ship between land- 
the image, texels and 
= hedge- & tree- 
ds ; M = buildings ; 
he landscape : 
otic volume ; BR = 
ace ; S = biotic 
cent pixels having 
ttern of texels. 
atial pattern of 
he image correspond 
P = polderland ; 
; G = region of 
W = Land van Waas. 
do not contain 
landscape. Their 
only by adding con- 
chieved automatially 
lowd in visual inter- 
structural informa- 
hierarchival rela- 
extures and struc- 
landscape. 
information about the 
d truth could be 
me : 
ification in the 
classical methods 
d geographical lands- 
significant for 
s of the landscape 
image for the 
using mainly tex- 
te the proportion 
he different land- 
3. - use these image attributes in a further typolo 
gical classification of the landscapes to ob 
tain units which are potential homogeneous 
strata for a statistical sampling of landuse 
categories using image classification ; 
4. - implement these strata in the classification 
procedure. 
General purpose landclassification methods have 
been used since the 1960's (Howard J. & Mitchell C., 
1980 ; Webster R. & Beckett P., 1970, Antrop M.,1983) 
The/are based upon a stereoscopic interpretation of 
medium to small scale air photographs. 
Stereoscopic satellite imagery as the one available 
from SPOT, will allow the use of the same methods. 
Photomorphic image classification on orbital imagery 
of a resolution level of Landsat MSS and TM, results 
in a classification of land units on scale levels 
of at least a landregion (J.Townsend, 1981 ; M.C. 
Pirard & C.M.Girard, 1985 ; V.Ackerson & E.Fish, 
1980). 
Geography offers techniques for a regional classi 
fication of such categorical landunits (A.Kilchen- 
mann, 1973). 
Landscape analysis offers a set of methods to des 
cribe and measure spatial structures in the landsca 
pe. Important ones are network analysis, shape and 
directional analysis. 
The proportion of pure pixels for any kind of ter 
rain feature can be based upon an estimation using 
a simulation pixel (M.Vauzelle, 1982 ; V.Aurelio, 
et.al., 1984 ; M.Antrop, 1985). The use of the 5 
groups described is suggested because of their 
constancy. 
A topological classification, based upon the attri 
butes measured, can be achieved by determining si 
milarities between the sample sites followed by a 
cluster analysis. In this case a mean squared 
Euclidian distance was used as similarity index. 
Finally, the clusters obtained should be related 
to terrain knowledge of the occuring landscapes. 
Landscape elements and components which cause im 
portant changes in the proportion of pure pixels, 
should be used for a chorological classification of 
the occuring geographical regions. Across-boundary 
similarity can be used for this and thus zones may 
be defined which have a constant structure according 
to the pixel size used. 
Although a small country, Belgium possesses great 
diversity of both natural and cultural landscapes. 
The long history of the occupation of the land by 
man, is characterized by : 
-different social groups and cultures acting in a 
different manner and different periods for the 
different natural regions ; 
-specific and drastic adaptations of the natural 
environment, which are also reflected in changes 
of the soil conditions ; 
-drastic changes occured since the early Industrial 
Revolution and are still going on today (house 
building, industrial land use, re-allotment). 
The high population pressure is illustrated by 
the average rural population density of 313 inhab./ 
sq. km, a motorway density of 13 km/100 sq.km, a 
railway density of 14 km/100 sq.km. The spatial 
differences in land use patterns between the major 
administrative regions of Belgium is evident even 
from general statistics (table 1.) 
In the regions of Flanders and Brussels resp. 
21 % and 77 % of the areas are more or less built- 
up. Linear constructions such as roads and water 
ways, occupy about 7 % and 20 % resp. of the area. 
The utmost splitted character of the land use is 
obvious. 
Table 1. Cadastral landuse in Belgium, the region 
of Flanders and Brussels (1980). Proportion (%) of 
the area occupied by the main categories (after 
H.Van der Haegen, 1982). 
Flanders 
Brussels 
Belgium 
agriculture 
67 
11 
61 
woodland, heathland 
12 
12 
23 
rurban 
4 
19 
3 
built-up areas 
10 
38 
7 
roads, waterways 
7 
20 
6 
Total area in ha 
1351143 
16178 
3051871 
About 12 % of the agricultural land has been refor 
med by re-allotment, mainly characterized by more 
open landscapes and larger fields. Traditional 
landscapes, which are given a large historical and 
ecological value, still occur more less well con- 
servated over an area of about 25 %. 
Although a great diversity in landscape typs exists, 
it is not possible to distinguish them on Landsat 
MMS image, not even by greatly deductive visual 
interpretation (I .Vandecasteele, 1979 ; M.Antrop, 
1985). In most cases there is no systematic rela 
tionship between the image structure formed by the 
texels and the landscape structures for the whole 
region of Flanders. Only macrostructures can be deli 
neated easily by photomorphicimage classification. 
They are formed by the settlement pattern at the 
scale level of villages and small towns, the pat 
terns formed by vast landelements (waterbodies and 
woods with an area of at least 1 sqkm), the main 
geomorphological units (valleys, polders, cuesta's, 
dunes) and the fieldpattern at the scale level of 
complexes. The orientation of the latter are espe 
cially important in the recognition process for 
regions with large fields bordered by ditches and 
tree- or hedgerows. 
4.2.The analysis based upon a simulated pixel samp 
ling 
A first analysis was based upon a landscape sampling 
using a simulated pixel upon orthophotographs on a 
scale of 1/10.000, according to the method already 
described. Sample sites have not been selected on 
a regional basis but on their structural type, 
especially when the structure is fine and complex 
(fig.2). Consequentely most sites are taken from 
the sandy region of Flanders between Ghent and 
Brussels, where the rural landscapehas been affected 
strongly by recent unorganized settlement, and 
which is also called now a 'rurban' landscape. 
The sampling used a simulated TM pixel size of 
30mx30m. The results are given in tables 2 and 3. 
The great diversity, complexity and the utmost 
fine structure of the different landscapes is evi 
dent even when only using the rather general cate 
gories of biotic volumes (BM, i.e. mainly woods), 
abiotic volumes (AM, i.e. mainly buildings), biotic 
space (BR, i.e. all the agricultural land and 
heathland, abiotic space (AR, i.e. water surfaces 
and open (industrial) wasteland and biotic screen 
(S, i.e. tree- and hedgerows). 
The number of landscape elements/sq.km, is extre 
mely high (table 2) and varies between about 1700 
(openfield and re-allotment landscapes) to about 
3500 (traditional enclosed landscapes with small 
fields and regions with open settlement of a high 
density). The proportionof estimated pure pixels 
(size 30mx30m) is accordingly very low : about 
50 %-80 % in the large scale open landscapes to 
only about 12 %-25 % in the fine structured enclosed 
and rurban landscapes. 
4. AN APPLICATION UPON THE REGION OF FLANDERS 
(BELGIUM) 
4.1.The general setting of the landscape 
5