Compiled and edited by
Scott M. Freundschuh and Jayant Sharma
Papers that were selected for inclusion in the meeting were grouped into six topical categories. Categorization was based on papers that addressed spatio-temporal issues either from a similar perspective, or with similar methods. The topical categories were:
Discussant: Nicholas Entrikin.
Separation of objects from subject
The paper by Helen Couclelis titled Making Space for Time: Towards a Truly Temporal Geography is concerned with putting time and process into GIS, focusing specifically on time in GIS. She used Zeigler's discrete-time and discrete-event model as a methodology for putting time into GIS. In this model, space and time are seen as taxonomic categories. Space and time are removed from human experience--we experience the world as a whole and cognition allows or enables us to isolate concepts such as space, time, and perception. Under this view, objects are separate from subject. This approach works well for physical geography, spatial analysis, and social sciences. Entrikin, however, questioned how space and time relate to particular phenomena within this framework. He took the position that one has to consider space and time as a connection between subject and object, if one wants to look at a broader space-time relationship.
Physical vs. human time
Couclelis further suggested that maybe we should not make such a big distinction between physicalist time (clock time) and human experience (body time), because we do experience physical time. Humans live in a physical as well as cultural worlds--the physical science view "gets our foot in the door."
Separation of space and time in GIS
In a decentralized view of space, GIS represents time from the perspective of physical sciences, while human (social) science centers time on individual experience. Entrikin suggested to also consider in this model place as experience of milieu over time--the value of place as an agent in the world. Similar to previous discussions at the San Miniato workshop, the question was raised whether we can talk about space and time in GIS as interrelated, not separate.
The same issue of the separation of space and time came up again in the discussion of Freksa's paper, though in a somehow different setting. Freksa argued that it is possible to represent space and time in some representational medium, like the computer. In computer science we have propositional representations (logic) vs. pictorial representations (something that we see). Do we really have these type of representations, and if so which one(s)? We can use pixels, which are propositional, to represent the world. We can look at individual pixels and can swap pixel for pixel, this really does not matter. But the resultant picture does not make sense. Where pixels actually belong so that the image does make sense is dictated by time. In other words, we need time to provide order to space for understanding. For this particular application, time and space cannot be separated.
Papagno in his paper Seeing Time, made the same argument, suggesting that we need space to talk about time, and vise-versa. Looking at the effects of people on the landscape over time is one way to model changes in space over time. His paper raised the question how do processes or human's activity shape or modify the landscape so that we can see time in this landscape.
The relation between formal models (e.g., rules) and reality
The paper by Freksa titled Temporal and Other Issues in Qualitative Spatial Reasoning explores qualitative spatial reasoning in a natural language system. It seems that the rules are more developed for time than for space. When developing formal spatial inference systems, how does the model relate to the reality being modeled? Were do we locate rules in the world? What do these rules mean? Should one look at human behavior as simply rule-based? Can the fundamental relations of human space be found in cognitive science? This is mostly an experiential view, i.e., knowing the rules without being conscious of them. Entrikin argued that geographers have a contribution to make here. The spatial-temporal reasoning question must be broader. The models must be integrated to human experience. Our cognitive categories pull the parts apart, while our investigations attempt to put them back together assuming an experiential view. GIS is limited to physical notions of space. We must view this in a broader context and discover how to merge culture and GIS, in order to make GIS trans-cultural. Entrikin concluded if GIS is used to redesign space, it has the potential to reshape our geographies.
Freksa warned to look very carefully at the representation of the real world. Our goal is to create representations to model aspects of our cognitive world, and we have to realize that these representations are undergoing continuos revision.
The discussions about formal aspects of spatio-temporal reasoning continued from the perspective of computational theories, focusing on the development of models for dealing with spatial and temporal information. Mike Worboys pointed out that this is the view taken by Computer Science, the science of the artificial, in that it deals with human constructions rather than the real world in the way geography does.
Designing computational tools
The paper by Cohn et al. titled Exploiting Temporal Continuity in Qualitative Spatial Calculus presents a logic-based framework for reasoning about spatial knowledge. The goal is an axiomatic theory of space and time that can provide a formal basis for explicit representation and reasoning in Artificial Intelligence systems. Their formalism works for either spatial or temporal regions and provides a suite of primitive relations and inference mechanisms. In particular, they define a composition table for the set of primitive relations. This table can be used for checking the consistency of spatio-temporal constraints or even the spatio-temporal evolution of the regions' properties. In the spatial domain their definitions of primitive regions and composition table are the same as those developed independently by Max Egenhofer and his colleagues (Egenhofer and Franzosa, 1991; Egenhofer et al., 1994). Thus while Cohn's group were not explicitly concerned with GIS, their work has potential application in the design and construction of such systems.
It was discussed whether Allen's (1983) work essentially solved the problem of representing temporal relations. This was not Allen's goal; rather it was an approach to temporal reasoning that has a solid formal foundation and hence proved to be a successful starting point for many researchers. The fundamental operational question is: Are we building tools or constructing models of human behavior? As yet we are dealing with tools, GIS being one of them. These tools could either support human reasoning or simulate it. Thus, Cohn pointed out, that while Allen's and their work does not necessarily solve the problem of representing temporal relations and temporal reasoning, it does provide a computational tool that has its definite uses.
Tobler observed that while graphics is considered as mainly a display mechanism, it actually is a powerful visualization tool that allows one to determine new patterns in data. Since human cognition is visual, the tool is important and powerful.
A finite set of spatio-temporal operations and relations
This led to another fundamental question of whether there exists a finite set of spatio-temporal operations and relations? The answer lies largely in the definition of the formalism. A simplified and restricted model could have a well-defined finite set of primitive relations and operations however their utility would be in question. Cohn stated that the group intended to test their ideas by applying them to various domains such as qualitative physics an GIS.
Chrisman asked people to consider the artifact of coordinates. These are simply a tool for encapsulating metric information that is so vital for much of spatial reasoning.
Bitemporal models for GIS
Mike Worboys' paper, A Generic Model for Spatio-Bitemporal Geographic Information, describes a model that handles objects that may be embedded in two orthogonal spatial and two orthogonal temporal dimensions. A purely spatial generic object model is extended using bitemporal elements that account for temporal referencing of the objects. An example of a bitemporal reference being the distinct event time, when an event occurred, and the database time, that is when the event was recorded in the information system.
The distinction between event and database time brought up the issue of how would an information system deal with legacy or historical data? The problems are: (1) legacy data may not fall neatly into the current data model; (2) the volume of data may be too large to successfully physically load it into the database; and (3) the legacy data has a certain semantics, context, and structure. How can this be preserved?
Collaboration between domain specialists and computer scientists
The paper by Liu et al. titled Spatio-temporal Reasoning in Atmospheric Science Databases describes techniques for tracking and detecting evolving physical phenomena in the very large datasets of atmospheric and ocean global models. The approach is mainly an engineering one, constructing a software system, and the project was at the time of the Specialist Meeting in the prototype stage. The particularly attractive aspect of this approach was that a team of computer scientists worked with a team of atmospheric scientists. They used primarily existing atmospheric and ocean global models to determine and define the spatio-temporal constructs.
Designing and implementing temporal GISs
The major concerns of this implementation are the sheer volume of data, and extending and integrating systems that allow rule-based spatio-temporal reasoning. The basic building block is a deductive database system, LDL++, extended with spatial and temporal construct suitable for the application domain of atmospheric sciences. The temporal reasoning capability is provided by an Event Pattern Language based on regular expressions and logic. The rationale behind the Event Pattern Language was providing a means for detecting and reasoning about sequences of temporal events. More specifically their interest was identifying spatial features and tracking them over time, an example being monitoring a cyclone and cloud formations. The question arose whether their event time model was sufficient for the application considering that many atmospheric phenomenon are cyclic in nature. The benefits of this effort are the practical and quantifiable experiences gained in design and implementation of a temporal GIS for a specific application domain.
Complementary aspects of implementing temporal GISs were described in Bill Hazelton's paper titled Some Operational Requirements for a Multi-Temporal 4-D GIS. He focused on a GIS that can handle various models of time, describing the functionality requirements, followed by an outline of a software architecture that would provide such functionality. The temporal functionalites considered were (1) the inclusion of multiple temporal concepts such as linear, cyclic, branching, or multidimensional time; (2) the treatment of topological integrity and sharing of data objects; (3) spatio-temporal indexing of the data to provide efficient retrieval times; and (4) dynamic modeling capabilities. Of interest was the concept of branching time, which Hazelton explained as the temporal concept in those situations where there is no total order between sequences of events or phenomenon. The similarity to versioning was recognized.
Discussant: David Mark
Spatial metaphors
The paper by Háj Ross titled Path, Points and Proforms explored paths and movements in time, and how they are expressed in space. In Ross's model, the description of a moving object's path is composed of the elements source, trajectory, direction, distance, speed, totality, goal, and mode. Mark noted that duration, an element which can be derived from the elements of distance and speed, is not listed explicitly in Ross's model. Ross commented that duration is not a general constituent or notion in language as compared to extent. Examples of extent are prices rising from $1.11 to $2.24; the sky changed from red to yellow through orange; it is 450 miles from Los Angeles to San Francisco; the play ran 5 days from Monday through Friday. How space is represented in language is relevant in understanding how we think about space. We put the world into motion to explain static objects, e.g., "the road runs through the mountain." Space is a common source for metaphors, e.g., dead ends, line of thought. The path schema is central to this idea. Is space the grounding base for the conceptual models that we are interested in here? If we use language to structure our data models, care should be taken so that we do not limit ourselves to studying only the English language. Must consider all languages.
Spatial processes
In the paper Temporal Dynamics and Geographical Information Systems Stephen Stead contends that the real world operates along a time line that is a single ordered line, upon which the temporal extent of data values are plotted. Stead focuses on systems that deal with modeling time in archeological applications. He offers an alternative method to the standard time-slicing technique, which stores snapshots of the landscape that are used as interpretive back-cloths. Stead's solution is to represent known data elements in four dimensions (x, y, z, and t). From these, real world processes can be modeled. Stead added that his interest is societies and human behavior and how they are dictated by the space they live in. Can we reconstruct space that people once lived in? How likely is the model's predictions of the reconstructed space to be true or close to the truth?
Irene Campari's paper Morphological, Topological and Chronological Time in Urban Development views the development of towns as a dynamic and continuous process, which is evident in human actions and events of limited durations. Time becomes a basis for the analysis of relationships between spatio-temporal events in urban space. Time is defined in terms of sequences of events. How do time and space constrain human spatial behavior? Both this paper and Papagno's paper suggest that processes leave traces in space as they evolve in time. Is this a valid view or way of linking time and space? Campari asked, "Does cognitive science look at the effects of space on human cognition?" In other words, if the space changes does the cognition change? How do people feel about the places where they live--e.g., a resident's view of the city vs. a traveler's view.
Discussant: Stephen Hirtle
Though the topic of the Specialist Meeting was on time and space, critical to the cognition of time and space is the representation of spatial knowledge. There is general agreement that spatial knowledge is acquired through temporal processes, but what is not fully understood is how spatial information is acquired, how spatial concepts are represented, and if the internal representation is dependent on the acquisition process.
Distinctions of landmark, route, and survey
Tversky, in her paper Acquiring and Updating Spatial Knowledge from Language, reported the results of several experiments, one of which explored spatial knowledge acquired from narrative (travel books) and from maps. Whether one learns a route or looks at a map, these are inherently different types of information and they lead to different performance. Does this distinction carry over if you acquire spatial information from narrative? The results of Tversky's experiment suggested that the answer to this question is "No"--the results argued for a single representational schema, regardless of how information is acquired. In this study, subjects read route or survey descriptions of naturalistic environments, after which they answered verbatim or inference questions that required either route or survey knowledge to answer. Verbatim statements were verified faster than responses to inference questions, but there were no differences in responses to inference questions dependent on the route/survey distinction. In a second set of studies, Tversky (with Franklin and Bryant) explored spatial knowledge acquisition from small scale scenes described in narrative where objects in the scenes were described in terms of up/down, front/back, and left/right of a protagonist. Results supported the spatial framework model which purports that space is conceptualized in terms of three axes corresponding to bodily position.
The paper by Montello titled A New Framework for Understanding the Acquisition of Spatial Knowledge in Large-Scale Environments argues for a modification of the simplistic landmark/route/survey distinction (which he termed the dominant framework) of spatial knowledge. Montello provided a new framework for spatial knowledge which consisted of five major tenets. The first tenet is that there is no stage at which only pure landmark or route knowledge that contains no metric information about distance and direction exists. Metric configurational knowledge begins to be acquired on first exposure to a novel place. The second tenet suggests that with increasing familiarity and exposure to places, there is a relatively continuous increase in the quantity (quantitative rather than qualitative shift), accuracy, and completeness of spatial knowledge. The third tenet indicates that the integration of knowledge about separately-learned places into more complex hierarchically-organized knowledge structures represents a significant and relatively sophisticated step in the microgenesis of spatial knowledge. The fourth tenet suggests that individuals with equal levels of exposure to an environment will differ in the extent and accuracy of their spatial knowledge. The final tenet indicates that linguistic systems for storing spatial knowledge provide for the existence of relatively pure topological knowledge, or at least non-metric knowledge; however, such non-metric knowledge exists in addition to metric spatial knowledge, not as a necessary precursor or intrinsic part of it.
Space-time relationships
The papers by Block and Freundschuh focused on space/time relationships. Freundschuh, in his paper Cognitive Distance at Various Geographic Scales analyzed when humans use time in lieu of space to estimate distances. The focus of this study was to explore how people normally think about and express cognitive distance in their everyday lives. This approach was in contrast to previous studies in which subject testing occurred in unnatural experimental situations, where subjects were constrained as to the kind of answers they could provide (i.e., provide distance estimate, provide time estimate). In the present study, subjects were queried during a typical conversation about how far it was to other places. The results indicated an inverted U-shaped curve that suggests that time is used for intermediate distance estimates (331km - 2,080km), and space is used for short (< 10km) and very long distances (8,000km).
Block, in his paper Psychological Time and the Processing of Spatial Information, turns the argument around and looks at how time perception can provide a framework for space perception. He makes important distinctions between position and duration, personal experiences vs. general knowledge, and prospective vs. retrospective. Block proposed a contextualistic model of temporal, and by extension, spatial experience. The model looks at how several variables interact to influence duration and other kinds of temporal experiences, behaviors and judgments. The variables included personal characteristics of the experimenter, contents of time period, activities during the time period and temporal behavior. This model can be extended to space.
Discussant: Waldo Tobler
People moving in geographic space
Two papers in this section studied how people move through space, at different levels of temporal detail. The paper by Stutz et al., Present and Future Diurnal Circulation of Population in a Large City, studied spatio-temporal parameters of population distribution. Travel diaries during a 24 hour period were kept for 2,754 households in San Diego, which recorded factors of trips by time of day and type of trip. These data were used to document the maximum population for census tracts. Stutz's finding is that most people travel at 11:00 AM.
The temporal level of detail and the methodology where different in Odland's paper Longitudinal Analysis of Migration and Mobility Behavior: Investigations of Spatial Choices in an Explicitly Temporal Context. Tobler pointed out a methodological aspect of space-time analysis in that longitudinal modeling contrasts with cross-sectional studies, which are very common in geography. The interaction table that would result from longitudinal modeling would be interesting, and big. The table would include location history of individuals, a cross-sectional treatment of location, the timing of migration (employment status is important for migration).
In terms of travel time, there is a difference between time-space and space-time--there is an asymmetry. This is easy to handle by imposing a vector field. Hägerstrand's work holds time as the third dimension. Constraints of space and time are imposed by society. Such data is expensive to collect.
In the paper Geometric Approaches to the Nexus of Time, Space, and Microprocesses: Implementing a Practical Model for Mundane Socio-Spatial Systems, Pip Forer views space as a series of snapshots. Access is always a function of time and space. In terms of defining action spaces, is it possible to define what could happen and what is likely to happen? Forer noted that Hägerstrand studied peoples' position in time. Can we have a discrete version of time? The raster approach is quite useful. A GIS view focuses on actual time, whereas Forer focuses on what individuals do, what freedom they have and what they can do. Solution is likely a mixture of such models as network and grid models. The aim is to achieve accessibility. What sort of query language should we use?
Movie vs. temporal GIS
GIS tends to represent a static environment. Of course, the environment is not static. When viewing a movie, how often do we only look at one frame? We usually do not. We view each frame in relation to the others and from that extract the plot. Odland replied that movies are an interesting analogy to time-movement in space. To make sense, there has to be continuity in many objects (not too much movement). Space-time snapshots would be interesting in a GIS context; however, in a census framework, for example, the snapshots are 10 years apart. It could be possible to label the change of pixels or objects by their time. This would help understand how many snapshots per time-frame to record.
Route planning
Steve Smyth's paper, A Representation Framework for Route Planning in Space and Time used path finding as an example. The basis of this paper was the methods used by people when determining a route. People were asked how they provided routes, so as to define the rules for route description (i.e., direction giving). These rules were then formulated into a computational model for developing route planning. The concepts were determined from language. In this paper, only one system (GIS) is referred to (i.e., AUTOMAP), therefore it is possibly too narrow in its approach. Tobler indicated that more realistic on-line routing is used with current GPS technology.
Discussant: John Herring
John Herring defined a temporal GIS as a filter and a macro scope that interprets, reduces, integrates, and presents temporally related, geographic data in a human consumable form. A systems that reduces spatial information to a scale that you can deal with. The concept of a paper map is static and limited--due to physical limitations. In a GIS, one can interact with the data--zoom in for more detail if it exists, shift (or pan), or change scales if the resolution of the data allows. Tobler noted that current GISs do not handle movement of objects through space and time very well. It seems there has been little focus on the development of temporal GIS by vendors, possibly because there has been no market for it. But there are probably other constraints as well.
Herring pointed out that the papers he reviewed addressed specific questions. Andrew Frank's paper Different Types of Time in GIS asked, "What is time?" and "How many answers might we find to this question?" The paper by Chris Weber, titled The Representation of Spatio-Temporal Variation in GIA and Cartographic Displays: The Case for Sonification and Auditory Data Representation asks, "How can time be represented in human computer interfaces?" Specifically, Weber's paper explores sonification. What other methods are possible? Nick Chrisman's paper Beyond the Snapshot: Changing the Approach to Change, Error and Process asked, "What is change, and how can it be detected and represented?" John Kelmelis's paper Process Dynamics, Temporal Extent, and Causal Propagation as the Basis for Linking Space and Time, on the other hand asked, "How does change work as a process?"