its type existing at the
time imagery was obtained, land-cover categories
were determined for assistance to classification
procedures. The area and percentages of each land-
cover category is given in Table 1.
land-cover pattern and
The study site covers an area of 94 280 hectares.
As a matter of fact 73. X of the non-irrigated
agricultural lands is devoted to Wheat and Barley
production, which is 20 560 hectares. Since the
ambient temperatures of 1984 plant growing season
are generally higher than previous years, cereals
namely wheat and barley were harvested before the
LANDSAT image was acquired. For this reason, most
of the agricultural lands of the study area on the
imagery was occupied by fallow. On the other hand,
87 % of the irrigated lands over the study area is
given for potato production. Remaining 13% of the
irrigated area is mainly covered by Sugar beets
(12 X) and 1% is left for legumes. For this
reason sugar beets and legumes were considered
into the same category in order to improve the
classification efficiency.
Nonvegetated areas was also considered in the same
category with Fallow.
for machine-
shown
classes
obtained as
land-cover/use
processing were
Consequently
assisted image
in Table 1.
4.2. Classification Accuracy Assessment
classification schemes
likelihood and minimum-distance to means
approaches) and unsupervised clustering are
summarized in contingency tables 2, 3 and 4.
Accuracy assessments results, ranked according to
Percent Correct, Commission Error and Kappa
statistics are presented in Table 5.
Results from supervised
(max imum
With respect to Bolu study area maximum likelihood
classifier provided the most accurate results.
This method has given the highest Percent Correct
value of 86.6 and the lowest Commission Error of
13.4. The greatest Kappa value of 0.809 is another
indicator that reveals the maximum |ikelihood
approach ranked the best classification scheme
Minimum-distance
Percent Correct
among the others for this study.
to means classifier has given a
value of 82.4, Commission Error of 17.6 and Kappa
value of 0.751. These accuracy symptoms ranked the
minimum-distance to means classification technique
after maximum likelihood.
that the more bands use
classifier, the
classification results would be.
Karteris (1990) reports that the
sensing information tool may be
the three-band combination
other three-band sensors, less cost of acquiring
and analyzing the data, good classification
results etc.). Owing to the above, he considered
that a comparison between the recorded accuracies
of six-band and three-band combinations for each
individual category would provide useful
information. In most cases the difference in
accuracy between them was negligible (a maximum of
3.9 percent). These findings encourage the use of
second
would seem
likelihood
Intuitively, it
in a maximum
better the
Nevertheless
basic remote
considered to be
(colour composites,
three-band combinations in natural resources
classifications and forest mapping projects.
However he also stated that band 4 should be
included in all the three-band combinations.
In this study it was thought that it may be reason
of why supervised classification schemes used
three band combinations (Band 4, Band 3 and Band
2) have given high classification accuracies.
Unsupervised clustering has displayed the lowest
Percent Correct of 73.1, Commission Error of 26.9
and Kappa statistics of 0.617. According to these
consequences unsupervised clustering was not so
successful as being to other schemes to classify
the study area of Bolu province. The reason for
this circumstances may be resulted from the
relatively large stand sizes and pure cover types
provided sufficient training site statistics to
characterize the existing study area cover
adequately.
4.3. Area Estimation Accuracy
The classification performance indicated by 80 %
correct recognition of test fields is believed to
be adequate for satisfactorily estimating crop
areas (Bauer et al., 1979). Therefore in this
study Percent Correct value of 82.4 on the basis
of maximum |ikelihood classifier for overall study
area can be assumed acceptable
correct recognition to estimate land-cover/use
acreage. Table 6 presents the comparison of land-
cover class area percentages with land-cover class
area estimates based on classification of LANDSAT
data for the study area of Bolu.
percentage of
Table 1. The Area and Percentages of Each Land-Cover Category
in The Study Area of Bolu, July 1984.
Category Category Mame Area Percent of
Mo (ha) total area
1 Coniferous 38 800 41.15
2 Deciduous 26 800 28.43
3 Water 300 0.32
4 Potatoes 5 405 5.75
5 Sugar beets + Legumes 1 595 1.697
6 Fallow + Monvegetated 21 380 22.68
TOTAL 74 280 100.00
292
