FUZZY CLASSIFICATION OF DIGITAL ORTHOPHOTOS 
FOR SPATIO-TEMPORAL LANDSCAPE MODELLING 
Wim J. Droesen and A. Peter van Deventer 
Grontmij Geogroep bv, Roosendaal, the Netherlands 
Mark van Til 
Amsterdam Water Supply, Amsterdam, the Netherlands 
KEY WORDS: Digital photogrammetry, fuzzy classification, landscape monitoring, vegetation dynamics. 
The paper describes the process of conceptual and concrete digital landscape modelling based on digital 
photogrammetry. The methods are integrated in a high resolution monitoring system that is able to represent both 
discrete and continuous terrain features, i.e. objects and fields respectively. The objects are constructed through 
crisp image interpretation techniques, while continuaus spatial variation is interpreted by fuzzy classification. The 
system is successfully applied for the mapping of vegetation dynamics in the Amsterdam Waterworks Dunes. 
1 INTRODUCTION 
In the last decade Gl-systems have become a standard 
component of the instrumentation in landscape 
ecological research (e.g. Turner and Gardner, 1990; 
Haines-Young et al., 1993; Johnson, 1990). For a longer 
period remote sensing data have been recognised to be 
an indispensable source of information for the 
landscape modeller. Initially, analogue aerial 
photographs were used. Nowadays digitised aerial 
photographs with a resolution less than 1 metre and 
other digital remote sensing data such as satellite 
images are frequently used (Quattrochi and Pelletier, 
1990). 
The digital elaboration of aerial photographs yields 
products with a high geometric accuracy, like topo- 
graphic maps and digital elevation models. When 
additionally to the geometric corrections some 
radiometric corrections are appiied, digital orthophotos 
are produced ready for quantitative evaluation. These 
orthophotos can be treated like any remotely sensed 
image. Consequently, digital interpretation techniques 
can be applied to obtain all sorts of high resolution 
thematic information. 
The availability of Gl-systems for a structured 
processing of environmental data and digital high 
resolution image processing as a tool for spatial data 
acquisition opens new possibilities for the modelling of 
natural landscapes. However, the application of these 
techniques not only facilitates the construction process 
of a digital landscape model, but also brings about the 
need to reconsider the concepts underlying the 
modelling process (Haines-Young et al., 1993). While 
working with Gl-systems and remote sensing, 
214 
ecologists often adhere to concepts and working 
methods, whereby the digital environment is utilised 
but not fully exploited. The objective of this paper is to 
introduce some new concepts for landscape ecological 
modelling which enable the construction of digital 
landscape models that more closely represent reality 
compared to most conventional landscape models. 
1.1 Landscape Monitoring System 
The process of ecological conceptualisation of a 
landscape (i.e. structure, function and change) to a 
discrete representation can be subdivided into several 
leveis of abstraction (fig 1). Kemp (1993) terms the 
models on the first level geographic models. Geographic 
models are conceptual models used by modellers ‘as 
they evolve an understanding of the phenomenon 
being studied and extract its salient features from the 
background of infinite complexity in nature’. Because 
we focus on ecological features in this paper, it is more 
appropriate to term these type of conceptual models 
landscape-ecological (LE) models. 
The second level of abstraction is represented 
by spatial models. Conceptual spatial models are 
formally defined sets of entities and relationships used 
to discretize the complexity of landscape-ecologicl 
reality (Goodchild, 1992). The entities in these models 
can be measured and the models completely specified. 
On the next level data structures describe details of 
specific implementations of spatial data models 
(Molenaar, 1994). Data structures and lower data layers 
are considered to be part of the instrumentation. 
The spatial model has to follow from the 
specifications in the  landscape-ecological model. 
However, in practise one works usually the other way 
around, starting from the spatial models readily 
implemented in commercial Gl-systems. In general 
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996 
thi 
ful 
ec 
Thi 
lan 
ori 
The 
Spe 
SUI 
ph 
Cru 
the 
spa 
exi: 
Sys 
À | 
dis 
of : 
wit 
exp 
Exa 
are 
Als: 
by: 
ren 
rep 
ava 
con