International Archives of the Photogrammetry, Remote 
Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004 
  
  
  
cauliflower. In addition, several pasture and clover fields and 
small townships are scattered throughout the area. 
40*00' 
  
  
30*00 
  
  
  
  
Cu 
e. 
Study Area Bursa 
    
Figure 1. Location of the study area, Karacabey, Turkey. 
  
500 M 
«Z 
© 
> 
v 
2 
= 
9 
N 
c 
= 
wv 
9 
= 
o 
zx 
— 
a"E 
# 
20'28"E 
  
28°20"2 
28° 
28°10'27"E 
  
  
  
  
  
  
  
Figure 2. A color composite of Landsat 7 ETM+ Bands 4, 5, 
o 
and 3 (RGB) overlaid with the agricultural parcel boundaries. 
Data Description 
As remotely sensed imagery, Landsat 7 ETM+ data was 
selected due its wider spectral coverage and availability of a 
high resolution band (ETM Band 8) for enhancing spatial 
resolution and features. Landsat 7 ETM+ data having Path: 180 
and Row: 32 scenes were acquired on three dates: May 15, July 
02, and August 19 2000. These scenes were found to be the 
suitable dates for monitoring vegetation development in the 
study area. From each scene, six visible infrared bands (Band 1- 
5 & 7) having 30 m resolution and one panchromatic band 
(Band 8) having 15 m resolution were used for the analyses. All 
three scenes were cloud free and of good quality. Since full 
scenes were not required for the achievement of this study, the 
subsets of the images were extracted to match approximately 
the same coverage with the vector parcel data (Figure 2). 
The ground reference data for the study area were collected 
through communicating with the farmers and making an 
accurate land cover survey in mid August 2000. Particular 
attention was paid to the selection of samples representative of 
the extent and distribution of the land cover categories in the 
study area. For each agricultural parcel, the current crop 
situation together with the previous crop grown (where valid) 
was recorded on the reference land cover map. Final ground 
reference data of the study was generated through combining 
information collected from farmers and the land cover survey in 
the field. The reference data were subsequently used for 
training and validating the classified outputs. In addition, the 
knowledge obtained about the relationships between the 
agricultural land cover classes and the agricultural parcel 
boundaries were utilized (i.e. sugar beet restricted zones) to 
improve classification accuracy. 
The cadastral maps of the study area were obtained as hard 
copy form from the cadastral office of Mustafa Kemal Pasa. 
The area covered by eleven 1:5,000-scale and one 1 :7,500-scale 
cadastral maps. 
ANALYSIS 
Preliminary Data Processing 
The preliminary data processing steps consist of digitizing 
agricultural parcel boundaries from 1:5,000 to 1:7,500-scale 
map sheets and relating the attribute data to existing database: 
and geometric correction following data fusion of the satellite 
images. 
Preparation of Vector Parcel Data: The cadastral maps were 
subsequently scanned and registered using available control 
points on the maps. The control points collected in a national 
datum were georefenced to UTM (Zone 35) projection using 
nearest neighbor resampling method. The maximum produced 
RMSE value of 3.60 m was found to be quite satisfactory, since 
it was less than a half pixel size of Landsat 7 ETM- imagery. 
Next, each raster map was digitized using a polygonal topology. 
When digitizing of raster maps were completed, edge matching 
was performed to obtain a continuous data layer of agricultural 
parcels (Figure 2). Together with agricultural parcel boundaries, 
the supplementary data that came from the cadastral map and 
the site visit were stored in a geographical information system. 
The database held parcel information in conjunction with the 
available land cover information in each parcel for further 
analyses. 
Data fusion: For each subscene, 30-m resolution Landsat 7 
ETM+ multispectral bands 1 to 5 and 7 were fused with the 15 
m resolution panchromatic band. The resulting fused bands 
were spatially enhanced while keeping the spectral 
characteristics close to the original multispectral bands. The 
algorithm used has been developed by Atlantic Aerospace. Its 
based on local correlation of edges to improve edges in loe 
resolution bands wherever a corresponding edge is found in the 
panchromatic band (Cheng et al., 2000). This is unlike the 
1086 
Internati 
Te ett 
popular 
spectral 
1999). 
Geomati 
2003). T 
subseque 
that the £ 
in case © 
the satell 
Geometr 
with the 
register t 
this pur 
Landsat 
layer. G 
ground « 
polynom 
method. | 
0.5 pixel: 
Parcel b: 
The pring 
based cla 
sensing a 
within th 
of agricu. 
the Land 
kept in a 
boundary 
using a bi 
spectral n 
misclassif 
the agricu 
For each : 
componer 
bands 4, 
monitorin 
The imag 
through q 
type were 
known la 
training v 
centre pai 
procedure. 
they woul 
number of 
amount ar 
throughou 
The Aug 
informatio 
Sugar bee 
Watermelo 
(Ric) and 
through e: 
classes | 
characteris 
Such as th 
The classi 
subclass 
Kiefer, 19«