International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B3. Istanbul 2004
that the orbitally georeferenced images are too much in the east
and usually in the south. It should be noted that these results are
the comparison to the reference, i.e., to the georeferenced ETM-
image using ground control points from the map.
Table 2. The comparison of the location of the images
georeferenced using orbital information to the reference image
georeferenced using ground control points
Error in east | Error in | Planimetric
direction north error
direction
MSS 1975 -915.0 -2.0 915.0
MSS 1983 -978.0 205.0 999.3
TM 1990 -1019.0 126.0 1026.8
ETM 1999 -414.0 1070.0 1147.3
. SRTM- -268.0 233.0 355.1
DEM
ASTER- -410.0 851.0 944.6
DEM I
ASTER- -898.0 481.0 1018.7
DEM 2
It is very likely that the positional accuracy of images
georeferenced using orbital information is higher than using
ground control points due to a small scale map used in the
ground control point collection. The master image was
georeferenced using 14 ground control points measured from
the 1: 500 000 scale maps (TPC G-4C, TPC G-4D, UK 1998)
and first degree polynomial transformation. The georeferencing
errors are very high due to the small scale map, the average
root-mean-square-error being 19.7 pixels.
3.4 Interpretation methods
Image interpretation methods were image clustering of the
Landsat-images to determine the general land cover, change
detection using the Landsat-images and image enhancement of
the QuickBird image to aid visual interpretation.
3.4.1 Clustering Landsat images: A general overview of land
cover was acquired using clustering analysis. In clustering
analysis, or unsupervised classification, the image pixels have
been divided into categories or clusters according to their
spectral similarities. The principle is that the spectrally similar
pixels belong to the same cluster and different pixels to different
clusters. When the clusters have been formed, they are
interpreted by comparing them to satellite image, reference
images like aerial images or maps, or ground survey data. The
Landsat MSS-, TM- and ETM-images were clustered into 30
clusters using k-means algorithm. (Richards, 1993). The clusters
were interpreted by comparing them to satellite image and
determining the most likely land cover.
3.4.0 Change detection using Landsat images: Change
detection was performed between the Landsat images in order
to determine the places of land cover changes. Change images
were formed by computing the differences of red and near-
infrared channels (ETM-channel — other image channel) and
clustering these two differences to 30 clusters using k-means
algorithm and the clusters were interpreted to categories no-
change and change. Also the type of change was interpreted if
possible.
Change images were computed using red channels (channel 5 in
MSS, 3 in TM and ETM) and near-infrared channels (channel 6
in MSS, 4 in TM and ETM). The purpose of using red channel
900
was to find changes concerning bare ground and mineral soil,
especially sand. The purpose of using near-infrared channel was
to find changes related to vegetation cover. The used pixel size
was 60 meters when comparing the MSS- and ETM-images and
30 meters when comparing the TM-and ETM-images. Ratio
images CH7/CH1 were also computed from TM- and ETM-
images and their difference image computed. It was hoped that
this channel ratio would be sensitive to sandy soils (see also,
Hofmann, K. — Geerken, R., 1998). It should be noted that the
seasons of images were different, so some changes are due to
the seasonal difference of vegetation. Also, some detected
changes can be due to the misregistration between images.
3.4.3 Image enhancement of the QuickBird image: Gradient
images were computed from the previously referred QuickBird-
image in order to enhance linear features like roads and paths,
or other signs about the human activity like houses or ruins.
Gradient images were computed using a Sobel gradient
(Gonzales and Wintz) with 3x3 filtering window for channels 1,
2 and 3.
4. THE CONSTANTLY
CHANGING DESERT BORDER
At Jebel Bishri environment largely dictates the possibilities to
carry out certain subsistence strategies, and this has also
affected the formation of particular types of archaeological
remains in the area which has been desert-steppe throughout the
Holocene, i.e., the past ten thousand years.
The specific archaeological remains left by different groups of
people are usually determined as tool-kits, assemblages or
technocomplexes (see definitions in Clarke, 1978). The desert-
steppe environment offers sites that have a good surface
distribution and are often visible for thousands of years on the
surface. However, the Mediterranean and semi-arid regions are
especially vulnerable to erosion. The limestone and sandstone
slopes of Jebel Bishri are easily eroded by winds, sand storms
and different run-off mechanisms caused by rains. It is known
that generally erosion causes decreasing artefact diversity from
slopes towards valleys (Evans - O Connor, 1999), but this is a
theoretical approach which does not take into account the
possible density of occupation and accumulation of
occupational layers at the sites beneath slopes.
The increasing desertification on the mountain is especially
detectable in the Landsat satellite data comparing the
movements of sands during the last thirty years. There was a
drastic increase of the sand cover from the year 1986 to 1993.
The peak of the sand movements was reached in 1993 (see
Meissner - Ripke, 1995). Some minor increase was still
detectable in the central area of Jebel Bishri and to the north
from the Landsat-7 ETM image analysis marking the difference
between the years 1993 and 1999. Monitoring these changes
allows us to understand the environmental changes, which have
affected the constrains to different livelihoods in the region.
The influence of a state in the livelihood of the nomads has been
detectable in our ethnoarchaeological study of the sedentarizing
nomads in the village of Shanhas in the western piedmont area
of Jebel Bishri. The CORONA satellite photographs also
provide evidence that agricultural initiations were carried out by
ploughing the ground also on the mountain in the 1960s.
Ploughing and agriculture have, however, been found to cause
catastrophic effects to the grazing grounds and the traditional
livelihood of the nomads. Well water does not suffice in the
Ir
d
Ci
dl
ni
th
th
th
ra
