International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004
Subsequently, the island has been undergoing rapid
development in terms of expanding road networks and the
construction of hotels and holiday complexes.
3. DATA AND METHODOLOGY
In this study, analysing of Gokceada islands land cover by
means of Landsat Enhanced Thematic Mapper (ETM) was
aimed. Data acquired on Jun 14 2000 was used as remotely
sensed data. Together with these satellite image, 1/25000 scaled
topographic maps and existing land use maps were used for
rectification and ground truth. Additionally, in order to produce
digital elevation model (DEM) 1/25000 scaled standard
topographic maps were used. Stages of digital image processing
techniques were followed in order to obtaine land cover
categories of study site.
In the first step, the simple image pre-processing was carried
out including image enhancement, and geometric correction.
Enhencement techniques were applied to satellite image in
order to increase visual distinctions between features and
increase the amount of information that can be visually
interpreted from the data. These procedure includes various
techniques. Landsat TM image was enhanced using high pass
filtering and histogram matching (for minimizing atmospheric
effects) enhancement techniques to improve interpretability of
image.
Second stage is rectification process. To locate ground features
on imagery, or to compare a series of images, a geometric
correction procedure is used to register each pixel to real world
coordinates (Jensen, 1996). Map to image registration was
applied on image in order to prepare them for an accurate land
cover classification. Landsat 7 ETM image dated 2000 was
transformed UTM coordinate system by means of 1:25000
scaled standard topographic maps by using first order
polinomial and nearest neighbour resampling method.
In the third phase, ISODATA classification technique was
applied to classify the Landsat images of the island. The aim of
the image classification process is converting image data to
thematic data. ISODATA (Iterative Self Organizing Data
Analysis Technique) which is clustering method, classify pixels
iteratively, redefine the criteria for each class, and classifies
again, so that the spectral distance pattern in the data gradually
emerge (Goksel, 1998). Seven land cover types for Gôkçeada
are identified and used in this study, including: urban or built-
up land, barren land, green areas, olive grove, forest, water and
sand. Figure 2 showes the visual results of classified images in
2000 and table 1 shows the statistical results of classification.
Classes 2000
ha %
Water 379.13 1.31
Forest 3548.40 12.30
Green Areas 5416.64 18.77
Barren Land 9722.32 33.69
Olive Grove 6768.68 23.46
Urban & Build up (Mix) 2479.08 8.59
Sand 542,92 1.88
Total 28857.17
Forest Green Areas Urban R Built up
Le Lnd
—
Sand Samen land Bear OH Gran SE Water
Figure 2. Classified 2000 date image
712
Table 1. Results of classified images
At the last stage of image processing calculation of accuracy
assessment of classification was performed. Accuracy
assessment is an important feature of land-cover and land-use
mapping, not only as a guide to map quality and reliability, but
also in understanding thematic uncertainty and its likely
implications to the end user. In this study, accuracy assesment
of classification was calculated using a error matrix (Lillesand
and Kiefer., 2000), which showed the accuracy of both the
producer and the user. The classification accuracy in remote
sensing shows the correspondence between a class label
allocated to pixel and true class. The true class can be observed
in the field, either directly or indirectly from a reference map
(Janssen, and Vander Well., 1994). For accuracy assessment,
250 pixels were randomly selected from the ground truth
coverage. Land use maps and photographs taken for
documentary purposes were used as reference data to observe
true classes. The overall accuracy and a Kappa analysis were
used to perform a classification accuracy assessment based on
error matrix analysis. For the 2000 dated image, overall
classification accuracy for the seven classes was established as
87.5% and the Kappa coefficient was computed 0.863 (Bektas,
2003). A standard for land-cover and land-use maps is set
between 85 (Anderson et al., 1976) and 90 % overall accuracy.
The accuracy is sufficient for delineating of land cover in order
to analyse.
DEM is defined as any digital representation of the continuous
variation of relief over space (Burrough, 1986). By means of
digitized contour lines of 1/25000 scaled topographic maps in
every 20m interval, DEM of the study area were performed by
using interpolation procedure. The classified image dated 2000,
superimposed with DEM in order to obtain better visualization
that are impossible by means of two-dimensional analysis are
given in figure 3. Reliability of data sources, frequency of the
points selected on the land and the mathematical method used
in conversion are important for the quality and accuracy of the
digital elevation model.
Interni
Fig
The D
order t
are pa
for de:
to hav
aspect
land is
to oth
Cover «
The sl
attache
oriente
Raster
a laye
classif
Raster
data.
simpli
smallei
occure
layer.
This |
technic
analysé
GIS te
plannir
and as]
the qu
categoi
a resu
thresho
urban
problai
Figur
