Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Designing and testing a prototypical landscape information interface for lay-people Salter, Jonathan D. 2005

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata


831-ubc_2005-0626.pdf [ 28.26MB ]
JSON: 831-1.0075011.json
JSON-LD: 831-1.0075011-ld.json
RDF/XML (Pretty): 831-1.0075011-rdf.xml
RDF/JSON: 831-1.0075011-rdf.json
Turtle: 831-1.0075011-turtle.txt
N-Triples: 831-1.0075011-rdf-ntriples.txt
Original Record: 831-1.0075011-source.json
Full Text

Full Text

DESIGNING AND TESTING A PROTOTYPICAL LANDSCAPE INFORMATION INTERFACE FOR LAY-PEOPLE  by  Jonathan D. Salter  A THESIS SUBMITTED IN PARTIAL FULFILMENT OF THE REQUIREMENTS FOR THE DEGREE OF  MASTER OF SCIENCE  in  THE FACULTY OF GRADUATE STUDIES  (Forestry)  THE UNIVERSITY OF BRITISH COLUMBIA  June 2005  ©Jonathan D. Salter  Abstract  Landscape management is increasingly moving towards managing for multiple values and involving the public in decision-making processes. To provide informed opinions about how they would like various aspects of the landscape to be managed, members of the public require information about the implications of given management actions. Ecological processes tend to occur over large spatial and temporal scales, and involve bothvisible and non-visible indicators of landscape condition. People have dif culty conceptualizing landscape processes at these large spatial and temporal scales. Conventional methods of communicating this type of landscape information in British Columbia have been criticized as being too technical and inaccessible for non-experts. Developing technologies, such as realistic landscape visualizations, have been heralded as possible solutions, because of their abilities to engage users and communicate effectively. Realistic landscape visualizations, however, emphasize the visual aspects of landscape change. This may exacerbate people's existing tendencies to evaluate landscape management based on what they see on the landscape. The purpose of this thesis was to design, build and evaluate a prototype landscape information interface to communicate spatial, temporal, visible and non-visible characteristics of the landscape to non-expert users. The resulting interface displayed landscape visualizations and indicator graphs in association with other contextual information. This interface was evaluated, using a questionnaire and an exploratory video analysis procedure, by investigating how non-expert subjects would use the interface to answer landscape related questions. Results of the evaluation suggest that subjects quickly understood the interface and the information it presented, and were able to combine the interface's representations to understand temporal and spatial aspects of the landscape. The subjects also demonstrated the ability to recognize relationships between the ecological indicators represented in the interface. The ndingsof the study suggest that such interfaces can be an effective means of communicating information about landscapes and the implications of management. Future directions are discussed, relating to the development and evaluation of landscape information displays.  ii  Table of Contents Abstract Table of Contents List of Tables List of Figures Acknowledgements Dedication  1  ii iii v vi viii ix  1  Introduction 1.1  1.2 1.3 1.4 1.5 1.6'  Background on Landscape Perceptions and Information 1.1.1 Landscape Preferences: The Scenic Aesthetic 1.1.2 Broader Views of Landscape Perception 1.1.3 Multi-Attribute Public Perceptions and Learning 1.1.4 Potential Role of Landscape Information Displays Thesis Research Objectives Background and Precedents for Landscape Information Display Study A p p r o a c h Thesis Outline A Note on Conventions Used i n the Thesis  1 2 5 6 9 12 13 15 16 16  2  Towards an Interface Design: Information Display Literature Review 2.1 Introduction 2.2 Data: Context for the Interface Design 2.2.1 The A r r o w IFPA M o d e l i n g Process 2.3 Information Display Techniques Literature Review 2.3.1 Cognitive Psychology Concepts 2.3.2 Landscape Visualization 2.3.3 Information Visualization 2.3.4 Graphs 2.3.5 M a p s 2.3.6 Text Information 2.3.7 D y n a m i c Representations 2.3.8 Interactivity 2.3.9 M u l t i p l e Representations 2.3.10 A Brief Note on H C I for the Interface Design 2.3.11 Conclusion  18 18 19 20 24 25 30 39 41 43 48 52 57 64 69 69  3  Interface Design and Implementation 3.1 Introduction 3.2 General Interface Principles and Attributes 3.2.1 General Interface Attributes 3.2.2 Required Technical Attributes 3.3 Interface Elements  71 71 71 73 76 79  iii  4  5  3.4 The Interface Explained 3.5 Evaluation of the Interface by Criteria 3.5.1 Simple and Minimal Interface Controls 3.5.2 User Interaction and Controls 3.5.3 Smooth Interactivity 3.5.4 Element Legibility 3.5.5 Coherent Information Display 3.5.6 Amenability to Evaluation 3.6 Conclusion  100 104 104 105 105 106 106 107 108  Evaluation  110  4.1 Introduction 110 4.2 Methods 114 4.2.1 Questionnaire Design 115 4.2.2 Subject Recruitment Methods 116 4.2.3 Physical Setup Methods 116 4.2!4 Session Methods 119 4.2.5 Analysis Methods 120 4.3 Results and Discussion 124 4.3.1 Task One 125 4.3.2 Task Two 131 4.3.3 Task Three 135 4.3.4 Task Four 140 4.3.5 Task Five 146 4.3.6 Task Six 151 4.3.7 General Findings Across Tasks 159 4.3.8 General Discussion of the Research Findings According to Dominant Themes . . 167 4.4 Relation of the Evaluation to the Thesis' Research Objectives 188 Conclusion  189  5.1 5.2 5.3 5.4  Landscape Perception, and the Effects of Cognitive Information Information Display Design and Implementation Evaluation 5.4.1 Methodology 5.4.2 Findings of the Evaluation 5.5 The Thesis' Research Objectives Revisited  190 191 194 197 197 199 203  5.6 Future Directions and Recommendations  205  Literature Cited  213  A  Study Questionnaire  228  B  A n Example Subject Spreadsheet from MacSHAPA  239  C  Subject MacSHAPA Timelines  249  iv  List o f Tables 3.1 The percentage of internet users utilizing different screen resolutions 4.1 Demographic information for the fourteen subjects from the study.  78 161  4.2 Frequency of non-directed selection for each Viewpoint by each of the subjects, as well as the mean number of non-directed selections for each of the Viewpoints 163 4.3 Frequency of non-directed selection for each Image Type by each of the subjects, as well as the mean number of non-directed selections for each of the Image Types 164 4.4 The amount of non-directed time each of the subjects spent on each Viewpoint, as well as the mean amount of of non-directed time for each of the Viewpoints 165 4.5 The amount of non-directed time each of the subjects spent on each Image Type, as well as the mean amount of of non-directed time for each of the Image Types 166  v  List of Figures 1.1 A representative example of realistic landscape visualization of the type used in this thesis. .  11  1.2 The understanding continuum. Understanding progresses from left to right, from data to wisdom  17  2.1 A map of the Lemon Landscape Unit in the Slocan Valley of southeastern British Columbia. 21 2.2 A schematic representation of the model relationships for the Arrow IFPA modeling project  22  2.3 The SIMFOR visualization process. Figure a) shows the SIMFOR raster data rendered in World Construction Set against a black background. Figure b) shows the completed SIMFOR image after it has been photomontaged in Adobe Photoshop (Adobe Systems Corp., 2002b)  24  2.4 A schematic diagram of the human memory systems applicable to learning from text and graphics 26 2.5 Examples of the four types of landscape visualization as described by McGaughey (1998). 36 2.6 A diagram explaining the increase in working memory capacity activated by the use of both the visual and auditive channels  49  2.7 Diagrammatic representations of Ainsworth and VanLabeke's (2004) dynamic representation types  57  2.8 A schematic representation of the locations of different presentation media along a continuum from low interactivity to high interactivity.  59  2.9 A schematic representation of the relationships between characteristics of different representation types and the available literature  70  3.1 Representative images of the four viewpoints used for the landscape visualization images in the interface  83  3.2 Images demonstrating the effects of resolution on tree images  84  3.3 The graphs of the non-visible indicators of landscape condition used in the interface (at year 100)  92  3.4 An example of a spatial overlay of GIS data draped on a realistic landscape visualization. 95 3.5 The context map used in the interface showing two different viewpoints  97  3.6 The Summary Information text area from the interface  98  3.7 The temporal slider bar used to control the time steps for the graphs and the landscape visualizations in the interface  99  vi  3.8 The dropdown menus used to control the landscape visualization viewpoint and the landscape visualization image type 100 3.9 An annotated diagram of the interface created for this project, with all of the interface elements labeled 102 4.1 The physical setup of the equipment for the study session  117  4.2 An annotated diagram of the MacSHAPA interface as used in this study.  122  4.3 A graphic matrix of the subjects' responses to questions that could be categorized into correct, incorrect, or conditional categories 126 4.4 The time lag relationship between the Young Forest (Figure a) and Swainson's thrush (Figure b) habitat graphs 136 4.5 The spatial relationship between the Young Forest (Figure a) and Swainson's thrush habitat (Figure b) spatial overlays on the landscape visualizations  137  4.6 The spatial relationship between the Old Growth (Figure a) and Swainson's thrush habitat (Figure b) spatial overlays on the landscape visualizations 138 4.7 Timeline examples of active and passive users. The black boxes outline the subjects' non-directed actions 157 5.1 A photograph of a public consultation processes conducted in the Landscape Immersion Lab for the Snug Cove density transfer study. 207 5.2 Images showing different forms of abstract landscape visualizations  vii  209  Acknowledgements I would like to thank my committee members: Dr. Stephen Sheppard, Dr. Hamish Kimmins and Mr. Don Luymes for their time and advice. I would also like to thank my external examiner Dr. Murray Journeay for his input. I am indebted to the members of the Collaborative for Advanced Landscape Planning (CALP) for their support. I would especially like to acknowledge Cecilia Achiam, Caitlin Akai, Ritchie Argue, Cam Campbell, Duncan Cavens, Howard Harshaw, John Lewis and Paul Picard, for help with this thesis and related projects.  viii  To my parents —for unwavering support on a sinuous journey.  ix  Chapter 1  Introduction 1.1  Background on Landscape Perceptions and Information  The intention of this thesis was to design, implement, and evaluate a prototypical interface to communicate temporal and spatial characteristics of the landscape. The purpose for creating the interface was to investigate techniques for information display that could effectively communicate ecological dynamics of forested landscapes to non-experts. The dynamics of natural landscapes are complex, and occur over large spatial and temporal scales (Forman, 1995). Management of these landscapes in Canada falls primarily to government agencies, and the majority (over 90%) are public lands. Members of the public both want and need to have a say in how these landscapes are managed (Jabbour and Balsillie, 2003). For members of the public to give informed opinions, or make informed decisions about how they would like those landscapes managed, they require knowledge about how the landscape functions, and how management actions affect it (Eaton, 1997). Forest management has evolved from the early focus primarily on the economic benefits of timber extraction, to a multiple value approach that takes into account non-timber economic values, ecological values, cultural values, scenic values, etc. (Bell, 2000; McCool and Guthrie, 2001; Sheppard, 2001a). One unifying concern across all of these values is the desire for the landscape to provide them in a sustainable manner. From an ecological standpoint, managing for sustainability means considering the spatial and temporal characteristics of the landscape (Kimmins, 2001). Kimmins (1999) describes sustainability in forest ecosystems as "a non-declining pattern of change". At the landscape scale this manifests itself as a shifting mosaic of ecosystem types, moving through successional stages and disturbance events. It is difficult to determine the sustainability of a landscape from assessing current conditions, even for expert forest ecologists (Kimmins, 1999). This is largely due to both the critical importance of nonvisible indicators of landscape condition, and the crucial role that temporal functions play in ecological sustainability (Kimmins, 2001). People, however, have difficulty conceptualizing future landscape 1  2 management issues at the spatial and temporal scales necessary for the consideration of sustainable management (Eaton, 1997). This inability to conceptualize landscapes at the necessary spatiotemporal scales, as well as the inability to assess ecological sustainability from current conditions, means that non-experts are often ill-equiped to assess the sustainability of landscapes, making informed feedback in public processes difficult. Exacerbating this problem are the methods which management agencies currently use to communicate the plans for management of the forest resource. These communication techniques are often opaque, and too technical for non-expert users (Forest Practices Board, 2000; Jabbour and Balsillie, 2003). New visual media such as realistic three-dimensional (3D) landscape visualizations, and interactive information displays show promise in improving this communication, and supporting learning. In the absence of accessible information, people tend to judge the management of the landscape based on what they see (Kimmins, 1999; Sheppard and Harshaw, 2001b; Shindler et al., 2005). This is problematic if the goal of management is sustainability, because there is an apparent disconnect between people's visual preferences for landscapes and several management techniques that are used in ecosystem management and sustainable forest management systems (Gobster, 1995, 1999; Ribe, 1989). This is applicable to real-world views, and may also manifest itself in representations of future landscapes constructed using new visualization media, potentially limiting their effectiveness or even biasing lay-people's input to landscape management decision-making based on aesthetic preferences. The following subsections describe the issues surrounding people's landscape perceptions and derive requirements for visual information presentation to convey more than just the aesthetic factors of landscapes. 1.1.1  Landscape Preferences: The Scenic Aesthetic  Preferences are thought to be low-level, emotionally-based responses to stimuli, that are very difficult to change (Zajonc, 1980). Research on landscape preferences has found consistency across cultures, occupations, ages, etc., in the preference for particular landscape characteristics (Aoki, 1999; Balling and Falk, 1982; Brush et al., 2000; Daniel and Boster, 1976). The basis for these strong consistencies are thought to be evolutionary, with humans having developed preferences for particular landscape characteristics that conferred advantages on our ancestors (Appleton, 1996; Kaplan, 1987). These evolutionary preferences have even been shown to have physiological effects on people, with the viewing of nature improving creativity, and demonstrably reducing recovery times from surgery (Parsons, 1991; Parsons et al., 1998; Ulrich, 1984). The result of the research on preferences has been a view of people's landscape preferences as being relatively consistent across demographics, and resistant to change. If true, this could make the implementation of particular ecosystem management and sustainable forest management practices difficult. For example, in mid-ground vista scenes, the percent  3  alteration caused by forest harvesting has been shown to be negatively associated with people's scenic preferences (McCool et al., 1986). Mimicking forest fire in fire dominated landscapes may result in the ecosystem management practice of amalgamating harvesting into large blocks, and consequently amalgamating leave areas. This amalgamation of harvest blocks, however, would apparently conflict with people's scenic preferences. Gobster (1995) indicates that some practices that enhance ecological sustainability work contrary to those that enhance visual quality. He speculates that the classical scenic aesthetic runs counter to such practices as providing wildlife trees, maintaining downed wood in cutblocks and silvicultural practices that maintain dense understory vegetation. If people's scenic preferences for landscapes do conflict with sustainable management practices, can something be done to alter those preferences to more closely align with "good" management? One suggestion has been to attempt to change people's landscape preferences by providing cognitive information. Preferences are, however, notoriously difficult to change (Daniel, 2001a). Zajonc (1980), in his pioneering work on preferences, stated that "preferences need no inferences". In other words, cognition is not a part of people's preference judgments, and is therefore unlikely to have a significant effect on people's preferences for landscapes. Little research has been conducted to date on the effects of information on landscape preferences. The research that has been conducted has been contradictory and inconclusive. Anderson (1981) looked at the effects of information in a very basic form: simply providing labels indicating the type of management that an image represented. The labels included wilderness area, national forest, national park, recreation area, leased grazing land and commercial timber stand. She found that the wilderness area and national park designations had a significant positive effect on people's perceptions of scenic quality, while commercial timber stand and leased grazing land both had a negative effect on people's scenic quality ratings. Anderson speculated that the labels she used implied a level of naturalness present in the scene, and therefore led to an emotional reaction that altered participants quality ratings. This study suggests that information, when presented in an easily understandable and relevant form, can have a significant effect on people's preferences for landscape scenes. The possible affective loading of the terms used, however, means that it is not a particularly strong case for the ability of purely cognitive information to alter evaluations. Brunson and Reiter (1996) looked at the effects of ecological information on the scenic evaluations of test participants. Forest scenes depicting a variety of ecosystem management and non-ecosystem management actions taken from Oregon were used to elicit scenic quality measures from two groups. One group was composed of Utah State University business students, and the other group was made up of employees of the Utah Department of Employment Security. Each group included a treatment group that was read information on ecosystem management, and a control group that was not provided with any information. Brunson and Reiter (1996) did not find a significant difference between  4 the pooled treatment groups and the pooled control groups. Hines (2001) examined the effects of both text based and visual (photographs) information on subjects scenic beauty ratings of natural landscapes. The study involved two paired conditions. In one condition, the treatment group was given information indicating that it would be ecologically beneficial to convert savannah landscapes in the Midwestern US back to woodland landscapes. The control group was given no information about the two landscape types. In the second condition, the treatment group was given information indicating that it would be ecologically beneficial to convert woodland landscapes back to savannah landscapes. Again, the control group was not given any information about the two landscape types. All of the subjects were then asked to rate a series of landscape slides for scenic beauty. These landscape slides represented both savannah and woodlands landscapes. The study found strong correlations for the scenic beauty of a given landscape type across both treatment and control conditions. Education, when taken as a complete condition (i.e. without considering which treatment group individuals fell into) did have a significant effect. Subjects that were provided with the educational material had significantly lower overall scenic beauty judgements than those that were not provided with the material. When the interaction between ecotype and education was considered, the results were not significant for either ecotype, indicating that the provision of information did not affect people's ratings of the different landscapes. These results are difficult to interpret. The educational material did affect peoples scenic beauty judgements, but not in a predictable or easily interpreted manner. Taylor and Daniel (1984) examined the effects of information in the form of brochures on peoples scenic beauty estimates, their willingness to recreate in an area, and their tolerance ratings for particular management actions. The participants were given information brochures that either contained general information about the Ponderosa pine (pinus ponderosa) forests that they were going to view, or brochures that contained information about the management practice of prescribed burning and the effects of light and severe fires on the overstory vegetation. The brochures did not significantly affect the participants' ratings of either scenic beauty or willingness to recreate in an area, but they did influence tolerance ratings. People who received the brochures with the fire information were more tolerant of prescribed burning management actions in post-test questionnaires than subjects that received the control brochure. While their landscape preferences were not altered, this study does indicate that members of the public may be more willing to tolerate management actions they do not prefer, if they are provided with relevant information. It is difficult to synthesize the current research on the effects of information on landscape preferences into a coherent picture. Anderson (1981) suggests that if the information is made salient and simple, it can have a significant effect on people's perceptions. Brunson and Reiter (1996), and Hines (2001) both indicate that provision of verbal and/or text-based information does not have a consistent effect  5 on people's perceptual judgements. Perhaps the most telling research from the literature on the effects of information on landscape preferences, was the work by Taylor and Daniel (1984), who found that information about forest fire did not affect people's preference judgments, but that it did affect their willingness to tolerate forest fire management practices. This suggest a role for information, perhaps not in altering landscape preference, but rather in altering perceptions of the acceptability of particular management actions. 1.1.2  B r o a d e r V i e w s of L a n d s c a p e P e r c e p t i o n  The application of the scenic aesthetic to forest land management developed as a response to public opposition to the practice of clearcutting in the 1970's, and the implementation of the United States' National Environmental Policy Act in 1969, which mandated the protection of scenic resources (Sheppard, 2001a). These developments led to the creation of visual resource management (VRM) practices in forestry, and the alteration of forest management to protect scenic values (Sheppard, 2001a). The focus on scenic resources led to significant research on people's preferences for particular landscape characteristics, such as the research mentioned in the previous sections. Traditional measures of landscape preference have focused on rapid responses to static images of landscapes (Gobster, 2001). The most common measure of landscape preference used in research has been the scenic beauty estimation method (SBE), developed by Daniel and Boster (1976). This method calls for users to rate the scenic beauty of static landscape scenes, most often displayed using representative photographs or slides. Because the focus of both research and management was on scenic preferences for current landscapes, VRM and perception research did not take ecology and sustainability into direct consideration. Where they did consider sustainability, there was a tacit belief that protection of visual resources was also good for the larger environment (Sheppard, 2001a). With the development of ecosystem management and sustainable forest management, the view of VRM has changed. No longer is it just a constraint on timber harvesting, now it may be in opposition to other non-commodity values like ecological sustainability. This change has been accompanied by a corresponding change in theories about aesthetic appreciation of landscapes, and how that appreciation is measured. Daniel (2001a), one of the creators of SBE has stated that the method is ill-equipped to deal with the spatial and temporal scales required in the discussion of ecological sustainability: To prefer an ecologically sustainable landscape must mean to prefer a particular geographic and temporal pattern of environmental conditions over other possible patterns. Such preferences cannot adequately be assessed by presenting individual scenes depicting particular places at particular points in time. Daniel (2001a) also states that assessment of people's perceptions of sustainable landscapes may more appropriately look at their willingness to accept management actions that are more sustainable,  6  rather than their preference for the resulting landscape conditions. This is more in line with the previously discussed "willingness to tolerate" found by Taylor and Daniel (1984) in relation to prescribed burning, despite the burning's negative scenic beauty ratings. Partially in response to the potential conflict between scenic preferences and ecological sustainability, some researchers and theorists subscribe to the notion that visual preference is only one part of the larger aesthetic appreciation or perception of landscapes. Carlson (2001) describes aesthetics as both seeing and knowing, and states that knowledge about the health and sustainability of the landscape contributes to the aesthetic appreciation of it. In this rationale, knowledge and information do not necessarily alter preferences, but may modify the effects of those preferences on the viewer's larger appreciation  of the landscape. Eaton (1997) agrees with Carlson that knowledge plays a fundamen-  tal role in the aesthetic experience, and that it can make aesthetic appreciation attentive to aspects of the landscape that might otherwise go unnoticed. Eaton (1997) states that appreciation of sustainable landscapes is not possible without knowledge of the functioning of those landscapes. This makes intuitive sense, as ecologists and landscape experts undoubtedly have a different aesthetic appreciation of the landscape than lay-people who lack the experts' knowledge about the landscape and its functions. Citing the limitations of the scenic aesthetic and drawing upon the more inclusive view of landscape aesthetics, several researchers have made normative arguments about the need to increase the public's knowledge about ecology, sustainability and landscape management. Gobster (1995,1999) argues that what is required is a wholesale societal move away from the scenic aesthetic towards an "ecological aesthetic". This ecological aesthetic prescribes that people adopt a knowledge-based way of looking at the landscape, indicating that people's aesthetic perceptions should include cognition and rational thought, rather than just being emotional responses to sensory information. Thayer (1989) promotes a similar concept in non-forested landscapes. He indicates that certain aspects of sustainable landscapes do not fit with the traditional scenic aesthetic. Particularly problematic are certain sustainable energy sources such as wind farms and solar power arrays, as well as water and waste management schemes (Thayer, 1989). Thayer argues that in order to foster acceptance of these technologies they must be made readily visible to the public. Traditional management in resource sectors has involved "hiding" extraction activities from view because they are deemed "ugly" by the public. In forestry, elaborate management schemes have been devised to mitigate the visible impacts of harvesting (Bell, 2001). According to both Thayer and Gobster, these practices can be detrimental to the acceptance of sustainable management practices. 1.1.3  Multi-Attribute Public Perceptions and Learning  Another mechanism for investigating people's perceptions and valuation of landscapes is to review research that has been conducted in public consultation processes for land-use planning. A recent  7 study by Sheppard and Meitner (2005) used a Multiple Criteria Analysis (MCA) with stakeholder groups in British Columbia's Slocan Valley. The stakeholders were asked to rank the importance they placed on different characteristics of the landscape. The cumulative (across-group) ranking of the different landscape resources was: 1. Biological richness 2. Water supply 3. Timber economic values 4. Forest/Soil productivity 5. Non-timber economic values 6. Recreation 7. Visual quality 8. Cultural resources 9. Safety This valuation indicates that the subjects in the study rated a number of resources, including ecological resources such as biological richness and forest/soil productivity, higher than visual quality. Similar results were found in a separate study of public involvement in forest planning in the central portion of BC, adjacent to Tweedsmuir Provincial Park. Jabbour and Balsillie (2003) conducted interviews with the local public advisory group (PAG) and a random telephone survey with local residents (the public sample population, or PSP) to determine their valuation of different landscape resources. While there were differences between the PAG and the PSP, the cultural and aesthetic resources consistently came out below environmental conservation, timber resources, fish and wildlife and hiking and recreation resources. Similar to the study by Sheppard and Meitner (2005), Jabbour and Balsillie (2003) found that, when taken in the larger context of a planning process, people placed a lower value on visual quality than they did on other, ecological and socioeconomic resources. Manning et al. (1999) conducted a survey of the environmental values of Vermont residents concerning the Green Mountain National Forest. Unlike the previously mentioned studies, they found that aesthetic values were rated as highly as ecological values (the difference between the two was statistically insignificant). Sheppard (2005) cautions that the low value placed upon visual quality in the two British Columbia studies mentioned above likely represents both an unwillingness on the part of participants to place a high value on something as 'fuzzy' as visual quality, and the fact that visual quality is a resource that has much larger impact when experienced in the landscape. Nevertheless, these results suggest that from a cognitive perspective, people are inclined to place a higher (or at least equal) value on the ecological functioning of the landscape then they do on its visual characteristics. It may also suggest that when people make negative judgments of landscape management based on visual appraisal, they are not  8 only reacting to visual quality, but also to what they think that quality says about the care and sustainability of the management practice. This would indicate that the problem is not so much one of incompatible preferences, but rather of misperceptions of the effects of particular management actions. Several researchers have identified ways in which these misperceptions might be addressed. Gobster (1995) outlines several strategies for fostering the adoption of an ecological aesthetic. One of these strategies is the presentation of a "conspicuous experiential quality" (Gobster, 1995). Gobster postulates that by making management activities readily apparent, and encouraging people to experience them, their acceptability ratings will be improved. This is similar to Thayer's assertion that sustainable landscapes must be made visible, but takes it one step further by indicating that people should be actively encouraged to experience and learn about the management practices. Nassauer (1997) suggests that managers need to provide "cues to care" to show people that the landscape is being managed with care and intent. Cues such as explanatory signs can be used to indicate that the visible management action is part of a larger plan for the landscape. Sheppard (2001a) expands this notion, by suggesting that managers employ "visible stewardship", making the careful management of landscape resources explicitly visible, so that commitment to, and respect for, the landscape and its long term health are clearly apparent in a manner similar to a farmer's care for the land he or she manages. These experiential methods may meet with initial resistance, because the management actions they address conflict with people's present evaluations of management practices. It will take time for people to experience and understand sustainable landscapes, cues to care and indicators of visible stewardship. A shorterterm mitigation, and one that may facilitate the implementation of the experiential methods is the improvement of information presentation and learning support provided to members of the public through public involvement processes and extension efforts. The improved provision of information about ecology and sustainability is another of the methods that has been advocated in the normative literature (Eaton, 1997; Gobster, 1995, 2001). Research in public participation processes supports the notion that participants want better and more accessible information. In their study of a public participation process Jabbour and Balsillie (2003) found that 86% of the public advisory group members felt that learning should be a major component of the process, and 50% defined the process in relation to information sharing. McCool and Guthrie (2001) reviewed two public participation processes in Montana, and conducted interviews with scientists, managers, and members of the public that attended the processes. The intended purpose of their study was to determine what attendees of the processes believed were determinants of successful or unsuccessful public consultation. The most commonly referred to determinant of success from members of the public was learning, indicating that the two-way flow of information was of crucial importance to public evaluations of success. Tuler and Webler (1999) conducted interviews with participants in public processes from New York and Vermont. They outlined seven categories of  9 qualifications for successful public processes. One of the seven categories was access to information, and further elaboration suggests that participants expected to be informed about management proposals. Wagner et al. (1998) in a review of both forester and public values in Ontario found that public confidence in government and industry managers was low, but that they strongly supported the use of scientific information. Wagner et al. (1998) suggest that this indicates that scientific information and the scientific rationale for management actions should be a key component of public outreach. These reviews of public processes and public values indicate that participants want information transfer and learning to be part of the process. Specifically, they want the two-way flow of information to be improved. Effective transmission of information from management to the public can improve the two-way information flow by informing the public of the intentions of management, and the science behind it. This can help to clarify the position of management agencies and focus the feedback from members of the public. 1.1.4  Potential Role of Landscape Information Displays  If ecological information is to be used to inform people's perceptions of ecosystem and sustainable forest management activities, what issues need to be considered? A review of the literature suggests that 1) the type of information that should be presented, 2) the method of presentation, and 3) the type of question being asked/measure being used are critical issues. The type of information to be presented is, in large part, dictated by the intended purpose. Some general principles may be formulated, however. Gobster (1999) states that it is not enough for people to understand bits and pieces of the landscape, they must understand how the pieces relate to the whole. This holistic approach to comprehending ecological systems is fundamental to ecosystem management (Franklin, 1997). Therefore, if the goal of providing information to people is to aid them in understanding the effects of landscape management actions, then it is not enough to give them information about limited landscape attributes in constrained geographic areas, and over short timescales. In ecological terms, one 60 hectare clearcut on a hillside is meaningless. Without the contextual information within which to place that particular cut, it is not possible to infer its ecological impact. As previously mentioned, people have difficulty visualizing at the spatial and temporal scales necessary for the consideration of ecological sustainability (Eaton, 1997). This is, however, the appropriate level at which to communicate if the intended goal is to foster an understanding of ecosystem and sustainable forest management practices. Sustainable management cannot properly function unless it is incorporated into a cohesive plan over a large geographic area (Bell, 2000). In order to properly understand this planning scope, people must have some conceptual understanding of how the landscape operates on that geographical scale. To this point, research into the effects of information on landscape preferences has used primarily  10 text-based or verbal means to communicate the information. Hines (2001) is the only study reviewed here that used images in its information package. These images, however, relied heavily on the associated text for explanation. Visual information is processed much more efficiently than other forms (Larkin and Simon, 1987; Tufte, 1983). The wealth of information in human societies that is directed at the visual system is a strong indicator of the emphasis that humans place on this modality (Feeney, 1994; Wainer, 1992). In effect, the research to date on the effects of information on landscape preference has been attempting to assess the impacts of a non-visual data source on a visual phenomenon. It is unlikely that this is an effective way to communicate information about visual changes over the landscape. Weidemann et al. (1997) found that judgements made about scenic quality based on verbal descriptions did not correlate with the same judgements made for a photographic scene representing the same location. They suggest that verbal representations cannot be used as valid surrogates for representing visual aspects of the landscape. Tahvanainen et al. (2001) found similar results in their studies of preconceptions. Verbally elicited responses to management actions such as clearcuts, understory removal and traditional management did not correspond to the scenic quality evaluations that subjects expressed when shown slides of the same management actions. Because participants in visual perception studies are being asked to evaluate visual phenomena, it seems logical that the information they are provided should be presented using visual means. One method for visual presentation of information about landscapes that shows promise is realistic landscape visualizations (see Figure 1.1). These landscape visualizations have been shown to provide engagement (Al-Kodmany, 2000; Campbell and Salter, 2004), and to effectively communicate the implications of development and management actions (Sheppard, 1989). Daniel (2001a) indicates that landscape visualization technology is capable of displaying the spatial and temporal scales necessary to represent ecological sustainability, and Gobster (2001) states that landscape visualization shows promise as one mechanism for communicating information to help improve the public's knowledge of ecology and sustainability. Sheppard and Meitner (2005) found that over 90% of the participants in a multiple criteria analysis process found the use of multi-temporal landscape visualizations helpful in assessing forest planning scenarios. These findings suggest that landscape visualization is one possible mechanism for visually communicating concepts of ecological sustainability at the spatial and temporal scales necessary. Visual presentation, however, may not be sufficient in and of itself. Heft and Nasar (2000) found that static information displays may not adequately represent environmental perception. They found that information retained in dynamic displays was not always retained in static displays. Daniel (2001a) suggests that static representations of landscapes are unlikely to be sufficient to represent the geotemporal variations of large landscapes. It might be necessary to move to time-lapse, fly-over or walkthrough animations, or real-time simulations in order to capture the spatial and temporal variations inherent in large scale ecological processes.  11  Figure 1.1: A representative example of realistic landscape visualization of the type used in this thesis. Studies on interactivity (to be discussed further in Chapter Two) appear to support the notion that interactive systems can increase engagement (Conroy and Gordon, 2004; Aldrich et al., 1998; Kettanurak et al., 2001). The presentation of pre-determined video or animation sequences are unlikely to have the same capability to draw peoples attention as interactive systems that give the participants control over the information display, and animation as an educational aid has thus far not shown promising results (e.g. Jones and Scaife, 2000; Lewalter, 2003; Lowe, 2003,2004; Tversky et al., 2002). The ability of participants to interact with the information provided about ecological processes over the landscape could greatly facilitate their investigation and subsequent understanding of the patterns and processes represented. The information presented should also be represented in the most appropriate form according to the available data, the intent of its presentation, and the need to avoid any misrepresentations of its basis and reliability (Wilson and McGaughey, 2000). Not all information is best represented in realistic visualizations or dynamic fly-overs of a landscape. Each data source should be represented according to its own underlying properties, such as inherent spatial characteristics, relationship to time, relationships to other variables, etc. The common availability of realistic visualization technologies has created a myriad of visual representations that have not been tested for validity or their ability to communicate (Wilson and McGaughey, 2000). The intention of communicating ecological information is to convey the most relevant information in the most appropriate manner. Finally, the information should be presented in such a way that it "surround(s) the truth" (Daniel, 2001a). This means that the participants should be presented with the information in such a manner that they are able to draw upon the relevant sources to formulate a coherent picture of how different aspects relate to the whole. This echoes earlier work by Appleyard (1977), suggesting that multiple media forms be used to triangulate on the truth and offset media bias. The idea of information immersion or triangulation can be difficult to achieve. Taylor and Daniel (1984) found that a brochure  12 which depicted fire information with both graphs and line drawings had a smaller effect on acceptance of prescribed burning than brochures that contained either graphs or line drawings. Their interpretation was that the subjects presented with the brochures that contained both graphs and line drawings were experiencing information overload, and as a result were not making use of the available information. The threat of information overload or, cognitive load as it is more commonly called, is a very real problem with information displays that represent a large amount of information (Chandler and Sweller, 1991; Sweller et al., 1998). Cognitive load will be discussed further in Chapter Two. To sum up, the presentation of information to assist in public planning processes, and educational awareness sessions about ecology and sustainable management, should be visual, dynamic, interactive, appropriate and immersive. The representations used should also follow ethical guidelines of production, to ensure they meet the requirements of accuracy, representation and defensibility as outlined by Sheppard (2001b). Evaluations of all of these criteria are necessary in order to determine how they should be implemented to produce information displays to communicate landscape information.  1.2  Thesis Research Objectives  Research into people's perceptions, and acceptance of sustainable management techniques is in its infancy. A considerable amount of research is required in order to determine if information can have an effect on these acceptance ratings. Research is also required to determine what information is appropriate, what manner it should be presented in, and how people's perceptions of landscapes should be measured. The research to date has been inconclusive, however the techniques used have been holdovers from landscape preference studies that are likely unsuited to the task (Daniel, 2001b,a; Eaton, 1997; Gobster, 2001). A move away from the use of static scenes and scenic beauty evaluations may be required in order to more adequately address the complex, dynamic nature of ecological information over the landscape, and the multi-faceted valuation of the landscape. The choice of the proper information to present will likely be an interplay between expert opinion and experimental research. Taken as a whole, a review of the research in this area raises a number of possible research tracks. There is the specific question of the effects of information on preference, which remains unclear, despite the research that has been conducted. There is the possible development of more inclusive measures for assessing people's valuation of landscapes, and management practices. There is the large question of determining what information is necessary and appropriate to present to public groups in consultation processes. Lastly there is the question of creating and evaluating better ways to portray landscape management and landscape processes to foster a more knowledge-based way of interpreting the landscape. It is this last question which is the focus of this thesis. The background information and problem framing information stated above led to the formation  13 of two research objectives for this thesis. General research objectives were derived, instead of more specific research questions, due to the exploratory nature of the study conducted in this thesis, and the relative lack of previous research in this area from which to draw specific hypotheses for testing. The research objectives developed for this thesis were: 1. To research, design and implement a tool to communicate information about ecological landscape conditions. The tool should show landscape visualizations in association with other information representations. The goal of the tool is to communicate information about the landscape's temporal and spatial aspects, and about visible and non-visible indicators of landscape condition to non-expert users. 2. To investigate how non-expert users interact with the tool; how they use the information provided to answer questions about the landscape, what they learned from interacting with the tool; and to assess the efficacy of the tool for communicating the spatial and temporal characteristics of the represented landscape.  1.3  Background and Precedents for Landscape Information Display  To address the issue of communicating information about the landscape, its dynamics, and its management, it is instructive to look at the larger picture of information display and communication. There are several research fields directed specifically at the communication of information. These fields include Information Design (e.g. Jacobson, 1999; Tufte, 1983,1990, 1997), Information Architecture (e.g Wurman, 2001), Information Visualization (e.g. Ware, 1999), and sub-fields within educational psychology and instructional science. Researchers in these fields investigate, design and evaluate information display techniques, often with complex information sources and non-expert users, in ways similar to that proposed in this thesis. Reviewing the literature in these fields, and applying what is known about information display to the design and implementation of the prototypical interface should improve the interface's communication capabilities. A review of the relevant information display literature will be conducted in Chapter Two. The rest of this section provides a more focused summary of recent work on forest landscape interfaces, as a baseline for further advances in information display on landscape management issues considered in this study. Reviewing specific instances of information interfaces created for the purpose of communicating landscape and forestry information can provide guidance to further development and testing. There have been a number of landscape and forestry information systems developed to communicate information about landscape management. Two of the most prominent systems have been Smart Forest (Orland, 1998) and the Landscape Management System (LMS) (McCarter et al., 1998). Both of these packages provide forest modeling characteristics, as well as the capability to visualize the effects of forest management actions. Smart Forest and LMS, however, are both expert-based systems, intended for use by managers and researchers in the evaluation of forest management plans. While their visualization output can communicate for-  14 est management effects in an accessible manner, operating the systems and generating that output requires expertise and knowledge of forestry and forest management. Two programs, Possible Forest Futures (PFF) (Kimmins, 2005) and its predecessor FORTOON (Kimmins and Scoullar, 1995), more closely approximate the intended interface for this thesis, as they were specifically designed to communicate concepts of forest dynamics and management. FORTOON is an interactive game for use by elementary and high school students to teach concepts of forestry and forest management. Possible Forest Futures is a more advanced communication tool that allows users to perform limited modeling activities and to examine the effects of those activities on a hypothetical landscape via animations of both numerical indicators on graphs, and abstract visualizations of the landscape's spatial characteristics. The main features that separate FORTOON and PFF from the interface proposed for this thesis are the intended audience, and the use of hypothetical versus realworld landscapes. Both FORTOON and PFF are intended as educational tools to communicate general concepts of forestry and forest management. They are not intended as tools to investigate forest management implications on specific landscapes, or for use in public consultation processes. Salter (2000) discusses the creation of an interactive computer interface to communicate the visual impacts of forest harvesting on the Pitt Lake Slope of the University of British Columbia's Malcolm Knapp Research Forest. This interface combined multiple information sources, such as static landscape visualizations, photographs, interactive panoramic images, animations, and viewshed maps. The intended purpose of this interface was to communicate the visual effects of the research forest's proposed harvesting to interested, non-expert, members of the public. The fundamental difference between this interface and the one proposed for the thesis was the former's focus solely on the visual aspects and impacts of the proposed harvesting. Perhaps the example of previous work most closely related to the intentions of this thesis, is that of the McGregor Model Forest (McGregor Model Forest Assoc, 2004). The Mcgregor Model Forest has created information displays that depict forested landscapes changing over time, through the use of realistic landscape visualizations, 3D drapes of Geographic Information Systems (GIS) data, and dynamic bar graph representations of numerical indicators of landscape condition. The resulting information displays represent real-world landscapes, showing both the temporal and spatial aspects of landscape dynamics, and the effects of management. The main difference between the McGregor Model Forest information display and the interface proposed for this thesis, is that the McGregor display is presented in video clips and on DVD's as pre-recorded animations, whereas the interface proposed for this thesis will provide the user with interactivity, allowing them to control the pace of information display, and to make limited changes to the information they are presented with. One of the common unifying features of the above-mentioned examples is their apparent lack of evaluation with potential users. Information displays, like those discussed above, can be difficult to  15 evaluate, because of the complex nature of users' interactions with them, and the number of information elements and interaction controls they provide to the user. Researchers in other domains have conducted evaluations of subjects' interactions with similar information displays and interactive interfaces (e.g. Ainsworth et al., 2002; Aldrich et al., 1998; Kettanurak et al., 2001; Yeo et al., 2004), however the knowledge domains, and several of the representational forms differed from the previously mentioned forestry and landscape information displays, making it difficult to draw parallels.  1.4  Study Approach  What the studies of information displays from other domains can provide, is some indication of the effectiveness of different information display techniques, and of how subjects may interact with these displays. These studies may also provide evaluation techniques for assessing the effectiveness of interfaces and information displays that provide landscape information. The first step in the study approach was to draw upon the research on information displays to aid in the design of an information interface to communicate spatiotemporal information about the landscape. This interface would be informed by the information display literature, and would utilize a combination of display elements to communicate visible and non-visible indicators of landscape condition over time. In order to truly determine the success of the interface created for this thesis, or any other information display, it is necessary to evaluate it with representative users. One of the keys to this thesis, therefore, was to evaluate the interface constructed for the thesis with users, to determine if it was successful at communicating information about the spatial and temporal aspects of landscape change, and the effects of landscape management on the environment. A review of evaluation methods used to study animations (Lowe, 2003), multiple representation displays (Kozma, 2003) and computer interfaces (Fisher and Sanderson, 1996) led to the decision to use a similar exploratory protocol to evaluate the interface in this study. The lack of research on landscape information displays also encouraged an exploratory approach, in order to elucidate hypotheses for further research. The resulting methodology used a think-aloud protocol and video analysis to capture the subjects' responses to the interface and the study's questions. A questionnaire was used to structure the study sessions, with specific, directed questions and open-ended questions. After the study sessions were completed, the video recordings of the sessions were transcribed and analyzed using MacSHAPA, a video annotation software package.  16  1.5  Thesis Outline  The thesis consists of five chapters in total, including the introduction chapter, and is organized in the following manner: - Chapter two is a literature review of research into the study of information displays. It discusses several available information display types in relation to the data set that was available for this thesis. The purpose of this chapter is to provide a research basis upon which to design the interface.  • Chapter Two  - Chapter three describes the design and implementation of the interface, reviewing the decisions that were made, and the principles upon which the design was based. Chapter three also provides a self-evaluation of the interface, discussing how it measures up to criteria set-out before its construction.  • Chapter Three  - Chapter four reviews the evaluation conducted with human subjects to investigate the efficacy of the interface for communicating landscape information. Chapter four discusses the methods, results and discussion from the evaluation process.  • Chapter Four  - Chapter five is the conclusion chapter, in which the findings of the thesis are revisited, and directions are set-out for future research.  • Chapter Five  1.6  A Note on Conventions Used in the Thesis  There are a number of conventions used in this thesis in respect to language that will benefit from clarification. The use of some words have a specific purpose that the reader should be made aware of before moving into the main portion of the thesis. The words that require clarification are: 1. The Interface - Because the prototypical interface that was created for this thesis will be referenced numerous times throughout the course of the thesis, it will most often be referenced as just "the interface". References to "the interface" in this thesis, therefore refer to the prototypical information display interface described in detail in Chapter Three. 2. Information - The use of the word information in the thesis has a specific connotation. Figure 1.2 describes the relationship between data, information, knowledge and wisdom, as put forth by (Wurman, 2001). Data refers to raw, uninterpreted results, such as a stream of individual temperature readings. Information requires that some form of organization be enforced on that data to facilitate interpretation (Winn, 1995). When data is organized and represented in a table, graph, visualization, etc., it becomes information. The majority of the discussion in this thesis relates to information, because it is organized into various displays and representations. As Figure 1.2 indicates, fostering knowledge of necessity involves the transfer of information. 3. Information Display versus Information Visualization - The term "information display" will most commonly be used in this thesis to describe the different methods of representing information, such as graphs, maps and landscape visualizations. While it is difficult to define, information visualization has a much more specific and constrained usage. Information visualization involves the visualization of large data sets that do not have an inherent spatial component (Ware, 1999). The purpose of information visualization is to capitalize on aspects of the human visual system to help communicate patterns in complex data sources. By this rough definition, information displays such as maps, graphs and landscape visualizations do not fit into the domain of information visualization, and therefore the term "information display" will be used throughout the thesis.  17  Research Creation Gathering Discovery  Presentation Conversation Contemplation Organization Storytelling Evaluation Integration Interpretation Retrospection  Figure 1.2: The understanding continuum. Understanding progresses from left to right, from data to wisdom. 4. Landscape Visualization - Landscape visualizations represent real-world locations with threedimensional perspective views, and varying levels of realism (Sheppard and Salter, 2004). They attempt to accurately depict landscape features and conditions, and are commonly used to represent the impacts of developments and management. Figure 1.1 shows a characteristic landscape visualization image, similar to those used in this thesis. 5. Landscape Preference - Landscape preference as used in this thesis refers specifically to people's aesthetic judgments or preferences for particular landscape forms. Scenic beauty, via the Scenic Beauty Estimation model (SBE) is the most commonly used measure of landscape preference (see Daniel and Boster, 1976). Landscape preferences have had significant implications for forest management, because of their central role in visual resource management (VRM). 6. Landscape Perception - The term landscape perception will be used in this thesis to refer to the broader perception of the landscape and its values. Landscape preference can be thought of as a constituent of landscape perception, as aesthetic values are but one of many values that make up landscape perception. Knowledge, culture, belief systems, etc., can all play a role in landscape perceptions. 7. Non-Visible Indicators - Kimmins (2001) refers to "easily visible indicators" of landscape condition as features that can be observed through visual inspection in midground scenes. Non-visible landscape indicators, then, are those features that are not readily visible in midground scenes. What is visible and non-visible differs between experts and non-experts; the focus here will be on non-visible indicators for non-experts. Features of forested landscapes such as soil nutrient levels, species habitat, forest ages, etc. are not readily visible to non-experts in midground scenes. Non-visible indicators equate to what Gobster (2001) refers to as extrasensory features, and what Eaton (1997) calls "nonperceivables". 8. Learning - Much of the literature to be covered in the Information Display chapter (Chapter Three) comes from the domains of educational psychology, cognitive psychology and instructional science. For this reason, there will be frequent references to learning during the course of the thesis. Learning in this sense refers to the transfer of information, and subsequent acquisition of knowledge. The use of the term learning does not in any way indicate that members of the public need to be taught how they should perceive or interpret the landscape. The provision of information is meant to inform subjects about the dynamics of the landscape and its management, not to attempt to make them value the landscape in a particular way.  Chapter 2  Towards an Interface Design: Information Display Literature Review 2.1  Introduction  Evidence has shown that landscape visualization can be an effective tool for public engagement processes (Al-Kodmany, 2000; Campbell and Salter, 2004; Sheppard, 1989). The strength of landscape visualization, particularly realistic landscape visualizations, appear to lie in their ability to represent real-world locations in a manner that is readily decipherable by people (Sheppard, 1989, 2000). This strength of landscape visualizations, however, is also a limitation if the intent is to communicate information other than that which is readily visible on the landscape. In other words, realistic landscape visualizations excel at representing visible aspects of the landscape, to the exclusion of those elements that are not visible. As outlined in the previous chapter, people have a tendency to judge the landscape based on what they see, rather than on non-visible, cognitive information about landscape condition. This tendency may be affective in nature, and resistant to change. The use of realistic visualization as a primary communication tool plays to this tendency, emphasizing the visual effects of landscape change, to the possible detriment of other important factors. This feature, and apparent limitation, of landscape visualizations was one of the main starting points that led to this thesis project. How could non-visible information sources, such as species habitat, soil productivity, species composition, and other ecological indicators about the landscape's condition be communicated in such a way that their impact and engagement could be increased to closely resemble that of visible attributes in landscape visualization? Also discussed in the previous chapter was the research to date on the effects of cognitive information on people's preferences for landscapes. To this point the research in this area has focused  18  19 predominately on text, presented either visually or verbally (e.g. Anderson, 1981; Brunson and Reiter, 1996; Hines, 2001; Jensen, 2000; Kearney, 2001; Tahvanainen et al., 2001; Taylor and Daniel, 1984). Even those studies that have used visual information in the form of example photographs (e.g. Hines, 2001) have relied upon text-based information for explanation. Given the widely accepted notion that visual information is more easily understood and more powerful for people (Larkin and Simon, 1987; Tufte, 1983), it would appear that the method for providing the non-visible cognitive information about landscape condition puts that information at a distinct disadvantage when compared with the use of photographs and/or landscape visualizations to communicate visual aspects of landscape change. If information about non-visible aspects of landscape condition is to effectively vie for people's attention, then it likely needs to be communicated in a visual way. It could be especially beneficial for this non-visible information to be paired with realistic landscape visualizations to draw upon that medium's demonstrated capacity to engage the viewer. The purpose of the creation of the interface for this thesis was to combine realistic landscape visualizations with other information display techniques to investigate the combination's capacity to improve the communication of non-visible indicators of landscape condition. The intention of pairing realistic landscape visualizations with other information displays of temporal and spatial characteristics of the landscape, as well as non-visible information, was to draw upon the engagement and salience of the landscape visualizations to improve the communication of the non-visible information. The fundamental requirements for the interface were the presence of realistic landscape visualization, a mechanism or mechanisms for displaying non-visible indicators of landscape condition, a means for the user to orient themselves within the landscape, and interactive control, such that the user could determine the information they wanted to view at any given time. Another key component in the creation of the interface was the requirement for a comprehensive data set that could provide the necessary information for the development of both the landscape visualizations, and the associated numerical information about non-visible aspects of landscape condition. The next section of this chapter will briefly discuss the data set used. The data will be discussed in greater detail in Chapter Three; however, in order to bound the discussion of the possible information display techniques, the data sources will be briefly reviewed here. The chapter will then turn to a literature review of the available research on different information display techniques which may be incorporated into the interface design.  2.2  Data: Context for the Interface Design  The construction of the interface required a comprehensive data set that could provide the necessary information to construct landscape visualizations as well as other information display components.  20 The intention to make the interface both temporally and spatially interactive also required specific attributes from the data set. Communication of landscape dynamics requires both spatial and temporal components, because of the scales (both temporal and spatial) at which ecological processes occur (Forman, 1995; Kimmins, 1999). In order to project landscape condition over time, it was necessary to have information from modeled sources that could predict future conditions. In order to portray information about a specific geographical location, it was necessary for some of the data to be derived from spatially explicit models. The landscape visualizations would require both spatial and temporal data to show the visible effects of landscape change over time. The information display component(s) required a temporal component, but did not specifically require a spatial component, depending on the display technique that was selected. The availability of spatial information for the information display component(s) would increase the range of possible display techniques, and could potentially improve the connection between the data about non-visible indicators of landscape condition, and the visible changes on the realistic landscape visualizations. Finally, all of the data to be used in the creation of components for the interface needed to be unified into a common framework, such that it covered the same spatial and temporal extents, over the same timescale. In other words, in order for the interface to provide a cohesive picture of what was occurring on the landscape, all of the information sources needed to cover the same information. The Arrow Forest District's Innovative Forest Practices Agreement (IFPA) project, and specifically the University of British Columbia Faculty of Forestry's IFPA sustainability project provided a unique opportunity for acquiring the necessary information to populate this interface, as it represented a groundbreaking interdisciplinary approach to the integrated modeling and visualization of forest conditions over time at the landscape scale. The Arrow IFPA project was undertaken by forest licensees and Forest Renewal British Columbia (FRBC) in the Arrow Forest District in southeastern BC. As part of the project, researchers from the UBC Faculty of Forestry were enlisted to conduct a unified modeling effort to investigate different tactical strategies for the management of forest resources in the Arrow Forest District. In order to standardize the data for the various models to be used, the Lemon Landscape Unit (see Figure 2.1) was chosen as a test-case for the model integration procedures. Over the course of the project three management scenarios were constructed for the Lemon Landscape Unit. The construction of these management scenarios, and the unified modeling of their ramifications for the landscape unit, provided the opportunity to populate this thesis' interface with the necessary data, for both the landscape visualizations and the non-visible indicators of landscape condition. 2.2.1  The Arrow IFPA Modeling Process  Figure 2.2 provides a schematic representation of the various modeling platforms, as well as their integration for the unified modeling effort and visualization generation conducted for the Arrow IFPA.  21  Figure 2.1: A map of the Lemon Landscape Unit in the Slocan Valley of southeastern British Columbia. This subsection will discuss the contributions of the different modeling packages to the overall modeling effort, and give a brief overview of the data types provided by each of the models. FORECAST FORECAST is a non-spatial model of forest stand attributes. FORECAST combines bioassay methods for the prediction of stand yield, with light and nutrient simulations that help determine future stand characteristics (Kimmins et al., 1999). FORECAST provides tabular data for a particular forest stand, or group of forest stands, but does not provide information about where those stands are geographically located, or about where individual trees are located within the stand. Data available from FORECAST include such stand attributes as tree species composition, stand density, tree height, soil nutrient composition and understory vegetation composition. Figure 2.2 depicts the FORECAST model's relationship to the other models in the Arrow IFPA modeling project. FORECAST was used to determine the proper tree images to be used in the landscape visualizations via a tree im-  22 age lookup file (Cavens, 2002), and a library of tree images. FORECAST data was also used to provide stand attribute information to the ATLAS landscape-level forest harvesting model.  FORECAST S t a n d Attribute Model  Library o f tree i m a g e s  I  CALP Visualization System  GIS data (terrain, streams, lakes, roads, etc.)  I  Renderer  Still Images and Animations  (World Construction Set)  ATLAS  Simfor  Harvest S c h e d u l i n g  Species Habitat  Model  Model  Figure 2.2: A schematic representation of the model relationships for the Arrow IFPA modeling project.  ATLAS ATLAS is a spatially explicit model that schedules forest harvesting in spatial polygons according to stand attributes and a series of prescriptive constraints, such as cutblock adjacency rules and visual impact restrictions (Nelson, 1999). In the Arrow IFPA modeling project, ATLAS derived its stand (or block) attributes from data provided by FORECAST. These stand attributes were used in combination with harvesting rules based on the tactical management scenarios created by the UBC sustainability project team. ATLAS then provided a spatial representation of landscape condition and tabular data on polygon attributes over a 265 year time-scale at decade-long intervals as Geographic Information System (GIS) map layers. SIMFOR SIMFOR is a landscape and habitat indicator model that allows users to model the effects of forest harvesting scenarios on particular landscape characteristics and species' habitats (Wells and Moy, 2002). In the Arrow IFPA project, SIMFOR took input data from ATLAS about harvest scheduling, and stand attribute data from FORECAST as its inputs. SIMFOR provides outputs in the form of a database, and spatially explicit raster GIS layers. Examples of SIMFOR outputs include predicted serai stage distributions for forest types, and song bird species' habitats. CALP Visualization System The CALP Visualization System (CVS) was created as a means of importing model data into a landscape visualization rendering package, so that the resulting realistic landscape visualizations could be produced based on the data provided by FORECAST, ATLAS and  23 SIMFOR. Cavens (2002) provides a detailed description of CVS, and its relationship to the ecological models used in the Arrow IFPA project. The CALP Visualization System is a series of Perl scripts and lookup files that translate model data into formats that are understandable by the associated landscape visualization rendering engine (Cavens, 2002). In the Arrow IFPA project the renderer used was World Construction Set (3DNature LLC, 2000), though CVS is not tied specifically to that package. FORECAST stand attribute information was used to determine the proper tree images to use, the relative proportion of each species in a particular stand, the density of trees in a stand, and the height of trees in a stand. ATLAS provided the spatial information about where stands were located on the landscape, as well as the schedule of harvesting and regrowth (driven by FORECAST stand curves). As a result, it was possible to depict realistic visualizations of the Lemon Landscape Unit, showing the harvesting and regrowth pattern over time. Other Visualization Information In order to complete the landscape visualizations, it was necessary to incorporate geographic information not provided by the ecological models. The GIS data required for the landscape visualizations was acquired from the BC Ministry of Environment, Lands and Parks' (MOELP) TRIM data, and the BC Ministry of Forests' (MOF) Forest Cover Data. TRIM data provided the necessary base data, such as elevation contour lines for the creation of the Digital Elevation Model (DEM), and lake, stream and road locations. ATLAS modeled only the productive forest land for the Lemon Landscape Unit, leaving other land cover types out of its calculations and predictions. This meant that a significant portion of the landscape unit did not have vegetative cover information necessary for the realistic landscape visualizations. Forest cover data from the MOF was used to fill in the land cover gaps presented by the ATLAS database. Forest cover mapping provides basic information on non-productive forest land, categorizing it into general types such as non-productive brush, rock, alpine forest, alpine, etc. These general types were used to create representative ecosystems in World Construction Set. These ecosystems did not change over time, unlike the productive forest ecosystems that were driven by the ecological models, but they did prevent the appearance of obvious data holes in the landscape visualizations. SIMFOR data could not be directly tied to the visualization images through CVS, and therefore was added using post-processing in Adobe Photoshop. Raster files generated by SIMFOR were converted into bitmap images, imported into World Construction Set and rendered against a black background for each of the selected viewpoints at each of the time periods. The resulting images depicted the SIMFOR raster information in a given colour conforming to the terrain data against a black background (see Figure 2.3a). Using Adobe Photoshop, the SIMFOR colour information was then selected out from the black background and overlayed onto the appropriate landscape visualization image (see Figure 2.3b). The resulting image depicts the location of SIMFOR data draped at the proper locations on the landscape visualization images.  24  Figure 2.3: The SIMFOR visualization process. Figure a) shows the SIMFOR raster data rendered in World Construction Set against a black background. Figure b) shows the completed SIMFOR image after it has been photomontaged in Adobe Photoshop (Adobe Systems Corp., 2002b). Available Data Set  From the various data sources discussed above, there were a number of possible data inputs to the interface. The available inputs derived by the integrated modeling included: • Ratio level numerical data from FORECAST and ATLAS on stand and landscape-level attributes (e.g. timber volume, forest age classes, area harvested, tree species distributions, etc.). • GIS data from the MOELP and the MOF, as well as from the spatial representations of both ATLAS and SIMFOR data. • Landscape visualizations from a number of different viewpoints, including both realistic landscape visualizations, and draped visualizations of SIMFOR data. With the exception of the provincial GIS data, all of the data were modeled, and therefore projected possible future conditions over the same time scale, and the same geographical area. Basing the interface on this data set, then, meant that the data available for presentation in the interface included landscape visualizations, spatial GIS information, and ratio level numerical information. The unifying theme of the available data was that they were landscape based, and ecological in nature. This theme would play a defining role in the construction of the interface, and the selection of the presentation mechanisms. The remainder of this chapter will review the relevant research on information display techniques, keeping in mind the data made available by the Arrow IFPA project discussed in this section.  2.3  Information Display Techniques Literature Review  Given the intended purpose of the interface, and the available data sources, the next step in the process was to design the interface itself. Due to the rapid progress of technology in computer related fields  25 (specifically visualization), development is often driven by what is possible, rather than what research has shown to be effective. From the inception of this thesis project, the intention was to draw upon research about information display techniques wherever possible, design guidelines where available, and intuition only as a last resort. The desire to draw upon research and established design guidelines led to an extensive literature review in a wide array of research areas. With the exception of research on realistic landscape visualizations, the predominate unifying feature of research into information display techniques has been a cognitive focus. Research into graphs, maps, animation, diagrams, interactivity, etc., has focused on the use of cognitive psychology principles and research techniques as methods for evaluation. This section of the chapter will, therefore, begin with a discussion of concepts of cognitive psychology that innervate the research on information display techniques, and create a theoretical basis for understanding the research results. The section will then turn to a review of the available literature on information display techniques pertinent to the design and construction of the interface. These techniques include: landscape visualization, information visualization, graphs, maps, text, dynamic representations, interactivity, and multiple representations. Where possible, the literature in these subject areas will be discussed as they relate to the cognitive psychology concepts that begin this section. The findings in this literature review relate to what is generally known about the application of the various information display techniques as a foundation for the design of the interface. A more specific application of the findings of the literature review to the actual design of the interface will follow in the Interface Design and Implementation Chapter. 2.3.1  Cognitive Psychology Concepts  As mentioned above, there are a number of key cognitive psychological concepts that arise continually in the relevant research, and which have formed the basis of research in the areas of information presentation, multimedia instruction, dynamic visualization and associated fields. Primary among those concepts are working memory, schema theory and cognitive load. Working Memory  The most fundamental concept underlying psychological research into learning and instruction is working memory. Figure 2.4 shows a schematic representation of the current conception of the memory systems pertinent to information display in the human brain. This figure will serve as a guide for the discussion of most of the cognitive psychology theories to be reviewed in this section, indicating the crucial role that memory systems (and particularly working memory) play in the perception and comprehension of information presentations. As can be seen in Figure 2.4, visual and verbal information from the outside world enters into the memory system through the eyes and ears and their temporally constrained sensory memory stores. If  26  Information Presentation  Iconic Memory  Visual Information  Eyes  Working Memory  Long-Term Memory  r! Visuospatial I Y ^—•' Sketchpad i  Selecting j Images  x  ,  1  J  R  '  Central Executive ^  Verbal Information  Ears  Selecting | Words  'TO ,  •  Input  Storage  Prior Knowledge/ Cognitive Schemas  Phonological! Loop  Figure 2.4: A schematic diagram of the human memory systems applicable to learning from text and graphics. this information is not passed on from sensory memory into working memory it will decay in a fraction of a second (Ware, 1999). As the figure depicts, the transfer of information from the sensory memory store to working memory involves the process of selecting images and words (or sounds) (Mayer and Moreno, 2002b). This selection process is driven by the attention of the individual perceiving the outside world, and in the working memory model is controlled by the central executive (Baddeley, 1992, 2003). The central executive is the site of manipulation, rehearsal and reasoning about information that is held in working memory, and is also considered to be the site of consciousness (Sweller et al., 1998). The central executive is the feature of the working memory model that most distinguishes it from earlier conceptions of short-term memory (Baddeley, 1992). Working memory is separated into three primary components (see Figure 2.4), the previously mentioned central executive, a visual memory store called the visuospatial sketchpad and an auditive memory store called the phonological loop (Baddeley, 1992, 2003). In a seminal paper by Miller (1956) it was postulated that approximately seven items (or chunks as Miller referred to them) of information was the functional limit of short-term memory. While there has been significant development in the understanding of working memory since then, this number has remained remarkably prescient (Baddeley, 1992). More recent work has built upon Miller's early research, indicating that the number of items that can be stored decreases significantly when manipulation of the information is required, likely to two or three items (Sweller et al., 1998). There is also evidence that the phonological loop and the visuospatial sketchpad have individual limits of approximately three to four items each (Baddeley, 2003), meaning that to tap the full capacity of working memory it is necessary to provide information in both visual and auditive formats. During reasoning in working memory, information can also be pulled in from long-term memory (see Figure 2.4), allowing previously acquired knowledge and cognitive schemas to be used (Ware,  27 1999; Mayer and Moreno, 2002b). As with information drawn from sensory memory, this information must be actively held, requiring attention and a portion of the limited resources of working memory. Information held and rehearsed in working memory can also move in the opposite direction, passing into long term memory for storage. The functioning and limitations of working memory are crucial to the effective and efficient transmission of information from the outside world to the individual for learning. Data presenters and information designers need to take the features of the working memory system into consideration when constructing their information displays, in order to ensure that viewers of their displays are not negatively affected by the limitations of working memory. The remaining cognitive phenomena to be discussed in this section are all derived from features of this memory model, and relate to limitations and opportunities that the model presents for the design of information presentation. Schema Theory  Schema theory was devised by Bartlett (1932) as an explanation for the storage of information as knowledge in long-term memory. "A schema is defined as a cognitive construct that permits one to treat multiple elements of information as a single element categorized according to the manner in which it will be used" (Marcus et al., 1996). Storage of schema in long-term memory can be thought of as a network of interconnected nodes of information, with smaller, lower-level schema combining to form higher-level, more complex schema. A common example of a schema is a person's conception of a restaurant (Armbruster, 1986). When the concept of a restaurant is introduced to people who have visited one, it conjures up images and associations of what one would expect to see in a restaurant: tables, servers, dishes, food, etc. Even though there are many elements of information involved in the schema of the restaurant, the schema itself can be held as a single element in memory. An example more tightly related to this thesis is the commonly held (in western society) schema for graph reading. Users of graphs have a schema indicating to them the purpose of the axes, the labels and the graphical representations in the coordinate space. The schema for graphs means that the user does not have to re-learn the conventions of a graph every time they see one, and therefore can focus on the information the graph presents. Schemas can be drawn from long-term memory into working memory for active processing of new information (see Figure 2.4). A schema can hold a huge amount of interconnected information, but it can be held as a single element in working memory (Kirschner, 2002). This allows for a significant reduction in the working memory load, and increases the total amount of information and processing that can occur in working memory at a given time. Expertise is thought to reside in long-term memory as a result of the combination of low-level  28 schemas into ever-more complex schemas (Sweller et al., 1998). Automation of these schemas can reduce working memory load even further by allowing schemas to be processed without the need for conscious effort (Pollock et al., 2002; Marcus et al., 1996). Automation occurs when an element of information that at one time required conscious attention to process no longer requires that attention. A good example is letter and word learning in children (Marcus et al., 1996). When children first learn letters, they must consciously process the shape of each letter. With practice they are able to perceive the letter as a single unit, without consciously processing the shape. When they move on to words, they initially see the word as a series of individual letters, until, again with practice, they come to recognize the whole word (letter recognition having become automatic). Learning to read means automating schemas for both letter and word recognition (Marcus et al., 1996). Highly complex and automated schemas, then, are the hallmark of expertise in a given domain, explaining why, after years of practice in a given field, an individual can undertake extremely complex tasks with little or no conscious knowledge of the operations they are performing. The promotion of schema generation is the key to designing materials intended to communicate new information to people. Fundamental to the fostering of schema creation is management of the limited resources of working memory, so that novel information can be processed in working memory and transferred into long-term memory as new schemas. The presence or absence of appropriate schemas to aid the assimilation and processing of new information is, therefore, a key determinant of cognitive load (Marcus et al., 1996). Cognitive Load Theory  Cognitive load theory is "a theory that emphasizes working memory constraints as determinants of instructional design effectiveness" (Sweller et al., 1998). The theory states that material that exceeds the capacity limits of working memory will be difficult to learn, because it will not allow sufficient resources to be applied to processing information and constructing schemas. Cognitive load can be separated into three different types: intrinsic, extraneous and germane (Sweller et al., 1998). Intrinsic cognitive load is the result of the inherent properties of the material to be learned. The most important aspect determining the intrinsic load of material is the amount of interactivity between its elements (Sweller et al., 1998). Information that requires the learner to simultaneously process multiple elements is high in element interactivity, and consequently high in intrinsic cognitive load (Carlson et al., 2003). Learning chemical symbols from the periodic table can be done serially (one at a time), and therefore is low in element interactivity, while learning the rules of sentence syntax requires the learner to simultaneously process multiple words and rules, and therefore is high in element-interactivity (Sweller et al., 1998). Truly intrinsic cognitive load is a function of the material itself, and is not amenable to change through alterations in instruction or information presentation  29 techniques. Extraneous cognitive load is the result of the presentation of material, or the activities undertaken in learning (Sweller et al., 1998). Extraneous cognitive load can be raised or lowered depending on how material is presented, and therefore is the primary focus of cognitive load theory. An example of extraneous cognitive load is the inferiority of text presentation of complex information when compared with diagrammatic presentation (Larkin and Simon, 1987). Several studies have shown that diagrams are superior to text when the material presented is complex (e.g. Carlson et al., 2003; Marcus et al., 1996; Mayer and Gallini, 1990). In other words, diagrams decrease the extraneous cognitive load, so that when intrinsic cognitive load is high, the combination of intrinsic and extraneous cognitive loads do not overwhelm working memory resources. When the intrinsic cognitive load of material is low (i.e. the information is not complex), representing the material as a diagram does not have a significant benefit because working memory resources are adequate for the task (Carlson et al., 2003). The final type of cognitive load is germane cognitive load. Germane cognitive load is caused by processing in working memory geared towards the creation of schemas (Sweller et al., 1998). When intrinsic and extraneous cognitive loads are sufficiently low, surplus working memory resources exist, and these resources can be used to process the incoming information for incorporation into schemas. The goal of applying cognitive load theory to the presentation of information is to decrease extraneous cognitive load to provide more resources for learning via germane cognitive load (Bannert, 2002). Research into the efficacy and application of cognitive load theory has looked at several different techniques for decreasing extraneous cognitive load. Studies have looked at the aforementioned use of diagrams (e.g. Carlson et al., 2003; Marcus et al., 1996; Mayer and Gallini, 1990), the use of example problems, or partially completed problem sets (e.g. Paas, 1992), the use of verbal as opposed to visual presentation of text (e.g. Jeung and Chandler, 1997; Kalyuga et al., 1999), the use of colour coding in text and associated diagrams (Kalyuga et al., 1999), and the integration of disparate data sources (i.e. text passages and illustrations) into cohesive representations (e.g. Sweller et al., 1990; Chandler and Sweller, 1991). As a result of the research into instructional techniques and cognitive load, several effects have been noted. Sweller et al. (1998) in their comprehensive review of cognitive load theory, have documented a number of general principles that have been elucidated by the research. While some of these principles do not directly apply to the contents of this thesis, others have a direct bearing on the design of the interface. The relevant effects are: • The Split Attention Effect - This effect occurs when different pieces of material that are necessary for comprehension of an instruction are presented in a non-integrated format. The classic example is the provision of a diagram tied to a text passage, where neither the diagram, nor the text passage is comprehensible without the other. Remedying the split attention effect normally involves the integration of material (e.g. the provision of text annotations on the diagram). • The Modality Effect - Based upon the dual memory structures (the phonological loop and vi-  30 suospatial sketchpad) found in working memory, it has been postulated that working memory capacity can only be fully utilized through the simultaneous use of verbal and visual information (Paivio, 1986). This effect suggests that when visual information such as illustrations or diagrams are presented, it is preferable to present accompanying text in a verbal, rather than a visual format. Several studies have been conducted which support this effect (e.g. Jeung and Chandler, 1997; Kalyuga et al., 1999; Mayer and Anderson, 1992; Mayer and Moreno, 2002a). • The Redundancy Effect - The redundancy effect states that when two (or more) pieces of material covering the same topic are fully intelligible on their own, then presentation of more than one of the sources will interfere with learning and understanding. This effect may seem counterintuitive, as the provision of material in multiple formats might be thought to provide a more comprehensive view of the information, however research into the redundancy effect has thus far corroborated its impact (e.g. Chandler and Sweller, 1991; Mayer et al., 2001). Instructional designs, and information displays should take these effects into consideration in order to minimize cognitive load and improve the transmission of information. To summarize, then, cognitive load deals with the limitations of working memory and with instructional mitigations which can be taken to compensate for those limitations. Intrinsic cognitive load is a function of the complexity of the material to be learned, and is not amenable to change. Extraneous cognitive load is a direct result of the method of presentation of information, and is the primary focus of the application of cognitive load theory to information presentation and instruction. Finally, germane cognitive load is the load generated by working memory processing that creates new schemas from incoming information. The main purpose of applying cognitive load theory is to decrease the extraneous cognitive load, freeing up resources for germane cognitive load and the consequent formation of schemas.  2.3.2  Landscape Visualization  The inclusion of realistic landscape visualizations was a precondition of the creation of the interface, due to their effectiveness in engaging non-experts in land-use planning processes (Al-Kodmany, 2000; Campbell and Salter, 2004; Sheppard, 1989). The use of landscape visualizations in the interface was intended to foster this engagement and provide a readily understandable medium for assessing the visible aspects of landscape change over time. It was hoped that the engagement and understandability of landscape visualizations would in turn improve the communication of the non-visible information about landscape condition included with the landscape visualizations in the interface. Uses of Landscape Visualization  While recent technological advancements in visualization software and computer hardware have increased the prevalence of landscape visualizations, as well as their possible applications, simulation of possible future landscapes is not a new phenomenon (Zube et al., 1987). Landscape simulations have been in use for centuries, and date back to the early Egyptians and Chinese (Zube et al., 1987).  31 One of the most famous applications of landscape simulations was Humphrey Repton's use of hinged illustrations to depict before and after representations of proposed developments in the 1700's. While the accuracy of many of Repton's representations have been questioned (e.g. Tufte, 1997), the basic premise on which they operated (the concurrent presentation of both existing and proposed conditions) presaged the dominant use of simulation media for centuries to come. Landscape simulations in resource management have predominately been used to convey information about the design and visual impacts of proposed developments or management actions (Appleyard, 1977; Sheppard, 1989; Zube et al., 1987). Visualizations are well suited to this task, due to their ability to portray visual impacts in a concrete manner that is readily understandable by non-experts. Other factors which have led to the limitation of landscape visualizations to static before and after representations of proposed developments have been the laborious and time-consuming process of constructing some of the dominant simulation media such as photomontage and physical 3D models, as well as the lack of explicit ties to underlying data. With the rise in availability, and improvements in technology, of computer-based visualization tools, new opportunities have arisen for the application of landscape visualizations. Efforts to link visualizations to underlying data from ecological models have led to more defensible visualizations, as well as providing new opportunities for viewers to query underlying data to improve transparency (e.g. Cavens, 2002; Orland, 1994). These explicit data ties, as well as advancements in visualization rendering and production, mean that new avenues have arisen for the use of landscape visualizations. These improvements in landscape visualization capability have coincided with the increased recognition being given to non-commodity forest values discussed in Chapter One, and the resulting more inclusive approach to forest management. Research into the use of landscape visualization as a mechanism for communicating the management effects of these more inclusive strategies is in its infancy. In the absence of extensive research on the use of landscape visualizations as communication mechanisms for larger (i.e. not strictly visual or scenic) management implications, there has been significant discussion and conjecture on their potential for this task. Bell (2001), and Daniel (2001b) both indicate that visualizations have the capacity to be used in public consultation beyond the conventional application as predictors of visual impact. Both authors stress the ability of landscape visualizations to depict landscape patterns as a means of communicating important ecological factors of the landscape in a salient manner. Daniel (2001b) further indicates that landscape visualizations can be used to communicate information about landscape changes at small spatial scales over long time periods, more appropriate to the scales at which ecological processes operate. This ability to communicate information about ecological patterns may be a valuable means of informing the public about the effects of current management options on future generations and future landscapes. Luymes (2001), and Sheppard (2004) have discussed the rhetorical strength of landscape visualiza-  32 tion images in shaping public opinion about landscape management options. Luymes (2001) discussed this rhetorical use in the context of forest management, and stressed the need for accompanying, transparent information to document the simulations' validity. Sheppard (2004) discusses landscape visualizations as a possible communication mechanism to convey the impacts of climate change, again drawing upon their ability to portray modeled data over long time scales in a manner that is readily understandable to non-expert users. Both authors acknowledge the ethical concerns of using landscape visualizations for rhetorical purposes, and stress the need for these simulations to be defensible and accurate (Sheppard, 2001b). Given the potential ramifications of environmental issues such as climate change, as well as the discrepancy between the scientific evidence and the coverage by mainstream media (see e.g. Boykoff and Boykoff, 2004), Sheppard (2004) argues that the possibility of using tools such as landscape visualization for influencing public opinion should, at least, be discussed. Finally, Cavens (2002) discusses the use of landscape visualization as a tool for collaborative design. He points out that the current set of design tools, such as Computer Aided Design (CAD), are cumbersome to use and require specific, expert knowledge to operate. Cavens (2002) postulates that by providing some basic interaction tools with the more easily understood metaphor that landscape visualizations provide, non-expert users can be involved in the design process. In essence the tools get out of the way and allow participants in a collaborative process to focus on the design. Types of Landscape Visualizations  Landscape visualizations are not a homogenous set of representations, with only one way to represent the features of the landscape. There are, in fact, a number of different landscape visualization technologies that represent the landscape in fundamentally different ways. McGaughey (1998) describes four main types of landscape visualization techniques. Those techniques include: - This technique uses computer programs such as Adobe's Photoshop to alter existing photographs to depict the possible impacts of proposed developments. This technique represents a digital evolution of previously existing techniques for the manual rendering (i.e. "painting"), and photomontage of photographs to accomplish the same effect (Sheppard, 1989). See Figure 2.5a) for an example of a video imaging simulation.  • Video Imaging  - Uses the same basic techniques as video imaging with the key difference being geometric video imaging's use of mathematical models of terrain, and/or terrain objects to accurately place photomontaged elements in a photograph. In this manner cutblock boundaries, proposed buildings, even individual trees, can be accurately positioned on a photographic background without the use of measurement and perspective calculation from the photographs themselves. See Figure 2.5b) for an example of a simulation created using geometric video imaging.  • Geometric Video Imaging  - Image draping utilizes geographic information associated with 2D images (orthophotos, satellite imagery, GIS layers, etc.) to "drape" those images over a 3D Digital Elevation Model (DEM) of the same location. This technique allows for the accurate 3D representation of geographic elements, with some associated distortion due to the lack of a third-dimension (e.g. the lack of 3D trees and buildings means that the visible area from a given viewpoint may be  • Image Draping  33 overestimated by the lack of occlusion). Figure 2.5c) shows an example of an image draped simulation. • Geometric Modeling - This is the most data-intense of the four main landscape visualization techniques. Geometric modeling uses mathematical functions to locate and render all of the elements of a landscape visualization, including the terrain, vegetation, buildings, water, sky, etc. While photographic elements such as trees, ground cover, and building facades may be used as textures in a geometrically modeled simulation, their placement, orientation and sizing are all determined by the mathematical relationships created by the computer software. Figure 2.5d) is an example of a geometric modeling simulation. While these four landscape visualization techniques represent the main categories, there are a number of different hybrid systems and techniques which allow visualization creators to combine strengths of the different methods. For example, certain image draping software packages allow for the placement of 3D elements such as trees and buildings, and geometric modeling simulations can provide improved realism by photomontaging foreground elements into the final image. Evaluation of Landscape Visualization  There has been limited empirical research into the efficacy and use of realistic landscape visualizations (Tyrvainen and Tahvanainen, 1999). To date research on landscape visualizations has focused primarily on realism, response equivalence and representational validity. This is likely a progression from previous research on the representational validity of photographs as a necessary precondition to their use as surrogates for real world locations in landscape preference research (see for example Shuttleworth, 1980; Stamps III, 1990). As mentioned earlier, the predominant use of landscape visualizations to date has been in the representation of development impacts, and therefore the elicitation of preference. In order to determine if landscape visualizations are viable surrogates for real world landscapes in preference research, their level of perceived realism and their representational validity need to be established. Different researchers have focused on different aspects of the realism/response equivalence/ representational validity questions. Appleton and Lovett (2003) and Lange (2001) investigated perceived levels of realism of landscape visualizations in response to various image elements. Lange (2001) found that draped orthophotography was perceived as highly realistic from background views, but less realistic from foreground views. Orthophotographs were rated as more realistic than the use of shaded relief texturing of the same landscape elements. Lange (2001) also found that geometric elements (such as the presence of 2D/3D trees) and buildings increased ratings of realism. Appleton and Lovett (2003) also investigated realism, but focused on looking for a 'sufficient' level of realism in landscape visualizations that would allow people to make decisions about the represented environment. For their study they used World Construction Set (3DNature LLC, 2000), a highly realistic and specialized landscape visualization package that is a true geometric modeler (allowing for the placement  34 of either 3D objects or 2D 'billboarded' objects). Appleton and Lovett (2003) did not find a 'sufficient' level of realism, but did detect a focus by image viewers on ground texture (specifically vegetation), especially in the foreground of the image. Bergen et al. (1995), and Daniel and Meitner (2001) used the Scenic Beauty Estimation (SBE) model, discussed in Chapter One of this thesis, to investigate the representational validity of landscape visualizations. The SBE is an established metric of people's scenic ratings for landscapes, and has been used in previous research to validate the use of photographs as surrogates for real world scenes. Comparison of landscape visualizations of a given location with photographs of the same location using the SBE method would then indicate the representational validity of the visualization images. Bergen et al. (1995) compared photographs of landscape scenes with Vantage Point images of the same scenes. Vantage Point was their own geometric modeler that, at the time, was limited to predominately 3D massings, without the ability to place 3D objects, such as foreground vegetation. Significant correlations were not found between the photographs and the Vantage Point images on SBE rating. The authors attributed the lack of correlation to the early prototype nature of Vantage Point, and its omission of such important image element capabilities as foreground vegetation, similar to the findings of Appleton and Lovett (2003) discussed above. Daniel and Meitner (2001) took a different approach to assessing the representational validity of landscape visualizations. While they also used the SBE method to investigate representational validity, their methodology focused on the alteration of realism levels for photographs of foreground landscape scenes. Subjects were presented with full colour, black and white, four bit colour, and black and white sketch (using a digital image software to create the 'sketch' effect), representations of the same scenes, and asked to give their SBE ratings. Daniel and Meitner (2001) found that SBE rating levels correlated poorly between the full colour images and each of the subsequent realism levels (grayscale, four bit colour, and black and white sketch), and suggest that this has implications for the effects of graphic realism on the representational validity of landscape visualizations. Sheppard (1989) investigated the response equivalence of a wide variety of landscape visualizations, such as photosimulation and computer line drawings, to photographs of the same scenes, and found generally low levels of response equivalence. Sheppard (1989) also found that response equivalence had a weak relationship with image accuracy, suggesting that some other factors were potentially influencing subjects' ratings. Bishop and Rohrmann (2003) also looked at the response equivalence of simulations, but compared them to the real world locations. For animated scenes of urban environments they found moderate levels of response equivalence between the simulations and the real world. Given the variations in technique, and the very different visual results that these techniques produce, it is difficult to draw conclusions about the response equivalence/realism/representational validity of  35 landscape visualizations as a whole. This suggests that researchers should investigate the representational validity of the particular technique they plan to use in future perception research before using visualizations to investigate scenic beauty, preference, acceptability, or other measures of visual perception of landscapes. Landscape Visualizations from a Cognitive Perspective  While the ability of landscape visualizations to act as surrogates for real landscapes, or photographs of those landscapes, is very important for perception and preference research, it is less important from a cognitive perspective. Abstract forms of representation such as maps, graphs, diagrams, charts, etc., are regularly studied in cognitive research as media for knowledge transfer and acquisition. Abstract representations of objects and concepts date back millennia, and in some specialized fields such as architecture, physics or chemistry can become highly stylized; almost indecipherable by non-experts. Landscape visualizations, then, do not necessarily need to have a one-to-one relationship to real world scenes in order to be useful for learning about those scenes, nor to study their effects on learning landscape-related information. A review of the available literature found no research on the cognitive effects of realistic landscape visualizations on information transfer. Cognitive research on other representational forms, as well as the theoretical underpinnings of the study of media, can provide some speculative basis for their effectiveness. From a theoretical cognitive perspective, high realism and response equivalence with photographs or real-world scenes should activate existing schemas for real landscapes, and/or photographs of landscapes. Schnotz and Bannert (2003) state that realistic pictures can make use of schemas from "everyday perception", whereas more abstract images must rely on schemas geared towards their specific symbolic representation. Experience of various landscape types, as well as landscape photographs, are prevalent enough that all sighted people likely have schemas for the understanding of landscapes (with the necessary caveats for local landscape variations, and cultures that do not possess or use photographs). Provided realistic landscape visualizations are 'close enough' representations of real world landscapes, they should activate these schemas, facilitating understanding of what the landscape visualizations portray by reducing cognitive load and freeing working memory capacity for active processing. Variations in the ability to utilize landscape visualizations to interpret information about the represented landscape could then be explained by the development of the individual's 'landscape schema'. A resource professional, such as a forester or ecologist, should have a much more developed schema for landscapes, with extensive links tying visual landscape patterns to probable ecological, hydrological, geological, etc. relationships. These more developed schemas, then, would free up more working memory capacity, and allow experts to derive complex information from realistic landscape visualizations. This theoretical explanation for the effectiveness of realistic land-  Figure 2.5 a: An example of video imaging. The building in the foreground is photomontaged into an existing background photograph to simulate the visual impact of a development proposal. Image courtesy of Grant Wilder, Salter Pilon Architects, Inc.  Figure 2.5 c: An example of image draping. A black and white aerial orthophoto is draped upon a 3D digital elevation model (DEM) using geographic coordinates.  Figure 2.5 b: An example of geometric video imaging. A 3D CAD model was used to accurately place the cutblocks which were then created using digital photo-painting techniques. Image courtesy of John Lewis, CALP, UBC.  Figure 2.5 d: An example of geometric modeling. All of the elements in the image (including sky, clouds, vegetation, water, etc.) are created by mathematical functions based on data provided by Geographic Information Systems (GIS).  Figure 2.5: Examples of the four types of landscape visualization as described by McGaughey (1998).  37 scape visualizations suggests that they could potentially be better for providing information about landscape conditions than novel display techniques (e.g. information visualization techniques) that lack an existing cognitive schema, and therefore exert extraneous cognitive load on the user, limiting the working memory resources required for information processing. Though research has not been conducted to test this theoretical explanation for the effectiveness of landscape visualizations, research into other representational forms from cognitive and educational psychology suggests how this could be performed. Realistic landscape visualizations could be compared with other representations of landscape information, such as 3D geovisualizations, 2D maps, diagrams, text passages, etc. using well developed methodologies to investigate subjects' ability to recall information and transfer it to novel problems (e.g. Mayer and Gallini, 1990; Mayer and Chandler, 2001; Moreno et al., 2001; Moreno and Mayer, 2002; Paas, 1992; Stern et al., 2003). Standardized techniques have also been developed to measure the cognitive load caused by representational media, and their instructional efficiency (e.g. Carlson et al., 2003; Marcus et al., 1996; Paas, 1992; Pollock et al., 2002; Tabbers et al., 2004). These standardized techniques could be applied to studies investigating the information transfer capabilities of landscape visualizations to determine if they decrease cognitive load, and consequently improve learning. While psychophysical and response equivalence approaches to evaluating landscape visualization are appropriate when the intention is to determine if landscape visualization images evoke the same perceptual and emotional responses as real-world landscapes, they do not explain the effectiveness, or lack thereof, of landscape visualizations in communicating cognitive information. If the intention is to communicate information other than the scenic or aesthetic attributes of the landscape, as discussed in Chapter One by authors such as Gobster (1994,1995,1996,1999), Thayer (1989), and Daniel (2001b,a), then a cognitive psychology methodology is more appropriate. If the ability of landscape visualizations to communicate cognitive information can be established, as well as their ability to operate as surrogates for real landscapes and photographs in perception research, then some of the large questions about the ability of cognitive information about landscape conditions to alter people's perceptions of land-use management, discussed at the end of Chapter One, can be addressed using the same media to represent both information streams. Draping of Spatial Overlays: Combining Realistic Visualizations with Geovisualizations  Draping refers to the conforming of spatial data in either vector (point, line or polygon) or raster formats onto a 3D surface; in the case of landscape visualization the 3D surface is the terrain. Draping provides a means of locating landscape information spatially in 3D. Figure 2.3 is an example of draping, with Swainson's thrush (Catharsus ustulatus) habitat draped onto a realistic landscape visualization image. Draping allows for the 3D spatial representation of information that may not otherwise  38 be visible on the landscape, and is a common practice in more abstract geovisualization applications. Similar to information visualization (discussed in the next section), geovisualization has received little investigation (Haug et al., 2001), and has been driven primarily by technology development, rather than demonstrated effectiveness. Unlike landscape visualization, geovisualization has been focused primarily on provision of expert tools for data analysis, rather than as a public communication tool (MacEachren and Ganter, 1990). A review of the literature found no examples of studies to determine the cognitive benefits of geovisualizations, either with expert, or non-expert users. Similar to research in other technology related fields (e.g. information visualization and virtual reality), development of new geovisualization technologies has led the evaluation of those technologies. Draping of GIS layers has often been anecdotally suggested as an effective means of communicating information about non-visual aspects of landscapes. The literature review conducted for this chapter, however, revealed no empirical studies to support this claim. Indeed, there are some findings in related literature which suggest that draped GIS data would not be the most effective means of communicating trends in landscape condition indicators over time. The first potential problem is that people do not accurately perceive differences in 2D areas, or 3D volumes (Cleveland and McGill, 1984; Spence, 1990; Teghtsoonian, 1965). Psychophysical research has found that people underestimate the magnitude of difference between objects of different area (Cleveland and McGill, 1985; Teghtsoonian, 1965). Because the quantity of a given landscape indicator when displayed by draping is represented by area, the above findings suggest that people may be unable to accurately determine trends in the indicator over time. This effect may also be exacerbated by a large number of irregularly shaped polygons draped across a large represented area, as well as by the effects of perspective and occlusion common in perspective views of landscapes. People may, therefore, have difficulty using draped GIS information to assess the amount of an indicator present on the landscape, as well as the amount of change in that indicator over time. Another potential problem with using draped information to convey trends in landscape indicators is the effect of moving draped images through time. Ainsworth and VanLabeke (2004) have categorized dynamic visualizations into three types: time persistent, time implicit, and time singular. A time persistent visualization is one that shows all previous instances of the visualization with the addition of each new instance. A line graph that proceeds from its time zero value to time n value, showing values of zero to n at all times during the progression, is time persistent (see Figure 2.7a). Time persistent visualizations are effective for portraying trends in data, because all previous data points are visible for reference (Ainsworth and VanLabeke, 2004). Draped GIS layer visualizations, on the other hand, are time-singular. Each frame (or time related image) in a time singular visualization contains only the information about that particular time period. In order for the viewer to compare information from previous time periods with the current  39 time period, they must hold the previously viewed information in working memory (Ainsworth and VanLabeke, 2004). The more complex the information, the more difficult it will be to make the mental comparison. This coincides well with the assertion that people are much better at making spatial comparisons (as they can with time persistent visualizations) than temporal comparisons (Hegarty, 2004; Tufte, 2001; Zakia, 1997). The time-singular nature of draped visualizations suggests that they may not be an effective means of communicating trends or patterns of change across the landscape over time. Time-singular visualizations are also useful in multiple representation displays (i.e. displays providing some combination of time persistent, time implicit, and time singular visualizations), because they allow the viewer to focus on the current value of the display, with alternate representations presenting the contextual information for other time periods (Ainsworth and VanLabeke, 2004). Ainsworth and VanLabeke (2004) will be further discussed in the dynamic visualization section later in this chapter. One final limitation of draping as a means of communicating information about landscape indicators is the limitation of real estate. If multiple indicators are draped simultaneously, then overlap, occlusion and visual clutter are all very real possibilities. While some of the interactions may be desirable (e.g. overlap of indicators suggesting a relationship between them), if the viewer cannot make sense of the information due to perceptual problems and cognitive load, then any benefit of draping will be obscured. The ability to interactively add and remove layers may alleviate this problem, and research into people's capacity to keep track of complicated multi-indicator patterns would clarify the problems and opportunities presented by multiple drapes. For all of the above-mentioned caveats and potential limitations of draping, it has one major strength that no other data display technique possesses: the ability to represent the spatial locations of landscape indicators in 3D. Maps can communicate the 2D spatial locations of indicators (and abstractly the 3D locations), and graphs and other information visualization techniques can portray trends and quantities, but only draping can quickly show the 3D spatial locations of indicators. If a given understory plant species is only found at high elevations on north facing slopes, draping is the most effective means of conveying that information. When used in a manner that capitalizes on this strength, and not its apparent weaknesses, draping can be an effective means of communication. Another strength of draping spatial overlays is their ability to form hybrid displays with other visualization types. So, for example, a spatial overlay of an indicator of landscape condition could be draped over a realistic landscape visualization to provide a hybrid display type that draws from the strengths of both representations. 2.3.3  Information Visualization  One of the fundamental design goals in the construction of the experimental interface was to pair some form of information visualization or information display with landscape visualizations to investigate  40 if the transmission of information about non-visual aspects of the landscape could be facilitated by this pairing. It was, therefore, a precondition of the interface that some form of information display would be used. The form of that display, however, was not clear. The intention was for the information representation to be easily understood by non-experts, suggesting that it should be either a form demonstrated to convey information quickly and effectively, or that it should be a form that was commonly used in everyday life, and who's graphical conventions were familiar to the majority of the potential users. The main strength of information visualization is its ability to provide organization and order to large data sets by capitalizing on the perceptual characteristics of the human visual system (Ware, 1999). Defining information visualization is a difficult task, because of the myriad of forms that it can take. Common to all of these forms, however, is the purpose of imposing a comprehensible structure on large data sets in order to facilitate understanding of the underlying patterns and connections in the data. Information visualization is most commonly differentiated from scientific visualization by the information's lack of an inherent spatial organization. Geovisualization (3D visualization of variables on a terrain surface) is a form of scientific visualization, because of its spatial dimensions. Developers of information visualization systems generally impose a spatial representation on the non-spatial data as a means of facilitating human perception of the data's inherent characteristics. User-centred research in the field of information visualization is sparse, and progress to date has been limited (Ware, 1999; Chen and Czerwinski, 2000). Research in information visualization has predominately been focused on the development of new technologies;' and the implementation of design ideas with specific data sets (Chen and Czerwinski, 2000). In the limited studies that have been completed on user interactions with information visualization systems, evidence has indicated that the technologies are more effective with experts in a given field who possess domain specific knowledge not shared by novices (Chen and Czerwinski, 2000; Graham et al., 2000; Sutcliffe et al., 2000). There are also aspects of cognitive load theory that suggest that information visualization displays will be more effective with experts that have domain specific knowledge and schemas. Researchers have postulated that instructions or information displays that require complex reasoning with novel information will be deficient (Kirschner, 2002; Pollock et al., 2002; Sweller et al., 1998). Because new material, especially high element-interactivity material, requires significant working memory storage, there is little room for complex reasoning and information manipulation in working memory. By their nature, information visualization techniques have high element-interactivity, because they are intended to convey patterns in densely represented data sources. If the viewer of an information visualization lacks schemas for both the information and the display technique, then the highly element-interactive data, along with the novel display technique will likely overwhelm their limited resources. Experts in a particular domain, on the other hand, have pre-existing schemas about the information in that domain, and  41  therefore can focus on understanding and using the novel information display technique, rather than having to reason about both the data and the display. Further research i n information visualization, specifically looking at the cognitive load created by the display techniques and the differences between expert and novice users may elucidate this issue. While there may be novel information visualization techniques that could effectively communicate the non-visible information about landscape change i n the interface, there is little or no supporting evidence to help select one of these techniques. One of the main purposes of the construction of the interface was to investigate the combination of landscape visualization and information display. In the absence of evidence for the efficacy of an information visualization technique for novice users, the inclusion of one of these techniques w o u l d likely have made it more difficult to determine the effects of the interaction between the information representation(s) and the landscape visualization. It w o u l d have been difficult to determine what portion of the effect was the result of the novel information presentation technique, and what part of the effect was the result of the interaction between the landscape visualization and the information presentation.  2.3.4  Graphs  The most common means of visual data presentation are graphs. W i l l i a m Playfair is generally attributed w i t h the creation of the graphing convention i n his Commercial and Political Atlas of 1786 (Feeney, 1994; Wainer, 1992). Since that time graphs have become ubiquitous i n the sciences and social sciences, and have been popularized i n western society by publications such as Time Magazine and U S A Today (Feeney, 1994; Shah and Hoeffner, 2002). This ubiquity means that graphs reach a large segment of the population i n western societies, and that their graphical conventions are, i n general, w e l l understood. M a n y people have pre-existing cognitive schemas about graphs, formed from previous experience w i t h them, w h i c h can aid i n graph comprehension (Kosslyn, 1989; Shah and Hoeffner, 2002). Graphs also have the benefit of having been extensively studied i n the fields of cognitive psychology and educational psychology (e.g., Carswell et al., 1993; Cleveland, 1994; Cleveland and M c G i l l , 1985; Guthrie et al., 1993; Kosslyn, 1989, 1994; Schutz, 1961a,b; Shah and Hoeffner, 2002; Shah et al., 1999; Simkin and Hastie, 1987; Tan and Benbasat, 1993; Zachs and Tversky, 1999). This research has led to the development of rules and guidelines i n the use of graphs and charts to present information (Cleveland, 1994; Kosslyn, 1989,1994; Shah and Hoeffner, 2002). There are numerous different types of graphs that have been developed since the time of Playfair. M a n y of these graphs have very specialized purposes, while others have more general applications. Probably the four most common graph types are bar graphs, line graphs, pie graphs and scatterplots. To greater and lesser extents each of these graph types has been explicitly studied to determine their  42 efficacy in the presentation of various data types, and for various purposes. Research suggests that bar graphs are best for communicating discrete quantities, generally in categorical data (Kosslyn, 1989; Shah and Hoeffner, 2002; Tan and Benbasat, 1993; Zachs and Tversky, 1999). Tan and Benbasat (1993) state that this is caused by the high axis anchoring of bar graphs, which makes it perceptually easier for users to read the appropriate numerical data for a bar graph off the corresponding axis. Zachs and Tversky (1999) indicate that the Gestalt principles of figural perception make bars effective for presenting categorical data, because they enclose and separate the data. Few (2004) states that bar graphs make use of one of the two preattentive capabilities of the visual system (capabilities that do not require any conscious attention) that can convey quantitative information. These two preattentive capabilities are the ability to determine 2D position, and the ability to determine line length. Bar graphs make use of line length to communicate the quantity of a category, and therefore perception of bar graph quantities can be done quickly, and with minimal effort. Pie graphs are most commonly used to communicate information about relative proportions (Simkin and Hastie, 1987; Kosslyn, 1994). Pie graphs are preferred to divided bar graphs for the presentation of proportion data because divided bar graphs often require the user to add segments from different parts of a bar in order to determine additive proportions (Simkin and Hastie, 1987). A difficulty of pie graphs for data presentations is that they use 2D area to communicate quantities. As mentioned earlier in the draping section, humans are not able to reliably differentiate differences in area (Few, 2004). Cleveland and McGill (1985) and Cleveland (1994) have suggested the use of dot charts as a replacement for pie graphs because they make use of the visual system's preattentive ability to judge line length, and therefore are easier to read. Of the four most popular graph formats, scatterplots have received the least amount of attention in perception research. Kosslyn (1994) indicates that scatterplots should be used when there are a large number of data points, and the intention is to convey a general impression of the data. Groups of points in a scatterplot form clouds of data which indicate a very general trend (e.g. increase in weight with height for a large number of individuals). Cleveland (1994) agrees, stating that scatterplots communicate the general relationship between two variables, but do not (without the assistance of a trend line or loess curve) accurately portray the true nature of a relationship. Scatterplots, therefore, are best used when the intention is to portray the general relationship between two variables for a large number of data points. When the intention is to communicate information about trends or patterns in data, then line graphs are the widely accepted preference (e.g., Carswell et al., 1993; Kosslyn, 1989, 1994; Schutz, 1961a,b; Shah and Hoeffner, 2002; Shah et al., 1999; Tan and Benbasat, 1993; Zachs and Tversky, 1999). According to Tan and Benbasat (1993) line graphs have high "entity anchoring" meaning that people perceive the line as one perceptual unit, rather than as a series of data points. This allows users of a line graph to  43 perceive the entire pattern of the line, rather than a number of disconnected values. This means that it can be held in working memory more easily, allowing for other information to be held simultaneously, and for cognitive operations to be performed on that information. The strong perceptual tendency to perceive line graphs as portraying continuous information about data trends may be a function of the human perceptual system that is reinforced by learning about graphs (Zachs and Tversky, 1999). Zachs and Tversky (1999) found that subjects that had no preconceived notion about the use of line graphs to portray continuous data and bar graphs to portray categorical data still followed these conventions in their assessment of data displays. In fact, when subjects were presented with line graphs that portrayed obvious categorical data, their interpretations represented it as continuous. For example, when height (y-axis) was plotted against gender (x-axis) in the form of a line graph, the subjects' interpretations indicated male vs. female gender as a continuous property (i.e. "the more male a person is, the taller they are") (Zachs and Tversky, 1999). This finding suggests that the tendency for people to look for continuous data trends in line graphs is very strong, even when they are unaware of the conventions governing graph types. The purpose of the information presentation of the non-visible information about landscape condition was to communicate trends and patterns in the chosen indicators, not to facilitate the reading of individual data points. One of the theoretical underpinnings of the design of the interface was that it is more important for non-experts and members of the public to have an overall conception of the patterns of change on the landscape over time, than for them to receive detailed facts about a particular indicator at a specific point in time. The information display, therefore, needed to be designed in such a way that the trends and patterns became the salient features, and users were not guided towards getting lost in the minutiae of individual values (though data transparency should be provided to allow for the assessment of assumptions). This design goal of focusing on trends and patterns of change, along with the lack of empirical evidence for the efficacy of particular information visualization techniques, and the wealth of research on the various graph forms, led to the selection of line graphs as the most likely choice for representing the non-visible indicators of landscape change. 2.3.5  Maps  Lloyd and Bunch (2003) state that "the fundamental role of a map is to represent real-world information as a set of abstract features with the goal of simplifying spatial information for map communication and reading comprehension". The use of maps to represent spatial aspects of the world dates back millennia, and has been tied to most cultures of the world (Thrower, 1999; Uttal, 2000). The basic ability to use maps has also been shown in children as young as four years old, even in the congenitally blind (Landau, 1986). This early, apparently unlearned, ability to use maps suggests that some map reading ability may be a product of an inherent spatial knowledge system (Landau, 1986).  44  G i v e n the widespread use of maps, and the apparent priming of the cognitive system for reading and creating them, what is k n o w n about how maps are understood and represented i n the brain's cognitive systems?  Map Learning and Cognition To date, the cognitive factors i n map comprehension and use are not w e l l understood (Lloyd and Bunch, 2003; Schofield and Kirby, 1994). This may, i n part, be due to the wide variety of map types and the complexity of many map forms. The lack of a comprehensive understanding of map cognition does not, however, indicate a lack of research i n the area, nor does it suggest a lack of a theoretical framework for map use and comprehension. In their review of research on map learning, K u l h a v y and Stock (1996) provide perhaps the most complete, and oft cited, model of map use and comprehension. They suggest that map learning is based on two factors: control processes, and characteristics of the human memory system. Control processes match the information presented on the viewed map to information already present i n the subject's memory (i.e. i n schemas), and also interpret the task demands associated w i t h learning the map (Kulhavy and Stock, 1996). Features of the memory system, previously discussed i n the Cognitive Psychology Concepts section of this chapter, also play a significant role i n map comprehension. The limitations of working memory, characteristics of the visuospatial sketchpad, and processing advantages provided by pre-existing schemas all play a role i n the acquisition of map information and resulting internal representation. Herrmann and Pickle (1996) have proposed a similar system, but have used three factors to describe the map learning process. They suggest that map learning is comprised of attributes of the map to be learned, the task demands placed on the map reader, and the cognitive attributes of the map reader. The distinction between the Herrmann and Pickle (1996) model and the K u l h a v y and Stock (1996) model is the former's assertion that attributes of the map play a role i n map learning. While this may seem an obvious factor i n map learning, there has been some argument on it i n the literature (see e.g. Gilhooly et al., 1988; K u l h a v y and Stock, 1996; Thorndike and Stasz, 1980). Gilhooly et al. (1988) found little difference i n map reading ability between lay-subjects and experts, except i n complex tasks such as recall of map contour line elements. They speculated that expert users have more complex map schemas, including schemas for contour line representations of elements such as valleys and ridges. Thorndike and Stasz (1980) found that the map learning strategies employed by subjects were tied to their map reading ability, and that these strategies were related to expertise in map reading. Subjects w i t h more expertise i n map reading employed more efficient map learning strategies, and therefore learned more from the presented maps. Lewis and Sheppard (2004), i n a study w i t h a First Nations band i n British Columbia, found that abstract or complex forest harvesting  45 maps were confusing to many subjects, causing errors in judgments about the locations of important landmarks. Kulhavy and Stock (1996), and Winn and Sutherland (1989) postulate that general map learning skills such as map feature identification and spatial location are not affected by expertise, but that more specialized skills (such as contour line interpretation) will be affected by previous knowledge and expertise. This suggests that general reference maps should be usable by the vast majority of users, while more complex representations may require expertise not prevalent in the general population. Researchers have generally separated maps into two forms of information: feature information and structural information (Verdi and Johnson, 1997; Rittschof et al., 1994; Kulhavy and Stock, 1996). Feature information indicates what something is, generally through the use of symbols, colours, etc. Structural information indicates where something is in relation to other map elements, and provides spatial, metric and border information. Structural information is thought to be more important to the efficient learning of maps than feature information, because the structural elements provide the map's "intactness" that allows it to be efficiently stored as a single perceptual unit in working memory (Stock et al., 1995; Verdi and Johnson, 1997; Rittschof et al, 1994; Kulhavy and Stock, 1996). Larkin and Simon (1987), in their seminal article on the cognitive advantages of images and diagrams for learning, state that the ability of the working memory system to hold an image as a single perceptual unit is what leads to their computational efficiency, and resulting advantage over text-based learning. Research supports the notion that map structural characteristics improve the ability of subjects to remember map information. Kulhavy and Stock (1996) found that the inclusion of structural information improved fact recall from an associated text passage by as much as 23% over presentation of map features without structural information. People have also been shown to remember more information from map areas adjacent to map boundaries, and to internal map boundaries such as streams and roads (Verdi and Kulhavy, 2002; Rossano and Morrison, 1996). Map boundaries are structural elements, key to the intact mental representation of maps in working memory. The improved learning of map features adjacent to map boundaries underscores the importance of structural information to map learning. Rittschof et al. (1994) state that structural information is encoded first when reading a map, and that it is implicitly rehearsed in memory as the map reader learns the feature information from the map. The primacy of structural information learning, as well as the benefits it confers on subsequent map learning suggests that it is an integral part of the creation of maps as learning aids. Provision of structural elements such as boundaries, streams, roads, etc. is very important to the effective communication of map information.  46 The Effects of Maps on the Learning of Associated Material  Much of the research into the use of maps in learning has focused on their ability to facilitate the learning of associated material, specifically text-based material. Raymond Kulhavy has developed a model to explain the ability of maps to improve the recall of text-based information from an associated passage. He has dubbed the model "conjoint retention" reflecting the idea that both the map and the associated text are held in working memory conjointly (Verdi and Johnson, 1997; Winn and Sutherland, 1989; Kulhavy and Stock, 1996; Kulhavy et al., 1993). The theory of conjoint retention is based upon the earlier theory of dual coding by Paivio (1986), which will be discussed in greater detail in the section on text to follow. The basic premise of Kulhavy's model is that the ability of maps to aid the learning of associated text material stems from their efficient storage in working memory and the referential ties that can be made between the resulting mental image and the incoming propostional text information. Because the map is held by the visuospatial sketchpad, and the text information enters through the phonological loop, working memory capacity is effectively expanded allowing for processing and references between the text and map information. This model has been tested and supported by a number of research studies (Verdi and Kulhavy, 2002; Kulhavy and Stock, 1996), and has been shown to occur with both reference maps (e.g. Johnson et al., 1995) and thematic maps (e.g. Rittschof et al., 1994). Maps, then, aid the learning of related text information by allowing referential links to be formed between the map's visual information and the accompanying text information. These links, formed in working memory, can then be passed into long term memory either as a new schema, or additions to an existing one. When recall is required, the map image in memory can serve as a mnemonic device through the previously formed referential links to help the subject recall the relevant text information. As a result, the presentation of a related map with text information can improve subjects' learning of that text information. In order to confer this advantage the map must be presented before or alongside the related text, so that it is present in working memory at the time of text encoding. A review of the literature did not reveal any research on the use of maps in learning studies in conjunction with other visual information displays (e.g. graphs or landscape visualizations). From the conjoint retention model, it is the presentation of information through both the visual and auditive memory systems that conveys the advantage. Presentation of two visual media sources would not necessarily show the same referential linking ability, due to their shared requirements for memory and processing from the visuospatial sketchpad. However, if both images were held efficiently in working memory, freeing up memory resources for information processing, then it is possible that similar learning advantages could occur.  47 The Use of Colour on Maps  One area of map design which has received significant attention is the use of colour, and more specifically colour schemes, on thematic maps. Because colour is commonly used on thematic maps to abstractly represent categories, and/or amounts of given thematic elements, their appropriate application is particularly important. It is also important to consider colour deficient vision when constructing map colour schemes. Approximately eight percent of males and one percent of females have colour deficient vision (Brewer, 1999). Almost all of these colour deficiencies are red-green in nature, though there are many different types (Olson and Brewer, 1997). Colour schemes, therefore, should be constructed in such a way as to communicate effectively to both colour proficient and colour deficient viewers. Where expressing quantities, colour ramps (variations in colour lightness and saturation without change in hue) do not pose a problem for colour deficient users, regardless of the hue selected (Olson and Brewer, 1997). In colour ramps, darker (lower lightness and/or higher saturation) colours are generally perceived to communicate higher values, and lighter colours (higher lightness and/or lower saturation) are generally perceived to communicate lower values (Brewer, 1999; Cuff, 1974; Olson and Brewer, 1997). Changes in hue, on the other hand, are usually used on maps to represent different categories of information. When only the lightness of a colour is varied, the human eye is limited to discerning a maximum of seven to eight shades, with shades of grey from white to black presenting the largest dynamic range, and therefore the best discernment (Muller, 1979). Lightness and saturation of colour tend to vary together, though either property alone (with the other held constant) can communicate changes in a quantitative value (Cuff, 1972). Of the two, however, lightness seems to be superior for communicating quantitative values, as people are not generally accustomed to comparing saturation levels (Brewer, 1999). Brewer (1999) contends that, in a quantitative scheme, saturation should be used to enhance the effects of lightness (thus increasing the discernable categories). Phillips (1982) investigated the use of colour schemes in the communication of elevation information. He compared spectral schemes that followed the visual spectrum from violet to red to communicate increasing elevation, to tonal schemes that held hue constant and varied saturation and lightness (light to dark communicating low elevation to high elevation). Phillips (1982) found that spectral schemes were better for the accurate reading of absolute values, while tonal schemes were better for reading relative values (i.e. judging relative elevations). In the study, some viewers had difficulty with the spectral schemes, because they found the portrayal of quantities by colour to be arbitrary (i.e. they did not understand the hue progression). Phillips (1982) also showed that the use of colour ramps were superior to contour lines for the rapid and accurate communication of elevation information. A study by Patton and Crawford (1978) demonstrated the potential for colour schemes to erroneously communicate unintended information. They investigated the common hypsometric colour  48 scheme of green to red to communicate increasing elevation. The Natural Resources Canada Atlas of Canada Glossary defines hypsometric tints as "A method of depicting relief (or depth of water) on maps by using a gradation of different colours or tints, usually between contour lines. Each band of colour represents a different range of elevation or depth in reference to a specified surface, or datum" (Natural Resources Canada, 2005). Patton and Crawford (1978) found that elementary, high school and college students could all use the colour scheme to determine elevation, but that the scheme also communicated unintended information. Patton and Crawford (1978) found that a significant portion of all three age groups equated the green of lower elevations to also mean "heavy vegetation". The subjects also interpreted the progression from green to brown to indicate a decrease in rainfall, and a weaker (insignificant) relationship between hue and temperature interpretations, with red (higher elevation) being perceived as warmer than green (lower elevation). These unintended communications of the green to red hypsometric colour scheme demonstrate the importance of the effective use of colour for cartography, and that reliable conventions, such as the use of tonal maps to communicate quantitative values, should be used when possible. When determining colour schemes for maps Brewer et al. (1997) indicate that the first priority should be clarity, with aesthetic preference coming in only when clarity has been achieved.  2.3.6  Text Information  Human acquisition and use of text information has received significant attention in the psychology literature. Treatment of the literature on text in this section will be limited to the combination of text information with images, such as diagrams, maps and illustrations. Research has repeatedly shown that pairing text information with related images such as maps and annotated diagrams improves learning over presentation of the text or image information alone (e.g. Mayer, 1998; Mayer and Anderson, 1992; Mayer and Moreno, 2002b; Mayer and Sims, 1994; Wiseman et al., 1985). Research has also shown that memory capacity can be expanded by the simultaneous use of the auditive and visual channels (e.g. Frick, 1984; Penney, 1989). For example, Frick (1984) found that recall memory was improved if some items were presented visually, and others were presented auditively, rather than all items being presented in one modality. Text information presented visually does pass into the auditive channel through rehearsal, but this additional step appears to reduce (though not negate) the benefits of presenting text and images. The Mechanism for the Dual Presentation Mode Benefit to Learning  The mechanism which accounts for the improvement in memory when both auditive and visual channels are used has been called "dual coding" by Paivio (1986). As mentioned in the Working Memory section of this chapter, working memory is separated into two distinct stores: the visuospatial sketch-  49  pad, and the phonological loop (Baddeley, 1992). The overall capacity of working memory is larger when both of these stores are utilized simultaneously, because they are separate, and posess their own capacity limits (Baddeley, 1992; Paivio, 1986). Therefore, when images, diagrams, illustrations, maps and other forms of visual information are presented visually, and text is presented auditively (either through narration or mental repetition of printed text), the capacity of working memory is increased, allowing for the integration of more information through processing in working memory (see Figure 2.6). Paivio (1986) called this integration the formation of "referential links" between information from the two working memory stores, and between working memory information and pre-existing knowledge in the form of schemas in long-term memory. Information flow Phonological loop  Information flow  Information  — i working  Intrinsic and Extraneous H Cognitive Load  Information  1 memory 1 capacity Cognitive Load  [  •  •  •  •  •  working capacity  Visuospatial sketchpad Visuospatial sketchpad  Single Mode Presentation  Dual Mode Presentation  Figure 2.6: A diagram explaining the increase in working memory capacity activated by the use of both the visual and auditive channels. One of the fundamental findings in the study of text and images has been the consistent experimental results indicating that text, when paired with images, is best presented auditively. A number of studies have found this "modality effect", and the results are consistently strong (e.g. Mousavi et al., 1995; Tindall-Ford et al., 1997; Kalyuga et al., 1999; Mayer et al., 2001; Leahy et al., 2003; Mayer and Chandler, 2001; Moreno and Mayer, 1999). Moreno and Mayer (1999) found that even the most physically integrated printed text and images did not lead to learning outcomes on par with visual images and auditive text. The mechanism for this effect is the largely independent visual and auditive memory stores discussed earlier (Tindall-Ford et al., 1997). Printed text must be read using the visual channel before mental rehearsal can translate it to the auditive channel. This increases the cognitive load in the visual channel, as well as not fully accessing the expanded memory capacity made available by dual mode presentation (Tindall-Ford et al., 1997). The result is decreased performance on learning related tasks for those subjects that receive all of their information through the visual channel. While this effect has been strong in the research conducted, researchers are quick to point out caveats and limitations. Kalyuga et al. (1999) state that purely visual information (text and images) has the advantage of permanence, such that the subject can refer to it repeatedly. They also point out that visual search (and the associated cognitive load) can be reduced by other mechanisms, such as colour coding  50 of text and associated image elements. In an experiment, Kalyuga et al. (1999) found colour coding to have similar benefits in learning outcomes to the use of dual mode presentation. Mayer and his colleagues (Mayer and Chandler, 2001; Mayer et al., 2001) point out that the modality effect is greatest when the material is presented rapidly, and that the effect would likely be reduced by interactivity and user-pacing of the presentation. Finally, Leahy et al. (2003) found that when an image and its labels or annotations provided all of the necessary information for learning, accompanying narration actually hindered learning, presumably by increasing cognitive load. While research has shown the consistently strong benefit conferred by presenting text auditively, researchers are quick to state that it is not a panacea, applicable in all learning situations. Text and Images From a Cognitive Load Perspective There has been significant crossover between research into cognitive load and research into the joint presentation of text and images. It is important to note that the pairing of text and images does not itself decrease the intrinsic (load dependent on information complexity) or extraneous (load resulting from presentation techniques) cognitive load of the information being presented, but rather expands the capacity of working memory to hold and process information (see Figure 2.6). There are, however, a number of characteristics of information presented as text and images that do have an effect on the extraneous cognitive load of presented information. Sweller and his colleagues have investigated the split attention effect, which occurs when people are required to attend to multiple information sources that must be mentally integrated before meaning can be derived from them (see e.g. Chandler and Sweller, 1991,1996; Tindall-Ford et al., 1997). This effect occurs when text and associated images are separated from one another either spatially or temporally. For example, if the text describing an illustration occurs on a separate page in a textbook. The physical integration of text and images, either by placing related text next to or on a diagram or image, or by having the associated text narrated at the appropriate point during the presentation of images, reduces the need to mentally integrate the information, and therefore reduces the associated cognitive load (Chandler and Sweller, 1991,1996; Tindall-Ford et al., 1997; Mayer et al., 1999). The benefit of dual mode presentation is also dependent upon the intrinsic cognitive load of the information. Tindall-Ford et al. (1997) investigated the effects of element interactivity (the amount that learning one piece of information is dependent on the learning of other associated information) on the benefit of dual mode presentation. They found that as the interactivity of information elements increased, the benefit of dual mode presentation also increased. For low element interactive information the presentation of both text and images through the visual channel, as opposed to text through narration and visual images, did not have a negative effect on learning. Low interactivity material, for example, would be a list of labeled symbols, where learning one item in the list was not dependent on  51 any other. High interactivity material could be represented by geometric diagram and their associated formulas, where comprehension of a diagram is dependent on learning the appropriate formulas, and vice versa. From a cognitive load perspective, then, the appropriate presentation of information utilizing the benefits of dual coding can both reduce the extraneous cognitive load caused by split attention, and ameliorate the effects of intrinsic cognitive load caused by information with high element interactivity. Mayer's Cognitive M o d e l of M u l t i m e d i a Learning  Perhaps the largest body of research on the combination of text and images has been conducted by Richard Mayer and his colleagues at the University of California at Santa Barbara (see Mayer, 1998; Mayer and Moreno, 2002b). From this research Mayer has developed seven principles of multimedia instruction relating to the presentation of text and images. These principles are: 1.  M u l t i m e d i a Principle - The multimedia principle states that subjects learn more from the combination of text and images than from text alone. This principle is based on the dual coding work of Paivio (1986), and the working memory research of Baddeley (1992), and has been supported by a number of studies (e.g. Mayer and Anderson, 1992; Moreno and Mayer, 1999; Mayer and Sims, 1994; Chandler and Sweller, 1991,1996).  2.  Spatial Contiguity Principle - This principle states that subjects learn more from printed text and associated images that are placed in close physical proximity to one another, than from those that are spatially separated. This principle, when coupled with principle three below, equates to the split attention effect described by Chandler and Sweller (1991). This principle also relates to the Gestalt principle of proximity, which states that humans are more likely to make comparisons and associations between items that are spatially or temporally proximal (Zakia, 1997). Studies have shown that spatial proximity of text and related images improves learning (see e.g. Moreno and Mayer, 1999; Chandler and Sweller, 1991,1996).  3.  Temporal Contiguity Principle  4.  Coherence Principle - This principle states that subjects learn better from material when extraneous words, sounds and images are excluded from a presentation. This principle also relates to the seductive detail hypothesis, which states that material included to increase interest in the subject matter, but that does not have immediate relevance to the material to be learned, is detrimental to the learning process (e.g. Harp and Mayer, 1998). Evidence for the coherence principle comes from experiments by Mayer et al. (1996), Mayer et al. (2001), and Moreno and Mayer (2000).  5.  M o d a l i t y Principle - The modality principle refers to the finding, discussed earlier in this section, that text-based information that is presented with images is learned better when narrated than when displayed as printed text. This principle has been supported by studies comparing learning outcomes from the visual-visual versus visual-auditive presentation of images and text (e.g. Moreno and Mayer, 1999; Chandler and Sweller, 1996; Tindall-Ford et al., 1997).  6.  Redundancy Principle - The redundancy principle is counterintuitive when viewed in relation to conventional wisdom, especially in light of the recent proliferation of Powerpoint (Microsoft Corp., 2001b) and other slide based presentation tools. This principle states that subjects learn  - Temporal contiguity in this application refers to the grouping of related images and auditive narration, such that the narration occurs at the same time as the corresponding image. Mayer and his colleagues (e.g. Mayer et al., 1999; Mayer and Anderson, 1992; Mayer and Sims, 1994) have shown that subjects learn better when images and related narration are presented simultaneously, rather than successively.  52 better from images and related narration than they do from images, related narration and printed text. According to this principle, the printed text is redundant, rather than reinforcing, and it competes for visual channel resources with the images. The printed text and the narrated text are also perceived at different rates, making the formation of referential links in working memory more difficult. The negative effects of the redundant presentation of text in both visual and auditive forms has been shown by Kalyuga et al. (1999) and Mayer et al. (2001). 7. Personalization Principle - This principle states that subjects will learn more from images and associated text if the text is presented in conversational, as opposed to formal language. This principle is based on the idea that subjects will work harder to learn material if they feel they are personally involved. Moreno and Mayer (2000) found a strong positive relationship between the use of conversational language and improved learning outcomes in a series of experiments explaining lightning and botany. In summary, then, learning is improved by the presentation of images (diagrams, illustrations, maps, etc.) in concert with related text. The learning effect is improved by the presentation of text as narration, to more fully engage the dual coding effect. Narrations should be temporally concurrent with the relevant images, whereas printed text should be spatially proximal or integrated with the relevant images. Learning materials should not include extraneous information, even if the intention is to create interest. Presentation of text should be limited to one channel (visual or auditive) to eliminate the interference caused by redundancy, and attempts should be made to personalize learning material by reducing formality, in order to engage the subject in the learning process. Application of these principles and research findings to the creation of educational multimedia environments should improve the comprehension of the presented material by the subjects that view them. 2.3.7  Dynamic Representations  Dynamic representations are representations that explicitly represent movement, and most commonly are used to represent changes through time and/or motion. Animations are thought of as the classic example of dynamic representations (Ainsworth and VanLabeke, 2004), but this category also includes interactive dynamic representations and real-time representations. "If a picture is worth a thousand words, are a thousand pictures worth a million words?"(Sheppard, 2000). This quote both summarizes and questions the conventional wisdom about dynamic representations; that being, if a static representation provides x amount of information, then does the sequential presentation of a series of static images n provide nx amount of information? While few would argue that there is a linear relationship between the number of frames (sequential static images) and the amount of information presented, there is a common belief that, by making movement and /or time explicit, dynamic representations are providing more information to the user (Hegarty, 2004; Hegarty et al., 2003; Jones and Scaife, 2000; Lowe, 2004; Rieber, 1991; Scaife and Rogers, 1996). When the portrayal of temporal or movement information is important to the communication of a given set of information, then the ability to explicitly represent those phenomena would seem beneficial.  53 The role of dynamic representations (specifically animation) in communicating dynamic information has been evaluated by a number of researchers with, at best, mixed results. The next subsection will discuss some of the relevant evaluative studies and their findings, and the final subsection in the dynamic representation section will discuss problems with dynamic representations, as well as how some researchers are attempting to reframe the application of dynamic representations and the questions asked in research. Evaluative Studies  A number of studies in the late 1980's and early 1990's compared static representations with dynamic representations on a variety of learning tasks. Some early results looked promising, with dynamic representations appearing to provide benefits to learning about dynamic systems. For example, Rieber (1991) found that animations of Newton's laws improved both the intentional learning, and the incidental learning (i.e. learning about aspects of the laws not explicitly covered by the study material) of the animation study subjects over the static representation group. Large et al. (1996) also found that animations facilitated leaerning. They found that animated diagrams of the human heart improved the learning outcomes of the subjects that viewed them when compared with a static diagram group. These studies, along with others, formed an early basis for the hypothesis that dynamic representations would help users to better understand information about dynamic systems. However, a review by Tversky et al. (2002) cast a critical eye on some of this early research. In a review of more than 20 studies comparing animation to static representations, the majority of studies found no benefits of animation over static representations. Perhaps more telling, in the studies that did show a benefit to animated representation, Tversky et al. (2002) determined that the animations and the static representations were not informationally equivalent (i.e. the animations were determined to contain more information than the corresponding static representations). The added information in the animations was not simply a reflection of their motion and the number of frames presented, but rather involved the addition of entirely new information elements. Therefore, when the information being presented was controlled for, there was no evidence from any of the available studies to suggest that animations improved learning. This review had a strong effect on the studies conducted after its publication, with researchers paying careful attention to the information equivalence of the research stimuli. The result has been a much less positive assessment of the efficacy of dynamic representations. In multiple studies on the use of animated weather maps to teach subjects about weather systems, Lowe (2003,2004) found no benefit from the use of animation. He found that subjects in the animation study condition were inclined to focus on perceptually salient material from the animations rather than thematically salient material. In other words, the subjects would focus on eye-catching changes in the display, rather than changes that depicted important weather phenomena. Lowe (2003) hypoth-  54 esized that in order for animations to be effective, thematically relevant aspects should be perceptually emphasized to attract the viewer's attention. Lewalter (2003) investigated the relative effects of pairing static diagrams and explanatory animations of light bending around planetary bodies with an expository text oh the subject. Lewalter was careful to maintain information equivalence between the static and dynamic representations by including directional arrows, movement cues, etc. on the static representation. In assessments of both recall and transfer Lewalter (2003) found no significant differences in learning outcomes between the static representation and the dynamic representation. Lewalter (2003) hypothesized that the transience of dynamic representations increases cognitive load, because the user must hold previously viewed information in memory in order to compare it to current information. This cognitive load may have prevented users in the animation group from learning more from the animated information. Jones and Scaife (2000) conducted a study using material similar to the Large et al. (1996) study (heart blood movement), but taking care to ensure information equivalence between the static and dynamic representations. They found no significant differences between the learning outcomes for the groups using the static representation versus the group using the dynamic representation. They did find that the animation group had artificially high confidence levels in their knowledge about the material, a result supported by a similar finding by Lowe (2004). This confidence suggests that users of animation may feel they are getting more, or more complete, information from animations, even if they are not successfully retaining that information. Problems with Dynamic Representations, and Possible Solutions  Perhaps the most commonly stated problem with dynamic representations is the increased extraneous cognitive load that they are thought to create (Bodemer et al., 2004; Hegarty, 2004; Jones and Scaife, 2000; Lewalter, 2003; Lowe, 2003, 2004; Ploetzner and Lowe, 2004). This is a conceivable byproduct of the "are a thousand pictures worth a million words?" phenomenon, as presenting more information does not necessarily lead to more comprehension. Lowe (2003,2004) calls the effects of the extraneous cognitive load caused by dynamic representations overwhelming. This overwhelming may cause viewers of dynamic representations to constrain their perception of the representations to particular elements, meaning that other elements are neglected (Lowe, 2004; Bodemer et al., 2004). As Lowe (2003, 2004) found, constraint on perception may lead to an overemphasis on perceptually, rather than thematically, relevant aspects of the representations. Both Lowe (2003) and Bodemer et al. (2004) advocate the use of perceptual cues in dynamic representations to draw the viewer's attention to particularly important information. Another likely contributor to the cognitive load caused by dynamic representations is their transience (Jones and Scaife, 2000; Lewalter, 2003). In order to compare information from different parts of  55 a dynamic representation, the viewer must hold the previously viewed information in working memory — potentially for an extended period of time. This recalls the previously discussed superiority of spatial versus temporal comparisons. Static representations allow the user to offload cognition onto the display, rather than holding aspects of the representation in memory. In this manner, multiple spatial comparisons can be made using static representations, drawing upon their permanence (Hegarty, 2004). One aspect of dynamic representations that may lead to extraneous cognitive load, but that can be controlled for, is the speed at which the dynamic representations run (Hegarty, 2004; Jones and Scaife, 2000; Tversky et al., 2002). In their review of studies on animation, Tversky et al. (2002) found that many of the animations were running too quickly for users to be able to accurately apprehend their content. In the field of landscape visualization, it is not uncommon for animated fly-overs to run at many times the speed of sound in order to maintain interest and engagement. It is likely very difficult for viewers of these animations to perceive subtleties of the terrain or land cover when traveling at these speeds. Unlike the amount of information and the transience of the display, however, the speed of a dynamic representation can easily be controlled for. Balancing engagement with rate of movement and/or display change will be the difficulty in determining the optimal speed of presentation for a given dynamic representation. A literature review revealed no research on the effects of dynamic representation speed on perception and learning. One interesting line of research conducted by Hegarty and her colleagues (Hegarty et al., 2003; Hegarty and Kozhevnikov, 1999) compares mental animation of static representations to external animations. The hypothesis of this research is that people mentally animate diagrams and images to infer how things work. These mental animations happen in a piecemeal fashion, with the user animating one element of the system, followed by another, etc. After animating the parts of the system, the user then integrates those parts into a cohesive picture of how the system works (Hegarty et al., 2003). External animations, on the other hand, tend to show the whole system in simultaneous motion. This simultaneous motion may be contrary to how cognitive systems normally operate, and therefore may impede the transfer of information. In studies of mental versus external animation, the presentation of explicit movement in external animations has not been found to improve learning outcomes over mental animation (Hegarty and Kozhevnikov, 1999; Hegarty et al., 2003). In a similar finding, Lewalter (2003) found that the use of a small series of static diagrams depicting the different states of a system were just as good as animations of the system in promoting learning. Presumably the subjects in the multiple static representation group were mentally animating between the static representations of different states to derive the workings of the system. Mayer and Moreno (2002a) promote the movement of research on dynamic representations away from the "does it facilitate?" approach and towards a "when does it facilitate?" approach. Research on  56 dynamic representations (especially animation) has been marked by a wide variety of representation types, subject areas, research subject pools, etc. Scaife and Rogers (1996) attribute this scattered approach with the failure to develop a cohesive cognitive model upon which to base research. They hold up cognitive load theory (Chandler and Sweller, 1991) and dual coding theory (Paivio, 1986) as examples of cognitive models that have given direction to research in educational and cognitive psychology (Scaife and Rogers, 1996). By moving to a more ordered approach to dynamic representation research, they believe that it will be possible to get a better picture of where and when dynamic representations will be useful. In a similar vein, Hegarty (2004) advocates the use of task analysis to determine the best fit between the information to be presented and the media of presentation. The over-riding theme of re-focusing dynamic representation research away from a "yes" or "no" answer, and towards a better understanding of "when" and "where" to apply it, may represent a change in the direction of dynamic representation research that will lead to a clearer picture of the benefits and limitations of dynamic representations. Finally, one interesting theoretical development that may have research implications in the application of dynamic representations is the recent work of Ainsworth and VanLabeke (2004), discussed briefly in the landscape visualization section of this chapter. They have broken dynamic representations down into three main types. The first type is time-persistent representations, which are representations that show the present value of a variable, as well as all previous values. A time series graph (see Figure 2.7 a) is an example of a time-persistent representation, because at any point in a dynamic representation it can show the given point and all previous points in the series. The presence of previous values in a time-persistent representation provides computational off-loading to the viewer, because they are not required to hold the information in memory (Ainsworth and VanLabeke, 2004). Timeimplicit representations show an array of values, but they do not indicate when those values occurred. In order to determine the onset or offset time of a value it is necessary for time-implicit representations to be run dynamically. A phase plot (see Figure 2.7 b) is an example of a time-implicit representation — with neither the x nor the y axis representing time explicitly. Time-implicit representations are appropriate for showing the relationship between two variables at a given point in time (Ainsworth and VanLabeke, 2004). Time-singular representations display one or more variables at a particular point in time. They are synonymous with the traditional conception of animation, with image elements appearing and disappearing in a transient fashion. A bar graph that changes value at each time period (see Figure 2.7 c for time one and 2.7 d for time two) is an example of a time-singular representation. Time-singular representations are often used to communicate complex information, where persistence would lead to confusion or clutter the display (Ainsworth and VanLabeke, 2004). The categorization of dynamic representations by Ainsworth and VanLabeke (2004) has interesting implications for the application of dynamic representations, and for the tailoring of content to repre-  57 C)  30r  Type A  Type B  Figure 2.7: Diagrammatic representations of Ainsworth and VanLabeke's (2004) dynamic representation types. sentation type. It also raises the possibility of combining the different forms to draw upon the strengths of each, and to provide multiple perspectives on the information. A schematic time-persistent representation could be used to represent the present and past values, while a time-singular representation is used to represent the information in a more complex or iconic format. The issue of combining multiple representations will be discussed further in its own section later in this chapter. A review of the dynamic representation literature did not reveal any examples of comparisons between real-time representations, animations and/or static representations. This is perhaps not surprising given the youth of real-time representations, and their (until recently) prohibitive technological requirements. It is likely that real-time representations would suffer from some of the same problems as animations, because of their similarities in the amount of information presented, the transience of information and the speed of presentation. One feature that real-time representations, as well as hybrid representations that possess limited user control, have that may make a difference in how they are perceived and used is interactivity. The use of interactivity has been hypothesized as a potential ameliorating factor to the problems presented by dynamic representations (e.g. Hegarty, 2004; Lowe, 2004; Tversky et al., 2002). 2.3.8  Interactivity-  Interactivity is an ill-defined concept, making it difficult to bound for the purposes of a particular research project. Traditionally, interactivity denoted face-to-face communication between people (Kettanurak et al., 2001). With the development of technology, from telephones to radios, televisions and computers, interactivity has taken on a much wider meaning. Indeed the entire field of Human-  58 Computer Interaction (HCI) has developed to address the numerous problems and research questions spawned by the need to create both hardware and software that humans can interact with. It is interactivity in this HCI sense which is the subject of this section; that is "a facility by which a user acts on a computer presentation, which in turn interprets the user's action and produces an appropriate response" (Narayanan and Hegarty, 2002). Even isolated to the domain of HCI, and more specifically the realm of human interaction with computer software, there has been little research or illumination on the contributions of interactivity to learning (Cairncross and Mannion, 2001; Kettanurak et al., 2001; Kirsh, 1997; Yeo et al., 2004). As with many technologies, the development has led the research, with the assumption that providing users with interactive computer software will aid in the learning of the material presented. Before discussing the available research on interactivity, the next subsection will attempt to further bound the discussion of interactivity within the confines of the types of interactivity intended for the experimental interface created for this thesis. Bounding the Discussion of Interactivity  This thesis was developed within the larger scope of research at CALP, and fits into a research thread on the use of landscape visualizations as planning and public involvement tools. The experimental interface developed for this thesis fits along a continuum of landscape visualization and information display which can conceptually be delineated, among other dimensions, according to the amount of interactivity each display mechanism provides to the user. Placing the interface within this continuum will allow for a more precise bounding of the wide-ranging literature on interactivity to more suitably address the use of interactivity in similar applications. Figure 2.8 is a schematic representation of the continuum of interactivity for landscape visualization and information display applications at CALP. Anchoring the low interactivity end of the continuum are static images. These images may take many forms, from photographs to simulations, maps, or graphical displays. Static images provide low interactivity because the information they present, and the format of that information, are not amenable to change by the user (Kettanurak et al., 2001). Static images have been used in a number of perceptual studies at CALP to investigate people's responses to different land management techniques (see e.g. Lewis and Sheppard, 2004; Picard and Sheppard, 2005). Linear slideshows consist of a series of images placed in a linear sequence, such that the user can advance the sequence at their own pace, but is limited to moving the images forwards or backwards in the prearranged sequence (e.g. Mayer and Chandler, 2001). This type of interface is commonly seen in slide presentation software such as Microsoft's Powerpoint. Limited control videos are animations or video-recordings that can be played on computers through the use of media players such as Apple's Quicktime (Apple Computer Inc., 2004) or Microsoft's Win-  59  ^  r>°"  cr  6?  <^ e  Low Interactivity  \  High Interactivity  Level of interactivity discussed in this literature review  Figure 2.8: A schematic representation of the locations of different presentation media along a continuum from low interactivity to high interactivity. dows Media Player (Microsoft Corp., 2003). The videos themselves are linear sequences of images, however, the media players allow the user to pause, advance and rewind the video content, providing a higher level of interactivity than the video possesses inherently. Schwan and Riempp (2004) discuss the use of limited control video in instructing subjects on learning to tie nautical knots. The shaded area in Figure 2.8 indicates the area of interactivity literature that will be covered in the next subsection. This area begins with the aforementioned limited control video, because of some of the similarities in control mechanisms between the media players and the interface designed for this thesis (e.g. slider bars for advancing and rewinding image frames). The next application along the continuum of interactivity is non-linear slideshows like those commonly created using Macromedia's Director software (Macromedia Inc., 2000). These slideshows are non-linear, because interface controls can be included which allow the user to jump from point to point within the slideshow without having to move in a linear fashion through the successive frames. Relatively complicated interaction schemes can be created using the non-linear slideshow technique, and software developers often use this technique to prototype interfaces for their applications. Salter (2000) is an example of a non-linear slideshow application created at C A L P to communicate the visual impacts of proposed forest harvesting at the Malcolm Knapp Research Forest. The non-linear interface point on the continuum marks the proposed placement of the interface created for this thesis. The interface includes non-linear interactivity with multiple representations and multiple interaction options to display different combinations of the representations.  Chapter  Four of this thesis will discuss the design and implementation of the interface in detail, and further  60 document the interactivity that it presents to the user. The non-linear interface represents the high end of the interactivity continuum to be discussed in the following literature review (as represented by the shaded rectangle in Figure 2.8). The next step along the interactivity continuum can, in some respects, be considered a break or branching of the continuum to a fundamentally different type of interactivity. Real-time movement indicates fluid interactivity within a 3D environment, where the user can position themselves freely at any point in the environment. Common movement modes in real time simulations include walking, driving and flying, which mimic those movement styles in real-world environments. This type of interactivity is qualitatively different than the interactivity discussed earlier along the continuum, which involved the selection of media with prearranged viewpoint and display characteristics. While real time movement is fundamentally different in its interaction style, conventional wisdom suggests that this type of interactivity, common in video games, would be placed higher on the interactivity continuum than those applications previously discussed. Campbell and Salter (2004) discuss CALP's use of a real time simulation of Snug Cove on Bowen Island to investigate residential densities with members of the Bowen Island community. Real-time movement can also be augmented with the use of other interaction strategies to provide even higher levels of interactivity. The next step along the continuum marries real-time movement with the ability for users to make real-time alterations to the environment represented, and the ability to query the underlying information upon which the environment was constructed. The user is not relegated to investigating the visual aspects of the present environment, but can instead query the nonvisual information and make changes to the environment to determine both their visual and non-visual effects. Cavens (2002) developed CALP Forester as a prototypical example of this type of application, allowing users to move through a forested environment, querying the underlying forest information and prescribing management actions to areas of the forest. The high end of the interactivity continuum is a moving target, driven by advancements in technology and understanding in HCI. Developments in Augmented Reality (AR), Virtual Reality (VR), haptic interfaces, as well as other, as yet unknown, developments will continue to drive the level of possible interactivity higher, with both quantitative and qualitative changes in interaction mechanisms. Cavens (2002) and Cavens et al. (2002) discuss the use of a laser pointer as a more natural interaction device for manipulating on screen displays in immersive and collaborative environments, and Hedley et al. (2001) discuss the use of mixed AR and immersive VR technologies along with complex computer vision tracking to provide users with augmented hologram-like displays of natural terrains. These two projects represent examples of how technological development in HCI is driving the high end of the interactivity continuum for landscape visualization and information display. One factor that bears further explanation in investigating the interactivity continuum displayed in  61 Figure 2.8 is the somewhat inverse relationship that exists between the level of interactivity of a given media and the amount of associated evaluation that has occurred. The perceptual effects of static media have been extensively studied. With the increase in interactivity comes a parallel increase in complexity of interaction effects that tends to make empirical evaluation of the individual effects of media, information and interaction mechanisms more difficult. At the high end of the interactivity continuum development typically leads evaluation, an effect common in related fields, such as information visualization. The purpose of this thesis was to find a point along the interactivity continuum that provided more interactivity than static images, linear slideshows, or standard animations, but that was situated such that research on similar applications could be consulted in the design phase, and such that it was amenable to evaluation itself. Interactivity Literature Review  Even with the scope of interactivity to be reviewed here limited to the bounded range depicted in Figure 2.8, it should not be assumed that the available literature is broad and conclusive. In fact, one of the predominant themes in the available literature on multimedia interactivity is the paucity of evidence either supporting or refuting the efficacy of interactivity (e.g. Cairncross and Mannion, 2001; Kettanurak et al., 2001; Yeo et al., 2004). In the available literature, however, there are some common threads, both beneficial and detrimental, that appear to be developing in the study of interactivity and its effects on learning. Benefits of Interactivity  One of the major themes that has arisen in research on interactivity is the  concept of engagement; the assertion being that provision of interactivity increases the user's engagement with the material, which subsequently leads to improved learning. While engagement can be a difficult attribute to quantify, a number of researchers have attempted to qualitatively assess the engagement provided by various media and levels of interactivity. Conroy and Gordon (2004) stated that providing participants at public involvement processes with interactive GIS and web-based material increased their level of engagement when compared with participants that attended standard oneway information flow meetings. Those participants who were provided with individual interactive GIS and web material had higher satisfaction ratings and showed a greater increase in understanding of the issues surrounding watershed management than the subjects that attended the more standard public meeting format. Aldrich et al. (1998) compared two educational CD-ROMs, one with limited interactivity and a predominately linear format, and the other with greater interactivity and less linearity. They found that the increased interactivity led to increased engagement and improved motivation to learn the material. They also found that interactivity led learners to take a more hands on and exploratory approach to learning. Finally, Kettanurak et al. (2001) directly measured user attitudes  62 when using multimedia instructional material of different levels of interactivity and found that high levels of engagement that accompanied highly interactive media led to improved satisfaction with the learning material and subsequently higher levels of performance on assessment tasks. It would appear, then, that by involving users' more in the learning process (i.e. active vs. passive learning) interactive systems increase the user's engagement and motivation, which in turn can lead to improved understanding of the displayed material. Another prevalent thread in the literature on multimedia interactivity is the notion that providing users with the ability to self-pace their instruction allows interactive systems to decrease cognitive load, thereby promoting learning. The rationale is that users can spend more time on the material that is most difficult, reducing the extraneous cognitive load caused by rapid or standardized pacing of instruction. Schwan and Riempp (2004) found that users who were given control over the pacing of video presentations of knot tying did not spend more time overall with the video material, but did spend more time reviewing the difficult portions of the knot tying video. Those subjects that were provided with interactive control showed a better understanding of the demonstrated knots in post session evaluations than subjects who watched the video with no interactive control. The results of this study suggest that providing users with the ability to self-pace their instructions can improve learning by allowing them to focus on particularly difficult (i.e. high cognitive load) material. Mayer and Chandler (2001) conducted an experiment in which the only difference between the subject groups was that one group could control the pace of the slideshow they were presented, while the other group could not. The researchers hypothesized that self-pacing would decrease cognitive load and aid in the development of component mental models of the formation of lightning that could later be incorporated into a more comprehensive model of the overall process. The results of their study showed no difference in the subjects' abilities to recall factual information, but a significantly greater ability for subjects from the self-paced group to transfer their knowledge to novel situations. Transfer is considered to be a measure of deeper learning of material, as it denotes a better understanding of concepts and their application (Mayer, 1998; Paas, 1992). The ability to self-pace the rate of instruction, especially for material with high intrinsic cognitive load, would appear to be a benefit of interactive multimedia. A possible benefit of interactive multimedia which has received less attention is its ability to be adjusted by the user to accommodate different learning styles. In their study on learning to tie nautical knots from interactive videos, Schwan and Riempp (2004) found large individual differences in the amount of time spent on given portions of the video, indicating that subjects were employing different learning strategies. Aldrich et al. (1998) found that teachers, when evaluating multimedia CD-ROMs, preferred the more interactive, less linear CD-ROM, in part due to its capacity to be adjusted to individual learning styles. Kettanurak et al. (2001), in their study of interactivity effects on attitude and  63 performance improvement, found that learning style moderated subjects' attitudes about interactive multimedia, such that subjects of given learning styles felt more or less positive about the experience of using the interactive systems. Because attitude was positively correlated to performance on postsession assessments, this indicates that learning style should have an effect on the efficacy of different levels of interactivity. Winn (1995) states that there are two fundamental factors that influence learning: a) how the information is presented, and b) how the learner interacts with the information. In other words, the two factors are the information and the learner. The importance of the learner, and what they bring to the learning process (i.e. their learning style), indicates the importance of providing material that is adaptive to the learner's needs. While more research is required, there is some evidence that interactivity can address individual differences better than traditional media by allowing the user to accommodate the information presented to their particular learning style. Limitations of Interactivity The incorporation of interactivity into instructional environments should not be seen as a panacea. Though interactivity may provide some inherent benefits, it may also confer some disadvantages. One limitation which has been found by a number of researchers is the tendency for users of interactive multimedia systems to move through instructional material rapidly, without spending significant time with particular aspects of the instruction. Yeo et al. (2004), in their study of an interactive interface for physics instruction, found that students often moved on to new material before sufficiently learning particular concepts. Researcher intervention remedied this problem, changing the students' interaction patterns and causing them to spend more time with individual physics concepts. In effect, students needed to be taught how to properly use the interactive media, and benefited from the presence of an instructor (Yeo et al., 2004). Kettanurak et al. (2001) found a similar effect, where highly interactive systems actually decreased post session achievement scores for subjects because they moved through the material without viewing all of it, or sufficiently learning specific concepts. Narayanan and Hegarty (2000, 2002) have called the conventionally accepted superiority of interactive multimedia to traditional media into question. In two separate studies they found that material prepared according to cognitive instructional guidelines showed no differences in learning outcomes, whether presented using traditional print media, or interactive multimedia (Narayanan and Hegarty, 2000, 2002). These findings corroborate the oft cited, and controversial, work of Clark (1983), that stated that media does not influence learning. This notion has often been disputed (see e.g. Kozma, 1991), but the fact that such a fundamental tenet of interactive multimedia is still in question indicates the relative immaturity of research in this area. From the available literature, then, there appears to be room for guarded optimism about the positive effects of interactivity in creating engagement, and providing the user with control of the material being presented. Significantly more research is required, however, and researchers should not accept  64 that the benefits of interactivity are well established. Interactive systems should be assessed according to the intended application, and user testing should be performed both to determine how the interaction mechanisms are being used, and whether or not their are aiding users in learning the material. 2.3.9  M u l t i p l e Representations  To this point in the literature review on information displays, the focus has been on individual display types, such as graphs, maps, text and animation. Each of these display types has its own inherent properties, and associated strengths and weaknesses. The promise of multiple representation systems is to draw upon the strengths of one or more representation types to augment the strengths of another representation type, or to overcome its weaknesses. An even loftier goal of pairing multiple representations is to create a synergistic effect, where the effectiveness of the collected representations is greater than the sum of the individual representation types (Kaput, 1989). Ainsworth (1999) states three main reasons for using multiple representations: 1. When different learners exhibit preferences for different representations (i.e. different learning styles). 2. When the user of the representations has different tasks to perform with them. 3. When learning more than one strategy or way of looking at the problem improves performance. Research into the use of multiple representations has been conducted primarily by two research groups: Robert Kozma and his colleagues at The Center for Technology in Learning (an independent think-tank in Menlo, California), and Shaaron Ainsworth and her colleagues at the University of Nottingham. Kozma has looked extensively at the use of multiple representations by experts and novices, primarily in the domain of chemistry (Kozma, 2003; Kozma and Russell, 1997). Ainsworth has studied the use of mathematical representations (Ainsworth et al., 2002), reviewed the use of multiple representations in other applications (Ainsworth, 1999), and developed theoretical models for the understanding of multiple representations (Ainsworth, 1999), and their application to dynamic displays (Ainsworth and VanLabeke, 2004). Reviewing the work of Kozma and Ainsworth, as well as studies by other researchers, reveals some common threads about the application of multiple representations. The remainder of this section will discuss these threads and their importance to the use of multiple representations. Translation Between Representations  One of the key factors in the effective use of multiple representations is the viewer's ability to translate between representations (Ainsworth, 1999; Ainsworth et al., 2002; Bodemer et al., 2005; Kozma, 2003). In order for multiple representations to aid comprehension of material, the viewer must be able to determine how the different representations correspond. For example, in order to understand the  65 correspondence between animation of tree growth and a graph showing increasing volume in a stand of trees, the viewer needs to be able to equate the growing trees with an increase in wood volume. In a study of expert chemists versus novice chemists (university undergraduates), Kozma and Russell (1997) found that experts were far superior at translating between different representations of the same chemical phenomena. When asked to form pile sorts of related groupings, experts placed more media types into the same group, while novices tended to group according to individual media types (Kozma and Russell, 1997). In think-aloud protocols, Kozma and Russell (1997) found that experts focused on conceptual aspects of the phenomena being represented, while novices focused primarily on surficial features of the media, making translation more difficult. This surficial focus is reminiscent of the perceptual problems that novice users experienced when viewing animations of weather systems in studies by Lowe (2003, 2004), and suggests that novices were not making optimal use of the graphical representations. Bodemer et al. (2005) speculate that the difficulty in attempting to translate between multiple representations of the same phenomenon may actually inhibit the formation of an appropriate mental model about that phenomenon, and therefore prevent effective learning. A study by Ainsworth et al. (2002) supports this assertion, as subjects that used multiple representations improved their distance estimations on math problems only after they stopped trying to translate between representations and focused on a single representation. Ainsworth et al. (2002), however, point out several caveats in the interpretation of their results. First of all, the subjects used in the study were children who were unfamiliar with one of the studies representations. Children in a subject group that only used that representation required longer to show accurate distance estimations than another subject group that used a more familiar representation. The interfaces required to control the representations also differed, with one representation being controlled though keyboard input, and the other through mouse input. Finally, the modalities of the two representation types differed as well, with one representation type being presented as text, and the other representation type being represented graphically. One of, or a combination of these factors may have played a role in the lack of translation demonstrated by the subjects in the study (Ainsworth et al., 2002). Brenner et al. (1997) did find that novice users could make translations between different media (graphs, tables, equations, etc.) in an in situ  classroom study. This study took place over a one month period, however, meaning that the  subjects had a significant period of time in which to learn to translate the representations. Given the crucial importance of translation to the effective use of multiple representations, further research on novice users' ability to translate representations is required. Thus far the research is equivocal, with support both for and against the ability of novices to make translations. The facility that expert chemists showed in translating between representations in the Kozma and Russell (1997) study suggests that familiarity with representations may play a role in translation, and Ainsworth et al. (2002) states that the more dissimilar representations are, the, more difficult translation will be. It may  66  be that effective use of cues and linking mechanisms, as well as common interfaces and familiar representations may aid the translation process. Interactions of Multiple Representations  Kozma (2003) and Kozma and Russell (1997) do not speak explicitly to the interactions between representations. They focus much less on the interactions themselves, and more on how the attributes of the viewers affect their use of multiple representations. Kozma (2003) does, however, speak to the importance of linking multiple representations, such that novice users can more easily connect the salient features of each display. Linking representations can be accomplished through the use of symbolic connections such as colour coding and labels. It can also be accomplished by ensuring that related actions in different representations have simultaneous onset, equal runtime and similar interface controls (Kozma, 2003). Cues such as colour coding may be able to capitalize on novice subjects tendency to focus on surficial and perceptually salient features of the representations to make connections between the representations and foster a conceptual understanding of the phenomena represented. Ainsworth (1999) and Ainsworth and VanLabeke (2004) cover the interactions between representations in a much more explicit fashion. Ainsworth (1999) delineates three functions of multiple representations: complementing, constraining and constructing. Constructing relates to the function of multiple representations in facilitating a deeper understanding of the represented phenomena. Construction will be discussed in the next subsection, as it represents the main goal of using multiple representations (and indeed all information displays). Complimenting and constraining functions, on the other hand, relate directly to the interaction between representations, and can each lead to the construction of knowledge if properly applied. Complimenting refers to the use of more than one representation to make up for limitations in a single representation (Ainsworth, 1999). The additional representation may provide more information, or it may facilitate different processes. For example, a mathematical equation and a graph of that equation are informationally equivalent (i.e. either representation can be derived from the other), but they facilitate very different interpretation processes. Ainsworth (1999) also outlines the differences between separate representations (i.e. they contain completely different information) and partially redundant representations. Partially redundant representations may be useful as a means of providing a different perspective on the presented information, and both partially redundant and completely separate complimentary displays may be useful for compensating for individual differences in users' representation preferences and learning styles (Ainsworth, 1999). Multiple representations can also be used to constrain the interpretation of one another, either to prevent misinterpretation, or to focus attention on task-specific information (Ainsworth, 1999). This constraint can occur in two main ways; first a familiar representation such as a map or graph can be  67 used to constrain the interpretation of an unfamiliar representation (e.g. a novel information visualization representation); and second, the display properties of a representation can be used to constrain the interpretation of another familiar representation (Ainsworth, 1999). An example of the second type of constraint is the use of a sketched map, with its inherent spatial representation, to constrain directions given verbally. The use of one or more representations to constrain interpretation of another representation may be an effective way to aid users in directing their attention towards the important information in a given display. This is especially important with novice users, given that they have a difficult time determining what information is relevant to a particular task (Lowe, 2003, 2004; Kozma, 2003; Kozma and Russell, 1997). In their paper discussed earlier in the dynamic representation section of this chapter, Ainsworth and VanLabeke (2004) related the different dynamic representation types they had identified to the complimenting and constraining functions of multiple representations. By their nature, time-persistent, timeimplicit and time-singular representations display different information (Ainsworth and VanLabeke, 2004). Time-persistent representations allow the user to offload computation onto the representation, because the persistence of values means that they do not need to be held in working memory. Timesingular representations, on the other hand, are transient, showing only one value for a given variable at a time. This feature of time-singular representations means that they are concise, and that values for a particular time period can be rapidly assessed (Ainsworth and VanLabeke, 2004). These different features of time-persistent and time-singular representations can be used to either compliment or constrain interpretation of one another. For example, the explicit representation of time in time-persistent representations can constrain the interpretation of the present value on a time-singular representation by placing it within the context of all previous values. The complimentary and constraining functions of multiple representations raise key methodological issues in the design of information displays. How can these functions be best utilized to enhance the communicative abilities of multiple representation environments? Constructing Knowledge from Multiple Representations  As mentioned in the translation subsection, Kozma and Russell (1997) found that experts translated easily between representations as a result of their ability to see through the surficial features of the representation to its conceptual information. Through a greater familiarity with the various representations, and a greater understanding of the knowledge domain, experts develop the ability to abstract from the specific conventions of a given representation and translate the underlying information to coincide with other representations (Kozma, 2003; Kozma and Russell, 1997). In effect, experts have more developed schemas which they can bring to bear on the comprehension of a given representation and the phenomena it represents. Novices, on the other hand, have difficulty constructing knowledge  68  from representations because they focus on the surficial, or perceptually salient features of the display (Lowe, 2003,2004; Kozma, 2003; Kozma and Russell, 1997). Novices, therefore, do not abstract the underlying information from the representation, preventing them from integrating the information from multiple representations into a conceptual understanding of the represented phenomena. Ainsworth (1999) also emphasizes abstraction as an important process in the construction of knowledge from multiple representations, however, she includes two other potential processes in her discussion: extension and relation. Extension, as the term implies, denotes extending knowledge already acquired to a new domain. For example, a person who has knowledge about the use of line graphs can extend that knowledge to reading and comprehending information from a subject area they are not familiar with. Use of familiar representations, then, should facilitate comprehension of information that is new to the user, because they can extend their knowledge of the representation to the new domain. Relation involves a subtle, almost circular process. The relational process improves understanding of the underlying information by requiring the user to relate (i.e. translate) multiple representations (Ainsworth, 1999). By presenting the user simultaneously with multiple unfamiliar representations, they are encouraged to decipher the various representations and to attempt to relate them. The need to relate the representations should lead to abstraction and a conceptual understanding of the knowledge domain (Ainsworth, 1999). The subtle difference here, is that the focus is on the representations themselves, and the comprehension of their symbol systems, rather than on the underlying information. The comprehension of the information comes as a byproduct of the need to relate the representations. Improving comprehension and conceptual understanding of the represented knowledge domain is the goal of utilizing multiple representations, and indeed, all information displays. Understanding how people construct deeper understanding from these displays, therefore, is extremely important. The work by Kozma (2003) and Kozma and Russell (1997) indicates that the development of expertise leads to the ability to derive conceptual information from multiple representations, and the ability to integrate that information into an understanding of the represented phenomena. If multiple representations are to be used to improve understanding among non-experts, however, means must be determined to accelerate the process of translation and knowledge construction. Ainsworth (1999) points the way to one possible solution through the extension of pre-existing knowledge and familiar display types. The methods of complimenting and constraining discussed in the Interaction of Multiple Representations subsection may also provide means for improving translation between images (Ainsworth, 1999). Finally, provision of cues to link multiple representations such as colour coding, simultaneous onset, and common interfaces may aid novice learners in translating between representations and constructing knowledge (Kozma, 2003). Designers of multiple representation environments and information displays should consider these tools in constructing their systems to attempt to facilitate the transfer of information.  69 2.3.10  A B r i e f N o t e o n H C I f o r the Interface D e s i g n  The main purpose of the interface created for this thesis was to provide an evaluation mechanism for the pairing of landscape visualizations and information display. The interface interaction mechanisms, therefore, were conceived to be a conduit to the presentation and manipulation of the media presented. The intention was not to create a novel interaction style, but rather to provide standardized controls with interaction styles that would be familiar to casual computer users. Researchers in the field of HCI work on the development of more natural computer interaction mechanisms (see e.g. Raskin, 2000; Shneiderman, 1998). Alterations to the metaphor of navigation and interaction with computer interfaces was beyond the scope of this thesis. The decision was made early in the process of developing the interface to utilize the common "direct manipulation" interface type, which is the dominant desktop computer metaphor in use today (Norman, 1988; Shneiderman, 1998). The Macintosh (Apple Computer Inc., 2005) and Windows (Microsoft Corp., 2002) operating systems are examples of direct manipulation interfaces. Direct manipulation denotes the use of an input device (e.g. a mouse or keyboard) to interact with on-screen elements such as buttons, menus, slider bars, etc. (Shneiderman, 1998). One of the fundamental ideas behind direct manipulation interfaces is that they provide affordances that mimic similar actions in the real world, and guide interaction (Norman, 1988). An example of an interface affordance is the visual stimulus provided by an on-screen button when it is depressed. Graphical, and occasionally auditive, indicators are used to signify to the user that the button has been activated, emulating the effects of pressing a button in the real world. While the metaphor behind the direct manipulation interface has been criticized (e.g. Raskin, 2000), it is the dominant mechanism for human-computer interaction in use today. In the previous discussion of information visualization it was stated that novel presentation formats could lead to cognitive load, and that they should not be paired with information from a domain that is unfamiliar to the user (Kirschner, 2002; Pollock et al., 2002; Sweller et al, 1998). With the focus of the interface for this thesis being the presentation of multiple information sources, and the evaluation of how those information sources were perceived and integrated, it was decided that the interaction mechanisms should be familiar. Further discussion of the design and implementation of the direct manipulation interface used in this project follows in Chapter Three. 2.3.11  Conclusion  The intended structure of this literature review was to move from the simpler aspects of information display such as individual graphs, maps and text, to the more complex features such as dynamic content, interactivity and multiple representations. In conducting the review it became apparent that there was an inverse relationship between the complexity of a representation and the amount of available research (see Figure 2.9). Therefore, there was an abundance of information about the perception  70 and use of different graph formats, and a dearth of information about the use of multiple combined representations.  Simple e.g. Static images  Complex e.g. Multiple representations  Abstract e.g. Line graphs  Realistic e.g. Landscape visualizations  High  Understanding and available research  Low  Figure 2.9: A schematic representation of the relationships between characteristics of different representation types and the available literature. There was a similar trend relating to the amount of abstraction or realism a given representation portrayed. There is a significant body of literature on schematic diagrams and simple graphs, but very little on the effects of realistic images, such as landscape visualization (see Figure 2.9). The difficulty in evaluating complex representations that has led to the paucity of information on some information display techniques has ramifications for the evaluation of the interface itself. These ramifications will be discussed in the Evaluation Chapter (Chapter Four).  Chapter 3  Interface Design and Implementation 3.1  Introduction  This chapter pertains directly to the first research objective for the thesis: To research, design and implement a tool to effectively communicate information about ecological landscape conditions. The tool should show landscape visualizations in association with other information representations. The goal of the tool is to communicate information about the landscapes temporal and spatial aspects, and about visible and nonvisible indicators of landscape condition to non-expert users. The chapter will draw upon the findings of Chapters One and Two to inform the detailed design and implementation of the information display interface. The chapter will begin with a discussion of the general design principles that have arisen from the literature reviews and research objectives, as well as some of the practicalities of implementing the design. The chapter will then turn to a discussion of the individual elements of the interface, their data requirements, and how they were designed and implemented. Once the design and implementation of the individual elements has been discussed, the completed interface will be reviewed. The chapter will then turn to the author's evaluation of the interface according to a set of criteria developed to assess its success for the intended application.  3.2  General Interface Principles and Attributes  Principle One: Combining Landscape Visualization with Other Information Representations  There were three general principles related to the thesis' research objectives which formed a conceptual framework within which the interface design would occur. The first principle, which has been discussed previously in this thesis, was to use landscape visualization as a means of improving the communication of non-visible indicators of landscape condition. The engagement and communica71  72 tion capabilities of realistic landscape visualizations have been documented (e.g. Al-Kodmany, 2000; Campbell and Salter, 2004; Sheppard, 1989), as have the difficulties presented by more traditional means of communicating landscape information (e.g. Forest Practices Board, 2000). Realistic landscape visualizations, however, are limited, in that they emphasize the visual aspects of landscape condition, and do not represent the non-visible aspects. The first principle of the design of the interface was to draw upon the engagement and communication capabilities of realistic landscape visualization to improve the communication of non-visible indicators of landscape condition being represented in complimentary information displays. Principle Two: Usable by Non-Expert Users  The second general principle influencing the interface design was that it was intended to be used by people who were not experts in resource management or landscape planning. This principle assumes a general interest in the information presented by the intended viewer, i.e. that the person who would use this type of interface would be the same type of person that would attend a public meeting, read a government mailout or visit the website of a landscape management agency. In other words, the intended audience for this type of interface would be people with an interest in how the landscape is managed, but without domain-specific knowledge or skills that would facilitate their comprehension of landscape management information. Research by Kozma (2003) and Kozma and Russell (1997) on the differences between experts and non-experts in the use and comprehension of information representations has definite implications for the representation of information for non-expert users. Research by Chen and Czerwinski (2000), Graham et al. (2000), and Sutcliffe et al. (2000) on information visualization suggests that novel information display mechanisms are less effective with users that are unfamiliar with a particular knowledge domain. Finally, Ainsworth (1999) states that the use of familiar display mechanisms can aid the comprehension of novel information by non-experts by allowing them to extend their understanding of the display (e.g. a bar graph) to the comprehension of novel information. The non-expert status of the intended audience, therefore, has specific ramifications for the design of the interface as a whole, and for the design of the individual interface elements. Principle Three: Evaluation of the Interface  The third general principle that guided the design of the interface was the intention to evaluate the interface with human subjects upon its completion. This meant that the interface was created for evaluation, rather than the evaluation being a post-implementation addition to the process. The intent to evaluate the interface had specific impacts on the interface's design. The more interface elements that were added, the more difficult it would be to discern which element or combination of elements were leading to a particular effect. It was, therefore, necessary to consider before constructing the in-  73  terface what aspects of the interface were most important to test. For a working prototype of an actual decision-support tool, the functionality would be very different than for a research tool that was created to investigate specific aspects and interactions of the interface elements. This desire to evaluate the interface led to a design goal of simplicity, limiting the included interface elements and interaction mechanisms to only those required to evaluate a particular interface functionality. The three general design principles stated above can be summarized as: 1. The combination of realistic landscape visualization with some form of information display depicting non-visible indicators of landscape condition. 2. An intended audience of interested, non-expert members of the public. 3. The intended evaluation of specific aspects of the interface, and the consequent requirement to keep the interface simple. Principle one and principle two relate directly to the first research objective for the thesis. Principle three relates to the first research objective in its design requirement for simplicity, and to the second research objective in its recognition of the need to evaluate the interface with non-expert users. Given these overriding principles, the next step in the process was to address important considerations pertaining to the overall design framework of the interface. 3.2.1  General Interface Attributes  Given the three general principles stated in the previous section, and the results of the information display literature review in Chapter Two, it was possible to make some decisions about general attributes the interface should possess. Detailed discussion of the interface elements and their implementation will follow; this section will discuss conceptual issues about the interface's layout and composition. Single Window Layout  The need for simplicity in the information display, resulted in the decision to maintain all of the interface elements in a single window. This meant that all of the available information types needed to be visible at one time, without relying on the use of dialog boxes or pop-up windows to access more information. The main benefits of this approach were simplicity of display, and assurance that all of the subjects in the subsequent evaluation would see the same information types (though not necessarily all of the same information). The simplicity of the single window approach was beneficial for the evaluation process due to the limitations of time and attention that human subject evaluations present. The intended audience of non-expert users also made the focus on simplicity an important consideration. To be beneficial to non-expert users in a time-limited public meeting or web-based format, an interface of this type would need to be easily understood, and to deliver information quickly. The simplicity of a single window display, while limiting the amount of information that could be presented,  74 should make the display easier to comprehend. From an evaluation perspective, the single window approach prevents the subjects in the study from viewing fundamentally different information. The subjects may view different information within a given display type (e.g. different years of a landscape visualization), but they would not view completely different information (e.g. one subject viewing all landscape visualizations, and another subject viewing all graphs). The decision to present the interface in a single window introduced limitations to the interface design as well, because it meant that all of the interface elements had to be presented to the user at one time on limited screen space. This meant that compromise was necessary between the inclusion of more elements and the size of each element. Legibility  Related to the trade-off between the number of interface elements and the size of individual elements is the issue of legibility. The legibility of individual elements is directly, though not exclusively, related to the size of the element. For text, font size is a key determinant of legibility (Bringhurst, 1999), and has implications for both the presentation of explanatory text, and for labels on interface elements such as maps and graphs. If the text used on interface elements was not sufficiently large and clear to be read, then significant information loss would occur. For image intensive elements such as maps and landscape visualizations, legibility is important because of the viewer's need to discriminate details. If improper element sizing, colour selection, lighting, etc., make a map or visualization image illegible, then important information can be lost. Legibility also has implications for interface controls, such as buttons, menus and slider bars. If interface controls are legible, then it should be apparent to the user what their intended purposes are. In the words of Norman (1988), the interface controls provide "affordances" which direct the user to their intended purpose. If the interface controls are illegible, then their purposes are obscured, and the interface may become confusing to the user. It should be noted that interface element legibility here refers specifically to the attributes of the interface itself, and not to attributes of the user. Attempts were made in the creation of the interface elements to address the common problem of colour deficient vision, but not to address problems of visual acuity. Colour deficient vision is a prevalent phenomenon in the human population, but with careful design it can be accounted for (Brewer, 1999; Meyer and Greenberg, 1988; Olson and Brewer, 1997; Ware, 1999). Efforts made in the design of interface elements to account for colour-deficient vision will be discussed in the relevant sections of this chapter. Problems with visual acuity, and other vision anomalies are more difficult to design for, and are beyond the scope of this thesis. Accessibility issues in interface design are a field of study in and of themselves, within the larger realm of HCI. In order to avoid the need to address visual acuity issues, the ability to interact with a graphical user interface was made a condition for participation in the evaluation of the interface.  75 Familiarity of Display Elements  As noted earlier, novel information coupled with novel display mechanisms can hinder comprehension for non-expert users (see e.g. Ainsworth, 1999; Chen and Czerwinski, 2000; Graham et al., 2000; Sutcliffe et al., 2000). Drawing upon familiar display mechanisms, on the other hand, may facilitate comprehension of novel information by allowing viewers to draw upon their knowledge of the symbol system of the familiar display (Ainsworth, 1999). Novel display mechanisms add to the cognitive load of the presented information by requiring the user to decipher their symbol system as well as the information they present. Familiar displays reduce cognitive load by allowing the user to access their pre-existing schemas for the familiar display to efficiently process the new information. Realistic landscape visualizations may not be familiar to viewers, however, they likely evoke the viewers' schemas for photographs of landscapes, which are a familiar display mechanism (e.g Shuttleworth, 1980; Stamps III, 1990). The prevalence of information graphics, both in education, and in everyday life have made reference maps, several graph formats, and diagrams familiar representations of information to a large portion of the population (Feeney, 1994; Wainer, 1992). Populating the interface with information display elements that leverage the familiarity many people have with common representations may serve to facilitate the transfer of information to the users. For this reason the inclusion of familiar display elements was a key design attribute for the interface. Simple, Standard Interface Controls  In keeping with the general design philosophy of simplicity, one of the key attributes of the interface was the use of simple, recognizable controls. The interface, therefore, used the direct manipulation metaphor made popular by its incorporation into the Windows and Macintosh operating systems. With a collective share of approximately 97% of the desktop computer market, the metaphor provided by these operating systems reaches a significant portion of the public (W3Schools, 2005). Providing interface controls that draw upon this familiarity, therefore, should facilitate the viewers' interactions with the interface. Just providing standard controls, however, would not ensure easy comprehension, because confusing arrangements and an excessive number of controls could hinder understanding of the presented information. The design of the interface controls, therefore, required not only the inclusion of familiar controls, but also keeping those controls minimal and easy to use. Interactivity  As discussed in Chapter Two, the research on the effects of interactivity on learning has been inconclusive. Interactivity has been shown to increase subjects' engagement with the learning material (e.g. Conroy and Gordon, 2004; Aldrich et al., 1998; Kettanurak et al., 2001), and the self-pacing of material provided by interactivity has been shown to improve learning (Mayer and Chandler, 2001; Schwan  76  and Riempp, 2004). However, the provision of interactivity may cause users to move through material rapidly, thereby missing important concepts (Kettanurak et al., 2001; Yeo et al., 2004). The provision in the interface of multiple landscape visualization viewpoints, multiple time steps, and multiple information representations meant that the only options for display were non-interactive dynamic representations (i.e. animations) or interactive dynamic representations. Research on animations has suggested that they present cognitive load problems for learners (e.g. Bodemer et al., 2004; Hegarty, 2004; Jones and Scaife, 2000; Lewalter, 2003; Lowe, 2003, 2004; Ploetzner and Lowe, 2004). Given the choice between non-interactive and interactive dynamic representations for the interface, then, the decision was made to included interactivity and self-pacing. The inclusion of interactivity provided a design cue for individual interface elements, as they all had to be made amenable to interactive control by the user. 3.2.2  Required Technical Attributes  In addition to the general attributes of the interface documented above, there were a number of interface-wide technical attributes that affected the design and implementation of the interface. Specific technical attributes of display elements (such as graph axes, or visualization resolution) will be discussed in their respective sections. The technical attributes to be discussed here have implications for the design of the interface as a whole. Screen Resolution  It is common in modern graphical user interfaces for the interface window to be sizeable/scalable so that the user can set the window size to the resolution of their display. The alteration of window size can either be accomplished by hiding interface elements when the window decreases in size, or scaling the interface elements to meet the size of the window. For two main reasons, the decision was made early in this project not to implement the sizing/scaling feature, but rather to standardize the interface at a particular resolution. The primary reason was the need to standardize the interface's display characteristics for the subsequent evaluation. By setting the interface to a particular resolution, the size and layout of the interface elements could be controlled for. A secondary reason for maintaining a given resolution was the image intensive nature of the interface. Scaling images to fit different display resolutions may have caused differences in the clarity of the images, which could have affected the perceptual response of the subjects in the evaluation (i.e. smaller images displaying less detail). Once the decision was made to standardize the interface resolution, the obvious question became what resolution to standardize to? In order to answer this question common monitor resolution statistics were investigated. There are a number of websites that track characteristics of computers that browse the internet. These statistics provide the best estimate of prevalence of different screen reso-  77 lutions, and therefore what resolution would be the most likely one that users of an interface like the one created for this thesis would be using. Table 3.1 shows recent screen resolution statistics. Table 3.1: The percentage of internet users utilizing different screen resolutions. Source (Channel Minds, 2005) Screen Resolution  Percentage of Users  640 X 480 pixels  1%  800 X 600 pixels  31.7%  1024 X 768 pixels  48.3%  1280 XI024pixels  13.6%.  1600 X 1200 pixels  1%  [Other I  7  ' .'; .'  • ' :[ :  4.2"/. (different aspect ratios, e.g. 1152 X 870 pixels) j  These statistics indicate that approximately half of current internet users work at 1024 X 768 pixel resolution, and that well over half (67.1%) use a resolution greater than or equal to 1024 X 768. Selecting a standard resolution involved a trade-off between the size and acuity of interface elements, and the likelihood that normal viewers would use this type of interface at a given resolution. A final consideration was the possibility of using the interface in CALP's Landscape Immersion Laboratory (LIL) for future evaluation. Each of the projectors in LIL run at 1024 X 768 pixels as their native, and therefore clearest, resolution. Taking all of these factors into consideration, the decision was made to standardize the interface at 1024 X 768 pixels, also known as XGA resolution. This resolution provided the best mix between common usage and interface element size, as well as the ancillary benefit of flexibility for future web and/or LIL applications. The chosen screen resolution had a significant effect on the design and implementation of the interface, as it dictated the number of display elements and their sizes. The Java Programming Language  Selection of the appropriate programming language to use in the implementation of the interface was another key concern that would have ramifications for the interface as a whole. The Java programming language was selected for three main reasons: 1. Java is cross platform - By using Java the interface is not tied to any particular operating system. The CALP labs include computers with Windows, Linux, Macintosh, and SGI IRIX operating systems. The interface will run on all of these operating systems without the need for further configuration. This gives significant flexibility in the mode of presentation of the interface to subjects for research. 2. Java allows for rapid development - One of the main strengths of the Java programming language is its modular construction. There are thousands of freely available libraries of code that  78  can be used to extend the capabilities of an application. Previously written code is also easily incorporated into new applications, this means that it is relatively fast and easy to modify the interface to reflect different display organizations and data configurations. 3. Java is designed for network applications - Java was designed to be used in client/server applications. This means that the interface could easily be modified to run as an applet in any Java enabled web browser, with the data being served from a central location. This would allow for distributed experiments in a controlled lab setting, with remote communities, or widely distributed over the Internet. The use of the Java programming language provided flexibility in the construction of the interface, as well as allowing for future implementations and evaluations with different delivery mechanisms such as web delivery and presentation to small groups in the Landscape Immersion Lab. Data  The data available for use in the interface was discussed in Chapter Two, and will be discussed in more detail in relation to the individual interface elements later in this chapter. The available data sources did, however, have a significant impact on the technical design and implementation of the interface, because they dictated what it was possible to display. The interface required a comprehensive data set, from which both spatial and temporal data sources were available. Spatial data needed to cover the same geographic area, and temporal data needed to run over the same time frame, and at the same time intervals. As mentioned in Chapter Two, the Arrow IFPA data set met these criteria, and provided a range of modeled data sources from which to populate the interface. The Arrow data set, however, was not originally constructed with the intention of enabling this type of information interface, and there were limitations to the data which had implications for how the interface was constructed. A prime example of the limitations of the Arrow data set were the time intervals used to generate information on ecosystem characteristics, ecosystem representativeness, and species habitat in the SIMFOR model. Because the SIMFOR model involves complicated calculations, and takes a significant amount of time to generate results, it was run on much longer time intervals than the ATLAS harvesting model, FORECAST stand attribute model and the CALP visualization system. This meant that some valuable ecological information that may have provided useful indicators for the interface was not available for inclusion. The scale of the modeling process also meant that re-running the models to correct time-scale discrepancies, thereby making the information available, was not a feasible option. Another limitation of the available data was the aggregation of data at the decadal scale. In order to simplify the modeling process, ATLAS ran the harvesting model at ten year intervals. From a numerical modeling perspective this had little effect on the validity of the results. However, from a landscape visualization perspective, the result was an image every ten years with the aggregated harvesting of the last decade represented as though it was all freshly harvested. The result was landscape visualiza-  79 tion images that appeared to display a higher concentration of harvesting than was actually occurring at any one time. While these limitations in the data set provided significant constraints on the construction and population of the interface, it should be noted that for the purposes of designing and evaluating the interface, the validity of the data was not of primary importance. The purpose of this project was not to conduct an actual public consultation process, where the veracity of the information presented was crucially important. Rather, it was to construct and evaluate a tool that represented landscape information in a fundamentally different way than it is normally presented. In other words, the purpose of this study was to investigate how the information was presented, not what information to present. While the specific data and indicators used in the interface were of less importance to the creation of the interface, the use of a cohesive data set that would make sense to the user was very important. The interface needed to present to the user a coherent picture of the landscape, with indicators of the landscape's condition that had some cohesive unifying theme. Given the available data sources from the Arrow IFPA modeling project, the decision was made to focus on ecological indicators, and specifically, on indicators of forest condition. The indicators themselves will be discussed as they relate to the individual interface elements; the emphasis here is on the constraining and organizing role that the available data sources had on the construction of the interface.  3.3  Interface Elements  Drawing upon the literature from Chapter Two, and the principles and attributes discussed previously in this chapter, it was possible to make decisions about what information display elements to include in the interface itself. One of the principles of the interface was to combine realistic landscape visualizations with other information display representations. Therefore, the inclusion of realistic landscape visualizations was a precondition for the interface. Because much of the information to be presented in the interface was spatial in nature, and came from a specific geographic area, a map was included to help orient the user. While the landscape visualizations provided spatial information, the viewpoints used were predominately perspective and orthographic views (see landscape visualization discussion later in this chapter). The perspective views did not cover the entire geographic extent of the study area, and therefore the user may become confused as to their location, view direction, etc. Reference maps are familiar representations in everyday life, and research studies have shown that simple reference maps do not suffer from an expert bias (i.e. there is no benefit in map reading ability conferred by expertise in a geographic or related field) (Gilhooly et al., 1988; Kulhavy and Stock, 1996; Winn and Sutherland, 1989). For the alternative information display elements that were to be combined with the landscape vi-  80 sualizations, the intention was to indicate non-visible aspects of landscape condition. It was also the intention to use familiar display mechanisms to convey the information about indicators of landscape condition. Line graphs have been shown to be an effective means of communicating trends in a variable over time (e.g. Carswell et al, 1993; Kosslyn, 1989,1994; Schutz, 1961a,b; Shah and Hoeffner, 2002; Shah et al., 1999; Tan and Benbasat, 1993; Zachs and Tversky, 1999). Because the information presented in the interface was to have a temporal component, demonstration of trends in a variable over time was a desirable feature. Line graphs are also time-persistent, with all previous values being represented at any one time (Ainsworth and VanLabeke, 2004). This time-persistence has the benefit of providing contextual information to the currently visible information on the landscape visualizations. A text area was also included in the interface, in order to provided additional explanatory information that was not available from the the landscape visualizations, map, or graphs. The inclusion of text also provided for the possible benefit of dual coding, with both the visual and auditive channels of working memory being employed. As another form of information display about the condition of particular indicators of landscape condition, draping of spatial GIS layers on the landscape visualizations was also employed. The intention of the draped information was to help the user visualize the spatial locations and extents of the indicators of landscape condition that were represented in a non-spatial manner on the line graphs. A secondary purpose was to help the user translate between the line graph representations and the landscape visualization representations. Translation between representations is a key factor in the successful use of multiple representations, and has been shown to be difficult for non-experts (Ainsworth, 1999; Ainsworth et al., 2002; Bodemer et al., 2005; Kozma, 2003). Finally, interactive controls were also included in the interface to allow the user to navigate along the temporal dimension, to select landscape visualization viewpoints, and to select which spatial information to drape on the landscape visualizations. The discussion will now turn to the individual interface elements, and a more detailed discussion of their design and implementation. Realistic Landscape Visualizations  The process of creating the landscape visualizations used in the interface was discussed in Chapter Two, along with the rest of the modeled data sources provided by the Arrow IFPA project. This section will begin with a brief review of that process, as well as a discussion of alterations made to the landscape visualizations generated for the Arrow IFPA for inclusion in the interface. As mentioned in Chapter Two, the ATLAS forest harvesting model generated the spatial information about forest stands and harvested area locations, and projected the spatial condition of the landscape over a 265 year time frame. This spatial information was used to generate the spatial pattern of the  81 forested landscape for the landscape visualizations. The FORECAST stand attribute model provided the visualization system with information on tree species composition, tree height and tree density within a particular stand. Through the data generated by ATLAS and FORECAST, then, it was possible to derive the locations and stand characteristics of all of the forested areas in the Lemon Landscape Unit. Over the course of the Arrow IFPA project, three different landscape management scenarios were constructed and modeled in ATLAS, FORECAST, SIMFOR, and the CALP visualization system. For the purposes of this thesis, the Forest Practices Code scenario was selected as the appropriate one to use. The Forest Practices code was selected because it reflected the current management practices at the time of the Arrow IFPA project, and because both the Zoning and the Criteria/Indicators-led scenarios included novel harvesting practices which may have confused the users of the interface. The purpose of this thesis was to evaluate the interface, and the combination of landscape visualization and information display techniques for communicating landscape information. The purpose was not to investigate the landscape management scenarios themselves. The main concern in selecting a scenario, then, was to standardize to a typical set of images and associated data. The main hurdle in utilizing the modeled information derived from the ATLAS and FORECAST models was translating that information into a form that the landscape visualization rendering engine could understand. A translating application called CECIL was developed to take ATLAS and FORECAST information and translate it into a series of scripts that 3DNature's World Construction Set could use to generate landscape visualization images (Cavens, 2002). Other necessary information for the construction of the landscape visualization images included terrain data, road locations, stream locations and lake locations. This information was provided by British Columbia Ministry of Enviornment, Lands and Parks' TRIM GIS data. Because ATLAS and FORECAST only projected data for the timber providing forest lands, it was also necessary to plug data holes on the landscape visualizations. This was accomplished using BC Ministry of Forests' Forest Cover GIS information about non-timber producing landscape elements such as rock, riparian areas, alpine areas, etc. The result of this process was a series of landscape visualization images from different viewpoints at decadal intervals over the time period modeled by ATLAS and FORECAST. Examples of the landscape visualization images generated for inclusion in the interface can be seen in Figure 3.1. The landscape visualization images generated for the interface are considered semi-realistic, in that they do not reflect the highest level of realism achievable using current technology. The normal technique used to generate landscape visualization images for perception research is to choose a small number of key viewpoints within a landscape, take base photography from those viewpoints, and then to match the generated landscape visualization images of the viewpoints to the representative photographs (Lewis and Sheppard, 2004; Salter, 2000). For the Arrow IFPA project, the modeled landscape area covered the entire Lemon landscape Unit, which is approximately 48,000 ha in size. The  82 project also called for the ability to view the landscape from any one of a number of viewpoints. The size of the visualized area, as well as the inability to focus on one, or a small number of viewpoints, meant that some of the techniques necessary to generate highly realistic images could not be performed. These techniques include photomontage of foreground elements, adjustment of lighting conditions, calibration of cloud and wave models, etc. As a result, the landscape visualization model took a more "averaged" approach, generalizing lighting conditions, avoiding the use of clouds, etc., to ensure that usable images could be generated from a number of viewpoints. The result was landscape visualization images that possessed a moderate level of realism (see Figure 3.1). A number of changes were made to the landscape visualizations originally created for the Arrow IFPA project in order to incorporate them into the interface. The first major change came about as a result of a review of the initial landscape visualizations generated for the Arrow IFPA project. The landscape visualization images represented forest densities which appeared much higher than those seen in photographs of the same locations. The landscape visualizations used density measurements generated from the FORECAST model, and, as such, should have closely reflected the densities seen on the ground. An investigation into the placement of tree images in World Construction Set revealed that it overemphasized the visual density of forest stands as a result of resolution scaling of the tree images. Figure 3.2 demonstrates the effect of scaling on tree images. The white squares on Figures 3.2 a) and 3.2 b) show the difference in detail levels between the higher (Figure 3.2 a) and lower (Figure 3.2 b) resolution images, and the averaging of pixel colours in the lower resolution image (Figure 3.2 b). World Construction Set represents black pixels (RGB value of 0,0,0) as transparent. When preparing a tree image for use in World Construction Set, it is placed on a black background, so that when it is rendered only the tree itself is visible. As the size of the tree images decreases with distance in the landscape visualization images, the pixel count per tree decreases. Pixels are then averaged together (i.e. a black pixel next to a green pixel becomes a dark green pixel). This averaging reduces the number of truly black pixels, and therefore reduces the amount of transparency surrounding the tree image. The reduction in transparency means that the viewer cannot see through the lower resolution trees and they "block up" to create the appearance of higher density. In order to compensate for this effect, it is necessary to reduce the actual density of trees being portrayed in the landscape visualization images by some factor. Through trial and error, the comparison of landscape visualization images to representative photographs, and consultation with landscape professionals, a 50% reduction factor was used on the images generated for the interface. As a result of the density change, the images generated for inclusion in the interface had a higher realism level then those generated for the initial Arrow IFPA project. It was also necessary to alter the aspect ratio of the landscape visualization images rendered for the Arrow IFPA project. The original images were rendered at 1024 pixels by 384 pixels, or an aspect  84  a) High Resolution  b) Low Resolution  Figure 3.2: Images demonstrating the effects of resolution on tree images. ratio of 8:3. This aspect ratio provided a panoramic view of the landscape, providing more contextual information to the viewer. The 8:3 aspect ratio, however, would not have worked within the interface, because it would have either taken up too much horizontal space, limiting the available space for other interface elements, or it would have resulted in landscape visualization images that were too small in the vertical dimension to provide sufficient detail. The decision was made to re-render the landscape visualization images at 700 pixels by 418 pixels, or an aspect ratio of 5:3. This size was arrived at as a trade-off between the need to display the landscape visualization images at a sufficient resolution to display their high level of detail, and the need to provide enough space for the other interface elements. The 5:3 aspect ratio provides a slightly wider field of view than the 4:3 ratio of standard digital photographs, and the 3:2 ratio of 35mm slide film photographs. Evaluation of the efficacy of this aspect ratio was beyond the scope of this thesis, however, because the purpose of the landscape visualizations was primarily to communicate cognitive information and not response equivalency for perception studies, it is unlikely that the aspect ratio had a significant impact on the results. One of the key decisions for the landscape visualization element of the interface was how many, and which, viewpoints to use. Because the chosen information set for the interface was ecological, and ecological processes occur over large land areas, it was necessary to select viewpoints that represented the entire landscape. Four viewpoints were selected for representation in the interface based upon their ability to represent a broad cross-section of the landscape unit. Those viewpoints were: 1. A perspective front country view of the Ringrose Face - This viewpoint was representative of how most viewers would see the landscape. It is a ground level perspective view (elevation - 604 metres, horizontal angle of view - 39°) looking east from a location near the city of Slocan. The viewpoint is representative of a typical "worst case" view for people traveling on public roads in the Slocan Valley through the Lemon Landscape Unit. Figure 3.1 a) shows the Ringrose Face perspective viewpoint.  85 2. A high oblique view along Lemon Creek - This viewpoint is a high oblique, or aerial view (elevation - 2680 metres, horizontal angle of view - 64°) looking east down the Lemon Creek drainage. The Lemon Creek drainage is the main drainage in the southern portion of the landscape unit. Significant logging activities occur in the Lemon Creek drainage, and it is an access route for recreation in the area. The high oblique view allows the viewer to see the entire drainage, but still provides a perspective view that allows the user to get a sense of the terrain. Figure 3.1 b) shows the Lemon Creek aerial viewpoint. 3. A high oblique view along Enterprise Creek - This viewpoint, like the Lemon Creek viewpoint, is an aerial view (elevation - 3450 metres, horizontal angle of view - 45°) looking east along the Enterprise Creek drainage. Enterprise Creek is a main drainage in the northern portion of the Lemon Landscape Unit, with significant logging activity, and an access route to Kokanee Glacier Provincial Park. Similar to the Lemon Creek viewpoint, the Enterprise Creek aerial viewpoint provides an overview of management activity in one of the landscape units main watersheds. Figure 3.1 c) shows the Enterprise Creek aerial viewpoint. 4. An Overhead view of the entire landscape unit - This viewpoint gives a broad overview of the entire landscape unit (elevation - 70 km, horizontal angle of view - 35°). The plan view looks straight down from a high elevation onto a realistic visualization of the landscape unit. Initially, thematic information was to be included on the map in the interface. However, limitations to the maps size (see discussion later in this chapter), and a desire to centralize the thematic information in one location led to the decision to create a plan view visualization. The location of the other three viewpoints were indicated on the plan view (see Figure 3.1 d) to help subjects locate those viewpoints on the landscape, and to tie the plan view visualization to the map, which also showed the viewpoint locations. The need to alter the density of the trees in the landscape visualization images, to change the images resolution, to c