nbul 2004 
ation of 
national 
Sy: 254- 
ıl image 
national 
(2), Dp. 
e affine 
f Photo- 
. Three- 
magery. 
| image 
- RPC 
, 23-27 
curacy. 
ng from 
OM). 
image 
ational 
26(3), 
sor ori- 
nal Ar- 
: 69-75 
00. An 
nagery. 
ensing, 
elation. 
tori, S. 
is of 
DV. In 
yrtland, 
l- 
ellite 
Re- 
ACCURACY EVALUATION OF DEM GENERATED FROM SPOTS HRS IMAGERIES 
M. S. Hashemian *, A. Abootalebi *, F. Kianifar “ 
“Department of Photogrammetry, National Cartographic Centre (NCC) of Iran, Meraj Ave., Azadi Sq., Tehran, Iran - 
(Hashemian, Abootalebi, Kianifar)@ncc.neda.net.ir 
KEY WORDS: Remote sensing, Hyper spectral, Generation, DEM, SPOT, Accuracy 
ABSTRACT: 
The present paper deals with the accuracy potential of SPOT-5 HRS satellite imageries for DEM generation. The goal is to derive a 
DEM from the along-track stereo data of the SPOT HRS sensor and to assess the accuracy by comparison with DEM data of superior 
quality. Following a brief description of the instrumentation and the data set used, the quality of the DEM generated from these 
imageries is assessed in several steps. For georeferencing of the stereo pair of images, some ground control points (GCPs), extracted 
from digital vector maps of the study area. A DEM of 10 meter grid size was generated from those images. And finally, the resultant 
DEM was compared with an existing and sufficiently accurate reference DEM of the study area. As can be seen below, the findings 
and results are very promising. 
1. INTRODUCTION 
DEM generation from satellite imageries scores over 
conventional methods of DEM generation, using topographic 
maps or aerial photographs. The global availability of satellite 
images allows for more quickly data processing for an 
equivalent area. In other words DEM generation using satellite 
images, draws continuous efforts mainly because of frequent 
imaging, global coverage and low cost. Though still lacking the 
accuracy and the amount of extracted information in 
comparison to the traditional aerial photogrammetry, the advent 
of ever higher resolution satellite images attempts to diminish 
the deficiency (Kim et al, 2001). In this regard, the present 
paper focuses on accuracy evaluation of DEM generated from 
SPOT-5 HRS satellite images. The derivation of terrain models 
from along track stereo data from space has up to now only 
been possible with the German MOMS-2P and the American 
ASTER sensor on TERRA (Reinartz, 2003). These sensors have 
lower resolution equal to 15-18 meter pixel size, but the new 
HRS sensor on SPOT-5 produces along track image stereo pairs 
with 5 meters ground sampling distance. This resolution is more 
suitable for obtaining higher accuracy in DEM generation 
process. In this study, an automatically generated DEM from a 
stereo pair of HRS images is compared with another reference 
DEM, generated from digital vector maps. In this way, the 
suitability of these images for DEM generation is investigated. 
2. INSTRUMENTATION 
SPOT-5 belongs to the latest generation of SPOT missions with 
significant improvements in terms of on-board instruments and 
autonomous system of positioning and attitude control that will 
enable high absolute location accuracy. HRS produces images 
with two optics looking forward and backward with 20 degrees 
with respect to the nadir direction. It has a spatial resolution of 
10 meter across track and a ground sampling distance of 5m 
along track. The swath of the HRS is 120 km with 12000 CCD 
elements. Forward and backward acquisitions cannot be 
performed at the same time. As a consequence, the maximum 
stereo segment that can be acquired is a little bit more than 60 
km. The size of pixels on the ground is 10m x 10m. However, 
HRS instrument has been designed for a ground sampling 
distance of 5 meters along the track. In a direction close to the 
epipolar planes, this along-track over-sampling allows higher 
altimetry accuracy of the DEM to be obtained. 
3. DATA USED 
The study area (Rasht region) located in the northern part of 
Iran. It includes both mountainous and flat terrain, fully covered 
with vegetations in the non-urban parts. It is essential to test the 
performance over diverse terrains and different conditions for 
scene acquisition. In particular, the terrain slope may affect the 
performance of automatic DEM generation. The terrain slope 
causes the layover effect and distorts an image. Such a varying 
relief terrain, allows for a more realistic evaluation of the 
results. The following data were used for this study: 
e 8 bit image data (size 12000 lines x 12000 pixels = 
120 km x 60 km) of the study area from two viewing 
directions in TIFF format. The data have been acquired 
on 2002-07-13 with a sun elevation of 69 degrees and 
nearly no clouds. 
e 1:25000 scale 3D map of the study area in digital 
format (Coordinate system: UTM; datum: WGS84). 
e DEM of 10 meters resolution produced from 1:25000 
scale topographic maps of the study area. These 
topographic maps were by Photogrammetric method 
using 1:40000 photographs in 1993 (Coordinate system: 
UTM; datum: WGS84). For the generated DEM, the 
RMSE of errors within 90% confidence level was found 
equal to 1.6 meters. This data is used as reference data 
for this accuracy assessment study. 
4. PROCESSING STEPS 
PCI Geomatica ver 8.2 was the main software package used for 
processing the data. In georeferencing step a total number of 99 
ground contro! points (GCPs) were used for each image of the 
stereo pair. They were extracted from 1:25000 scale digital 
vector maps of the area. Also, to improve matching between 
stereo images, a total number of 100 tie points (TPs) were 
collected. Table 1 shows the accuracy results on these GCPs for 
each image. The distribution of GCPs and TPs in the two 
images has been shown in Figure 1.