maceachren@psu.edu; www.GeoVISTA.psu.edu

kraak@itc.nl

E.Verbree@geo.tudelft.nl

Introduction

In this paper, we consider the integration of cartographic and virtual reality (VR) methods for representing geospatial information B within geospatial virtual environments (GeoVE). We define a GeoVE as any virtual environment used to represent geospatial information (either measured or simulated). Thus GeoVEs are virtual environments that represent characteristics of the world (or possible worlds) at scales from the experiential (e.g., a neighborhood) to the global.

Integration of cartography and VR within GeoVE offers exciting possibilities for enhancing the understanding of our world. At the same time, this integration generates a tension between maps as abstractions of reality and VR as simulations of (or replacements for) reality, thus as iconic Asign vehicles@ for aspects of the real world (or for worlds we can only imagine). Success in cartographic design for GeoVE is likely to be a function of our ability to capitalize on this tension (melding advantages from both perspectives), rather than allowing the tension to result in competition among environment components and, thus, confusion on the part of users. Our goal in this short paper is to outline some key issues to be considered in a research program directed to cartographically enhanced virtual environments and to use some examples from our own work as a base for discussing these issues.

Virtual Reality is a widely used term with a broad range of interpretations (Heim 1998). The common thread that seems to link developments labeled VR is that VR applications involve computer-generated three-dimensional representations supporting dynamic change in user viewpoint. Viewpoint manipulation, as one kind of interactivity, corresponds to one of the four AI@s of GeoVR proposed by MacEachren, et al., (1999) -- who adapted and extended the Athree I=s of VR.@ first proposed by Heim, (1998). The four AI@s are defined as:

Here, with emphasis on the first two factors above, we distinguish two primary categories of virtual environment: (a) viewable/manipulable 3D worlds experienced from the outside (generally non-immersive with only manual control of viewpoint B such as web-based VRML, see: Brutzman 1998) and (b) experiential 3D worlds that users can enter and explore from the inside (generally at least semi-immersive with both manual and automatic viewpoint control, the latter using head, eye, or body tracking B such as those achieved through virtual workbenches, head-mounted displays/HMD or CAVEs, see: Cruz-Neira, Sandin, and DeFanti 1993). To simplify discussion below, we use non-immersive and immersive as shorthand labels for these categories. [While we simplify discussion here by distinguishing two categories and emphasizing one factor in these labels, we recognize that a full typology of GeoVE will be much more complex that this dichotomy implies B involving all four factors identified above with each being a composite of several subfactors. For example, immersion has, at a minimum, a separate component representing each human sense.]

Non-immersive B viewable/manipulable (external) virtual worlds

With respect to 3D map-based visualization, Kraak (1994) described the functionality required for an interactive modelling environment for 3D maps. Some of the functions called for by Kraak (1994) (and implemented a few years ago with custom workstation-based routines) can be found in the GeoVRML worlds that are used currently to disseminate three-dimensional representations over the World Wide Web (see MacEachren, 1998). Fairbairn and Parsley (1997) describe an integration of CAD, cartography, and VRML to produce a representation of the built environment. The project resulted in a model of their university campus through which users can explore the campus layout as well as enter virtual buildings. Dykes, et al, (in press) focus on integration of cartography, GIS, and VRML methods for web-based dynamic representation of the natural environment B as part of their efforts to design a virtual field course for study of physical geography.

Immersion is itself a multifaceted factor in VR (as noted above). It can be stimulated by a range of perceptual and kinesthetic cues, including: stereoscopic display, sound, real time interaction, and direct coupling of viewing position with the image on the display. An HMD provides a sense of immersion while effectively isolating users from the real environment and from other individuals, making communication with other users difficult. A better approach (particularly for multi-user environments) may be the CAVE (Cruz-Neira, Sandin, and DeFanti 1993). This is a multiple screen projection display system that offers stereoscopic surround projection to several users simultaneously. Below, we describe use of this kind of environment. Immersive VR gives the user the illusion of being part of the projected geo-world. As a result, the illusion of 'immersiveness' also depends of the the actual position of the user in the GeoVE. For a full sense of immersion, the user must obtain the impression of being in a natural position with respect to the scene (e.g. 1.80 m above street level in a simulation of the built environment). In their discussion of GeoVE developments, Neves and Camara (1999) contend that the ability to >feel= results of particular actions, achieved through immersion, leads to a clearer and more natural understanding of those actions.

In an immersive GeoVE that represents the large-scale built environment, interaction with the scene will usually be limited to navigating and simple queries. At this scale, manipulation of objects is not expected to be useful (e.g., being able to pick up and rotate a building that appears to be real-world size is not an intuitive operation). In contrast, when a GeoVE is used to represent small-scale natural environments, the position of the user can be compared with a bird=s-eye view. In this case, the built-up environment will be abstracted to simple 3D-objects that resemble toy replicas situated on or above the abstracted DTM or TIN of the surface. Research has shown that when natural environments are represented from the bird=s eye perspective, a full range of cartographic tools and techniques can be used to give the user a better idea about the presented information (Kraak 1994). Interactivity within a virtual environment should make it possible, not only to manipulate the scene viewpoint and other parameters, but also to manipulate objects within the scene and the data-to-display mappings that determine presence and appearance of these objects. Such manipulable toy world style representations should Aafford@ actions applied to objects in them, such as being picked up, turned around, and moved. We anticipate that semi-immersive Virtual Workbenches and ImmersaDesks have advantages over non-immersive GeoVRML or fully immersive CAVEs for this kind of visualization and interaction.

Here we propose a distinction between two fundamental categories of GeoVE application: (a) GeoVE that simulates the tangible world (a goal of which is to allow users to exerience places distant in space and/or time -- e.g., students in the U.S. could experience a Brazilian rainforest environment or urban planners could experience potential implications of their planning decisions), and (b) GeoVE to represent non-tangible (non-visible) aspects of the world B thus an environment to enhance cartographic abstractions (a goal of which is to allow users to explore complex multivariate relationships in geospatial information B e.g., in a context such as integrated regional assessment in which economic, social, and environmental factors must be integrated).

The terms virtual reality an virtual environment both suggest a simulation of real environments. The prototypical virtual environment for many people may be the fictional Aholodeck@ from the Star Trek television show and movies. Typical Star Trek holodeck applications fit the definition of GeoVE provided above). The goal in design of such an environment (whether geographic or architectural in scale) is to simulate, as accurately as possible, characteristics of tangible objects in the real world, or that might exist in some real, or imagined, world (e.g., (Bishop and Karadaglis 1994)). Ultimate success in design would be a virtual environment so compelling that users could not tell it apart from the real thing.

Figure 1. Views associated with the three stages in the planning process
http://karma.geo.tudelft.nl/.

An initial category of abstract GeoVE involves use of space to represent space, time to represent time, and graphic variables to represent attributes abstractly (a category that corresponds to typical 2D animated maps of temporal information B such as animation of disease diffusion, see: MacEachren and DiBiase 1991). An example of this category of GeoVE would be a VRML world in which the three spatial dimensions are used to depict space (e.g., in the form of a 3D terrain representation), display time is used to depict real world time through which change occurs (e.g., in land use across that terrain), and color hue is used to depict each land use. The result is a 3D (or 2.5D) dynamic thematic map that is enhanced by interactivity (particularly in relation to user control of viewpoint), and intelligence of display objects (that change their level of detail or provide additional information when the user gets close to them B thus a map enhanced through application of two of the factors in virtuality noted above). GeoVRML used in this way corresponds to the Amodel view,@ described above for use in depicting the tangible environment. An example corresponding to the Aworld view@ is development of flow visualization methods implemented in immersive or semi-immersive environments, (e.g., the Virtual Chesapeake Bay, a semi-immersive virtual environment for understanding interaction between hydrologic and biotic systems in the Bay, Wheless et al. 1996).

Between the two extremes detailed above, are many combinations of spatial, temporal, and attribute abstractness/iconicity that can be used in a GeoVE (space does not allow us to develop the full typology here). One combination that we have had some success with uses two spatial dimensions of the GeoVE to represent geographic location (latitude and longitude) and the third to represent both geographic location (elevation) and time. This is accomplished by mapping elevation to a small subset of the z-axis, with time mapped (linearly) to the remainder of this axis. Into this space-time cube, attributes are mapped using a variety of abstract representation forms that include use of color and location (the latter in the form of isosurfaces). One of our research groups (the Apoala Project within the GeoVISTA Center at Penn State) has implemented this form of representation in a semi-immersive ImmersaDesk environment (figure 2). [An ImmersaDesk uses a large format screen, 3D projection, and head tracking of the Adriver@ to provide users with a limited sense of being Ain@ the environment and its size allows small groups to use the system for same-time same-place collaboration.]

Figure 2. A scene from a demonstration project in which one of our research teams (the Apoala Project B see acknowledgments for url) collaborated with researchers at Old Dominion University to develop a same-time different-place visual exploration of two environmental data sets using linked IDesks. The demo was part of an AInternet2 Day@ on the Penn State University Campus in November, 1998 B an activity designed to illustrate the potential of high speed Internet connections for supporting science and education (see: MacEachren et al. 1998). The data used in the Penn State component of the demonstration, shown here, are extracted from a much larger climate data set for the Susquehanna River Basin of Pennsylvania, New York, and Maryland -- specifically daily maximum temperature and precipitation extending from May through July, 1972. Among the features that the resulting dynamic environment highlights are the relationship of temperature with both topography and precipitation B one of the more dramatic is substantially reduced temperature across the basin following Hurricane Agnes, as the huge quantities of water dumped on the region slowly evaporated.

To interact with virtual environments requires a good set of interface tools. One key distinction here is between those tools needed to navigate the environment and those needed to work in that environment. The first category should include options to move through the virtual environment by manipulating the user viewpoint, thus tools to allow 'walking,' 'flying,' >teleporting,= etc. As a complement to tools that enable movement, tools that facilitate orientation are also important. It is relatively easy to get lost in three-dimensional environments. Different options for orientation aids are possible. A multiple view approach were the user can see the environment from different perspectives seems particularly promising.

GeoVirtual Environments pose a major research challenge for cartography B as well as a challenge to the traditional map as a form of representation and communication. We believe that a key component of this challenge is to find ways to productively integrate abstract and iconic forms of representation.

Parts of this work were carried out within the Apoala Project, supported by: the U.S. Environmental Protection Agency (grant # R825195-01-0 B Donna J. Peuquet and Alan M. MacEachren, Co-PIs) (www.geog.psu.edu/apoala). Support of the Penn State Center for Academic Computing is also appreciated. We also would like to acknowledge contributions to examples described here made by: Robert Edsall, Daniel Haug, George Otto, Raymon Masters, and Liujian Qian at Penn State and Gert van Maren, Rick Germs, and Frederik Jansen at Delft University.