)04 
'ell 
ter 
ng, 
ms. 
and 
of 
e 10 
lion 
ical 
1an- 
tion 
s in 
rsity 
PARCEL-BASED CROP MAPPING THROUGH MULTI-TEMPORAL MASKING 
CLASSIFICATION OF LANDSAT 7 IMAGES IN KARACABEY, TURKEY. 
M. Arikan 
METU, Graduate School of Natural and Applied Sciences, GGIT, 06531 Ankara Turkey — marikan@usa.net 
KEY WORDS: Remote Sensing, Agriculture. Classification, Integration, GIS, Landsat, Multitemporal 
ABSTRACT: 
This study describes the parcel-based classification of agricultural crops using multi-date Landsat 7 ETM+ images acquired in May, 
July, and August 2000. The study area is located in North-West of Turkey with an area of about 170 km? and 
crops. The objective was to map the summer (August) crops w 
based on a multi-temporal masking of Landsat 7 ETM+ images. First, a supervised per-pixel classification of the three im 
July, and August 2000) was performed using a maximum likelihood classifier algorithm. The 
computed by comparing them with the ground truth information. Those classes th 
the August image was re-classified using the unmasked classes only, 
masking technique was applied to overcome the problems caused by th 
completing the classification process, the multi-temporal classified out 
grows a variety of 
ithin the agricultural land parcels. The classification methodology is 
ages (May, 
accuracy of classified outputs was 
at meet the threshold values were masked out and 
excluding the masked fields from the classification. The 
e spectral overlaps between the information classes. After 
put of the August image was analyzed in a field specific 
manner in the integration of remote sensing and geographic information system (GIS). In each parcel, the percentages of classified 
pixels were computed and the modal class label was assigned to the parcel. The analysis results were fed back to a GIS database for 
immediate update. The resulting classification accuracy of the multi-temporal masking technique was 81%, which was 10% more 
accurate than the classification of the August image only. 
INTRODUCTION 
Automated image classification is one of the most widely used 
techniques to extract thematic information from remotely 
sensed data of the earth. The thematic information is 
increasingly being used in varies fields, such as in agriculture to 
monitor and estimate crop development and its spatial 
distribution. Thus, up-to-date and accurate classification results 
are required for analyses which provide basis for deciding and 
implementing policies and plans for management of agricultural 
crops in local, regional and global scale. 
Classification can be subdivided into two methodologies pixel- 
based and parcel-based classification. Unfortunately, the land 
cover classes do not have uniform spectral response due to 
atmospheric effects, noise, heterogeneity within the land cover 
lypes, and mixed pixels present at the boundaries (Ioka and 
Koda, 1986; Janssen ct al., 1990; Lunetta et al., 1991). Thus, 
resulting classification often includes misclassified land cover 
types. In order to overcome this problem and increase the 
reliability of classification accuracy for land cover 
identification in agricultural lands, the classification can be 
performed using a parcel-based approach. 
In parcel-based classification, the remotely sensed imagery is 
integrated with vector parcel boundaries that explicitly provide 
the spatial context of agricultural parcels. By including the 
Spatial context in the classification, individual parcels are 
classified instead of single pixels. Thus, the uncertainty caused 
by per-pixel classification is eliminated and the classification 
accuracy can be increased up to a certain extent (Janssen et. al., 
1990; Johnson, 1994;. Aplin et. al, 1999). Parcel-based 
Classification techniques are applicable through integration of 
[mote sensing and geographic information systems which 
Provide integrated analysis for raster imagery, vector graphics 
and attributes. An improvement in the classification accuracy 
can be achieved by extending the classification method by 
means of incorporating expert knowledge and ancillary data 
(Mason et al., 1988: Middelkoop and Janssen, 1991). In the 
absence of spatial context information, multi-temporal 
classification together with decision rules yields acceptable 
accuracies. Such the studies were performed by Beltran and 
Belmote (2001), and Lanjeri et al (2001). 
This study introduces a methodology for integrating remote 
sensing and gcographic information systems to accurately 
classify agricultural crops. Parcel-based image classification 
with expert knowledge is carried out to better discriminate 
agricultural land cover classes. The selected study area is an 
agricultural land located in Karacabey, Bursa. 
STUDY AREA AND DATA 
Study Area 
The selected study area is situated in Marmara region, near 
Karacabey, Bursa (Figure 1). The region is a level plain (within 
10 m) agricultural land, which is one of the most important 
agricultural areas in Turkey. The area covers an agricultural 
land of approximately 170 km? of nine villages namely: 
Hotanli, Kucukkaraagac, Yolagazi. Sultaniye, Eskisaribey, 
Ortasaribey, Yenisaribey, Akhisar, and Ismetpasa fall within the 
study site. 
The area is characterized by rich, loamy soils which, in addition 
to the excellent weather conditions, make agriculture the main 
land use in the region. The agricultural land is predominantly 
used for the cultivation of arable crops that are wheat, corn, 
tomato, sugar beet, rice, pepper, and watermelon as well as 
other crops of secondary importance that are pea, onion and 
1085