anbul 2004
ne thirds of
the contour
racy of the
table using
ılated from
ence of the
DEM using
SE, or root
ates. Using
> calculated
lopographic
1
ccording to
[able 3). In
? accuracy
valuated as
adel (DSM)
). Accuracy
nous areas
ferred from
at elevation
5 the slope
1d slope are
topography
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B2. Istanbul 2004
Test Nub. Of Min/Max Mean | Standard
Areas Points deviation
Flat 442525 -18,776 / 58.369 S 756 8,732
Flat 259488 -16,800 / 35.512 3133 7,537
Flat 159600 -31,121 / 65.109 17,888 16,955
Flat 73369 -8,155 / 18.853 4,101 3,485
Slope 101592 | -37.509 / 135 586 | 24.580 18.813
Slope 101592 |] -62.112/137.722 | 30.785"] "28.164
Slope 117196 | -37.509/135.857 | 37.406 | 21.089
Slope 108724 | -37.509 / 135.857 | 42.962 | 24.536
Rough 190350 | -82.079/ 273.138 | 67.910 | 47.264
Rough 190350 | -116.527 / 237,497 | 19.340 | 35.326
Rough 190350 | -232.174 / 172.750 | 2.489 41.679
Rough 190350 | -158.581 / 256.810 | 69.867 | 57.341
Table 3. Minimum/maximum errors, mean and standard
deviations of chosen test areas in different
topography.
Elevation accuracy is decreasing dramatically especially for
rough areas with high altitudes. For smaller slopes (up to 6°)
better (significant) results can be obtained since the radiometric
disparities between Fland F5 images of stereo pairs are small
due to their high resolution. According to the evaluation of the
two of the extracted radargrammetric DEMs (i.e., 2L and 4M
solutions) our results revealed that 2L solution is better for flat
and moderately sloped areas. However for high altitudes larger
deviations, which can be accepted as gross errors, are observed.
On the other hand, using 4M-detailed DEM, better elevations
are calculated for high altitudes whereas inferior elevations are
calculated for flat areas.
3. CONCULUSIONS
As a result of imaging geometry of radar, elevation accuracy is
strongly related to the relief type and slope. Execution of DEM
from SAR imagery is a dilficult task due to the characteristics
of SAR imagery and conflicting requirement of stereo DEM
extraction. SAR images respond very strongly to the terrain
slope. Radiomatically, slopes facing the sensor are very bright
duc to the direct reflection, while slopes facing away from the
sensor are dark. Geometrically, mountain peaks are shifted
towards the sensor, causing foreshortening of the slopes facing
the sensor and stretching of slopes facing away from sensor. In
extreme cases tops of mountains are imaged before bottoms and
back slopes are completely shadowed. Automatic DEM
derivation based on image matching requires that the same area
looks similar in two images of the stereo pair. Radiometric
disparities and geometric distortions of SAR images may cause
too large differences between images for a successful matching.
In addition to strong radiometric terrain induced distortions,
SAR images are corrupted by random speckle noise. The noise
may cause spurious matches that forces the use of relatively
large templates for matching and also decreases the sharpness
of the determined peaks. All these factors contribute to the
lower quality of SAR DEMs (PCI, 2001). In this research
RADARSAT fine beam images offer separations (convergence
angle) 8°. Due to the high resolution of FI-F5 images, for the
flat areas with small slopes (0% to 3%) effect of radiometric
and geometric disparities are less and the quality of DEM is
better. It is determined that standard deviation is ranging from
4m to 8m for small slopes. Standard deviation increases up to
20 m for medium slopes (396-15 9$). However, for steep slopes
(higher than 15%) the stronger geometry of F1-F5 is completely
cancelled out because of too large geometric disparities, and
calculated elevations are not significant. In the research area
there is a linear relation between orthometric heights and
extracted stereo DEM. The major characteristic of the research
area is that slope increases as altitude increases. This nature of
the relief causes significant deviations for calculated elevations
of rolling topography of the research area. Consequently, for
flat and small slopes difference value of stereo and topographic
DEMs gives better information about land use types than the
terrains with strong slopes on high altitudes.
REFERENCES
Hijazi J.,2001. Elevation extraction from satellite data using
PCI software. In:First Symposium on Space Observation
Technologies for Defence Applications,Abu Dhabi, United
Arab Emirates, 19-20 August 2001.
Greenfield, J. S, 1991. An operator-based matching system.
Photogrammetric. Engngineering and Remote Sensing
Leberl, F. Maurice, K., Thomas, J.K., MillotM., 1994.
Automated radar image matching. experiments. /SPRS journal
of Photogrammetry and Remote Sensing.
Marra, M., Maurice, K. E., Ghiglia, C. D., Frick, G. H., (1998).
Automated DEM Extraction Using RADARSAT ScanSAR
Stereo Data, IEEE.
Min-Ho Ka and Man-Jo Kim, (2001). DEM Generation Using
SAR Stereo Technique with RADARSAT Images Over Seoul
Area. Korean Journal of RS, 17 (2) pp.155-164.
PCI Geom. Grp, 2001. RADARSAT DEM Componant Ver.8.2
Toutin, Th., 1995. Generating DEM from stereo images with a
photogrammetric approach: examples with VIR and SAR data.
EARSeL Advances in Remote Sensing
Toutin, Th., 1995a. DEM Generation with Photogrammetric
Approach: Examples with VIR ve SAR. FARSeL Journal of
Advances in RS, 4 (2), pp 1 10-117
Toutin, Th., 1995b. Multisource Data Fusion with An Intgrated
and Unifiedgeometric Modelling, EARSel Advances in Remote
Sensing,.A (2-X).
Toutin, Th., 1998. Elevation de La Précision des Images de
RADARSAT. Jour. Can. de Télédétection, 24 (1), pp 80-88
Toutin, Th., 2002. Evaluatin of radargarammetric DEM from
RADARSAT images in high relief areas. /EEE Trans. on
Geoscience and Remote Sensing 38 (2), pp 782-789.
ACKNOWLEDGEMENTS
We would like to thank TUBITAK for funding this research.
