The complexity of the natural environment, its polythetic and dynamic character, requires appropriate new methods to represent it in GISs, if only because in the past there has been a tendency to force reality into sharp and static objects. A more generalized spatio-temporal data model is required to deal with fuzziness and dynamics of objects. This need is the motivation behind the research reported in this thesis. In particular, the objective of this research was to develop a spatio-temporal data model for objects with fuzzy spatial extent. This thesis discusses three aspects related to achieving this objective: identification of fuzzy objects, detection of dynamic changes in fuzzy objects, and representation of objects and their dynamics in a spatio-temporal data model. For the identification of fuzzy objects, a six-step procedure was proposed to extract objects from field observation data: sampling, interpolation, classification, segmentation, merging and identification. The uncertainties involved in these six steps were investigated and their effect on the mapped objects was analyzed. Three fuzzy object models were proposed to represent fuzzy objects of different application contexts. The concepts of conditional spatial extent, conditional boundary and transition zones of fuzzy objects were put forward and formalized based upon the formal data structure (FDS). In this procedure, uncertainty was transferred from thematic aspects to geometric aspects of objects, i.e. the existential uncertainty was converted to extensional uncertainty. The spatial effect of uncertainty in thematic aspect was expressed by the relationship between uncertainty of a cell belonging to the spatial extent of an object and the uncertainty of the cell belonging to classes. To detect dynamic changes in fuzzy objects, a method was proposed to identify objects and their state transitions from fuzzy spatial extents (regions) at different epochs. Similarity indicators of fuzzy regions were calculated based upon overlap between regions at consecutive epochs. Different combinations of indicator values imply different relationships between regions. Regions that were very similar represent the consecutive states of one object. By linking the regions, the historic lifelines of objects are built automatically. Then the relationship between regions became the relationship or interactions between objects, which were expressed in terms of processes, such as shift, merge or split. By comparing the spatial extents of objects at consecutive epochs, the change of objects was detected. The uncertainty of the change was analyzed by a series of change maps at different certainty levels. These can provide decision makers with more accurate information about change. For the third, and last, a process-oriented spatio-temporal data model was proposed to represent change and interaction of objects. The model was conceptually designed based upon the formalized representation of state and process of objects and was represented by a star-styled extended entity relationship, which I have called the Star Model. The conceptual design of the Star Model was translated into a relational logical design since many commercial relational database management systems are available. A prototype of the process-oriented spatio-temporal data model was implemented in ArcView based upon the case of Ameland. The user interface and queries of the prototype were developed using Avenue, the programming language of ArcView. The procedure of identification of fuzzy objects, which extracts fuzzy object data from field observations, unifies the existing field-oriented and object-oriented approaches. Therefore a generalized object concept - object with fuzzy spatial extent - has been developed. This concept links the object-oriented and the field-oriented characteristics of natural phenomena. The objects have conditional boundaries, representing their object characteristics; the interiors of the objects have field properties, representing their gradual and continuous distribution. Furthermore, the concept can handle both fuzzy and crisp objects. In the fuzzy object case, the objects have fuzzy transition or boundary zones, in which conditional boundaries may be defined; whereas crisp objects can be considered as a special case, i.e. there are sharp boundaries for crisp objects. Beyond that, both the boundary-oriented approach and the pixel-oriented approach of object extraction can use this generalized object concept, since the uncertainties of objects are expressed in the formal data structures (FDSs), which is applicable for either approach. The proposed process-oriented spatio-temporal data model is a general one, from which other models can be derived. It can support analysis and queries of time series data from varying perspectives through location-oriented, time-oriented, feature-oriented and process-oriented queries, in order to understand the behavior of dynamic spatial complexes of natural phenomena. Multi-strands of time can also be generated in this Star Model, each representing the (spatio-temporal) lifeline of an object. The model can represent dynamic processes affecting the spatial and thematic aspects of individual objects and object complexes. Because the model explicitly stores change (process) relative to time, procedures for answering queries relating to temporal relationships, as well as analytical tasks for comparing different sequences of change, are facilitated. The research findings in this thesis contribute theoretically and practically to the development of spatio-temporal data models for objects with fuzzy spatial extent.
