1 2004
7071
dinates
e then
id. the
n error
? error
shold
indomly
ined by
S84 and
Helmert
ervation
nutes at
ferior to
the GPS
able to
:levation
standard
itline the
e SPOT-
using 14
dow, and
n RMSE)
using 13
selection
esult was
algorithm
for SPOT
1 lists the
(SD), its
abled the
allows the
ch DEM.
iate error
rtographic
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
Based on the results obtained in this study, the generation of
DEM from SPOT-HRV stereo-images can be done with
methods of digital restitution, leading to RMSE values less than
the pixel size. The sampling interval is one of the factors that
influences the quality of the DEM: The best results are obtained
for a cell size twice the pixel size (i.e., 20 m from SPOT-HRV ).
Increasing of this distance among sampled points is not a good
strategy because it is equivalent to a progressive generalization
of the DEM structure.
The influence of software is not obvious from the experiments
carried out. The accuracy of SPOT-DEM is similar for both
Erdas and Socet Set.
Finally, SPOT-DEM have been compared with the DEM
generated from a topographic map at a 1:25.000 scale. This
process implies the comparison of 2.200.000 points.
Comparing DEM was done by means a simple difference map
algebra operator. Table 4 shows the basic statistics. The
accuracy statistics of the cartographic DEM are similar to those
of SPOT-DEM. Comparing DEM was done by means a simple
difference map algebra operator. We can observe the small
differences in SPOT-DEM and cartografíc DEM.
Error (m)
Source data Software a n c
ME RMSE SD CI?
Ortho Base 1.5 7,7 7.4 +0,6
SPOT-HRV en 2 i 2
s Socet Set -4.6 8,6 7,3 +0,6
Cartographic 1 7,9 7.8 +0.6
* Mean Error
* Root Mean Square Error
* Standard Deviation
4 Confidence Interval for SD (95%)
Table 4. Error statistics for DEMs
5.2 DEM depuration results
We have conducted the depuration process based on the
hypothesis of a certain correspondence between correlation and
data reliability: The presence of a low correlation value is not a
definitive proof of poor quality, but is a valid warning signal
and has statistical significance. If this hypothesis is true, we can
carry out a cleaning procedure of the potential inaccurate points
without a significant loss of quality.
Figure 3 shows the errors of the depuration of the DEMs as a
function of the chosen correlation coefficient threshold. The
huge DFM (with no points yet removed) was denoted as
MDEOO. (her DEM were generated by previously deleting
those points whose correlation coefficient was less than a
threshold value (Table 3). For example, MDESO was the result
of taking a threshold value 0.50 for the correlation coefficient.
It can be noted that error did not rise significantly when the
number of eliminated points is increased, at least until a
correlation threshold of 0.93 (standard deviation, SD=7.9) or
0.94 (SD=8.0) is reached. On moving to 0.93, the quality of the
DEM significantly dropps (SD=12.2). MDE94 contains only
18.5% of the points of the massive original DEM (MDE00),
while the MDE93 contains 23%.
We emphasize that the depuration process does not imply an
improvement in accuracy statistics, but it contributes to making
the structure much more manageable in a GIS environment.
259
Error (m)
Wo
71. a ati BRE KEN
>95 >94 >93 >92 >91 >90 >85 >80 >75 >50 all
Threshold value 9/o
[= 2«- - 5p —Á4— RMSEz |
Figure 3. Errors in the DEMs according to the threshold
value of the minimum acceptable correlation coefficient
(test of 315 CPs). It’s possible to reduce the initial TIN to
only 19% of the points without any statistically
distinguishable loss of quality.
6. CONCLUSIONS
Automated DEM extraction using cross-track SPOT satellite,
has been known for 17 years. We concluded that SPOT images
will provide the opportunity for the generation of DEM with
RMSE Z-values less than the pixel size. We cannot conclude
that the accuracy results are affected by other factors.
Digital photogrammetric procedures generate points which, in
certain conflictive zones, may not be very reliable. These zones
are characterized by low correlation values due to the
radiometric differences between images or because they are in
the shade where the stereo-matching algorithms do not work
correctly. The presence of a low value of the correlation is not a
definitive proof of poor quality, but it is a warning signal.
Occasionally the converse may be the case: the existence of a
high value for the correlation may be accidental. The usual case,
however, is for a certain correspondence between correlation
and quality of the data.
Hence, the depuration of a DEM by means of threshold values
of the correlation coefficients seems to be a simple but effective
way of reducing the size of the data structure without significant
loss of quality. The tests that we performed supported this
hypothesis, and in our working zone we were able to reduce the
initial TIN to only 19% of the points without any statistically
distinguishable loss of quality.
It is to be expected that the optimal correlation threshold will
depend on such factors as the radiometric quality of the images,
the geometrical resolution, and even on the stereo-matching
algorithms used in the DPW. Since quality control procedures
are always required, however, it is not any great extra burden to
carry out the type of tests described in the present work in order
to "lighten" the DEM before it can be regarded as a finished
product.
One of the problems that can arise is the deficient localization
of the ground control points. While these points should be
spread out over all types of relief, the usual case is to take them
in the more readily accessible zones. Such deficient sampling
