The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008 
1333 
3. METHODS 
3.1 Data processing 
The Cartosat-1 data processing was performed with the 
commercial off-the-shelf software ENVI 4.3® and involved the 
following aspects: i) image pre-processing, ii) optimization of 
the DTM’s extraction procedure and, iii) the analysis of the 
influence of GCPs in the modelled DTM. 
Before generating the DTMs from the Cartosat-1 images it was 
studied the influence of the parameters involved in the 
generation process (i.e., number of tie points used, search 
window and moving window sizes, correlation coefficient and 
terrain detail). A sensitivity analysis led to the final optimal 
configuration as shown in Table 1. 
The 60 tie points used in all the tests have been automatically 
detected using a regular grid scheme which covered the entire 
images. By increasing their number it was not observed any 
improvement in the Y parallax (1.08 pixel). 
Regarding the computational time it is to be said that reducing 
the terrain detail level, from level 7 to level 6, it was obtained a 
decrease of the processing time by a factor of 4.6 and without 
any sensible decrease in the final DTM’s accuracy. 
Parameter 
Value 
Number of tie points 
60 
Search window size 
800 pixel 
Moving window size 
30 pixel 
Correlation coefficient 
>0.8 
Terrain detail 
level 6 (over 7 levels) 
Table 1. Optimization of the DTM’s automatic extraction 
procedure. 
3.2 Ground control points selection and their accuracy 
This study investigated the multi-resolution DTM’s generation 
with respect to the influence of the number and configuration of 
the GCPs used. Thus, different configurations were tested in 
order to reach the best results with the fewer number of GCPs 
(Lang, 1999). 
The GCPs supplied by the PI have been measured with GPS 
instruments and have a vertical accuracy of 0.05m (Kay and 
Zielinski). 
3.3 Validation of results 
Results obtained from the data processing of the Cartosat-1 
stereo images were checked with respect to: 
■ A set of independent check points (ICPs) extracted 
from the original C-SAP dataset. These data were 
used for all the DTMs generated; 
■ The ADS40 DTM; 
■ The NASA’s SRTM DTM. 
All the dataset were first converted into the UTM-WGS84 
F31N reference system. From the high-resolution ADS40 DTM 
four downsampled DTMs were generated in the range of 2.5m - 
10m. In addition, a medium-resolution DTM with 25m grid 
spacing was derived from the high-resolution DTM. For the 
validation of the low-resolution DTM, two reference DTMs 
were used: i) a downsampled version of the ADS40 DTM (90m 
grid resolution) and ii) the SRTM DTM resampled from its 
original resolution (60m x 90m) to 90m grid resolution. 
Regarding the accuracy of reference DTMs, the Linear Error 
(LEa) is a measure typically used for accessing the overall 
accuracy of elevation data at the fixed confidence level a. It is 
assumed that the North and East coordinates are correct and a 
comparison in the elevation is performed towards reference 
measures: an LE90=2.5m implies that 90% of the measures to 
be tested vary from the reference measures by 2.5m or less. 
Accuracy of references DTMs are as follows: 
■ ADS40 DTM: this is a high resolution DTM with a 
RMSE=0.6m in elevation, as from Pi’s specifications; 
■ SRTM DTM: absolute LE90=6.2m in elevation for 
Eurasia (Rodriguez et al., 2006). 
Another term of comparison used for evaluating the DTM’s 
performances were the requirements for IGN’s and Spot 
Image’s Reference 3D®. The Reference 3D® is a lare second 
DEM extracted by means of automatic correlation from SPOT 
S/HRS stereo data. Its specifications are given in Table 2 
(Gachet and Favé, 2006). 
Parameter 
Specifications 
DEM resolution 
larc second (~30m on the Equator; 
21m at 45° of latitude) 
Planimetric 
absolute accuracy 
15m at 90% confidence level 
Altimétrie 
absolute accuracy 
10m at 90% confidence level, for 
slopes lower than 20% 
18m at 90% confidence level, for 
slopes included in 20% and 40% 
30m at 90% confidence level, for 
slopes greater than 40% 
Planimetric 
relative accuracy: 
10 m at 90% confidence level 
Altimétrie 
relative accuracy 
5m at 90% confidence level, for slopes 
lower than 20% 
15m at 90% confidence level, for 
slopes included in 20% and 40% 
28m at 90% confidence level, for 
slopes greater than 40% 
Table 2. Reference 3D® specifications. 
4. RESULTS AND DISCUSSION 
4.1 High-resolution DTM generation 
Four high-resolution Cartosat-1 DTMs were generated in the 
range of 2.5m - 10.0m. Because of the huge amount of data 
involved, the accuracy analysis of the high-resolution DTMs 
was performed on two subsets with different characteristics: i) a 
flat area with a dominant land cover of cultivated/crop fields 
and ii) a hilly area with an elevation range from 53m to 470m 
and a dominant forest land cover type. The subsets were 
selected as being representative samples of all the image 
content. 
The best results were achieved using five GCPs (four in the 
comers and one in the middle of the image), regardless the 
resolution of the DTMs generated. On the other hand, a