4 
Landunits on these scale levels can be recognized and 
delineated easily by visual interpretation of small 
scale imanery, such as those produced by satellite 
remote sensing (V.Ackerson & E.Fish, 1980 ; C.M. 
Gerards & M.C.Girard, 1973, 1975, 1985 ; H.Antrop & 
L.Gaels, 1977 ; M.Antrop, 1982). 
Landelements and -facets can be recognized indivi 
dually upon a satellite image when their size, shape 
ant orientation can be resolved and when their spec 
tral signature at a given time differs sufficiently 
from the one of the adjacent features. When this is 
not the case, the landscape structure to which they 
belong, will be distorted and may not be recognized. 
Landscape ecology and -planning often group the 
landscape elements and -facets according to two pro 
perties : the biotic significance and the importance 
of their spatial dimensions, i.e. the height and 
the size of the objects. Thus five groups are 
defined : biotic and abiotic volumes, biotic and 
abiotic spaces and biotic screens. Examples of each 
group are : forests, built-up areas, agricultural 
land, water- and barren surfaces, hedges and tree- 
rows. These groups are fairly independent from the 
phenology and remain constant throughout the year. 
They also reflect the typological composition of 
the landscape as well for the content (genotypical 
composition) as for the physiognomic appearance or 
the "scenery" (phenotypical composition). For the 
image interpretation these groups are also signifi 
cant. Volumes and screens are shadow giving objects. 
Biotic volumes occupy mostly vast areas and thus 
are registered with a large proportion of pure 
pixels. Biotic screens are seldom resolved in the 
image but cause a lot of interference with the spec 
tral reflectance of the adjacent fields. Biotic 
space constists of the vast areas of agricultural 
land containing complex structures of field patterns 
and a varying diversity of land uses. Both this 
group and the one formed by the abiotic volumes give 
large proportions of mixed pixels. Although these 
groups are not completely significant for land use 
interpretation (all agriculture in one group), they 
have the advantage of being fairly constant in time 
(no seasonal variation) and being good indicators 
for the landscape structure. 
Even with a well differentiated spectral signature, 
the possibilities of identification of each object 
category and of inventoring its areal distribution 
on a remote sensed image depend upon three additio 
nal factors : 
1. - the size of the objects vs. the pixel size ; 
2. - its shape (compactness and orientation) vs. 
the pixel shape ; 
3. - its areal proportion in the scene. 
The first two factors determine the proportion 
between the pure and mixed pixels for each object. 
The probability of having pure pixels in a given 
category increases with the object size, the com 
pactness of its shape and the proportion of the 
area it covers. These three parameters may vary a 
lot in the geographical space. 
In visual image interpretation, a remote sensed 
image is to be considered as an holistic feature 
showing more or less structured image primitives. 
These are made of aggregates of adjacent pixels 
having the same photographical density. In analogy 
with the airphoto-interpretation these primitives 
form the texture of the image and may be called 
texels. Their significance is fundamentally diffe 
rent from the concept of a pixel and from the con 
cept of texture in digital image processing. They 
are formed by a real spatial structure and are not 
mathematical constructions between pixels. 
The size, shape orientation of the texels as well as 
the patterns they may form, contain distorted infor 
mation about the landscape structures. 
A pixel should be considered as a discrete sample 
out of the landscape continuum. No classification 
problems with the pure pixels as far as the diffe 
rent categories have distinct image properties. 
Figure 1. Hierarchical relatioship between land 
scape structure, structure of the image, texels and 
pixels. Landscape elements : A = hedge- & tree- 
rows ; B = woodland ; C = fields ; M = buildings ; 
D = wasteland. Components of the landscape : 
BM = biotic volumes ; AM = abiotic volume ; BR = 
biotic space ; AR = abiotic space ; S = biotic 
screens. Texels formed by adjacent pixels having 
the same DN. 
Texture formed by a spatial pattern of texels. 
Micro-structure formed by a spatial pattern of 
textures. Macro-structure of the image correspond 
to the main landscape types : P = polderland ; 
H = Houtland ; M = Meetjesland ; G = region of 
Ghent ; L = Land van Lokeren ; W = Land van Waas. 
On the contrary, mixed pi els do not contain 
any real information about the landscape. Their 
classification can be improved only by adding con 
textual information. This is achieved automatially 
in the deductive procedure followd in visual inter 
pretation, using textural and structural informa 
tion from all over the scene. 
There is an space-dependent hierarchival rela 
tion between pixels, texels, textures and struc 
tures in the image and in the landscape. 
Fig. 1 illustrates this. 
3. THE METHOD 
The integration of structural information about the 
landscape with the other ground truth could be 
achieved in the following scheme : 
1. - make a typological classification in the 
area of study, combining classical methods 
of landclassification and geographical lands 
cape analyss. Attributes significant for 
the structural properties of the landscape 
should be used ; 
2. - sample the remote sensed image for the 
landscape units obtained using mainly tex 
tural parameters. Estimate the proportion 
of the pure pixels for the different land 
scape components ; 
3. - use the 
aical c 
tain ur 
strata 
categor 
4. - imp!erne 
procedu 
General pur 
been used sin 
1980 ; Webste 
The/are based 
medium to sma 
Stereoscopic : 
from SPOT, wi' 
Photomorphic 
of a resoluti( 
in a classific 
of at least a 
Girard & C.M.C 
1980). 
Geography of 
fication of sl 
mann, 1973). 
Landscape an 
cribe and meas 
pe. Important 
directional an 
The proporti 
rain feature c 
a simulation p 
et.al., 1984 ; 
groups describi 
constancy. 
A topologica 
butes measured 
mi lari ties bet\ 
cluster analys 
Euclidian distc 
Finally, the 
to terrain knov 
Landscape eleme 
portant changes 
should be used 
the occuring ge 
similarity can 
be defined whic 
to the pixel si 
4. AN APPLICATI 
(BELGIUM) 
4.1.The general 
Although a smal 
diversity of bo 
The long histor; 
man, is characti 
-different soc 
different mant 
different natt 
-specific and c 
environment, v 
of the soil cc 
-drastic change 
Revolution anc 
building, indi 
The high popul 
the average rura 
sq..km, a motorw 
railway density 
differences in 1 
administrative r 
from general sta 
In the regions 
21 % and 77 % of 
U P- Linear const 
ways, occupy aboi 
The utmost split 
obvious.