Intern
ANALYSIS OF DIGITAL ELEVATION MODELS DETERMINED BY HIGH RESOLUTION
SPACE IMAGES The d
have b
least <
Kocak, Güven* , Büyüksalih, Gürcan*; Jacobsen, Karsten "x indepe
epipol
* Karaelmas University Zonguldak, Turkey match
** University of Hannover, Germany it is n
calcul:
WG IV/7 orientz
the qu
KEY WORDS: matching, high resolution, satellite, SAR, analysis have
CORIE
ABSTRACT: betwee
depenc
Digital Elevation Models (DEM) generated by automatic image matching of high resolution space images are used often for very expres:
high resolution orthoimage production. The published accuracies of such DEMs are not in any case realistic and do represent often accura
only the quality which can be reached in flat areas with good contrast and without influence of buildings and vegetation. In the h/b * :
mountainous area of Zonguldak, Turkey, Digital Surface Models (DSM), representing the height of the visible surface have been ground
pixels -
generated by means of stereo models taken by TK350, ASTER, SPOT 5 and IKONOS. As reference, GPS control points and a DEM
of the topographic maps 1:25 000 are available like also the DSM of the SRTM X-band and C-band.
The ac
The DSMs have been reduced by an automatic process to DEMs and are analysed separately for forest and open areas and as a prograt
function of the terrain inclination. The image matching of TK350 photos did require a scratch removal and a low pass filter because layer ir
of several scratches and the film grain. Nevertheless, the results could be accepted only in the open areas. In the forest the image not nec
contrast was not sufficient and this was leading to large blunders and not accepted point matching. ASTER is using the near infrared inclinat
band (NIR) for the stereo coverage and here we do have a good contrast also in the forest. Caused by this, the image matching was detailec
without any problems also in the forest. In the available SPOT 5 level 1A and level 1B images, the grey value variation in the forest
is limited. With both products approximately the same accuracy has been achieved. Two IKONOS scenes taken with a time interval The ref
in the (
of 2 month have been used also for the DEM generation. The difference of 20? in the sun elevation caused quite different shadows,
so a matching was not possible in the forest areas, but also in the build up parts it was difficult. Quite different results have been
reached in another area where the IKONOS images have been taken from the same orbit; here the results have been excellent like
using t
|
|
|
|
|
|
|
|
|
| Depen
|
|
|
|
|
|
|
|
|
|
| like als
also with QuickBird images taken with a time interval of just 10 days. The height models of the Shuttle Radar Topographic Mission ellipsoi
(SRTM) are also referring to the visible surface, so in the forest areas a height shift can be seen. In addition there is a strong transfor
dependency upon the terrain inclination. The SRTM elevation models are better like the DEMs based on medium resolution space | nationa
images. | known,
| the hei
adjustm
| also so
1. INTRODUCTION 2. TEST AREA SRTM-
Digital elevation models are a basic requirement for any Most of the following investigations have been made in the area |
mapping and GIS. With the availability of SRTM C-band of Zonguldak, Turkey located north of Ankara at the Black Sea. /
DSMs (http://edesgs9.cr.usgs.gov/pub/data/srtm/) for large The area 1s Mmounialhous with an average inclination of 23%.
parts of the world, the situation has been improved, but Approximately 45% of the land area is covered by forest wil | With In
: ; : blished. In tree heights in the range of just 10m. In the north-west direction | Shuttle
up to now ihe data are not completely pu $ed parts of the Black Sea are included. As reference a DEM from | the maii
addition the accessible SRTM DSM has only a spacing of the topographic map 1 : 25 000 is available as well as some | South ar
3", corresponding to approximately 90m and some gaps control and check points determined by GPS. Bp — Shuttle
exist in very steep areas. Not for every purpose the iw, | system
accuracy and details are sufficient. | width oi
of only
DSMs can be generated with different space images and | covered
accuracies corrésponding to the ground pixel size, height to | the land
base relation and radiometric quality. Not in any case the | multiple
claimed and published accuracy of the DEM generation | DSMs +
is realistic. Sometimes the DEMs have been checked only | DSMs b
at well defined points not taking care about areas with not | the wav
optimal contrast, not taking care about the influence of n ; the used
the vegetation and also the terrain inclination. The detailed A E vx top of tl
analysis of DEMs generated by different space images will be Figure 1: test area Zonguldak (SRTM C-band DEM) with the | are Let
shown. main investigation area | The C-b
| ground)
| to 54°,
636