A typical GIS today is a general-purpose tool utilized by a skilled operator. What it lacks in capacity for automatic generalization is often made up for by the specialized skills of the user, and the audience for the output. GIS maps by their nature are usually designed to answer very specific questions, and familiarity with the issues at hand can often overcome suboptimal cartography.

When GIS is introduced to mass audiences, however, the equations change. No longer can the operators rely on the familiarity of the audience with a given subject to interpret maps that may be difficult to understand. The cognitive load must be reduced some other way.

One method is to make a system narrow in purpose, and incorporate a specially-designed dataset. This is the approach that is used for GIS-like devices such as car navigation systems. Given a highly circumscribed set of operations (i.e. just scale change) and some hierarchical rules for structuring and displaying data, these displays can be useful navigational aids. The more functions these systems offer, however, the more likely it is that the cartographic objectives of the designer will be lost in the output. A dynamic means of controlling the appearance of the map display is required.

An example of a cognitive question about these displays that has been well addressed by research is how they should be oriented for the user to most effectively extract navigational information (i.e. Aretz 1992). This is a question that is fundamentally different for electronic displays than for paper maps, since electronic displays offer almost infinite possibilities for changing the display on the fly. The general answer to this cognitive question is that they should be oriented track-up, in the direction the user is traveling.

Another point that must be taken into account is the readability of an electronic display compared to a paper map. Gooding and Forrest (1990) compared the comprehension of paper maps with the same map stored as an image on a video disc. They found that performance with paper maps was significantly better than with video screens displaying images of the same maps. While this particular example presents just one dimension of the paper map vs. screen question, it points to the fact that significantly different cartographic design parameters are needed for video displays.

A more fundamental cognitive question about map design is how much information should be presented on the page. That is because access and processing time are related to the complexity and clutter on the map, although not always in a straightforward way. There is always tension between the goals of presenting complete information while not obscuring pertinent objects with unnecessary detail. Most navigation maps are designed to err on the side of presenting complete information at the expense of accessibility because they are often the sole resource for navigation. While it may take considerable time and attention to extract needed spatial information, that is judged preferable to faster access. Some notable exceptions to this design tendency are in aeronautical charts, particularly those designed for instrument approach procedures. That is because they are specifically designed for one phase of flight where the pilot is subject to high workload and other demands on his attention.

A similar environment exists inside an automobile on an urban expressway. Car navigation systems that display route information on a screen can reduce the cognitive load on a driver by solving the where-am-I problem (via GPS or some other mechanism) and displaying route information that helps the driver make important navigational choices. Too much detail, however, can make such a display an expensive drain on attention to interpret. That can limit its usefulness, or worse, take away needed attention from safe driving.

Since a car navigation device or other specialized GIS must display many different maps, it is impractical to view them all and adjust cartographic parameters for each before it is displayed to the driver. Instead, some kind of prediction must be made as to the appropriate map information to be displayed before it is actually rendered. This is a specific part of the map generalization problem which has been termed "model" generalization (McMaster & Shea 1992). Data is reduced in volume per unit geographic area.

This reduction is apparent in the way map series of multiple scales are produced. The number of streets depicted per unit area on an Interstate Highway map of the United States is some small fraction of those shown on a state highway map, and in turn of those on a county map. This hierarchical arrangement is not only a cartographic convenience, however; it may also reflect how we navigate and think about space.

People use hierarchical structures in order to improve performance and reduce cognitive loads in solving spatial problems (Car & Frank, 1994). Ideally, we would have explicit knowledge of the elements represented in those hierarchical levels, and create our map displays accordingly. Failing that, we can attempt to tailor them so they are both interpretable and congruent with our current need for spatial information.

One means of measuring the information content of a map has been to use information theory. This is based on the idea presented by Shannon and Weaver in The Mathematical Theory of Communication in 1949. But while this idea has had great success in other applications, it has generally failed in cartography. The basic objection has been that maps contain more information than is represented by the sum of their parts; the spatial relations between objects add complexity that cannot be accounted for by simply adding up the number of elements displayed. (It is notable, however, that some attempts have been made to overcome this objection. Neumann, 1994, is a good example.)

Even if it were possible, however, to measure the information content of a map by counting objects, there are other obstacles to using this as a measure of clutter. The emergent perceptual properties of groups of objects can make a large number appear to be a simple figure. The Gestalt psychologists described a number of properties of visual grouping, but none were applicable to quantitative measures. Monmonier (1974) reviewed the work of psychological measures of complexity, and developed and compared different measures of complexity. He came up with three basic types of complexity measures; pattern fragmentation and autocorrelation, size inequality among map regions, and spatial trend. None of these measures, however, seems to help in developing criteria for navigational maps, since they were designed for choropleth maps.

MacEachren (1982) examined the complexity of both choropleth and isopleth maps. He points out that map complexity may have an adverse effect on map effectiveness, and points out that this relationship is an area for future research. In this particular work, he makes a distinction between denumerable measures of complexity that deal with individual features, and structural measures that focus on relationships between parts. While they both clearly have an impact on perception and cognition, the denumerable measures offer the most promise for database selection approaches to the problem.

A comparison of different measures of complexity was undertaken by Bregt and Wopereis (1990), also on choropleth maps. Like MacEachren, they compared mathematical measures with subjective judgments made by human subjects. Their conclusion was that a fragmentation index was the best measure for choropleth maps, because of its reliability and simplicity. This, however, is a structural measure that offers little promise for guiding feature selection.

Another tack taken in the search for measures of complexity has been in eye-movement studies. Castner and Eastman (1984-85) studied perceived map complexity by this method. But while they found some correlation between eye movement parameters and subject reports of perceived complexity, they were not able to come up with a quantitative measure that could be used by cartographers as a guideline. They did, however, discuss how functional complexity is related to the map, the map user, and the environment in which the map is perceived. From that they determined the perceived complexity as reported by users in a given situation.

A more direct approach to investigating the optimal number of symbols on a map was taken by Phillips and Noyes (1982). They investigated topographic symbols on geological maps, and found that there was a clear improvement in performance with a reduction in visual clutter. Defining clutter, of course, is one of the central difficulties in designing a study of this type. They found that perceived clutter operated quite separately for lines and points within the same map, and that there was greatest interference between symbols of the same color. This conclusion is reminiscent of the work of Anne Triesman, a psychologist who has published on the perception of groups and the relative effects of different shapes and colors.

In a similar vein, Lloyd (1988) wrote on the cognitive processes involved in searching for map symbols. He determined that searching for map symbols on a display is a serial process, and that the search time was roughly proportional to the number of symbols on the map to be considered. This result supports the idea that a navigational map would be more useful under time pressure if the number of symbols is reduced from the maximal set that will fit on the map, to an optimal set that is the minimum necessary to include the critical features.

Despite this lack of quantitative measures of appropriate detail and complexity, however, cartographic practice indicates that complexity and clutter matter. Different map series have different levels of complexity and symbol density depending on the kind of navigation problems they are designed to solve. Freeway system maps have relatively sparse representations of street networks, partly because freeway systems are themselves relatively simple, and because they are designed to be used in cars when there is little attention safely available to devote to map search and comprehension. Instrument approach plates for aircraft navigation are also highly simplified representations designed for use during a specific phase of flight. For more general navigation and planning, both street maps and aeronautical charts show an increased degree of density and complexity, in part because of their additional role as repositories of spatial data that are designed for planning as well as execution of navigational tasks.

In an experiment on aircraft pilots, Williams (1996) studied how much scene detail contributed to performance on a navigational task. While the degree of detail was controlled on a 3-D perspective display rather than a map, the results may have some relevance to this question. There was no significant difference in performance between those pilots presented a low vs. high level of detail in training. While the authors caution against generalizing from their study, it is additional evidence supporting the thesis that more is not always better.

Oviatt (1996) describes a system for map display that starts with a minimal representation of an area for real estate information. She describes how users can add information through a multimodal query interface (voice as well as screen pointers) to display. This differs significantly from the typical process of abstracting information from maps, where the user must search through large amounts of information to find the points of interest.

Another system has been demonstrated by Arikawa (1991) that uses a multi-window system where different scale maps are presented at the same time. Users can query a reference map by area and see a more detailed version simultaneously. This system adds detail according to a set of importance rules that rank features both with and between feature type. In addition, label type and placement are dynamically controlled to prevent overlap.

These efforts are considerably different than traditional approaches to automated cartographic generalization. They assume the user is not a cartographer or mapmaker, and that time is at a premium. They use rules to predict what might be the most appropriate information to put on the map. In some respects, this might be considered an outgrowth of the idea of amplified intelligence for cartographic generalization described by Weibel (1991) and others. The questions that are raised in the construction of a map that predicts what the user needs now, however, are cognitive and ergonomic as well as cartographic.

As MacEachren (1991) suggests, it will be intriguing to determine the influence that varied views have on problem solving efficiency. One way to approach this problem is to let users themselves manipulate interactive maps to their own satisfaction for the performance of navigational tasks. While this may give us insight into general preferences and cognitive processes, it also opens the possibility for mapping software to be tuned to individual users. Both the construction of tools for such experiments and the analysis of results may yield insights into how to best select and display spatial information for navigation and other tasks.

References:

Aretz, A. J., Wickens, C. D. 1992. The mental rotation of map displays. Human Performance. 5(4).

Arikawa, M., Kambayashi, Y. 1991. Dynamic name placement functions for interactive map systems. Australian Computer Journal. 23(4):133-147.

Bregt, A. K., Wopereis, M. C. S. 1990. Comparison of complexity measures for choropleth maps. The Cartographic Journal. 27:85-91.

Car, A., Frank, A. 1994. General principles of hierarchical spatial reasoning - the case of wayfinding. Proceedings of SDH: Scotland, September 1994.

Castner, H. W., Eastman, J.R. 1984. Eye-movement parameters and perceived map complexity-I. The American Cartographer. 11:2.

Castner, H. W., Eastman, J.R. 1985. Eye-movement parameters and perceived map complexity-II. The American Cartographer. 11:2.

Goodchild, M. 1988. Stepping over the line: Technological constraints and the new cartography. The American Cartographer 15(3):311-319.

Gooding, K., Forrest, D. 1990. An examination of the difference between the interpretation of screen based and printed maps. The Cartographic Journal. 27:15-19.

MacEachren, A. M. 1995. How maps work. New York:Guilford.

MacEachren, A. M. 1991. The role of maps in spatial knowledge acquisition. The Cartographic Journal. 28:152-162.

MacEachren, A. M. 1982. Map complexity: comparison and measurement. The American Cartographer. 9:1.

Martel, A., Ward, A. W. ???? Ergonomic development of digital map displays. ? (Authors employed by British Aerospace Defence)

McMaster, R.B., Shea, K.S. 1992. Generalization in digital cartography. Washington, D.C.:Association of American geographers.

Monmonier, M. 1974. Measures of complexity for choroplethic maps. The American Cartographer. 1:2

Neumann, J. 1994. The topological information content of a map / an attempt at a rehabilitation of information theory in cartography. Cartographica. 31:1

Oviatt, S. 1996. Multimodal interfaces for dynamic interactive maps. Proceedings of CHI 96 Vancouver BC Canada.

Phillips, R. J., Noyes, L. 1982. An investigation of visual clutter in the topographic base of a geological map. The Cartographic Journal. 19:2.

Weibel, R. 1991. Amplified intelligence and rule-based systems. In Map Generalization, ed. Buttenfield, B., McMaster, R. B. London:Longman.

Williams, H. P. et al. 1996. A comparison of methods for promoting geographic knowledge in simulated aircraft navigation. Human Factors. 38:1