'hniques
ies. One of
effectively
ome duties
one of the
;hio, grape,
ws spatial
ters in this
ind applied
"eld Crops
inated field
round truth
ilture (field
been given
an not be
n that the
gradually.
naintain its
nts of the
of current
er planning
f statistical
y has been
ral Affairs.
ng to the
1e cadastral
differences
te sensing
; including
sed on the
surface by
|! computer
about the
tained and
sector has
> of remote
sthods can
areas and
A griculture
1er sectors.
International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
Reliable information is one of the most important components
for efficient and effective management.
2. Materials and Methods
2.1. Materials
This project has been implemented by Geographical
Information Systems and Remote Sensing Department of
Central Research Institute of Field Crops (CRIFC) in General
Directorate of Agricultural Research. The study conducted
between the years 2000 and 2001. In the realization of project,
there has been administrative and financial supports of local
government authorities.
Pentium IM work station and Unix based HP series computers
have been used as computer hardware. Erdas Imagine and
Arc/Info software were employed to process the images and to
digitize the topographical maps, respectively. For the district
based studies of project, a database with 1:25000 scale was
chosen to comprise the city and district boundaries and village
centers of whole country. For this aim, topographical maps with
1:25 000 scale were obtained from the General Commandership
of Mapping of Turkish Army. Then, digital elevation model
was developed by digitizing these topographical maps.
Satellite images are the main materials of the project. Multi
band LANDSAT-7 ETM and single band IRS images covering
the whole city with high resolution have been purchased. For
implementing the project, satellite images, field data, digital
cartographic maps with 1/25 000 scale including province and
county borders, and statistical data were used. Parts of two
LANDSAT-7 ETM scenes are needed to provide complete
coverage for the study area. For this aim, full two scenes having
path-row of 174-34 and 173-34 were chosen (Figure 1). It is
expected that the phenological differences of a specific plant
would give the advantages in determining the pistachio areas by
means of remote sensing imagery (Campbell, 1987). To select
suitable image dates where most changes occur, a phenological
calendar was prepared with regard to knowledge and
experience of local people and government persons. Based on
this calendar and cloudiness, the scene dates were determined
as 10" of September-10" of August for fall season and 26^ —
27^ March for spring season in the year 2000. Good quality and
cloud free images that contain the both of these two periods
were purchased. The satellite data had been already rectified to
UTM coordinate system with WGS 84 datum.
Collection of field data was performed from mid September to
the end of October in 2000, and GPS devices were used to
collect ground truth data in the field. Field data was planned to
be used for
- final reprojection
- image analysis and classification
- accuracy assessment
Collecting reference data is performed not only for pistachio
areas but also other land cover classes such as; crop, pasture,
urban, bare lands etc. Since the study area was large, county-
based sampling is done by taking into consideration.
The following criteria were considered;
- selection of sample sites was based on
accessibility (closeness to the road network)
- intensity of pistachio farming in the county,
161
- mostly point sampling, sometimes polygons
were created by visiting all of the corners of the
field
- representation of other major classes of each
county
Total 3440 GPS records were taken from the study
area. Approximately, 30% of total records was set aside for
accuracy assessment. Remaining GPS data were used to select
training samples in classification process. All GPS data were
downloaded to the PC and exported to the GIS format (Arc/Info
file). As for cartographic data, digitized administrative
boundaries and village locations in county level of Gaziantep
province were used. This data had been digitized from 1/25 000
scale base maps of Turkey and provided as Arc/Info vector
coverage file. This is only digital vector layer we have in our
data at this scale level. It is quite accurate for national and
regional but not as accurate as in provincial and county level
(Figure 2).
Agricultural statistics for the main crops grown in the study
area were taken from the provincial directorate of agriculture.
In basic, these agricultural statistics had been collected by
direct contact with farmers and field survey accomplished by
government department for the year 1999.
2.2 Method
The method employed for this project was basically dependent
upon the supervised classification of image data by using GPS
data and then testing of classification performance.
LANDSAT7-ETM image for fall season were base data used
for classification. Two different date scenes were mosaiced to
output image from which final data covering the study area was
obtained. This data was produced by subsetting the mosaiced
image with Arc/Info coverage of province borders (Figure 1).
Figure 1. Study area.
Even though the province image is in the UTM coordinate
system, collected field data (in UTM projection system) did not
match with the GCPs over image. Average shifting was 240 m
for x, and 180 m for y coordinates. The image was reprojected
one more time using GPS coordinates (Cook and Pinder, 1996),
and resampled to 30 meter resolution with affine coordinate
transformation (Verbyla, 1995). In the final reprojection, total
