For the SAR data the achieved accuracy is worse, as 
documented by the RMS residuals being about twice as 
much as for the optical images. Due to the nominal pixel 
resolution of about 12.5 meters for the JERS-1 and 
ERS-1 SAR images, RMS residual lengths of about 30 
meters result therefrom. 
2.2. Stereoscopic Accuracy Assessment 
An a-priori stereo mapping accuracy may be evaluated 
from residuals between measured GCP coordinates and 
intersected 3D ground control point coordinates, the latter 
being calculated in a 3D stereo intersection procedure 
from the respective stereo pixel measurements (Raggam 
et al., 1990 [2]). Again, statistics of the individual 3D point 
residuals may serve for a stereoscopic accuracy 
assessment. 
The respective statistical values of the resulting 
coordinate differences, such as minimum, maximum or 
RMS-residuals, are summarised in Table 2. For the 
optical JERS-1 stereo model the resulting RMS height 
residuals are 54 meters. For image pairs of the SPOT 
sensor comparative values may be around some 10 
meters for panchromatic image data and stereo 
dispositions with a base-to-height ratio of around 1.0 
(Raggam et al., 1993). The worse accuracy of JERS-1 
results from a worse pixel size (18 meters versus 10 
meters) and a worse bas-to-height ratio, which is around 
0.3. 
OPS-1/ OPS-2 
(Q=~15°) 
SAR-1 / SAR-2 
(Q=-2.5° - 3°) 
  
Table 2: Statistics of a-priori stereo mapping for the 
optical (above) and the SAR (below) stereo model. 
The geometric disposition of JERS-1 SAR data does not 
propose a high accuracy. Due to the small intersection 
angles © of about 3 degrees maximum the range 
geometry is very sensitive to even small pixel errors, 
Which cause serious displacements in across track 
direction, i.e., mainly in Easting and Height. For the 
JERS-1 SAR stereo model the respective RMS 
accuracies are 170 and 140 meters. For comparison, 
accuracies of some 80 meters in planimetry and around 
40 meters in height were achieved in a similar experiment 
related to ERS-1 SAR stereo models (Raggam et al., 
1993). 
673 
2.3. Sensitivity Analysis 
Numerical simulation was used to analyse the effects of 
certain pixel errors in the stereo images, as they are 
usually evident from manual point measurement as well 
as automatic image correlation, onto the accuracy of 
corresponding ground coordinates, being achieved from 
3D intersection of projection lines. Disparities of +/-1 pixel 
were assumed in along-track and across-track for the left 
and/or the right image of the stereo pairs. These errors 
basically cause a shift of the projection lines/curves to be 
intersected for the determination of the ground 
coordinates. 
The point displacements resulting from the numeric 
simulation are summarised in Table 3. From this Table 
one basically can deduce a verification of the values 
given in Table 2 for the a-priori stereo mapping accuracy. 
The most severe effects are caused by the pixel errors 
corresponding to the orientation of the stereo baseline, 
which is in along-track (Ax) for the optical and in across 
track (Ay) for the SAR stereo data. For instance, 1-pixel 
disparities cause height errors of 65 meters for the optical 
and more than 100 meters for the SAR data. In the 
second case also a considerable displacement in Easting 
of around 120 meters is achieved. 
Comparative values were elaborated for SPOT and ERS- 
1 stereo models in a similar experiment (Raggam et al., 
1993). For SPOT again a height displacement of only 
some 10 meters was achieved, while the height errors for 
ERS-1 SAR stereo models were around 30 meters. 
Hence, SPOT and ERS-1 image pairs in general show a 
superior geometric imaging disposition than optical and 
SAR data from JERS-1. 
-/+ 16.7] -/+65.5 
+/- 1 | +/-128.9| -/+ 23.9] -/+ 107.5 
  
Table 3: Coordinate mismatch (in meters) caused by one- 
pixel errors in the course of stereo-intersection. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996