id 
r 
m 
je 
in 
e 
id 
» 
t 
)T 
of 
ps 
on 
me 
y. 
a, 
sal 
"a- 
91. 
the 
ing 
L7= 
cal 
nts 
ra- 
ol. 
AGRICULTURAL LAND-COVER AND FOREST CLASSIFICAT ON 
OF LANDSAT DATA IN THE PROVINCE OF BOLU 
A.Nejat EVSAHtBtOGLU Mehmet TANKUT 
Assoc.Prof.Dr. Manager of Remote Sensing 
The University of Ankara and GIS 
Agricultural Engineering AS-1SLEM Remote Sensing 
Depar tment, TURKEY Engineering Co.Ltd., 
ISPRS Commission No: IV Ankara, TURKEY 
ABSTRACT 
The main purpose of this study was to analyze LANDSAT TM data by computer processing for mapping of 
Agricultural and Forest cover types in the province of Bolu, Turkey. Primary LANDSAT 5 TM data used for 
evaluation was portion of scene Path 178, Row 32 (Center geographic coordinates N 40*20'00" E 31*43'00") 
acquired July 16, 1984. Reference data used to support the analysis was consisted of identification of 
ground observation and recording on large-scale map the crop type and land-use of necessary number of 
fields in the test areas. Supervised and Unsupervised approaches were utilized for classification. Digital 
analysis was performed using ERDAS image processing system. Classification schemes were evaluated using 
contingency tables and were ranked using KAPPA statistics. The study results indicate Supervised 
Classification technique was more accurate than Unsupervised Clustering over the study area. Maximum 
Likelihood classification method was superior to any digital classification approach for the area. The 
study results also display that classification accuracy is dependent on the selection of a particular 
classification scheme and that remote sensing techniques utilizing computer-aided image analysis methods 
can be used to identify the Agricultural Land-Cover and Forest pattern in the study area. 
KEY WORDS: Classification, Image analysis, Image Processing, Landsat, Remote Sensing Application. 
1. INTRODUCTION 
applications, imagery types and analysis 
techniques. Satellite imagery is also recognized 
Sound Land Management requires timely and accurate as a useful too! for forest mapping and inventory 
information about the type, amount, availability (Buchheim et al., 1985; Borry et al., 1990). 
and condition of renewable resources being 
produced (DeGloria and Benson, 1987). The objective of this study was to evaluate 
several classification schemes for identifying and 
In all disciplines involving land management there mapping agricultural and forest cover types in the 
exist a need for timely, reliable information on province of Bolu, Turkey from Landsat TM data. 
which to base resource management decisions. One 
of the most important types of resource 2. LITERATURE REVIEW 
information required for comprehensive planning is 
a current data base of the vegetation/land/water The use of remote sensing for the investigation of 
surface cover, subsequently'referred to as "cover terrestrial resources has become increasingly 
type" (Fleming, 1988). common since the launching of the first Landsat 
satellite in 1972. Conventional aerial photographs 
Land-cover are used for many purposes. In land have of course been of great value for many years, 
consolidation projects and in environmental and and continue to provide important data. However, 
hydrological studies, accurate, up to. data remotely sensed data from satellites have a number 
information about land-cover on a regional scale of important advantages leading to their 
is often required. Knowledge of changes in land- increasingly wide adoption (Townshend, 1981). 
cover is becoming increasingly important from both 
the ecological and economical point of view Since Landsat MSS data became available, many 
(Janssen et al., 1990). efforts have been made to demonstrate the utility 
of remotely sensed data is developing land-use and 
Particularly in the last twenty years, remote crop area statistics on regional and national 
sensing employing digital Landsat data has level (Holko and Sigman, 1984). 
developed at a rapid pace. |t has became a 
practical tool for monitoring the environment and Gautam and Chennaiah (1985) have oriented their 
assessing our natural resources in a number of efforts toward studying the changes in land-use 
application areas. Nevertheless, computer - and land-cover in Tripura, India, using  LANDSAT 
generated land-cover classifications require images of two different dates and to see how well 
significant improvement in both their accuracy and data obtained help in the study of geographical 
specificity in order to be used operationally in phenomena with special reference to land-use and 
many applications. One faced to the solution of cover. Ninety percent accuracy of each land-use 
this problem is to improve the quality of the raw category has been achieved when compared with 
data. This has been initiated with the launch of existing data compiled on ground surveys by the 
the Landsat Thematic Mapper and SPOT satellites working plan of the Division of Forest Department. 
(Lo et al., 1986). Equally important is to decide 
the most proper methods to analyze and classify Many studies for developing land-cover and forest 
the data for a particular region. information have shown that digital processing of 
Landsat MSS data requires some interpretation 
On the other hand, timely and accurate knowledge and/or integration using aerial photography and/or 
of forest composition and condition can be an some type of ground observations (Hoffer and 
invaluable forest management tool. The ability of Staff, 1975). Most reports on forestry 
remote sensing analysis to augment traditional applications of TM imagery have used Thematic 
forest resource evaluation procedures has been Mapper Simulator (TMS) data (Franklin, 1986; 
demonstrated by researchers for a diverse range of Vogelmann and Rock, 1986). In general, TM-type 
289 
———— 
Se 
>