Many authors claim that geographic visualization (geovisualization) tools with multiple and linked views facilitate the understanding of a complex reality represented by multivariate data sets (e.g., Edsall 2003a; Jem et al. 2007; Ho et al. 2012). If this claim is true, one should expect a wide use of such tools among stakeholder groups who want to reach a better understanding of a multifaceted reality. However, although coordinated and multiple views have been already broadly discussed in the literature, such tools are still rarely used outside academia, and many practitioners are still not aware that multiple linked views might be useful and that coordination can facilitate solving their real-life tasks (Andrienko and Andrienko 2007). We therefore revisit the general issue on whether multiple-view geovisualization tools are useful in order to understand our world. For this purpose, we use multi-hazard assessment and geovisualization as a case.
Strategies in assessing natural-hazard risks have recently evolved toward approaches that integrate the exposure to several hazards with various aspects of vulnerability (Cutter 1996; Cutter, Mitchell, and Scott 2000; Greiving, Fleischhauer, and Luckenkotter 2006; Tate, Cutter, and Berry 2010; Rod et al. 2012; Rod, Opach, and Neset 2014). As a result, there is a demand for visualization tools that mitigation practitioners and decision-makers can use to make sense of multivariate data on hazard exposure and vulnerability (Tate et al. 2011). From this, a novel application domain has emerged, where highly interactive tools with multiple linked map and data displays might be desirable. On the basis of a study on the functionality of a particular web-based geovisualization tool called 'ViewExposed' (Opach and Rod 2013), our aim with this article is to derive more general conclusions on whether multiple-view geovisualization tools consisting of choropleth maps dynamically linked with a parallel coordinate plot (PCP) may facilitate the understanding of multivariate spatial characteristics. More specifically, we have formulated the following three research questions:
(1) Do choropleth maps linked with parallel coordinates help people understand where the most vulnerable locations are and why these locations are vulnerable?
(2) Do parallel coordinates sparklines (small plots the lines of which replicate the polylines from parallel coordinates) help understand the information provided in a PCP?
(3) Might a multiple-view geovisualization approach be intuitive enough to be useful for both expert users and those unfamiliar with sophisticated geovisualization tools?
The article contains six sections. The first section describes the background on geovisualization tools representing multivariate data with an emphasis on the visualization of natural hazards. Thereafter, we introduce the ViewExposed tool and describe briefly its content and dynamic linking functionality. In this context, we also reflect on whether choropleth maps linked with PCPs work. In the third section, we present the empirical study that was carried out in order to verify theoretical considerations. The fourth section provides the data analysis and results. In the last two sections, we discuss the results and conclude.
Cartographic representation of multivariate spatial characteristics
Traditional cartography has mainly considered maps as a medium for effective communication of facts to a wide public. In this traditional approach, however, there are no truly successful solutions for the representation of multivariate spatial characteristics. The use of maps changed significantly in the 1990s (MacEachren 1994; MacEachren and Kraak 1997; Andrienko and Andrienko 1999), when maps became a tool in geovisualization fostering 'visual thinking' (Kraak 1998). Visual thinking has since become a concept encapsulating hypothesis generation and visual data analysis (Andrienko et al. 2001), marking a shift in the use of maps from communication to exploration. In order to enhance maps' visualization capabilities, various visualization techniques, such as parallel coordinates and interactive manipulation techniques, well known from graphic exploratory data analysis (EDA), have been adapted toward application using georeferenced multidimensional data (e.g., Dykes 1997; Andrienko and Andrienko 1999; Harrower, MacEachren, and Griffin 2000). As a result, the use of tools with multiple linked views for complex geovisualization interfaces such as Descartes (Andrienko and Andrienko 1999), CommonGIS (Andrienko et al. 2002), Geo VISTA Studio (Gahegan et al. 2002), the VIS-STAMP visualization system for space-time and multivariate patterns (Guo et al. 2006), or the generic GeoAnalytics Visualization component toolkit (Jem et al. 2007; Ho et al. 2012) has increased within communities where sophisticated geovisualization approaches for multivariate spatial characteristics are desirable. In these usually highly interactive tools, map displays often equipped with multivariate point symbols such as star plot glyphs (Takatsuka and Gahegan 2002) or 'utility symbols' (Andrienko, Andrienko, and Jankowski 2003) are linked with data displays equipped with EDA visualization and interactive manipulation techniques in order to enhance knowledge acquisition and construction.
It is well known that in order to facilitate knowledge construction, visualization tools need to be interactive (MacEachren and Ganter 1990). Typically, knowledge construction is facilitated by allowing the users to visualize multivariate data in many different ways that are dynamically linked, for example, maps linked with scatterplots, scatterplots linked with PCPs, or choropleth maps linked with PCPs. In the latter example, while a choropleth map is mostly used to represent the values of a single attribute assigned to geographic areas, a PCP is convenient for simultaneous representation of more than two variables (Inselberg 1985). Thus, in such a combination, a PCP may help to get a deeper insight into the data shown on a choropleth map because it offers information unavailable on the map. Furthermore, it might be useful for visual data exploration and mining (Inselberg 1998). Many authors have emphasized the advantages of coordinated and linked views (Roberts 2005), but there is no common consensus that tools with multiple linked views are successful for knowledge acquisition (Andrienko et al. 2001).
Multivariate geovisualization for hazard-related data
In spatial planning or hazard and risk management, hazard maps have proved to be efficient regarding the visual representation of hazard assessments (Kunz and Humi 2008; Kunz, Gret-Regamey, and Humi 2010). However, besides maps presenting exposures to various natural threats, more specific approaches can be found. Vulnerability mapping, for example, gives the users an overview of local adaptive capacities (Cutter 1996; Cutter, Mitchell, and Scott 2000). Integrated vulnerability mapping in turn goes further, because it combines adaptive capacities with exposures to natural hazards (Tate, Cutter, and Berry 2010; Tate et al. 2011; Rod et al. 2012; Rod, Opach, and Neset 2014). In comparison with simple univariate hazard maps, where various threats are shown individually, designing interactive visualization tools for integrated vulnerability assessments is more challenging. Even if vulnerable areas are shown on a map and users are able to identify these places, it is difficult to interpret the information and obtain a proper understanding. Then, the following question arises: how can the multivariate spatial characteristics on integrated vulnerability be successfully represented on a map?
As opposed to hazard maps showing one or two natural threats, to our knowledge, there are few examples in the literature demonstrating the use of geovisualization for multivariate hazard-related data. Hinkel and Klein (2009), for instance, introduce the DIVA tool for assessing vulnerability to sea-level rise on national, regional, and global scales. The tool allows the end users to conduct their own assessment--select data and scenarios, run model simulations, and display the results in the form of a map. Tate et al. (2011) in turn describe a prototype of the Integrated Hazards Mapping Tool for the state of South Carolina. Although the tool is composed of a map view and a tabular display and offers only a basic functionality, its assessment with nine emergency managers and regional planning officials from Georgetown County confirmed that it might be useful in hazard mitigation planning. It has been well tuned toward meeting its predefined goals; therefore, as the authors claim, it is not daunting because it has not been equipped with a plethora of interactive functions.
Both the preceding mentioned projects advocate the use of interactive tools to facilitate the understanding of spatial characteristics of natural hazards. Thus, they fit Chesneau's recommendation (2004) to implement more extensive interactivity and animations into natural-hazard visualization systems. Nevertheless, the functionality offered is rather basic. The tools use maps to display georeferenced hazard data and do not incorporate other visualization techniques to strengthen data exploration. Thus, to our knowledge, there is still no study that supports the claim that coordinated and linked views juxtaposing to map and data displays, such as choropleth maps and PCPs, may facilitate the understanding of multivariate data on natural hazards.
Choropleth maps linked with parallel coordinates
A PCP is a geographic 'multidimensional detective' (Edsall 2003b), which can be integrated into various visualization environments. For instance, PCPs linked with choropleth maps have been applied in many different contexts, with a variety of data content and geovisualization complexity (e.g., Andrienko and Andrienko 2001; Edsall 2003a, 2003b; Robinson et al...