ABSTRACT
With the development of the computer and related technologies has come revolutionary changes in how we collect, process, analyze, and view data about the world around us. The potential use of “virtual reality”, or computer generated simulated environments, for the display of geographic data raises issues regarding the appropriate representation of spatial data in support of spatial learning and decision making. Understanding the relationship between the methods of presenting spatial data and the acquisition and representation of that data in human spatial knowledge structures is at the core of cognitive studies in cartography and GIS. This paper reports on research investigating the effect of iconicity/abstractness of areal symbols on the acquisition of geometric and thematic information during fly-through navigation of virtual environments.
INTRODUCTION
Cartographers have historically concerned themselves with the production of paper maps; however, Robinson’s (1952) The look of maps: An examination of cartographic design, signaled the beginning of the era of scientific, cartographic research. From the 1950’s through the early 1980’s this research largely employed a psychophysical research paradigm which was of little practical utility in either the design of maps, or the understanding of the map-use process (Petchenick, 1983). Since that time, however, the focus of cartographic research has shifted from a goal of finding the rules for creating perfect maps employing a communication model, to understanding the relationships between the cognitive processes of map users and the physical properties of mapped information using an information processing approach.
The recent advances in digital computing technologies, including both the hardware and software domains, have presented cartographers with opportunities and challenges including how “best” to represent spatial information for use and analysis. “Until recently… the rendering of GIS results primarily has been restricted to the same set of display techniques used in manual cartography” (Berry, Buckley, and Ulbricht, 1998, p. 47). These techniques, the result of hundreds of years of cartographic experience and decades of cartographic research, have proven appropriate for conventional map products but have yet to demonstrate their efficacy in simulated environments and in modes of acquisition that include movement of the user’s point-of-view.
Application and development of these technologies falls under the general heading of “visualization,” “scientific visualization,” or “geographic visualization.” Visvalingam (1994, p.19) describes visualization as “primarily a mental process which serves a variety of purposes, including visual analysis. Visual analysis refers to the use of visualization as a distinct method of inquiry for provoking insight and for concept refinement.” A slightly different description of visualization comes from Haber and McNabb (1990, p. 75) who refer to visualization as a process composed of “transformations that convert raw simulation data into a displayable image. The goal of the transformation is to convert information into a format amenable to understanding by the human perceptual system.” McCormick et al. (1987, p. 3) further defines the discipline of scientific visualization as a “discipline concerned with developing the tools, techniques and systems for computer-assisted visualization. It studies those mechanisms in humans and computers which allow them in concert to perceive, use and communicate visual information.”
While a great deal of interest exists in the potential utility of visualization technologies and techniques in the exploration of aspatial phenomena and concepts, nowhere is the interest greater than in the spatial domain; specifically, in the display of geographic data. MacEachren (1995, p. 361) suggests that, “GVIS (geographic visualization) has the potential to help us cope with the flood of information that technology increasingly provides, and to stimulate insightful discovery – but only if those creating the GVIS environments understand how visual-cognitive representations interface with the sign systems of cartographic representation. At this point, there are few certain answers about this interface in a dynamic interactive environment, or therefore how to design GVIS environments.”
This paper reports on research examining the effects of the level of abstraction of the representation of surface features on the integration of spatial knowledge acquired in fly-through navigation of large-scale virtual environments. Specifically, this paper addresses questions regarding the effects of the level of iconicity/abstractness of the symbols used to represent landcover classes (“draped” over a terrain model) on the integration of spatial information including: directional/angular information, distance/length information, and thematic (land use/land cover) information.
BACKGROUND
Spatial Knowledge Acquisition
Research investigating the use of various media intended to facilitate the
sequential acquisition of spatial information has shown mixed results in the
acquisition of landmark and route knowledge. In research utilizing isolated
photographs of an environment as the spatial data source (Hershberger, 1975;
Winkel and Sasanoff, 1966), the photographs were found to be inadequate for the
accurate acquisition of route sequences or the spatial relationships among
landmarks. Likewise, research using sequences of photographs taken at frequent
intervals along a route (Allen, Kirasik, Siegel, and Herman, 1980; Allen, 1979;
Allen, Siegel, and Rosinski, 1978) were found to “… not provide sufficient
information to support the learning of relative positions of landmarks” (Goldin
and Thorndyke, 1982, p. 458). Other studies, utilizing films of both natural and
simulated environments, and computer simulations of both large and small scale
space, however, support the notion that simulated environments can be used for
the accurate acquisition of spatial knowledge (Craik, 1977; Craik, 1978; Ciccone,
Landee, and Weltman, 1978; Cohen, 1980; Goldin and Thorndyke, 1982). Much as
Craik (1977, 1978) compared the effect of media and “environment” on the
acquisition of spatial knowledge (color vs. black and white film, and the
navigation of real vs. “model” environments), this research has focused on how
subjects’ performance in certain tasks is affected by the type of area symbols
used as a thematic “drape” during fly-through navigation of a virtual
environment. Unlike Craik (1977, 1978) however, no comparison was made between
the quantity or quality of spatial knowledge acquired from different sources
(e.g., simulated environments vs. maps, “real-world” navigation, written
descriptions, films, etc.), leaving questions regarding the appropriate use of
these technologies open for future research.
While recognizing the important role maps play in spatial learning in a
variety of task domains, it is nonetheless acknowledged that the mode of
acquisition and spatial knowledge acquired from maps is different in character
and utility from knowledge acquired from other sources. Thorndyke and Hayes-Roth
(1982, p. 560) for instance, in a comparison of spatial knowledge acquired from
maps and navigation suggest that:
"From a map, people acquire survey knowledge encoding spatial relations. This
knowledge resides in memory in images that can be scanned and measured like a
physical map. From navigation people acquire procedural knowledge of the routes
connecting diverse locations…With moderate exposure, map learning is superior
for judgements of relative location and straight-line distances among objects.
Learning from navigation is superior for orienting oneself with respect to
unseen objects and estimating route distances. With extensive exposure, the
performance superiority of maps over navigation vanishes."
Sholl, in a comparison of maps and natural environments as sources of spatial
knowledge, suggests reasons for these differences.
"While both the environment and maps provide information about interrelations
among elements comprising space, they differ in ways attributable to the
symbolic function served by maps, including but not limited to, differences in
scale, perspective, boundedness, and topological relation to the viewer. Maps
are small in scale, providing an aerial perspective of the layout of the
environment, and have distinct boundaries…In contrast, environments are large in
scale, are viewed from the ground, and are essentially unbounded (although they
do contain natural and artificial barriers that can obstruct travel)." (Sholl,
1995, p. 177)
Simulated Environments
The interest in using simulated environments for spatial, and procedural
learning purposes, is probably most evident in the field of flight simulation.
Kleiss (1995), for example, employed multidimensional scaling techniques to
identify the relative importance of the graphical properties used in simulated
low-altitude flight to pilot’s simulator control capability. Rather than trying
to identify relevant cognitive factors, this study focused on identifying the
elements of the environment that could best be manipulated to increase user
performance. Using a multi-dimensional scaling technique Kleiss found that in a
dynamic presentation condition terrain vertical development was the most
important scene property necessary for controlling altitude in flight
simulators. While some research (Engle, 1980; Kleiss and Hubbard, 1993; Martin
and Rinalducci, 1983) indicates that performance in flight simulators improves
with increases in object density, Kleiss (1995) found that irregular clustering
of objects and variations in the size of objects in clusters resulted in optimum
performance. These findings, consistent with the results of Barfield, Rosenberg,
and Kraft (1989), indicate that “the important property of terrain shape relates
to smaller-scale terrain elements rather than large vertical obstructions.”
Other studies of human perception/cognition or performance in simulated environments investigate: the interaction between visual target detection and the presence or conspicuity of borders in the simulation (Monk, 1981); the alignment effect when navigating in a virtual environment with map-acquired knowledge (May, Peruch, and Savoyant, 1995); the task and information variables that contribute to human performance in low altitude flight (Flach and Warren, 1995); the importance of textural features to image classification tasks (Haralick, Shanmugam, and Dinstein, 1973); the use of virtual environments for learning ‘complex scientific concepts’ (Dede et al., 1997); and the effects of level of immersion on the accuracy of spatial knowledge acquired in a simulated architectural environment (Henry and Furness, 1993).
METHODS
In this study, land use/land cover information was “draped” over a 3-dimensional terrain relief model using four different methods: 1) geospecific texture, defined by Suter and Nuesch (1995, p. 91) as “icons of a part of the earth’s surface, i.e. satellite images and aerial photographs…recorded directly by a sensor.” 2) geotypical texture (Suter and Nuesch, 1995, p. 91), “small photo patches representative of some type of landcover” 3) cartographic-iconic symbols as textures (e.g. tree shaped symbology representing the class “forest”), and 4) cartographic-abstract symbols as texture(e.g. arbitrary symbology such as dot patterns or colors to represent different categories). These four methods represent distinctly different levels of abstraction ranging from what Robinson and Petchenick (1976, p. 61) call a mimetic image symbol (one in which the map mark “retain[s] some graphic characteristic that can be visually or conceptually related to the referent”) to arbitrary symbology.
As these four methods represent different levels of abstraction, they also offer, or afford, different visual cues, which are used by subjects to perform the various tasks required in this research. Clearly, along the continuum from the “mimetic” (Robinson and Petchenick, 1976, p. 61) air-photo, to the arbitrary, and symbolic use of contrasting hues to identify “type”, the quantity, meaning, and utility of many of the “to whom it may concern” messages (Golledge and Stimson, 1987, p 1) are altered. Just how this alteration occurs, and what effect this has on the utility of the information being translated into knowledge by individuals will surely be a matter of great interest and debate as visualization using simulated environments becomes more widespread.
This paper reports on research examining the effects of the level of
abstraction of the representation of surface features on the integration of
spatial knowledge acquired in fly-through navigation of large-scale virtual
environments. Specifically, this paper addresses the following questions. Does
the level of iconicity/abstractness, or the symbols used as thematic “drapes” in
fly-through navigation of simulated environments affect:
1) Subjects’ ability to accurately recall relative directional changes while
involved in a route learning task?
2) Subjects’ ability to accurately estimate distance/path-length in standard
units while involved in a route learning task?
3) The quantity and accuracy of land use information recalled by subjects while
involved in a variety of spatial learning tasks.
Tasks performed by subjects included: distance estimation (both relative and based on a standard scale), angular estimates (recall of the visible intersection between “legs” of the path) as described graphically on a “compass rose”, and memory of the land use of polygons along each leg of the fly-through.
Subjects
Twenty-four subjects, 16 male and 8 female, were recruited for participation in
this research. Twenty were graduate students from the Department of Geography,
San Diego State University, two were undergraduates from the same department,
and two were college graduates of other institutions with extensive experience
with maps, as well as, experience in air-photo interpretation. To participate in
this research, subjects were required to have a minimum of 60 days of experience
using: maps, remotely sensed data or satellite imagery.
Subjects were paid $30.00 to participate in the research, which consisted of four, forty to forty-five minute “blocks.” Between each “block” subjects allowed to rest for as long as necessary before proceeding with the experiment.
Materials
Selection of the study area for this research was based on the following
criteria:
1) The study area must contain six different land use classes at no more than
the Anderson II level of classification (a level of classification readily
available from the air-photos used in this research),
2) The study area should be of sufficient size to allow a five minute
fly-through at a simulation speed of at least 100 miles per hour (at scale),
3) The study area should have sufficient topographic relief to permit the use of
topographic features as landmarks or anchors,
4) The study area should be of sufficient topographic relief and be of
sufficient size to force a sequential acquisition of spatial knowledge. Subjects
must not be able to see more than one change in route direction (choice point)
ahead of any other choice point,
5) The study area should not be familiar to subjects involved in the research.
Based on these criteria, an area was selected from the Tijuana River Watershed Database, a joint database development project of the Department of Geography, San Diego State University and El Colegio de la Frontera Norte, Baja California. The study area covers approximately 72 square kilometers with dimensions of 12,000 x 6,000 meters, containing a small portion of the urban environment of southeastern Tijuana, Mexico with the remainder being made up of the rural and natural landscape south of the city.
The terrain of the study area includes two major valleys running in a predominantly east/northeasterly direction flanked on either side by hills and mountainous terrain with relative relief of several thousand feet. The land use of the area is a mix of six types, divided into 86 polygons (categorized on a modified Anderson Level 1 system), ranging from approximately 4% to 44% coverage (Table 1) (Figure 1).
Table 1. Land use categories, polygons, and coverage in the study
area.
Figure 1. Land use of the study area.
Four thematic “drapes,” a terrain relief model, and a single path was used in this research. To allow comparison between subject’s responses for each of the four rendering methods, it was necessary that each rendering method (test block) afford similar quantities and configurations of spatial information without allowing subjects to transfer learned information between test blocks. To achieve this end, the “drapes” for two rendering methods (with the path in a single spatial orientation) and the terrain model were mirrored on an east-west axis (Figures 2,3,4, and 5). Additionally, direction of travel along the path was alternated resulting in a unique combination of travel direction and land use/terrain orientation for each of the four rendering methods (Figures 2,3,4, and 5).
Figure 2. Geospecific rendering method, representing the “iconic” end
of the continuum. Geospecific texture, defined by Suter and Nuesch (1995, p. 91)
as “icons of a part of the earth’s surface, i.e. satellite images and aerial
photographs…recorded directly by a sensor.”
Figure 3. Geotypical rendering method, less “iconic” than geospecific.
Geotypical texture (Suter and Nuesch, 1995, p. 91), “small photo patches
representative of some type of landcover.” Contrasting hues were also added to
distinguish categories.
Figure 4. Cartographic-iconic symbols as textures utilizing
contrasting hues and iconic patterns to represent land use categories (e.g.,
green hue with tree shaped symbols arranged in a random configuration
representing the class “natural”).
Figure 5. Cartographic-arbitrary symbols used in this “drape”
represent the lower end of the iconic/arbitrary continuum. While the assignment
of hues to represent specific land use categories was not totally arbitrary
(i.e., green represented “natural” and “agriculture,” “water” was blue, etc.)
the hue itself presents no pictoral (iconic) association with the category.
Procedures
Prior to the data collection phase of this research subjects were given a brief
introduction to the general purpose of the study, the tasks that they would be
asked to perform, and the sequence of the data collection procedures. Prior to
each test “block,” and each fly-through (within each “block”) subjects were
reminded of the foci and tasks to be performed. Each subject participated in a
series of four test “blocks,” each utilizing a single rendering method as
described previously. While the larger data collection methodology included
three fly-throughs in each test block, the questions addressed in this paper are
concerned with the first two fly-throughs, the first, in which angular/distance
estimates were collected (Figure 6), the second, conducted immediate after
completion of the first, in which memory for thematic information were collected
(Figure 7).
Figure 6. Fly-through 1, conditions and tasks.
Figure 7. Fly-through 2, conditions and tasks.
Figure 8. Report form 1: Length and Angle (actual size 8.5” x 11”).
Figure 9. Report form 2: Land use/landform (actual size 8.5” x 11”).
To avoid possible bias due to “block” order, a counterbalanced approach was employed. Thus, utilizing 24 subjects, all possible “block” sequences were represented.
Subjects’ responses regarding angular changes in bearing were determined through direct measurement of their graphic depiction of the recalled angle, using a protractor, and were recorded to the nearest degree. Length/distances estimates were converted from decimal miles and kilometers to meters. Since subject’s length estimates are subject to “personal scale transformations,” a method (Figure 6) was employed to normalize values, thus, allowing within and between subject analyses (Montello, 2001). Subjects’ “land use/landform (topo) forms” were first “graded” for correct identification of land use polygons, then counts were made of the total number of land use polygons identified, and the number of those polygons correctly identified.
Figure 10. Calculation of “scale free” length estimates.
e = “scale free” error of estimate for the test “block”
X = actual length of each “leg” (route traveled)
Y = verbal estimate of length of each “leg” (in meters)
Data collection for this research was conducted in the Visualization Lab at the San Diego Supercomputer Center, located on the campus of the University of California San Diego, using a Silicon Graphics “Octane” computer. Subjects viewed the fly-throughs on a high resolution, wide aspect (12” x 18”) color monitor, offering a “window on the world” VR experience (Figure 7). The simulation was created from a C++ application written (APPENDIX II) for the World Tool Kit release 9 (Engineering Associates Inc., 1999), a virtual reality development package. Based on the results of the pilot study, the parameters used in subject’s fly-throughs were set as follows: the elevation was “fixed” at 300 meters (at simulation scale), with an apparent speed of approximately 120 miles per hour, and a 40o viewing angle (orthogonal view would be 90o). For the purposes of this research, subject’s control during data collection consisted exclusively of directional control (using left and right arrow keys).
Figure 11. “Window on the world” view, geospecific rendering method.
Analytical Design
The statistical approach used in the analysis was a multivariate approach to
repeated measures, included in SPSS 10 for Windows (SPSS inc., 2001) as a
“General Linear Model.” This research employed a 1x4 factorial design, with 24
repeated measures. Employing this approach, the within subject effects were
tested at a .95 level of confidence, using the “Pillai’s Trace” statistic.
Post-hoc tests were used to identify, in a pair-wise fashion, between which of
the four cases of independent variables significant differences occurred. The
same approach was applied to the between group variable, “order” of the test
“blocks”, for each dependent variable.
RESULTS
Angular Estimates
As a means of answering the first research question, i.e. does the level of
iconicity, or the symbols used as thematic “drapes” in fly-through navigation of
simulated environments affect subject’s ability to accurately recall relative
directional changes while involved in a route learning task, total error of
estimates for each treatment were analyzed. These estimates, measured to the
nearest degree from graphic representations of the recalled angle, represent the
total absolute-error per “block” (treatment), of the recalled angles collected
approximately ten seconds after viewing each angle (four angles total per
“block”) (Table 2).
In the analysis of angular recall by treatment, a p value of .80 was calculated, using “Pillai’s Trace” test of significance. At the 5% rejection level, therefore, significance was not found, thus, retaining the null hypothesis (that there was no difference in the total angular error based on treatment).
Table 2. Descriptive statistics: total error in angular estimates, by
treatment.
Since the order in which the four “blocks” (each block tested a different treatment) were administered was counterbalanced among subjects there was no need to account for the potential effect of order of treatment on the total error of angular estimates. While the order of treatments was accounted for in the research design, the order of testing was not. Therefore, an analysis was conducted to determine if the order of testing had a significant effect on subject’s total error of angular estimates (e.g., did subject’s accuracy improve due to familiarity with the test, etc. or did accuracy decline, due to fatigue, etc?) (Table 3). As in the previous analysis, a multivariate approach to repeated measures was applied to the angular error data. This time, however, the independent variable being “block order” rather than treatment. In the analysis of angular recall by order, a p value of .38 was calculated. At the 5% rejection level, therefore, significance was not found, thus, retaining the null hypothesis (that there was no difference in the total angular error based on order).
Table 3. Descriptive statistics: total error in angular estimates, by
“block” order.
Since no significant difference was found in subjects’ total error in graphically representing the angles found at “turning points” along the paths, based on the symbolic iconicity/abstractness (treatment), it can be concluded that overall, background textures acted as neither distracters or aids in angular recall in this study. While it might be expected that a regular pattern (streets, etc.) in imagery or cartographic symbols might prove useful in an angular estimation task, the general lack of such “reference features” in the imagery and cartographic symbols directly adjacent to the “choice points” may have prevented any possible benefit.
Length/Distance Estimates
While quantifying the angular error of subjects’ estimates used in the previous
section allowed direct measurement of subjects’ graphic depictions of a recalled
angle, determination of length/distance error was not as straightforward. Since
the instruction given to subjects during testing was to “Estimate the length of
the path [route] you just flew in feet, yards, meters, miles or kilometers”, the
estimate elicited was not of the path that was visible on the ground, rather, it
was an estimate of the route learned as subjects followed the path. The routes
subjects traveled were recorded as a series of coordinates into ASCII files as
subjects flew through the simulated environments in the first fly-through of
each “block”, and were generated into ArcInfo coverages for use in a number of
analyses. For the purposes of this research, error in the estimated length for
each treatment was established as the sum of the absolute difference between
subject’s length estimate for each leg and the length of the actual route
traveled for each leg.
During the analysis of subject’s data it was discovered that the application program prematurely terminated the recording of coordinates in the ASCII files. The resulting ArcInfo coverages confirmed the problem, with each path missing approximately ½ of the final “leg.” While only the final segments of subject’s routes was affected, to maintain geometric equivalence between each of the four “blocks” (with direction of travel reversed for two “blocks”) both the first and fifth “leg” had to be omitted from the analyses.
The analysis of “scale free” length/distance estimates resulted in a p value of .38. At the 5% rejection level, therefore, significance was not found, thus, retaining the null hypothesis (that there was no difference in the “scale free” estimated length of the “legs” of the routes based on treatment) (Table 4).
Table 4. Descriptive statistics: “scale free” length estimate error,
by treatment.
Results of the analysis of the “scale free” error in length/distance estimates by “block” order, however, showed an overall increase in accuracy with each test “block” (Table 5). With a calculated p value of .03, employing “Pillai’s Trace” test of significance at the .05 rejection level, we must reject the null hypothesis, recognizing that overall, subjects were able to more accurately estimate lengths/distances with repeated exposure to the test procedures or practice. Further analysis, employing Bonferroni’s pairwise comparisons, at the 5% rejection level, shows the only significant difference is between the first and fourth “block” (p = .02). While the results in Table 7 show consistently increasing accuracy through the four “blocks”, the high error in the first “block” suggests that the increases in accuracy may, to a large degree, be the result of familiarity with the test procedures, and only minimally to increased ability over time.
Table 5. Descriptive statistics: “scale free” length estimate error,
by order.
Land use/Land cover Features
Unlike the estimation tasks involved in the “geometric analysis”, involving the
calculation of cognitive distance and angular perception and recall, the
“landmark analysis” investigated the volume and composition of subject’s spatial
knowledge, insofar as they related to the general categories of “land use
features” and “topographic features.”
In addition to the differences in tasks subjects performed in the first two fly-throughs it must be recognized that there existed a substantial difference between the tasks performed in the geospecific method and all other methods. The geospecific method of rendering (use of an air-photo) prevented the use of prototypical or homogeneous symbology to represent specific classifications. Even with the inclusion of polygon boundaries delineating the various land use polygons, subjects were confronted with the task of performing air-photo interpretation, into predetermined categories, literally “on-the-fly.” Without any further explanation of the potential problems associated with this type of task, it is obviously a different task than performed in the other three “blocks.”
In each of the other three test “blocks” the task was simply recognition of the six symbols and the memorization of the classification of particular polygons, for later recall. While it is true that the topographic/landform feature identification task was also an air-photo interpretation task it should be recognized that this task was held constant in each of the four test “blocks.” Additionally, subjects were not asked to “fit” the landforms into predetermined categories, as was the case with land use identification task for the geospecific test “block.”
Again, supporting the notion of the increased difficulty (or subject’s lack of confidence in their classifications) of the landmark task in the geospecific “block”, the total number of land use features recalled in the geospecific “block” was significantly lower than in the other three methods. With a mean value of 23.5 land use features recalled in the geospecific “block” vs. 31.3, 31.5, and 33.5 land use features recalled for the geotypical, cartographic-iconic, and cartographic-arbitrary test “blocks” (differences of over 25% between the geospecific method and each of the other methods) (Table 6). With a calculated p value of less than .01, employing “Pillai’s Trace” test of significance, at the .05 rejection level, the null hypothesis was rejected, thus indicating a potential link between treatment and the total number of land use features recalled. Further analysis, employing Bonferroni’s pairwise comparisons, at the 5% rejection level, indicated that the null hypothesis was rejected for the geospecific vs. each of the other methods, i.e. the total number of land use features recalled for the geospecific “block” was significantly different from all other treatments. The null hypothesis was, however, retained in pairwise comparisons between all other methods, indicating no significant effect of symbolic iconicity on the number of land use features recalled for the geotypical, cartographic-iconic, and cartographic-arbitrary test “blocks” (Table 7). Additionally, analysis of test “block” order on the total number of land use features identified, resulted in a p of .74, thus retaining the null hypothesis (that “block” order has no effect on the total number of land use features recalled).
Table 6. Descriptive statistics: total land use/land cover features
recalled, by treatment.
Table 7. Bonferroni pairwise comparisons with significance at the .05
level. From the repeated measures test of total land use features recalled, by
treatment.
Perhaps more important than the total number of landmark features recalled (reported), are the number of correctly identified land use/landcover features identified, and the proportion of land use features that were correctly identified. For the purposes of this research, correctly identified land use features were polygons on the test form (Figure 5) labeled with the correct land use “type.” Since subjects were tested to ensure their ability to remember the six land use classes used in this research prior to the beginning of each test “block”, only polygons containing these classes were counted as correct (e.g., “forest” or “scrub” is not acceptable for the class “natural”). An examination of the descriptive statistics associated with subject’s accuracy in recalling land use features from fly-throughs reaffirms earlier findings regarding the relative difficulty (or uncertainty) when performing land use identification and recall tasks employing the geospecific “drapes.” Previously, it was noted that the total number of land use features recalled in the geospecific “block” was significantly lower (approximately 25%) than each of the other test “blocks”; this analysis indicates that of the reduced quantity recalled, an average of less than 57% of those features are accurately recalled (Table8). A statistical analysis of both the total number of correctly recalled land use polygons and the percent of correctly identified land use polygons [(total land use polygons recalled/correct)*100], employing “Pillai’s Trace” test of significance, at the .05 rejection level, resulted in calculated p values of less than .01 for each. Therefore, the null hypothesis was rejected in both cases, confirming the significant effect of treatment on both the total number of correct land use polygons and the percentage of correctly recalled land use polygons. Further analysis of the total number of correct land use polygons identified by subjects, employing Bonferroni’s pairwise comparisons, at the 5% rejection level, showed a significant main effect in all but the geotypical and cartographic-iconic comparison (Table 9). Additionally, Table 16 shows that the percent land use features correctly identified in the geospecific test “block” was significantly different (lower), than all other methods (see Table 10); the geotypical “block” was significantly different than the geospecific and cartographic-arbitrary (higher than geospecific and lower than cartographic-arbitrary (see Table 10); the cartographic-iconic “block” significantly different (higher) than the geospecific “block”, and; the cartographic-arbitrary “block” higher than both “image based” “blocks.” With calculated p values of .59 for the total number of correct land use polygons recalled, and .44 for percent correctly recalled land use polygons, “block” order was found to have no significant effect on results.
Table 8. Descriptive statistics: total number of correct land use
features and percent land use features correctly identified, by treatment.
Table 9. Bonferroni pairwise comparisons with significance at the .05
level. From repeated measures test of the total land use features correctly
recalled, by treatment.
Table 10. Bonferroni pairwise comparisons of repeated measures test of
the % correct of total land use features recalled, by treatment, with
significance at the .05 level.
DISCUSSION
In the geometric analyses included in this study, subject’s angular estimates
and length/distance estimates were examined to determine whether the level of
iconicity, or the symbols used as thematic “drapes” affected:
1) Subjects’ ability to accurately recall relative directional changes while
involved in a route learning task, or;
2) Subjects’ ability to accurately estimate distance/path-length in standard
units while involved in a route learning task (research questions 1 and 2).
Employing a multivariate approach to repeated measures analysis at the .05 rejection level, no significant effect of treatment was found for either the angular recall task or the distance/path length estimation task. The lack of significant difference (between treatments) in the angular recall task, which measured subject’s ability to visually perceive a single angle, store the image of that angle in memory, and subsequently recall and reproduce the angle as a graphic representation, indicates that overall, background textures acted as neither distracters or aids in angular recall in this study. While the statistical analysis of the “scale free” errors of subject’s length estimates resulted in a calculated p value of .07, at the .05 rejection level, no statistically significant effect was found. Thus, the level of iconicity did not affect subject’s ability to accurately estimate distance/path-length in standard units while involved in a route learning task. In the analysis of the effect of “block” order, however, it was found that order had a significant (p = .03) effect on accuracy of estimates. Descriptive statistics (Table 5) show the largest increase in accuracy occurring in the second block, with smaller, but consistently improving accuracy in subsequent “blocks.” These results seem to indicate that subjects accuracy resulted from familiarity with the testing (refining the test taking strategy after the first “block”), and that with practice, accuracy in distance estimation tasks can increase.
The analyses conducted regarding land use/land cover features and
topographic/landform features focused on how the level of iconicity, or the
symbols used as thematic “drapes” affected the quantity or accuracy of
topographic/landform information and land use information recalled (research
question 3). Overall, treatment was shown to have a significant effect on:
1) the total number of land use/land cover polygons recalled,
2) the number, and proportion of land use/land cover polygons correctly
identified.
Moreover, the geospecific treatment was found to result in subjects recalling
significantly fewer landmarks overall (total of land use and topographic
features), fewer land use polygons, a lower percentage of correctly identified
land use polygons to total recalled land use polygons, a reduced number of
topographic features (compared to the cartographic-iconic method), and a lower
proportion of topographic features to total features recalled.
In addition to the significant differences in subject’s (landmark) task performance in the geospecific “block” vs. other methods, significant differences were also identified between other treatments, in several of the tests. The use of the cartographic-abstract symbology resulted in a significantly higher number of correctly identified land use polygons than in the other three methods, and a markedly higher proportion of correctly identified polygons than in the other three methods. Table 8 shows that the total number of correctly identified land use polygons increases as the level of iconicity decreases, as does the percentage of correctly identified land use polygons. Whether these results indicate a decrease in ambiguity of the symbology (e.g., differences in relationships between the sign-vehicles (symbols) and their interpretants, requiring a more complex “translation” schemata as might be suggested in a semiotic analysis), the increased efficiency and accuracy of perceptions (searches) when fewer visual “dimensions” are used (as explanations derived from Feature Integration Theory or Guided Search Theory might propose), or simply a recognition of the dominance of color in the perception and cognition of visual information, the practical implications are the same. In fly-through applications using simulated environments, where the task domain includes the recognition and recall of nominal scale (areal) data, the use of hue to establish “type” is the most effective method.
CONCLUSIONS
In this research, four methods of rendering thematic data, occupying four different positions along the arbitrary-iconic continuum were tested. In the various tests used to establish the acquisition and recall of spatial data from these environments, no pattern of significant differences was found that could be considered related to the ordinal continuum. Rather, the results show that particular rendering techniques are more effective than others for performing specific tests of spatial knowledge content.
While highly iconic displays, as in the case of photographic drapes, may prove visually compelling, and may lead to subjects becoming “engaged” in the fly-through process, for example, this research has shown that for tasks requiring the identification of thematic classifications the geospecific rendering method is the least effective. For angular estimates, length/distance estimates the geospecific rendering method is not significantly different than any other method tested. For the identification and recollection of thematic data, cartographic techniques, especially the use of contrasting hues to distinguish nominal scale categories is the most effective.
There are a number of issues that should receive future attention. First, with the seemingly exponential increase in the cost of achieving more realistic simulations, serious consideration should be given to the benefit/cost relationship between spatial data display techniques and technologies. To this end, there is a need to establish which spatial learning tasks benefit from the properties of virtual reality (VR) display techniques and which of these tasks can be more economically performed using other tools such as maps, verbal descriptions, or direct environmental experience.
Another area of needed work, is ongoing research into the impact of “cutting-edge” VR capabilities and technologies on the effectiveness of spatial learning and task performance. The efficacy of “state-of-the-art” hardware including the latest rendering engines, motion-tracked head mounted display units, holographic display devices, and haptic and auditory input and output devices should be investigated. Questions including, how motion-coupled head mounted display devices that allow persons to “look around” much as they would in a natural environment, affect performance, what effect display resolution has on spatial knowledge acquisition, and how peripheral vision affects spatial learning in simulated environments, require investigation.
Research into the impact of data scale, resolution, and precision on spatial knowledge acquisition in simulated environments is needed. As the fly-through mode of data acquisition was examined in this research, it must be recognized that the data needed for other scales of analysis, and modes of acquisition, will differ. While aerial-photography might be adequate for fly-through simulations, what types of data are required for walk-through, or drive-through simulations? Additionally, with the cost of implementing highly detailed simulations, an understanding of the requirements for various task domains and modes of acquisition is essential. While clearly, the highest expression of simulated environments, the artificial imagery (AI) and special effects used in motion pictures is often justified (if the film makes money), at what cost, and under what circumstances should simulated environments be the tool of choice for cartographers?
With the certainty of increases in data quantity, resolution, and hopefully precision, in the future, and the possibility of new types of data that will enhance our abilities to discover and understand processes and patterns in the world around us come new challenges for representing those data. While the limitations of current technologies control the extent to which we can use simulated/virtual environments as research or learning tools, advances in spectral and hyper-spectral display capabilities, the use of audio and haptic devices as both control and output devices, as well as the potential for the incorporation of smell and taste in simulations, open new vistas to researchers. To take full advantage of what Sutherland (1965) dubbed the “ultimate display”, computer generated environments that “look real, act real, sound real, and feel real”, we must understand more about how humans assimilate information from different media, and from the natural environment. From this knowledge, we can begin to design learning interfaces and environments that will promote insightful learning and discovery. While much can be said for the trial and error approach to development of design principles, it will be through the thoughtful understanding of the process by which people acquire and use spatial knowledge that the promise of visualization using virtual environments will be realized.
ACKNOWLEDGEMENTS
I would like to thank Dr. Richard Wright (S.D.S.U), Dr. Reginald Golledge (U.C.S.B.), Dr. Daniel Montello (U.C.S.B.), and Dr. Elisabeth Nelson (U.N.C.G.) for their guidance and support in the successful completion of this research.
This research was supported in part by the NSF through computing resources provided by the National Partnership for Advanced Computational Infrastructure at the San Diego Supercomputer Center, and funding provided through the San Diego State University Research Council.
BIBLIOGRAPHY
Allen, G.L., 1979. A Developmental Perspective on the Organization of Spatial Knowledge. Ph.D. Thesis, University of Pittsburgh.
Allen, G.L., Kirasik, K.C., Siegel, A.W., and Herman, J.F., 1980. Development Issues in Cognitive Mapping: the Selection and Utilization of Environmental Landmarks. Child Development.
Allen, G.L., Siegel, A.W., and Rosinski, R.R., 1978. The role of Perceptual Context in Structuring Spatial Knowledge. Journal of Educational Psychology: Human Learning and Memory. (4), 617-630.
Barfield, W., Rosenberg, C., and Kraft, C., 1989. The Effects of Visual Cues to Realism and Perceived Impact Point During Final Approach. In Proceedings of the Human Factors Society 33rd Annual Meeting, (pp. 115-119), Santa Monica, CA.: Human Factors and Ergonomics Society.
Berry, J.K.,Buckley, D.J., and Ulbricht, C., 1998. Visualize Realistic Landscapes. GIS World. 11(8): 42-47.
Ciccone, D.S., Landee, B., and Weltman, G., 1978. Use of Computer-generated Movie Maps to Improve Tactical Map Performance. Technical Report PTR-1033-S-78-4, Perceptronics, Inc., Woodland Hills, CA.
Craik, K.H., 1977. Environmental Psychology and Environmental Planning: Collaborative Research at the Berkley Environmental Simulation Laboratory. Paper presented at Conference on the Federally Coordinated Program of Research and Development in Highway Transportation. U.S. Department of Transportation, Ohio State University, Columbus, OH., November.
Craik, K.H., 1978. Impressions of a Place: Effects of Media, Context, and Personality. Paper presented at the American Psychology Association Meetings. Toronto, November.
Dede, C., Salzman, M.C., Loftin, R.B., Sprague, D., 1997. Multisensory Immersion as a Modeling Environment for Learning Complex Scientific Concepts. In N. Roberts, W. Feurzig, and B. Hunter (eds.), Computer Modeling and Simulation in Science Education, Berlin: Springer-Verlag.
Engineerining Associates Inc., 2000. Engineering Animation Incorporated, Sense 8 Product Line, 100 Shore Hwy, Suite 200, Mill Valley, CA.
Engle, R.L., 1980. Use of Low Altitude Visual Cues for Flight Simulation. Unpublished master’s thesis, Arizona State University, Tempe.
Flach, J.M., and Warren, R., 1995. Low-altitude Flight. In Hancock, J. Flach, J.K. Caird, K. Vincente (eds.), Local Application of the Ecological Approach to Human Machine Systems. Hillsdale, N.J.: Lawerence Erlbaum Associates.
Franklin, N., 1996. Language as a Means of Constructing and Conveying Cognitive Maps. In J. Portugali (ed.), The Construction of Cognitive Maps, Netherlands: Academic Publishers.
Franklin, N., Tversky, B., 1990. Searching Imagined Environments, Journal of Experimental Psychology: General. 119: 63-76.
Freundschuh, S.M., Mercer, D.J., 1995. Spatial Cognitive Representations of Story Worlds Acquired from Maps and Narrative, Geographical Systems. 2: 217-233.
Goldin, S.E., and Thorndyke, P.W., 1982. Simulating Navigation for Spatial Knowledge Acquisition, Human Factors. 24(4): 457-471.
Golledge, R.G., and Stimson, R.J., 1987. Analytical Behavioral Geography, London: Croom Helm.
Haber, R.B., and McNabb, D.A., 1990. Visualization Idioms: a Conceptual Model for Scientific Visualization Systems. In Nielson, G.M., Shriver, B., Rosenblum, L.J. (eds), Visualization in Scientific Computing, IEEE Computer Society Press.
Haralick, R.M., Shanmugam, K., and Dinstein, I., 1973. Textural Features for Image Classification. Manuscript prepared under contract NAS5-21822, NASA Goddard Flight Center, and contract DAAK02-70-C-0388, Geographic Sciences Laboratory, U.S. Army Engineer Topographic Laboratory.
Henry, D., Furness, T., 1993. Spatial Perception in Virtual Environments: Evaluating an Architectural Application, Proceedings: IEEE Virtual Reality Annual Symposium, Sept. 18-23, pps. 33-40.
Hershberger, R.G., 1975. The Representation and Evaluation of Environment. In B. Honikman (Ed.) Responding to Social Change. Stroudsburgh, PA.. Dowden Hutchinson & Ross, 109-116.
Kleiss, J.A., 1995. Visual Scene Properties Relevant for Simulating Low-Altitude Flight: A Multidimensional Scaling Approach, Human Factors. 37(4): 711-734.
Kleiss, J.A., and Hubbard, D.C., 1993. Effects of Three Types of Flight Simulator Scene Detail on Detection of Altitude Change, Human Factors. 35(4): 653-671.
MacEachern, 1995. How Maps Work: Representation, Visualization, and Design. New York: The Guilford Press.
Martin, E.L., and Rinalducci, E.J., 1983. Low-level Flight Simulation: Vertical Cues (AFHRL-TR-83-17, AD-A133612). Williams Air Force Base, AZ.: Operations Training Division, Human Resources Laboratory.
May, M., Peruch, P., Savoyant, A., 1995. Navigating in a Virtual Environment With Map-Acquired Knowledge, Ecological Psychology. 7(1): 21-36.
McCormick, B.H., DeFanti, T.A., and Brown, M.D., 1987. Visualization in Scientific Computing. Special issue ACM SIGGRAPH Computer Graphics, 21(6).
Monk, T.H., 1981. The Interaction Between the Edge Effect and Target Conspicuity in Visual Search, Human Factors. 23(5): 615-625.
Montello, D.R., 2001. Personal communications.
Petchenick, B.B., 1983. A Map Maker’s Perspective on Map Design Research. In D.R.F. Taylor (ed.), Graphic Communication and Design in Contemporary Cartography, Chichester: John Wiley and Sons.
Robinson, A.H., 1952. The Look of Maps. Madison: University of Wisconsin Press.
Robinson, A.H., and Petchenik, B.B., 1976. The Nature of Maps. Chicago, University of Chicago Press.
Sholl, M.J., 1995. The Representation and Retrieval of Map and Environmental Knowledge, Geographical Systems. 2: 177-195.
Suter, M., and Nuesch, D., 1995. Automated Generation of Visual Simulation Databases Using Remote Sensing and GIS. Proceedings, Visualization ’95, IEEE Computer Society Press, Atlanta, GA.
Sutherland, I., 1965. The Ultimate Display. Proceedings of the IFIP Congress 65, Vol. 2, W.A. Kalenich (ed.), Washington D.C.: Spartan Books.
Thorndyke, P.W., Hayes-Roth, B., 1982. Differences in Spatial Knowledge Acquired from Maps and Navigation, Cognitive Psychology. 14: 560-589.
Visvalingam, M., 1994. Visualisation in GIS, Cartography, and ViSC. In H.M. Hearnshaw and D.J. Unwin (eds.), Visualization in Geographic Information Systems, Chichester: John Wiley and Sons.
Winkel, G.H., Sasanoff, R., 1966. An Approach to an Objective Analysis of Behaviors in Architectural Space. Seattle, WA.: University of Washington, College of Architecture and Urban Planning, Architecture Development Series, No. 5, August.