tanbul 2004 
is a general 
SAR is not 
ible surface 
| ls again a 
ave InSAR. 
moved by a 
OR. This is 
ith a clear 
atistics, the 
iews of the 
rd deviation 
ached with 
mean value 
p to 26.1m 
n has to be 
p area, the 
up to 3.8m 
n the forest 
n top of the 
the height 
llax of 0.7 
uracy level 
lels. Under 
ected from 
overing the 
removed if 
listance are 
t can never 
atic image 
> Synthetic 
her details 
| Elevation 
> of Digital 
Nashington 
Handbook, 
  
  
ASSESSMENT OF DEM ACCURACY DERIVED FROM SPOT-5 HIGH RESOLUTION 
STEREOSCOPIC IMAGERY 
W. Kornus, R. Alamüs, A. Ruiz, J. Talaya 
Institut Cartografic de Catalunya (ICC), Parc de Montjuic, E-08038 Barcelona 
wkornus@ice.es, ralamus@icc.cs, toni@ice.es, talaya@icc.es 
Commission I, WG 1/2 
KEY WORDS: SPOT, Space Photogrammetry, Orientation, Adjustment, DSM/DTM, Comparison, Accuracy 
ABSTRACT: 
This paper describes the derivation of Digital Surface Models (DSMs) from 3-fold along-track stereoscopic SPOT-5 imagery in the 
scope of the HRS (High Resolution Stereoscopic) study, organized by the Centre Nacional d'Études Spatiales (CNES) and the 
International Society of Photogrammetry and Remote Sensing (ISPRS). The orientation of SPOT-5 is reconstructed by bundle 
adjustment using a functional model based on correction polynomials. It resulted in an RMS-error of 2 m in Easting, Northing and 
Height at 17 check points. DSMs are produced for 4 test sites, which are located in different terrain types (mountainous, moderate and 
urban). An automatic region growing image matching process generates a dense point cloud in image space, which later is rigorously 
  
  
  
   
   
   
  
   
   
  
   
  
   
  
   
  
   
   
   
   
   
   
transformed into the object space and converted into a regular spaced DSM. The comparison with a digital terrain model (DTM) of 
superior accuracy yields standard deviations better than 5 m (16) in flat and moderate terrain and better than 10 m (16) in mountainous 
regions. An additional DSM covering the entire image scene (approx. 60 km x 80 km) is produced with a standard deviation of 
approximately 8 m using the commercial software ISAE and rational functions. The sigma values include all errors of the automatic 
matching process as well as the differences between the surface and the terrain model and therefore must be regarded as conservative. 
All results are finally summarized and conclusions are drawn from the study. 
1. THE SPOT-5 SATELLITE 
SPOT-5 is the first satellite of the SPOT family with along- 
track stereo imaging capability. The two HRS cameras are 
tilted by +/- 20 degrees and acquire nearly simultaneous 
stereopairs (at a 90-second interval) of 120 km swath, along 
the track of the satellite, with a B/H ratio of 0.84. A 
continuous strip of 600 km length is covered stereoscopically 
with 10 m ground sampling distance (GSD) across track and 
with 5 m GSD along track. The nadir looking panchromatic 
HRG instrument provides imagery in the mono-spectral 
bands HMA and HMB at 5 m GSD, in the multi-spectral 
bands XS1, XS2, XS3 at 10 m and SWIR at 20 m GSD. The 
ground pixels of HMA and HMB scenes are interleaved to 
enable the interpolation of so-called THR ‘supermode’ 
images(SPOT Magazine, 2000), having a nominal GSD of 
2.5m. A summary of the SPOT-5 payload and mission 
characteristics is given in (Fratter et al., 2001). The size of an 
image scene is 12000 x 12000 pixel (24000 x 24000 pixel in 
case of the THR ‘supermode’ image). 
2. INITIAL DATA 
The study was conducted using an image scene covering 
Barcelona and the surrounding area. The data set provided by 
CNES comprised 5 images (HRS1, HRS2, HMA, HMB and 
the THR supermode image) and auxiliary data (time series of 
orbit positions, velocities and attitude angles, look angles for 
each CCD element, etc.). The ICC provided the reference 
data set consisting of a regular DTM of 15 m grid size and 
1.1 m accuracy (1 opqy) covering the total area of the images 
and of 32 orthoimages of 0.5 m pixel size and 0.5 m accuracy 
(16) for 8 well distributed, approximately 10 x 14 km wide 
test sites (TS) (see figure 1). Figure 2 shows the two 
radiometrically improved and strongly reduced images of the 
THR and the HRS1 channels. The different resolutions of the 
HRSI image in scan and flight direction are clearly visible. 
  
  
  
  
aae dd 
mun 
  
  
  
Figure 1: Location 
i P2 1 P í 3 
es aie i v bo edi ejl it V I 
ISI AN EE E 
of 
the 
reference DTM 
rectangle) and the 8 test sites 
   
Figure 2: THR image (left, GSD: 2.5m x 
(d 
image (right, GSD: 10m x 5m) 
(light blue 
2.5m) and HRSI