ion zone
e Pamir-
occur at
| Fergana
Ld
i 41 + 00
N
40° 30°
N
f ground
he 29" of
)9. In the
for high
eige et al,
ed lenses
ns within
ick stereo
strips are
it ratio of
iprising 2
Roessner, Sigrid
Three scenes (60, 61, 62) MOMS-2P data take T0906 include the northernmost part of the Fergana Basin (elevation
below 700m N.N.) and its rim between 700 and 2000m where the majority of landslides occur. Since cloudiness is
mainly limited to higher areas, a large part of the terrain affected by landslides is visible in the MOMS-2P scenes.
MOMS-2P data acquisition took place from an altitude of 372 km. Each scene covers an area of 100 by 100km. Stereo
imaging mode D combines two inclined panchromatic stereo channels (6, 7) and two nadir looking spectral channels in
the blue (1) and near infrared (4), whose spatial resolution is 17m.
2.2 GPS measurements
Precise 3-D coordinates of ground control points (GCP) and independent check points (CP) were obtained by geodetic
GPS measurements within an area of 150km length and 60km width (Fig. 1) using LEICA (SR9500) receivers. Network
planning aimed for a GPS accuracy of one meter in X, Y, Z and a mean distance between locations of 25 - 30km. GPS
observations took place simultaneously for a fixed reference station and mobile rover points. Because of the size of the
study area four reference stations had to be established (Fig. 1). Based on experiences from previous campaigns
(Roessner, 1997), observation time was 30 minutes for trigonometric points and 15 minutes for other field points.
Possible locations for GCP were preliminarily selected based on MOMS-2P stereo channel 7 and topographic maps. For
their identification in the field, black-and-white hardcopies (approx. scale of 1:25.000) were used which allowed single
pixel identification. During the GPS campaign 59 object points were observed (total of 83 points). Correspondence
between image points in the MOMS-2P data and object points in the field could be achieved with an accuracy of about
one pixel. In some cases, steep relief and absence of man-made structures led to uncertainties in determining overall
point positions. Such points become evident and can be eliminated in stereo data processing (3.2).
Precise geographic WGS-84 coordinates were obtained during post-processing of the GPS observations with the
software GPSURVEY (TRIMBLE GPS). Reference station coordinates were determined independently at the GFZ
Potsdam with the EPOS software (Angermann et al., 1997) achieving an absolute accuracy of a few centimeters in X,
Y, Z. These coordinates and precise ephemeris were introduced into GPSURVEY baseline processing. In the result
average internal accuracy for all rover points amounts to 0.08m in X, 0.07m in Y and 0.04m in Z leading to an average
point error of 0.11m. Assuming a factor of 10 between internal and external GPS accuracy, the results are in the
expected range of accuracy and represent high-quality control points.
3 STEREO DATA PROCESSING
In the past it was demonstrated that stereo processing of MOMS-2P 3-line imagery allows for DEM generation with
10m accuracy and better (Kornus et al, 1999). The main steps of the developed approach and results of its
customization for the study area in Kyrgyzstan are described in the following sections.
3.1 Image matching for tie point generation
DLR software for automatic image matching - developed for the stereo scanner projects MEOSS, MOMS-02, and
MOMS-2P - was used to generate tie points between the image strips of the 3 looking directions. Though massive
numbers of tie points are generated by this software (in our case about 424,000 points), the emphasis was put on the
quality of the input points for the bundle adjustment. The actual matching schemes are described by Lehner and Gill,
(1992) and Lehner and Kornus, (1995). Subpixel accuracy was achieved by local least squares matching (LLSQM). For
input to the photogrammetric adjustment a coarse grid of best points (here 3557 points) was selected. Criteria for this
selection are the final correlation coefficient after convergence of the LLSQM (threshold used is 0.8) and the stability of
the results while interchanging the template role of the images in the stereo pairs (threshold here is 0.1 pixel). The
thinning is done so to get a fairly homogeneous distribution for the whole strip.
3.2 Bundle adjustment
The functional model of the photogrammetric bundle adjustment used here is based on the principle of orientation
images (OI), as originally proposed by Hofmann et al., (1982). The extended functional approach used for MOMS
image orientation is comprehensively described in Kornus, (1999). Recent investigations still showed some problems
introduced by high frequency attitude variations during MOMS-2P data recording, which could be partly modeled by
means of a dense GCP distribution (25km - 30km mean distance) (Kornus et al, 1999). The real variations have
apparently been filtered out in the provided gyro data files. With correctly reprocessed gyro data height, accuracies of
7m and better are expected. In the bundle adjustment four groups of input information are entered. These are image
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000. 1261