Chapter 4: Case Studies of Simulation Models of Recreation Use · 2008. 5. 8. · Chapter 4: Case Studies of Simulation Models of Recreation Use David N. Cole used to monitor crowding-related

USDA Forest Service Gen. Tech. Rep. RMRS-GTR-143. 2005 17

Computer simulation models can be usefully ap-plied to many different outdoor recreation situations.Model outputs can also be used for a wide variety ofplanning and management purposes. The intent ofthis chapter is to use a collection of 12 case studies toillustrate how simulation models have been used in awide range of recreation situations and for diverseplanning and management applications.

The types of recreation situations included in thesecase studies vary in the size and remoteness of thearea being modeled, as well as in the type of recreation.Case studies include:

• large backcountry areas: Bighorn Crags in theFrank Church River of No Return Wilderness,Misty Fjords National Monument, ColoradoRiver through Grand Canyon National Park,Humphrey’s Basin in the John Muir Wilder-ness, and Isle Royale National Park;

• smaller frontcountry areas: Yosemite Valleyin Yosemite National Park, Alcatraz Island inGolden Gate National Recreation Area, ArchesNational Park, Acadia National Park, Para-dise Meadows in Mount Rainier National Park,and the Twelve Apostles in Port CampbellNational Park, Australia;

• overnight hikers: Bighorn Crags, Humphrey’sBasin, and Isle Royale;

• day hikers on trails: Yosemite, Arches, Acadia,and Mount Rainier;

• visitors to facilities: Alcatraz Island and theTwelve Apostles;

• bicyclers: Acadia National Park;• whitewater rafters: Grand Canyon National Park;• visitors on cruise ships and ocean kayakers:

Misty Fjords;• visitors in automobiles: Acadia.

Perhaps the most basic use of computer simulationmodeling is as a tool for describing current use pat-terns. The purpose of the first four case studies (Big-horn Crags, Misty Fjords, Grand Canyon, and JohnMuir) is to illustrate this use. Simulation can also be

Chapter 4:Case Studies of SimulationModels of Recreation UseDavid N. Cole

used to monitor crowding-related indicators, such asnumber of encounters, persons-at-one-time or per-sons-per-viewscape, either to describe the currentsituation or to determine whether standards for suchindicators are being violated. The Bighorn Crags andJohn Muir case studies illustrate the estimation ofencounter rates, while persons-at-one-time or persons-per-viewscape measures are estimated in the Yosemite,Alcatraz Island, Arches and Acadia case studies.

Computer models can also be used for predictivepurposes. In the Isle Royale case study, simulationwas used to help planners identify alternative man-agement actions that would be effective in reducingcampsite sharing. This was accomplished by predict-ing the effects on campsite sharing of reductions inamount of use, changes in the spatial and temporaldistribution of use and increases in behavioral restric-tions and number of facilities. At Yosemite, AlcatrazIsland, and Arches, simulation was used to predict themaximum amount of use that can be sustained withoutviolating crowding-related standards. At Alcatraz andArches, the effect of alternative public transportationsystems on maximum allowable use was also predicted.

Simulation can also be used to prepare for thefuture. In the Acadia and Twelve Apostles case stud-ies, use is steadily increasing, and future use levels areforecast to be much higher than they are presently. Inthese case studies, simulation models are used topredict the effect of increased use on crowding-relatedvariables. In the Twelve Apostles case study, theeffects of changes in park infrastructure are alsopredicted. Finally, the Mt. Rainier case study showsan innovative way to collect data on visitors in achallenging situation.

The case studies included in this chapter suggestthat there is reason to be enthusiastic about thepotential of computer simulation modeling as a visitormanagement tool. However, this enthusiasm must betempered with appropriate realism and caution. Theapplication of simulation to outdoor recreation issuesis still in its infancy, and there is much need for morelearning and development. As with so many things,

18 USDA Forest Service Gen. Tech. Rep. RMRS-GTR-143. 2005

the more we learn, the more we recognize the need toknow even more. The models presented in the casestudies have only been partially validated. Simulationoutputs have not been statistically compared to real-world observations. Numerical estimates and predic-tions are presented without numerical estimates ofhow much confidence one should have in these metrics.More work is needed to ensure that the models devel-oped are as valid as possible, simulation runs areconducted correctly and outputs are appropriatelyinterpreted. However, these case studies demonstratethe useful outputs that valid models can produce.

Recreation Visitation and Impacts inthe Bighorn Crags Portion of theFrank Church—River of No ReturnWilderness _____________________

Randy GimblettSuzanne CableDavid N. ColeRobert M. Itami

Purpose

This case study demonstrates the use of agent-basedmodeling and simulation to describe recreation usepatterns in a popular portion of the Frank Church—River of No Return Wilderness in central Idaho. Thecase study will demonstrate the data collection meth-ods, modeling, and simulation of backpackers and rec-reational stock users on multiple day trips. Particularattention is given to estimating encounter rates in theinterior of this wilderness because encounter rates canaffect the experience of wilderness visitors. Conse-quently, wilderness managers commonly want to moni-tor encounter rates and frequently develop standardsfor maximum acceptable number of encounters.

Study Area

The Frank Church—River of No Return WildernessArea is the largest contiguous wilderness area in theUnited States, outside of Alaska. The most popularportion of this wilderness for backpackers is an areaknown as the Bighorn Crags. The area also receivessubstantial use by groups traveling with pack and saddlestock. Rugged and remote, this country offers adventure,solitude, and breathtaking scenery. Like other popularwilderness areas, physical impacts from dispersed visi-tor use are evident throughout the area, and socialimpacts to visitor experiences are likely, but currentlynot well documented. To develop the information neededto better manage recreation in the Bighorn Crags, weconducted inventories of all recreation impacts in the

Crags (official trails, user-built trails, and campsites),and collected the data needed to build a computer simu-lation of the distribution of recreation use.

Data Collection Procedures

To build the computer simulation, data were col-lected on both visitor demographics and site charac-teristics. Site characteristics include a map of travelnetworks and popular destinations.

Visitor Characteristics—Trip diaries were used tocollect the data on visitor itineraries needed to con-struct the simulation. Visitors were asked to take adiary with them and record on one side of the diary theirgroup size, mode of travel, and date and time of entryand exit. They were also asked a series of attitudinalquestions about trail, campsite, and management con-ditions. On the other side of the diary, a map wasprovided so a group could record route information.Specific instructions for the map were as follows:

Please indicate on the map where you camp, thenumber and type of encounter(s) you have and a notationat the edge of the map anytime you leave and re-enter thewilderness area. Please locate each of the campsites youvisit as accurately as possible on the map. Place a ‘C’beside the campsite and a number that indicates thenight of the trip that you camped at that location (Ex-ample C2 denotes the place where you camped on the 2ndnight of trip). In addition to camp locations we would likeyou to indicate the number and type of encounters youhave with other parties throughout your trip. Place an ‘E’to mark any encounters you have along the trail as theyoccur or while at camp at the end of the day. Associatedwith the ‘E’, provide one or more of the following notationsto denote the type of encounter(s) you had, ‘O’ (OtherParty Camping), ‘P’ (Packstock) or ‘B’ (Other Backpacker)followed by a number which indicates the number ofpeople in the group encountered. (Example EP10 meansencounter with a packstock group of 10 people).

To create trip itineraries in the format required byRBSim, all spatial and relational data from the diarywere entered into an Access database via a Web-basedinterface developed specifically to enter both types ofdata. Figure 1 provides an example of this interface andone of the trips that was entered into the database. Thiswas a 4-day trip into the area. C1, C2, and C3 were thethree locations where the group camped, and DH repre-sents day hikes taken from each of the campsites. Torepresent this trip in RBSim, data were transformedinto a sequence of travel routes and destinations (ge-nerically referred to as links and nodes). The sequencewas determined by the number on the left side of thedestination notation. For example, the trip in figure 1camped the first night at Harbor Lake. The next daybegan with a day hike to Bird Bill Lake, followed by abackpack to and camping at Sky High Lake. The nextday began with a day hike to Terrace Lakes, followed bya backpack to and camping the third night at ReflectionLake. The fourth day began with a day hike to Lost Lake


Figure 1—Interface for capturing spatial data about the trip.

and then a return to the trail head. No other visitorswere encountered on the trip. This sequence of links,nodes, and durations at campsites are conformed into adata sequence that RBSim reads and uses to replicatethis trip in the trace simulation.

Site Characteristics—Global Positioning System(GPS) tracks were used to inventory trails and camp-sites to create a travel route and destination network.All trails in the study area were inventoried during thesummer of 2003. Link data were then converted intoArcView format, and then converted into a format thatconforms to standards established for RBSim. In somecases, it was clear that recreationists traveled offtrailbetween destinations. These informal routes weredigitized and represented as part of the trail network.

Campsite inventory data were also collected in thefield. This consisted of creating a photographic recordof the sites, using a GPS to locate recreation sites, andtaking quantitative measures of variables such as soilerosion, vegetative cover, and amount and type ofdisturbance (Cole 1989). Similar to the trail data,campsite and trailhead nodes were inserted into theArcView map. Since most of these campsites weresituated beside lakes, a single node was establishedfor each lake, and all campsites at the lake wereassociated with that node. Consequently, campsite

statistics are provided for lakes rather than individualcampsites. We were not confident that visitors couldidentify the exact campsite they used at a lake.

Modeling Characteristics

RBSim, a recreation behavior simulator (Gimblettand others 2001) was used to describe patterns ofvisitor use across the landscape. The modeling ap-proach used in this study is a trace simulation usingbaseline fixed itineraries derived from the trip diaries.A typical trip for the Bighorn Crags simulation isdescribed by an entry node, an exit node to the net-work, an arrival curve, a probability distribution ofagent types, a list of destinations, and trip duration.On execution, the simulator reads all the trip itinerar-ies, schedules all trips, and then an agent representingthe type of trip moves from one destination to the nextacross the travel network using the shortest traveltime between the two points. A large set of informationfor each agent is collected and stored in an Accessdatabase for later processing.

Simulation Outputs

Simulation output can be used to describe use levelson trails and at lakes as well as encounter levels on


trails and at camping areas around lakes. There isconsiderable interest in the estimation of encounterlevels because they are a common indicator and aredifficult to measure in the field. To derive encountersalong trails and at campsite destinations, a simulationwas constructed. The simulation was run for a total of89 days, from July 3 through September 29, 2003. Atotal of 75 trips through the Bighorn Crags weresimulated. Average trip length was 3.6 days. Visitorstypically arrived on Friday or Saturday and departedthe wilderness on Sunday or Monday. Peak use oc-curred in late July and early August. Average groupsize was 2.8. Model outputs were visitor use levels andencounters by trail segment and camping area (lake).Statistics are the mean of 10 replications of the simu-lation. Trail encounters varied substantially betweenreplications, while trail use, camp area use, and camp-site encounters did not.

Trail Use—Table 1 shows simulation output forthree different trail segments. Total use is indicatedby the total number of groups that traversed the trailsegment during the simulated sampling period. Themost heavily used trail segments are from the trail-head at Bighorn Crags campground to the trail inter-section that branches off to Welcome and WilsonLakes. Encounters are recorded when one group over-takes another group in the simulation or where twogroups pass each other in opposite directions. Table 1shows the total number of encounters that occurredduring the 89 day season, the mean number of encoun-ters per day, as well as the number of days on whichencounters occurred.

There is not a simple linear relationship betweenuse levels and encounter levels. Nor is there a linearrelationship between the average number of encoun-ters and the days on which encounters occur. Thisillustrates the value of the computer simulations.Segments 64 and 66 have similar encounter levels, butboth use levels and the number of days with encoun-ters are much higher for segment 64. Trail segment 64is much shorter than segment 66. Apparently, as thelength of trail segments decreases, so does the ratiobetween number of encounters and amount of use.Figure 2 displays the distribution of trail encountersacross the Bighorn Crags. The maximum mean dailyencounter level was just 0.51 encounter per day, andon many of the trail segments, no encounters occurred.

Table 1—Simulation output on use and encounter levels for selected trail segments.

Trail Total Total Encounters Days withsegment groups encounters per day encounters

61 2 0 0 064 104 7 0.08 5.566 26 6 0.07 1.4

This suggests that crowding on trails is not a seriousissue in the Bighorn Crags.

Campsite Use

Table 2 provides information about use levels andencounters between groups at the more popular camp-ing areas in the Bighorn Crags. Camping areas areentire lakes or other destinations. Ship Island andWilson Lakes are the most popular camping destina-tions in the Bighorn Crags (figure 3). Someone wascamping at these lakes about 30 nights during the 89-day season. By dividing nights occupied by 89, one canderive the likelihood of camping with at least one othergroup at each camping area.

The simulation also generates output on the totalnumber of groups that camped in each area over theseason. Dividing total number of groups by 89 gener-ates an estimate of mean groups per night (table 2). Ifmore than one group is camped in a camping area onthe same night, campsite encounters occur. All groupscamped in the same area were assumed to encounterall other groups in that area. Encounters are derivedfrom simulation output as the difference between thetotal number of groups that camp in the area and thenumber of nights that anyone camps in the area. Whendivided by 89, an estimate of mean campsite encoun-ters per night is generated (table 2).

Conclusion

This case study illustrates how a computer simula-tion model can be used to describe the distribution ofuse across a large complex of backcountry trails andcampsites. It can generate estimates of encounterlevels, useful indicators of potential crowding prob-lems that are costly to monitor directly. Simulationoutput suggests that crowding in the Bighorn Crags istypically not a serious problem at present.

References

Cole, D. N. 1989. Wilderness campsite monitoring methods: asourcebook. Gen. Tech. Rep. INT-259. Ogden, UT: U.S. Depart-ment of Agriculture, Forest Service, Intermountain ResearchStation. 57 p.

Gimblett, H. R.; Richards, M. T.; Itami, R. M. 2001. RBSim: Geo-graphic simulation of wilderness recreation behavior. Journal ofForestry. 99(4): 36–42.


Figure 2—Mean trail encounters per day inthe Bighorn Crags over an 89-day season.

Table 2—Simulation output on use and encounters at camping areas.

Nights Likelihood of a Mean groups Mean encountersCamp area occupied camp encounter per night per night

Ship Island Lake 31 0.35 0.54 0.19Wilson Lake 29 0.33 0.39 0.07Airplane Lake 27 0.30 0.30 0Welcome Lake 24 0.27 0.33 0.05Reflection Lake 23 0.26 0.33 0.07Cathedral Lake 18 0.20 0.28 0.08Birdbill Lake 13 0.15 0.18 0.03Terrace Lakes 11 0.12 0.15 0.02Barking Fox Lake 9 0.10 0.11 0.01Sky High Lake 9 0.10 0.12 0.02Heart Lake 8 0.09 0.09 0Wilson Creek—offtrail 6 0.07 0.07 0South of Cathedral Rock 3 0.03 0.03 0Cathedral Rock 3 0.03 0.03 0Gentian Lake 3 0.03 0.03 0Buck Lake 3 0.03 0.03 0Ramshorn Lake 2 0.02 0.02 0Wilson Creek west of Buck Lake 2 0.02 0.02 0


Recreation Visitation in Misty FjordsNational Monument in the TongassNational Forest _________________

Randy GimblettRobert M. ItamiSuzanne Cable

Purpose

This case study illustrates the utility of agent-basedmodeling and simulation in describing recreationaluse patterns at Misty Fjords National Monument inthe Tongass National Forest. The case study demon-strates data collection, modeling, and simulation ofmultiple travel modes including arrivals and depar-ture of helicopters, floatplanes, and cruise ships onmultiday trips. Particular attention was given to as-sessing potential impacts on wildlife at certain criticallocations along the coastline.

Figure 3—Nights per year of camping at lakesin the Bighorn Crags, over an 89-day season.

Study Area

Located in southeast Alaska, Misty Fjords NationalMonument is 35 km east of Ketchikan and about 1,100air km from Seattle. Created in 1978, by PresidentialProclamation, the Monument encompasses about920,000 ha within the Tongass National Forest. In1980, Congress passed the Alaska National InterestLands Conservation Act (ANILCA) and designated allbut 60,000 ha as Wilderness. Monument status pro-tects ecological, cultural, geological, historical,prehistorical, scientific, and wilderness values. Remoteand wild, the Monument is a primarily mountainous,nearly untouched coastline, characterized by deep salt-water fjords, and is home to mountain goats, brown andblack bears, and a host of fish and marine mammals.

The Behm Canal, a deep, long waterway of thenortheastern Pacific Ocean, leads to the heart of theMonument. Places such as Walker Cove and RudyerdBay, characterized by rock walls jutting 1,000 m abovethe ocean, provide flightseers, cruise ship passengers,


boaters, and hikers with photographic opportunities.Commercial services are allowed, under permit, toprovide basic public services in keeping with Monu-ment designations. The Forest Service and the State ofAlaska manage fish and wildlife habitat coopera-tively. Very little recreation use actually occurs on theland due to the steep, inaccessible terrain, but sincethere is limited regulation of air or water, encroach-ment of tourists from anchored ships and commercialflights on wildlife is of serious concern.


Data were collected on both visitor demographicsand site characteristics. Site characteristics include amap of travel networks and popular destinations, aswell as those areas wildlife biologists understand to berich estuaries that provide needed forage and protec-tion during particular times of the year for wildlifepopulations.

Visitor Characteristics

Trip itineraries for the simulation were constructedfrom visitor data acquired from a number of sourcesfor the 2000 tourist season. These included recordsfrom cruise ship tours, commercial fishing and hunt-ing records, harbor master records of water-basedtravel of large cruise ships entering the harbor, and inparticular, privately owned recreational luxury boatsand sailboats. In addition, records were obtained forfloatplane tours and a range of administrative tripsregularly taken by the Forest Service. Informationcollected included the type of activity, size of water-craft, name of company, number of visitors on tour,duration of visit, time of departure from origin of trip,and time of arrival at selected destinations. Typicalboat, helicopter, and floatplane speeds and commonflying altitudes were also collected. This informationprovided enough background to characterize the typeand frequency of trips as well as common destinations.

While trip logs are an excellent source of data tocharacterize the trip, many routes can be taken to anydestination. So in addition to trip itinerary data,travel companies and commercial floatplane opera-tors were asked to map their preferred and typicaltravel routes, stopovers at lakes, and any other infor-mation that could be used to accurately describe thepattern of travel. Anecdotal information was alsocollected on events that might cause variations intypical travel routes. For example, travel patterns inAlaska are frequently dependent on weather.

Site Characteristics

All the mapped travel route data were entered intoa Geographic Information System (GIS) and used to

construct a network of all possible travel routes. Anetwork was created for water travel (including cruiseships, privately owned recreational fishing and luxuryboats, sailboats, administrative water craft), floatplanes(commercial and administrative; fig. 4), and helicopters(commercial and administrative). These networks in-cluded all major destinations identified in the trip logsand in interviews with tour operators.


RBSim, a recreation behavior simulator (Gimblettand others 2001) was used to examine the seasonalpatterns of visitation that occur in Misty Fjords Na-tional Monument. The modeling approach used in thisstudy is a trace simulation using baseline fixed itiner-aries. Central to the trip planning behavior of a recre-ation agent is the typical trip. The concept of a typicaltrip is based on the premise that visitors have commonpatterns of use. For example, day use visitors arrivingduring weekdays will have a different arrival pattern,a different duration of stay, and perhaps a differentpattern of destinations than a traveler arriving on aweekend or an overnight visitor.

A typical simulated trip has a global trip plan, whichspecifies an intended list of destinations the agentwishes to visit. A global trip plan consists of total tripduration, an entry node, an arrival time, an exit nodeand a sequence of intermediate nodes (destinations)between the entry and exit nodes. Each destinationnode has a visit duration.

The simulator reads all the trip itineraries andschedules all trips, then moves each agent from onedestination to the next across the travel network usingshortest travel time between the two points. Informa-tion for each agent is collected and stored in a databasefor processing.

Simulation Outputs and ManagementRecommendations

A total of 2,149 trips were entered into the simula-tion for the 2000 visitor use season for Misty FjordsNational Monument. The simulator produced outputon encounter levels for all destinations on the net-work. Encounters are the number of other agentsvisible within a specified radius of the destinationcategorized by craft type. Figure 5 shows total numberof encounters at the 12 most frequently visited desti-nations. A number of destinations received very littleuse, while others would be considered to be high usedestinations. Rudyerd Bay and associated dockingfacilities (1,301 combined encounters) are the mostpopular destinations in the Monument. Visitationpeaks in August, generating the greatest potential forconflict between different modes of travel (fig. 6).


Figure 4—Floatplane network, includ-ing destinations for Misty Fjords.

Figure 5—Number of encounters at 12 frequently visited destinations.

Village Island

Rudyerd Bay

Hyder

Pt. Eva

Portage

Mink Bay

Wrangell

Ketchikan

Pt. Alava

Ella Lake

Alava Bay

Pt. Sykes

Wasp CoveWard Cove

Wilson Arm

Grant Cove

Martin Arm

Pt. Whaley

Behm Canal

Kah Shakes

King Creek

Anchor Pass

Bold Island

Walker Lake

Walker Cove

Mirror Lake

Sargent Bay

Skull CreekNarrow Pass

Halibut Bay

Clear Creek

Edward Pass

Hidden Inlet

Black Island

Verdue Point

Wilson River

Manzanita Bay

Fillmore Lake

Humpback Creek

Fillmore Inlet

Channel Island

Wilson Narrows

Chickamin River

Hugh Smith Lake

Unuk River Post

Valentine Creek

Rudyerd Bay South

Tombstone Rv-Hyder

New Eddystone Rock

Unuk River - mouth

0100200300400500600700800900

Ala

va

Bay

Big

Goat

Lak

e

Hy

der

Ket

chik

an

Man

zon

i

Lak

e

Pt.

. A

lav

a

Pu

nch

bo

wl

Lak

e

Pu

nch

bo

wl

Co

ve

Ru

dy

erd

Dock

Ru

dy

erd

Bay

Un

uk

Riv

er

- M

outh

Wal

ker

Co

ve

Destinations

Nu

mb

er o

f E

nco

un

ters


Figure 6—Monthly use levels for different user types.

To gain insight into potential conflict between recre-ation and wildlife, a more detailed analysis was per-formed at both Walker Cove and Rudyerd Bay. Moni-toring agents were placed at both areas and thesimulations rerun for the month of August (peakperiod of visitation) to determine the temporal distri-bution of use. Wildlife disturbance can be responsiveto temporal use distributions. We were interested inpeak use days in August, as well as the hours duringthose days that visitors are within sight of wildlife.

At Rudyerd Bay, visitation varied from day to day,increasing slightly as the month progressed (fig. 7).The number of fixed-wing aircraft frequenting thearea was lower than expected. The hourly assessmentrevealed a cyclical pattern of visitation over the periodof a day (fig. 8) with the number of ships increasingduring the evening hours. Many of these cruise shipsanchor in the cove at night and then leave early in themorning (between 4:00 and 10:00 a.m.).

These daily and hourly patterns of visitation revealthat there is very little time during the month of

August when there is not a cruise ship or commercialfloatplane in Rudyerd Bay.

Conclusion

This case study illustrates the application of com-puter simulation modeling in a marine environmentused by many different types of users. The spatial andtemporal explicitness of outputs is helpful in assessingmanagement concerns about the potential for impactson known concentrations of waterfowl, shorebirds,and harbor seals. Such impacts are highly space andtime dependent.

References

Gimblett, H. R.; Richards, M. T.; Itami, R. M. 2001. RBSim: Geo-graphic simulation of wilderness recreation behavior. Journal ofForestry. 99: 36–42.

0

10

20

30

40

50

60

70

10 11 12

Month

Nu

mb

er o

f V

isit

ors

Cruise Ship

fixed-wingaircraft

Boatdisplacementhull

Kayaks

Boat planninghull

54 6 7 8 9


Figure 8—Hourly encounters, for sea kayaks, cruise ships, and fixed-wing aircraft, at NorthArm of Rudyerd Bay, August 2000.

Figure 7—Daily encounters, for cruise ships and fixed-wing aircraft, at North Arm of Rudyerd Bay,August 2000.

0

5

10

15

20

25

1 3 11 13 15 17 19 21 23

Hour

Num

ber

of

Enco

unte

rs

Sea Kayaks

Cruise Ship

fixed-wingaircraft

5 7 9

0

1

2

3

4

5

6

7

8

9

10

8/1

/20

00

8/2

/20

00

8/3

/20

00

8/4

/20

00

8/5

/20

00

8/6

/20

00

8/7

/20

00

8/8

/20

00

8/9

/20

00

8/1

0/2

00

0

8/1

1/2

00

0

8/1

2/2

00

0

8/1

3/2

00

0

8/1

4/2

00

0

8/1

5/2

00

0

8/1

6/2

00

0

8/1

7/2

00

0

8/1

8/2

00

0

8/1

9/2

00

0

8/2

0/2

00

0

8/2

1/2

00

0

8/2

2/2

00

0

8/2

3/2

00

0

8/2

4/2

00

0

8/2

5/2

00

0

8/2

6/2

00

0

8/2

7/2

00

0

8/2

8/2

00

0

8/2

9/2

00

0

8/3

0/2

00

0

8/3

1/2

00

0

Date

Num

ber

of

Enco

unte

rs

Cruise Ship

fixed-wing

aircraft


Simulation of Recreation Use Alongthe Colorado River in Grand CanyonNational Park ___________________

Randy Gimblett

Purpose

The purpose of this case study is to demonstrate theuse of agent-based modeling and simulation ofmultiday river trips along the Colorado River, tripsthat are in high demand and for which crowding is amajor concern. Model outputs not only illustrate usepatterns along the river, but also predict how modi-fications of trip scheduling can affect use distributionand crowding.

Study Area

The increasing popularity of whitewater trips onthe Colorado River through Grand Canyon NationalPark threaten to impact sensitive, dynamic ecosys-tems and degrade the quality of experience of humanvisitors. Over 23,000 individuals and another 3,700guides, researchers, and park staff travel throughthe Grand Canyon each year. Visitors travel on over600 commercial or privately organized river trips ona variety of watercraft powered by oars, paddles, ormotors. Trips range from 6 to 18 days (to DiamondCreek) in the primary use season, and up to 30 daysat other times. Because of fluctuating water levelsimposed on the river from Glen Canyon dam, tripspeed can vary substantially through particularreaches of the river. When water levels are low,progress is slower (especially for oar-powered trips),and lost time is usually made up by skipping orreducing the time allocated to attraction sites and/orby spending more time rowing down the river. Highwater results in faster progress, which usually trans-lates to more stops and longer times at both attrac-tions and campsites, resulting in more crowding andgreater environmental impacts.

Many of the major drainages and side canyonsalong the 450-km river corridor in Grand CanyonNational Park provide opportunities for hiking andswimming. Well-known attractions and destinationsare regular stops for nearly every river trip thatpasses through the canyon. Crowding and encounterlevels along the river at attraction sites are oftenextreme and have been shown to affect the characterand quality of the visitor experience (Shelby andNielsen 1976). Simulation modeling was proposed asa means of better describing existing use patternsalong the river, as well as to predict whether changesin launch schedules could be used to decrease crowd-ing on the river.


Data were collected on both visitor characteristicsand site characteristics. Site characteristics includeriver travel networks and popular destinations.


In 1998 and 1999, hundreds of trip leaders fromprivate and commercial trips were asked to completetrip reports (fig. 9). These trip reports asked the tripleader to record the time in and time out for everyreasonable day or overnight stop for some 250 desig-nated sites between the launching spot at Lees Ferryand Diamond Creek. Other information collected tocharacterize the trips included: (1) river flow ratesduring the trip, (2) type and size of craft, and whetheroar or motor power, (3) number of passengers on thecraft, (4) experience of the guide, (5) number of encoun-ters with other parties, and (6) planned or actual timeallocated to specific sightseeing, rest, and campingstops during the trip.

Approximately 500 trip diaries were collected, rep-resenting about a 50 percent return rate for the com-mercial trips and a 30 percent return rate for theprivate trips. The trip diaries represent trips of alllengths and propulsion types (motorized, non-motor-ized). Itineraries were created from each of the diariesreturned. The trip itineraries were subsequently usedto simulate individual trips moving down the river,stopping at and starting from designated sites.

In addition, 15 river guides were interviewed. Theseguides collectively represented many years of experi-ence running the Colorado River either non-commer-cially (privately) or as guides for commercial outfit-ters. They had experience at various river flows andwith all types of watercraft (oars, paddle boats, dories,and motor boats). The intent of the interviews was tolearn as much as possible about the logic employed bya river guide when taking a trip down the ColoradoRiver. Questions were open-ended. For example, tounderstand how a guide might choose a campsite weasked questions such as, “when do you start thinkingabout camping for the evening?”, “what campsites doyou like and why?”, “which campsites do you try toavoid and why?”, and “what factors go into the processof choosing a campsite, and explain why each factor isimportant?” These insights were used to create a list ofpossible reasons regarding whether or not a campsitewas likely to be selected.


Data for the Colorado River were entered into aGeographic Information System (GIS) and used toconstruct a network for water travel on the river. Thenetwork included launch and take out sites, major


Figure 9—A portion of the trip diary used to collect spatial and temporal tripinformation.

rapids, and attraction sites. In addition, data on over200 beach sites, obtained from the Grand CanyonNational Park, were included. Beach site data in-cluded information on beach size at high water, esti-mated campsite capacity, and location by river mileand right or left side of the river.


The Grand Canyon River Trip Simulator (GCRTS)used a trace and terminating simulation approach(Daniel and Gimblett 2000) (refer to chapter 3 for moredetail). Central to the trip planning behavior of arecreation agent is the typical trip. The concept of atypical trip is based on the premise that visitors havecommon patterns of use. A typical trip has a global tripplan, which specifies an intended list of destinationsthe agent wishes to visit in a given sequence, anddurations. A global trip plan consists of a total tripduration, an entry node, an arrival time, an exit node,

and a sequence of intermediate nodes (destinations)between the entry and exit nodes. Each destinationnode has a visit duration.

While this is a trace simulation, there is minimaldecisionmaking the agents must do when confrontedwith certain situations on the river. While their tripplan specifies that they are to travel to a specificcampsite, that campsite might be full that night. Inthis case, the agent might choose to move to the nextunoccupied site. The simulation uses elements of fuzzylogic in the decision structure to weigh factors orvariables to provide the agent with some autonomy inadaptive decisionmaking. For example, when the agentis choosing a campsite, the current conditions of theriver and the individual trip play a role, as does thehistorical popularity of a campsite under consideration.This fuzzy logic approach takes into account all thesefactors and weighs them appropriately so that eachagent’s campsite decision represents a reasonable out-come for that given, particular set of circumstances.


Figure 10—Number of commercial trips per year that stop at selected sites along the Colorado River.

On execution, the simulation engine reads in a launchschedule (either the current launch schedule or a pro-spective calendar created by the user), then it createsand launches the trips from Lees Ferry. These simu-lated trips execute their days on the river by choosingattraction sites for hikes or other activities, by stoppingfor lunch, and by selecting an appropriate campsiteeach night. Certain trips must be at given locations oncertain times (for example, some trips exchange pas-sengers at Phantom Ranch). The trips are managed bythe simulator to meet these fixed points as scheduled.Moreover, a planning algorithm helps each simulatedtrip plan out an optimal schedule that will, for example,include stops at the key attraction sites and ensure thatcampsite selections are appropriate. A record is kept foreach simulated trip—where and when it encountersother trips, where it chooses to engage in an activity orto stop to camp, how long it stays, and so on. A large setof information for each trip is collected and stored in adatabase for processing.

Simulation Outputs

Figure 10 shows the popularity of selected attractionsites along the river with commercial groups. Similaroutputs can be produced for noncommercial groups, as

well as for campsites along the river. None of the sitesalong the river are used more than 60 nights duringthe primary use season.

To explore how different management options mightaffect crowding levels, a scenario was developed to testhow changes in the launch schedule would affect thedistribution of visitors and number of contacts along theColorado River. Current management policy directivestates that during the primary use season there shouldbe an 80 percent probability that a party will makecontact with seven parties or less per day. There shouldbe an 80 percent probability that contacts will notexceed 90 minutes per day and 125 people per day. Abaseline simulation was run using the data collected onriver trips launching between July 1, 1998, and August31, 1998 (peak season), and tested against a secondsimulation where launch restrictions were imposed onthe trips. All original trips that launched before noonwere randomly assigned a time before 10:00 a.m. and alltrips that previously had launched after 12:00 wererandomly assigned a launch time after 2:00 p.m. Wewanted to test whether this fixed launch schedulewould more evenly disperse trips along the river, lead-ing to an increase in the percentage of trips that metmanagement objectives for contacts.

Activity Site Visits for Commercial Trips Along the Colorado

River - Grand Canyon 1998

0

10

20

30

40

50

60

11

R,

SO

AP

19

.9 L

, 2

0 M

ILE

23

.7 L

, L

ON

E

33.6

L, B

ELO

W

39 R

, R

ED

BU

D

43.8

L, P

RE

S

51.2

L,

53

.2 R

, L

WR

59 R

, B

ELO

W

65

.5 R

, L

AV

A

71 L

,

75

.6 L

,

84

.4 L

, A

BO

VE

92

.3 L

, A

BO

VE

96.1

L, S

CH

IST

102.8

R, N

EW

10

9.4

R,

11

0

11

9.2

R,

12

0.1

R,

LW

R

12

4.3

L,

AB

OV

E

13

1.8

R,

13

3.9

R,

136.3

L, O

C'S

13

8.4

L,

14

4.2

R,

14

8.5

L,

15

8.2

R,

15

8

16

4.8

L,

168 R

, F

ER

N

17

4.4

R,

LW

R

17

9.7

R,

18

2.8

R,

18

5.5

R,

18

8.2

R,

LW

R

19

4.1

L,

206.6

R, IN

DIA

N

21

1.5

R,

UP

R

21

9.9

R,

22

3.4

L,

22

4-

235 L

, B

RID

GE

24

1.5

L,

Commercial

Trips


Figure 11—Number of contacts per river mile for scenarios with and without launch restrictions.

Of the trips launched between July 1, 1998, andAugust 31, 1998, only 9 percent actually met themanagement objective of coming into contact withseven or fewer parties, and 14 percent of those tripsexceeded the maximum duration of contact. Alteringthe launch schedule to impose restrictions after earlymorning and not allowing departures until early after-noon on a daily basis during the same time period didimprove the situation: 24 percent of trips met themanagement objective of coming into contact withseven or fewer parties, with only 8 percent of thosetrips exceeding the maximum duration of contact.Comparing the two simulations illustrates that, byimposing restrictions on launch times, managers canincrease the number of trips that meet managementobjectives.

Figure 11 shows the total number of river contactsby river mile for both simulation runs for the periodJuly 1, 1998, through August 31, 1998. This providesthe manager with an estimate of the total number ofthe contacts and where those contacts occur. Imposinglaunch restrictions reduces the total number of con-tacts along the river. At miles 34, 58.6, and 160, impos-ing launch restrictions reduces the number of rivercontacts by nearly half. Imposing launch restrictions

has a surprising effect on river contacts in the middlesection of the river, but has little effect towards the endof the trip.

More work still needs to be done to calibrate andvalidate the model. However, as this case study sug-gests, trace, fixed itinerary, terminating simulationsas described in this paper, can provide useful informa-tion for park managers on the spatial and dynamicdistribution of river trips. Simulating baseline condi-tions and comparing them to simulated scenarios canprovide ways to test and derive more meaningfulmanagement objectives and standards for visitor con-tacts and capacity. Altering launch schedules, discon-tinuing exchanges, and limiting use at camp andactivity sites are all conscious management actions,one of which has been shown here to significantlyreduce numbers of contacts.

References

Daniel, T. C.; Gimblett, H. R. 2000. Autonomous agent model tosupport river trip management decisions in Grand Canyon Na-tional Park. International Journal of Wilderness. 6(3): 39–42.

Shelby, Bo; Nielsen, J. M. 1976. Use levels and crowding in theGrand Canyon. Tech. Rep. 3. Colorado River Research ProgramReport Series. Washington, DC: U.S. Department of the Interior,National Park Service.

0

10

20

30

40

50

60

70

0 8

27

.2

30

.4 34

40

.7 41

47

.3

58

.6

61

.5

65

.7

67

.5

74

.3

87

.8

10

3

10

8

12

1

13

1

13

6

14

5

15

7

16

0

18

3

18

6

20

4

21

0

22

0

22

3

23

3

24

6

25

6

27

7

River Mile

To

tal N

um

be

r o

f R

ive

r C

on

tacts

Du

rin

g P

ea

k

Se

aso

n

No

Restrictions

Launch

Restrictions


John Muir Wilderness: Describingthe Spatial Distribution ofBackpacking Use on a Systemof Trails and Campsites __________

Steven R. LawsonRobert M. ItamiRandy GimblettRobert E. Manning

Purpose

This case study demonstrates the utility of a com-puter simulation model in describing the spatial dis-tribution of visitor use and encounters between groupsin a dispersed recreation setting. A simulation modelof backcountry hiking and camping use was developedfor a portion of the John Muir Wilderness in the InyoNational Forest, California (Lawson and others 2004).The John Muir Wilderness covers 235,000 ha in theSierra Nevada Mountains. The area is characterizedby snowcapped mountains with hundreds of lakes andstreams and lush meadows. The simulation modelestimated hiking and camping use and encounters bytrail segment and camping location. Outputs from thesimulation model were imported into a GeographicInformation System (GIS) database and mapped. Thiscase study should be considered exploratory and illus-trative. There were weaknesses in the data collectionphase that suggest results do not accurately representcurrent use distribution in the study area. For ex-ample, response rate to the visitor survey was low, andrespondents were not necessarily representative of allvisitors to the area. However, since these weaknessescan be easily corrected, the study illustrates the poten-tial value of this tool.

Data Collection

Visitor Characteristics—During the 1999 visitoruse season, diary-like questionnaires were distrib-uted to backcountry visitors in a portion of the JohnMuir Wilderness. Questionnaires were distributed attrailhead self-registration stations and at ranger sta-tions when visitors picked up their agency-issuedpermit. Randomly selected self-registration stationswere periodically attended by data collectors whodistributed diaries to visitor groups and collectedcompleted questionnaires from groups as they fin-ished their trips. In addition, questionnaires weredistributed by commercial packstock outfitters, fol-lowing instructions given by the research team.

The diary-like questionnaire included a series ofquestions concerning group and trip characteristicsand a map of trails and natural features. Respondents

were instructed to record their route of travel duringtheir visit, including the trailhead(s) where they startedand ended their trip, and their camping location oneach night. Respondents were also asked to report theduration of their visit, the number of people in theirparty, and their mode of travel. The response rate was32 percent, resulting in a total of 324 competed diaries.The low response rate is a primary reason we questionthe validity of model outputs for other than illustra-tive purposes.

Site Characteristics—Data on the trail networkwithin the study area were provided by the InyoNational Forest as a GIS overlay. The data includedall trail segments and intersections within the studyarea. These data were supplemented with informationfrom a campground inventory completed in the sum-mers of 1999 and 2000. Campsite “clusters” werecreated from the visitor surveys by grouping campinglocations based on proximity and common access. Asingle campsite cluster was comprised of all reportedcamping locations that were within a subjectivelydetermined reasonable distance of each other. Thecampsite clusters were used to determine campingencounters within the travel simulation model. Spe-cifically, groups camping at locations within the samecampsite cluster were considered to be within closeenough proximity to be within sight and/or sound ofeach other.

Travel Simulation Model Design andAnalysis

The travel simulation model was developed usingthe commercially available general-purpose simula-tion software, Extend, developed by Imagine That,Inc., and is a probabilistic, steady-state simulation(see Chapter 3 for information about probabilistic andsteady-state simulations). Two simplifying assump-tions worth noting were built into the model. First,only data for overnight hikers were included in themodel; packstock trips and day trips were excluded.Second, the travel simulation model was only appliedto a small portion of the larger study area. The sectionof the study area for which the model was developed isreferred to as the Desolation Lake Locale (fig. 12).

The majority of visitor use occurs during the sum-mer months. Therefore, the computer simulation modelwas designed to focus on the “peak” period of thevisitor use season, July 1 through September 30, 1999.The simplifying assumptions described in the previ-ous paragraph, coupled with the decision to focus onthe summer months of the visitor use season, resultedin a total of 190 usable trip itineraries included asinputs into the travel simulation model.

Outputs—The simulation model was designed togenerate four outputs concerning visitor use densities


Figure 12—Larger study area, with the Desolation Lake locale inside the box.

and hiking and camping encounters. First, averagehiking use per day is calculated for each trail segmentby summing the number of groups that pass througheach trail segment during the course of the simulationand dividing by the total number of days simulated.

Second, average hiking encounters per group perday are calculated for each trail segment on each daythat at least one group passes along the trail segment.Two types of hiking encounters were calculated(Schechter and Lucas 1978). “Overtaking encounters”are defined as one group passing another group whiletraveling in the same direction along the trail while“Meeting encounters” are defined as two groups pass-ing each other while traveling in opposite directions.The average number of hiking encounters per groupper day is calculated for each trail segment by sum-ming the total number of hiking encounters along thetrail segment throughout the simulation and dividingby the total number of groups that hiked the trailsegment during the simulation.

Third, average camping use per night is calculatedfor each campsite cluster by summing the number ofgroups that camp at the campsite cluster during the

course of the simulation and dividing by the totalnumber of nights simulated. Fourth, average campingencounters per group per night are calculated for eachnight that a campsite cluster is occupied by one ormore parties. The number of camping encounters agroup has on a given night is equal to the number ofother groups camping in the same campsite cluster onthe same simulated night. The average number ofcamping encounters per group per night is calculatedfor each campsite cluster by summing the total num-ber of campsite encounters throughout the simulationand dividing by the total number of groups that campedat the campsite cluster during the simulation.

Baseline Simulation—The simulation reportedhere was designed to generate the outputs describedabove for baseline visitor use levels and existingmanagement practices, where the baseline level ofvisitor use is assumed to be equal to the number ofgroups that completed the diary questionnaire duringthe sampling period. The reader is cautioned that onlyabout one-third of the groups completed the question-naires. Therefore, the outputs reported underestimateuse substantially—perhaps by a factor of about three.


Model Verification—Several model verificationtechniques were conducted to ensure that the simula-tion model was implemented correctly. First, indi-vidual components of the simulation model were de-veloped and debugged to ensure that they functionedproperly. Modules for the campsite clusters, trail seg-ments, trail junctions, and trailheads were developedand tested individually before being combined in thefull model. Second, the authors conducted a structured“walk-through” of the simulation model to review thelogic and operation of the simulation model. Third,Extend’s built-in debugging functions were used todetect and correct data coding errors in the full modelthat resulted in incorrect routing of simulated trips.Fourth, Extend’s animation function was used to visu-ally verify that the model did not contain errors. Fifth,the trip itinerary data reported by respondents to thediary questionnaire were used as the basis for aquantitative verification technique. Specifically, theproportion of camping use for each campsite clusterwas calculated for the diary trip itinerary data andcompared to the proportion of camping use for eachcampsite cluster. This technique provided a methodfor comparing the output of the model with knownanalytical results (Schecter and Lucas 1978).

The most powerful technique for validating a com-puter simulation model is to compare model outputdata to data from the actual system the model isdesigned to replicate (Law and Kelton 2000). Due to alack of data from the actual system (for example,hiking encounters per group and hiking use per trail-head in the area), this quantitative technique couldnot be used to assess the validity of the computersimulation model. However, validation techniquesbased on intuitive judgement were used to assess thecontent and face validity of the simulation model byexamining the reasonableness of the inputs and out-puts. Furthermore, sensitivity analyses were con-ducted to test whether the model outputs change inthe expected direction in response to changes in se-lected input parameters (for example, total use).

Results

Table 3 reports mean camping use per night andmean camping encounters per group per night for eachcampsite cluster. The map in figure 13 portrays thespatial distribution of camping use within the studyarea. Table 4 reports mean hiking use per day andmean hiking encounters per group per day, by trailsegment. The map in figure 14 portrays the spatialdistribution of hiking use within the study area.

The results reported in table 5 suggest that thecomputer simulation model has been constructedcorrectly and that operating errors have been elimi-nated through the debugging and verification tech-niques described earlier in this paper. Specifically, as

Table 3—Mean camping use and encounters, by campsitecluster.

Mean number of Mean campingCampsite camping groups encounters perclusters per night group per night

7 0.86 0.9036 0.12 0.1437 0.74 0.7538 0.05 0.0639 0.15 0.1240 0.05 0.0341 0.26 0.2242 0.32 0.3344 0.44 0.4345 0.13 0.1246 0.48 0.5147 0.31 0.2548 0.14 0.1449 0.04 0.0050 0.12 0.1551 0.07 0.0452 0.02 0.0053 0.04 0.1056 0.10 0.0957 0.14 0.1380 0.11 0.0981 0.07 0.02

indicated in table 5, there is no substantive differencebetween the distribution of campsite cluster usereported in the diary survey and the simulated trips.

Due to the low response rate to the questionnaire,the content validity of the model, which is concernedwith the reasonableness of the model inputs, is low.Furthermore, statistical comparisons of model out-puts to actual system data are not possible due to alack of “ground-truthing” data. However, results of aseries of simulations conducted with the model sup-port the internal and face validity of the model. Spe-cifically, as the total simulated use of the study area is“ramped up,” estimates of hiking and camping use andencounters increase or remain unchanged for all trailsegments and campsite clusters, respectively.

Conclusions

This study illustrates the potential usefulness ofcomputer-based simulation modeling in describingthe spatial distribution of recreational use in back-country and wilderness landscapes. Dispersed recre-ation in such areas is inherently difficult to observedirectly. However, the study findings suggest that bycollecting representative data by means of trailheadcounts and a diary survey of a sample of visitor groups,it is possible to estimate levels and patterns of visitoruse. The model can also estimate encounters between


Figure 13—Use levels of camp-site clusters in the Desolation Lakelocale.

Figure 14—Use levels of trailsegments in the Desolation Lakelocale.


Table 4—Mean hiking use and encounters, by trail segment.

Mean number of Mean hikingCampsite hiking groups encounters perclusters per day group per day

2 3.51 0.203 0.08 0.004 3.51 0.115 3.43 0.346 0.58 0.037 0.14 0.038 0.04 0.009 3.35 0.11

10 3.28 0.1011 3.20 0.0512 0.12 0.0013 0.20 0.0114 0.80 0.0415 2.95 0.2016 1.10 0.0217 2.47 0.1118 2.41 0.0519 0.15 0.0120 0.99 0.0121 0.90 0.0322 0.77 0.0623 0.09 0.0024 0.13 0.0125 2.31 0.0726 0.15 0.0227 1.08 0.0628 0.15 0.0129 0.45 0.0230 1.29 0.0131 0.68 0.0332 0.63 0.0533 0.04 0.0034 1.87 0.0935 0.07 0.0036 1.43 0.0837 0.29 0.0238 0.88 0.0639 1.29 0.2140 0.22 0.0141 1.25 0.07

132 0.06 0.00

Table 5—Percentage of total camping use in each clusterbased on survey data and model output.

Campsite cluster Survey data Model output

7 0.18 0.1836 0.02 0.0237 0.16 0.1638 0.01 0.0139 0.03 0.0340 0.01 0.0141 0.05 0.0542 0.07 0.0844 0.09 0.1045 0.04 0.0346 0.10 0.1047 0.06 0.0648 0.03 0.0349 0.01 0.0150 0.03 0.0251 0.01 0.0152 0.00 0.0053 0.01 0.0156 0.02 0.0257 0.03 0.0380 0.02 0.0281 0.01 0.01

groups, which is particularly difficult to monitor di-rectly. This information can be used for several pur-poses, including monitoring indicator variables to en-sure that standards are not violated, identifyingpotential bottlenecks or congested sites, schedulingmaintenance and patrol activities, and educating visi-tors about the conditions they are likely to experience.

Finally, the results of this study demonstrate that,by integrating computer simulation and GIS technolo-gies, it is possible to map existing recreational use

patterns. This capability provides managers with atool to communicate more effectively with the publicregarding existing visitor use management issues. Byimporting computer simulation results into a GISdatabase, it is possible to conduct overlay analyseswith resource data to examine relationships betweennatural resource characteristics (for example, resourcefragility, resource impacts, and so forth) and existingvisitor use patterns. Furthermore, integrating GISand computer simulation technologies provides man-agers with a means to illustrate with maps the poten-tial effect of alternative visitor use policy decisions andmanagement practices on visitor use patterns andnatural resources within a dispersed, backcountrysetting.

References

Law, A.; Kelton, W. 2000. Simulation modeling and analysis. 3d ed.Boston, MA: McGraw Hill.

Lawson, S.; Itami, B.; Gimblett, R.; Manning, R. 2004. Monitoringand managing recreational use in backcountry landscapes usingcomputer-based simulation modeling. In: Policies, methods andtools for visitor management: proceedings of the second interna-tional conference on monitoring and management of visitor flowsin recreational and protected areas. Rovaniemi, Finland: FinnishForest Research Institute: 107–113.

Shechter, M.; Lucas, R. 1978. Simulation of recreational use for parkand wilderness management. Washington, DC: Resources for theFuture.


Figure 15—Study photograph of the trail toYosemite Falls, representative of moderateuse levels.

Frontcountry Trails and AttractionSites in Yosemite National Park:Estimating the Maximum Use ThatCan Be Accommodated WithoutViolating Standards of Quality _____

Robert E. ManningWilliam A. ValliereBenjamin WangSteven R. LawsonPeter Newman

Purpose

Many park managers are concerned about overuseof trails and attraction sites resulting in low qualityvisitor experiences. Using contemporary planningframeworks, such as Limits of Acceptable Changeand Visitor Experience and Resource Protection, theyoften formulate indicators and standards of quality.Indicators of quality are measurable, manageablevariables that reflect the quality of the visitor expe-rience. Standards of quality define the minimumacceptable conditions of indicator variables. Onceindicators and standards of quality have been formu-lated, management actions can be implemented toensure that standards of quality are maintained.Computer simulation modeling can be used to esti-mate the maximum level of visitor use that can beaccommodated along trails and at attraction siteswithout violating crowding-related standards of qual-ity. This case study provides an illustration of theutility of computer simulation modeling for this pur-pose at heavily used frontcountry trails and attrac-tion sites in Yosemite National Park (Manning andothers 2003).

Description of Study Area

Yosemite Valley is the scenic heart of YosemiteNational Park, arguably the first National Park in theUnited States, and certainly one of America’s bestknown and most popular National Parks. YosemiteValley is a glacially carved area of approximately 20km2 and features sheer granite walls of up to 5,000 feetand several of the world’s highest waterfalls. YosemiteNational Park draws over 4 million visits annually,and carrying capacity has been a long-standing andcontroversial issue. Several specific visitor attractionsites were chosen for this study: the trail to VernalFall, Yosemite Falls (including the base of the falls andthe trail leading to it), Bridalveil Fall (including thebase of the fall and the trail leading to it), GlacierPoint, and the trail to Mirror Lake.

Potential Standards

The Yosemite Valley project involved two phases. Thefirst used a visitor survey to generate a range of potentialstandards for maximum people-at-one-time (PAOT) atattraction sites and persons-per-viewscape (PPV) alongtrails. Using computer-edited photographs illustrating arange of PAOT and PPV at study sites, respondents wereasked about the level of use they (1) prefer, (2) consideracceptable, (3) consider so unacceptable they would nolonger visit, and (4) think the National Park Serviceshould allow (fig. 15). Mean responses for each of thesefour evaluative domains (denoted as preference, accept-ability, tolerance, and management action, respectively)provide a range of potential standards from the perspec-tive of current visitors. Reponses were compiled forvisitors on four trail segments and at three popularviewpoints (table 6).

Computer Simulation Modeling

The second phase, the primary subject of this casestudy, involved using computer simulation to predictmaximum use levels that could be sustained withoutviolating these potential standards. For this purpose,detailed counts and observations of visitor use werecollected at each of the seven study sites. Variablesincluded length of trails, length of typical trailviewscapes (how far along the trail a hiker can typi-cally see), number of visitors arriving per hour, visitorgroup size, length of time visitors stop at attractions,and the speed at which visitors hike trails.

Using the input data described above, simulationmodels were developed for each of the study sites. Themodels were built using the commercial simulationpackage, Extend, by Imagine That, Inc. Extend is anobject-oriented, dynamic simulation package that has


Table 6—Alternative crowding-related standards of quality for all study sites.

Trail to Trail to Base of Trail to Base of Glacier Trail toNormative standard Vernal Fall Yosemite Falls Yosemite Falls Bridalveil Fall Bridalveil Fall Point Mirror Lake

of quality (PPV) (PPV) (PAOT) (PPV) (PAOT) (PAOT) (PPV)

Preference 11 18 43 7 8 19 10Acceptability 26 40 92 18 20 42 24Management action 30 46 100 20 19 49 26Tolerance 39 60 126 26 25 61 34

Figure 16—Schematic diagram of computer simulation model of Glacier Point.

been used extensively in business, manufacturing, andelectronics applications to improve quality and effi-ciency. The object orientation makes code-writing un-necessary, and the programming algorithm can easilybe expressed by the graphic display of objects andconnections. For example, figure 16 shows the top layerof the simulation model developed for Glacier Point.The model was built with a series of “blocks” to repre-sent components of this study site. The block labeled“Hiker Generator” in the top left section of the model iswhere simulated visitors are generated. Connected to itare groups of blocks that provide data concerning em-pirical visitor arrival rates and group sizes. (These dataare derived from field observation.) The block “TotalUse” allows the researcher to specify the daily total visituse level to be simulated. The simulated visitors thenspend a randomly assigned length of time (based onfield observations) at the “Viewing Area” block that

represents the Glacier Point visitor attraction. Theassociated blocks (those underneath the Viewing areablock) measure the number of people-at-one-time(PAOT) at the Glacier Point viewing area. The simula-tion models developed in this project were probabilistic(based on probabilities of simulated visitors selectingavailable travel itineraries) and were terminating (theymodeled an entire day).

Model output could be generated in several graphicand numerical forms. For example, figure 17 tracesminute-by-minute PPV levels along the trail toBridalveil Fall over the duration of a day. This particu-lar model run was generated using an average sum-mer day total use level of 1,415 visitors (derived fromthe counts of visitor use taken to help construct themodel). To predict the maximum total daily use levelthat could be accommodated at each study site andeach evaluative domain (preference, acceptability,


Figure 17—Person-per-viewscape (PPV) along the trail to Bridalveil Fall.

management action, and tolerance), models were runmultiple times. If standards were violated, modelswere rerun at successively lower use levels until stan-dards were no longer violated. In addition, all modelswere run multiple times to “average out” the random-ness associated with each individual model run. Therange of estimated daily maximum use levels for eachstudy site and evaluative domain is shown in Table 7.

Conclusion

This project illustrates the utility of computer simu-lation modeling related to planning and managementof trails and attraction sites in the front country ofparks. Specifically, it illustrates an important applica-

Table 7—Range of daily maximum use levels at all study sitesa.

Trail to Trail to Base of Trail to Base of Glacier Trail toNormative standard Vernal Fall Yosemite Falls Yosemite Falls Bridalveil Fall Bridalveil Fall Point Mirror Lake

Preference 2,100 4,500 3,000 1,200 700 1,500 1,800Acceptability 6,000 9,500 5,500 3,200 1,700 4,000 5,000Management action 7,000 11,000 5,900 3,500 1,700 4,800 5,500Tolerance 9,300 13,000 7,300 4,800 2,300 6,300 7,700

a Daily maximum use levels were calculated to allow standards to be exceeded a maximum of 10 percent of the time.

tion of modeling for estimating the maximum uselevels that can be sustained without violating stan-dards related to crowding. Planners can use this infor-mation to evaluate the consequences of deciding amongalternative standards. Such predictions can also belinked to transportation planning. For example, oncestandards are selected, a visitor transit system couldbe designed for Yosemite Valley to deliver the “right”number of visitors to the “right” places at the “right”intervals.

References

Manning, R.; Valliere, W.; Wang, B.; Lawson, S.; Newman, P. 2003.Estimating day use social carrying capacity in Yosemite NationalPark. Leisure: The Journal of the Canadian Association forLeisure Studies. 27(1–2): 77–102.

0

5

10

15

20

25

30

35

40

45

Time of Simulation Day

PP

V

7:00 am 8:00 pm


Alcatraz Island: Estimating theMaximum Use That Can BeAccommodated Without ViolatingStandards of Quality _____________

William A. ValliereRobert E. ManningBenjamin Wang

Purpose

The purpose of this case study is to illustrate how acomputer simulation model can be used to estimatethe maximum number of people that can visit AlcatrazIsland without violating crowding-related standardsof quality for the cellhouse on the Island (Manning andothers 2002). The model acts as a translator that takessimple inputs (such as the number of visitors per ferryboat servicing the island and the frequency of boatarrivals) and estimates the persons-at-one-time(PAOT) on Michigan Avenue (as representative of thewhole cellhouse). This approach provides an efficientand economical way of monitoring an indicator ofquality to measure against standards of quality.


Alcatraz Island is an historic site within GoldenGate National Recreation Area, a unit of the USDINational Park System, and is a heavily visited touristattraction. The island, located in San Francisco Bay, iswidely known for its history as a Federal prison forincorrigible criminals. This history has been romanti-cized and popularized in several books and movies;consequently, the demand to visit Alcatraz is high.Visitation now exceeds several hundred thousandannually, and continues to grow rapidly.


Data collection consisted of two phases. During thefirst phase, visitors were asked questions regardingthe acceptability of different levels of use in the cellhouse (fig. 18). Using computer-edited photographs,respondents were asked about the level of use they(1) prefer, (2) consider acceptable, and (3) think theNational Park Service should allow. Mean responsesfor each of these three evaluative domains (denoted aspreference, acceptability, and management action,respectively) were 25, 44, and 44 persons-at-one-time(POT). These values provide a range of potentialstandards from the perspective of current visitors.

Data collection for the second component of theresearch consisted of gathering the information neces-sary to construct the computer simulation model.

Figure 18—Study photograph of Michigan Av-enue in the prison cellhouse, representative ofa moderate use level.

Inputs gathered to build the basic model includedcurrent numbers of visitors per ferry boat, currentboat schedule, the lengths of time between debarka-tion and arrival into the cellhouse audio tour ticketline, and the length of time that visitors take to gothrough the audio tour. Boat schedules and the num-ber of visitors per boat were easily obtained from Parkrecords. The time between debarking and arrival atthe audio tour and the time it takes visitors to finishthe audio tour were determined by research staff overa 3-day period in October 1998. On October 27 through29, three researchers stationed at the dock and vari-ous points in the cellhouse handed time assessmentcards to randomly selected visitors. The researchersrecorded on the cards the times of the day when thevisitors arrived and departed from selected places. Atotal of 179 time assessment cards were collectedduring the 3-day period.

Model Characteristics

The software package selected for the simulationwas Extend by Imagine That, Inc. Figure 19 is the toplayer of the model, and is what a researcher or man-ager would encounter after opening the model. Theprison cellhouse model was built with “blocks” thatrepresented specific components of the Alcatraz Is-land visitor system. The block labeled “Dock” in thetop-left section of the model is where debarking simu-lated visitors are created. Double clicking on it wouldreveal an interface that allows the researcher or man-ager to specify the boat schedules and boat loads to besimulated. The block labeled “Up the Hill” representsthe areas where the visitors are between the time theydebark and the time that they line up for the audiotour. The amount of time that the simulated visitorsare delayed here was determined by the 3 days of


Figure 19—The Alcatraz cell house visitor research model.

empirical time assessment exercises described above.The average time taken was 32 minutes, with a stan-dard deviation of 19 minutes. The blocks labeled“Audio tour line” and “Cashier” simulate the queuethat visitors wait in to pay for their audio tours.

The next series of blocks simulate portions of theaudio tour in the cell house. Each simulated visitortravels through the blocks labeled “Tour beginning,”“Tour middle,” and “Michigan Ave.” Both the audiotour program and the estimated time that visitorspaused the cassette during their tour determine theamount of time visitors take to move through thesystem. The length of time that visitors took to travelthrough the entire system was measured during the3 days of time assessments. The average time takenwas 44 minutes with 13 minutes standard deviation.When compared with the cassette running time ofapproximately 35 minutes, these results indicatethat visitors frequently pause their cassette playersduring the tour. The ratio between these times wasused to estimate that the total time each visitor spentout in the open on Michigan Avenue was about 5.4minutes.

Model Outputs

A typical simulation model run with current sum-mer boatloads and schedule yields a line graph show-ing the numbers of simulated visitors on MichiganAvenue through an entire simulated day (fig. 20).Figure 20 shows that PAOT on Michigan Avenue atany one time can fluctuate substantially, anywherefrom about 50 people to about 90 people. Figure 21provides a different look at this fluctuation in the formof a histogram. Each bar represents a range of PAOTconditions on Michigan Avenue through the samesimulated day.

The average condition on Michigan Avenue during10 computer runs was about 70 PAOT. The modelestimates that visitors encounter more than 70 PAOTfor about 44 percent of their tour time. They encoun-tered more than 80 PAOT for about 15 percent of the timeand more than 90 PAOT for about 3 percent of the time.

The model was run multiple times (to “average out”the randomness associated with each individual modelrun) to estimate the maximum total daily use levelsthat could be accommodated on the Island withoutviolating each of the potential crowding-related stan-dards more than 10 percent of the time (table 8). For


Figure 21—Simulated people-at-one-time (PAOT) conditions on Michigan Avenue.

Figure 20—People-at-one-time (PAOT) in Michigan Avenue over the minutes of asimulated day.

0

10

20

30

40

50

60

70

80

90

100

110

Time of Simulation Day

PA

OT

9:30 am 5:30 pm

0

5

10

15

20

25

0 to

55

to 1

010

to 1

515

to 2

020

to 2

525

to 3

030

to 3

535

to 4

040

to 4

545

to 5

050

to 5

555

to 6

060

to 6

565

to 7

070

to 7

575

to 8

080

to 8

585

to 9

090

to 9

595

to 1

0010

0 to

105

105

to 1

10

PAOT Ranges

Pe

rce

nta

ge

of

Tim

e P

AO

T R

an

ge

Oc

cu

rre

d


this range of potential standards, maximum allowableuse on Alcatraz Island could range between approxi-mately 2,500 and 4,800 visitors per day. A daily uselimit for Alcatraz Island could be implemented rela-tively easily through management of the ferry systemserving the Island.

A final analytical approach used in this study em-ployed the computer simulation model of visitor use toexplore the effect of alternative ferry schedules. Cur-rently, ferries depart San Francisco for Alcatraz Is-land every half hour from 9:30 a.m. to 5:30 p.m. If ferrydepartures were reduced to every hour, maximumallowable use would also be substantially reduced.Relatively large numbers of visitors arriving at thesame time, however, would result in many visitorsseeing relatively large PAOT in Michigan Avenue.

For example, with the current ferry schedule, approxi-mately 4,800 visitors per day can be accommodated onAlcatraz Island without exceeding 44 PAOT more than10 percent of the time (table 8). However, if ferries wereto depart only every hour, the computer simulationmodel estimates that only 3,200 visitors per day could beaccommodated without exceeding 44 PAOT more than10 percent of the time. Increasing the frequency of ferryservice would not increase maximum allowable uselevels substantially. For example, increasing the fre-quency of departures to every 15 minutes would increasethe maximum use level that would not exceed 44 PAOTmore than 10 percent of the time from approximately4,800 visitors per day (under existing ferry service) toapproximately 4,900 visitors per day.

Conclusions

This case study illustrated the utility of a computersimulation model in dealing with crowding-relatedissues in an urban park. Specifically, it provided ameans of predicting maximum use levels that could beaccommodated without violating specified standards.In addition, it provided insight into how varying publictransportation schedules might influence maximumuse levels.

References

Manning, R.; Wang, B.; Valliere, W.; Lawson, S.; Newman, P. 2002.Research to estimate and manage carrying capacity of a touristattraction: a study of Alcatraz Island. Journal of SustainableTourism. 10: 388–404.

Table 8—Alternative daily maximum use levels at AlcatrazIsland.

Standards of quality Maximum daily use level

Preference (25 PAOT) 2,560Acceptability (44 PAOT) 4,800Management action (44 PAOT) 4,800

Arches National Park: DescribingVisitor Use Patterns and Predictingthe Effects of AlternativeTransportation Systems __________

Steven R. LawsonRobert E. ManningWilliam A. ValliereBenjamin Wang

Purpose

The purpose of this case study is to illustrate how acomputer simulation model can describe daily visitoruse patterns and levels throughout Arches NationalPark’s road and trail network and at selected attrac-tion sites. The simulation model provides managerswith estimates of “minute-by-minute” vehicle trafficlevels along all road segments and at all parking areaswithin the park on a “typical” peak season day, as wellas the number of visitors along selected trail segmentsand at significant park attractions (for example, Deli-cate Arch). Further, the model is used to predict thelikely effects of alternative transportation systems oncrowding and congestion. Specifically, simulation wasused to predict the effect of a mandatory shuttle bussystem on the number of visitors that can be accommo-dated at Delicate Arch.

Arches National Park

Arches National Park is located in southeasternUtah, 8 km north of the town of Moab. The park isapproximately 30,000 ha in size and is known for itsunique sandstone features, including Delicate Arch,The Windows, and Devil's Garden. The predominantrecreation uses of the park include hiking, photogra-phy, and scenic driving, and the majority of recreationuse is concentrated along trails and at attraction sitesadjacent to or near the park’s road network. Mostvisitors enter the park through an entrance gate at thesouthern end and access the interior via the park’sroad system. Currently, there are no restrictions onthe number of vehicles allowed to enter the park, andthere is no required alternative transportation sys-tem. The park is undergoing an alternative transpor-tation planning process, however, and a shuttle bussystem is among the ideas being discussed by parkstaff and the public.


A variety of methods were employed to gather thedata used to build the simulation model of visitor use.A traffic counter at the entrance to the park recorded


the number of vehicles entering and the time eachvehicle entered. These traffic data were collected dur-ing the 7-day period August 19 to 25, 1997. Total dailyentries for these 7 days averaged 1,346 vehicles.

Data on visitor characteristics and travel patternswithin Arches National Park were collected through aseries of onsite surveys administered to visitors dur-ing the summers of 1997 and 1998. Vehicle travelroute questionnaires were administered to 426 visitorgroups in private automobiles and 160 tour bus driv-ers as they were exiting the park. Respondents wereasked to report their group’s size, the amount of timethey had spent traveling on the park roads, and whereand how long they paused during the visit (for at least5 minutes). Finally, with the aid of the interviewer,they were asked to retrace the route of their trip on amap of the park. The vehicle travel route question-naires were administered to visitor groups on 6 daysduring the period August 14 to 30, 1997, and to tourbus drivers on 42 days between July 9 and October 22,1998. For safety concerns, each sampling day startedat 7:00 a.m. and ended at dusk.

A second questionnaire was administered during thesummer of 1997 to a total of 180 visitor groups return-ing from hikes to Delicate Arch. One visitor from eachgroup was asked to report the group’s size, the amountof time they had spent on the trail and at the Arch, andwhere and how long they paused during the hike (for atleast of 5 minutes). These questionnaires were admin-istered on 3 days during the period August 15 to 24,starting at 7:00 a.m. and ending at 10:00 p.m.

Hiking questionnaires were also administered dur-ing the summer of 1998 at The Windows and Devil’sGarden sections of the park. Similar to the hikingquestionnaire administered at Delicate Arch, visitorgroups at The Windows and Devil’s Garden areas wereasked to report information about their group size, theroute they hiked, and the places and amount of timethey paused during the hike. A total of 245 question-naires were completed by visitors returning from theirhikes around The Windows on 5 days during theperiod July 18 to August 3, and 320 questionnaireswere administered to hikers returning from theirhikes in the Devil’s Garden on 5 days during the periodJuly 5 to August 6. Surveys in both locations started at7:00 a.m. and ended at 10:00 p.m.

The sampling period for the visitor surveys wasdesigned to ensure that an adequate number anddiversity of vehicle and hiking routes were sampled.For example, a greater number of sampling days wasallocated to collecting tour bus routes than personalvehicle routes because fewer tour buses enter the parkeach day than personal vehicles. In addition, thesampling period was selected to occur during the peakperiod of the visitor use season. Lastly, while thesampling occurred over a 2-year period, there is noreason to believe that the distribution of visitor travel

routes changed substantively from the first year to thesecond year as there were no changes to park infra-structure, such as roads and trails.

Data needed to validate the output of the travelsimulation model were gathered through a series ofvehicle counts conducted at selected parking lots inthe park. The number of vehicles in the Delicate Arch,The Windows, and Devil’s Garden parking lots (thepark’s three major attraction areas) were counted 11times a day between 6:00 a.m. and 10:00 p.m. on 4 daysduring the period from August 19 to 25, 1997. The totalnumber of vehicles entering the park was recordedwith traffic counters on each of the days that parkinglot counts were conducted.


Using the data inputs described above, a probabilis-tic, terminating simulation model of park use wasdeveloped (refer to Chapter 3 for more informationabout terminating and probabilistic simulation mod-eling). A terminating simulation has a known initialstate (usually zero) and a known ending state, whichis well suited for replicating a system’s behavior overa discrete period of time, such as a single day. Since theprimary use of the simulation model developed in thisproject was to describe recreation use patterns andlevels on a “typical” peak-season day, terminatingsimulation was adopted.

Simulation Model Runs

Use Levels and Patterns—To estimate existingrecreation use levels and patterns (vehicle and pedes-trian), a series of runs was conducted to simulatecurrent average “peak” daily use. Twelve replicationsof the current daily use simulation were conducted toaccount for stochastic variation of model inputs andoutputs. As noted earlier, safety concerns preventedvehicle and tour bus travel route surveys from beingadministered after dark. Therefore, each run simu-lated a single day of park use from 5:00 a.m. to 4:00 p.m.

Effect of Alternative Transportation—Previ-ous research has led to establishment of selectedindicators and standards of quality for major attrac-tions within Arches National Park (Manning andothers 1995, 1996a,b; National Park Service 1995).For example, to avoid unacceptable levels of crowding,the number of people-at-one-time (PAOT) at DelicateArch should not exceed 30 more than 10 percent of thetime. An initial set of simulations was conducted withthe Arches computer simulation model to estimate themaximum number of visitors that can be allowed tohike to Delicate Arch before this standard of quality asviolated (referred to in the remainder of this paper ismaximum allowable use at Delicate Arch). Next, aseries of simulations were conducted to predict whether


implementing a mandatory shuttle bus system wouldincrease the maximum allowable use of Delicate Archand, if so, by how much.

To estimate maximum allowable use for DelicateArch without a shuttle bus system, a series of simu-lations was run in which the total number of visitorshiking to the Arch was varied. The average percent oftime that PAOT at Delicate Arch exceeded 30 (inother words, the maximum acceptable level of PAOTat Delicate Arch) was recorded for each use levelmodeled. An iterative process of increasing or de-creasing the daily number of visitors hiking to theArch was followed until PAOT at Delicate Arch ex-ceeded 30 an average of 10 percent of the time. Forexample, a series of 12 simulations was run for aselected level of visitor use and the average percent oftime that PAOT exceeded 30 was calculated from thesimulation results. If PAOT exceeded 30 an averageof more than 10 percent of the time, the next set of 12simulations was run at a lower use level, while ifPAOT exceeded 30 an average of less than 10 percentof the time, the next set of 12 simulations was run ata higher level of visitor use.

A series of model runs was conducted to simulate theoperation of a shuttle bus system designed to delivervisitors to the Delicate Arch parking lot at regularlyscheduled time intervals. Separate model runs wereconducted to simulate alternative shuttle bus sched-ules designed to arrive at Delicate Arch every 15, 30,and 60 minutes. For each shuttle bus system simu-lated, the number of visitors riding the shuttle bus andhiking to the Arch was varied to estimate the maxi-mum allowable use of Delicate Arch. The shuttle bussimulations were based on the assumption that theamount of time visitors spend hiking to the Arch andat the Arch itself would not change as a result ofimplementing a shuttle bus system. Although it wouldhave been possible to alter the amount of time visitorsspend at the Arch in response to a shuttle bus system,there is no reason to believe that this would be morevalid than assuming no change. Because the deliveryschedule was fixed for each shuttle bus system consid-ered, the simulations were deterministic (in otherwords, the results of multiple simulations of a givenshuttle bus delivery schedule would always be thesame).

Model Validation—A series of 48 model runs wasconducted to validate the simulation model output.The number of vehicles entering the park was variedto match the number of vehicles entering the park onthe 4 days that parking lot counts were conducted. Themodel runs were repeated 12 times for each of the fouruse levels to capture stochastic variation. For each ofthe four total use levels modeled, the average numberof vehicles in selected parking lots was calculated andcompared to the actual parking lot counts.

Model Output

Use Levels and Patterns—Estimates of the num-ber of people at one time (PAOT) at Delicate Archthroughout the hours of a “typical” peak season dayare presented in figure 22. Estimates of the number ofvehicles along a selected road segment and in theDelicate Arch parking lot throughout the hours of a“typical” peak season day are presented in Figures 23and 24, respectively.

Alternative Transportation System—Without amandatory shuttle bus system, the model estimatesthat a maximum of 315 people can be allowed to hiketo Delicate Arch between the hours of 5:00 a.m. and4:00 p.m. without exceeding 30 PAOT at Delicate Archmore than 10 percent of the time. Results of simulationruns conducted to test the effect of implementing amandatory shuttle bus system are reported in table 9.The data in the third and fourth columns suggest thatthe maximum allowable use of the Arch could beincreased by 29 to 68 percent if visitors were requiredto ride shuttle buses to the Delicate Arch parking lot.For example, the model estimates that a shuttle bussystem designed to deliver visitors to Delicate Archevery 60 minutes would increase the maximum allow-able use of the Arch from 315 hikers between the hoursof 5:00 a.m. and 4:00 p.m. to 407 hikers. Further, theresults suggest that smaller, more frequent shuttlebuses would increase the maximum allowable use ofDelicate Arch to an even greater extent.

Model Validation—Table 10 presents validationresults based on comparisons between actual parkinglot counts and model estimates for three parking lotsand parkwide. The fact that cars were only counted for4 days precludes statistical comparison of observed andpredicted counts. It is also not possible to assess theaccuracy of model outputs, such as those in table 9.However, the relatively small difference in means forobserved and predicted counts suggests that the modelperformed as expected.

Conclusion

This case study demonstrates how computer simu-lation modeling can be used to describe visitor usepatterns and levels throughout a dispersed recreationarea. The findings also illustrate the capability ofsimulation modeling to conduct a more comprehensivemonitoring program than may be feasible by relyingsolely on on-the-ground monitoring techniques. Moni-toring indicators such as PAOT by means of on-the-ground counts can be very time consuming and expen-sive. However, once a computer simulation model ofvisitor use has been developed, PAOT can be esti-mated relatively easily.


Figure 22—Model estimates of people at one time at Delicate Arch.

Nu

mb

er o

f V

ehic

les

0

10

20

6am 8am 10am noon 2pm 4pm

Figure 23—Model estimates of vehicles at one time on a 2.4 km segment of road.


Table 9—Estimated maximum allowable use of Delicate Arch with mandatory shuttle system.

Arrival interval Estimated maximum Percent increase in(minutes) Passengers daily use allowable use

60 37 407 2930 21 462 4715 12 528 68

Table 10—Parking lot car counts—number ob-served and number predicted by themodel.

Observed Model

Delicate Arch 29 25The Windows 25 30Devil’s Garden 67 69All parking lots 40 41

Figure 24—Vehicles at one time in the Delicate Arch parking lot.

The findings from this project suggest that requir-ing visitors to ride a regularly scheduled shuttle busto the Delicate Arch parking lot would increase themaximum allowable use of the Arch. More generally,this project demonstrates the capacity for computersimulation modeling to assist managers in assessing

the effectiveness of alternative transportation sys-tems in a manner that is more cost effective and lesspolitically risky than on-the-ground trial-and-errorapproaches.

References

Manning, R.; Lime, D.; Freimund, W.; Pitt, D. 1996a. Crowdingnorms at frontcountry sites: a visual approach to setting stan-dards of quality. Leisure Sciences. 18: 39–59.

Manning, R.; Lime, D.; Hof, M. 1996b. Social carrying capacity ofnatural areas: theory and application in the National Parks.National Areas Journal. 16: 118–127.

Manning, R.; Lime, D.; Hof, M.; Freimund, W. 1995. The visitorexperience and resource protection (VERP) process: the applica-tion of carrying capacity to Arches National Park. The GeorgeWright Forum. 12: 41–55.

National Park Service. 1995. The visitor experience and resourceprotection implementation plan: Arches National Park. Denver,CO: U.S. National Park Service, Denver Service Center.

25

50

4pm0

Nu

mb

er o

f V

ehic

les

6am 8am 10am noon 2pm


Isle Royale National Park:Estimating the Effectiveness ofAlternatives for Managing Crowdingat Wilderness Campsites _________

Steven R. LawsonRobert E. ManningAnn Mayo Kiely

Purpose

The purpose of this case study is to demonstrate howa computer simulation model can be used to identifymanagement actions that are effective in bringingconditions into compliance with standards. The studywas conducted at Isle Royale National Park, located inthe northwest corner of Lake Superior, approximately120 km from Houghton, Michigan, and 30 km fromGrand Portage, Minnesota (Lawson and Manning2003a; Lawson and others 2003). Approximately 99percent of the park’s land base is designated wilder-ness. The park has a system of 36 campgrounds, witha total of 244 designated tent and shelter sites dis-persed along lakeshores, and a network of 260 km oftrails. Primary recreation activities at the park, whichis open to visitors from mid-April until the end ofOctober, include hiking, camping, and boating.

In recent years, recreation use of Isle Royale Na-tional Park has increased steadily. The park is par-ticularly popular for backpacking and backcountrycamping, and as a result, overnight backcountry usedensities are among the highest in the National ParkSystem (Farrell and Marion 1998). Isle Royale Na-tional Park’s approach to backcountry camping man-agement is designed to maximize public access andmaintain visitors’ sense of spontaneity and freedom.Visitors interested in backcountry camping at IsleRoyale National Park are required to obtain a permit,but there is no limit on the number of permits, andvisitors are not required to follow fixed itineraries.While visitors do have the option to obtain specialpermits for offtrail hiking and camping, the vastmajority choose to camp at the designated camp-ground sites (Farrell and Marion 1998). Consequently,backcountry campground capacities are commonlyexceeded during peak periods, causing visitors to haveto “double-up” at campsites with other groups. Be-cause previous research at the park suggests that forsome visitors, having to “double-up” detracts from thequality of their experience (Pierskalla and others1996, 1997), the park has considered establishingstandards for the maximum proportion of backpack-ers who would have to “double-up” at campsites. Man-agers needed insight into what actions might have tobe taken to achieve these standards.

The simulation model was designed to achieve threerelated objectives: (1) to describe the current spatialand temporal extent of campsite sharing in the park;(2) to assist park managers in identifying manage-ment actions capable of reducing or eliminating camp-site sharing; and (3) to enhance the effectiveness ofpublic involvement processes by providing managerswith precise, quantitative estimates of the effective-ness of management alternatives.

Data Collection

Backcountry camping permits issued by park staffduring the 2001 season provided the primary source ofdata needed to construct the computer simulationmodel. Data needed to test whether the simulationmodel outputs are valid estimates of on-the-groundconditions were gathered through a series of camp-ground occupancy observations conducted throughoutthe park’s 2001 visitor use season. Specifically, back-country rangers and campground hosts recorded thenumber of groups camping in selected campgroundson several nights throughout the visitor use season.

Constructing Model Inputs From PermitData

Information from the permits concerning each back-country camping group’s starting date, camping itin-erary, and group size were used as inputs to constructthe simulation model. Starting dates reported on thebackcountry camping permits were used to determinethe total number of trips starting on each day of the2001 visitor use season. From these data, the averagenumber of trips starting on weekdays and weekenddays from Rock Harbor, Windigo, and all other loca-tions was computed. Within the model, for each of thethree simulated entry points, an exponential distribu-tion with a mean equal to the mean number of tripstarts for the corresponding location and day of theweek was used to generate simulated backcountrycamping trips.

Information about group sizes reported on the back-country camping permits was used to compute thepercentage of small (six or fewer people) and large(more than six people) visitor groups during the 2001season. Within the simulation model, this informationwas used to assign visitors a group size (small orlarge).

Trip itineraries recorded on the permits were seg-mented by trip starting location and group size toconstruct a set of six probability distributions ofsimulated backcountry camping trips (small and largegroup itineraries for each of the three simulatedentry points). After being assigned a group size withinthe model, simulated visitor groups were assigned a


Table 11—Management alternatives quantified, based on simulation model output.

Wilderness Status Permit Fixed Campsite Temporalvalues quo quota itineraries construction redistribution

Public Current 22% 30% Current use Current useaccess use reduction in increase in (shift 22% of

July/August July/August peak)use use

Facility No new No new No new 13 new No newdevelopment campsites campsites campsites campsites campsites

Visitor No fixed No fixed Fixed No fixed No fixedfreedom itineraries itineraries itineraries itineraries itineraries

Camping 9% of 5% of <1% of 7% of 5% ofsolitude— groups share groups share groups share groups share groups shareJuly and August sites/night sites/night sites/night sites/night sites/night

Camping 0.4% of 0.4% of <1% of <1% of 1.4% ofsolitude— groups share groups share groups share groups share groups sharelow-use period sites/night sites/night sites/night sites/night sites/night

camping itinerary from the distribution itinerariesthat corresponded to their starting location and groupsize.

Computer Simulation ModelCharacteristics

Using the data inputs described above, a probabi-listic, steady-state simulation model of backcountrycamping use was developed (refer to Chapter 3 formore information about steady-state and probabilis-tic simulation modeling). The commercial simulationsoftware package, Extend, developed by ImagineThat, Inc., was used to develop the simulation model.The primary use of the simulation model developedwas to replicate backcountry camping use at the peakof the visitor use season in order to address theobjectives outlined earlier. Consequently, steady-state simulation was adopted for this study becauseit is designed to replicate system behavior over thelong run at a given level of production or capacity.

Simulation Model Runs

Simulation runs were conducted to estimate theextent of campsite sharing in the park under statusquo conditions, and to estimate the effectiveness ofmanagement actions at reducing or eliminating camp-site sharing. Management alternatives evaluated withthe simulation model include reducing use throughquotas on permits, requiring visitors to follow pre-scribed fixed itineraries, and building additional camp-sites within existing backcountry campgrounds.

A series of simulations was also conducted to vali-date the simulation model outputs. Data concerning

the average number of groups per night camping inselected campgrounds were gathered from 20 modelruns. These data were compared with campgroundoccupancy count data collected by park staff for thesame campgrounds during the 2001 visitor use sea-son. In addition, a workshop was conducted to instructpark staff how to use and modify the simulation modelto continue meeting their planning needs. The parkstaff’s use of the simulation model is ongoing, allowingthem to evaluate management strategies as new ideasemerge throughout the park’s backcountry and wil-derness planning process.

Model Output

Table 11 summarizes the results of simulation runsconducted to estimate the current extent of campsitesharing, and to estimate the effectiveness of alterna-tive strategies for reducing or eliminating campsitesharing. The alternatives outlined in table 11 wereselected for analysis with the simulation model be-cause they reflect a range of management approachesthat emphasize campsite solitude, visitor freedom,public access, and facility development to varyingdegrees.

Park managers have the option of managing back-country camping to maintain status quo conditions.Under this alternative, an average of about 39 per-mits would be issued per day, there would be no newcampsite construction, and visitors would not berequired to follow prescribed itineraries. Park plan-ners were interested in a standard of no more than 5percent campsite sharing. Simulation results for the“Status Quo” alternative suggest that under the park’scurrent management approach, an average of about


Table 12—Mean campground occupancy rates, with numberof observations in parentheses, as observed andpredicted by the model.

Campground Observed Model

Daisy Farm—weekend 15.0 (15) 15.8 (16)Daisy Farm—weekday 15.1 (14) 13.3 (40)Belle Isle —weekend 3.1 (12) 3.8 (16)Belle Isle—weekday 3.1 (19) 3.1 (40)

9 percent of groups are required to share campsitesper night during July and August, with 24 percentsharing during the busiest 2 weeks of this period.Less than 1 percent of groups are estimated to sharesites during the low-use period of the season.

Simulation runs were conducted to assess the effec-tiveness of a permit quota at reducing or eliminatingcampsite sharing. Under the “Permit Quota” alterna-tive, there would be no new campsite construction andvisitors would not be required to follow prescribeditineraries. However, the average number of permitsissued per day during July and August would bereduced to ensure that an average of no more than 5percent of groups share campsites per night (a stan-dard for campsite sharing that the park is consider-ing). Such an approach would continue to emphasizevisitor freedom and limit facility development in wil-derness, while allowing for greater camping solitudethan the status quo for those groups able to obtain apermit. However, fewer individuals who want to takea backcountry camping trip during July or Augustwould be able to. The simulated “Permit Quota” alter-native suggests that the park would need to reducevisitor use during July and August by nearly 25percent to ensure that an average of no more than 5percent of groups share campsites per night.

Decisions to limit public use of public lands areinherently controversial. To avoid this controversy,park managers could institute a fixed itinerary sys-tem, rather than a permit quota, to reduce or eliminatecampsite sharing. Under this approach, visitors wouldbe assigned to campgrounds that had open campsites,and no new campsites would be constructed. However,visitors would have fewer choices of itineraries andwould lose the freedom to spontaneously alter theircamping itinerary during the course of their trip. Theresults of the simulated “Fixed Itineraries” alterna-tive suggest that, by requiring visitors to follow pre-scribed camping itineraries, the park could issue ap-proximately 30 percent more permits than they didduring the 2001 visitor use season, and at the sametime virtually eliminate campsite sharing.

Rather than institute a permit quota or requirevisitors to follow prescribed itineraries, managerscould try to reduce or eliminate campsite sharing bybuilding new campsites. The park’s recently adoptedGeneral Management Plan allows for construction ofup to 13 additional campsites in specific campgrounds.If this “Campsite Construction” alternative were to beadopted, the simulation results suggest that, withoutinstituting any limits on use, the park could reducecampsite sharing by about 2 percent, resulting in anaverage of approximately 7 percent of groups sharingcampsites per night.

As the results of the simulated “Status Quo” alterna-tive indicate, campsite sharing is a problem primarily

during the months of July and August, while there isvirtually no campsite sharing during the low useperiod of the season. Further, results of the “PermitQuota” alternative suggested that park managerswould need to reduce the number of permits issuedduring July and August by about 25 percent toensure that an average of no more than 5 percent ofgroups share sites per night. However, rather thanturning those visitors away completely, park manag-ers could shift “surplus” peak season use to the low-use period. This “Temporal Redistribution” approachwould allow managers to maintain season-wide visi-tor use levels, reduce campsite sharing during Julyand August, avoid building new campsites, and main-tain visitor freedom with respect to camping itinerar-ies. Results of the simulated “Temporal Redistribu-tion” alternative suggest that campsite sharing wouldincrease from an average of approximately 0.4 per-cent of groups per night during the low use period ofthe season, to just over 1 percent of groups per night.

Another option is to redistribute use spatially.Simulations conducted to estimate the effect of redis-tributing visitor use evenly across the two primarystarting locations for backcountry camping trips(Windigo and Rock Harbor) would not reduce camp-site sharing. Finally, redistributing use evenly acrossthe days of the week also would have no effect oncampsite sharing. The results of these simulationsare not included in table 11.

Results of simulation runs conducted to test thevalidity of the model outputs are summarized intable 12. There were no substantive differences be-tween the observed campground occupancies and thecorresponding model output. This suggests that thecomputer simulation model accurately representedbackcountry camping conditions at the park duringthe 2001 season.

Park staff’s use of the simulation model is ongoing.For example, park staff have used the model toestimate the effect of shifting some use to secondaryentry points, differentially altering the visitationlevels of hikers, paddlers, and powerboaters, andsetting alternative standards for campsite sharing atdifferent times of the season. In addition, park staffhave used the model to estimate where and how


many new campsites would need to be added to elimi-nate campsite sharing during peak season demand.Using simulation results as a guide, park staff con-ducted site visits to determine the feasibility anddesirability of campground development needed tomeet peak camping demand, based on considerationsof physical constraints of wetlands, fragile habitatsand topography as well as appropriate size of camp-grounds in different areas of the park. The number ofnew sites that would be needed to accommodate peakdemand, according to model estimates, is greater thanthe number of sites that could be added, given theconstraints listed above. However, the number offeasible new sites would mitigate campsite sharingduring the peak period of the season to some extent.

Conclusions

Findings from this project demonstrate how simula-tion modeling can be used to identify effective manage-ment actions and avoid those that are less effective.Furthermore, by providing park managers with pre-cise, quantitative estimates describing the effective-ness of management alternatives, the computer simu-lation model serves as a communication tool. Parkplanners have used simulation model data to betterinform the public of the costs and benefits of differentmanagement options, resulting in more effective pub-lic involvement processes. An additional use of themodel data is its incorporation in further research. Forexample, the model data generated in this study servedas the basis in designing a visitor survey in whichrespondents were asked to evaluate a range of man-agement alternatives (Lawson and Manning 2003b).The model data provided an important element ofrealism to the choices presented to respondents.

References

Farrell, T.; Marion, J. 1998. An evaluation of camping impacts andtheir management at Isle Royale National Park. Resources Man-agement Report. U.S. Department of the Interior, National ParkService, Research.

Lawson, S.; Manning, R. 2003a. Research to inform management ofwilderness camping at Isle Royale National Park: Part I—de-scriptive research. Journal of Park and Recreation Administra-tion. 21(3): 22–42.

Lawson, S.; Manning, R. 2003b. Research to inform management ofwilderness camping at Isle Royale National Park: Part II—prescriptive research. Journal of Park and Recreation Adminis-tration. 21(3): 43–56.

Lawson, S.; Mayo Kiely, A.; Manning, R. 2003. Integrating socialscience into park and wilderness management at Isle RoyaleNational Park. George Wright Forum. 20(3): 72–82.

Pierskalla, C.; Anderson, D.; Lime, D. 1996. Isle Royale NationalPark 1996 visitor survey: final report. Unpublished report on fileat: Cooperative Park Studies Unit, University of Minnesota, St.Paul, MN.

Pierskalla, C.; Anderson, D.; Lime, D. 1997. Isle Royale NationalPark 1997 visitor survey: final report. Unpublished report on fileat: Cooperative Park Studies Unit, University of Minnesota, St.Paul, MN.

Acadia National Park CarriageRoads: Estimating the Effect ofIncreasing Use on Crowding-Related Variables _______________

Robert E. ManningBenjamin Wang

Purpose

The purpose of this project was to create and applya computer simulation model to estimate use patternsalong the carriage roads in Acadia National Park in away that was meaningful in understanding and man-aging the quality of the visitor experience. Recreationuse of the park’s carriage road system has increaseddramatically in recent years, and this has causedconcern among both park users and managers. Domi-nant uses have evolved from horses and carriages (forwhich the nearly 80 km of roads were originally con-structed in the early part of the twentieth century) tohiking and, more recently, mountain biking.

Because of the complex, intersecting nature of thecarriage road system and its inherently dispersed use,direct, systematic observation of recreation use isdifficult. Thus, it is challenging to quantify recreationuse levels and patterns. How many visitors are usingthe carriage roads? How often do visitors encounterone another along the carriage roads? How manyvisitors can be accommodated on the carriage roadsbefore the quality of the recreation experience is di-minished to an unacceptable degree? The simulationmodel of the carriage road system was designed andused to help answer these and related questions.

Description of the Study Area

The carriage roads of Acadia National Park, onMaine’s Atlantic Coast 65 km southeast of Bangor, area unique system of more than 80 km of beautifullydesigned and highly engineered gravel roads builtunder the direction of John D. Rockefeller, Jr., in theearly 1900s. Although the roads were built for horse-drawn carriages, they are now used mainly by bicy-clists and walkers, providing a welcome escape fromautomobile traffic and access to many undevelopedareas of the park. Equestrian use is now low anddeclining. Longtime observers agree that carriageroad use increased greatly with the rise in popularityof the mountain bike in the 1980s, although no data oncarriage road use were collected during that timeperiod. However, the park fielded an increasing num-ber of complaints from visitors and area residentsduring that time about “crowding” and “conflict” onthe carriage roads.



Surveys of carriage road users indicated that thenumber of other visitors seen while hiking and bikingwas important in defining the quality of the recreationexperience (Jacobi and Manning 1999; Manning andothers 1998). Since use of the carriage roads is rela-tively high, the concept of persons-per-viewscape (PPV)was developed as an appropriate measure of carriageroad use. PPV was defined as the number of carriageroad visitors per 100 m of the trail system, the averagedistance that can be seen along the generally windingcarriage road system. Thus, the simulation model wasdeveloped to estimate PPV levels across the carriageroad system.

A variety of methods were employed to gather thebaseline data necessary for building the simulationmodel (Wang and Manning 1999). These were visitorcensus counts, onsite visitor surveys, Geographic In-formation System (GIS) analysis, a field visit, exami-nation of engineering maps, and computer timing ofvisitor arrival patterns. These are described in moredetail below.

Data on where, when, and how many visitors en-tered the carriage roads system were gathered usingfive 1-day visitor censuses. From 9:00 a.m. to 5:00 p.m.on each of the 5 days, observers stationed at all eightmain entrance points recorded the number of visitorsthat entered per hour. Census data suggest that about3,000 visitors per day use the carriage roads on anaverage summer day.

Information on visitor characteristics and travelpatterns were gathered with an onsite survey. A totalof 514 questionnaires were administered at the eightmain exit points. The number of surveys gathered ateach site was proportional to that site’s share of totalexits from the entire system as determined by a censusof visitor use conducted the previous summer. Onevisitor from each group was asked about their groupsize, their mode of travel (hiking versus biking), thetotal amount of time they had spent on the carriageroads that day, and where and how long they pausedduring the visit. Equestrian-related cases were notincluded in the study since these uses constitute sucha small percentage of all carriage road use. Finally,with the aid of a map, respondents were asked to list,in order, all of the carriage road intersections theypassed during their trip. This produced a total of 381unique travel routes along the carriage road system.

The lengths of carriage road sections between inter-sections were calculated from a digital coverage usingGIS analysis. The lengths of each unique route thatrespondents traveled were then calculated with ashort ARC Macro Language program.

A field visit to the carriage roads and an examina-tion of engineering maps determined that the lengthof a typical viewscape—the length of carriage road

that can be seen at one time—was approximately100 meters.

The precise timing of visitor arrival patterns wasmeasured using a Pascal computer program. Datawere gathered over four half-hour periods with similaruse levels near one entrance point on the carriageroads. Each time a visitor group arrived, a key was hiton the computer to record the time of arrival (down toa second) and travel mode (hiking or biking). Sixty-fivedata points were recorded. These data were gatheredto verify the use of a decaying exponential distribu-tion to simulate arrival patterns. The use of anexponential distribution assumes that visitor groupsarrive independently.


The simulation model was built using Extend, acommercial, general simulation software package de-veloped by Imagine That, Inc. The structure of themodel was built with hierarchical blocks that repre-sented specific parts of the carriage road system. Thethree main types of hierarchical blocks that comprisedthe model were entrance/exit blocks, intersectionblocks, and road section blocks.

The entrance/exit blocks were built to generate thesimulated visitor parties. Visitor parties were gener-ated using an exponential distribution varying aroundmean values from the census counts. The parties werethen randomly assigned travel modes (hiking or bik-ing) and group size, both according to probabilitydistributions derived from the visitor survey. Simu-lated visitor parties were then randomly assignedtravel speeds according to a lognormal distribution.The mean and standard deviation of the distributionwere calculated from the travel times reported bysurvey respondents and the lengths of their travelroutes. Lastly, visitor parties were randomly assigneda route identification number according to frequenciesof actual routes reported by survey respondents.

The intersection blocks were built to direct simu-lated visitor parties in the proper direction when theyarrive at carriage road intersections. Lookup tablesunique for each intersection direct each party towardthe correct next intersection as indicated by theirroute identification numbers and how many times, ifany, they have been through that intersection.

The road section blocks were built to serve twofunctions. The first was to simulate travel through theroad section by delaying simulated visitor parties forthe appropriate period of time, according to theirassigned travel speeds. The second function was togather PPV data. Within each road section block, asimulated 100-m road segment was built. The numberof visitors traveling on that 100-m segment was countedand recorded once every simulated minute.


The simulation model developed in this project wasprobabilistic (based on probabilities of simulated visi-tors selecting available travel itineraries) and wasterminating (an entire day was modeled).

Model Runs

The model was run 135 times to generate informa-tion on average PPV. Each run simulated carriageroad use from 9:00 a.m. to 5:00 p.m. (the hours of peakuse), but only recorded output from 10:00 a.m. to 5:00p.m. Output from the first hour was considered unre-liable because people who would have entered thecarriage roads before 9:00 a.m. were not included inthe simulation. The completed model was run for sixlevels of total daily use of the carriage roads (multiplesof the current use level of 3,000 visitors): 375, 750,1,500, 3,000, 6,000, and 12,000 visitors per day. Themodel runs were repeated five times at each use levelto capture stochastic variation and generate statisti-cal confidence intervals. PPV conditions were recordedfor four use zones: the entire carriage road system,road sections designated as high-use zones, road sec-tions designated as low-use zones, and the road sectionbetween intersections 6 and 9 (a particularly heavilytraveled section).

The model was also run for validation purposes.Data on how many simulated visitor parties exited ateach exit point each hour were gathered for 20 modelruns. Data on how many parties passed by the westside of Eagle Lake (the site of a permanent infraredtrail counter) were gathered for 16 runs.

Model Outputs

Results are presented first on PPV conditions, andthen on the results of model verification and valida-tion. Figure 25 summarizes results estimating PPVconditions across all carriage road sections in thesystem for different total use levels. The results of thesimulated days are expressed in the number of min-utes out of an hour that a typical visitor will seecertain selected numbers of PPV. For example, infigure 25, when 1,500 visitors use the roads in a day,a typical visitor would see no one else (0 PPV) for 48minutes out of an hour, one to five other visitors for11 minutes out of an hour, and six or more othervisitors for 1 minute out of an hour. Each data pointshown represents the mean values, rounded to thenearest minute, from five model runs. Standard de-viations calculated for these mean values range from1.10 minutes (total use 6,000, 0 PPV) to 0.03 minutes(total use 12,000, 21 to 30 PPV).

Figure 26 provides a comparison of PPV resultsamong the different use zones when 3,000 visitorsuse the system in a day (the approximate current use

level on the average day). The use zones are (1) low-use road sections, (2) all of the road sections in thesystem, (3) high-use road sections, and (4) the roadsection between intersections 6 and 9, the mostheavily used road section in the system. Each datapoint shown represents the mean values, rounded tothe nearest minute, from five model runs. Standarddeviations calculated for these mean values rangefrom 1.58 minutes (highest-use road section, 0 PPV)to 0.10 minutes (all road sections, 11 to 15 PPV).

Figure 27 compares observed data with theoreticaldistributions used in the model. Figure 27A shows theempirical distribution of interarrival times (amountof time between arrivals of visitor parties) gatheredwith the laptop computer at entrance points alongsidea theoretical exponential (M = 103.94) distribution.Figure 27B shows the distribution of average bikertravel speeds calculated from visitor surveys along-side a theoretical lognormal (M = 0.799 min/100 m, SD= 0.431) distribution. Figure 27C shows the distribu-tion of hiker travel speeds gathered with visitor sur-veys and calculated with a GIS program alongside atheoretical lognormal (M = 1.615 sec/100 m, SD =0.879) distribution. Visual comparison of these distri-butions suggests there is little reason to conclude thatthe data are poorly fitted by the theoretical distribu-tions used in the simulations.

Figure 28 summarizes output validation results.Results are shown for three comparisons betweenobserved data and model outputs: (A) the distributionof visitors across exit points, (B) the distribution ofvisitor exits across time, and (C) the distribution ofvisitors who passed by the Eagle Lake infrared trailcounter site through the hours of the day. Visualcomparison of the distributions suggests there is noreason to conclude that the observed data are poorlyfitted by the model output.

Conclusions

This case study illustrates the utility of computersimulation as a tool for estimating how hiking andbiking use are spatially and temporally distributedand how these use patterns vary with a range of uselevels. PPV outputs for different total use level condi-tions and use zones provide a sophisticated view ofcarriage road use that would be difficult to observedirectly, but that is strongly related to the quality ofthe visitor experience. A second phase of this projectinvolved a survey of carriage road visitors to helpdetermine PPV-related standards of quality (Jacobiand Manning 1999; Manning and others 1999, 2000).The computer simulation was then used to estimatethe maximum number of visitors that can be accom-modated on the carriage roads without violating PPV-related standards (Manning and others 1998).


Figure 25—Persons-per-viewscape (PPV)distributions for six total use levels, for allroad sections in the system.

Figure 26—Comparison of persons-per-viewscape (PPV) distributions among differ-ent use zones with 3,000 total system use.


References

Manning, R.; Valliere, W.; Minteer, B.; Wang, B.; Jacobi, C. 2000.Crowding in parks and outdoor recreation: a theoretical, empiri-cal, and managerial analysis. Journal of Park and RecreationAdministration. 18(4): 57–72.

Jacobi, C.; Manning, R. 1999. Crowding and conflict on the carriageroads of Acadia National Park: an application of the visitorexperience and resource protection framework. Park Science.19(2): 22–26.

Manning, R.; Valliere, W.; Wang, B.; Jacobi, C. 1999. Crowdingnorms: alternative measurement approaches. Leisure Sciences.21(2): 97–115.

Wang, B.; Manning, R. 1999. Computer simulation modeling forrecreation management: a study on carriage road use in AcadiaNational Park, Maine, USA. Environmental Management. 23(2):193–203.

Manning, R.; Jacobi, C.; Valliere, W.; Wang, B. 1998. Standards ofquality in parks and recreation. Parks and Recreation. 33(7): 88–94.

Figure 28—Frequency comparisons for validation ofmodel outputs.

Figure 27—Frequency comparisons for theverification of input distributions.

(A) Inter-arrival Times

0

5

10

15

20

251

5

45

75

10

5

13

5

16

5

19

5

22

5

25

5

28

5

31

5

34

5

Average Time Between Arrivals of Visitor Parties (sec)

Pe

rce

nta

ge

Fre

qu

en

cy

Observed Data

Exponential Distribution

(B) Biking Travel Speeds

0

5

10

15

20

25

30

35

0.1

25

0.6

25

1.1

25

1.6

25

2.1

25

2.6

25

3.1

25

3.6

25

4.1

25

4.6

25

Average Time Taken to Travel 100 meters (min)

Pe

rce

nta

ge

Fre

qu

en

cy Observed Data

Lognormal Distribution

(C) Hiking Travel Spe eds

0

5

10

15

20

25

30

35

0.2

5

1.2

5

2.2

5

3.2

5

4.2

5

5.2

5

6.2

5

Average Time Taken to Travel 100 meters (min)

Pe

rce

nta

ge

Fre

qe

ncy Observed Data

Lognormal Distribution

(A) Exit Points

0

5

10

15

20

25

30

35

40

Eagle

La

ke

Vis

itor

Ce

nte

r

Park

-

ma

n

Mt.

Place of Exit

Perc

enta

ge o

f T

ota

l E

xits Census Data

Model Output

(B) Exit Times

0

5

10

15

20

10

:00

-

11:0

0

11

:00

-

12:0

0

12

:00

-

1:0

0

1:0

0-

2:0

0

2:0

0-

3:0

0

3:0

0-

4:0

0

4:0

0-

5:0

0

5:0

0-

6:0

0

Time of DayP

erc

enta

ge o

f T

ota

l E

xits

Census Data

Model Output

(C) Eagle Lake Counter

0

5

10

15

20

10

:00

-

11:0

0

11

:00

-

12:0

0

12

:00

-

1:0

0

1:0

0-

2:0

0

2:0

0-

3:0

0

3:0

0-

4:0

0

4:0

0-

5:0

0

5:0

0-

6:0

0

Time of Day

Perc

enta

ge F

requency

Observed Counter Data

Model Output


Acadia National Park Scenic Roads:Estimating the RelationshipBetween Increasing Use andPotential Standards of Quality _____

Jeffrey C. HalloRobert E. ManningWilliam A. Valliere

Purpose

Automobile use is the primary means of transporta-tion to and through most National Parks, and formany visitors it is also the predominant means bywhich the park is experienced. Some National Parkunits, such as the Blue Ridge National Parkway, aredesigned primarily for automobile use. Others, suchas Acadia National Park on the Maine coast, rely onscenic roadways interspersed with pull-offs or sideroads to vistas to provide “the national park experi-ence” to a majority of users. Despite the importance ofautomobile use in the National Parks, little researchhas been conducted to examine crowding, automobilecongestion, and social carrying capacities on parkroads. This case study illustrates how a computersimulation model of automobile traffic in Acadia Na-tional Park can be used to help estimate the maximumnumber of vehicles that can be accommodated on thepark road system.


Established as the first eastern National Park, Acadiacurrently receives more than 3 million visits annually.The Schoodic Peninsula is one of three geographicallyseparate areas of Acadia, located approximately anhour’s drive from the main Mount Desert Island por-tion of the park. Frazer and Schoodic Points are themost frequently visited areas of this 900 ha portion ofthe park.

A scenic road is the sole travel route through SchoodicPeninsula. The road enters the park near Frazer Pointand follows the shoreline of the peninsula, reachingSchoodic Point prior to leaving the park. The roadpermits only one-way vehicle travel, except on accessroads to both scenic points. A segment of road betweenFrazer and Schoodic Points was modeled in this study.This segment was selected because it represented ageneralized section of the park scenic road.


Data collection consisted of two phases. During thefirst phase, visitors were asked questions regardingthe acceptability of different levels of road traffic.

Using computer-edited photographs, respondents wereasked about the level of use they (1) prefer, (2) consideracceptable, (3) consider so unacceptable they would nolonger visit, and (4) think the National Park Serviceshould allow. Mean responses for each of these fourevaluative domains (denoted as preference, accept-ability, tolerance, and management action, respec-tively) provide a range of potential standards from theperspective of current visitors.

Phase 2 of the study involved the collection of travelroute information using commercially available, hand-held Global Positioning System (GPS) units. Routedata were collected during 9 days in July, 2003. Occu-pants of the first 10 cars to enter Schoodic Peninsulaat the start of the daily data collection period (9 a.m.)were asked to participate in the study by carrying aGPS unit in their car. Participants were instructed notto leave the GPS units on their automobile dashboardssince preliminary trials had determined that unitswere susceptible to overheating and malfunction inthis circumstance. The GPS units were collected fromparticipants at the Schoodic Peninsula exit, and theroute data were downloaded to a computer. The GPSunits were then returned to the park entrance forcollection of additional route data throughout thesampling day.

The GPS units (GARMIN eTrex Legend) were set toautomatically record route points. Numerous pointswere collected for every route, with each point containinga latitude, longitude, and time. Specifications for theGPS unit state that location information is accurate towithin 3 m. An Arc Macro Language (AML) script wasthen used to extract data points of interest from the manyrecorded. The script did this by first setting circularzones around each intersection or point of interest. Theearliest and latest recorded data falling within thesezones were then extracted from the data set.

The final aspect of data collection, during phase two,involved counting the number of cars arriving duringhalf-hour intervals between 9 a.m. and 7 p.m. Theseheadway counts were used to determine meaninterarrival times for each half-hour interval.


A simulation model of the Schoodic Peninsula wasbuilt using Extend, a commercial, generalized simula-tion software package produced by Imagine That, Inc.Car travel through the Schoodic Peninsula was set upas a terminating, discrete event simulation model. Inother words, the simulation model was designed torelease individual cars over a 10-day peak period intothe system, each with its own unique set of informa-tion (in other words, attributes). The model, built fromExtend’s libraries of standard blocks, was organizedinto hierarchical blocks and overlaid on a map of the


Schoodic Peninsula to provide a more accurate visualrepresentation.

Cars are generated, released into the model, andassigned the necessary attributes in the entranceblock. The entry of cars follows a decaying exponentialdistribution, based on mean interarrival times for thehalf-hour intervals corresponding to the time of day inthe simulation. Once in the system, one of 193 uniqueroutes collected using GPS units are randomly se-lected and that route’s attributes are assigned to asimulated vehicle. The route attributes specify whethera simulated vehicle visits or bypasses Frazer andSchoodic Points, the amount of time spent at each ofthe two points, and the time required to travel on thepark road between Frazer and Schoodic Points. Theselection of a travel path based on its frequency ofoccurrence in the data set of empirically derived routesis characteristic of a probabilistic model. However, themodel was also set up with an option to route cars tothe next location specified in the route if the physicalcapacity of either of the points’ parking lots wasexceeded. In this way the model represents a hybrid ofboth probabilistic and rule-based models.

Model Outputs

The simulation model was run five times (50 days) atthe automobile density-related standards of qualityfor preference, acceptability, management action, anddisplacement (table 13). None of the standards for theroad segment were violated at the sampled use level(M = 415 cars/day). At increased multiples of thesampled use level, standards of quality were progres-sively violated, as measured by the percentage of timeout of standard (table 14).

These outputs can be used in many different ways.For example, additional runs of the simulation be-tween two and three times the sampled use levelshowed that at approximately 2.1 times the sampled

Table 13—Standards of quality for the scenic park road col-lected during Phase 1.

Standards of qualityType of standard (No. cars/quarter mile)

Preference 2.5Acceptability 7.5Management Action 8.5Displacement 12.7

Table 14—Percentage of time each standard of quality was violated on the park road.

Multiple of Managementcurrent use Preference Acceptability action Displacement

1 0.0 0.0 0.0 0.02 7.6 0.0 0.0 0.03 45.1 0.0 0.0 0.04 69.4 0.0 0.0 0.05 80.1 0.1 0.0 0.06 86.3 3.4 0.3 0.07 91.3 19.0 5.4 0.08 93.4 40.4 20.8 0.09 95.2 52.6 39.9 0.1

10 97.3 60.3 50.7 1.5

use level, the preference standard is violated 10 per-cent of the time. When multiplied by the sampled uselevel this indicates that 871 cars per day could beallowed into the Schoodic Peninsula without violatingvisitors’ preference for crowding on the park road morethan 10 percent of the time. Similar outputs were notgenerated for acceptability, management action, andtolerance standards because the multiples of currentuse at which the 10 percent threshold would be crossedwere beyond that deemed likely to occur, at least in thenear future.

The effect of the Frazer Point parking lot size oncrowding along the park roadway was also an outputof the simulation model. By tripping (virtual) digitalswitches located on the bottom right of the model, aphysical capacity could be imposed at both Frazer andSchoodic Point parking lots. When the parking lotsreached their designated capacity, cars were prohib-ited from entering and were forced to continue on thepark road to their next destination. However, becausethe Schoodic Point parking lot is located beyond thesegment of park road modeled for this study, it doesnot affect the crowding results. Restricting the num-ber of cars at Frazer Point to 24 (the current numberof spaces in the parking lot) had no statistically signifi-cant effect on roadway crowding at any multiple of thesampled use level (p > 0.05).


Conclusion

This case study illustrates how computer simulationmodels can estimate baseline use levels on park roadsand estimate how much use can be accommodatedbefore potential standards are violated. Results sug-gest that crowding on the road is not a problem atexisting use levels, even when cars are forced to bypasstheir visit to Frazer Point. However, if use doubled,managers may need to limit use if managing for apreference-based standard of quality.

Port Campbell National Park,Australia: Predicting the Effects ofChanges in Park Infrastructure andIncreasing Use__________________

Robert M. Itami

Purpose

The purpose of this case study is to demonstrate theuse of agent-based modeling and simulation to predictthe outcomes of increasing use and different manage-ment scenarios at a popular day-use site at Port CampbellNational Park in Victoria, Australia. The case studydemonstrates the modeling and simulation of multipletravel modes, including arrivals by bus and privateautomobile, and onsite pedestrian behavior. It illus-trates how to construct agent-based visitor modelsusing RBSim for existing conditions and for a newmaster plan with new parking facilities, pedestrianwalkways, visitor center, and toilet facilities. Finally,simulation modeling is used to examine the perfor-mance of a new master plan over a 10-year time periodand compare the results with a “do nothing” alternativewhich maintains the existing site management regime.In this case study, the “do nothing” alternative isreferred to as “Scenario 1”and the new master plan isreferred to as “Scenario 2.” This case study is describedin detail as an example of rule-based simulations.


Port Campbell National Park and the associatedBay of Islands Coastal Park are located on the GreatOcean Road, approximately 250 km west of Melbourne.Comprising 65 km of rugged and spectacular coastalscenery, the two parks are protected in a strip rangingin width from a few meters in the Bay of Islands Parkto 2 km within the Port Campbell National Park. Theparks have World Conservation Union (IUCN) rat-ings of Category II (National Parks) and Category III(National Monuments) and are designated for eco-system conservation and appropriate recreation and

protection of outstanding natural features, educa-tion, research, and recreation respectively. The PortCampbell National Park attracts large and steadilyincreasing numbers of visitors. The park’s popularityis enhanced by its proximity to Melbourne and thelarge number of tour buses that visit the site daily.

Modeling Two Scenarios at TwelveApostles

RBSim was used to examine the impact of changesin park infrastructure and increasing visitor ratesover a 10-year period on the Twelve Apostles site.Some 701,000 people visited the site in 2001/2002, andby 2006/2007 the number is expected to be 864,000.Figure 29 shows the layout plans for the two scenarios.Scenario 1 shows the conditions before development offacilities in the new master plan. Car and bus trafficentered the site from the Great Ocean Road andtraveled southwest onto a loop road, which was config-ured with car and bus parking around the perimeter ofthe road. Pedestrians then entered the viewing plat-forms from the loop road. They proceeded along woodenwalkways and viewing platforms located at the top ofthe cliffs overlooking the view of the Twelve Apostles.On busy days the loop parking filled quickly, causingdrivers to park along the entry driveway and alongGreat Ocean Road with resultant safety hazards.

Scenario 2 shows the layout anticipated in the newmaster plan. In this scenario the viewing platformsand walkways in the viewing areas remain the same,but vehicular parking is located to the north of theGreat Ocean Road. A pedestrian underpass allowspedestrians to pass safely under the Great Ocean Roadto the viewing platforms. Major facilities provided inthe new master plan include a new parking area forbuses, cars and trailers, a new visitor center withpublic toilets, and a new walkway extending from thevisitor center to the viewing platforms.

Model Inputs

Data inputs for RBSim models fall into two catego-ries: site characteristics and visitor characteristics.Site characteristics include a GIS map of the road andtrail network, a Digital Elevation Model (DEM), aninventory of visitor facilities (including location, ca-pacity, and typical visit durations), and the rate ofarrival of visitors in cars, buses, or other travel modes.Visitor characteristics include arrival curves, rules ofbehavior, and typical trips.


Road and trail network—The existing road andtrail network for Scenario 1 was imported from ParkVictoria’s corporate GIS database. RBSim has utili-ties for importing GIS data from either MapInfo tab


Figure 29—Network layout and facilities for Scenario 1 and Scenario 2.

files or ESRI Shape files. The new road and trailnetwork for Scenario 2 was provided in the form of anAutoCAD drawing file. The network for car and busparking, pedestrian walkways, and visitor facilitieswas simplified and generalized (while maintainingpositional accuracy) and hand digitized with RBSim’sdigitizing functions. RBSim uses network algorithmsfor calculating travel time, travel distances, andshortest path from one destination to the next. Thisrequires that the travel network is structured in sucha way that the endpoints of all links connect to nodes,and that there is a node at the intersection of three ormore links. The RBSim network import utility at-tempts to construct this topology by breaking links atintersections and inserting nodes, joining the end-points of links according to user-specified tolerances,and building a tabular database that describes thetopology of the network in the form of a forward stardatabase. RBSim provides network editing functionsfor correcting network topology and adding or deleting

links and nodes while maintaining network topology.Once the network topology is corrected, the networklinks are attributed for the maximum travel speed. Inthe case of the main highway (Great Ocean Road) themaximum speed is 90 km/hr. For entry roads themaximum speed is 15 km/hr, 10 km/hr for parkingareas, and 4 km/hr for trails.

Digital elevation model (DEM)—The DEM repre-sents the elevation of the land surface and is stored asa grid matrix of elevations. The digital elevationmodel for Port Campbell National Park was gener-ated in ESRI ArcInfo from contour elevations andspot heights derived from 1:25,000 topographic mapswith 10-m contour intervals. Elevations were inter-polated on a 20-m floating point grid and importedinto RBSim in ESRI grid export format. The DEM isused by RBSim to calculate uphill and downhillslopes to determine walking speeds for pedestriansand to calculate line-of-site across terrain forintervisibility calculations. If a DEM is not specified,


Table 15—Facilities at the Twelve Apostles locale before (sce-nario 1) and after (scenario 2) the implementation ofthe new master plan.

Facility Scenario 1 Scenario 2

Viewing platform 345 people 345 peopleInformal lookout 5 people 30 peopleBus park 6 buses 12 busesCar park 28 cars 165 carsVisitor centre None 100 peopleToilet None 29 peopleTrailer park None 12 cars

RBSim assumes a flat surface with an elevation ofzero.

Facilities inventory—Results of an onsite facilitiesinventory to determine the type and capacity of visitorfacilities were entered into the Scenario 1 network asnode attributes. For Scenario 2, parking spaces forcars, buses and trailers were counted from the draw-ings provided by Parks Victoria. The number of toiletsand the capacity of the visitor center were determinedfrom design documentation for the 2001 master plan.The facilities for both scenarios are enumerated intable 15.

Locales—A locale is a collection of one or more nodeswith associated facilities that have a shared identityand can be grouped based on proximity or commonaccess. The Twelve Apostles visitor site, includingviewing platforms, boardwalks, parking lots, and trailsis an example of a locale. Locales are defined in thetravel network by selecting all nodes to be included inthe locale using the RBSim network editing functions,and then assigning the “Locale” attribute to the nodes.This function assigns all the selected nodes the samelocale name. The locale is important so that site-specific rules can be created for agents. Also, agentrules are only active when an agent enters a locale,otherwise the agent follows its global trip plan. Eachlocale must also have at least one entry node. In thecase of the Twelve Apostles locale, the entry node isassigned to the intersection of the Great Ocean Roadand the entry road for both Scenarios 1 and 2.


Agents—The visitor simulation for Twelve Apostlesutilizes RBSim’s capability of using autonomous in-telligent agents to simulate onsite visitor behavior.For both Scenario 1 and Scenario 2, two agents weredefined to represent visitors arriving by bus andthose arriving in private automobiles. This reflectsthe source of visitor data from road traffic counts,which differentiates between buses and cars. In thissimulation, because of the simplicity of the site, themain constraint on site behavior is the duration of

stay. Agents are assigned two different travel modes.The “default” travel mode is walking by foot. Agentsare assigned a second “Arrival” travel mode as partof the typical trip. In this case the arrival travel modefor automobiles is car, and the bus arrival travelmode is bus. Agents are also assigned a maximumtravel speed for their default travel mode. In bothcases the maximum travel speed is 4 km/hour.

Arrival curves—Parks Victoria has a policy of man-aging visitor use for the 95th percentile busiest day.This means facilities should be planned for the 19thbusiest day of the year. From several detailed trafficanalyses conducted over a period of 5 years, it wasdetermined that on the 95th percentile day (EasterSaturday), 44 buses visited the Twelve Apostles. In2006, at 7 percent growth per year, this is estimated torise to 62 buses (40.3 percent growth from 2001). In2011, visitation is estimated at 87 buses (96.7 percentgrowth from 2001).

Car projections are factored for the average growthof Twelve Apostles visitation (3.5 percent). On the95th percentile day (Easter Saturday), 1,589 carsvisited the Twelve Apostles. In 2006, at 3.5 percentgrowth per year, this number is estimated to rise to1,887 cars (18.8 percent growth from 2001). In 2011,visitation is estimated at 2241 cars (41.1 percentgrowth from 2001). Figure 30 shows detailed arrivalcurves for cars; similar data were developed for buses.

Agent rules—Agent rules are a set of user-definedbehaviors that are defined using a stimulus/responseor event/action framework. RBSim exposes runtimeproperties of the network, agent, and global events.Each of these properties will have a state or value thatcan be defined as a stimulus or event. Boolean logic canbe used to combine two or more stimuli to createcomplex conditions for behavior. Behavior is definedas a directive to search for a facility. An example of acomplex rule is:

If (TravelMode = ‘Car’ AND Locale = ‘12 Apostles’AND LocaleEntry = True) THEN Find Carpark

Agent rules are assigned to agents in the manage-ment scenario builder. The user can specify the orderin which the agent considers rules for execution. Forinstance, an agent should always park a car beforegoing to a visitor center. Agent rules are a set of user-defined agent actions that are triggered by changes inthe agent’s location, travel mode, or other characteris-tics of the network or agent. An interface for designingrules exposes the set of simulation properties that canact as triggers. Agent rules are active within thecontext of a locale.

Table 16 shows the agent rules used in the scenarios.Scenario 1 used rules 1, 2, and 5; and Scenario 2 usedall five rules. These rules direct the behavior of theagents once they enter the Twelve Apostles locale.


Figure 30—Car arrival curves used in the simulations for Twelve Apostles, for 2001Easter Saturday traffic count data as well as projections for 2006 and 2011.

Table 16—Rules used by agents during the simulation runs.

RuleID Rule

1 If Arriving at a locale in a car then find a car park.2 If arriving at locale entry in a bus then find bus parking.3 When at Twelve Apostles find visitor centre.4 At Twelve Apostles find toilet.5 At any locale find viewing platform (repeatedly).

Table 17—Typical trips, with durations generated randomly by the simulator between theminimum and maximum duration (in minutes) for cars and buses.

Description Min duration Max duration Travel mode

Buses Easter Saturday 2001 60 80 BusCars Easter Saturday 2001 30 40 Car

Note that special rules (3 and 4) are needed in Scenario2 for the new visitor center and toilet facilities.

Typical trips—The typical trip represents a patternof visitation within the park. The typical trip is definedby an entry and exit node to the travel network, one ormore destinations, a travel mode (car, bus, and so on),

a set of rules, trip duration and an arrival curve. Twotypical trips were assigned to each scenario—one forbuses, the second for cars. The trips to Twelve Apostlesare all round trips, so the entry and exit nodes are thesame (just offsite). The destination node is the en-trance to the site on Great Ocean Road. Buses followrules 2, 3, 4, and 5. Cars follow rules 1, 3, 4, and 5. Table17 shows the duration for each type of typical trip. Theduration for the trip varies randomly within the rangespecified. The durations are estimated from analysisof traffic count data.

Model characteristics—All agents follow a globaltrip plan. These plans however provide only a generaltrip itinerary. Once the agent begins its trip, changingconditions of the network (facilities becoming full),global events (rain storms), and agent states (agentfitness, running short on time), can all act together tochange the behavior of the agent according to rules


and the internal way-finding logic of the agent. Theway finding reasoning of an agent is influenced by thefollowing factors: (1) available time (defined by timeelapsed subtracted from total trip duration), (2) travelmode as it affects travel time, (3) agent preferences, (4)list of rules and their order, (5) currently executingrules, (6) internal state of the agent, (7) current loca-tion of the agent, (8) condition of the network includingavailability of facilities, access restrictions, and traveltime to destinations on the network, and (9) previouslyvisited destinations.

When an agent arrives at a locale, it checks to see ifthere is a duration set for this locale in the global tripitinerary. If the duration is greater than 0 then theagent checks to see if there is enough time left in itstotal trip duration by subtracting the time elapsedsince the beginning of the trip and the time to travel tothe exit node. If the remainder is positive and greateror equal to the duration set for this locale, the agententers the locale and performs an initialization proce-dure. Once the locale network has been initialized, theagent then evaluates all possible combinations ofdestinations from its current node location. The agentthen evaluates each path and rejects any path thatexceeds the available locale visit duration. The re-maining paths are then ranked to maximize the sitepreferences and contain facilities that are on theagent’s current rule list. A gravity model is used toweight the paths so paths with high-priority facilitiesare ranked higher for facilities close to the agent’scurrent location.

Once the preferred path is selected, the agent loadsit as its current trip itinerary. The agent then traversesthis itinerary as far as it can in the current time step.If the agent encounters a node that contains facilitiesthat are on its current rule list, the agent changes itsinternal state to “visiting facility” and generates avisit duration for that facility. If the facility at the nodehas no available capacity (for example, the parking lotis full), the agent “looks ahead” on its itinerary to seeif a facility of the same class is available. If there is, theagent then continues its trip toward that node. If thereis no other facility of the same class, the agent will thenchange its state to “queuing” and waits until thefacility becomes available.

During each iteration of the simulation, the agentmust check its available trip time, current travelmode, current rule list, and current state. Any of thesecan trigger a change in behavior. The agent mayabandon its current trip and calculate a path back toits car, or to the exit. If the conditions have notchanged, then the agent continues to execute its cur-rent behavior.

Although there are many more details to this behav-ior, the above reflects the overall logic behind the

agent’s way-finding logic. When implemented, thelogic produces behavior that appears “smart” in thatthe agent generates logical paths and exhibits behav-ior that is humanlike.

Simulation Outputs and ManagementRecommendations

Several model outputs were produced to both pre-dict the future effect of increasing use on currentfacilities and to compare Scenarios 1 and 2. TheRBSim simulation runs for 2001, 2006, and 2011produce visitor loads and capacities for each facilityin the simulation including car and bus parking,visitor center, restrooms, and viewing platforms. Thestatus of each facility is written to a database at eachtime step of the simulation. These outputs are thensummarized for each hour of the day.

Trip Completion Rates—Trip completion ratesmeasure the failure rate of visits to the Twelve Apostlessite. If all parking is full, agents cannot enter the site,and the trip is counted as failed.

Trip completion rates in 2001 are around 100 per-cent, primarily because the available capacity (aver-age or minimum) never reaches zero (fig. 31). In 2006,with average car park capacity projected to fall to 5percent with current facilities and minimum capacityprojected to reach zero for at least three hours of theday, trip completions fall to the 91 to 95 percent range.As visitors arrive at the facility and find no parkingspaces, they are forced to leave the locale. By 2011,with average car park capacity around 3 percent for 3hours of the day, and minimum capacity at zero forapproximately 5 hours, trip completions fall to 80percent during peak loading.

Visitor Encounters—Visual encounters are gen-erated using line-of-sight calculations between eachagent and all other agents within a 200-m radius.Screening effects of terrain are also incorporated intothe calculation. The summaries can be generated forany link, node, or facility, or summaries can be madefor each individual agent. Figure 32 shows the averagevisual encounters per visitor at lookouts. The numberof encounters peaks around 3:00 in the afternoon witharound 60 people visible at one time.

Queuing Times for Parking—RBSim simulatesqueuing behavior for parking with current facilities.The simulator records for each agent the total time inminutes that each party waits for parking. Projectionsin figure 33 show that average queuing, by 2011, willbe double that of 2001. Average peak queuing time isexpected to rise to approximately 1.25 minutes by2006 and to almost 2 minutes by 2011.


Figure 31—Trip completion rates in 2001 and projected rates for 2006 and 2011.

Figure 32—Average visitor encounters at lookouts in 2001, 2006, and 2011.

Current Facility Projections - 2001 v 2006 v 2011

Trip Completion Rate

0.0%

10.0%

20.0%

30.0%

40.0%

50.0%

60.0%

70.0%

80.0%

90.0%

100.0%

7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

Hour of the Day

Tri

p C

om

ple

tio

n R

ate

(%

)

Current Facility - 2001




Average Visitor Encounters at Lookouts

0

5

10

15

20

25

30

35

40

45

50

55

60

65

7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23

Hour of the Day

Avera

ge V

isit

or

En

co

un

ters





Figure 33—Average queuing times at car parks in 2001, 2006, and 2011.


Average Queuing Time at Car Parks

0.00

0.10

0.20

0.30

0.40

0.50

0.60

0.70

0.80

0.90

1.00

1.10

1.20

1.30

1.40

1.50

1.60

1.70

1.80

1.90

2.00

7 8 9 10 11 12 14 15 16 17 18 19 20

Hour of the Day

Avera

ge Q

ueu

ing

Tim

e (

min

ute

s)




13

Scenario 1 vs Scenario 2

Average Available Car Park Capacity

0%5%

10%15%20%25%30%35%40%45%50%55%60%65%70%75%80%85%90%95%

100%

7 8 9 10 11 12 13 14 15 16 17 18 19 20 21

Hour of the Day

Scenario 1 -2001

Scenario 2 -2001

Scenario 2 Parking: 165 spacesScenario 1 Parking: 28 spaces

Av

era

ge

Ca

pa

cit

y A

va

ila

ble

(%

)

Figure 34—Average available parking capacity for Scenario 1 and Scenario 2 for 2001 traffic data.


Car Park Capacity—Figure 34 shows the averageavailable capacity for the car park over the day. At 100percent, the car park is empty; at 0 percent, all spacesare taken. The figure shows that for Scenario 1 theparking lot is full from 11:00 a.m. until 5:00 p.m. InScenario 2 there currently is excess capacity through-out the day.

Management Implications

Simulation results predict that bus parking will beinadequate during the busiest time of the day between2:00 and 4:00 p.m. by the year 2006. This shortage isexacerbated by the year 2011, as bus parking is inad-equate for the whole period from 3:00 to 5:00 p.m. By2006 the car park is full from 1:00 to 4:00 p.m. By 2011the car park is full from 12:00 to 5:00 p.m.

The longer walk from the new parking facilities tothe viewing platforms extends the average length ofstay an average of 6 to 7 minutes. As the number ofvisitors increase, there is increasing pressure on view-ing platforms and lookouts. Crowding increases be-cause of the increased duration of stay and the in-creased capacity of car parks. It is expected that visitorsatisfaction will decrease with an increase in visitors.This is caused by increased queuing times at parkinglots, an increase in the length of stay, the number ofvisual encounters, especially at viewing platforms,and the number of visits that fail because of lack ofparking at peak periods. This can partially be resolvedby increasing the capacity of viewing platforms, butthe long-term solution will require redistributing thevisitors to other sites, especially at peak periods. Busparking will need to be managed between 3:00 to 5:00p.m. within 5 years (for example, use informal spacesnear the visitor center). Car arrivals will have to belimited after 1:00 p.m. in 10 years or an extension tothe car park will have to be built. Viewing platformswill have to be increased in capacity in the 5- to 10-yeartime horizon if the overflow car park is used or if thecar park is extended further.

Mount Rainier National Park:Collecting Data to Model VisitorUse on a Complex FrontcountryTrail System ____________________

Mark E. Vande Kamp

Purpose

The purpose of this case study is to illustrate noveldata collection procedures that were developed tomodel a complex frontcountry trail system in MountRainier National Park.

On clear summer days when the avalanche lilies arein bloom and blue ice is visible in the glacial crevasseshigh above, it is easy to understand why ParadiseMeadow is the most heavily used area in MountRainier National Park. In 2003 an estimated 370,000people drove, rode, or hiked to the lodge, visitor center,and climbing facilities at 5,400 feet on the south flankof the mountain (NPS 2004). Of those, about 70 per-cent took walks or hikes on the system of paved andgravel trails (fig. 35) in the subalpine meadow north ofthe visitor center and lodge (Vande Kamp 2001). Thosewalks and hikes are an important aspect of manyvisitors’ experiences of Mount Rainier (Johnson andothers 1991). At the same time, the level of visitationin the meadow has created negative impacts on thephysical resources and the quality of visitor experi-ences found there. Offtrail hiking has damaged veg-etation in many areas of the meadow (Rochefort andSwinney 2000), and at peak times, visitor movementon popular trails is impeded by high visitor density(Vande Kamp and Zweibel 2004).

To set policies that protect the physical resourcesand the quality of visitor experiences, managers ofMount Rainier seek to understand the relationshipsbetween manageable aspects of visitation (where visi-tors go, what visitors do, how many visitors are present)and important resources or experiences directly threat-ened by visitation. Toward this end, Mount Rainier isdeveloping computer simulation models of visitationpatterns in Paradise Meadow that can be used:

• To estimate potentially informative measuresof visitation that are difficult to observe di-rectly (such as the square cm of trail per hiker)and to relate those estimates to more easilymeasured and managed measures of visita-tion (such as the number of vehicles in theparking lot).

• To identify “bottlenecks” in the trail systemwhere hiker density is highest and changes intrails or hike routing information might bemost effective.

• To explore the impacts of hypothetical changesin management policy, trail construction, orvisitation yet to occur in the real world (whileacknowledging the assumptions necessary forsuch prediction).

The Challenge of Modeling a ComplexTrail System

Simulation models have generally been applied toeither complex trail systems that are remote andextensive, such as wilderness areas, or simple trailsystems that are readily accessible, such as overlooksand visitor facilities (refer to other case studies inthis report). The itinerary information necessary to


develop simulation models of those environmentshas most often been collected by contacting visitorsafter their hikes and recording the route and timeinformation that they recall. This method can yieldaccurate itinerary data for both types of situationsbecause hikers in remote and complex trail systemsusually pay close attention to navigation, and hikersin simple systems don’t have complex itineraries torecall. An alternate method of collecting itinerariesin which visitors to remote and extensive systems aregiven map-diaries before their hikes and asked to

Figure 35—Trail system and waypoint sign locations at Paradise Meadow.

record their movement has also been used to obtainaccurate itinerary data.

It is more difficult to collect accurate itinerarydata in Paradise Meadow because the trail system isboth complex and readily accessible. Many visitorsdo not have a clear destination when entering themeadow, and they meander through the trail sys-tem without maintaining their geographic orienta-tion. Visitors to specific locations such as Alta Vistaor Myrtle Falls often depend on directional signs forguidance and do not know which of several possible


routes they followed. Recall of hiking itineraries insubsequent interviews is often vague and inaccu-rate. Providing a map-diary and asking visitors tokeep a detailed record of their movement is alsoproblematic. Hikers who keep such itinerary recordsare required to pay considerable attention to boththe map and the environment. Such attention islikely to alter their hiking itineraries and thus leadto simulation models that accurately estimate thevisitation patterns of hikers who attend to maps butnot the visitation patterns of typical visitors.

A method of collecting itinerary information is neededthat (1) can provide detailed route and time informa-tion and (2) poses a small burden on hikers that isunlikely to alter hiking behavior. The rest of this casestudy describes a method of collecting itinerary infor-mation that is particularly well suited for use incomplex and readily accessible trail systems like Para-dise Meadow. The waypoint card method combineswaypoint signs with recording of the time when visi-tors pass strategic locations to create a method ofcollecting itinerary data that is simple for visitors tocomplete but does not require large numbers of surveyworkers.

Methods

The goal of sampling was to produce a randomsample of visitor parties entering Paradise Meadowvia four access points leading from the parking lots orvisitor center. Figure 35 shows the trail system withthe four entry points labeled A, B, C, and D. A surveyworker was stationed at each entry point and in-structed to approach every third party who passed. Ifthe number of survey workers is smaller than thenumber of entry points into the trail system, a morecomplex sampling schedule in which workers rotate tothe different entry points can be used to produce arepresentative sample of hiker itineraries. The samplewas collected on August 23 and 24, 2003, between 9:00a.m. and 6:00 p.m.

When approaching a hiking party, survey workersintroduced themselves, stated that they were conduct-ing a study of visitor movement for Mount Rainier, andasked one member of the party to participate in thestudy. A laminated copy of the Office of Managementand Budget disclosure statement was handed to thosevisitors who agreed to participate. Then the surveyworker asked for the visitor’s home zip code, the size ofhis or her party, the number of party members whowere less than 18 years old, and the party’s hikingdestination. The survey worker then instructed therespondent how to complete the remainder of thesurvey by explaining:

There are waypoint signs like this one (pointing to theentry waypoint) placed along trails throughout themeadow. Every time you pass one of these signs, please

write the letter of the sign and the time on your card.I’ll fill out this first one for you (writing down the timeand entry waypoint letter on the card). If you pass thesame point more than once, please write down the timeeach time you pass. There are pencils and clocks oneach sign. When you are done hiking, please leave yourcard with the survey worker where you leave themeadow, or place the card in the survey box like thisone (point to exit sign and box) if no one is there.

Most respondents returned their cards directly to asurvey worker. When they did, the worker asked themif there were any places that they stopped for morethan 5 minutes during their hike. If so, the workernoted the location and duration of the stop on thewaypoint card.

Figure 36 shows the front and back of the waypointcards on which data were written. The cards wereprinted on waterproof card stock and included a twist-tie that could be used to attach them to a belt loop orbackpack strap so that they would be handy for re-spondents to write on as they hiked. Figure 37 showsthe front of a waypoint sign. Each sign was printed on8.5- by 11-inch card stock, attached to a metal stake,and covered with a clear plastic bag to protect it fromrain or fog. A small LCD clock was attached to thecenter of the sign. The clocks on all signs were synchro-nized. The back of each sign was identical to the front,with the exception that the clock area included textexplaining that a clock could be found on the otherside.

Waypoint signs were placed at locations in themeadow selected to define most common hiking itiner-aries. For example, common sequences of waypointspassed included C, I, I, D for hikes to Myrtle Falls, orC, G, K, P, R, M, I, C for the loop hike to PanoramaPoint. The sign locations were selected to define gen-eral boundaries for the lower, middle, and uppermeadow. The number of signs necessary to provideprecise information about hiking routes in the lowermeadow (locations below the F, G, H, and I signs) waslikely to impose an unreasonable burden on hikersbecause the signs would have been only yards apart.Because the trail system becomes simpler as onemoves away from the parking areas, longer hikes intothe upper meadow were more precisely defined by thewaypoint card data than were shorter hikes in whichvisitors did not leave the lower meadow.

Exit signs (fig. 38) also measuring 8.5 x 11 inchesand attached to metal stakes were also placed at theentry/exit points. Plastic boxes were attached to thebase of each staked exit sign where respondents coulddeposit cards if the survey worker was not presentwhen they finished hiking.

Results

Of the 351 parties asked to participate in the study,17 refused at the initial contact and 265 returned


Figure 36—Front and back of the waypoint card respondents used to record the waypoints theypassed while hiking.

Figure 37—Typical waypoint sign placed alongsidethe trail.

Date: _____ Card # _____ Int.: _____ Zip ________ Party Size: ____ No.<18yo: ___ Dest.: ____________________ Stop: _____________________

INSTRUCTIONS: For each way-

point sign you pass, please write

the letter of the way-point and the

time you passed it on the other

side of this card.

If you pass the same way-point

more than once, WRITE THE

LETTER AND TIME EVERY

TIME YOU PASS.

After your hike, please give this

card to the worker at the trailhead

or drop it in the “Study Card” box

at the trailhead.

Way Point Time

OMB # Expiration

WAY POINT

FF

The Current Time Is:

Please Do Not Move or Otherwise Alter

This Temporary Sign

Mt. Rainier National Park is conducting a study of the routes

people hike on these trails. A random sample of hikers are

recording the times that they pass these way-points today.

Please Enter the Way Point and Time on Your Card

Way Point Time

A 11:38

Time

AExample

Example

CLOCK


Figure 38—Exit sign placed at entry/exit pointsinstructing respondents how to return waypointcards when the survey worker was not present.

Table 18—Duration of stops at three locations in ParadiseMeadow.

Duration of stopsMean Standard

Locations minutes deviation N

Myrtle Falls 10.6 3.7 20Alta Vista 11.7 4.1 7Panorama Point 17.2 13.7 16

completed cards, yielding a total response rate of 76percent. Anecdotal reports from survey workers indi-cated that respondents found it easy to follow thesurvey instructions. An initial check of the 265 com-pleted itineraries found that 48 (18 percent) includedsequences of waypoint signs that were impossible toencounter. However, closer examination of those im-possible routes suggested that 27 occurred becauserespondents overlooked a single waypoint sign. If theoverlooked signs are replaced in those itineraries, theincidence of impossible routes falls to 8 percent. Intotal, 70 percent of all parties approached for the studyreturned waypoint cards that traced possible hikingroutes.

Many visitors found it difficult to recall the locationsand durations of the times they stopped hiking formore than 5 minutes. In addition, survey workersoften found it difficult to talk with returning partieswhile maintaining the sampling interval for enteringparties (particularly at the busiest entry point,waypoint D). Thus, a relatively small number of re-spondents described the places they stopped and theduration of their stops. Table 18 summarizes the

WAY POINT STUDY

IF YOU HAVE A WAY POINT CARD

AND YOU ARE FINISHED WITH

YOUR WALK OR HIKE, PLEASE:

1. Write this last way point and the current

time on your card.

2. Recall the locations where you stopped

hiking for more than five minutes. Write

each location and the amount of time

you spent there on your card.

For example: “Alta Vista — 18 min.”

3. Put your card in the box below.

THANK YOU

duration data for three locations at which hikerscommonly stop.

Discussion

The high response rate, low incidence of impossibleroutes, and anecdotal reports all suggest that respon-dents found it easy to accurately record and report theirhiking routes by filling out the waypoint cards. Thus,waypoint cards offer a clear advantage over retrospec-tive interviews when applied in Paradise Meadow.

The relative usefulness of the waypoint card methodin comparison to map-diaries is currently more diffi-cult to determine because this initial study was notdesigned to compare and contrast those data collectionmethods. For example, a comparison of the degree towhich participating in the waypoint card study or amap-diary study would alter hiking routes is notpossible. (The question of whether the waypoint cardmethod altered hiking itineraries at all will be testedby future work in which simulation models based onthe waypoint card itineraries are constructed andtheir validity is tested.) Other comparisons of thewaypoint card and map-diary methods of collectingitinerary show a mixture of advantages and disadvan-tages. The data provided by the waypoint cards areideally formatted for the building of computer simula-tion models, and are much easier to code and enter intoa computerized database than the data from map-diaries. However, the waypoint data do not provide thesame level of detail as a complete map-diary. In per-haps the most extreme case from the current study,figure 35 shows at least seven possible hiking routesthrough the lower meadow for a hiker whose waypointcard began with points B, G.

When the results of the waypoint study were used toconstruct a simulation model of use for the observedweekend, three shortcomings of the data were identi-fied: (1) a lack of detail describing hikes in the lowermeadow (discussed above), (2) poor information aboutthe durations and locations of hiking breaks, and (3)lack of information about hiking patterns on the trailloops at Alta Vista, Glacier Vista, and PanoramaPoint. For the construction of the preliminary model,these shortcomings could be addressed by making


assumptions about hiker behavior. However, futureapplications of the waypoint card method could beimproved by addressing each shortcoming.

The data describing hiking breaks were limited, butwere generally consistent with observations that mosthiking breaks occurred at a few locations in ParadiseMeadow (Myrtle Falls, Alta Vista, Panorama Point).The validity of the computer simulation would beimproved if the distribution of stop durations at thoselocations could be more accurately described. In futurestudies, visitors at those sites will be systematicallyobserved to collect such descriptive data.

More detailed description of hikes in the lowermeadow can most easily be obtained by addingwaypoints in that area. The initial success of themethod and verbal reports from respondents indicat-ing that participation was not a concern suggest thatsuch additions will not pose a significant burden.Thus, waypoint signs will be added at several morejunctions in the lower meadow.

The specific changes necessary to obtain more de-tailed description of hiking patterns on trail loops hasnot been determined. Waypoint signs could be added,systematic observations could be made, or some com-bination of those two techniques might be adopted.Inspection of each trail loop prior to the next waypointcard study will determine the alterations or additionsto the study design.

Currently available GPS receivers offer a means ofcollecting hiker itinerary data that is superior to thewaypoint card method or any other paper-and-pencilinstrument. Unfortunately, the receivers are not yetcheap enough that they can be distributed to a largenumber of hikers without unacceptable costs due tobreakage, loss, and pilfering. Nonetheless, small num-bers of GPS receivers might be used to collect informa-tion that could be used to validate the waypoint card

information (for example, by asking some visitors toboth fill out a waypoint card and carry a GPS receiver).Such validation studies could develop procedures use-ful in future large-scale studies of hiker itinerariesusing GPS receivers.

Conclusion

Computer simulation modeling holds great promiseas a management tool. As models are developed for avariety of recreational settings, novel problems willinevitably arise. This case study discusses how a newdata collection procedure was used to address theproblem of collecting information describing use of acomplex trail system at Paradise Meadow in MountRainier National Park.

References

Johnson, D. R., Foster, K. P.; Kerr, K. L. 1991. Mount RainierNational Park 1990 general visitor survey. Technical Report.Seattle, WA: University of Washington, College of Forest Re-sources, Pacific Northwest Cooperative Ecosystem Studies Unit.137 p.

National Park Service. 2004. National Park Service public usestatistics office Web site. ttp://www2.nature.nps.gov/stats/

Rochefort, R. M.; Swinney, D. D. 2000. Human impact surveys inMount Rainier National Park: past, present, and future. In: Cole,D. N.; McCool, S. F.; Borrie, W. T.; O’Loughlin, J., comps. Wilder-ness science in a time of change conference—Volume 5: wilder-ness ecosystems, threats, and management. Gen. Tech. Rep.RMRS-P-15-VOL-5. Ogden, UT: U.S Department of Agriculture,Forest Service, Rocky Mountain Research Station: 165–171.

Vande Kamp, M. E. 2001. Paradise parking lots: hiker vs non-hikeruse. Technical Report. Seattle, WA: University of Washington,College of Forest Resources, Pacific Northwest Cooperative Eco-system Studies Unit. 3 p.

Vande Kamp M. E.; Zweibel, B. R. 2004. A regression-based modelof visitor use in the paradise developed area of Mount RainierNational Park. Technical Report. Seattle, WA: University ofWashington, College of Forest Resources, Pacific NorthwestCooperative Ecosystem Studies Unit. 38 p.


Documents

Chapter 4: Case Studies of Simulation Models of Recreation Use · 2008. 5. 8. · Chapter 4: Case Studies of Simulation Models of Recreation Use David N. Cole used to monitor crowding-related