ory in 
spatial 
ws an 
ng the 
ch an 
cover 
e first 
ng the 
image 
ied. in 
crop 
cover 
28). It 
in the 
es in 
ige of 
  
  
y the 
T and 
ithout 
Another multispectral classification experiment was carried 
out using SPOT images of 1996 for the Kursk test site. 
Data processing was carried out in the frame of the TACIS 
project "Development of System for Crop Assessment and 
Monitoring in Russia" together with the French companies 
Scot Conseil and Sotema. Figure 5 shows the SPOT image 
of July 7, 1996. The SPOT image was processed using a 
supervised classification into 32 classes. The subsequent 
expert analysis offered the possibility to merge classes 
within field boundaries. The result of such classifications 
are given in Table 3. A similar procedure was used for 
every SPOT image obtained during the vegetative growing 
season. Cross-analysis of the images allows correction of 
the results of single image classifications (Table 4), 
improving the accuracy of the classification. 
The combined method based on using multispectral SPOT 
information together with TK-350 data was also tested for 
the Kursk test site. The first step was a rectification of the 
SPOT image with lower spatial resolution (20 m) into the 
TK-350 image with a spatial resolution of 10 m. It was 
performed using a set of 35 ground control points in both 
images, resulting into accurate matching (Figure 6). 
Good results were also obtained using the panchromatic 
TK-350 image of high resolution for primary stratification 
of the test territory and then classification of the 
multispectral scanner image separately for each of the strata 
with more uniform natural conditions than the whole image. 
Table 3. Classification results for the Kursk test site (SPOT 
image 07.07.1996). 
6. CONCLUSIONS 
e  NOAA-AVHRR data can be used at continental scale to 
specify agricultural regions. 
.*  MSU-E data are complementary to SPOT-XS data. 
e Anintermediate resolution sensor (e.g. MSU-SK) can 
be very valuable for further differentiation of the 
agricultural regions. 
e Ahighresolution sensor (e.g. TK-350 or KVR-1000) 
can be very valuable for monitoring detailed local 
phenomena. 
Table 4. The results of cross-classification of the SPOT 
images for the Kursk test site (SPOT images 20.06.1996 
and 07.07.1996). 
  
  
  
  
class area area 
(ha) (%) 
fallow land 25689.3 16.05 
cereals 20351.4 12.72 
winter cereals 11830.3 7.39 
winter wheat (in good condition) 7.0 0.00 
spring cereals 9334.4 5.83 
spring barley 11812.7 7.38 
harvested fields 10313.5 6.44 
buck wheat 18544.2 11.59 
oats 3140.5 1.96 
natural vegetation + perm grass 20541.2 12.84 
woods 18544.2 11.59 
urban areas 14493.6 9.06 
water 1209.4 7.55 
clouds 508.9 0.32 
total 160,000 100.00 
  
  
  
class area area 
(ha) (96) 
fallow land 9049.4 5.65 
winter cereals 18588.6 11.62 
(wheat + barley + others) 
winter wheat + buck wheat + 25712.8 16.07 
harvested fields 
winter wheat 714.6 0.45 
(in good condition) 
buck wheat 19579.6 12.24 
barley + fields with weeds 7840.9 4.90 
spring barley 2193552 13.72 
summer crops + winter rye 5315.2 3.45 
oats 8036.2 5.02 
natural vegetation + permanent 335199 20.95 
grassland 
woods 9486.7 5.93 
total 160,000 100.00 
  
  
Figure 5. SPOT-XS image of 7 July 1997 for the Kursk test 
site (only a black-and-white copy is presented here). 
Intemational Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 101