Visualizing Geospatial Information Uncertainty: What We Know and What We Need to Know

Posted: July 5th, 2006 | No Comments »

Alan M. MacEachren and Anthony Robinson and Susan Hopper and Steven Gardner and Robert Murray and Mark Gahegan and Elisabeth Hetzler. “Visualizing Geospatial Information Uncertainty: What We Know and What We Need to Know“. In Cartography and Geographic Information Science,, Vol. 32, No. 3, pp. 139–160, July 2005.

The papers reviews and assess progress toward visual tools and methods to help analysts manage and understand information uncertainty. First the authors note that there is no comprehensive understanding of the parameters that influence successful uncertainty visualization. In turn, without this understanding, effective approaches to visualizing information uncertainty to support real-world geospatial information analysis remain elusive.

Conceptualizing Uncertainty
Uncertainty is an ill-defined concept, and the distinction between it and related concepts such as data quality, reliability, accuracy, and error often remaining ambiguous in the literature. According to Hunter and Goodchild (1993), when inaccuracy is known objectively, it can be expressed as error; when it is not known, the term uncertainty applies. Pang et al (1997) delineated three types of uncertainty related to stages in a visualization pipeline: collection uncertainty due to measurements, and models in the acquisition process, derived uncertainty arising from data transformations, and visualizations uncertainty introduced during the process of data-to-displac mapping.

Decision Making with Uncertainty
The research seems to take for granted that visual depictions of uncertainty are useful for decision making. Tversky and Kahneman (1974) not a conflict: some experts are dependent on statistical analyses to incorporate uncertainty into their decision, but lay users tend to ignore or misinterpret staticstical probalilities and instead rely on less accurate heuristics when making decisions. This divergence prompts 2 questions related to my topic:

  1. will providing information about data uncertainty in an explicit visual way help a lay or expert map reader make different decisions
  2. if they do make different decisions, will provision of information about data uncertainty lead to better, more correct, deicsions or simply cause analysts to discount the unreliable informations

Cliburn et al. (2002) address the idea of making decisions based on uncertain data with the help of uncertainty representations. They list the depiction of uncertainy as a drawback, because policy makers (the users for their study) typically want issues presented with not ambiguity. One participant in their study suggested that a depiction of uncertainty could be used to discredit the models rather than having the intended effect of signaling unbiased results.

Topology of Uncertainty
The literature makes it clear that there are a variety of kinds of uncertainty and to be useful, representations of uncertainty, visual or other, must address this variety. Therefore, there are there have been efforts to delineate the components of information uncertainty and related them specifically to visual representations methods. The earliest conceptual framework for geospatial uncertainty separated error components of value, space, time, consistency and completeness. All the approaches have in common the observation that uncertainty itself occurs at different levels of abstraction.

Most of the efforts to formalize an approach to uncertainty visualization within geovisualization (and GIScience more generally) derive from longterm work on Spatial Data Transfer Standards (SDTS) (Fegeas et al. 1992; Moellering 1994; Morrison 1988). The focus of the initial SDTS effort was on specifying categories of “data quality” which were to be encoded as part of the metadata for car tographic data sets: The categories of data quality defined as part of the SDTS are:

  • Lineage
  • Positional accuracy
  • Attribute accuracy
  • Logical accuracy
  • Logical consistency
  • Completeness

Gahegan and Ehlers (2000) focused on modeling uncertainty within the context of fusing activities between GIS and remote sensing. Their approach matched five types of uncertainty—data/value error/precision, space error/precision, time error/precision, consistency, and completeness—against four models of geographic space: field, image, thematic, and object, as shown below:

Gahegan And Ehlers Uncertainty

From an InfoVis rather than SciVis perspective, Gershon (1998) took a very different approach than Pang, focusing on kinds of “imperfection” in the information about which an analyst or decision maker might need to know. His argument is that imperfect information, while involving uncertainty, is more complex than typically considered from the viewpoint of uncertainty alone.

Gershon Taxonomy Imperfect Knowledge

Building on the typology efforts, Thomson et al. 2004 propose a typology of uncertainty relevant to geospatial information visualization in the context of intelligence analysis:

Thomson Uncertainty2004

Visual Signification of Uncertain Information
The most basic methods of visually representing uncertainty are available through direct application of Bertin’s (1983) visual variables, following guidelines already used in traditional cartography. The original set of variables includes location, size, color value, grain (often mislabeled as texture), color hue, orientation, and shape. In work focusing specifically on uncertainty visualization, Davis and Keller (1997) asserted that using color hue, color value, and “texture” are the “best candidates” for representing uncertain information using static methods. However most of the uncertainty visualization research includes an implicit assumption that users of uncertainty information are homogeneous.

Testing Use and Usability
Very little has been done to empirically evaluate whether the proposed applications work, or whether the theoretical perspectives lead to supportable hypotheses. In subsequent research, MacEachren et al. (1998) tested three methods of representing reliability, i.e., certainty, of health data on choropleth maps and again found that color saturation, counter to their prediction, was less effective for signifying uncertainty than the alternatives tested. Results from Schweizer and Goodchild (1992) , based on user performance on tasks ranging from simple value look-up to overall map comparisons, indicated that reliability information can be added successfully to choropleth maps without inhibiting users’ map-reading ability. Leitner and Buttenfield (2000) went beyond this to consider the impact of different representation methods on map interpretation for decision making. Map detail was found to have limited impact on results, but the maps that depicted uncertainty led to significantly more correct location decisions than those that did not. Response times were similar with and without uncertainty representation, from which the authors conclude that representation of uncertainty acts to clarify mapped information rather than to make the map cluttered or complex.

There is a need for a more systematic approach to understanding:

  • the use of information uncertainty in information analysis and decision making, and
  • the usability of uncertainty representation methods and manipulable interfaces for using those representations.

Discussion
We cannot yet say definitively whether decisions are better if uncertainty is visualized or suppressed, or under what conditions they are better; nor do we understand the impact of uncertainty visualization on the process of analysis or decision making. There is little agreement in the literature about the best way to represent uncertainty. A great number of these methods seem to have potential for displaying attribute certainty on static and dynamic data representations, but only a few of them have been empirically assessed and the results have not been studied in depth,

Challenges
The paper concludes by identifying seven key research challenges in visualizing information uncertainty, particularly as it it applies to decision making and analysis. The ones that relate to my work are:

  • Understanding the components of uncertainty and their relationships to domains, users, and information needs
  • Understanding how knowledge of information uncertainty influences information analysis, decision making, and decision outcomes
  • Understanding how (or whether) uncertainty visualization aids exploratory analysis
  • Developing methods for capturing and encoding analysts’ or decision makers’ uncertainty
  • Assessing the usability and utility of uncertainty capture, representation, and interaction methods and tools

Key references are:
Gahegan, M., and M. Ehlers. 2000. A framework for the modelling of uncertainty between remote sensing and geographic information. ISPRS Journal of Photogrammetry and Remote Sensing 55(3):176-88.

Gershon, N. D. 1998. Visualization of an imperfect world. Computer Graphics and Applications (IEEE) 18(4): 43-5.

Relation to my thesis: I am interested in methods to help users of location-aware system manage and understand information uncertainty. Methods can be to design adaptable core system or like in this case appropriate uncertain information visualization. This paper first acknowledges that there is no comprehensive understanding of the parameters that influence successful uncertainty visualization, and actually the approaches to support real-world geospatial information analysis remain elusive. The authors call for more systematic approach to understanding the usability of uncertainty representation methods and manipulable interface for using those representations. This is an area I might want to cover. A research question could be to know if decisions are better if location uncertainty is visualized or suppressed, and under what conditions they are better. The challenges mentioned by the authors are also part of my research domain such as “understanding the components of information uncertainty and their relationships to domains, users, and information needs and assessing the usability and utility of uncertainty capture, representation, and interaction methods and tools.

Finally, I am very interested in the different topologies of uncertainty, as I have already tried to sketch in the past. The approach by Thomson et al. 2004 is inspiring. Part of my work is to define what (spatial) uncertainty is in the context of ubiquitous computing.