421 
distances . Our preliminary tests have shown that this method works quite well. 
The estimation of some changes such as thinning may require other methods, cf. Thomas 
(1990). The spatial information of the image can be taken into account in the feature 
choice and/or in the postprocessing. Segmentation techniques or Gibbsian random field 
modelling, for example, are possible postprocessing methods (see Besag (1986) and 
Tomppo (1987,1989)). Bias and the accuracy of the estimates must of course be kept 
under control. 
4. DATA MANAGEMENT 
Special attention must be paid to data management; total single date coverage of 
the country with TM imagery involves about five Gbytes of data. The output of the 
image analysis will be compatible with the the forest management planning system 
(MELA) used by the Finnish Forest Research Institute. This system allows simulation 
of the development of the forest between two ground measurement (and image analysis) 
timepoints. In addition, themes can be presented in a raster form and the maps can be 
produced at different scales. 
The system allows the use of digital map data of other organizations, for example 
the National Board of Survey and the National Board of Forestry. In the future, the 
output will be formatted to be suitable for the databases of those organizations. 
5. PRELIMINARY RESULTS 
The above methodology and input data have been applied in estimating NFI - 
variables in a few areas in eastern Finland. Sum characteristics of ordinary stand wise 
data, measured for forest management planning purposes, are available as comparison 
material. These data (referred here as FBI) -data) are based on visual ground estima 
tion and information from false colour aerial photographs. The data are measured by 
the local Forestry Board District and only non-company private forests are included. 
All the above mentioned digital map data are so far available only from the area of two 
communes, Tohmajarvi and Vartsila, the total area of non-company private land being 
52 000 hectares. 
The digital communal boundaries were applied in order to restrict the test area. 1 he 
boundaries of lands of two forest companies were digitized in order to remove non-private 
forest areas from the test site, because the forest characteristics of company-ovned land 
were not known. 
Table 1 shows the satellite image-based estimates ( NFI) and the estimates of Forestry 
Board District (FBD) for some mean characteristics. (Note that the total number of 
NFI -variables is about 200.)