oral 
ing 
(in 
the 
ra, 
lite 
ied 
' to 
ing 
of 
aps 
On 
jor 
ale 
of 
00 
ce 
ale 
hic 
he 
selection of the most suitable satellite data: 
e the crop calendar for the areas of interest 
e the availability and quality of satellite data. 
Additionally, in the control zone AITO, three LANDSAT 
TM images were used for the control of 300 declarations 
of the sample, that the declared plots were arable land 
during the period 1990 -1991. 
In Table 1 the technical characteristics of the used 
images are given. 
2.3 Ground Samples 
The ground samples were selected along road transacts. 
For the control zone AITO 26 transacts were defined, 
whereas for the control zone PYRG, 32 transacts. The 
length of the transacts was between 500 m and 2500 m. 
The main factor that were taken into account for the 
selection of the transacts was to include a large number 
of different crops, covering each one a big amount of area 
in order to eliminate the effects of "mixed-pixels" along 
their boundaries. The exact location of the transacts were 
defined by photo-interpretation of the images on the 
screen. 
The Ground Survey was executed within a time period 
suitable for the identification of all the crops of interest. 
Colour hardcopies in A4 format of the first geocoded 
SPOT-XS image of each zone, in scale 1: 10,000 were 
used for the registration of the crops' boundaries by the 
investigators. Additionally, were used B/W hardcopies in 
A4 format of the geocoded SPOT-P image in scale1: 
25,000 and maps in 1: 50,000 scale have been used for 
the orientation of the investigators and easy approach of 
the transacts. 
The samples of the Ground Survey were used not only for 
the definition of the crops’ appearance during the photo- 
interpretation procedure but also for the classification of 
the images. 
3. DATA PROSSECING 
3.1 Data Input - Digitisation and 
Localisation of the Plots 
The declarations were delivered in analogue form (lists) 
and the data of all declarations had been entered into the 
alphanumeric Data Base by the contractor. 
The main problems during data input, had to do with the 
deficiency of the cartographic reference, the declared 
area and the kind of crop. During the input of the data the 
errors and the deficiencies had been registered into lists. 
In agreement with MoA the half-finished declarations and 
the error lists were send back for their completion. 
For the digitisation of the plots the PC Arc/Info and the 
Autocad 13 were used. All the available consolidation 
maps of MoA were digitised in advance. Finally, the 
alphanumeric and the geometrical data base using the 
vector module of the ERDAS Imagine 8.2 System on 
SPARC stations, were integrated in a flexible GIS. 
739 
The localisation of the plots on the digitised maps or 
orthophotomaps has been executed with a unique code 
for every plot and with the name of the map and the 
community, that the plot belongs. 
3.2 Pre-processing of Satellite Images 
The pre-processing of the satellite data includes two 
types of corrections, the radiometric and the geometric. 
The method of the histogram shifting was applied for the 
radiometric corrections. It is based on the fact that the 
effects of the atmospheric Scattering can be somewhat 
minimised by shifting the histogram “to the left" (JENSEN 
J., 1986). 
Using the Topographic Maps of MoA in scale 1: 5,000 
and 1: 2,000 as reference, the SPOT-P images were 
rectified. The rest of the satellite data (LANDSAT-TM, 
SPOT-XS) was rectified by applying “image to image” 
registration. Thirty seven (37) to fifty six (56) Ground 
Control Points (GCPs) were used for every image. For the 
geometrical correction it was decided to apply a second 
order polynomial transformation. For the control of the 
geometrical correction accuracy, 14-18 Control Points 
were used (Control Points were different from the GCPs). 
The RMS errors of the rectification and the control 
accuracy for the satellite images are given in Tables 2 
and 3. 
All SPOT images were resampled to 10x10 m? pixel size 
using the Restoration module of ERDAS Imagine. The 
LANDSAT images were resampled to 10x10 m? pixel size 
by using the cubic convolution method. The Restoration 
algorithm take into account the height and the azimuth of 
the sun, the individual characteristics of the sensor and 
produces sharper, crisper rectified images by preserving 
and enhancing the high spatial frequency component of 
the image during the resampling process and improves 
the classification accuracy and the radiometric quality of 
the images (CHIESAC., TYLER W.,1994). 
For the optimum visual interpretation a contrast 
enhancement was applied with the method of linear 
stretching. 
4. CONTROL OF THE DECLARATIONS 
In order the declarations to be classified into “accepted”, 
"rejected" and "doubtful" where checked into three stages 
(EAGGF,1995) : 
e Parcel Level 
At this stage each plot is checked for its area and land 
use, against the satellite image displayed on the 
computer screen and a "control code" is given to each 
field depending on the conformity with the photo- 
interpretation. Table 4 indicates the various codes and 
the proposed guidelines. 
e Group Level 
At this stage the total declared area (Dg) within a 
group of similar type of crops, is checked against the 
one which derives from the interpretation (measured, 
Mg) of the satellite image by summing the area of the 
individual plots coded in the previous stage. Table 5 
indicates the rules for classified the crop groups. 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996