3 
Symposium on Remote Sensing for Resources Development and Environmental Management / Enschede / August 1986 
Structural information of the landscape as ground truth 
for the interpretation of satellite imagery 
M.Antrop 
State University of Ghent, Belgium 
ABSTRACT : The possibilities for a detailed and accurate land use interpretation using satellite imagery, 
depend for a great part upon the landscape structure. The selection of the training sites for image classifi 
cation should follow a stratified scheme based upon a landscape classification which uses structural indica 
tors. An analysis is made of the landscapes of Flanders based upon such indicators. An estimation of the 
occurence of pure pixels (TM size) for different landscape components and types is given, as well as the 
compactness of the field shapes and their orientation in relation to the pixel size and scan direction. 
RESUME : Les possibilités pour une interprétation détaillée et accurate de l'utilisation du sol sur des 
images satellites, dépendent en grande partie de la structure du paysage même. Les sites pour l'élaboration 
des clefs d'interprétation devrait être sélectionnés à partir d'une classification du paysage, basée sur 
des indicateurs structurels. Un exemple d'une pareille analyse est donné pour les paysages de Flandre. 
Un estimation est faite de la nroportion des pixels purs (TM) pour différentes composantes du paysage, ainsi 
qu'une analyse de la forme et de l'orientation des champs en relations avec la taille du pixel et la direction 
du balavage. 
1. INTRODUCTION 
It is a basic knowledge in classical airphoto-inter 
pretation that an interpretation key has a restricted 
validity. It remains valid and guarantees a tested 
accuracy for a given application only for one set 
of photographs (with a constant scale, emulsion and 
season) and for one reoion which can be considered 
as homogenous with resnect to geographical and eco 
logical relations. Large study areas show a great 
environmental diversity and there it becomes neces 
sary to differentiate the interpretation key on the 
basis of the geographical structure of the terrain 
observed. A first and fast assessment of this can be 
achieved by a photomorphic image analysis, which ma 
kes a holistic approach of the landscape information 
contained in the image. In fact, it can be conside 
red as a natural way of an unsupervised visual image 
classification. 
The influence of the geographical diversity of the 
environment on the interpretation of an image beco 
mes even more important for orbital remote sensing 
because very vast regions are observed. Consequente- 
ly, the decree of detail and the accuracy by which 
thematic information can be extracted from satellite 
imagery - and thus the economic rentability -, 
does not depend only upon the characteristics of 
the remote system used. From the technical point of 
view, spatial, spectral and temporal resolution of 
the system can still be improved, as well as tech 
niques for image processing and enhancement. From 
the methodological point of view, the interpretaion 
accuracy depends upon the selection of the training 
areas, the classifier used and how successfelthe 
extrapolation of this local knowledge can be achie 
ved for the whole area of interest. Crucial for this 
are the variations of the spatial, spectral and 
temporal characteristics of the terrain itself. 
As F.Henderson (1980) put it very clearly : the 
omnipotent role of the environment. 
In most parts of Western Europe, we have to deal 
with complex and fine structured landscapes, charac 
terized by important human pressures. Even the power 
ful ground resolution of second generation sensors as 
TM of Landsat and HRV of SPOT, remain rather coarse 
for thematic inventorization and is, for small areas, 
still not competitive to classical ground methods. 
Therefore, a geographical landscape classification 
may be helpful to give a regional and structural ba 
sis for a stratified samp! ing of the training areas 
and a more intelligent extrapoation of the results 
obtained there. 
2. BASIC CONCEPTS 
Landscapes should be considered as holistic phenomena 
which continuously change on the 3 dimensional space 
of the earths surface. They reflect the efforts made 
by man through history to adapt, shape and organize 
the natural environment to its cultural needs. 
Consequentely, landscapes do not vary by the natural 
conditions of the environment alone, but also by 
cultural zones. Generally spoken, cultural factors 
determine more the regional variation then the na 
tural ones in areas with a high human pressure 
(high population density, high technological level 
and a long history). 
Landscape studies make the distinction between 
landscape elements, components and structures. 
This is mainly based upon their spatial and topolo 
gical characteristics. Each of them is described 
using attributes which can have different levels of 
measurement. Landscape elements are discrete objects 
(houses, single trees, fields, etc.) of relative 
small size and consequentely they are characterized 
by high spatial frequencies. Landscape components 
(relief, microclimate, etc...) change continuously 
and gradually through space. They may remain almost 
constant within some areas which can be small or 
large and which are called landfacets according the 
already classical system of landclassification 
(Howard J. 5 Mitchell C., 1980 ; Webster R. & 
Beckett P., 1970 ; Mabbutt J., 1968). Their spatial 
frequency is lower then for landscape elements, but 
may vary a lot. Landscape structures (fieldpatterns, 
road networks, relief- and dratnagepatterns, etc.) 
are spatial arrangements of landscape elements and 
landfacets and can be described by typological 
parameters as density, connectivity, orientation , 
and so'on.On small scale maps, zones wth a similar 
and a constant structure are delineated by immate 
rial boundaries to form the high order landunits in 
the hierarchical landclassification systems (landsy- 
stem, -region, -province and -division) .