In addition to the achievements of this work, it is important to consider its limitations, and to determine where further improvements might be required. The work within this thesis could be extended in many directions; thus, looking at these limitations will help evaluate the extensions that would be relevant.
The work presented in this thesis placed particular focus on the representation of ge- ographic phenomena which can be described in terms of changing spatial extensions of geographic features. Moreover, the kinds of geographic features that can be represented are limited to those features whose extension at a particular time can be defined as a 2- dimensional polygon corresponding to some portion of the earth’s surface. Clearly, these geographic phenomena represent only a small portion of the geographic domain. There- fore, the approaches developed here may not be applicable to represent other kinds of
addressed by Devaraju and Kuhn [27], where a process is regarded as having physi- cal objects and substances as their participants. The representation of those phenom- ena involves considering certain physical and chemical transformations that might occur amongst events and process participants. An example of a phenomenon of this nature is ‘evapotranspiration’.
Moreover, there are other kinds of phenomena that can be described in terms of changes of values of attributes observed for a particular region, without regard to changes in the spatial extension of the region. The representation of these kinds of phenomena might be based, for example, on geo-sensors applied to a particular region whose spatial extension is static over time. An example of geographic phenomena of this nature is de- scribed by Kulik et al. [56], in which vegetation modification events are examined. The logical framework presented in this thesis can potentially be extended to deal with other kinds of phenomena, without much modification to most of its formal apparatus, includ- ing the approach to modelling temporal aspects of events and processes, to determining the relationships between them and the method of handling temporal vagueness.
Although this work concentrates on the representation of geographic features whose spatial extensions are subject to change over time, the focus was not placed on the de- velopment of a logical language which can represent many different spatial changes that might affect these features. Rather, this thesis presents logical definitions of some spatial changes to illustrate how they can be specified within the framework; and then it explores one of them (i.e., expansion) to carry out experiments using the system prototype. Spatial changes affecting 2-dimensional polygons have already been extensively discussed in the literature. Therefore, in this work, efforts have been directed to design the framework in such a way that additional spatial changes can be defined with no impact to the rest of the semantics. Therefore, an extension of this work would be to provide definitions for a larger number of spatial changes which may affect spatial extensions of features. Other changes that might be included are, for instance, deformation and rotation, as suggested by Claramunt et al. [21].
Amongst the most important limitations of this work is the restricted variety of re- lationships between events and processes. The framework presented in this thesis only provides a way to represent events as chunks of processes, and to represent processes in terms of their constituent events. Further expansions to the logical framework presented in this thesis could incorporate other relationships between these concepts. Several rela- tionships that could potentially be incorporated to the framework are described by Galton [36] (e.g., transitions, repetition, composition, specification).
resenting 1 to n relationships between different event and process classifiers. For ex- ample, an event could be determined by a chunk of two different process that proceed in parallel. Methods of specifying relationships between different event classifiers and between distinct process classifiers are also desired. Moreover, even more complex situ- ations could be represented by incorporating methods of modelling relationship patterns between events and processes, similar to as developed within the semantic formalism pro- posed by Claramunt and Th´eriault [20], which incorporates the Event Pattern Language (EPL) [42, 43] to model changing elements of geographic space. Using such a kind of language, occurrences of an event associated with a certain classifier can be identified by matching patterns of occurrences of events associated with other classifiers. In these lan- guages, event/process patterns are specified using expressions which resembles regular expressions1. However, additional capabilities to represent certain temporal aspects (e.g., duration) are still the subject of further investigation.
In this thesis, a process is regarded as an entity which is subject to change over time. However, the approach to representing these changes is considerably limited in the frame- work proposed in this work (i.e., the representation of process change is mostly based on the concept of process activeness). Further expansions to this work could therefore con- sider a number of different properties that could be ascribed to processes. For example, a process may be described as being constant, or intermittent, or slowing down, or acceler- ating. The representation of these changes requires dealing with different kinds of vague- ness, and standpoint semantics appears to be applicable to most situations. The incorpo- ration of an improved representation of process properties (together with the provision of methods of specifying relationship patterns between events and processes, described above) would make the logical framework an important resource for the development of theories of causality for geographic phenomena. For example, as described by Kulik et al. [56], deforestation caused by different agents leads to different impacts on the vegetation. Therefore, if the cause (i.e., origin) of a phenomenon is unknown, it might be inferred by analysing its impact on geographic space.
The approaches described in Chapter 3 to modelling spatio-temporal data and rep- resenting geographic features have some limitations and therefore could be improved in several forms. First, the geometric representation of Stars is restricted to 2-dimensional polygons. Therefore, the model could be improved to allow the representation of other geometric types (e.g., points and lines), as well as to enable 3-dimensional representation of space.
Another limitation within the approach to representing spatio-temporal data is that the
representation of compound geographic features is currently determined by part-of rela-
tions which may hold between homogeneous coverage attributes and one heterogeneous attribute. Although this is sufficient for representing many different types of geographic features, this could be improved to represent more complex scenarios. As discussed earlier, features are maximal well-connected extents of their corresponding coverage at- tribute; however, given the limitation of this model, no other feature can be proper part of this (neither of different type or of the same type). Therefore the system does not al- low, for example, a city to be part of an island (i.e., a maximal chunk of urbanisation to be part of a maximal chunk of land). Thus a potential enhancement would be to allow features to be represented based on a multiple-level attribute hierarchy, where geographic features could contain other features of different types. The current version of the pro- posed framework is based on a polymorphic relation CC, which relates a pair of attributes of different types. Observe that it works well for the reduced variety of scenarios which can currently be represented. However, for representing more complex scenarios, the use of distinct relations with different properties would become essential (such as CPcc, CPcf,
and CPff, relating, respectively, a pair of coverage attributes, a coverage attribute and a
feature attribute, and a pair of feature attributes).
Moreover, an improvement to the approach to representing geographic features would affect the method of inferring the type and the spatial extension of geographic features. It might incorporate other existing approaches to handling spatial vagueness. For example, Bennett et al. [12, 13] proposes a method of handling vagueness in which the geographic feature type can be inferred based on different geometric characteristics (e.g., a water body can be classified as ‘river’ or ‘lake’ depending upon its level of ‘linearity’). More crucially, a more complex representation of a feature life should be developed. The pro- posed model is significantly limited in this aspect, and consequently is not capable of representing effectively with splits, merges, and trajectories affecting features. It should be observed that this affects directly the interpretation of the identity of events and pro- cesses. For example, consider two disconnected regions which undergo urbanisation, characterising two distinct processes going on. Then, at a certain time, these urbanised regions get connected to each other, therefore characterising a single process going on for the whole region.