an entirely new and separate entity. Consequently, there are no logical dependencies or relationships captured. Of 
particular importance is the ability to maintain details of the process by which the (spatial representation of) features 
are defined for the following reasons: (i) their derivation can be subject to analysis, and possibly refinement; (ii) 
justification can be given as to the suitability of a feature for a given task and (iii) a comprehensive statement of 
uncertainty introduced by the featurisation process can be given. 
The following list describes the desirable properties for an integrated geographic data model, addressing the above 
shortcomings and adding some further functionality that becomes possible only when an integrated approach to 
spatial data handling is adopted. 
1. Ability to take in and view data in a number of ways, including in its primary (raw) form and in feature 
(processed) form. A consequence of building a fully integrated GIS is that the potential user-base is 
significantly widened. The resulting system may be used by people and application programs with a wider 
variety of backgrounds, skills and agendas. Consequently, a number of different views must be provided onto 
the underlying data that are appropriate for each user-base. This paper introduces four such views, appropriate 
for the integration of remote sensing technology. A formalisation of the integration process is presented in the 
form of high level description of view transformations and a lower level object schema and an extended Object 
Retrieval Calculus (Worboys, 1990), (Gahegan 1994b), in the second part. 
2. Ability to divorce details of implementation and storage from any logical relationships in the data, implying the 
hiding of these details from the user (logical abstraction). Other researchers, for example Albrecht (1994), 
stress the need to construct GIS functionality around the tasks that the user might wish to carry out, and not 
around a particular data structure that happens be used to represent the data. Although it takes a lot of effort to 
solve the deep semantic issues that this approach raises, it also presents the opportunity to design a geographic 
data model that is at last logically independent of the physical layer. 
3. Ability to support multiple spatial representations of features where required. Multiple spatial representations 
are a natural consequence of scale, uncertainty, temporal change and different approaches to data processing 
that routinely affect the data we use, thus they should play an important role in many types of spatial analysis. 
This point is taken up in Section 2.4.2 below. 
4. Ability to describe the formation of all derived objects. All derivations should be repeatable and 
communicable. Many of the feature descriptions used within GIS (and computer vision in general) can 
unfortunately be described as unrepeatable, unrefutable and subjective science 1 , in that little is known 
concerning the reliability or applicability of the results in terms of the features produced. As Davis and 
Simonett (1991, p. 200) point out, within GIS “...the processes and data used to generate the cartographic 
information are usually unknown or irretrievable...". This in turn affects the validity of the results produced by 
an unknown amount. The problem arises in part due to the separation of roles between data producers and data 
consumers; here the remote sensing specialist and the GIS specialist (Rhind & Green, 1988). All too often, the 
meta-data that is required for the correct and appropriate application of data is lacking. The integrated 
approach described here alleviates this problem. 
5. Ability to reason with feature formation details. If the formation of a feature is known then its applicability in a 
given set of circumstances may be calculated. In addition, it becomes possible to impose constraints on the way 
in which features may be processed and combined, allowing the enforcement of consistency checking across 
various domains (temporal, spatial, thematic, uncertainty and scale). 
1 There is a movement in some circles, (e.g. Haralich, 1994; Prechelt, 1994) to make such experiments repeatable, possibly by 
others, so offering a much broader basis for evaluation and comparison.