The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008
1373
aDTMs behaved similarly and in near flat terrain they showed a
high accuracy (RMSE<4.52m and LE<7.00m). With increasing
the slope the RMSE and LE90 rapidly increased until
unacceptable values (greater than 22.00m for the RMSE and
greater than 35.00m for the LE in the range of 30°-40° for
slopes). It is to be noted that for the Salon de Provence test site
94.12% of samples had a slope less than 20°, consequently the
resulting DTM’s accuracy was better than 10.74m for RMSE
and better than 17.00m for LE90.
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 DTM
generated using automatic correlation from SPOT-5/HRS data.
Its specifications are given in Table 4 (Buillon et al., 2006).
As shown in Table 5, the Cartosat-1 aDTM fulfilled the
Reference 3D® requirements while the rDTM did not.
4.2 Low- resolution DTM generation
The performances of the Cartosat-1 satellite were also tested for
the generation of low-resolution DTM (90m grid resolution).
As term of comparison it were used:
- The NASA’s SRTM DTM;
■ The MNTDBTOPO® DTM downsampled to a 90m
cell resolution;
■ A set of ICPs extracted from the original C-SAP
dataset.
Results show that the 90-meters Cartosat-1 aDTMs performed
better than the SRTM DTM and overcome the SRTM’s
specifications, while the 90-meters Cartosat-1 rDTM poorly
performed and did not met the SRTM’s specifications.
Regarding the comparison to the higher precision ICPs, the best
results were obtained for test #14 (p=-0.12m, o=3.16m,
RMSE=3.09m and LE90=4.09m), followed by test #5
(p=0.69m, <T=3.32m, RMSE=3.32m and LE90=5.34m) and the
SRTM performed worse (p=0.51m, cr=3.37m, RMSE=3.33m
and LE90=7.38m).
Even if looking at mean values of residuals, standard deviations
and RMSEs the Cartosat-1 aDTM and the SRTM have similar
values, the former is more accurate in terms of LE90: that
means that the Cartosat-1 low-resolution aDTM has fewer gross
errors than the SRTM.
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
I I slopes greater than 40% |
Table 4. Reference 3D® specifications.
When comparing all the low-resolution DTMs to the
downsampled MNT DBTOPO®, it seems that the SRTM has a
better accuracy (p=-0.87m, cr=5.92m, RMSE=5.99m and
LE90=8.00m) than the Cartosat-1 low-resolution aDTMs
(respectively p=-0.41m, cr=6.47m, RMSE=6.48m, LE90=9.00m
for test#5 and p=- 1.21m, a=6,82m, RMSE=6.93m and
LE90= 10.00m for test# 14), but these results may be due to an
inaccurate estimation of the elevation data in the downsampled
MNT DBTOPO® DTM.
As expected, the Cartosat-1 low-resolution rDTMs showed
inaccurate results (p=4.20m, cr=13.51m, RMSE=14.14m and
LE90=19.00m).
When considering the requirements for IGN’s and Spot Image’s
Reference 3D®, the Cartosat-1 low-resolution DTMs (both
relative and absolute models) met the requirements only for
slope less than 20%, while for higher slopes they did not met
the requirements.
5. CONCLUSIONS
For the Salon de Provence test site, it was fully investigated the
potentialities and limits of the 2.5m Cartosat-1 stereo images
for generating DTMs using commercial off-the-shelf software,
so from the point of view of a typical user.
When generating relative DTMs, if not properly geocoded it
was observed an error in the elevation values of some hundreds
of meters. After georeferencing, the RMSE was between 9m
and 14m and the LE90 between 16m and 19m.
When generating absolute DTMs, the optimum number of
GCPs was found to be nine, with a regular geometric
distribution (i.e., three in the upper part of the image, three in
the centre of the image and three in the lower part of the image),
while the minimum number of GCPs needed to get good DTMs
was found to be three (i.e., one in the centre, one in the lower
left and another in the lower right). In this cases, the absolute
DTMs fulfilled the standards required for the IGN’s and Spot
Image’s Reference 3D® and could be used for deriving such
product.
Finally, the comparison of the downsampled Cartosat-1 DTMs
with the NASA’s SRTM showed that the former fulfilled the
SRTM’s specifications and could be successfully used to fill
gaps in SRTM’s DTMs. A similar conclusion was also found
investigating the Mausanne les Alpilles test site (Gianinetto,
2008).
Slope
range (%)
Reference 3D®
requirements (LE90)
rDTM
(LE90)
aDTM
(LE90)
0-20
10.00
12.00
6.00
20-40
18.00
36.00
14.00
>40
30.00
58.00
27.00
Table 5. Accuracy of the Cartosat-1 relative and absolute DTMs.
Statistics on residuals (LE90) computed with respect to
Reference 3D® requirements