The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bl. Beijing 2008
thought also using fewer GCPs (tests #13, #11 and #9) similar
results were achieved (Table 2). Even if using only 4 GCPs (test
#9) led to better result in terms of LE90 (1.91m), altogether it
gave worst performances and produced more biased results (p=-
1.03 m, CT=1.59m, RMSE=1.67m) than the previous.
The minimum number of GCPs necessary to obtain a good
DTM was determined to be three (test #5). Using this
configuration it was obtained a mean value of residuals of
1.15m, a standard deviation of 1.97m, a RMSE of 2.24m and a
LE90 of 3.23m, which are 58.9% and 16.6% worst if compared
to the best one, respectively in terms of RMSE and LE90.
Comparing the Cartosat-1 aDTM with the MNTDBTOPO®
DTM it was observed a similar behaviour than that described
above. Generally speaking, increasing the number of GCPs the
aDTM’s residuals decreased until the values of p=- 1.22m,
<7=6.79m, RMSE=6.90m and LE90=10.00m (for test #14).
Once again, the use of three GCPs (test #5) gave good results.
In any case, no significant improvement was observed using
more than 9 GCPs both when comparing results with the ICPs
and with the MNT DBTOPO® DTM. Table 3 shows a
summary of results.
4.1.3 The influence of the terrain’s elevation and slope:
Once optimized both the algorithm’s parameters and the
configuration of the GCPs used in the generation process, two
aDTMs were selected as ‘best DTMs’ (test #5 and test #14) and
further analysis was carried on to study the DTM’s accuracy
with respect to the terrain’s elevation and slope characteristics.
Regarding the first issue, both the aDTMs behaved similarly
and in near flat terrain they showed a high accuracy
(RMSE<4.49m and LE<7.00m). With increasing the elevation
the RMSE and LE90 rapidly increased until values of 17.74m
and 29.00m for test #14 and 16.45m and 28.00m for test #5,
respectively.
Regarding the influence of the terrain’s slope, again both the
Test
ID
GCPs
(nr.)
GCPs configuration
ICPs
(nr.)
p(m
)
ct (m
)
M
(m)
RMSE
(m)
LE90
(m)
1
2
UR, LL
20
-1.50
4.49
3.47
4.63
7.91
2
2
UL, LR,
20
-2.46
3.90
3.36
4.53
8.91
3
2
LR, LL
20
-2.80
4.47
3.78
5.17
8.91
4
3
UL, LL, LR
19
-2.12
2.15
2.43
2.98
4.91
5
3
C, LR, LL
19
1.15
1.97
1.77
2.24
3.23
6
3
C, UL, UR
19
1.82
2.63
2.82
3.14
4.70
7
4
UL, UR, LL, LR
18
-1.52
1.80
1.78
2.32
3.97
8
4
C, UR, LL, LR
18
-0.63
2.24
1.95
2.27
2.98
9
4
C, UR, LL, UL
18
-0.31
1.59
1.31
1.58
2.97
10
5
C, UL, UR, LL, LR
17
9.26
2.50
9.26
9.57
13.38
11
6
UL, UR, LL, LR, RC, LC
16
-1.03
1.36
1.21
1.67
1.91
12
7
C, UL, UR, LL, LR, RC, LC
15
-1.46
1.68
1.59
2.18
4.70
13
8
UL, UR, LL, LR, RC, LC, C, CL
14
-0.12
1.64
1.26
1.58
3.77
14
9
UL, UR, LL, LR, RC, LC, C, CL,
CR
13
0.47
1.39
1.23
1.41
2.77
15
22
all
0
0.47
1.61
1.39
1.64
2.38
C=centre image, UR=upper right, UL=upper left, LR=lower right, LL=lower left, RC=right centre, LC=left centre, CU=centre upper, CL=centre
lower, all=all the 22 available GCPs.
Table 2. Accuracy of the Cartosat-1 absolute DTM. Statistics on residuals computed with respect to the Independent Check
Points.
Test
ID
GCPs
(nr.)
GCPs configuration
ICPs
(nr.)
p(m
)
ct (m
)
M
(m)
RMSE
(m)
LE90
(m)
1
2
UR, LL
20
-1.84
8.09
8.30
12.00
-1.84
2
2
UL, LR,
20
-2.86
7.77
8.28
12.00
-2.86
3
2
LR, LL
20
-3.14
7.52
8.15
12.00
-3.14
4
3
UL, LL, LR
19
-3.57
6.70
7.59
11.00
-3.57
5
3
C, LR, LL
19
-0.43
6.43
6.45
9.00
-0.43
6
3
C, UL, UR
19
-0.02
7.47
7.47
10.00
-0.02
7
4
UL, UR, LL, LR
18
-3.08
6.98
7.63
11.00
-3.08
8
4
C, UR, LL, LR
18
-1.94
6.87
7.14
10.00
-1.94
9
4
C, UR, LL, UL
18
-1.72
7.10
7.31
10.00
-1.72
10
5
C, UL, UR, LL, LR
17
8.06
10.00
12.85
16.00
8.06
11
6
UL, UR, LL, LR, RC, LC
16
-2.24
7.10
7.44
10.00
-2.24
12
7
C, UL, UR, LL, LR, RC, LC
15
-2.81
7.07
7.61
11.00
-2.81
13
8
UL, UR, LL, LR, RC, LC, C, CL
14
-1.69
6.98
7.18
10.00
-1.69
14
9
UL, UR, LL, LR, RC, LC, C, CL,
CR
13
-1.22
6.79
6.90
10.00
-1.22
15
22
all
0
-0.76
6.81
6.86
9.00
-0.76
C=centre image, UR=upper right, UL=upper left, LR=lower right, LL=lower left, RC=right centre, LC=left centre, CU=centre upper, CL=centre
lower, all=all the 22 available GCPs.
Table 3. Accuracy of the Cartosat-1 absolute DTM. Statistics on residuals computed with respect to the French DTM
1372