1Jacques Bertin can be considered as the founder of an important theoretical framework regarding the representation of geographic data (Bertin, 1967/1974). The overall framework is about the translation of characteristics of geographic data into signs or signals that can somehow be perceived by human beings. Bertin's Sémiologie Graphique deals witha part of that framework : the translation into signs that can be visually perceived, and are mainly meant to be used in 2D, static graphics for representation purposes. The influence of this work has been considerable, and it inspired many cartographers and geographers to attempts to adapt and extend the original framework.
2In this paper some of the ongoing efforts to extend and adapt Bertin's original framework will be dealt with first. The scope is limited to maps, other graphics are not considered. The paper then focuses on attempts to extend the framework for dynamic visualization variables in cartographic animations. Research plans of the author and some initial thoughts about those variables will be presented. Finally, a discussion on the extent of cartographic involvement in the representation of geographic data by non-graphic signs or signals is presented, and some conclusions will be given.
3Bertin (1967/1974) deserves credit for the introduction of a graphic sign language in cartography. He proposed seven variables as the building blocks of the language, and syntactic rules to match the variables to characteristics of geographic data. His work needs no further elaboration here, but it is clear that it can be considered as a solid basis for a more extensive framework for the representation of geographic data by signs or signals in maps.
4Extensions to the graphic variables are proposed by several authors. They sometimes consider the variables as composites rather than primitives. MacEachren (1995) mentions the following examples, often provided by other authors. Morrison - in 1974 - suggested to add colour saturation. Although saturation was mentioned by Bertin, he did not separate it from hue. Most graphics software, however, allows individual manipulation of all three aspects of colour : hue, value (or lightness) and saturation. Caivano - in 1990 - found that texture has three aspects, all referring to variation in the texture elements : directionality (ratio of length to width), size and density. Density, as illustrated in figure 1, more resembles Bertin's variable value. MacEachren himself proposed pattern as a graphic variable at a higher level. Patterns consist of elements that have shape, orientation, texture, size and arrangement. Pattern arrangement (or regularity ; figure 2) was originally introduced by Morrison in 1974, and added - together with colour saturation - to Bertin's list of variables. In 1992 MacEachren introduced focus (the sharpness of details) as a composite variable. Later he changed it into clarity (figure 3). It consists of the variables crispness (of symbol edge and/or infill, in figure 3 only illustrated by a vague edge), resolution (or level of abstraction of the base map) and transparency (of an intervening layer, giving an illusion of various degrees of fog).
Figure 1 : Example of the composite visual variable texture, as proposed by Caivano in 1990 (after MacEachren, 1995).
Figure 2 : Example of the visual variable arrangement, as proposed by Morrison in 1974 (after MacEachren, 1995).
Figure 3 : Example of the composite visual variable clarity, as proposed by MacEachren (after MacEachren, 1995).
5Adaptations to the syntax are also suggested. Morrison's nine variables were matched to only two data levels : either nominal or ordinal/interval/ratio. On the other hand, he distinguished three degrees of appropriateness for application of the variables : useable, possible and impossible. The 'possible' category was dropped later, in 1984. Based on his own research and on evaluation of the findings of others, MacEachren's system - as proposed in 1995 - contains twelve variables which can be matched to nominal, ordinal or interval/ratio data with three degrees of appropriateness. In addition, he indicated the capability of the variables to isolate data visually, and to distinguish visual levels (figure 4). In his classification texture seems better suited for nominal than for ordinal data. Perhaps MacEachren's conception of texture deviates somewhat from Bertin's. More surprisingly however, the variable size is indicated as 'good' for the whole range of data, includingnominal level.
6Other researchers have also attempted to adapt elements of the graphic sign language, sometimes for particular applications only. And numerous people have applied it for the representation of data, and even meta data, or data qualityaspects (e.g. Van der Wel et. al., 1994, MacEachren, 1994a).
7So far, reference has been made to sign systems which are basically designed for 2D representation of data. If applied to create 3D maps, the viewer may encounter interpretation problems if the third dimension is simulated on a 2D surface by depth cues. Examples of depth cues are perspective, colour, shading, texture, overlap, etc. A number of them resemble the graphic variables. Kraak (1988) investigated whether the perception properties of the graphic variables still hold in 3D maps on screen. In his experiments he found that depth cues had no positive effect on interpretation speed, they sometimes even slowed down the processing. The effects on interpretation quality varied from positive, via not significant, to negative influence.
8The sign systems mentioned above can be used to represent spatial and temporal aspects of geographic data within the available spatial dimensions of the paper or the screen. In cartographic animations a temporal dimension, display time, can be added. Within display time a number of variables can be applied to control the animation. Several so-called dynamic variables (figure 5) have been distinguished (DiBiase et.al, 1992 ; MacEachren, 1994b).. Attempts have been made to develop a syntax as well, but some questions remain to be answered (see below).
Figure 5 : Illustration of the dynamic visualization variables according to DiBiase et.al. (1992) and MacEachren (1994).
9Characteristics of geographic data can also be translated into signs or signals for other modes of perception than sight. Vasconcellos(1993)tried to define the tactual equivalents of Bertin's graphic variables for visually handicapped people. She concluded that colour (hue/saturation) is the only variable that cannot be applied, but different textures can be used instead :
'...with equal value for qualitative data or with different values to express order' (Vasconcellos, 1993, pp. 997).
10Elevation is an additional tactual variable. Similar syntactic rules can apparently be applied as in the visual domain. Some comments can be made. Texture 'with equal value' may be confused with symbols consisting of elements having different shapes. Furthermore, deformation of the finger skin is fundamental for tactual perception, and the degree and speed of deformation are mainly influenced by elevation, texture and sharpness of the tactual symbols (Jansson, 1972). Elevation is commonly applied to differentiate between point, line, area symbols and Braille. Therefore, its use to express measurement levels of data is limited. In the literature, the tactual variable texture is usually considered to refer to the degree of roughness or coarseness of the surface. Both variables can be used to create contrast between symbols in general, but if they are used to create an ordered range of symbols, they better seem to match Bertin's value than texture, because the ratio of elevated to non-elevated texture elements (in visual terms : black to white) is not constant. Sharpness refers to the transition of elevated symbols to the ground : steep angles are sharper than gradual transitions. Sharpness increases with increasing roughness and coarseness, but it is also one way to create contrast in shapes. Roughness, coarseness and sharpness are not expressed in Vasconcellos' system. The visual language might therefore too literally be translated into a tactual system.
11MacEachren (1995) mentions that sound variables have been proposed by Krygier. Most of them can be considered as sonic equivalents of the graphic and dynamic variables of the visual system. The related syntactic rules (leading to an aural sign language) refer to the representation of geographic data at two levels : nominal and above nominal. Kimerling & Buckley (1997) seem to attempt to even further extend the framework. They are conducting experiments with choropleth maps to establish taste and smell variables (which may lead to the establishment of a gustatory and olfactory language). At the moment these variables can only be used inhard copy maps, but it may well be possible to use smell, and perhaps even taste variables in a computer-generated representation of geographic data in the future.
12The examples mentioned above are mainly limited to Anglo-Dutch research. It is, however, not intended to produce a complete overview of all attempts made to build on Bertin's work in this paper, but to show that the original framework has almost continuously been adapted and extended. This happened under the influence of the same factors which according to Kraak & Ormeling (1996, pp. 2) contribute to the '...changing world of spatial data handling...' : technological, conceptual and user-oriented developments. Technological developments include those in hardware, software and in networks like the Internet. New techniques and applications could be developed based upon these, like 3D mapping, animation etc. Examples of conceptual developments are the recognition of composite, and the definition of dynamic variables. User-oriented developments include the growing awareness within the cartographic community of the distinction between map use for presentation (where the map is supposed to communicate 'knowns' to a wider audience), and for analysis/exploration (where the map is used as a visual tool to stimulate thinking, and the user tries to discover 'unknowns'). Application of cartographic i.e. syntactic rules may be different, although there may also be rules which can be applied to the whole range of map use. MacEachren (1994) suggests that application of visual logic in the translation of data is very important for presentation, while visual contrast seems to be particularly relevant for exploration. Further research in exploratory map use, and in the cognitive aspects involved in visual information processing, is required to gain more insight. In figure 6 the developing framework for the representation of geographic data is presented. The fact that the sign language may work differently for presentation to the 'outside' world and for exploration to gain 'insights' is also represented. In the next section, an example of research in map use for exploration will be given.
Figure 6 : A developing framework for the representation of geographic data.
13Dynamic visualization variables have been distinguished (see above), and are also applied in several experiments. However, some questions remain to be answered, such as : what are their characteristics and relationships, what is the relation with the graphic variables, how can they be successfully applied, and for what ? In an attempt to contribute to the extension of the theoretical framework for the representation of geographic data, research has been initiated in which dynamic visualization variables are used for a particular application : the monitoring of spatial dynamics. Many earth scientists are interested in monitoring. This interest may stem from a wish to understand processes, to interfere in undesired developments, or perhaps to extrapolate to future spatial situations. Monitoring can therefore be used as instrument in various kinds of decision making processes. The perspective taken in the research is that of a domain specialist, like an environmental scientist or planner, who is involved in the processing of spatio-temporal data for monitoring purposes in a GIS-environment. The specialist wants to reveal unknowns, and the visualization goal is exploration.
14Spatial dynamics can be observed from changes in the characteristics of spatial objects in time, and an obvious way to visualize those objects in their spatial and temporal context is through animated maps or images. If a series of slightly different graphics is displayed as frames in an animations at a speed of at least 24-30 frames per second, smooth dynamic transitions are perceived, so users do not have to rely on memory deduction to perceive the dynamics. Animations are therefore very suitable for the display of dynamic phenomena. However, exploring an animation for changes is not easy : many changes may occur in display time, and all over the display area. Therefore, it is likely that a viewer will miss at least some of them. Not seeing in visualizations is a common error (MacEachren & Ganter, 1990). The author wants to investigate whether not seeing the changes in spatio-temporal data can be reduced if the dynamic variables are manipulated - or used as real design instruments - to control the animation in display time.
15First, an attempt will be made to define the dynamic vizualization variables in the context of the research, and to determine their characteristics and relationships. Next, animations will be produced in which the variables are manipulated in several ways, and the ability of those variables to assist domain specialists in the detection of spatial dynamics will be tested. One of the results expected is a methodology for the application of dynamic visualization variables for the monitoring of spatial dynamics. This can be considered as a contribution to the developing framework described above. In addition, recommendations for the design of functionalities (animation tools) that allow domain specialists in a GIS environment to manipulate the variables in ways that prove to be relevant are expected.
16In a next phase, efforts could be directed towards the design and implementation of a prototype with functionalities based on the recommendations mentioned above, including the necessary interface. A prerequisite for this phase, however, is the timely availability of a link between the visualization environment and the GIS data base. It is beyond the scope of the research to cover that aspect, but software vendors may well have succeeded in providing the link by that time.
17The dynamic visualization variables have been defined by MacEchren (1995, pp 281-287) in the following way (see also figure 5) :
display date (or moment of display) : the time at which some display change is initiated ;
order : the sequence of frames or scenes (a scene is a number of sequential frames with no changes) ;
duration : the length of time between two identifiable states ;
frequency : the number of identifiable states per unit time ;
rate of change : the difference in magnitude of change per unit time for each of a sequence of frames or scenes ;
synchronization (or phase correspondence) : the temporal correspondence of two or more time series.
18Synchronization refers to the possibility to run two (or more) temporal animations simultaneously, and shift them in time so that patterns are in phase, and relationships between data sets can be discovered (e.g. the pattern between emission levels of pollution and the occurrence of certain diseases may be similar, but may only become clear if the time lag has been removed). The question arises whether synchronization can be seen as a variable. According to the definition, a variable should have at least two identifiable states. Therefore, this author prefers to consider synchronization as an analysis instrument. Phase can perhaps be considered as a variable (with the states in phase and out of phase), but its characteristics seem to deviate from the other dynamic visuzalization variables in at least two ways : it is a dichotomy, and two (or more) temporal animation have to run simultaneously to apply it. Therefore, it may not be considered in the research mentioned above.
19MacEachren (1995) has suggested a syntax for the dynamic visualization variables (figure 7). In a small experiment, Yaman attempted to link Bertin's perception properties to five dynamic visualization variables, synchronization was excluded (Kobben & Yaman, 1996). The idea was mainly to obtain a first impression and to stimulate discussion. Not considering the question here whether it is correct to assume that the properties of the graphic variables are also exhibited by the dynamic variables, his results have been transformed to allow comparison with MacEachren's suggestions. Bertin's syntactic rules are used to match properties of variables to measurement levels (figure 8).
Figure 7 : MacEachren's syntax proposal for dynamic variables. The variables have been rearranged based on their effectiveness for the different measurement levels (after MacEachren, 1995).
Figure 8 : A transformation of Kobben & Yaman's syntac for dynamic variables (after Kobben & Yaman, 1996).
20However, these schemes cannot be used in the same way as the ones for the graphic variables. The measurement levels in figure 7 do not seem to refer to attributes of the data which have to be represented, but to the effect of the variable. For example, display date means that a phenomena is on (initiation of a change) or off, and this is an effect at nominal scale, so it is indicated as 'good' at nominal level. Duration can be measured at an ordinal or interval/ratio scale, and therefore these levels are indicated as 'good'. If we look at measurement levels of data, however, it seems that the dynamic visualization variables mentioned above can all at least be used for nominal and higher than nominal levels. An example will be given to illustrate this. If changes in the extent of a tropical rainforest area have to be detected from an animated sequence of images, the data are at nominal level (forest or some other type of land use). Display date can be used to show presence or absence of forest at various locations in time ; the temporal sequence of the frames can be reordered to first show changes in forest extent in the wet season over a number of years, followed by the frames of the dry seasons ; and if forest has disappeared in an earlier contiguous area, duration, frequency or rate of change can be applied to emphasize these changes more than those at the fringe. And if data are at a higher than nominal level, e.g. vegetation indices, similar examples can be thought of.
21For the author's research, a link between types of spatio-temporal changes and dynamic visualization variables seems more appropriate.
22The advent of the Internet, with its potential for the dissemination of data via the World Wide Web, allows us to think about the use of signs and signals for the representation of geographic data in cyberspace. Virtual worlds can be experienced through multiple senses. In the most extreme case of fully immersive virtual environments, sight, hearing and touching are currently possible, and in the future it may well include smell. Taste may be more difficult. At the less sophisticated end of the spectrum, Web-pages accessed on a normal PC currently address sight and hearing, but extension to other senses can be expected.
23Cartographers are currently most familiar with the representation of geographic data by signs intended for visual perception, and have limited experience with the use of audio signals. Tactile symbols have only been used in the context of mapping for visually handicapped people, but they may have broader application possibilities in fully immersive virtual worlds. Finally, only a start has been made to address the other two senses.
24The question may arise whether involvement beyond the graphics domain is still the cartographer's business. There seems to be no doubt for touch (at least if used for mapping
25for visually handicapped people), given Taylor's definition of cartography of 1991 :
'the organization, presentation, communication and utilization of geo-information in graphic, digital or tactile form...' and Board's definition of a map (1990) : 'a representation or abstraction of geographic reality. A tool for presenting geographic information in a way that is visual, digital or tactile.' ( as cited in Kraak & Ormeling, 1996, pp. 43).
26Sound is by some cartographers already considered 'at the fringe', and signs or signals for the other two senses are then perhaps considered to be even further away from the cartographic discipline. However, if the core business of cartography is the representation of spatial data, and the cartographic discipline wants to retain its role, the signs and signals for all senses have to be included. And definitions of 'cartography' and 'map' have to be developed that go beyond the ones given above.
27The following conclusion seems justifiable. Bertin's Sémiologie Graphique, published 30 years ago, laid the foundation for a theoretical framework for the representation of geographic data by signs or signals that can be perceived by human beings. The framework is a dynamic construction, it has been - and will be - extended and adapted. The author, for example, wants to contribute further to the dynamic part of the framework, and other extensions can certainly be expected as a response to technological, conceptual and user-oriented developments. It may therefore never be finished. But ongoing development seems to be in the interest of all those who work with geographic data.
I would like to thank my colleague Wim Feringa for providing the illustrations.