An agent-based approach to providing tourism ...

3 downloads 0 Views 1MB Size Report
seasonal nature of tourism, and an increasingly competitive landscape are only a few of the uncertainties that affect tourist arrivals (Baggio, 2008; Farrell and ...
Environment and Planning B: Planning and Design 2011, volume 38, pages 486 ^ 504

doi:10.1068/b35148

An agent-based approach to providing tourism planning support Peter A Johnson, Renee E Sieber

Department of Geography, McGill University, 805 Sherbrooke Street West, Montreal, Quebec H3A 2K6, Canada; e-mail: [email protected], [email protected] Received 15 December 2008; in revised form 2 July 2010

Abstract. Agent-based modeling (ABM) is a computer simulation approach that can be used to represent real-world systems and create planning scenarios to examine possible future outcomes of present-day decisions. This approach can be applied in tourism planning, where destinations are exposed to a variety of externalities, and must develop strategies to adapt to changing operational conditions. We describe the development of TourSim, an ABM of tourism dynamics set in the Canadian province of Nova Scotia. We present an overview of the data sources and techniques used to inform agent behavior and the destination landscape, as well as consider aspects of system representation and validation and how these may affect the use of TourSim. TourSim is used to generate three scenarios of tourism dynamics; a base-case scenario, one that simulates the effect of a decrease in visitation from American markets as a result of economic crisis, and the use of advertising as a response to this lower level of visitation. These scenarios are used to evaluate ABM in comparison with other computer-based methods of modeling tourism, namely geographic information systems and system dynamics models.

1 Introduction Recent global economic and political events have impacted the tourism industry throughout the world. Fuel price fluctuations, new security regulations, the highly seasonal nature of tourism, and an increasingly competitive landscape are only a few of the uncertainties that affect tourist arrivals (Baggio, 2008; Farrell and TwiningWard, 2004; Russell and Faulkner, 1999). These impacts are difficult to predict, as some destinations may benefit from shifts in visitor markets whereas neighboring ones struggle to adapt. Strategies such as advertising to raise awareness, new product development, and improved transportation linkages can be employed, each with their own positive and negative trade-offs. Key to the development of a tourism industry more resilient to these global impacts are methods that can better represent how the processes that form tourism unfold across time and space. One computer modeling method that can capture and analyze these aspects of tourism is agent-based modeling (ABM) (Bonabeau, 2002; Grimm and Railsback, 2005; Zellner, 2008). ABM represents system components as individuals, or `agents' that interact upon a spatially referenced landscape. Rather than focus on the outcomes of the tourism phenomenon, such as bed nights sold or aggregate expenditures, ABM facilitates the study of the processes that create a given system (Epstein and Axtell, 1996; Parker et al, 2003). By modeling the processes of tourism, such as how individuals and destinations interact to form patterns of impact, planners and researchers can identify points in a system where planning actions can be best applied. For tourism planning, an ABM can then be used to develop and experiment with different strategies, comparing their impact throughout a system, rather than at a fixed sample location or aggregate scale. In this study we have two main goals: the first being to describe the development of TourSim, an ABM of tourism dynamics. TourSim is used to examine the impact of a decline in visitation from the American market on destinations within the Canadian

An agent-based approach to providing tourism planning support

487

province of Nova Scotia. Nova Scotia occupies a unique geographic position in North America, located at the furthest eastern mainland point. This remoteness is a source of tourism product but also imposes a transportation premium compared with competing destinations such as Maine, New England, and Prince Edward Island, all located closer to major tourist markets. These factors expose the Nova Scotia tourism industry to fluctuations in visitation from the United States. This issue of exposure has broad implications for tourism development, making this case study a relevant example for tourism planners operating in many locations. The second goal is to place ABM within the range of existing computer approaches used to model tourism, identifying the degree to which each approach can be used to represent the process-based nature of tourism and ultimately assess the potential of ABM as it can be used to capture dynamics of interest to tourism academics and planners. 2 Computer-based methods for modeling tourism A better understanding of the processes of tourism and how changes to those processes impact the formation of travel patterns can be developed through the use of new computer-based modeling approaches. The following section compares approaches to modeling tourism that have some ability to represent processes and are applied within tourism studies. We focus on computer-based modeling approaches that can be used to facilitate the systematic testing of structural and behavioral relationships among tourism components, both in space and over time. Three approaches are investigated: geographic information systems (GIS), system dynamics (SD) models, and ABM. GIS is increasingly used to study many types of spatial phenomenon and has been applied and critiqued within many planning environments. SD models and ABM are less widely known approaches; however, early research has pointed to several advantages when used in a tourism context. Though these latter two approaches show potential, their use in tourism is still not fully explored. In the following subsections a literature review gives an overview of each of three modeling approaches, indicating the degree to which each is currently used to represent how tourism patterns and impacts are formed. 2.1 GIS applications in tourism

GIS can be defined in a narrow sense as a computerized system used to facilitate the collection, storage, retrieval, analysis, and distribution of spatial information (Chrisman, 2002). Bahaire and Elliot-White (1999) present an overview of GIS application areas within tourism; creating an inventory of resources, identifying ideal areas for development, the identification and monitoring of tourism indicators over time, integrating social and environmental datasets, and providing decision support to planning. Despite this wide range of possible application areas, there are few examples of tourism GIS applications that consider how patterns and impacts are formed. Rather, much research focuses on the visualization of tourist travel patterns. These travel patterns are based on individual-level data collected with global positioning system units, radio frequency identification tags (Bishop and Gimblett, 2000; Shoval and Isaacson, 2007), or time diaries and other survey instruments (Becken, 2005; Connell and Page, 2008; Van der Knaap, 1999). Connell and Page (2008) use GIS to create visitor itinerary maps collected with a post-trip mail survey to identify routes taken and stops made through a national park in Scotland. This study results in a visualization of the quantity and directionality of tourist flows and stopping points. GIS is used in this study as a visualization tool öthe temporal representation of tourist movement is minimal, and the resulting static map analysis shows aggregate quantities of flows over the total study time period.

488

P A Johnson, R E Sieber

These flows are compared with previous studies in an attempt to construct a typology of different tourist-route patterns, including the identification of main gateways into the national park area and stopping points that had potential to become congested during peak visitation times. This study represents a typical use of GIS in tourism research, focusing on visualization of quantities of tourists (McAdam, 1999; Stewart et al, 2008), as opposed to analysis of how these patterns are formed or changed. This type of study cannot extrapolate these patterns over an altered destination landscape or represent patterns generated by new tourism marketing trends. The continually shifting nature of tourism precludes the ability to consider a small sample of tourist travel patterns to be both accurate and durable over time and across various conditions. The popularity of some destinations grows rapidly and for others it declines due to a confluence of factors that are not often represented in a tourist trip diary. These externalities are included only through the way in which they may have influenced the travel patterns of the sample and cannot be independently identified or manipulated. This makes the use of GIS in this manner adequate as a visualization of a limited series of patterns over a certain time period, but inappropriate to examine the underlying variables that influence tourist pattern generation. 2.2 System dynamics models in tourism

SD models, also called `stock and flow' models, are a comparatively new modeling approach used within tourism studies. SD models conceptualize system components as a series of stocks (a container that holds a quantity of a variable), connected to a number of flows that lead both into and out of the stock, modifying its value. The tourism futures simulator (TFS) is one of the most developed applications of SD tourism. It is a platform used to develop tourism scenarios that facilitate communitybased planning (Walker et al, 2005). The TFS uses data collected by community members to parameterize stocks that represent tourism components, such as money spent on destination marketing, numbers of tourists at an attraction, and tourist expenditures. This modeling approach is inherently temporal, as components interact according to a system clock that controls the timing of actions (Peterson, 2000). Though recent studies exploring the use of SD models in tourism show potential in examining the function of tourism components, two issues constrain this application. First, SD models are commonly aspatial. The lack of a spatial component to the representation of tourism ignores one of the key aspects of tourism as a phenomenon. Second, SD models typically represent entities as homogeneous stocks, where individual behavior is masked or aggregated. When considering the wide range of tourist behavior and destination characteristics, removing this aspect from a model provides an incomplete representation. 2.3 ABM in tourism

ABM is an emerging modeling approach with potential within tourism research and planning (Baggio, 2008). Within an ABM, system components are represented as individuals, or agents, that interact upon a landscape. ABM can be used to model systems characterized by interacting processes and individual behaviors over time and space (Epstein and Axtell, 1996; Parker et al, 2003). These individual agents are assigned general rules that govern their behavior and interactions, using data sources such as sample surveys, laboratory experiments, GIS or remotely sensed data, participant observation, and companion modeling (Robinson et al, 2007). Though a popular approach for modeling systems such as urban development (Waddell, 2002), use of water resources (Zellner, 2007), retail markets (Heppenstall et al, 2006), and recreation behavior (Gimblett and Skov-Petersen, 2008), there has to date been little use of

An agent-based approach to providing tourism planning support

489

ABM within tourism or tourism planning, despite its benefits in representing system processes and interactions over time and space. There are several benefits from using an ABM to represent the processes and interactions of the tourism phenomenon. In creating a tourism ABM, tourist agents can be parameterized with behaviors to govern activity or accommodation preference, as well as spatial characteristics, such as travel distance. Tourist destinations can also be implemented as agents, expressing behaviors such as price setting, product development, and advertising. These interactions between destination and tourist can play out upon a georeferenced landscape, with space as an explicit variable. An ABM can also include a variety of temporal behaviors, allowing each component to operate at a different time scale. This means that a tourist may be modeled as spending two days at a certain destination, but this can comprise one leg of a longer total trip. Destinations may also make decisions on the basis of monthly or yearly feedback about tourist arrival numbers. In this way, time is a function of each object, rather than a global time present throughout the entire simulation. This use of object-dependent time allows ABM to reflect more accurately the dynamic interactions of tourism. Considering these factors, ABM has particular relevance to representing the characteristics of tourism, as it facilitates the representation of individual-level spatiotemporal interactions that have the potential to generate system-level effects. In the following sections we focus on testing the application of ABM to tourism. 3 Development of TourSim TourSim takes a traditional supply ^ demand approach to conceptualizing tourism, where tourist agents move across a landscape of destinations, attempting to satisfy their accommodation and activity preferences. The landscape used in TourSim is spatially referenced and representative of Nova Scotia, with destinations populated by types of accommodation and activity. Tourist agents possess characteristics, such as preference for accommodations, activities, length of stay, and maximum daily travel distance. These characteristics are drawn from a direct survey of tourists and the manner in which tourist agents are able to meet their preferences in negotiation with the supply available at each destination determines the pattern of their visitation within TourSim. The following subsections provide details on the construction of TourSim. 3.1 The destination landscape

TourSim represents thirty five of the most commonly visited tourist destinations in Nova Scotia as fixed georeferenced points (figure 1). These are a variety of locations, from small towns to major cities, and are selected using major groupings of tourist accommodations drawn from an annual inventory of tourism businesses collected by the Nova Scotia Department of Tourism, Culture, and Heritage (NSDTCH) (http:// www.gov.ns.ca/tch/pubs/insights/), research division. Each destination is parameterized with accommodation and activity types, on the basis of data from the provincial tourism website listing of businesses (http://novascotia.com). There are data aggregated to correspond with categories identified in the Canadian Travel Survey (CTS) and the International Travel Survey (ITS), two multiyear tourism surveys conducted by Statistics Canada (http://ww.statcan.gc.ca/) also used to parameterize tourist-agent behaviors. Table 1 shows a listing of select destinations with the accommodation and activity categories at each, presented in a binary presence or nonpresence (absence) format. Using two different scales of data created challenges in parameterizing the model. Destination data were at the individual level, representing a specific accommodation activity. Tourist data were recorded not for a specific accommodation or activity (eg the Art Gallery of Nova Scotia), but rather aggregated into the seven accommodation

490

P A Johnson, R E Sieber

Figure 1. Map of destinations and regions in Nova Scotia, Canada.

and twenty-two activity types listed in table 1. As such, destination amenities are aggregated into the same range of categories. One implication of this approach is that TourSim treats accommodation and activity types as being identical between destinations, recording only the presence or nonpresence of each. Though this limits the richness included for each destination, the alternative would require a subjective assessment of the quality or attractiveness of an amenity. In the absence of a measure of these, we kept all accommodations and activities on an equal playing field, in that a hotel at one destination is equal to one at another destination, although this is not the case in reality. In addition to accommodation and activity characteristics, each destination is assigned an `awareness' value to represent the relative position of the destination in the minds of tourists. This adds realism to the interaction between destination and tourist, simulating how tourists make decisions with imperfect information, conferring a competitive advantage to better-known destinations. The awareness function assigns a value from 0 to 100 to each destination, where those with lower values are less well known compared with those with higher values. These values are based on the 2004 Nova Scotia Visitor Exit Survey (http://www.tourismvc.com/research/nstat.html), a survey that recorded the percentage of all tourists to the province who stopped at that destination. These percentages were translated directly into awareness values.

An agent-based approach to providing tourism planning support

491

Table 1. Destination accommodation and activity categories: Y ˆ present; 0 ˆ not present. Category

Destination Baddeck

Canso

Halifax

Shelburne

Windsor Wolfville

Accommodation Hotel Motel Bed and breakfast Resort or lodge Camping With friends or relatives Cottage

0 Y Y 0 Y Y Y

0 Y Y 0 Y Y 0

Y Y Y 0 0 Y Y

Y Y Y 0 Y Y Y

0 Y Y 0 Y Y 0

Y 0 Y 0 Y Y Y

Activity Convention Visit friends or relatives Shopping Sightseeing Festival or fair Cultural event Museum or gallery Historic site Zoo Sports event Bar or nightclub Casino Theme park National park Sports or outdoor activities Golfing Fishing or hunting Boating or swimming Walking or hiking Cruise or excursion Bird or wildlife viewing Cycling

0 Y Y Y 0 0 Y Y 0 0 0 0 0 Y 0 Y Y Y Y 0 Y Y

0 Y 0 Y 0 0 Y Y 0 0 0 0 0 Y 0 0 0 Y Y 0 0 0

Y Y Y Y 0 Y Y Y 0 Y Y Y Y Y Y Y 0 Y Y Y Y Y

0 Y Y Y Y Y Y Y 0 0 0 0 0 0 0 Y 0 Y Y 0 0 0

0 Y 0 Y Y Y Y Y 0 0 0 0 0 0 0 0 0 0 Y 0 0 0

Y Y Y Y 0 Y Y Y 0 0 0 0 0 0 0 0 0 Y 0 0 0 0

For example, if 10% of tourists stopped at a real-life destination, the destination in the model was assigned the value 10. This awareness function mediates the frequency with which the tourist agent considers each destination, a process described in greater derail later in this section. 3.2 Tourist agent behaviors

TourSim contains three types of tourist agents based on their home market: domestic; American; and international. Each type of tourist agent is informed with preferences for accommodation, activity, maximum travel distance, and length of stay, according to data drawn from the CTS and ITS. Table 2 shows the accommodation and activity categories that each tourist agent can prefer and the percentage of each tourist type that chose each accommodation and activity. For example, 14% of domestic tourists stay in hotels, compared with 50% of international tourists. These values are used in the agent decision-making process as a main driver of tourist visitation. Length of stay is an important tourist-agent behavioral characteristic included in TourSim. Figure 2 shows the range of length of stay assigned to each tourist type. The distance that a tourist agent is willing to travel per day is the final tourist-agent characteristic. Figure 3 shows the distance traveled by tourists as recorded by domestic tourists. As distance-traveled data were not collected for American and international tourists,

492

P A Johnson, R E Sieber

Table 2. Tourist accommodation and activity profiles as percentages. Market

Category

domestic

American

international

Accommodation Hotel Motel Bed and breakfast Resort or lodge Camping With friends or relatives Cottage

14 6 1 1 10 58 10

33 39 8 0 14 0 7

50 19 14 0 12 0 5

Activity Convention Visit friends or relatives Shopping Sightseeing Festival or fair Cultural event Museum or gallery Historic site Zoo Sports event Bar or nightclub Casino Theme park National park Sports or outdoor activities Golfing Fishing or hunting Boating or swimming Walking or hiking Cruise or excursion Bird or wildlife viewing Cycling

4 28 16 12 2 2 2 2 1 2 3 1 1 6 9 1 1 3 3 1 0 0

0 6 15 16 2 4 11 14 3 1 4 3 1 11 5 1 1 2 0 0 0 0

0 10 15 14 3 3 9 11 3 1 6 2 2 9 6 1 1 3 0 0 0 0

50 Domestic Percentage of sample

40 International American

30

20

10

0

1

6

11

16 21 26 31 Tourist length of stay (days)

Figure 2. Tourist length of stay.

36

41

An agent-based approach to providing tourism planning support

493

100

91

Percentile of sample

81

71

61

51

41

31

21

11

0

0

490

980

1470

1960

2450

Distance per day (km)

Figure 3. Tourist distance traveled per day.

the domestic tourist-distance value was assigned to all tourist categories. This can be considered to represent the majority of travel around the province of Nova Scotia, with over 90% of trips being of less than 500 km. 3.3 Tourist-agent decision making

For a tourist agent to visit a destination, a matching is required between preferences and potential destination using data from the CTS and ITS. Each day of model time, tourist agents of each type are added to the model at a seasonally variable monthly rate drawn from data collected by the NSDTCH. Tourist agents enter Nova Scotia via one of seven airport, highway, and ferry access points. Model time proceeds in hours, with each tourist agent evaluating destinations and moving daily when a match is found. This decision making occurs asynchronously throughout a 24-hour period, with all agents currently active in the model making a movement decision within that time frame. Tourist-agent decision making is processed according to a flow chart controlled by a heuristic decision-making strategy. The sequence of rules is outlined in figure 4. Once a tourist agent is introduced, it selects one of the thirty-five destinations to evaluate at random. This ensures that tourist agents do not evaluate destinations in a consistent order, thus avoiding conferring an advantage to certain destinations. The next step is to assess the awareness level of the candidate destination. For example, the provincial capital of Halifax, being a major tourist destination, has a higher

494

P A Johnson, R E Sieber

Tourist agent enters simulation

Assigned to a port of entry

Select a random destination

not aware Check awareness

is aware

unsuccessful advertise

outside range

Evaluate activity and accommodation Successful Increase awareness Evaluate distance to travel

continues less than 30%?

within range

Visit destination

Destination occupancy

Check length of stay finished

Tourist agent leaves simulation

Figure 4. Tourist-agent decision-making flowchart.

awareness value than a tertiary destination and therefore the majority of tourist agents who randomly select Halifax will continue to evaluate it for a preference match. The remainder of the time, tourist agents will select another destination to evaluate, first for awareness. Once a tourist agent has successfully passed through the awareness check, it then evaluates the candidate destination for a match between accommodation and activity preferences and the local supply of tourist amenities. Following Berger and Schreinemachers (2006), we use a Monte Carlo method of assigning behaviors

An agent-based approach to providing tourism planning support

495

to a population of agents using a set of survey data. Using this method, tourist agents are assigned preferences by drawing a random number from 0 to 100, which corresponds to a specific accommodation and activity preference (see table 2). If a match under both categories is found, the tourist agent then evaluates the distance required to travel from its current location to the candidate destination. If this distance is less than a daily distance threshold, the tourist agent makes a successful trip. This decisionmaking process continues within a loop, with many thousands of actions occurring every day of model time. As model time progresses, a tourist agent will accumulate a predetermined number of successful matches, spread out over a number of simulated days. When this number of visits is reached the tourist agent is removed from the model, representing the completion of its trip. 3.4 TourSim operational interface

TourSim was developed with an Internet application component where scenarios can be run in a web browser. Scenarios are preset and key variables in the simulation can be altered to compare multiple model runs. The TourSim interface consists of a setup screen and a runtime screen. The setup screen (figure 5) presents a description of the simulation, a set of blank charts where results are later added, and a range of useradjustable variables depicted as slider bars. These variables include the rate of tourism growth and options for viewing results for a specific tourist destination. Once the user has set the variables, the model can be run. This launches the runtime screen and begins the advancement of model time from 1 January. The runtime screen (figure 6) show a map of Nova Scotia with tourist travel routes. These routes thicken according to the number of tourist agents who travel on that route, giving a visual indication of tourist travel patterns. Continuously updating charts track the values of total and per day visitation. If the user has selected to view results for one destination, pie charts show the accommodation and activity choices selected by tourist agents for that destination. Once TourSim ends, data are added to the blank charts on the setup screen. The user can now alter any of the variables and rerun the model. The results of this second run are added to the setup screen (figure 7), facilitating comparison of the model runs.

Figure 5. TourSim setup screen.

496

P A Johnson, R E Sieber

Figure 6. TourSim runtime screen.

Figure 7. TourSim setup screen with results. 3.5 Verification and validation

The degrees to which a model functions as intended and behaves in a similar fashion to the real-world target system are important considerations in the ABM development process. These evaluations serve as a type of `reality check' on the model and have two main components, verification and validation. Verification determines whether model elements function as expected, in that there are no errors in specification. Verification can be achieved through testing procedures including code reviews, comparative visual tests, and spot checks of code sections (Gilbert and Troitzsch, 2005; Grimm and Railsback, 2005; Parker et al, 2003). To verify TourSim we ran the model

An agent-based approach to providing tourism planning support

497

first with a minimum number of agents (10 in total). We configured TourSim to report the output at each point within the tourist decision-making process, ensuring that, for example, a tourist agent that had visited the destination of Halifax had actually been randomly assigned to that destination and possessed the correct preferences. Next, we forced tourist agents to follow certain behaviors, testing if model output would match. We decreased the tourist length of stay value to 1, representing a maximum trip length of one day. This had the effect of preventing tourist agents from visiting more than one destination before leaving the model. Using counters within the model we observed that the number of agents entering corresponded with the sum of the number of visits to each destination and the number of agents exiting. We also experimented with `turning off ' various aspects of TourSim, altering agent preference and destination supply to force all tourist agents to visit one destination, and placing a destination beyond all possible distance thresholds. In each case, TourSim demonstrated the expected outcome, confirming that the tourist-agent decision making and destination interactions function correctly. The second step in our test of TourSim was validation given the model goals and balanced by the need for simplicity in model building (Crooks et al, 2008; Rykiel, 1996). Validation can take several forms, such as statistical or predictive analysis between modeled and independent data, and historical reconstruction of known events (Rykiel, 1996). For models that are more exploratory, validation methods such as pattern matching with independent data (Bankes et al, 2002) or qualitative checking with experts can be employed (Becu et al, 2003; David et al, 2005). As discussed by Parker et al (2003), ABMs are made to address different goals, ranging from explanatory, or `proof of concept' theoretical models, to more highly detailed and descriptive for use in policy environments. TourSim falls in between these two types; it moves beyond a purely theoretical model by incorporating real-world data to parameterize agent and landscape; however, it does not address a sufficiently wide range of tourism dynamics and processes to enable predictive use. Following a number of authors (Janssen and Ostrom, 2006; Rykiel, 1996), we validate our model to a qualitative level. Our intent for TourSim is not a predictive model, but a framework to support the exploration of tourism dynamics. To validate TourSim, we compare modeled results with two independent datasets provided by the NSDTCH, one of provincial and regional tourist visitation and one of destination occupancy. At the provincial scale we compare total visitation levels generated by TourSim with 2007 actual visitation (table 3). At this aggregate scale TourSim overpredicts the total number of tourists by approximately 5%. Table 3. Comparison of modeled with actual data for regional scale validation. Region

Modeled visitation

Actual visitation in 2007

Percentage difference

Annapolis Valley Cape Breton Eastern Shore Fundy Shore Halifax/Dartmouth Northumberland Shore South Shore

228 780 639 300 30 080 97 780 1 161 905 345 425 164 215

273 000 410 000 22 000 140 000 1 307 000 190 000 199 000

ÿ16 ‡56 ÿ37 ÿ30 ÿ11 ‡82 ÿ17

Total

2 667 485

2 541 000

‡5

498

P A Johnson, R E Sieber

Breaking this down to a regional level shows differences in visitation range from an underprediction of 11% at the major destination of Halifax/Dartmouth, to a substantial overprediction of 82% for the Northumberland Shore region. Several aspects of the model structure and source-data reporting method call into question the appropriateness of this regional comparison. In TourSim we define only the thirty-five most visited locations in Nova Scotia and these are represented as discrete points. To determine regional visitation, destinations are grouped according to the regional divisions shown in the NSDTCH data. This results in a situation where destination visitation may not be counted in the correct region, but rather attributed to another destination or neighboring region. A more detailed, destination-level validation compares TourSim visitation data with a 2007 occupancy survey of all fixed-roof accommodations, grouped into destinations. This survey is limited in that it does not include campground accommodations (ie not fixed-roof ) and tourists who stayed with friends and family, negating direct comparison between the number of modeled tourist visits and the number of rooms sold. To overcome this limitation, we compare the share of tourist visits that each destination receives as a percentage of the total provincial visitation in both the modeled TourSim scenarios and the occupancy survey. Using this destination-level comparison, a similar distribution between modeled and real data is evident (table 4). One notable exception is the major tourist destination of Halifax, which TourSim underpredicts by 12.8%. This is an issue of both the scale of data collection and the coding of destinations within TourSim. Halifax is often considered to comprise a very large regional municipality that includes many smaller towns and villages, some of which are represented in TourSim, or grouped with other nearby destinations. Table 4. Destination-level validation. Destination

Modeled provincial visit share (% of total)

Actual visitation share in 2007 (% of total)

Percentage difference a

Amherst Antigonish Baddeck Canso Cheticamp Dartmouth Digby Halifax Ingonish Louisbourg Lunenburg New Glasgow North Sydney Sheet Harbour Sydney Truro Windsor Wolfville Yarmouth

1.6 7.3 5.2 0.4 2.2 7.8 2.9 34.8 3.6 1.3 2.1 2.2 2.0 0.5 6.5 4.2 0.7 0.9 2.7

2.0 3.2 2.8 0.2 1.2 10.7 1.8 47.6 1.7 0.7 1.5 2.7 2.7 0.7 6.6 3.7 0.4 0.8 2.4

ÿ0.4 ‡4.1 ‡2.4 ‡0.2 ‡1.0 ÿ2.8 ‡1.2 ÿ12.8 ‡1.9 ‡0.6 ‡0.5 ÿ0.5 ÿ0.5 ÿ0.1 ÿ0.1 ‡0.5 ‡0.3 ‡0.1 ‡0.3

a Variations

in percentage differences from reported to actual due to rounding.

An agent-based approach to providing tourism planning support

499

4 Scenario comparison results To demonstrate the use of TourSim to model tourism dynamics, we develop three scenarios to investigate the effects at the destination level of a downturn in visitation from American markets and the use of advertising as an adaptation strategy. This issue is of interest to tourism planners and businesses within Nova Scotia and also has implications for any destination facing issues such as the effects of new border passport requirements, rising gasoline prices, economic upheaval, and currency-exchange fluctuations. The first scenario uses averaged tourist arrival numbers for 2000 ^ 07 and holds growth from all three markets (American, international, and domestic) constant, creating a base case for comparison. For the second scenario, the number of American tourists entering Nova Scotia declines at a rate of 10% per year. The third scenario examines the impact of advertising as a way to compensate for this reduction in visitation. This is represented by increasing the awareness value of destinations in response to low levels of tourist occupancy. Occupancy is calculated as the number of tourists visiting the destination, compared with a total seasonally adjusted supply of accommodation. Every month that a destination experiences less than a 30% level of occupancy, it increases in awareness by a value of 1, to a maximum of 100, representing the efforts of destinations to advertise. This 30% cutoff was chosen after consulting the NSDTCH 2007 occupancy survey, which shows that the majority of regions have a greater than 30% average annual occupancy rate. Six destinations were selected for this comparison: Baddeck, a major rural tourism location and the centre of the Cape Breton Island tourism region; the remote village of Canso; the mid-sized regional tourism locations of Shelburne, Wolfville, and Windsor; and Halifax, the provincial capital. Halifax has the largest and most diverse supply of tourism accommodation and activity in the province, with Baddeck and Wolfville possessing an average range of amenity, whereas Shelburne, Windsor, and Canso contain a limited and seasonal range. These locations were chosen as representative of the main types of destinations in Nova Scotia. We chose to run each scenario for five simulated years, as this allowed visitation patterns to develop and is similar in length to a typical tourism planning time horizon. Each scenario is repeated twenty times with results averaged to account for randomness in the model structure. 4.1 Change in number of American visitors

The 10% decline in American visitation represents a reallocation of tourists within the province, in that for one destination to show gains, losses must occur elsewhere. Destinations have a level of exposure to the American market based on their available accommodation and activity options, distance to ports of entry, and spatial relationship with other destinations. These factors combine to create an emergent pattern of visitation. The effect of the decline in American tourism can be seen in the results of the base-case scenario (table 5). Note how the impact of this decline varies by destination, indicating that reliance on American visitation is a product of the range of accommodation and activity choices at a destination and how this matches (or does not match) with the preference of American tourists. If a destination is overly reliant on a specific tourist market, a small change in arrivals can create a large impact. The implications of this finding for tourism planning is that destinations concerned about changes in the American market must consider multiple factors of location, product range, and neighboring destinations when developing a response. We now adjust a microlevel characteristic of the destinationsötheir awareness valuesöto simulate the impact of advertising. By raising the awareness values of destinations with low levels of occupancy, tourists may consider these locations more frequently.

500

P A Johnson, R E Sieber

Table 5. Comparing the percentage change in destination visitation under different scenarios. Destination

Baddeck Canso Halifax Shelburne Windsor Wolfville

Percentage change in visitation 10% loss scenario

advertising scenario

ÿ2 ÿ4 ÿ1 ÿ4 ÿ2 ÿ3

ÿ5 ‡26 ÿ5 ‡59 ‡9 ‡11

Awareness increase (%)

0 30 0 50 23 0

This does not guarantee that a tourist agent will visit a particular destination, but simply increases the likelihood that it considers it as a destination. Within this advertising scenario there are several variables that have been altered, affecting the way in which patterns emerge. First, there are fewer American tourist agents entering the simulation. This lowers visitation to destinations favored by Americans, potentially triggering advertising. Second, destinations can increase their level of awareness in response to low levels of occupancy. And finally, tourist agents now have an altered landscape of awareness upon which to make their decisions. The changes made in this scenario demonstrate the experimental questions that can be framed using an ABM approachöquestions of complex interactions that take place on a shifting landscape populated by heterogeneous individuals. The results of the advertising strategy show how destinations can raise awareness as a response to a decline in American visitation. As in the base case, the impacts of changes to the system create effects that vary between destinations. Shelburne (‡59%) and Canso (‡26%), two comparably unknown destinations, advertise and see a dramatic benefit. Three locations did not need to advertise to maintain a greater than 30% level of occupancy: Baddeck (ÿ5%), Halifax (ÿ5%) and Wolfville (‡11%). Baddeck and Halifax show losses in visitation, attributable to increased competition driven by the rise in awareness of other destinations. The case of Wolfville shows surprising behavior, in that, although it did not advertise, it saw an increase in visitation. This is further complicated by the fact that the neighboring destination of Windsor advertised and saw an increase of 9% in visitation. The advertising at Windsor is potentially generating the spin-off visitation seen in nearby Wolfville. This scenario comparison shows that in terms of how TourSim models visitation, increasing awareness has a positive effect in driving more visitation to both the destination that advertised, but also to neighboring destinations. 5 Discussion The purpose of this research is twofold: first to examine the impacts of a decline in American-market visitation to the Canadian province of Nova Scotia, through the use of TourSim, an ABM of tourism dynamics, and second, to place ABM as an approach within others used in tourism. We present the results of the case study, including insights into the dynamics of destination adaptation and competition, and the implications this holds for tourism planning in Nova Scotia. We also discuss the development and use of TourSim itself, identifying how ABM can be used to represent the processes that form tourism and the limitations and benefits of this approach compared with other tourism-modeling approaches.

An agent-based approach to providing tourism planning support

501

5.1 Nova Scotia case study

The Nova Scotia case study examines the characteristic of destination awareness and its role in increasing destination visitation as a response to a downturn from the American-tourist market. By observing how the pattern of tourist visitation varies in response to changes in destination awareness, we can develop a better understanding of the dynamics of destination competition. This was tested for destinations that met the criteria of low levels of occupancy and awareness. Though each destination that raised its level of awareness showed an increase in visitation, other destinations that did not increase awareness also showed increased visitation. This was an unexpected finding, indicating that a type of spillover effect exists, based on the spatial location of destinations that advertise to those that did not. This increased awareness at the destination level brought more tourists into the region, showing the effect of interacting variables that influence the use of advertising as an adaptive strategy within a competitive tourism environment. That the spatial relationship of destinations within a geographic area can influence the competitive dynamics between them has particular relevance to tourism planning within Nova Scotia and for other tourism destinations throughout the world. For community-based planners, the spatial relationship of a given destination with other destinations may suggest development options that look to link together nearby destinations, pooling advertising resources for mutual benefit. This effect is also relevant for regional planners, who must consider issues such as regional brand identity, product development, and also the way in which tourists move through a region and between destinations. The directionality and flow of tourists of different types from destination to destination and the ways in which these can be either harnessed for modified to increase benefit are both areas of interest to regional and communitybased planners. For provincial-level tourism planning within Nova Scotia, this type of study is one that can be extended to examine the position of Nova Scotia within national and international competition. As a travel destination that imposes a transportation premium for tourists from major North American markets, travel to Nova Scotia is sensitive to the distance between destination and tourist-generating location. Further research into how Nova Scotia fits within these larger networks could yield valuable results on the influence of distance on destination visitation. 5.2 ABM as a way to model tourism

Through the development of TourSim, we have identified areas of advantage and constraint to the use of ABM for modeling tourism, compared with GIS and SD modeling approaches. TourSim has facilitated a process-oriented view of tourism, allowing the identification and representation of the individual-level spatial and temporal behaviors of tourists and how these interact upon a destination landscape. This approach has helped to identify how the spatial relationship between destinations is a factor in the use of advertising as an adaptation to shifts in tourist visitation. Our experiences developing and testing TourSim further define the ways in which ABM can be used to realize a process-based representation of tourism. This is particularly relevant when compared with other computer-based approaches, notably GIS and SD models that each fail to some degree in the representation of the temporal and spatial aspects that form the processes of tourism. Parker et al (2003), indicate that the process-orientated nature of ABM allows it to represent the underlying processes and dynamics that form a given system. Implicit in this process orientation are the interactions of components within time and space. These attributes of ABM are clear when comparing TourSim with GIS models of tourism, where TourSim represents the process as it unfolds through time as opposed to historical reconstructions of static time slices.

502

P A Johnson, R E Sieber

Though GIS models of tourism provide excellent representation of the spatial aspects of tourism, both in distribution of destinations and flow patterns of tourists, there is currently no use of GIS to represent the processes of tourism as they unfold over time. Though both process orientation and temporal characteristics are found in SD models of tourism, the aggregation of individual behaviors into simple stocks and flows avoids representation of the variety of individual tourist behaviors and the correspondingly diverse supply of tourism products. By modeling at an aggregate level, the diversity of tourist behaviors and preferences no longer has an impact on their travel patterns. Differentiation in spatial characteristics and interactions cannot be captured and modeled adequately using this approach, reducing opportunities for observing and experimenting with the fullest range of possible dynamics and scenarios (North and Macal, 2007). Compared with SD models, ABM as evidenced through this implementation of TourSim can show the individual-based processes as they change over time and through space. When comparing these three approaches it is clear that GIS and SD models do not support the same type of process-based approach as ABM. Both GIS and SD models omit a key variableötime in the case of GIS and space in the case of SD model. As tourism is a spatial process, the inclusion of these characteristics is essential in any model. Indeed the results derived in the Nova Scotia case study would not have been possible with either GIS or SD models. Comparably, ABM is an approach that can represent the spatial, temporal, and process-based nature of tourism, as evidenced through our development of TourSim. In the representation of tourism, this presents a new and potentially valuable approach to understanding the phenomenon and is a research direction with potential to generate further insight into the functioning of tourism. Despite the strengths of using ABM to represent tourism, there are several constraints in the development and use of ABM. This research emphasizes the importance of validation as a check on the appropriateness of ABM as a representation of a system. In particular, the degree of model validation is related to the modeling goals, with a planning support model requiring a more stringent degree of validation than a proof of concept model. We used two independent datasets to validate TourSim. The different data-collection methodologies and tourist-destination-area definitions used in the validation data made direct comparisons with modeled data difficult. As a result, there are some inconsistencies between the two, particularly when comparing modeled data with destination-level validation data. The sourcing of data for validating an ABM is a significant issue. For ABMs requiring high levels of validation it may be more appropriate to gather primary validation data rather than rely on secondary datasets. 6 Conclusions Tourism is a type of economic and social phenomenon that is uniquely exposed to forces and externalities. Recent global economic and geopolitical issues have served to impact the viability of tourism in many parts of the globe. For tourism researchers and planners looking to better understand the nature and generation of tourism patterns and impacts the investigation of ABM as an approach can uncover advantages in the representation of spatial and temporal characteristics of tourism. ABM provides significant benefits compared with traditional tourism models and approaches, namely GIS and SD models. Future areas of refinement of TourSim must focus on the sourcing of improved datasets for parameterizing tourist and destination data. Recent work on the tracking of individual tourists (Shoval, 2007; Shoval and Isaacson, 2007) can be used as a source of data for agent parameters, though the scaling up of these data to a provincial

An agent-based approach to providing tourism planning support

503

level remains problematic. Traditional methods of data collection such as trip diaries provide a valuable source of microlevel data on tourist behavior, but the time-consuming nature of collecting even a limited dataset makes this an inappropriate source of data for parameterizing a model of a broad range of tourist behavior. It is possible that, in the future, user-generated content provided via travel-review websites and drawn from tourist-supplied `digital traces' can be used to inform an ABM, though this remains an underdeveloped research area (Girardin et al, 2008; Johnson et al, 2011). This research also has implications for the use of computer models within tourism research and planning. Of significant importance in the future development of TourSim is the evaluation of its use from the perspective of tourism planners themselves. This evaluation must focus on issues of suitability to planning task and weigh benefits and constraints of adoption. In the development of planning-related technologies, these considerations are essential to ensuring that such tools are appropriate to the niche that they intend to serve (Vonk et al, 2005; Zellner, 2008). Currently the penetration of ABM within tourism planning is limited, though with increased development of casestudy examples the profile of this technology is improving. ABM, though emerging, should be considered to have both academic and applied value as a way to represent how tourism processes, operating under various scenarios, generate patterns of impacts in ways that are not adequately addressed by traditional methods. References Baggio R, 2008, ``Symptoms of complexity in a tourism system'' Tourism Analysis 13 1 ^ 20 Bahaire T, Elliot-White M, 1999, ``The application of geographic information systems (GIS) in sustainable tourism planning: a review'' Journal of Sustainable Tourism 7 159 ^ 177 Bankes S, Lempert R, Popper S, 2002, ``Making computational social science effective: epistemology, methodology, and technology'' Social Science Computer Review 20 377 ^ 388 Becken S, 2005, ``Towards sustainable tourism transport: an analysis of coach tourism in New Zealand'' Tourism Geographics 7 23 ^ 42 Becu N, Perez P, Walker A, Barreteau O, Page C L, 2003, ``Agent based simulation of a small catchment water management in northern Thailand: description of the CATCHSCAPE model'' Ecological Modelling 170 319 ^ 331 Berger T, Schreinemachers P, 2006, ``Creating agents and landscapes for multiagent systems from random samples'' Ecology and Society 11(2) 19 Bishop I, Gimblett H R, 2000, ``Management of recreational areas: GIS, autonomous agents, and virtual reality'' Environment and Planning B: Planning and Design 27 423 ^ 435 Bonabeau E, 2002,``Agent-based modeling: methods and techniques for simulating human systems'' Proceedings of the National Academy of Sciences 99 7280 ^ 7287 Chrisman N, 2002 Exploring Geographic Information Systems (John Wiley, New York) Connell J, Page S J, 2008,``Exploring the spatial patterns of car-based tourist travel in Loch Lomond and Trossachs National Park, Scotland'' Tourism Management 29 561 ^ 580 Crooks A, Castle C, Batty M, 2008, ``Key challenges in agent-based modelling for geo-spatial simulation'' Computers, Environment and Urban Systems 32 417 ^ 430 David N, Sichman J S, Coelho H, 2005, ``The logic of the method of agent-based simulation in the social sciences: empirical and intentional adequacy of computer programs'' Journal of Artificial Societies and Social Simulation 8 4 Epstein J M, Axtell R, 1996 Growing Artificial Societies: Social Science from the Bottom up (The Brookings Institution, Washington, DC) Farrell B H, Twining-Ward L, 2004, ``Reconceptualizing tourism'' Annals of Tourism Research 31 274 ^ 295 Gilbert N, Troitzsch K G, 2005 Simulation for the Social Scientist (Open University Press, Maidenhead, Berks) Gimblett H R, Skov-Petersen H, 2008 Monitoring, Simulation and Management of Visitor Landscapes (University of Arizona Press, Tucson, AZ) Girardin F, Dal Fiore F, Ratti C, Blat J, 2008, ``Leveraging explicitly disclosed location information to understand tourist dynamics: a case study'' Journal of Location Based Services 2 41 ^ 56 Grimm V, Railsback S F, 2005 Individual-based Modeling and Ecology (Princeton University Press, Princeton, NJ)

504

P A Johnson, R E Sieber

Heppenstall A J, Evans A J, Birkin M H, 2006, ``Using hybrid agent-based systems to model spatially-influenced retail markets'' Journal of Artificial Societies and Social Simulation 9 2 Janssen M A, Ostrom E, 2006, ``Empirically based, agent-based models'' Ecology and Society 11 37 Johnson P, Sieber R E, Magnien N, Ariwi J, 2011, ``Automated web harvesting to collect and analyse user-generated content for tourism'' Current Issues in Tourism doi:10.1080/13683500.2011.555528 McAdam D, 1999, ``The value and scope of geographical information systems in tourism management'' Journal of Sustainable Tourism 7 77 ^ 92 North M J, Macal C M, 2007 Managing Business Complexity: Discovering Strategic Solutions with Agent-based Modeling and Simulation (Oxford University Press, New York) Parker D C, Manson S M, Janssen M A, Hoffmann M J, Deadman P, 2003, ``Multi-agent systems for the simulation of land-use and land-cover change: a review'' Annals of the Association of American Geographers 93 314 ^ 337 Peterson G D, 2000, ``Political ecology and ecological resilience: an integration of human and ecological dynamics'' Ecological Economics 35 323 ^ 336 Robinson D T, Brown D G, Parker D C, Schreinemachers P, Janssen M A, Huigen M, Wittmer H, Gotts N, Promburom P, Irwin E, Berger T, Gatzweiler F, Barnaud C, 2007, ``Comparison of empirical methods for building agent-based models in land use science'' Journal of Land Use Science 2 31 ^ 55 Russell R, Faulkner B, 1999, ``Movers and shakers: chaos makers in tourism development'' Tourism Management 20 411 ^ 423 Rykiel E J, 1996, ``Testing ecological models: the meaning of validation'' Ecological Modelling 90 229 ^ 244 Shoval N, 2007, ``Sensing human society'' Environment and Planning B: Planning and Design 34 191 ^ 195 Shoval N, Isaacson M, 2007, ``Tracking tourists in the digital age'' Annals of Tourism Research 34 141 ^ 159 Stewart E, Jacobson D, Draper D, 2008, ``Public participation geographic information systems (PPGIS): challenges of implementation in Churchill, Manitoba'' The Canadian Geographer 52 351 ^ 366 Van der Knaap W G M, 1999, ``Research report: GIS-oriented analysis of tourist time ^ space patterns to support sustainable tourism development'' Tourism Geographies 1 56 ^ 69 Vonk G, Geertman S, Schot P, 2005, ``Bottlenecks blocking widespread usage of planning support systems'' Environment and Planning A 37 909 ^ 924 Waddell P, 2002, ``UrbanSim: modeling urban development for land use, transportation and environmental planning'' Journal of the American Planning Association 68 297 ^ 314 Walker P A, Lee D, Goddard R, Kelly G, Pedersen J, 2005 Regional Tourism Modelling: The South West Tapestry (Griffith University Sustainable Tourism Cooperative Research Centre, Southport, QLD) Zellner M L, 2007, ``Generating policies for sustainable water use in complex scenarios: an integrated land-use and water-use model of Monroe County, Michigan'' Environment and Planning B: Planning and Design 34 664 ^ 686 Zellner M L, 2008, ``Embracing complexity and uncertainty: the potential of agent-based modeling for environmental planning and policy'' Planning Theory and Practice 9 437 ^ 457

ß 2011 Pion Ltd and its Licensors

Conditions of use. This article may be downloaded from the E&P website for personal research by members of subscribing organisations. This PDF may not be placed on any website (or other online distribution system) without permission of the publisher.