The enormous advances in computer technology in recent years made it possible and necessary to give consideration to temporal geographic information systems. Several conceptual frameworks have been proposed, and a few partial implementations have been reported, but it will take a big step until off-the-shelf temporal GIS will be available. Despite these initial advances, the technical and conceptual difficulties still require a large amount of attention (Ramachandran et al., 1994; Peuquet, 1994; Langran, 1993).
Langran (1988) examined a concept called dimensional dominance, where access to the data is classified as either predominantly spatial or predominantly temporal to optimize data and algorithms. She examines four representational models for spatio-temporality based on existing non-temporal data models (Langran, 1992b): (1) Space-time cube, (2) Sequential snapshots, (3) Base state with amendments and (4) Space-time composite.
Except for the first representation, there are no implicit temporal relations between objects and states involved. 2) and 3) contain time as separate data layers, whereas 4) handles the temporal aspect separately in the non-spatial attribute database (Kienast et al., 1991; Langran, 1992b).
Peuquet (1994) proposes an integrated approach called triad representational framework, which is an extension of the dual representational framework described earlier (Peuquet, 1988). It integrates temporal, locational and object-related semantic aspects of the data using spatial learning and knowledge-based scene interpretation.
The efforts to build prototype temporal GIS were taken to satisfy specific needs and therefore were implemented only with partial support of temporal aspects. Beller et al. (1991) developed a prototype temporal GIS as a proof of concept to carry out global change research. The central concept is the temporal map set object (TMS), which is a collection of GIS maps representing the same area and theme at different times. The system is capable of performing interpolation between time slices and can incorporate events as separate binary TMSs. Langran (1993) describes implementation issues to consider when building temporal GIS, citing specific problems from automated hydrographic and aeronautical charting systems (Langran, 1990) and a forest GIS (Langran, 1992a). Main aspects seem to be representation, incremental updates, temporal generalization and longevity. Incremental updates produce artifacts and edge effects which can cause data to be inconsistent. The Environmental System Research Institute, Inc. (ESRI) has developed a temporal GIS application for a private forestry management firm (L. Montgomery, pers. comm.). The implementation is based on an area event system which maintains valid times and allows historical queries.
The application of GIScience in the domain of changing phenomena seems to have a promising future. Wachowicz and Healey (1994) stated that ``by producing a lineage of data to track the historical information associated with real-world phenomena, temporal GIS will provide analytical tools for the recognition of patterns of change through time as well as the prediction of future changes, by implementing dynamic simulations''. Today this is still more likely to be a wish than reality. In contrast to the numerous prototypes of temporal database systems most current GIS products remain snapshot-oriented systems capable only of static representations of data (Bagg and Ryan, 1997). In contrast to the database research the need for adequate representation of temporal data becomes an urgent need in GIS, because large databases are being built and updated for monitoring the continuously changing environment. In these databases spatial and temporal inconsistencies are not allowed, because today's analysis methodology is not able to cope with inconsistent data. Today's temporal GIS research is mostly concentrating on data representation and query (Yearsley and Worboys, 1995; Worboys, 1994; Ramachandran et al., 1994; Kemp and Kowalczyk, 1994; Güting et al., 1998; Peuquet and Wentz, 1994), but the analysis part is not receiving much attention. Güting et al. (1998) provide a semantic foundation for handling time dependent geometries. They are the first authors considering moving objects (e.g. airplanes) in their work for founding their framework. This seems to be a major step (even though still on a theoretical level) to overcome the event centered view used by most other researchers. Most prototype applications are using time only as an additional selection criterion, but neglect the possibility of completely new analysis methods for temporal phenomena.
In the early 90ies, a frame free geographical representation was postulated as a research aim in GIScience (Kemp, 1993; Tobler, 1989). Although this has not been realized yet, it has become a prominent issue in handling temporal geodata.