52
was performed to ensure that only pixels selected from within large homogeneous forest and
non-forest regions were used. Up to 500 forest and 500 non-forest pixels were sampled
randomly from each of the 82 forest regions.
For both maps the Digital Chart of the World database (DCW) (ESRI, 1993) was used to
define urban classes that could be represented as polygons at l:lMillion scale.
Validation of the classification results was performed by comparison with land cover
summary statistics defined by the Statistical Office of the European Union (EUROSTAT).
The Nomenclature of Territorial Units for Statistics (NUTS) was established to provide a
uniform breakdown of territorial units for the production of regional statistics (EUROSTAT,
1995). The NUTS regions use a common land use nomenclature and are hierarchically
defined at different scales based on the institutional divisions in force in the Member States.
METHODOLOGY
The monthly maximum value NDVI and Ts data were classified in each region independently
in an attempt to reduce regional climatic and biophysical variations in the NDVI and Ts data.
Regional processing was also performed to facilitate automation of the classification
procedure. Figure 1 shows a flow diagram of the generic classification methodology used to
produce both the land cover and forest cover maps. The left hand side of the flow diagram
illustrates the supervised classification procedure used to produce the forest cover map and the
right hand side illustrates the unsupervised classification procedure used to produce the land
cover map. The land cover map was composed of 13 ecosystem regions and the forest cover
map was composed of 82 homogeneous forest regions.
The supervised procedure involved the maximum likelihood classification of an optimal
multitemporal combination of NDVI and Ts data into forest and non-forest classes. Maximum
likelihood classification rules (James, 1985) were defined independently for each region using
the forest and non-forest training data held in the georeferenced database. The classification
rule was then applied to the remaining pixels in the forest region (Kennedy et al., 1995, Roy et