It is well recognized that the world we live in is a dynamic place where change and movement over space and through time is the norm. Indeed, enhancing our ability to decipher complex dynamics in human and environmental systems is crucial to our understanding of the world in general (Yuan and Hornsby 2008).
Dynamic GIS can certainly be applied to representing changing environments, but layered upon that, the spatial behavior of the sentient "denizens" of our changing environment is increasingly a factor in understanding the world and its interactions. It is the community of sentient objects (SOs), animals and people, that is the wider focus of this research. SOs are complex, influential, and dynamic actors in our biosphere, and their movements and activities have profound impacts on our environment. Therefore, the movement of sentient objects (MoSOs) requires special attention. SOs are capable of thinking; they have desires that lead them to conduct various activities to reach their goals and their decision-making has an immediate relationship with the environment. These factors generate additional layers of complexity for representing the MoSOs. However, in other ways, the question is simplified because of the crisp, unambiguous, and persistent physical nature of individual organisms that allows SOs to be modeled as moving point objects, which are more straightforward than vague linear or areal objects.
The MoSOs consists of three main dimensions - space, time, and activity (s,t,a)--and each dimension opens to various properties and expressions. When representing and visualizing sentient movement, space, time, and activity should be considered together. However, activity is uniquely located with SOs; it is a privileged dimension which drives movement and is the key to comprehending movement patterns (Zhao, Forer, and Harvey 2008).
Interactive and dynamic visualization has been widely recognized as an essential and thought-provoking tool for understanding dynamic phenomena (Andrienko and Andrienko 2007b; Kraak 2008), since "the human eye and brain can scan and interpret information on a map more rapidly than the most elegant software can process the equivalent digital data" (Goodchild 1988, 311). Over recent decades, movement data, in large volumes and with "seductive" precision, are increasingly available and affordable. This has opened up numerous research topics and enabled the development of sophisticated data mining and pattern recognition techniques that identify patterns and anomalies within many movement fields (Forer and Zhao 2011). However, our inability to explore and visualize massive movement data-sets is pronounced. It is necessary to advance visualization approaches and to integrate visualization with geocomputational techniques, such as data transformations, generalization, and other computer-based operations, in order to extract knowledge from such movement data (Andrienko and Andrienko 2007b; Dykes, MacEachren, and Kraak 2005).
Movement is a scale-dependent phenomenon, with an inherent hierarchical structure. According to hierarchy theory, higher-level processes impose a great influence on movement systems and provide contexts, boundaries, and constraints for lower-level processes, which in turn provide the mechanisms, materials, and energy to sustain the higher-level processes. Therefore, in order to fully comprehend movement at one scale, it is necessary to analyze it at three scales: the focal scale, one scale higher, and one scale lower. This approach takes into consideration the contexts that afford and constrain the sustainability of the movement (Yuan and Hornsby 2008, 28). Those levels can be identified by spatial, temporal, and activity scales, so generalizing patterns of the entire three dimensions of sentient movement data at various scales promises a better comprehension of such data.
Space, time, and activity are interwoven in the movement process. Given the richness and complexity of (s,t,a) movement data and its scale dependency, it is almost impossible to effectively visualize all the dimensions of movement, their relationships and interactions, and their generalized representations at various scales, in any single representation. Therefore, coordinated and multiple-view (CMV) techniques are needed to explore and visualize (s,t, a) movement data. CMV can provide richer, reinforced, and complementary information and insights (Hangourt 2004) and can facilitate an understanding of relationships between different representations and comparisons between views (Aoyama et al. 2007). Cognitively, multiple views can be a powerful aid in visualization, taking advantage of human abilities to discover patterns and filter information and allowing more intuitive manipulation (Green, Ribarsky, and Fisher 2009, 3). Multiple views need to be linked and coordinated (Carr 1999) so that the information displayed in each representation can be visualized interactively and integrated into a coherent image of the data as a whole, since the whole is greater than the sum of its components (Buja et al. 1991).
This research anticipates that the deliberate integration of space, time, and activity through coupled generalization and CMV visualization of sentient movement data will provide deeper and more powerful insights into our understanding of the MoSOs. The next section focuses on exploring CMV visualizations of (s,t,a) tourist movement data based on a set of individual timelines and results derived from a variety of (s,t,a) aggregation techniques. Detailed (s,t,a) aggregation techniques, which can be seen in works by Zhao, Forer, and Harvey (2008) and Zhao (2012), will not be discussed in this article.
Visualizing (s,t,a) tourist movement data with coordinated multiple views
The West Coast Tourism Flow Survey (WCTFS)
The main tourist data-set used in this article is a significant subset of the West Coast Tourism Flow Survey conducted from December 1999 to June 2001 on the West Coast of the South Island of New Zealand (called the West Coast hereafter). As Figure 2 best indicates, the geography of the populated West Coast is long and thin with three narrow entry and exit portals, namely Haast, Arthur's Pass, and Murchison (see Figure i). Using these portals as control points for survey sampling, some 2630 domestic and international tourists participated in the survey, providing data through diary or recall survey instruments. Information was collected on their itineraries (including times of stopping and overnighting, the purpose and location of any stops over 5 minutes, and activities at stops), time use, expenditure, and other profile attributes (Forer, Fairweather, and Simmons 2003). The data-set is characterized by its rich information about the process of being a tourist. Such genuine itinerary data are rare to find and very valuable in identifying the impacts associated with tourism at a local scale as well as a sound base for visualizing the time geography of tourist movement and activity (Forer 2005).
The visualizations that follow are largely based on a 500-respondent subsample of the full set of survey respondents, namely those who had chosen to spend 3 days on the West Coast and whose records were complete or whose gaps in recording were enhanceable by using space-time constraint analysis to generate internally consistent itineraries (Sun et al. 2013). In this process, GIS and known data regarding the respondents' activities enabled the estimation of each person's location of intermediate points of presence during the day. This in turn allowed a wider range of data coding for each person and thus enabled a wider range of analyses of tourist presence and activity on any given day. Detailed activities recorded by respondents were informally classified into 12 classes.
Multiple-view geovisualization of tourist activities
Multiple views generally refer to multiple distinct views (a distinct view is a specific visualization of the data) of the same or related data displayed simultaneously. Multiple views can display multiple- or same-form representations with different display parameters or attributes; they may be represented at an individual or an aggregated level, and at the same or different spatial, temporal, and activity scales. Depending on the relationship between two or more representations, we classified multiple views into six...