4th International Conference on Integrating GIS and Environmental Modeling (GIS/EM4):
Problems, Prospects and Research Needs. Banff, Alberta, Canada, September 2 - 8, 2000.

Grand Canyon River Management

Simulating Rafting the Colorado River through Grand Canyon National Park Using Spatially Explicit Intelligent Agents

GIS/EM4 No. 56

H. Randy Gimblett
Terry C. Daniel
Catherine Roberts

Abstract

The goal of this paper is to discuss an effort to develop and test a statistical computer-implemented model for estimating the movement and interactions among river trips. The Grand Canyon River Trip Model combines spatially explicit intelligent agent modeling with Geographic Information Systems (GIS) in an interactive system to provide Grand Canyon National Park managers and their constituents with an effective decision support tool for representing and evaluating the distribution and volume of use along the river. The river trip simulation system and management implications will be discussed.

Keywords

Computer Simulation, Intelligent Agent Modeling, Geographic Information Systems, Human-Environment Interactions, Natural Resource Management, River Management

Introduction

The Grand Canyon is truly one of the wonders of the world. While millions of visitors travel to the Grand Canyon every year from all corners of the world, only few will venture into the canyon to explore it’s beauty. Even fewer will be provided with the opportunity to float the river and experience the canyon from a totally different perspective. Out of the over four million visitors to the Grand Canyon, only 20,000 get to experience the canyon on a float trip annually. Under current management about 300 commercial trips, with a maximum of 36 people per trip, carry a total of about 19,000 visitors annually (approximately 80% of total visitors). Over 250 of the 300 commercial trips use motorized watercraft, accounting for virtually all of the short trip (6 – 8 days) opportunities on the river. Trips are relatively expensive ($1500 - $1800 per person for a 7 day trip), they require a minimum of one week of recreational time, and reservations must be confirmed at least one year in advance. Still, spaces available on commercial trips are often taken within hours of being offered.

Privately organized trips are limited to 16 people per party, and account for about 3,000 visitors per year. These trips require considerable investment in equipment and a high level of watercraft expertise. They are typically longer than commercial trips and only very rarely employ motorized craft. Private trips also require a remarkable amount of advanced planning and commitment--the typical waiting time on the reservation list is currently over 15 years.

The challenge for the Park is to balance the public demand for recreation use on the river with mandated environmental protections and increasing calls for the river corridor to be managed as (and perhaps officially designated as) wilderness. Wilderness policies could prohibit the use of motorized watercraft, eliminating the shorter commercial trips that currently accommodate 70% of all visitors. Because longer trips require more time and are more costly, this policy would almost certainly limit the river trip experience to an even more exclusive set of the public. It is less clear whether pursuit of wilderness policies would result in substantial, or any, reductions from the current numbers of trips or total numbers of visitors (visitor-days). High demand insures that visitors displaced by the loss of shorter, motorized trips will be replaced by equal numbers willing to afford the longer trips. Restrictions on numbers of trips and/or total visitors would have to continue to be rationalized on the basis of environmental impacts and impacts on the quality of visitor experiences.

Current policies are based on the plausible assumption that encounters with other parties generally have a negative effect on quality of river trip experience, especially wilderness experience. The existing research base is not sufficient, however, to justify the specific quantitative limits/targets for encounters set out in some park management policies. Moreover, relating per-trip encounters to annual or seasonal limits on numbers and timing of river trips (launch schedules) and/or total numbers of visitors (or visitor-days) is at least problematic. Yet, the launch schedule has been, and remains the predominant means for managing impacts, both social and environmental, in the canyon.

Grand Canyon River Trip Simulator (GCRTS)

Development of the Grand Canyon River Trip Simulation system (GCRTS) was motivated by the need to better assess the effects of river trips on environmental and social conditions in the canyon and to enable better projections of the effects of changes in launch schedules and other management policies. GCRTS takes advantage of recent developments in simulation modeling that employ a number of different artificial intelligence approaches in combination with Geographic Information Systems (GIS). These systems are designed to address complex human-environment interactions, with special attention to spatial and temporal dynamics. Agent-oriented modeling has recently gained popularity for capturing behavioral conditions and allows for intercommunication among and within agents that coexist in an environment. Intelligent agents are being used by researchers to solve problems in complex domains where behavior evolves over time and individuals adapt to an ongoing set of changing environmental and social conditions. A growing number of studies have explored emergent behavior in individuals or societies using the power of GIS for representing the spatial worlds in which these societies reside and interact. A number of investigators (eg. Kohler et al. 1996; Westervelt 2000; Bishop and Gimblett 1998; Gimblett et al. 2000a; Box 2000) have taken advantage of the spatially-explicit IBMs, intelligent agents and GIS, each using various techniques to link simulation models and GIS.

The Grand Canyon River Trip Simulator was developed as a special application of a GIS/intelligent agent modeling system developed by (Itami 2000; Itami and Gimblett 2000). The Recreation Behavior Simulator (RBSim) was a specific application of this more general system designed to simulate behavior of human recreators in high use natural environments (Gimblett 1998). RBSim was used as the foundation for the current river trip modeling and visualization system. RBSim is programmed using object-oriented classes, properties and object structure.

Development of the system began with a basic conceptualization of the required inputs, outputs and processes to be represented. An intensive data collection phase collected detailed trip diaries/itineraries for over three hundred trips. A number of commercial outfitters also provided trip summaries extending over several years. While substantial, this empirical database was not adequate for complete specification of the complex processes that affect the hundreds of trips that interact on the river. Empirical data from the trip reports was augmented by interviews with experienced river guides, trip outfitters and private boaters. These interviews outlined typical trip “profiles” and identified a number of general “rules” or guidelines that affect trip progress. Statistical analyses of the trip report database and interpretation of the expert interviews guided the development of the basic river trip model, which employs an intelligent agent architecture (Bieri & Roberts 2000; Gimblett et al. 2000b). This data, including the rule were utilized as input into the simulation model.

The model has been implemented in the form of an interactive computer system accessed through a graphical/windows interface. Support is provided for many types of queries and reports primarily designed for park managers, but also useful for trip outfitters and for individual trip planning. Many of the features of the system and its development and implementation have been described previously (e.g., Gimblett et al 2000b; Roberts & Gimblett 2000).

The Simulation system

The management requirements of the Park dictated much of the conceptualization of river trips that underlies the GCRTS. Inputs to the system were largely defined by and limited to those conditions over which managers have some control, and/or about which they have some means of knowing. These factors are largely operationalized by the launch schedule, which specifies which trips can be launched on what days, the take-out day and location, and any scheduled passenger exchanges along the way. Various regulations were also considered such as the maximum distances trips are allowed to travel within a day, camping and layover restrictions, and areas in the canyon where stopping is prohibited. Beyond these few restrictions, trips are free to individually decide their progress—where to stop, for how long and how far to float on the river on a given day. Management requirements also dictated the representation of encounters between trips as a central output of the model, including where encounters occur, between what parties (what types of trips) and the nature of the contact (on-river, off-river, visual only). Encounters with other trips is also one of the key variables affecting trip progress, as trips respond to the presence of other parties at attractions and especially at campsites.

The above considerations dictated the “trip” as the primary modeled unit. Because many trips are composed of more than one watercraft, the model actually projects the locations of idealized “trip centroids.” There are restrictions regarding the extent to which the individual watercraft associated with a given trip can disperse on the river, and all members of a party are required to camp together each night. So this simplification was not deemed unreasonable, but it may result in some estimation errors. For example, within the model “encounters” must be defined by relations between trip centroids, rather than between individual watercraft.

The GCTRS is composed of two interacting but standalone modules. The RiverTrip Data Base component uses an ACCESS database to collect, analyze and report on parameters of actual and/or simulated river trips. This component contains a set of algorithms for querying individual or multiple trips to obtain summary statistics and to make comparisons among actual and/or simulated trips. This provides the quantitative data for the validation of the simulation model, as well as providing important insights into the characteristics of actual and simulated trips.

The second major module is a simulation system, programmed using object-oriented classes, properties and object structure. GCRTS is comprised of sets of interconnected agent classes designed to simulate trip behavior and represent the associated river environment. These agent classes are programmed to interface with the RiverTrip database, reading and writing information into a variety of formats. The Trip Class stores and passes information to the simulation environment that describes attributes of river trips such as length of trip, trip id, name of outfitter, number of boats, type of boats, boatman, number of passengers, number of crew etc. This class was designed to provide maximum flexibility in creating agents to represent actual commercial and private trips that leave Lees Ferry and travel along the Colorado River thru the Grand Canyon National Park.

The RiverClass stores information about the physical (camp and attraction) sites along the river. Beach locations are stored by name and associated location along the Colorado River expressed as river mile starting from Lees Ferry - Mile 0 and ending down river with take outs at Diamond Creek – Mile 225.7 and Lake Mead Takeout - mile 279. Information is read from a GIS database consists of geographic features rasterized into 90-meter cells. This data structure allows the decision maker to accurately represent geographic features of the river environment and to easily update features as changes occur (e.g., eroding beaches, varying water levels, addition or deletion of attractions or campsites, river flow rates). This class stores attributes such as river mile, beach locations (river right or river left), carrying capacities, attractiveness of camps and attraction sites (based on historic data), and number of people currently occupying a beach. In addition this class keeps track of numbers of trips on the river and computes the time it takes a trip to float through a 90-meter cell.

The Encounter Manager class keeps track of the number of encounters each trip has with other trips on and off of the river. This class monitors numbers and types of encounters that are occurring on a cell/cell basis, and writes data out to a database. Encounters are time marked and stored for each trip, for each cell and for each camp/attraction site.

The Visualization class provides a way to view and control elements of simulations on the river. GIS data, shaded relief maps and dynamic pan and zoom utilities display the progress of trips as they move down the river. In addition to watching the movement and interaction of trips on the river, the user can query individual trips and beaches/attraction sites at any time in the simulation to obtain information about current locations, conditions, and cumulated numbers of encounters.

The Simulation class is the heart of the simulator. It provides the interface for running the simulation, controlling trip movement, reading trip and launch schedules and it controls an enormous amount of file management tasks. Trip simulation outcomes are stored in the RiverTrip Data Base for subsequent analysis.

The software “engine” that was developed simulates raft-parties (trips) moving down river. The launch schedule is the primary mechanism in the simulation for directing how boats will travel down the river. The 1998 launch schedule was provided to the research team by the National Park Service). The 277 mile stretch of the Colorado River that is used for Grand Canyon river trips is divided into 90-meter “cells” and each trip is dealt with on a cell by cell basis. The underlying structure of the engine is the use of a priority time structure. This structure permits attention to only one trip object at a time and prioritizes the trips to be handled next. This one-by-one analysis is dependent on the time at which each trip will next enter a new cell in the river. The trip that is at the top of the priority stack determines, using artificial intelligence algorithms to be described later, what action to take (to stop or not and, if so, for how long). It then returns control to the engine. As each trip passes or stops at an identified site, the information is logged for subsequent analysis and display.

Geo-temporal scale

Determining the geographic resolution of the model was a challenge. On the one hand, modeling 226 river miles favored a relatively coarse geographic scale. On the other hand, to achieve reasonable accuracy in determining encounters between trips, including visual encounters, a relatively fine scale was required. As the river winds and turns, contact between two parties can depend upon tens of meters of separation. The GCRTS divides the 226-mile (363,702-meter) river into over 4,000 90-meter cells, and each trip is modeled as passing into and out of these 90-meter segments of the river. Similarly, riverside attractions and campsites are assigned to the nearest cell(s), with the addition of a river-right, river-left designation where appropriate.

Time presented similar scope and resolution issues. The model needed to accommodate trips from 6 to 30 days in duration, occurring over the 11 months of the primary and secondary use seasons. At the same time, encounters have typically been defined in terms of intervals as short as 5 minutes. When two trips make contact, an encounter is recorded. If the trips separate for 5-minutes or more and then again make contact, this is recorded as a second encounter. Further, watercraft could pass through individual 90-meter river cells in a matter of a few minutes. Thus, the time step for the model was set to handle minute-to-minute calculations of trip locations. The GCRTS, then, represents a 7-day trip as 10,080 minutes.

Modeling trip progress

The trip diary data and the expert profiles were combined to define “plans” for the different trip types in the model. The applicable trip-type plan defined the “initial conditions” for individual trips. That is, to model a particular trip (e.g., a 12-day commercially outfitted oar trip) the trip is first assigned the appropriate type-plan. Each trip is assumed to follow the assigned plan (with some random variations in parameters such as start times, float speed, time into overnight camp, etc) until an event/situation is encountered that triggers one of the rules (e.g., the campsite specified in the plan is reached and found to be occupied by another modeled trip). The applicable rule is then applied (“fired”) to affect an appropriate deviation from the plan.

The autonomous agent modeling architecture represents each trip as an “agent” (object) that responds (e.g., stop, slow down, speed up) according to an assigned set of rules. Rules are based on time relative to the assigned trip plan, to features of the river environment (campsites, attractions) and to features of other trips that are encountered. The criteria on which trip decision rules are based are hierarchically organized. Behavior of the trip agent depends upon how given temporal, environmental, and social conditions relate to the ordered set of decision criteria

Trips proceed down the river encountering potential campsites, attractions and other trips, evaluating the relevant criteria to arrive at decisions that probabilistically determine whether the trip stops or extends time on the river to make more progress. Actions affect the progress of the trip, moving it away from and then back toward the overall “plan” assigned to that trip. Decisions made up stream concatenate to affect conditions/criterion values down stream, which in turn affects down-stream decisions and actions. As the end of the day (the time for getting into a campsite for the night) or the end of the trip (the designated take-out day) approaches, the opportunities for stops at attractions become more constrained.

A Example Simulation Run

While there are many ways to use GCRTS as a management tool, pertinent to this paper and in particular this conference is to examine the effects of modifying launch schedules on encounters along the river using simulation. To demonstrate the use of GCRTS the following scenarios have been developed for the river. Two simulation runs were undertaken to examine encounters by river mile using the 1998 launch schedule. Simulation runs consisted of all commercial 8-day trips launching between July 1, 1998 and August 31, 1998. These time periods were selected because they represent primary, peak season use. Two simulations were constructed, one where launch restriction were imposed and the other without. An imposed launch restriction is one where trips must launch in the early morning or late afternoon. Launches with no restrictions impose no constraints on trips allowing them to leave, camp and undertake other activities according to their planned itinerary or whenever they choose on that day. Figure 1 illustrates a plot of both simulation runs showing what would happen on the river with no intervention and management of the resource versus some intervention. The y-axis represents the number of river contacts, while the x-axis represents the spatial location along the river specified by river mile where the contact occurred. The solid lines reveal heavy numbers of encounters along the river illustrating what would happen on the river with no intervention and management of the resource. It is clear that there are consistently more that forty river contacts at certain places along the river that well exceeds the management standard. South Canyon, LCR, Phantom Ranch, Deer Creek and Diamond Creek areas are some of those noted for contacts exceeding standards.

Figure 1 – Comparison between Restrictions (heavy dashed lines) and No Restrictions (thin, solid lines) to Launch Schedule

The heavy dashed lines are the same simulation conditions except there have been launch restrictions imposed on the trips. It is quite evident from figure 1 that contact levels at South Canyon, LCR, Phantom Ranch, Deer Creek and Diamond Creek noted as ‘hotspots’ of high contact levels have been dramatically decreased. By setting constraints on the launch schedule the simulated trips meeting management objectives have increased by 15% and average daily contacts have been reduced by 50% without setting any other constraints on camp or attraction sites. If our objective were to reduce river contacts and keep within management objectives this option would look more appealing. While this is a simple application of the model, it does illustrate how powerful simulation can be in developing, running and test a variety of scenarios to management human/environment interactions.

State of the System

The Grand Canyon River Trip Simulation system is currently functional, but not yet adequately tested. Of particular concern is the lack of sufficient tests to confirm the veracity of river trip projections provided by the model. Caution is particularly required when launch schedules and other specifications of input conditions differ from those in force when the 1998-1999 trip itinerary data were collected. Still, the system does provide a considerable advance in river management decision support for the Park, as well as confirming the benefits of the autonomous agent approach for modeling complex social-environmental interactions in wilderness recreation settings.

Conclusions

This paper has briefly outlined some initial attempts at integrating multi-intelligent agents and rule-based decision-making algorithms within a GIS-based environment to model complex human-environment interactions. Multi-agent models have many advantages compared with previously applied modeling approaches. Using agents to represent individuals or parties, incorporating GIS to represent the environment and utilizing intelligent agent technology in modeling of river management issues provides a number of important advantages. Agents used to represent trips can be programmed with strategies, goals, intentions and negotiation strategies that learn about and adaptive to their surroundings and others they encounter. Using GIS to represent a geo-referenced environment provides the manager with an effective way to view agent interactions and assess the number of encounters and where they occur. In addition a GIS approach incorporating visualization makes the simulation model easy for policy-makers, planners, managers and the public to understand and generate and test a variety of alternatives. Using agent technology as a visitor management tool allows decision makers to develop and test “what if” scenarios and explore options that will guide management decisions in resolving recreation use interactions, aid in establishing limits of use and standards and evaluate alternatives in terms consequences of policy actions and social, environmental and economic impacts.

More important and pertinent to this special issue is the notion of visual encounters. While there is much controversy in literature as to whether excessive human encounter levels in recreation settings have adverse effects on recreation experience, this research provides a tool to begin to examine this issue. The GCRTS provides a mechanism to identify and examine where excessive levels of encounters occur in space and time, as well as provide ways to evaluate whether these encounter levels currently meet management standards and alternatively a way to alter and test new, perhaps more realistic standards. More importantly, GCRTS provides a way for managers to utilize the management mechanisms that they have control of such as launch schedules, user-day quotas, permits and site restrictions to reduce encounter levels and minimize social and ecological impacts. The experiments discussed in this paper clearly illustrate how important these management mechanisms can be in minimizing encounter levels. Above all, these simulation experiments illustrate how simple management decisions or rules applied to a situation can have severe cascading effects downstream. Encounters levels are an important example of this effect and are still a powerful indicator of visitor use patterns and experience quality in recreation settings.

Finally this paper has described the current stage of development of an agent-based simulation system for modeling complex human-environment interactions in the context of river rafting trips. Such a system, and information about how to construct such a system could contribute to the solution of a wide range of scientific and practical problems raised in human-environment interactions.

Acknowledgements

The authors are grateful to the many members of the Grand Canyon Outfitters and the Private Boaters Associations who provided valuable data and expert advise to support the modeling effort. The contributions of our collaborators Susan Cherry, Dana Kilbourne and Michael Meitner (University of Arizona); Michael Ratliff, Doug Stallman, Ryan Bogle, Joanna Bieri and Gary O’Brian (Northern Arizona University); and Linda Jalbert, Jennifer Burns, and Lauri Domler (Grand Canyon National Park) are also very gratefully acknowledged. This work was supported in part by a cooperative research agreement between the University of Arizona and the National Park Service (1443-CA8601-97-006) and by McIntire Stennis project ARZT-139238-M-12-142.

References Used

Bieri, J. and Roberts, C. 2000. Using the Grand Canyon River Trip Simulator to Test New Launch Schedules on the Colorado River. Association for Women in Science Newsletter, August.

Bishop, I. D. and Gimblett, H.R. 1998. Modelling Tourist Behaviour: Geographic Information Systems and Autonomous Agents. 1st International Scientific Congress on Tourism and Culture for Sustainable Development. Athens ’98. May 19-21. Athens, Greece.

Box, P. 2000. Spatial Units as Agents: Making the Landscape an Equal Player in Agent-Based Simulations. In H. R Gimblett, editor. Integrating GIS and Agent based modeling techniques for Understanding Social and Ecological Processes (spatially-explicit simulation modeling). Oxford Press (In Press).

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., Itami, R.M. and Richards, M. 2000a. Simulating Wildland Recreation Use and Conflicting Spatial Interactions using Rule-Driven Intelligent Agents. In H. R Gimblett, editor. Integrating GIS and Agent based modeling techniques for Understanding Social and Ecological Processes (spatially-explicit simulation modeling). Oxford Press (Forthcoming).

Gimblett, H.R., Roberts, C., Daniel, T. C., Ratliff, M., Meitner, M. J., Cherry, S., Stallman, D., Bogle, R., Kilbourne, D., Bieri, J. and O’Brien. G. 2000b. 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.

Itami, R. M. 2000. Mobile Agents with Spatial Intelligence Intelligent. In H. R Gimblett, editor. Integrating GIS and Agent based modeling techniques for Understanding Social and Ecological Processes. Santa Fe Institute/Oxford Press. 1999. (Forthcoming).

Itami, R. M. and Gimblett, H. R. 2000. Intelligent Recreation Agents in a Virtual GIS World. Complex Systems 2000 Conference. University of Otago, Dunedin, New Zealand. November 19-22, 2000.

Kohler, T. A., Van West, C. R., Carr, E.P. and Langton, C G. 1996. Agent-based Modeling of Preshistoric Settlement Systems in the Northern American Southwest. Third International Conference Integrating GIS and Environmental Modeling. January 21-25. Santa Fe, NM.

Roberts, C. and Gimblett, H. R. 2000. Computer Simulation for Rafting Traffic on the Colorado River. Proceedings of 5th Biennial Conference of Research on the Colorado Plateau, US Dept. Interior, U.S. Geological Survey.

Westervelt, J. 2000. GIS and Agent Based Modeling. . In H. R Gimblett, editor. Integrating GIS and Agent based modeling techniques for Understanding Social and Ecological Processes. Santa Fe Institute/Oxford Press. 1999. (Forthcoming).

Authors

Randy Gimblett, Professor, School of Renewable Natural Resources.
The University of Arizona. Tucson, AZ 85721.
Email: Gimblett@ag.arizona.edu, Tel: 520-621-6360, Fax: 520-621-8801

Terry Daniel, Professor, Department of Psychology & School of Renewable Natural Resources.
The University of Arizona. Tucson, AZ 85721.
Email: tdaniel@u.arizona.edu, Tel: 520-621-7453, Fax: 520-621-8801

Catherine A. Roberts, Associate Professor, Department of Mathematics and Statistics and Computer Science
Northern Arizona University, Flagstaff, AZ 86011-5717.
Email: Catherine.Roberts@NAU.EDU, Tel: 520-523-6884, Fax: 520-523-5847.