GIS/EM4 - Ecological change under pressure of urban development

RBSIM 2: Simulating human behavior in National Parks in Australia

Integrating GIS and Intelligent Agents to predict recreation conflicts in high use natural environments

GIS/EM4 No. 57

Robert M. Itami
Glen S. MacLaren
Kathleen M. Hirst
Robert J. Raulings
H. Randy Gimblett

Abstract

This paper describes advancements in recreation management using new technology that couples Geographic Information Systems (GIS) with Intelligent Autonomous Agents to simulate recreation behaviour in real world settings. RBSim is a computer program that allows park management to explore the consequences of change to any one or more variables so that the goal of accommodating increasing visitor use is achieved while maintaining the quality of visitor experience. RBSim provides both a qualitative understanding of management scenarios by the use of map graphics from a GIS as well as a quantitative understanding of management consequences by generating statistics during the simulation. Managers are able to identify points of over crowding, bottle necks in circulation systems, and conflicts between different user groups.

RBSim is designed to be easy to use by park management staff. This is facilitated through a tight integration with MapInfo GIS which allows a practical solution for quickly building complex simulation models. Simulation techniques provide methods for evaluating details of management decisions as they impact visitors and the environment. No other analysis tool that achieves this level of understanding is currently available.

The paper describes RBSim focussing on the parameters required to generate simulation models for existing and proposed park management scenarios.

Keywords

Intelligent Agents, Mobile Agents, Geographic Information Systems, Recreation Behaviour Simulation.

Introduction

The increasing demand for outdoor recreation worldwide is the consequence of people having more leisure time, greater mobility and more disposable income. The proliferation of new types of recreation such as mountain bike riding, snow boarding, canyoning and other emerging activities are often in conflict with more traditional outdoor activities. As visitor numbers increase, there is a simultaneous increase in environmental impacts, crowding, and conflicts between different recreational types and users. These circumstances make recreation management a complex problem. Managers of natural areas must accommodate increasing visitor use while at the same time, maintaining environmental quality and assuring visitors have the high quality experience they anticipate.

Conventional methods used in the design and planning of park management facilities have focussed on the use of user surveys and traffic counts to estimate the requirements. However these methods fall far short of the real needs of managers who need to comprehensively evaluate the cascading effects of the flow of visitors through a sequence of sites and estimating the effects of increasing visitor flows through time. In addition, managers need to know if designed capacities for parking, visitor centers, roads, camping areas, and day use facilities can accommodate projected visitor numbers. Crowding, conflicts between different recreation modes, impacts on environments and seasonal effects such as day length and weather area all factors park planners must consider in the design and location of new facilities.

There are many options available to park managers to deal with heavy visitor use. New sites can be opened up, a system of reservations can be implemented; areas can be closed so sites can recover from over use; facilities can be expanded to accommodate large numbers of visitors. Each of these strategies will have different impacts on the overall system. The complex inter-relationships between these decisions are almost impossible for a manager to predict. It is in this context where simulation of recreation behavior is of real value. This paper describes a computer simulation methodology that uses intelligent agents to simulate recreation behavior, coupled with Geographic Information Systems to represent the environment.

Computer simulation has typically been utilized to study biophysical environmental processes (Goodchild et al., 1993) with more recent attention being focused on "human dimensions" of environmental systems. Many of these simulation modeling efforts employ a number of artificial intelligence techniques combined with Geographic Information System (GIS) functions to address human-environment interactions (e.g. Slothower, 1996; Gimblett et al. 1996a, 1996b; Briggs et al. 1996). Exploratory studies (eg. Berry et al. 1993; Gimblett et al. 1994; Saarenmaa et al. 1994; Gimblett and Itami 1997) and Itami (1988) have suggested the use of cellular automata (CA) as a method for simulating dynamic environmental processes over large scale landscapes, and applications of this approach have been successfully demonstrated (e.g., Green et al. 1987; Manneville et al. 1989). Individual-Based Models (IBM) have recently been applied to develop spatially-explicit models of ecological phenomena. IBMs are "organisms-based models capable of modeling variation among individuals and interactions between individuals" (Slothower et al. 1996). IBMs offer potential for studying complex behavior and human/landscape interactions within a spatial framework. One form of individual-based modeling approaches, "agent-oriented programming," facilitates representation of dynamic interactions among multiple agents that coexist in an environment. Included in this approach is the study of complex adaptive systems, where tools and techniques are being developed to study emergent behavior, for example Swarm (Hiebeler, 1994, Langton et al. 1995); Echo (Forrest et al. 1994), and GENSIM (Anderson and Evans 1995). The combination of spatially-explicit IBMs, reactive agents, artificial intelligence (AI) and Geographic Information Systems (GIS) offer a powerful alternative to previous modeling techniques for exploring emergent, complex, evolutionary processes. This paper describes the use of agent-based systems coupled with GIS to simulate human recreation behavior in natural landscapes.

Multi-Agent simulation of outdoor recreation

RBSim 2 (Recreation Behavior Simulation) (Gimblett & Itami, 1997; Gimblett, 1998; 1998a; Gimblett et al. 1999; Itami et al., 1999) is a computer simulation tool, integrated with a Geographic Information System (GIS) that is designed to being used as a general management evaluation tool for any park. This capability is achieved by providing a simple user interface that will import park information required for the simulation from a geographic information system. Once the geographic data is imported into RBSim 2, the park manager can change a number of variables including the number and kind of vehicles, the number of visitors, and facilities such as the number of parking spaces, road and trail widths and the total capacity of facilities

RBSim 2 allows park management to explore the consequences of change to one or more variables so that the quality of visitor experience is maintained or improved. Statistical measures of visitor experience are generated by the simulation model to document the performance of any given management scenario. Management scenarios are saved in a database so they can be reviewed and revised. In addition, the results of a simulation are stored in a database for further statistical analysis. The software provides tables and graphs from the simulation data so park managers can identify points of over crowding, bottle necks in circulation systems, and conflicts between different user groups.

Specifically RBSim 2 uses concepts from recreation research and artificial intelligence (AI) and combines them in a GIS to produce an integrated system for exploring the complex interactions between humans and the environment (Gimblett et al. 1996a; Gimblett et al. 1996b, Gimblett 1997a, Gimblett and Itami, 1997, Itami et al., 1999). RBSim 2 joins two computer technologies:

Geographic Information Systems to represent the environment and;
Autonomous human agents to simulate human behavior within geographic space.

RBSim 2 uses autonomous agents to simulate recreator behavior. An autonomous agent is a computer simulation that is based on concepts from Artificial Life research. Agent simulations are built using object oriented programming technology. The agents are autonomous because once they are programmed they can move about their environment, gathering information and using it to make decisions and alter their behavior according to specific environmental circumstances generated by the simulation. Each individual agent has it's own physical mobility, sensory, and cognitive capabilities. This results in actions that echo the behavior of real animals (in this case, humans) in the environment.

Simulation Object Model

RBSim 2 uses object oriented software technology to model components of the overall simulation system. These software objects include:

Network Object Model - contains network topology for roads, trails and other linear features organised as a forward star network with associated attributes and methods for calculating travel time and distances across the network. RBSim imports vector data from MapInfo GIS and constructs the network topology automatically.
Terrain Model - contains elevation data represented as a regular grid of elevations. Since MapInfo does not support raster data, data is imported from ESRI’s grid export format.
Graphics Engine - provides visualisation of the park as a map showing current location of recreation agents.
Agent object - represents the recreator's personality, mobility characteristics, and reasoning system.
Typical Trips - represent the gross travel patterns of visitors to a park. Typical trips are defined by entry/exit node pairs, travel mode, arrival times, and agent type. Typical trips also include hourly arrival schedules at the entry node for the trip. Typical trips can be defined from analysis of traffic survey data, or they can be defined by park managers familiar with visitor use patterns.
Locale Trips - represent the idiosyncratic travel paths of agents travelling by foot. These trips are determined by time available, fitness of the agent, preferences for different types of landscape attractions (scenic, historic, educational, etc), terrain characteristics, and the behavior of other agents. Locale trips are generated incrementally by agents at run time in response to the above states.
Arrival Schedule - contains the exact arrival times for each agent for each typical trip for the duration of the simulation. Each arrival is defined by the date and time of arrival to the nearest second, the agent type, travel mode, trip duration and number of passengers. The Arrival Schedule is generated by the simulation engine at run time.
Simulation Engine - generates the arrival schedule of agents, controls simulation events such as weather, road opening and closure, seasonal events and other user defined events.
Output Object - stores run-time states for agents and the network at intervals specified by the user.
Output Analysis - generates summaries of runs for network links and nodes, agent encounters, and the number of visitors at each facility. Summary statistics are output in text files, maps are automatically generated showing the magnitude of visitor use at each node.

It is beyond the scope of this paper to describe the entire simulation system. Specifics of the Agent model is described in Itami (in Press), Itami and Gimblett (in Press). More detail on the GIS integration can be found in Itami et al. (1999). In this paper we focus on parameterization of the model using traffic count data, and the user paradigm for building models based on expert judgement.

Parameterizing the RBSim 2 model

The RBSim 2 simulation model requires 3 classes of parameters:

Environmental parameters

Road and Trail Network - as described earlier, the road and trail network are derived from a GIS. RBSim currently links to MapInfo GIS. MapInfo does not support topologically structured networks. RBSim has automated utilities for creating the network topology. The network database is structured in a forward star network. This data structure allows analysis using network algorithms for finding shortest distance and shortest time paths between any pair of origins and destinations. These algorithms are used by agents in their reasoning system for planning trips. Once the forward star data structure is generated the user then manually parameterizes the network with attributesincluding: speed along links; turning impedance; and park facilities (including description, capacity, and estimated duration of visit).
Terrain - terrain factors are important especially for pedestrians navigating difficult terrain. Terrain is also used as a parameter in the calculation of visibility between agents. RBSim uses grid elevations as a terrain model and imports data from ESRI’s grid export format.

Typical Trips

Throughout the development of RBSim we were cognizant of the duo nature of a simulation system that will be applied in a real-world management setting. The simulation system must be able to simulate from field collected data as a means to validating the simulation model on the one hand. On the other hand, it must also meet the needs of recreation managers who wish to develop new management scenarios where no user data is available.

A critical aspect of simulating large numbers of users in a travel network is the development of methodologies for generating trips for each agent. It is possible to generate transition probabilities from node to node from field collected data, however, this methodology does not generate trips that are representative of real-world trips. In fact this methodology is just as likely to produce many nonsensical trips where agents go back and forth repeatedly between pairs of nodes, or traverse the network in non-optimal paths. A critical problem with probabilistic transition tables, is that they cannot be used to generate trips for new network configurations for which there is no data. The concept of typical trips acknowledges the fact that many visitors will follow similar routes through a park. The variability between trips is determined by the entry node to the park, the exit node, the travel mode, the intent of the trip and the duration of the trip. Park managers who have worked in a park for a period of time will develop a good understanding of these visitor patterns. Generally the "main route" along main roads is fairly easy to predict. However there is considerable variation within secondary trips off the main routes. Secondary trips to specific destinations or "locales" may have considerable more variation because of personality preferences, level of fitness, duration of visit and travel mode. RBSim uses agent technology to simulate both levels of this hierarchy.

At the general level a typical trip is defined by the following:

Entry Nodes
Exit Nodes
Destinations
Arrival Rates
Occupancy (number of passengers in cars or buses)
Trip duration

These parameters can be derived from traffic survey data, however in the cases where this data may not exist, Park Managers can specify typical trips based on field experience. There is a simple user interface that allows the user either to generate typical trips from traffic data, or manually.

Locale Trips

At the level of a locale (generally where visitors leave their car and proceed on foot), there is much more variation. This variation is simulated using agent technology. This requires that the agent have a reasoning system to generate the locale trip, navigate across varying terrain and to decide on how long to visit different site attractions based on preferences. The reasoning system is described in Itami (in press) and Itami and Gimblett (in press).

The complex relationships between preferences, fitness levels, travel speeds and terrain variables are still a matter of good social science research (see Gimblett et al, 1996a and b, Gimblett, 1998). However a framework for relating preferences to site attractions has been implemented in RBSim 2. Personality preferences are generated by the user using the Analytical Hierarchy Process (AHP) (Saaty, 1995) to generate weights for preferences for site attractions including scenery, history, environment or other site interests. The user then provides travel mode, average travel speed, and fitness level for the agent:

These parameters are stored in a database for each agent personality. These personalities may then be associated with typical trips to define the parameters necessary for a complete trip through the park.

Discussion and Conclusion

The agent simulations are an excellent method for modeling recreator encounters and conflicts. The agent simulations provide a dynamic view of encounters between agents and identify the spatially explicit locations where they occur. The effect of these encounters on the overall recreational experience is still unknown. However, this simulation environment provides a way to test and evaluate many scenarios of recreational use. Figure 1 shows the simulation interface used in the RBSim 2 software.

Figure 1. Interface to the simulation environment within RBSim 2

The challenges relating to the use of simulation modeling in the context of applied recreation management demands that the simulation can be parameterized with field collected data where this exists. However it must also support concepts and techniques for allowing non-technical managers to provide estimated parameters based on field experience. RBSim accommodates both of these instances allowing for great flexibility in its application.

RBSim 2 is a significant step forward to make linkages between GIS, multiagent systems and recreation behavior modeling. By generalizing the software to parameterize a simulation from existing GIS datasets, the RBSim 2 simulation platform can be used to study management alternatives in any park. The tight integration with GIS also allows ancillary models such as environmental impact models and economic models to be integrated into the simulation. Park managers can then test many different management assumptions in both qualitative and quantitative terms before committing resources to expensive construction projects.

Acknowledgements

We would like to acknowledge the considerable contribution of Dino Zanon, of Parks Victoria, Australia was invaluable in the development of requirements for Park management. Parks Victoria supported the implementation of RBSim 2 for Port Campbell National Park and Bay of Island Coastal Park.

References used

Anderson, J. & M. Evans. (1995). A Generic Simulation System for Intelligent Agent Designs. Applied Artificial Intelligence, Volume 9, Number 5, October 1995. pps. 527-562

Berry, J.S., G. Belovsky, A. Joern, W.P. Kemp & J. Onsager. (1993). Object-Oriented Simulation Model of Rangeland Grasshopper Population Dynamics. In Proceedings of Fourth Annual Conference on AI, Simulation, and Planning in High Autonomy Systems, Tucson, AZ. September 20-22, 1993, pp. 102-108.

Briggs, D.; Westervelt, J.; Levi, S.; Harper, S. (1996). A Desert Tortise Spatially Explicit Population Model. Third International Conference. Integrating GIS and Environmental Modeling. January 21-25. Santa Fe, NM.

Forrest, S. & T. Jones. (1994). Modeling complex adaptive systems with Echo. In Russel J. Stonier and Xing Huo Yu. editors, Complex Systems. Mechanism of Adaptation, pps. 3-20. IOS Press, 1994.

Gimblett, H.R.; Durnota, B.; Itami, R.M. (1996a). Spatially-Explicit Autonomous Agents for Modelling Recreation Use in Complex Wilderness Landscapes. Complexity International Journal. Volume 3.

Gimblett, H.R.; Itami, R. M.; Durnota, B. (1996b). Some Practical Issues in Designing and Calibrating Artificial Human Agents in GIS-Based Simulated Worlds. Complexity International Journal. Volume 3.

Gimblett, H.R.; Itami, R. M. (1997). Modeling the Spatial Dynamics and Social Interaction of Human Recreationists’ Using GIS and Intelligent Agents. MODSIM 97 - International Congress on Modeling and Simulation. Hobart, Tasmania. December 8-11.

Gimblett, H.R. (1998). Simulating Recreation Behavior in Complex Wilderness Landscapes Using Spatially-Explicit Autonomous Agents. Unpublished Ph.D. dissertation. University of Melbourne. Parkville, Victoria, 3052 Australia.

Gimblett, H. R. C. Roberts, T. Daniel, M. Mitner, S. Cherry, D. Kilbourne, M. Ratliff, D. Stallman, R. Bogle. Intelligent Agent Modeling for Simulating and Evaluating River Trip Scheduling Scenerios for the Grand Canyon National Park. In H. R Gimblett, editor. Integrating GIS and Agent based modeling techniques for Understanding Social and Ecological Processes. Santa Fe Institute/Oxford Press. (In Press).

Goodchild, M., B.O. Parks & L.T. Steyaert. (1993). Environmental Modeling with GIS. Oxford University Press, 1993.

Hiebeler, D. (1994).The Swarm Simulation system and individual-based modeling. In J. Michael Power, Murray Strome, and T.C. Daniel, editors, Decision Support 2001. 17th Annual Geographic Information Seminar and the Resource Technology '94 Symposium, volume 1, pps. 474-494. American Society for Photogrammetry and Remote Sensing, 1994.

Itami, R.M. (1988). Cellular Worlds: Models for Dynamic Conceptions of Landscapes. Landscape Architecture, 78(5): 52-57, 1988.

Itami, R. M. (1994). Simulating Spatial Dynamics: Cellular Automata Theory. Landscape and Urban Planning 30. pps. 27-47.

Itami, R. M. (in Press) Mobile Agents with Spatial Intelligence. In H. R Gimblett, editor. Integrating GIS and Agent based modeling techniques for Understanding Social and Ecological Processes. Santa Fe Institute/Oxford Press.

Itami, B, Randy Gimblett, Rob Raulings, Dino Zanon, Glen MacLaren, Kathleen Hirst, Bohdan Durnota (1999): RBSim: Using GIS-Agent simulations of recreation behavior to evaluate management scenarios in Proceedings of AURISA 99, 22-26 November 1999, Australasian & Regional Information Systems Association, Inc. Ewan G. Masters, Editor. Canberra, ACT (Refereed paper)

Langton, C., N. Minar & R. Burkhart. (1995). The Swarm Simulation System: A tool for studying complex systems. Santa Fe Institute, Santa Fe, New Mexico, USA, http:// www.santafe.edu/ projects/swarm/ swarmdoc/swarmdoc.html, 1995.

Manneville, P., N. Boccara, G.Y. Yichniac & R. Bidaux. (1989). Cellular Automata and Modeling of Complex Physical Systems. Springer, New York.

Saarenmaa, H.; Perttunen, J.; Vakeva, J.; Nikula, A. (1994). Object-oriented modeling of the tasks and agent in integrated forest health management. AI Applications in Natural Resource Management. 8 (1), pps. 43-59.

Saaty, Thomas L. (1995) Decision making for leaders: the analytical hierarchy process for decisions in a complex world. Published Pittsburgh, Pa. RWS Publications.

Slothower, R. L.; Schwarz, P. A.; Johnson, K. M. (1996). Some Guidelines for Implementing Spatially Explicit, Individual-Based Ecological Models within Location-Based Raster GIS. Third International Conference Integrating GIS and Environmental Modeling. January 21-25. Santa Fe, NM.

Authors

Robert M. Itami, Digital Land Systems Research
22 Dunstan Avenue, Brunswick, Victoria, Australia 3056
Email: dlsr@one.net.au
and
Senior Research Fellow, Department of Geomatics
University of Melbourne, Victoria, Australia 3010.
Email: b.itami@eng.unimelb.edu.au, Tel: +61-3-8344-4243, Fax: +61-3-9347-2916.

Glen S. MacLaren, Environmental Systems Solutions
Email: gmaclaren@goconnect.net
and
Research Fellow, Department of Geomatics
University of Melbourne, Victoria, Australia 3010.
Email: g.maclaren@eng.unimelb.edu.au, Tel: +61-3-8344-6773, Fax: +61-3-9347-2916.

Kathleen M. Hirst, Consultant, GIS Applications Pty Ltd.
PO Box 113, South Caulfield, Victoria AUSTRALIA 3162
Telephone: + 61 3 9523 6323
Facsimile: +61 3 9523 6323
Mobile: 0414 803 658
Postal Address: PO Box 113, South Caulfield, Victoria AUSTRALIA 3162
Email: khirst@gisapps.com.au Tel/Fax: + 61-3-9523-6323, Mobile: 0414 803 658

Robert J. Raulings, Director, eFirst
Level 1, Century House, 132 Gwynne St, Richmond, Vic, AUSTRALIA 3121
Telephone: +61 3 9427 9233
Email: rob@efirst.com.au

H. Randy Gimblett, Professor, School of Renewable Natural Resources
University of Arizona
BSE 325, Tucson, Arizona USA 85721
Email: gimblett@Ag.arizona.edu