The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008
1346
Quality
% of vertices
Excellent
32
Good
37
Fair
15
Isolated
11
Suspicious
5
Table 6: Quality Report of DEM
4.2 DEM assessment
As can be seen in Table 7 and Figure 4, the DEM gives good
performance (Standard Deviation [SD] 3.83m) for the low altitude
area category. The SD at GCP locations is around 4.55m. As the
terrain altitude increases, the SD is also increasing. The results
degrade with increased altitude. However, this is expected while
using automatic DEM generation techniques. The SD value is
around 7m when compared with reference DEM. As the DEM
generated in the present study is surface DEM, the high values of
SD are encountered compared to reference DEM, i.e. bare earth
DEM.
The DEM analysis is carried out at two modes, i.e. point mode
and surface mode. At point mode, accuracy is tested at GCP
locations (old and new provided by PI), random locations with
different terrain conditions. Surface mode comparison is made
with reference to SRTM DEM and reference DEM provided by PI.
The assessment is carried out using the “raster to vector”
approach, using bilinear interpolation to determine the elevation
for each CartoSAT-1 DEM 1 Ox 10m cell position in the reference
data set, by interpolating from the nearest 4 vertices of the 2x2m
grid. The standard deviation and mean values of the elevation
differences (between the CartoSAT-1 DEM point and the
corresponding interpolated vertex) are calculated and stored for
analysis. The overall results of the comparison with the higher-
grade reference DEM is presented in Table 7 and Figure 4.
Description
Mean (m)
Std. Dev (m)
Old GCP Locations
-3.133
5.235
New GCP Locations
-0.718
4.546
Random Locations
-2.27
6.86
Less than 100m
3.01
3.83
100-300 m
-2.30
6.25
More than 300m
-7.05
7.33
Wrt SRTM DEM
-0.350
7.047
Wrt Reference DEM
-3.259
7.473
Table 7. DEM Accuracy Assessment
The mean values for all the three categories of altitude are ranging
from -7.05m to 3.01m with respect to the Standard Deviation
results recorded. The mean value when compared with SRTM
DEM is not significant, i.e. around zero, as it indicates both are
surface DEMs. When compared with reference DEM, which is
bare earth DEM, the difference is around -3.259m. Hence the
DEMs generated with Cartosat-1 are meeting the specified
accuracies. These DEMs are extremely useful for applications of
watershed management. The difference image of CartoSAT-1
DEM and reference DEM is illustrated in Figure 5.
0
8070966
8H—
1
15823971
2
3
15164332
14148240
—
4
13139328
—BBB
5
12206929
■Hi
6
11115134
7
9667716
8
8002272
9
6380709
H—
10
4901222
0
I
Figure 5. DEM Difference Image along with legend
Contours have been generated at 10m interval and overlaid on the
reference DEM which shows clear match (Figure 6) and also
compared with contours generated from reference DEM.