890 
The stratification of the whole area was done by using soils map 
and a color composite of the Landsat MSS data. County and strata 
boundaries were digitized and stored in the RDS data base. Digital 
field boundary map (DFBM) of 35 farms out of 100 was also created 
previously and stored into RDS. After linking this geocoded DFBM 
to the coverage codes retrieved from a non—geographic database a 
digital reference map resulted (Fig. 3) in the GDMS data base. 
This provided for selecting training and test fields to each of 
the thematic classes, e.g. the major crops, water bodies, forest 
etc. 
From this stage on all the analysis steps - training with 
clustering, spectral—user's classes correlation, spectral classes 
configuration assessment, the per point classification and error 
checking — were done using Landsat data restricted to an area that 
had been defined by a mask derived from GDMS data. The resulted 
crop map and the derived crop area estimates were close to the 
figures of the Hungarian Central Statistical Office which was 
responsible for providing statistics and used traditional ground 
collection plus area sampling (Table 2). The cooperation between 
the IAS and GIMS was low level and did not completely exploit 
potential. 
Categories 
Statistical 
Office 
(Ha) 
FöMI RSC 
(Ha) 
i! 
Difference 
, 11 
il 
Winter cereals 
111.124 
114.290 
+2.8 
II 
Maize 
117.372 
114.218 
-2.7 
h 
Alfalfa + pasture 
168.852 
176.860 
II 
+4.5 
H 
Total 
397.384 
405.363 
II 
+ vJ . 
»I 
II 
II 
II 
II 
II 
II 
Table 2.Comparison of area estimates of Hungarian Central 
Statistical Office and FOMI Remote Sensing Centre using field 
based area sampling and bitemporal Landsat MSS data respectively. 
Advanced crop survey methods strongly relying on GIS/GIMS 
In general it is worthwhile to involve as many of the relevant 
available ¿i priori information into the thematic classification 
process as possible at a low cost- These lead us to a more 
reliable classification method that results in high accuracies 
with high confidence values. Up to now that particular fact has 
been exploited that the observed change rate in cooperative and 
state farm field boundaries was low, expected to be about 2—ZV. 
annually. In addition, boundary change detection method based on 
the DFBM and actual Landsat TM or SPOT data has been devised to 
update the DFBM. One of the hardest points of the task is the