The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
742 
The satellite position and velocity is given as precision orbit 
data in the SUP-file, ancillary 8, both in ECI and ECR 
coordinates for every minute. The position at imaging time is 
calculated using a Hermite interpolation, considering the 4 data 
points around the point of interest. 
x(t) = J^(x i F 0i +x i F u )Hf 
/=1 
where 
j*i h ‘j 
F \i = t-tj 
(17) 
(18) 
where x is the orbit position vector and t the time of interest. 
5.3 Tests 
To check the correctness of our processing, the coordinates of 
the upper left pixel are calculated without the modifications that 
were made to adapt the data to DLR software. This coordinate 
is then compared to the coordinate calculated using the 
coefficients given in the LED-file, ancillary 1. These 
coefficients were e.g. used for Cal/Val tests (Saunier, 2007). 
For the Catalonian test site, the difference is approximately 240 
m, mainly in flight direction. For the French test site, the 
difference is smaller, but still in the order of approximately 60 
m, also mainly in flight direction. This difference results 
probably from the different versions of processors used. 
Especially the pointing alignment parameters given in the SUP- 
file, ancillary 13 have changed. 
We thus decided to use GCPs to estimate boresight angles using 
the DLR developed software ESTIMATE. 
5.4 RPC Generation 
Since for a RPC-based approach no new software had to be 
developed, the first idea was to generate RPCs for the PRISM 
images. Therefore, a three dimensional grid of control points is 
generated over the whole image from the exterior and interior 
orientation. This is done by a modification of the software 
ORTHO, developed at DLR (Müller, 2005). The estimated 
boresight angles are used as input for ORTHO. 
The RPCs are then computed as described in (Lehner, 2007), 
using XDIBIAS RPC generation software, developed at DLR. 
To check the RPCs, coordinates of the control points were 
recalculated using the RPCs and compared to the original 
coordinates. Figure 3 shows the plot of these residual vectors 
for a nadir image of a test site in Catalonia/Spain. 
The results are similar for forward and backward images as well 
as for other test sites (Germany, France). The residuals are 
smaller than one pixel; however, they may affect the DEM- 
generation. Regarding the residual behavior in row direction, 
there seems to be an oscillation with an amplitude of 
approximately one pixel. In order to find the reason for this 
oscillation, the attitude angles are examined. 
ir 
12000 .-/I 1 
an 
6000 
2000 1 j J 
'if. I, 
.ill.) In,. 
,llljllm 
/fflttttn 
,ttutp9n 
s'titfn,. 
»ntntlS 
m/rnn 
nftffPH 
• mummunm 
////////////✓✓✓//// 
'////////////////// 
'/✓///✓///////✓//// 
muiiitntuu 
*uuitumu44 
*Ui4Uinn*444 
I 
III I 
•Miiiumm 
•ni un mm 
•mnnmm 
mimi 
tin 
ut It fin 
ill Itti ft 
Ilf ittm 
I if if ft !H 
Utftltn, 
,. » I m,,. 
;i| 
numi 
mum 
mum 
mum 
mtiiitttltnnmitmm 
ntltnnttfttimitttttln 
mtttfttftmm nttntm 
mm 
mil/ 
mill 
mm 
mu 
mu 
tim 
nm 
mm 
mm 
mm 
I 1 Pixel: 
L 
2000 4000 6000 8000 10000 12000 14000 16000 
Figure 3: Residuals between original control point coordinates 
and those calculated by RPCs for nadir image of 
Catalonian test site. 
lines 
Figure 4: Residuals of pitch angle after subtracting a second 
degree polynomial for nadir image of Catalonian 
test site. Values are in degree. 
When plotting the attitude angles for an image, they seem to 
have a linear behavior. However, when we estimate a second 
degree Legendre polynomial as trend line and subtract it from 
the original values, an oscillation is clearly visible. Figure 4 
shows the residuals in pitch angle for the nadir view of the test 
site in Catalonia/Spain. The amplitude of the oscillation is small, 
however, in the images, it results in deviations of up to one 
pixel. For the yaw angle, the plot looks similarly, while for the 
roll angles, the residuals are slightly smaller. The oscillation 
may result from vibrations caused by the satellites momentum 
wheels. A similar oscillation is also known from other cases, e.g. 
MOMS-2P (Lehner, 2003). Due to the high resolution of