elementary objects at a lower level; e.g. at a municipality 
level a GIS may contain houses, streets and parks, while at a 
national level a GIS may contain towns and urban areas. 
Another aspect is how the data will be used: is it for 
monitoring the terrain situation, or is it for the analysis of 
the situation, or will the data be used for planning purposes 
etc. Each field of activities puts its own requirements the 
type of terrain description, although there are often overlaps 
between these requirements. A fourth important aspect that 
in many disciplines the relevance of data changes with time. 
In agriculture for example, the requirement for soil 
information has changed during the last decades. Originally 
the main interest was to analyze the suitability of soils for 
sustaining different crops. At present the interest changes to 
e.g. the capacity of the soil to store chemical elements which 
could do harm to the environment. 
These observations show that the decision of what data are 
relevant is always context dependent, i.e. the elementary 
objects with their classes at the different hierarchical levels 
and linked to these classes the attributes with their value 
domains. For these elementary objects it should be decided 
also to which geometric type they belong. This will depend 
on the geometric role they play in the terrain description, 
this role should not be confused with the appearance of these 
objects in reality. It may be that a river is treated as a line 
object in a data base for hydrological purposes, whereas the 
same river is treated as an area object in the data base of the 
authority responsible for its maintenance. Similarly a town 
may appear as an area object in a data base for demographic 
purposes, whereas the same town appears as a point object 
in a data bases containing air traffic routes. The decision 
which geometric aspects of an object are important is always 
made in a particular context and that implies the decision 
whether it should be treated as a point-, line- or area object. 
Within such a context one should also decide what are the 
relevant object aggregates and associations that should be 
constructed from the elementary objects. 
Data modelling should be done with care, the discussion of 
the structural and semantic aspects and the context 
dependency of the data explains why. Therefore tools should 
be developed to assist us in the process of spatial data 
modelling, to find out what can be modelled and what not. 
The Modul-R Formalism of [Bedard e.a. 1992] is one 
example of such tools. 
In the explanation of some of the concepts that play a role in 
a geo-information theory reference was made to elementary 
and aggregated terrain objects. These are high-level 
concepts, very close to the conceptual level a GIS user 
would like to think at. Operational GISystems and 
DBMSystems require that these high-level objects are broken 
down into low-level data types that match the logical data 
structures of these systems. The data management in these 
systems has in most cases been organised so that the user is 
forced to think and operate at this low level. This often 
frustrates the use of these systems, because the user may 
have great problems analyzing his complex high-level 
problems when he forced to work at such a low level. Data 
management tools should be developed that bridge the cap 
between these two levels, [Lee e.a. 1992] discus this 
problem and propose a solution. 
752 
UNCERTAINTY 
Data models are prescriptions how objects should be 
represented in an information system. Once these models 
have been defined we are faced with the problem of data 
acquisition to fill the database. This means that terrain 
objects should be identified, they should be classified and 
their geometry should be measured and their attributes 
evaluated. This should be done by measuring operations and 
it is through these operations that uncertainties of different 
types are introduced in the data. Three major types of 
uncertainty will be described shortly here (see for a general 
treatise [Klir e.a. 1989]. 
The criteria for assigning terrain objects to a certain class 
might be fuzzy: e.g. the definition of nature districts is not 
always clear. Does it mean that people do not interfere with 
the development of flora and fauna? If it means that there is 
only a limited interference of people, then how little should 
that be. No sharp criteria can be formulated. 
The geometry and the attributes of the terrain objects should 
be evaluated through measuring procedures or through the 
processing of measuring data. Measuring operations 
introduce in general stochastic components in the observed 
data. These stochastic components propagate through the 
processing steps applied to these data. This type of 
uncertainty is generally expressed in accuracy models in 
terms of variances and probabilities. 
The third type of uncertainty is related to evidence theory. It 
may be that the object classes are clearly defined and the 
object data are accurate, but still the data do not contain 
sufficient information to decide whether a particular object 
belongs to a certain class or not. This case is well known in 
remote sensing image classification. If a classification is 
made to determine the landcover of an area, then the 
landcover classes might be well defined. Still the spectral 
information in the image might not give sufficient evidence 
to assign the parcels with certainty to those classes. 
In [Brimicombe 1992] the problem of uncertainty in GIS is 
treated and proposals are formulated how to deal with it. 
CONCLUSION 
The introduction of information technology does not only 
confront organisations with technical problems, but it does 
have a significant organisational impact. Consequently it 
requires a redefinition of the organisational structure and of 
staff expertise and responsibilities. More direct is the effect 
on the information flow through the organisation. This 
concerns the technical and structural aspects of the 
information flow, but also (and not the least) the definition 
of the data and data processing models. 
Tools should be made available to experts involved in this 
task. For organisations dealing with Geo-Information these 
tools should be embedded within the framework of a theory 
dealing with the structure or syntaxis and semantics of GEO- 
information and the structure of its processing models. 
Several structural and sematic aspects of data modelling have 
been explained in the previous sections. [Bedard e.a. 1992] 
[Lee e.a. 1992] and [Brimicombe 1992] explain how tools 
for some of these modelling tasks can be developed. 
LIT 
[Als 
[Be« 
[Bri 
[Bro 
[Bro