International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
TK350 RMSZ | Bias RMSZ - bias F(slope)
[m] [m]
Open areas 23.3 22 20.0 023.9 *tano
Forest S13 7.3 49.0 * 11.4 * tana
check points 6.6 0 4.7, 22 * tan a.
C-band DSM DZ> | RMSZ | Bias RMSZ - bias
50m [m] [m] F(slope)
All points 1.19% 11.7 -5.4 8.3 +7.8* tan
Open areas 2.11% 9.9 -3.9 7.5 6.6 * tan ao
Forest 0.03% 13.6 -8.3 27T7.3*12n0
check points 0 9.4 -2.0 | 404 122* tana
Table 2: RMS discrepancy of the SRTM C-band DEM against
the reference DEM from the map 1 : 25 000 and check points
A direct comparison of the X-band DSM with the C-band DSM
shows root mean square Z-discrepancies of 8.4m or RMSZ=
4.8m + 12.2 * tan o and only a negligible bias. For flat terrain
the RMS discrepancy of 4.8m has to be divided by 1.4 under
the condition of the same accuracy of both data sets leading to
3.4m accuracy for each data set. A similar accuracy has been
reached in a smooth area in Arizona where the RMS
discrepancies of the SRTM C-band DSM against the DEM of
the USGS has been 3.7m or 2.7m + 22.7 * tan a. Also the X-
band DSM in a rolling area in Germany has been in the range of
RMSZ = 3m (Koch et al 2002).
A M hee
Y UU tS mL S e
OUS SERENO
CA ERAN QE, CSS
: N ^? iN S
: M
d
A
J
{ Len TN |
Table 3: RMS discrepancy of the TK350 DEM against the
reference DEM from the map 1 : 25 000 and check points
The achieved results in the forest areas cannot be accepted, but
also in the open areas the results are far away from the results
published by Chekaline and Fomtchenko 2000 talking about a
Z-accuracy of 5m to 7m. Only at the location of the well
defined control points this has been reached in flat areas.
6. ASTER
The Japanese ASTER sensor on the US EOS AM-1 platform
has in the near infrared band a vertical and by separate optics
also an inclined view with 27.2? nadir angle enabling a
stereoscopic coverage within the orbit with a height to base
relation of 1.7. The short time interval of just 55 seconds
between both images avoids problems of a change in the object
space. The pixel size is 15m on the ground. The near infrared
spectral range has the big advantage of good contrast also in the
forest areas, so with the exception of the areas covered by water
no larger gaps have been in the generated DSM and the
correlation was very high. 72% of the points are correlated
better than r=0.95 and 99% better than r=0.6.
Figure 3: contour lines based on SRTM elevation models
left: C-band right: X-band
The accuracy of both SRTM elevation models is similar;
nevertheless, the C-band data are available only with a spacing
of 3" corresponding to approximately 90m while the X-band
data are available with a spacing of 1” corresponding to
approximately 30m. The differences in the details can be seen
at the contour lines shown in figure 3.
5. TK 350
The TK350 are perspective Russian space photos flown with an
endlap of 60%. Film copies of photos taken from a flying height
of 220km have been available. The very large photo size of
460mm x 300mm leads with the focal length of 350mm to a
covered area of 284km x 189km or a size of the stereo model of
79km x 189km with a height to base relation of 2.0. The
ground resolution is announced corresponding to a pixel size of
5 to 10m, but an edge analysis of the used photos showed only
an effective pixel size on the ground of 13m, nevertheless, the
photos have been scanned with 16um corresponding to 10m on
the ground.
The image quality of the used TK350 photos was disappointing
— an extreme high number of scratches could be seen like also a
strong film grain requiring a scratch removal and low pass
filter. The poor contrast in the forest area was leading to larger
gaps and large errors. In the open areas the contrast was quite
better, but nevertheless the results of the generated DSM at the
location of the control points having optimal contrast was quite
better like against the reference DEM. For the analysis Z-
discrepancies exceeding 150m have not been respected.
ASTER DZ > | RMSZ | Bias RMSZ - bias
100m [m] [m] F(slope)
All points 0.15% | 28.7 20 |25.3:-13.2* tana
Open areas 0.1699 | 25.0 0.7.1. 21.7. 14.5.* tana
Forest 0.22% 31.2 4.5. 1.27.9 £18.5.* tana
check points 0 12.7-14| 0
Table 4: RMS discrepancy of the ASTER DEM against the
reference DEM from the map 1 : 25 000 and check points
The RMSZ in the open areas and flat parts of 21.7m
corresponds to a Spx of 0.85pixels. This is a satisfying
operational accuracy for such a sensor. Of course in the inclined
parts the quality is reduced like also in the forest parts. Like
usual, in the forest parts there is a bias of -4.5m (matched points
are located above the reference DEM). In addition the ASTER
DSM has been tested against check points. The RMSZ at the
check points, which have been used also as control points for
the scene orientation, corresponds to Spx = 0.5 pixels. In a
second test, GPS check points distributed over the city area of
Zonguldak were used — this resulted in SZ=13.9m. As next,
points were digitized from 1:2000 scale road maps produced by
the Highway Department of Turkey. This resulted in root mean
square differences for Z of 14.0m; however these values are
influenced by steep slopes, just beside the roads in the
mountainous region. All the check point results are quite better
like the result in relation to the reference DEM. Even in flat
areas there is a relation of 1 : 1.5 between the quality figures.
7. SPOTS
SPOT 5 images do have a pixel size of 5m on the ground. The
stereoscopic view is generated by viewing across the orbit, so
there is always a remarkable time interval between imaging
638
Internc
both s
change
SPOT
Resolu
model
These
— SPC
genera
genera
classic:
dimens
In the
August
interva
clouds
the san
|
_ =
Figure
le
The pa
with th
the left
14^ wi
24.3 gr
of the
similar
Figure :
Matche
The sce
-16.65°
theory :
a relat
window
surface
smooth
like mo
In figu
coeffici
points :
