The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
0 (due north) to 360 (again due north). The value of each
location in an aspect dataset indicates the direction the surface
slope faces. The slope is presented in Figure 4 and the aspect
regions defined from the reference CDED level 1 are by
geographic directions.
Figure 4: Slope classes of the reference CDED level 1
3.3 Land cover classes
Claudia C. Carabajal and David J.Harding. (2005) were among
the first to evaluate the influence of the vegetation on digital
elevation models. Canada’s forests are vast-nearly 50% of the
total landmass of the country (Natural Resources Canada, 2001).
The EOSD data used are represented by many classes. All the
classes were not used. The main classes were considered. Those
are: Coniferous, broadleaf, mixedwood and herbs. The
distribution of the species, dominated by coniferous-open
justifies the adoption of such classes. We combined coniferous-
sparse, coniferous open and coniferous dense to form the class
of coniferous. The same idea was used to create the broadleaf
and mixedwood classes. The herbs class was maintained.
3.4 Error Statistics
Statistics are computed for the differences between CDED level
1 and SRTM model per each segmented terrain classes as for
the study area. Several descriptive statistic measures were
employed, among which the mean, the standard deviation for
the both the error and the absolute value. We determined also
the root-mean-square error (RMSE). If RMSE is normally
distributed then, we can compute the linear error (LE) at 95%
confidence level (Maune et al. 2001). This indicates that the
95% of CDED level 1 points represent the true value with |error|
< LE. LE = 1.96 * RMSE. Also from Maune et al. (2001), the
contour interval (Cl) is related to RMSE by the relation: C.I =
3.2898 * RMSE. After normalizing the distribution of all the
differences by filtering, on the base of three times the standard
deviation we, respectively computed the LE and Cl. All the
results obtained are presented in table 1. The same statistics
were made for the four species of the land cover (Table 2). The
influence of the slope was removed since we considered only
the first slope classe (Slope < 5°); we filtered the other slope
classes.
Statistics
Land cover classes
Herbs
Broadleaf
Coniferous
Mixedwood
Accuracy
RMSE
9,0
12,7
10,7
12,0
LE
17,7
24,8
20,9
23,6
Cl
29,7
41,7
35,0
39,5
No of points
25
4412
33527
3460
Percentage
100%
99,10%
99,30 %
99,10%
Error
Minimum
-13,0
-37,0
-35,0
-37,0
Maximum
22,0
42,0
31,0
39,0
Mean
-1,5
2,8
-1,8
1,4
S.D
9,2
13,0
10,8
12,1
|Error|
Mean
7,7
10,3
8,2
9,4
Tableau 2: Differences in function of land cover in meters
between CDED and SRTM
Concerning statistics on ICESat’s points/profiles, the skewness
which characterizes the degree of asymmetry of a distribution
around its mean and the kurtosis which describes the relative
peakedness or flatness of a distribution compare with the
normal distribution were added to the statistics presented in
table 2. ICESat data were filtered to remove samples that might
have been contaminated by cloud cover or other atmospheric
interference. Figure 5 below presents that correlation between
CDED level 1 and ICESat.
CDED Elevation (m)
Figure 5: Correlation between CDED level 1 and ICESat data
3.4.1 Altitude Error distribution
The altitude errors (differences) between CDED level 1 and
SRTM model are respectively in the range - 50 to 47 m, while
the standard deviation of the altitude is 15.6 m. In the following,
it is examined if the difference in mean error is statistically
significant for CDED level 1 versus SRTM model. For this, the
first task is to determine the equality of sample variances. We
used the R software; R. Version 2.6.2 (2008-02-08). The
observed F-Statistic equals to 0.9968 for CDED level 1 versus
SRTM model. With a p-value of 0.5541 at 95% confidence
interval, the true ratio of variances is not equal to 1. In this case
it was not possible to presume the equality of variances. This
indicates that the means errors of both DEM are significantly
different.
1426