IntroductionRecreation and tourism is a rapidly growing industry. As the number of visitors to anyparticular environment or cultural site increases, effective management of the visitorsbecomes more important to ensure that the attraction itself is not damaged and thatthe pressure of numbers does not substantially reduce the level of satisfaction of thetourists' objectives. Effective management depends upon knowing what visitors areseeking, how they will behave, and how this behaviour may be modified by thepresence of others or by particular management strategies. Some of this informationcan be gathered by traditional visitor surveys. Information on likely visitor levels maybe the subject of various demand studies such as gravity modelling. Managementof visitor behaviour, however, depends upon models that allow for interaction betweenvisitors, changes in movement patterns or time allocations as a product of existing siteconditions, estimation of resultant visitor-satisfaction levels, and estimation of theeffect of the visitors on the site.

GIS are increasingly used in site assessment for many kinds of activity: urban devel-opment, ecological modelling, agricultural suitability, etc. In the tourist and recreationalcontext there are GIS-based implementations of models such as the recreation oppor-tunity spectrum (ROS) (Itami and Raulings, 1994), approaches to estimation of visualqualities or visual impacts (Bishop and Hulse, 1994), and, less directly, environmentalimpacts such as erosion (Liao and Tim, 1994).

Management of recreational areas: GIS, autonomous agents,and virtual reality

I D BishopCentre for GIS and Modelling, University of Melbourne, Parkville, Victoria 3052, Australia;e-mail: i.bishop@engineering.unimelb.edu.au

H R GimblettSchool of Renewable Natural Resources, The University of Arizona, Tucson, Arizona 85721,USA; e-mail: gimblett@ag.arizona.eduReceived 9 June 1999; in revised form 10 October 1999

Environment and Planning B: Planning and Design 2000, volume 27, pages 423 ^ 435

Abstract. Management of recreational activity in areas that are culturally or ecologically sensitiverequires knowledge, and effective management, of recreationists' behaviour. In this paper we explorethe role of spatial information systems, spatial modelling, and virtual reality in the analysis andprediction of visitor location and movement patterns. The quantitative modelling of the time spentby visitors on various aspects of the site attractions and of visitor conflict has not been widelyattempted, having only recently become possible because of greater computer power, better spatialdata storage options, and new modelling paradigms. Rule-driven autonomous agents can be used assurrogates for human visitors. Behavioural rules can be derived and calibrated from visitor surveys.This is, however, an expensive and time-consuming process and testing of people's decisions in avirtual environment may provide sufficient information for rule definition. Once a rule-set is deter-mined, the autonomous agents move over a GIS-based model of the landscape. Rendering algorithmsdetermine what an individual agent is able to `see'. Based on the established rules, this and otherfactors (such as tiredness) determine behavioural choice. Recording of model runs to file allowsmanagers to undertake additional analysis to quantify and explore the influence of alternativemanagement options on recreationist movement, congestion, and crowding. Through the GIS,impacts such as erosion can also be modelled. In the longer term the combined models can becomepart of a decision support system for sustainable tourism in fragile environments.

DOI:10.1068/b2637

Computer-based storage of information about a recreational area is the firstrequirement of effective modelling of visitor behaviour. However, it is not sufficientin itself. Also required is a mechanism for linking the behaviour to the site character-istics. This may be the product of statistical analysis such as regression or ANOVA.However, an empirical model will not accommodate interaction between the visitorsthemselves and may not be capable of responding to changes in managementregimes. A more complete model will include interactions between individuals andwill take account of individual responses to specific environmental or site conditions.Increasing computer power, better GIS software, and new paradigms in agent model-ling have now introduced the prospect of modelling at the level of the individualvisitor.

The concept of autonomous agent modelling (AAM) is that each agent (individualvisitor) has his or her own set of rules that describe his or her behavioural (andpotentially emotional) response to particular sets of circumstances. After placing anumber of agents in a siteöas described in a GISöthe simulation is run with eachagent acting autonomously. After a period of model operation a picture emerges ofcollective behaviour.

It is proposed here that:(1) Rules of behaviour can be derived by observation of choices made in a virtualenvironment;(2) These rules can become the basis for effective AAM of recreational behaviour;(3) That visitor satisfaction levels can be estimated from the agent modelling;(4) Environmental impacts and their spatial distribution can be mapped; and(5) The potential exists for extension to a more complete spatial decision supportsystem (SDSS).

3

Environmental impact Satisfaction levels

Sustainability

Erosionmodel

Agentsimulationengine

Visualanalysis

WaterqualitymodelVisual

impactmodel

Slope Elevation Vegetation Built form

3 3 3 3

GIS

Management options

33

Virtual environment

Visitor surveys(in virtual environment)

Agent typeA

Agent typeB

Agent typeC

Behaviouralrules

3

3

3

3 3

3

"

3

3

3

"

3

3

3

33

3

3

3

3

3

3

Figure 1. The concept of integrated GIS and autonomous agent modelling for sustainable naturalarea recreation management. Management options are processed through the system to generatevisitor satisfaction levels and environmental impact levels. The sustainability of the proposal canthus be assessed and adjusted. Additional GIS data sources, agent types or environmental impactmodels can be added as appropriate and available.

424 I D Bishop, H R Gimblett

Thus the effect of a particular management decision will influence the movementoptions. These in turn will influence the environmental effects of the visitors. Figure 1shows the possible configuration of such an SDSS.

GIS modellingData and data structureIn landscape and environmental planning applications the data sets included in theGIS database typically include elevation, slope, aspect, land use, vegetation, soils, andinfrastructure. In urban applications the typical data will include cadastral parcels,roads and utilities, land use, and slope. These data may be stored in vector or rasterform.Vector structures store points, lines, and areas by their precise positions in spacewith attributes attached to each graphic entity in a relational database. Raster data arestored by values on a lattice of cells. A raster structure, though not as locationallyprecise as the vector option, is better suited to many modelling applications. The choiceof structure is significant to the implementation of agent-based models on a GIS.

A number of common techniques of static analysis or modelling are of interest inthe tourism or recreational context. The mechanisms and application of these in visualmodelling, erosion modelling, and the recreational opportunity spectrum are brieflydescribed.

Visual quality and visual impact modellingAssessment of the visual resource is seen as a key element in environmental planning(Brown, 1994). Tourist routes, outdoor recreation areas, etc depend for their interest ontheir visual qualities. It is appropriate to seek to model the intrinsic visual qualities(Bishop and Hulse, 1994) and the visual impacts of developments of land managementpolicies (Hadrian et al, 1988). A key aspect of such modelling is viewshed mapping.Many GIS packages allow for the determination of seen area from any particularviewpoint on the basis of a raster grid of terrain height values (a digital terrain model).After assessment of the seen area the distribution of water, vegetation, infrastructure,etc within the view can be computed and used as input to models of visual quality.

Analysis of the seen environment is a key input to the AAM whenever the behav-ioural rules of agents include reference to visual encounters with other visitors or tolandscape qualities. In the GIS context the visual analysis can be approached throughline-of-sight calculations based on a digital terrain model. Unfortunately the viewshedanalysis used in GIS-based modelling can be computationally slow and becomesrapidly slower as the grain of the data set becomes finer. It also has great difficultydealing with specific information such as the location of a building or a clump of treeswhich are identified in land cover mapping but which have a small spatial extent incomparison with the units of the terrain model. A potential solution is to use the three-dimensional (3-D) rendering algorithms that are integral to the creation of virtualenvironments. This option is discussed further below.

Recreation opportunityThe ROS model (Driver et al, 1987) provides a conceptual framework for relatingopportunities for particular behaviours and experiences to specific settings. The ROStool has been widely used in the USDA Forest Service who are slowly shifting from fieldinventory and hand-mapping techniques to automated mapping. A raster GIS-basedimplementation of the ROS model has been developed by Itami and Raulings (1994).This means that changes to forest resources, such as logging, can be modelled for theirrecreational effect, and alternative strategies generated (Gimblett et al, 1995). As withvisual modelling, the output of ROS modelling can be part of a set of behavioural rulesfor autonomous agents.

Management of recreational areas 425

Erosion and other impactsIn land-use suitability assessments, mapping of erosion potential is important in selec-tion of sites for forest, road building or cropping activities. In these contexts themodels are well established. The universal soil loss equation (USLE) is a popularoption for erosion potential mapping and has been fully implemented in a GIS byseveral researchers (for example Liao and Tim, 1994). Intensive recreational activity ina natural setting can also provoke erosion problems. Erosion potential is thereforeimportant for recreational management (for example, trail planning) whereas changesin traffic levels flowing from agent-based modelling can be translated into predictionsof erosion. Similar analysis might apply to other effects of recreation in the naturalenvironment such as weed invasion, pest infestation, or effects on wildlife (Harris et al,1995).

Thus these static GIS-based analyses have value in their own right but also asinputs to or extensions of agent-based modelling (see figure 1).

Autonomous agentsConcept and recent applicationsAn autonomous agent is a computer simulation that is based on concepts fromartificial life research. Agent simulations are built by using object-oriented program-ming technology. The agents are autonomous because once they are programmed theycan move about the landscape like software robots. The agents can gather data fromtheir environment, make decisions from this information, and change their behaviouraccording to the situation they find themselves in. Each individual agent has its ownphysical mobility, and sensory and cognitive capabilities. This results in behaviour thatechoes that of real animals in the environment.

Individual or agent-based models have been developed in the 1990s as a popularapproach to modelling spatially explicit ecological phenomena. Much of the initialdevelopment was based on creatures with simpler behaviour rules than human tourists.These rules involved responses to the immediate environment and to other nearbyagents. The agents were not given goals or any ability to learn about their environment.More recently researchers have introduced complex agents whose behaviour develops(or emerges) as each agent learns more about its surroundings and which are capableof adapting their complex goals. Several software products to support agent modellinghave been developed (Forrest and Jones, 1994; Langton et al, 1997). The architecture,according to Kinney et al (1995) ` consists of a database containing current beliefs orfacts about the world; a set of current desires (or goals) to be realised; a set of plansdescribing how certain sequences of actions and tests may be performed to achievegiven goals or to react to a particular situation; and an intention structure containingthose plans that have been chosen for [eventual] execution.''

The process of building an agent is iterative and combines knowledge derived fromempirical data with the intuition of the programmer. By continuing to program knowl-edge and rules into the agent, watching the behaviour resulting from these rules, andcomparing it with what is known about actual behaviour, we can observe a rich andcomplex set of behaviours emerge.

Recent studies by Drogoul and Ferber (1995) and Findler and Malyankar (1995)clearly demonstrate the potential for agent-based modelling techniques to examinehuman ^ landscape interactions. As Drogoul and Ferber (1995, page 128) state: ` ... weare interested in the simulation of evolution of complex systems where interactionsbetween several individuals at the micro level are responsible for measurable generalsituations at the macro level. When the situation is too complex to be studied analyt-ically, it is important to be able to recreate an artificial universe in which experiments

426 I D Bishop, H R Gimblett

can be done in a reduced and simulated laboratory where all parameters can becontrolled precisely''. In other words, use of agent modelling will not necessarilyreplicate the behaviour of any one particular human but the overall patterns of collec-tive behaviour can emerge which cannot be predicted by other analytical processes.

In the recreational context there are two stages in the process of establishingbehavioural rules. First, there needs to be an assessment of agent types. Clearly eachindividual in the real world behaves differently in response to stimulus. However, itmay be possible, and appropriate, to consider people as belonging to certain types.These types may be defined by, for example, mode of travel or purpose of visit. Thesecond requirement is that specific responses to possible environmental conditions bedetermined. These conditions may be physical (land slope, vegetation density) orperceptual (number of other visitors in view, tiredness).

Surveys are necessary to determine the visitor types and their conditional responses.These surveys can be based on field interviews and questionnaires or (perhaps moreconveniently) on people experiencing a virtual reality.

Virtual environmentsInteractive visual simulationA virtual reality (VR) has been defined as an environment created by the computer inwhich the user feels present. Offerings may range from landscape simulation generatorssuch as VistaPro (Berger et al, 1996), to animation (Hetherington et al, 1993), to highlyinteractive environments (Graf et al, 1994; Lange, 1994), to the full immersion condi-tion with head-mounted displays (HMD) or a CAVE environment.

Previous reasons for applying VR to landscape have been to communicate thequalities of a specific plan or design, to provide for interactive manipulation of designelements (Orland, 1993) or to undertake experiments in perception (Bishop and Rohr-mann, 1995). In general, the greater the level of interactivity the lower the level of visualrealism and so a trade-off must be made.

The validity of the response to a visual simulation is a key issueöespecially if VRis to be used to study behavioural responses. Validity studies in relation to landscapepresentation media have been going on for a long timeöbeginning with photographs(Shuttleworth, 1980), moving to video (Vining and Orland, 1989), and more recentlydealing with computer graphics (Bergen et al, 1995; Oh, 1994; Watzek and Ellsworth,1994). To date, the immersion option, and whether people respond in a `natural' waywhen wearing an HMD or in a CAVE, has not been widely explored.

Testing behavioural responseVirtual environments have been used for some time in activities such as pilot trainingand the behavioural response is assessed as part of their performance evaluation. Weare not, however, aware of the use of VR for less structured testing in the natural orbuilt environment. Darken and Sibert (1996) have reported their experiments in way-finding in virtual worlds. They conclude:

` ... we will need to determine whether or not a human's conception of an abstractspace is analogous to that of a physical space. Early indications are that this is thecase.''

Subjects in this case used a CRT-based display mounted on a counterbalanced mechan-ical arm to see into the virtual world. They were required to search for specific objectsin the environment.

In preliminary experiments we are using a normal computer monitor as thewindow on the virtual world. Software has been prepared by using IRIS Performer

Management of recreational areas 427

(Rolf and Helman, 1994) that moves the viewer automatically along a path (figure 2)until a decision point is reached (see http://surprise.sli.unimelb.edu.au/�ck/pfv/pfv.html).At this point the subject can either choose the right, centre, or left path option by usingthe mouse button. In one case various objects (a brightly coloured modern sculpture, akiosk, and a group of people) have been added to the park environment near to thechoice points but clearly on one fork or the other. In a second case, in the Scottishcountryside, paths have been flanked by, or pass through, different vegetation typesand densities. At the first fork in the urban park, of the twelve subjects who saw anintroduced feature all of them chose the path towards that feature. In the Scottish trialit was found that the preference for one path over the other, when the whole samplewas considered, was significant at three of the ten choice points. Separation of thesample into two groups based on cluster analysis showed clear within-group preferenceat a further four choice points. A visual review of the dominant choices showed thatvery few people were inclined to select either the very enclosed or the very openoptions (preferring lightly wooded pathways). The separation into subgroups appearsto distinguish between those who chose to head towards the nearest high ground andthose who were perhaps attracted by the valley floor and/or more distant hills.

These results suggest that rules derived by using this approach can be added to thebehavioural profile of the autonomous agents so that their behaviour changes inresponse to management decisions that affect their visual environment. These rulesmay, in some cases, prove to be generic. In others they will be site specific. There is agood deal of experimentation to be done before such details emerge. We also need toremember that virtual environments appeal to a limited set of human senses (primarilyvisual and auditory). We can visit the cafe but we cannot enjoy the virtual drink, norcan we breathe a salty breeze from the sea. A choice of uphill journey will not bephysically taxing without access to haptic feedback mechanisms.

Visual analysis for agentsThe difficulties with using GIS for analysis of the field of view of the agents werediscussed above. An alternative is to use the Z-buffer hidden surface algorithm that isintegral to 3-D rendering. This determines not only what can be seen in the view butalso how far away it is.

Figure 2.View from the urban park simulation being used to test behavioural choices. Thesubject chooses the left or right path. Various objects, which may influence the decision, areintroduced.

428 I D Bishop, H R Gimblett

A series of linked studies undertaken by Wherrett (1998) have shown that:(a) good predictive models of landscape preference can be based on land form andland cover parameters measured from photographs; and(b) these parameters can also be measured from simulations of the same landscapesgenerated from GIS-based data.

Thus the software used to generate 3-D views of landscape may overcome thedifficulties of attempting to run visual quality analysis within the GIS environmentitself. This is particularly true when hardware-based acceleration of the renderingroutines occurs: increasingly the case as 3-D graphics processing boards which imple-ment graphics libraries (such as OpenGL) become available at low cost. Such boardsmake calculation of important variables, for visual analysis, such as visible surfacesand surface shading easy and fast. Image enhancement techniques, such as anti-alias-ing that identifies local `horizons' may also be harnessed for visual analysis. Thesetechniques also work on true 3-D models rather than the 2.5-D limitation of GISsoftware. Hence the difficulties presented by local buildings or vegetation no longerapplyöprovided they are recognised in the database.

A case studyA few researchers have now sought to combine spatially explicit agent modelling withGIS in order to model behavioural outcomes in a realistic environment (Gimblett et al,1996; Kohler et al, 1996). In this paper we use one such study (Gimblett, 1998) as anexample. This study is based on recreational behaviour in Broken Arrow Canyon,Arizona, a popular tourist centre for desert landscape experience.

Agent type identificationIn a particular tourist setting, the first modelling task will be to identify a set of agenttypes. These may be distinguished by their mode of movement, their goals, their pref-erence for crowds or isolation or other characteristics that will shape their behaviour.

In the case of Broken Arrow Canyon, the identified agent types were hiking agents,mountain-bike riding agents, and organised jeep-tour agents. The hikers and bikerswere subdivided, on the basis of extensive visitor surveys and factor analysis, into socialexperience-oriented (gregarious) and landscape experience-oriented (isolationist).

Agent behavioural rulesBehavioural rules specify how an individual member of an agent set will respond toparticular situations, their movement speed, their need for rest, their reaction toencounters with other agents, their length of stay at particular attractions, etc.

Ideally all these rules will be derived empirically from observation. In the BrokenArrow study only some of the rules had direct supporting evidence; others were basedon `reasonable' suppositions. The major rules are described briefly below.

(1) Range of movementThe first parameter that must be set for all agents is their potential range and availablelocations within that range. In some conditions of agent modelling it will be appro-priate to permit the agents to range across the whole database. In other cases it willbetter reflect actual behaviour and management constraints if they are confined tospecific paths.

This choice may also influence the configuration of data in the supporting GIS. Forfree-ranging agents, a raster data structure will permit movement from the current cellto any adjacent cell. With constrained paths it may be more accurate to define these invector format and thus retain more precise placement for the individual agent. This canbe of particular importance in determining visual contact in complex terrain. In the

Management of recreational areas 429

Broken Arrow study, hikers, bikers, and jeeps had their own fixed trails although thesewere modelled as linked available cells in a raster database of 10 m resolution. The trailsintersect as shown in figure 4.When more than one agent is in a cell, this is a physicalencounter. When agents can see each other but are not in the same cell, this is a visualencounter.

(2) Speed of movementDifferent agent types move at different speeds. A particular agent type may vary his orher speed depending on special circumstances. In the Broken Arrow Canyon study, theaverage agent speeds for hikers utilised earlier studies on recreationist movementpatterns by Wagtendonk (1980). As shown in table 1 these varied considerably withslope and slope direction.

Speeds for bikers and jeep travellers were based on total trail journey times asrevealed by the visitor survey.

(3) Variations in speedThe speeds of the different agent types were also varied by their age, their recentenergy loss or gain (by rest), and their tendency to join with others or to avoidcompany.

The last factor depends not only on proximity of the agent in question to otheragents but also on the existence of line-of-sight between the two. This depends on theterrain, the vegetation, and any buildings that may be in between. Ideally all three arepart of the database. In the Broken Arrow study there were no buildings in the vicinityand the vegetation was generally short and sparse. Consequently, visibility modellingwas based on the terrain model alone. The raster terrain model has heights at 10 mintervals across the landscape.Visibility was determined to a threshold of 1 km. Beyondthis distance it was assumed that the agent would not respond to the sight of anotheragent. If an agent was found to be visible within 1 km then other specific rules wereapplied.

Figure 3 shows some of the rules employed and also indicates that these can eachbe rendered active or inactive for model testing purposes.

Model operationsThe model runs under the RBSim software (http://nexus.srnr.arizona.edu/�gimblett/rbsim.html) (Gimblett et al, 1999; Itami, 1999) developed by using routines drawnfrom the raster GIS software package SAGE (Itami and Raulings, 1994). The initialdialogue box gives the user access to a number of independent set-up windows. Theuser determines how many agents of each type are to be used in the modelling process,what age groups they fall into, and the time intervals between them setting out. Thesecan also be randomised within constraints. The user also establishes the total durationof the simulation.

Table 1.Variation in recreational walking speed as determined by Wagtendonk (1980).

Slope Slope direction Mean speed, m sÿ1

Gentle Up 0.87Down 0.91

Moderate Up 0.65Down 0.82

Steep Up 0.53Down 0.85

Note: standard error of mean in each case is approximately 0.05 m sÿ1.

430 I D Bishop, H R Gimblett

The simulation engine generates recreation agents from scheduling parameters setby the user. It then moves forward in time and adjusts agent positions according to therules followed by each agent. The new positions are displayed on screen as shown infigure 4.

Figure 3. Some of the rules which the user may choose to include in the simulation.

Figure 4. Screen presentation of visitor locations during model operation.

Management of recreational areas 431

As the model runs over the specified simulation duration it also records statistics(at user-specified intervals) about the run. These include the position of each agent,their energy levels, the time spent viewing landscape features, and the number ofperceived and actual encounters with other agents.

ResultsBased on the summary statistics it is possible for each model run to map, for example,the number of encounters between one group of agents (such as hikers) and anothergroup (such as jeeps) which are generally regarded as conflicting activities [figures 5(a)

(a)

(b)

30

20

10

0

Trailhead Chicken Point Submarine Rock Trailhead

Total hikers

Total bikers

Total jeeps

101 201 301 401 501 601 701

Hikers trail sequence, cells

Figure 5. (a) Graphical and (b) on-terrain views of the encounters by hikers on a specific modelrun. Cells in the digital terrain model that are on the trail have been raised by an amountproportional to the number of encounters at that point to create the on-terrain view. The chart(a) covers the full extent of the hiker trailöout and return. The peak in numbers of encounterswith bikers near the centre of the plot corresponds to the distant peak near the furthest point ofthe trail in figure (b).

432 I D Bishop, H R Gimblett

and 5(b)]. Based on the surveyed objectives of different visitor types it is also thereforepossible to graph visitor satisfaction levels during their visit. Using these outputs wecan assess different trail configurations or permitted visitor numbers or activities fortheir contribution to overall tourist satisfaction. Reductions in encounter levels, andhence satisfaction, were shown to be very possible.

ConclusionsIn general, tourists are not confined to a specific trail. If a variety of trail options areprovided for the agents, or if they are allowed to roam freely across the database, thenadditional rules would need to be developed to simulate choice of direction. Such rulesmay be based on the relative length or steepness of the available trail options, onaccess to specific visitor attractions, on the type or density of vegetation or on thegeneral visual attractiveness of the optional ways forward. Opening up the directionalpossibilities for agents also complicates the process of rule derivation in the virtualenvironmentöbut not nearly as much as it would for surveys in the real environment.Subjects in a virtual world can be given free rein to explore and their movementslogged by the computer. With sufficient subjects involved the logs can be analysed forpatterns of behaviour and choice.

Development of the rules will require extensive additional research of both ageneric nature (what are the factors which influence movement patterns?) and alsospecific to any particular area of study (how are the general factors affected by localconditions or attractions?) Virtual environments can make a significant contribution toresearch into both these questions. We see therefore increasing linkages between 3-Drepresentation of virtual environments, the elicitation of responses to environments,and the modelling of behaviour in, attitudes to, and decisions about the environment.

The technologies will also increasingly work with common data sets (as suggested infigure 1) and also with common algorithms. For example, a GIS is able to store dataand calculate a viewshed in most conditions in the natural environment. However, theviewshed algorithm is not particularly rapid and a GIS is not equipped to deal withcomplex built form. In a substantially built environment the built form would have to bemodelled with greater complexity and a CAD structure used to store the data. In eitherenvironment, we envisage use of hardware-based rendering algorithms (using low-cost3-D graphics cards as used by computer games) for rapid visual assessment allowingautonomous agents to make judgments of visual quality and hence environmentalresponses and decisions. The outcome can be better procedures for planning andmanagement of fragile landscapes under pressure from tourism or other development.

Acknowledgements. Robert Itami made SAGE available and was instrumental in the developmentof the RBSim software. This research was supported in part by funds provided by the RockyMountain Forest and Range Experiment Station, Forest Service, US Department of Agriculture.We thank the Experiment Station and the Coconino National Forest Service for their supportand assistance in facilitating the Broken Arrow Canyon project.

ReferencesBergen S D, Ulbricht C A, Fridley J L, Ganter M A, 1995, ` The validity of computer-generated

graphic images of forest landscape'' Journal of Environmental Psychology 15 135 ^ 146Berger P, Meysembourg P, Sales J, Johnston C, 1996, ` Towards a virtual reality interface for

landscape visualization'', in Proceedings of the Third International Conference/Workshop onIntegrating GIS and Environmental Modelling Sante Fe, NM; http://www.negia.ucsb.edu/conf/SANTE-FE-CD-ROM/main.html

Bishop I D, 1996, ` Comparing regression and neural net based approaches to modelling ofscenic beauty'' Landscape and Urban Planning 34 125 ^ 134

Bishop I D, Hulse DW, 1994, ` Predicting scenic beauty using mapping data and geographicinformation systems'' Landscape and Urban Planning 30(1-2) 59 ^ 70

Management of recreational areas 433

Bishop I D, Robey M, 1994, ` Implementing an environmental impact model within a geographicinformation system'', in Proceedings of AURISA94 Australasian Urban and RegionalInformation Systems Association, Executive Office, Stephen House, 32 Thesiger Court,PO Box 307, DeakinWest, ACT 2600

Bishop I D, Rohrmann B,1995, ` Visualisation tools: their role in understanding human responses tothe urban environment'', in Proceedings of Fourth International Conference on Computers inUrbanPlanningandUrbanManagementDepartment ofGeography and Environmental Studies,University of Melbourne, Parkville,Victoria

Brown T, 1994, ` Conceptualizing smoothness and density as landscape elements in visual resourcemanagement'' Landscape and Urban Planning 30 49 ^ 58

Darken R P, Sibert J L, 1996, ` Wayfinding strategies and behaviors in large virtual worlds'',Proceedings of CH196,Vancouver, BC; http://www.acm.org/sgichi/chi96/proceedings/papers/Darken/Rpd-txt.htm

Driver B L, Brown P J, Gregoire T G, 1987, ` The ROS planning system: evolution, basic conceptsand research needed'' Leisure Sciences 9 201 ^ 212

Drogoul A, Ferber J, 1995, ` Multi-agent simulation as a tool for studying emergent processes insocieties'', in Simulating Societies: The Computer Simulation of Social PhenomenaEds N Gilbert, A Doran (UCL Press, London) pp 127 ^ 142

Findler N V, Malyankar R M, 1995, ` Emergent behavior in societies of heterogeneous, interactingagents: alliances and norms'', in Artificial Societies: The Computer Simulation of Social LifeEds N Gilbert, R Conte (UCL Press, London) pp 212 ^ 236

Forrest S, Jones T, 1994, ``Modeling complex adaptive systems with Echo'', in Complex Systems,Mechanism of Adaptation Eds R J Stonier, Xing HuoYu (IOS Press, Amsterdam) pp 3 ^ 20

Gimblett H R, 1998 Simulating Recreation Behaviour in ComplexWilderness Landscapes UsingSpatially-explicit Autonomous Agents PhD thesis, Department of Architecture, Building andPlanning, University of Melbourne, Parkville,Victoria

Gimblett H R, Richards M T, Itami R M, Raulings R J, 1995, ` Using interactive, GIS-basedmodels of recreation opportunities for multi-level forest recreation planning'', in Proceedingsof Geomatics 95 Natural Resources Canada

Gimblett H R, Durnota B, Itami R M, 1996, ` Some practical issues in designing and calibratingartificial human-recreator agents in GIS-based simulated worlds'', Proceedings of ComplexityInternational http://www.csu.edu.au/ci/vol3/gimetal3/gimetal3.html

Gimblett H R, Itami R M, Richards M, 1999, ` Simulating wildland recreation use and conflictingspatial interactions using rule-driven intelligent agents'', in Integrating GIS and Agent BasedModeling Techniques for Understanding Social and Ecological Processes Ed. H R Gimblett(Oxford University Press, Oxford) in press

Graf K C, Suter M, Hagger J, Meier E, Meuret P, Nuesch D, 1994, ` Perspective terrainvisualizationöa fusion of remote sensing,GIS and computer graphics''Computers andGraphics18 795 ^ 802

Hadrian D R, Bishop I D, Mitcheltree R, 1988, `Automated mapping of visual impacts in utilitycorridors'' Landscape and Urban Planning 16 261 ^ 283

Harris L K, Gimblett H R, ShawW W, 1995, ` Multiple use management: using a GIS model tounderstand conflicts between recreationists and sensitive wildlife'' Society and NaturalResources 8 559 ^ 572

Hetherington J, Daniel T C, Brown T C, 1993, ``Is motion more important than it sounds? Themedium of presentation in environment perception research'' Journal of EnvironmentalPsychology 13 283 ^ 291

Itami R M, 1999, ` Mobile agents with spatial intelligence'', in Integrating GIS and Agent BasedModeling Techniques for Understanding Social and Ecological Processes Ed. H R Gimblett(Oxford University Press, Oxford) in press

Itami R M, Raulings R J, 1994 SAGE Reference Manual Digital Land Systems Research,Parkville,Victoria

Kinney D, Georgeff M, Bailey J, Kemp D B, Ramamohanrao K, 1995, `Active databases andagents systems: a comparison'', technical report 53, Australian Artificial IntelligenceInstitute, 171 Latrobe Street, Melbourne; http://www.aaii.oz.au/research/techreports/tnlist.html

Kohler T A,VanWest C R, Carr E P, Langton C G, 1996, `Agent-based modelling of prehistoricsettlement systems in the Northern American southwest'', in Proceedings of the ThirdInternational Conference on Integrating GIS and Environmental Modelling Sante Fe, NM;http://www.ncgia.ucsb.edu/conf/SANTA-FE-CD-ROM/main.html

434 I D Bishop, H R Gimblett

Lange E, 1994, ` Integration of computerized visual simulation and visual assessment inenvironmental planning'' Landscape and Urban Planning 30 99 ^ 112

Langton C, Burkhart R, Ropella G, Daniels M, 1997, ` SWARM: the swarm simulation system'';http://www.santafe.edu/projects/swarm/

Liao H-H, Tim U S, 1994, ` Interactive modelling of soil erosion within a GIS environment'', inProceedings of URISA94 Urban and Regional Information Systems Association,1460 Renaissance Drive, number 305, Park Ridge, IL 60068

Oh K, 1994, `A perceptual evaluation of computer-based landscape simulations'' Landscape andUrban Planning 28 201 ^ 216

Orland B A, 1993, ` Computer video-imaging in visual simulation'', in Environmental Simulation:Research and Policy Issues Eds R Marans, D Stokols (Plenum Press, NewYork) pp 213 ^ 252

Rolf J, Helman J, 1994, ` IRIS Performer: a high performance multiprocessing toolkit for real-time 3D graphics'', in Proceedings of SIGGRAPH 94 Association of Computing Machinery,1515 Broadway, NewYork, NY 10036

Shuttleworth S, 1980, ` The use of photographs as an environmental medium in landscape studies''Journal of Environmental Management 11 61 ^ 70

Vining J, Orland B, 1989, ` The video advantage: a comparison of two environmental representationtechniques'' Journal of Environmental Management 29 275 ^ 283

Wagtendonk J W, Benedict J M, 1980, ` Travel time variation on backcountry trails'' Journal ofLeisure Second Quarter 99 ^ 106

Watzek K A, Ellsworth J C, 1994, ` Perceived scale accuracy of computer visual simulations''Landscape Journal 13(1) 21 ^ 36

Wherrett J R, 1998 Natural Landscape Scenic Preference: Techniques for Evaluation and SimulationPhD thesis, Robert Gordon University, Aberdeen

Management of recreational areas 435

ß 2000 a Pion publication printed in Great Britain