The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008
1334
significant difference was observed between the flat and hilly
areas, the former being more accurate.
When comparing the Cartosat-1 high-resolution DTMs to the
high precision ICPs, it was obtained a mean value of residuals
(p) between 0.29m for (the 2.5m, 7.5m and 10m grid cell size)
and 0.37m (for the 5m grid cell size), a standard deviation (a)
between 1.62m (for the 10m grid cell size) and 1.81m (for the
5m grid cell size), a RMSE between 1.61m (for the 10m grid
cell size) and 1.81m (for the 5m grid cell size) and a LE90 of
2.60m (for all the resolutions).
When comparing the Cartosat-1 high-resolution DTMs to the
ADS40DTM, for the flat terrain it was assessed a RMSE
between 2.87m (for the 10m grid cell size) and 3.39m (for the
2.5m grid cell size) and a LE90 between 4.60m (for the 10m
grid cell size) and 5.50m (for the 2.5m grid cell size), while for
the hilly area a RMSE between 11.96 (for the 7.5m grid cell
size) and 12.89 (for the 2.5m and 5m grid cell size) and a LE90
between 18.20 (for the 7.5m grid cell size) and 19.80 (for the
2.5m and 5m grid cell size). The overall accuracy was between
6.30m and 7.20m for RMSE and between 8.10m and 9.70m for
LE90. Table 3 shows a summary of results.
When comparing the Cartosat-1 high-resolution DTMs to the
Reference 3D® specifications, there is little difference between
the DTMs generated at different resolution, and the Reference
3D®’s requirements are always fulfilled (Table 4).
Terrain’s
type
Resolution
(m)
F
(m)
a
(m)
RMSE
(m)
LE90
(m)
Flat
2.5
-1.11
3.20
3.39
5.50
5.0
-1.08
3.02
3.20
5.20
7.5
-1.07
2.83
3.02
4.90
10.0
-1.07
2.66
2.87
4.60
Hilly
2.5
-0.55
12.88
12.89
19.80
5.0
-0.59
12.87
12.89
19.80
7.5
-0.42
11.95
11.96
18.20
10.0
-0.55
12.56
12.58
19.30
Overall
2.5
-0.80
7.20
7.20
9.70
5.0
-0.90
7.10
7.10
9.50
7.5
-0.90
6.20
6.30
8.10
10.0
-0.90
6.50
6.50
8.40
Table 3. Accuracy of the Cartosat-1 high-resolution DTMs.
4.2 Medium-resolution DTM generation
Starting from input data, a 25m resolution DTM was generated
and its accuracy was validated both using a set of high precision
ICPs extracted from the original C-SAP dataset and using a
downsampled version of the ADS40 DTM.
The best medium-resolution DTM was obtained using five
GCPs. The accuracy evaluation based on the ICPs led to the
following results: p=-0.02m, cr=1.68m, RMSE=1.65m and
LE90=2.37m. On a global scale, the comparison of the
medium-resolution Cartosat-1 DTM with the downsampled
ADS40 DTM led to the following results: p=-1.45m, <j=7.76m,
RMSE=7.91m and LE90= 10.20m.
Table 5 shows the influence of the GCPs (number and geometry)
on the final Cartosat-1 image processing. Results have been
compared to the ADS40 DTM downsampled at 25m resolution.
DTM
resolution
( m )
Slope
range
(%)
Reference 3D®
requirements
(LE90)
Cartosat-1
DTM
(LE90)
2.5
0-20
10.00m
5.50m
20-40
18.00m
17.20m
>40
30.00m
21.90m
10
0-20
10.00m
5.20m
20-40
18.00m
13.70m
>40
30.00m
22.40m
Table 4. Accuracy of the Cartosat-1 high-resolution DTMs.
Statistics on residuals (LE90) computed with respect to
Reference 3D® requirements.
4.3 Low- resolution DTM generation
The investigation also tested the accuracy in the generation of
low-resolution DTMs from Cartosat-1 data. In this case, results
where tested also with respect to i) a set of high precision UCPs
extracted from the original C-SAP dataset, ii) a downsampled
version of the ADS40 DTM and iii) the NASA’s SRTM,
widely used in the remote sensing community.
With respect to the high precision ICPs, both Cartosat-1 and
SRTM DTMs showed similar results: 2.64m RMSE and 2.67m
RMSE, respectively for Cartosat-1 and SRTM, and 3.80m LE90
for both.
With respect to the downsampled ADS40 DTM, the Cartosat-1
DTM showed better results than the SRTM DTM. For the
formed it was observed a RMSE of 5.28m and a LE90 of 7.60m,
while for the latter a RMSE of 6.50m and a LE90 of 10.00m.
As it was observed also for the Salon de Provence (France) test
site, when comparing Cartosat-1 and SRTM low-resolution
DTMs the former shoes better overall accuracy (Gianinetto,
2008).
5. CONCLUSIONS
For the Mausanne les Alpilles test site it was fully investigated
the potentialities and limits of the 2.5m panchromatic Cartosat-
1 stereo images in generating DTMs using commercial off-the-
shelf software, so from the point of view of a typical user.
For the high-resolution DTMs (in the range of 2.5m-10.0m),
regardless the DTM’s grid resolution, the investigation pointed
out that in flat areas the accuracy is about 4-5 times better
(about 3m RMSE and about 5m LE90) than that in hilly areas
(about 12m RMSE and about 20m LE90). In any case, the
accuracy of Cartosat-1 DTMs fulfilled the requirements of the
Reference 3D®.
For the medium-resolution DTM (25m), the overall accuracy of
the Cartosat-1 DTM was about 8m RMSE and about 10m LE90.
Finally for the low-resolution DTM (90m), as it was observed
also for the Salon de Provence (France) test site, the Cartosat-1