Guiding Users in the Visualization of Geographic Data

Volker Jung
Darmstadt Technical University, Interactive Graphics Systems Group, Wilhelminenstra*e 7, 64283 Darmstadt, Germany

One of the most important components of Geographic Information Systems (GIS) is visualization. This is because it can help users to understand complex spatial relationships as well as being the most efficient medium to communicate them. Today, a large number of people use GIS Visualization to produce thematic maps and graphic presentations, but not all of these users have experience in graphic design. Consequently, an increasing number of graphical presentations are being produced that, while looking professional and convincing, do not really provide insight into the data. The GIS Visulization Component has been characterised as function-rich and knowledge-absent, and it has been repeatedly suggested that this gap can be closed by providing knowledge-based assistants that guide and advise non-experts in the visualization of geographic information.

A number of prototypes for such assistance systems have been built recently, but there is still a long way to go before they will be widely used as part of a general-purpose GIS. One of the restrictions of these prototypes is that they cover only a (varying) part of the graphic design process. This is most likely an inherent problem, since a truly generic, automated system would require an immense effort to formalise the complete design process (symbolisation, generalisation, etc.) and to acquire knowledge about all application domains. In many practical GIS applications the visualization design space is much more limited, though, because only certain visualization methods are appropriate or because the scale range is restricted. In these cases, assistance systems can be particularly helpful even today.

Shortcomings of current visualization assistants
Apart from their limited visualization design space, current systems have other shortcomings that severely hinder their acceptance by users:

  1. They do not consider important visualization conditions. Visualization design must not be based on the data set characteristics alone. Other conditions such as the visualization goals, the output medium characteristics and of the characteristics of the visualization audience can drastically influence the effectiveness of a design.
  2. They operate under a "closed world" assumption. Current systems behave as if they had all the knowledge of the world at their disposal and consequently could find the only valid or most effective design in any particular case. Therefore, they do not offer alternative designs or let users modify proposed ones. In reality, application knowledge can only be partially formalised and users often need to adapt the designs made by the system.
  3. They do not explain design decisions. Experience with expert systems in many application areas has shown that users mistrust the decisions of the system if they are not made transparent. Users want to know what conditions have led to design decisions and what rules have been applied in the design process. Only if they understand why the system makes specific choices are they ready to accept them and also learn from them.

VIZARD: a user guidance system
The purpose of the research presented here is to create a paradigm for a new type of user guidance system meeting these requirements. VIZARD is an operational prototype of such a guidance system. It designs visualizations of spatially-referenced data, taking into consideration all important visualization conditions. VIZARD designs various alternative visualizations of data sets, evaluates their effectiveness, and presents them to the user in the estimated order of effectiveness. Users can browse through alternative designs, select appropriate ones and also adapt them to specific requirements by editing their features. VIZARD explains to users why a particular design is expressive for the data set and what conditions are favourable or unfavourable to its effectiveness.

Knowledge-based visualization design. The visualization design methodology decomposes data sets into simple relations that can be expressed using primitive visualization methods (e.g. thematic maps, statistical graphs). Partial visualizations are repeatedly composed using a set of composition operations (e.g. overlay, animation) until all complete visualizations of the data set have been found. In each of these steps, design knowledge is used, e.g. expressiveness rules for each of the primitive visualization methods and composition rules for all composition operations.

Fuzzy effectiveness evaluation. Subsequently, all visualization designs found to be expressive for the data set are evaluated according to their effectiveness. The evaluation scheme assigns fuzzy effectiveness ratings to the designs based on a set of effectiveness rules for the various visualization methods and conditions. Fuzzy ratings are used here because they represent the whole range of effective and ineffective aspects of a design, as opposed to 'crisp' ratings which express only averages. In addition, fuzzy sets are a well-known way of representing linguistic variables, allowing the formulation of pragmatic effectiveness rules.

Presenting and explaining designs. Previews of those designs that have a sufficient degree of effectiveness are presented in a graphical browser. Here users can compare alternative designs and select the one that matches their requirements best. An explanation component generates reports, describing why a specific visualization design is expressive and what aspects of the design are effective or ineffective. These reports are hyperdocuments, containing links from the textural explanations to an on-line tutorial on graphic design that explains the visualization methods and design rules in depth. By using these explanations together with the tutorial, users gradually become aware of design principles and, consequently, increase their graphic literacy.

Modifying designs and automatic verification. Users can always adapt the proposed visualization designs by editing parameters such as the data classification or the visual variables and features used. VIZARD's 'design critique' component verifies that modified designs are effective and do not violate well-known graphic principles.

Implementation. VIZARD's knowledge base, together with the design and evaluation algorithms, are implemented in the CLIPS expert system shell. The knowledge base consists of about 200 rules covering 11 visualization methods. Only two-dimensional methods (thematic maps, statistical graphs and graphic scripts) are supported by the prototype. Graphical user interface, visualization rendering and geo-data processing are implemented using C in a UNIX and X/Motif environment, while NCSA's Mosaic serves as a browser for hypertext explanations and the on-line tutorial.