tuning the GIS to the application needs. Only the knowledge 
which is transferred to the developper will be available in the 
system later. Knowledge acquisition, however, is not only the 
transfer of expertise. Knowledge acquisition is a creative process 
in which the systems builder constructs qualitative models of 
human behaviour. This makes it necessary to find a structured 
approach for knowledge acquisition. In artificial intelligence well 
known methods of knowledge acquisition are : 
e interviews such as interviews in standardised form or inter- 
views focussed on a specific problem, 
group discussions, where several specialists are discussing 
about problem solving, 
observations and own experience, when joining the applica- 
tion domain, 
contents analysis, which includes study of literature as well 
as study of the manuscripts of interviews and group discus- 
sions, 
learning methods (cf. E. Rich (1988)) such as routine learn- 
ing, conceptual learning, learning by analogy, learning in a 
general problem solver, learning by exploring etc, which are 
often used for machine learning purposes. 
others. 
From those methods especially conceptual learning seems to be 
an appropriate way to gain knowledge about an application do- 
main for GIS purposes. The idea of conceptual learning is to 
construct classes of objects, where the typical properties are iden- 
tified and collected. A class may then be defined as a structure 
of the common properties or components of all the instances 
(PART.OF-relation). Objects in these classes may have rela- 
tions to other objects in a hierarchical manner (IS. A-relation). 
This shows a similarity with the entity-relationship approach and 
with structural and behavioural object-oriented database design, 
which makes it a usefull tool for the data modelling part in a 
GIS. Besides the procedural - or behavioural - approach in the 
application domain may as well be analysed and the knowledge 
about this may as well be captured using conceptual learning. 
One may distinguish two types of knowledge acquisition ap- 
proaches, which are the model-based and the incremental knowl- 
edge acquisition. In model-based knowledge acquisition the 
knowledge acquisition step is separated from the design and im- 
plementation phase. So the analysis of the expertise is indepen- 
dend from the knowledge representation later chosen for imple- 
mentation. The result is a model of the expertise, its methods 
and their cooperation. In incremental knowledge acquisition the 
process of knowledge acquisition is regarded as an incremental, 
approximative and faulty process. The model of the expertise 
is produced during knowledge acquisition, which demands for a 
stepwise modelling, detection and correction of errors. An explo- 
rative prototyping - the prototype is only used for gaining the 
knowledge and designing a model of the expertise — is very usefull 
for that process of analysing the expertise making the design and 
implementation of the system independend from the prototype. 
The design of the appropriate data model for the application and 
the design of the procedural flow including the adaptation of own 
methods on the GIS-product may be seen as two important steps 
where knowledge acquisition method could become very power- 
full tools. In practise usually the step of knowledge acquisition 
and representation is very often ignored or executed in a less 
detailled way. Instead of modelling the application domain an 
immediate data structuring and procedural flow is implemented 
taking the specialised GIS-product characteristics into account. 
738 
The disadvantage of this procedure is that the knowledge about 
the application domain is lost (or hidden). In particular one can- 
not easily recognize what part of the application domain is real- 
ized according to the specifications and what parts of the usual 
workflow is not supported because of system restrictions. The re- 
sult of such an approach is very often that a solution exists, which 
is only partially related to the applications problem. It is pro- 
posed to use knowledge acquisition and representation methods 
for analysing and designing the application domain before im- 
plementing it on a GIS which is only able to manage procedural 
knowledge. The model-based knowledge acquisition is assuming a 
complete model of the application domain; it can be set up when 
the expert knowledge is transferred. Such an approach should 
be taken if the application domain is well known and production 
cycles are well established. The practical experience on the other 
hand shows that an incremental approach could become a bet- 
ter way to implement GIS in a new application field. But here 
usually the method of implementation is based on evolutionary 
prototyping, which means that at the end the prototype results 
in the production system. This very often has the disadvantage 
that production is restricted to the capabilities of the GIS. 
The problem of all knowledge acquisition methods lies within the 
interpretation and analysis especially with respect to complete- 
ness and complexity. How does a human expert reach his problem 
solving feasibility ? Expertise also is based on human capabilities 
and implicit knowledge, that is not easy to query. Experts also 
rely on unconcious knowledge, even if they present it differently. 
Experts are experts in doing things not in explaining them. 
2.4 Knowledge representation 
The term knowledge representation’ is defined by U. Reimer 
(1991) as the describing of symbols — so called representation 
structures — which fit in a recognizable manner to an extract 
of the world to be represented. Furtheron an interpretation in- 
struction is needed which allows the formulation and evaluation 
of queries on that structure. A knowledge representation is stand- 
ing for a set of circumstances — the represented part of the world 
— and is in itself a model of this part of the world. The knowledge 
base consists of the knowledge used by the system which is ex- 
plicitely coded and is not implicitely hidden in the programming 
code. For further details in general see again U. Reimer (1991) 
and in a GIS-context see R. Bill (1991). There are two types of 
knowledge representation : 
e Rule-based representation formats handle the rules in the 
form of statements. Well known representation schemata are 
models based on logic theory or on production rules. The 
whole knowledge is written down in the rules and statements; 
the data base itself is very often an unstructured and passive 
set of facts. In particular of interest for GIS is the fuzzy-logic 
which in comparison to the binary or boolean logic, where 
only true (0) and false (1) are allowed, supports a set of 
states. The fulfillment of a rule or statement may range in 
the intervall [0, 1]. An example of fuzzy-modelling is given in 
the application chapter in this paper (Figure 5). A further 
example for production rules is given there, too. 
Object-centered representation formats are placing the ob- 
ject itself in the foreground. Here ideas such as inheritance, 
methods etc. are included, these ideas are also found in 
object-oriented data modelling. Well-known types of object- 
centered representation formats are semantical networks, 
frames and scripts. For more detail see again U. Reimer 
(1991) in general and R. Bill (1991) with respect to GIS- 
applications. Semantical networks are ideal representations