orbit 
the MIR 
ting orbit 
vas set to 
nsformed 
se to the 
obtained, 
] Kornus, 
/ery pixel 
check for 
'esults 
5 pixel) 
ion using 
sed. Here, 
an image 
LTS into 
yrgyzstan 
RC/INFO 
using the 
1999). 
\ (Linder, 
pographic 
letail and 
ster DEM 
minimum 
5x5, 7x7 
alculating 
DEM and 
Roessner, Sigrid 
in analyzing their frequency distributions and statistical 
achieved for a DEM interpolation of 50m raster size with 
the whole study area (Fig. 1). For the Maili Suu test area 
from the dataset with 1 pixel minimum point distance (1 
parameters. Best representation of original mass points was 
subsequent 3x3 filtering. This parameterization was applied to 
a DEM with 25 m raster size and 3x3 filtering was generated 
Hinimurr 7 m). Comparisons of both DEM were based on interpolation 
accuracy and slope frequency distributions. The 25m-DEM shows a slightly better representation of original mass point 
elevations. Slopes derived from the 50-m DEM are lower in comparison to the 25m-DEM indicating a greater 
morphological detail contained in the latter one. 
3.6 Derivation of orthoimages using DEM and estimated interior and exterior orientations 
After the estimation of the interior and exterior orientations by bundle adjustment and the derivation of the DEM, the 
imaging geometry is fully known (to the indicated accuracy). Thus, orthoimages of all 4 MOMS channels can be 
generated. This has been done for the Maili Suu test area based on the 25-m DEM. The result is a multispectral 
orthoimage fitting to the DEM. In case of MOMS stereo mode D this is a big improvement, since nadir looking 
channels 1 and 4 and stereo channels 6/7 are now available as geometrically identical channels. This way channel 
combinations 1-6-4 / 1-7-4 cover almost the whole range of the VIS/NIR part of the spectrum between 440 and 810nm 
and allow true RGB visualizations leading to a much better spectral discrimination between different objects. 
4 ASSESSMENT OF ABSOLUTE HEIGHT ACCURACY BASED ON GPS CHECKPOINTS 
Absolute height accuracy of the 3x3 filtered and 50m raster size DEM was determined within a 150 by 60km part of the 
DEM where GPS observations of independent check points (CP) were obtained during the 1999 and a previous 1998 
GPS campaign (Fig. 1). Spatial consistency between both campaigns was checked based on 3 locations with double 
observations. The average difference between the independently determined coordinates amounts to 0.06m in X, 0.1m 
in Y and 0.04 m in Z. From all possible CP only those were selected which were situated in areas with mass point 
coverage. In the result, for a total of 36 CP elevation differences (d) between GPS heights and heights interpolated from 
the raster DEM at the same X, Y positions were calculated. For an empirical estimate of height accuracy mean (dmean) 
and standard deviation (dggey) Were determined based on these elevation differences (Tab. 1). For all 36 points, height 
accuracy amounts to 1.1m +/- 17.3m. This result is in the expected range of accuracy taking into account problems in 
image-based identification of GCP (2.2) and the highly dissected relief of the study area leading to a stronger influence 
of positional accuracy (X, Y) on height accuracy (Z). Tab. 1 shows a statistical analysis of the CP in dependence on 
their relief position (Fig. 1). Best height accuracy is obtained for points situated in plains approaching accuracy of 10m 
which had been achieved in previous studies (Kornus, 1999). Largest standard deviations occur for points along hill 
slopes and valleys. The latter ones show a tendency of being filled in contrast to hill tops being lowered. The limited 
number of CP reduces the statistical significance of the results. However, the obtained height accuracy satisfies the 
needs of landslide analysis since it is of subordinate importance in comparison to morphological correctness of the 
DEM which will be discussed in the following section (5). 
  
  
  
  
  
  
  
control points number of points dmin in Mm dmax In m dmean IN M dstdey IN M 
all 36 -38.7 29.3 1.1 17.3 
lain 15 -11.6 19.2 3.3 77 
valley 8 -27.2 18.5 -2.2 15.5 
hill slope 9 -38.7 29.3 7.0 27.3 
hill top 4 11.5 22.5 18.3 4.7 
  
  
  
  
  
  
  
Tab. 1 Height accuracy in GPS points (CP) for 50m-DEM (3x3 filtered) in dependence on relief 
5 DEM-BASED CHARACTERIZATION OF KASHGARTA LANDSLIDE 
The Kashgarta landslide in the Maili Suu test area (Fig. 2) is the biggest mass movement in the Upper Maili Suu river 
basin. The landslide developed in spring 1994 and has displaced a total mass of about 10 million tons until the end of 
June 1998. The Maili Suu area characterized by high landslide intensity is situated in the transitional zone between the 
Variscan consolidated basement block in the north (Arslanbob block as part of the Southern Tien Shan) and the Fergana 
Basin in the south. In the area of the Kashgarta landslide the northern rim of the Fergana Basin is formed by an anticline 
consisting of weakly consolidated sediments (Upper Cretaceous up to Paleogene) which is folded along an E-W axis. In 
  
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 1263