Scholarly article on topic 'Optimising recreation services from protected areas – Understanding the role of natural values, built infrastructure and contextual factors'

Optimising recreation services from protected areas – Understanding the role of natural values, built infrastructure and contextual factors Academic research paper on "Social and economic geography"

CC BY
0
0
Share paper
Academic journal
Ecosystem Services
OECD Field of science
Keywords
{}

Abstract of research paper on Social and economic geography, author of scientific article — E.C. Heagney, J.M. Rose, A. Ardeshiri, M. Kovač

Abstract Effective management of recreation within protected areas requires a comprehensive understanding of the drivers of site visitation. To date, large multi-site studies that compare recreation demand for protected areas in response to underlying site attributes are rare, and have generally been restricted to high-profile, high-visitation sites. Our study, undertaken in south-eastern Australia, is the first to use random utility travel cost methods to explore recreational preferences across all sites within a large protected area network. We applied a novel zero-inflation statistical correction to identify the value of recreation demand arising in response to a broad range of site attributes, including protected area size, remoteness, natural values and built infrastructure. We find a strong influence of built infrastructure on recreation demand, but only a subset of the 9 infrastructure types modelled consistently generated recreation demand across the protected areas network. Other infrastructure contributed positively or negatively to tourism demand depending on contextual factors like site remoteness and the availability of recreation substitutes. We discuss the implications for protected area management at both the site- and network- scales, and as well as implications for designing more effective travel cost studies that allow the robust transfer of study findings to other protected area sites.

Academic research paper on topic "Optimising recreation services from protected areas – Understanding the role of natural values, built infrastructure and contextual factors"

Ecosystem Services xxx (2017) xxx-xxx

ELSEVIER

Contents lists available at ScienceDirect

Ecosystem Services

journal homepage: www.elsevier.com/locate/ecoser

Optimising recreation services from protected areas - Understanding the role of natural values, built infrastructure and contextual factors

E.C. Heagney a,b* J.M. Rosec, A. Ardeshirid, M. Kovaca

a NSW Office of Environment and Heritage, PO Box A290, Sydney South, NSW 1232, Australia

b School of Environment, Science and Engineering, Southern Cross University, Military Rd, Lismore, NSW 2480, Australia c UTS Business School, University of Technology, 15 Broadway, Ultimo, NSW 2007, Australia d Institute for Choice, University of South Australia, 140 Arthur St, North Sydney, NSW 2060, Australia

ARTICLE INFO

ABSTRACT

Article history:

Received 31 May 2017

Received in revised form 29 September

Accepted 11 October 2017 Available online xxxx

Effective management of recreation within protected areas requires a comprehensive understanding of the drivers of site visitation. To date, large multi-site studies that compare recreation demand for protected areas in response to underlying site attributes are rare, and have generally been restricted to high-profile, high-visitation sites. Our study, undertaken in south-eastern Australia, is the first to use random utility travel cost methods to explore recreational preferences across all sites within a large protected area network. We applied a novel zero-inflation statistical correction to identify the value of recreation demand arising in response to a broad range of site attributes, including protected area size, remoteness, natural values and built infrastructure. We find a strong influence of built infrastructure on recreation demand, but only a subset of the 9 infrastructure types modelled consistently generated recreation demand across the protected areas network. Other infrastructure contributed positively or negatively to tourism demand depending on contextual factors like site remoteness and the availability of recreation substitutes. We discuss the implications for protected area management at both the site-and network- scales, and as well as implications for designing more effective travel cost studies that allow the robust transfer of study findings to other protected area sites.

© 2017 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://

creativecommons.org/licenses/by/4.0/).

1. Introduction

Recreation and tourism in protected areas make an important economic contribution to individuals, and to the community more broadly. For individuals, the benefits from accessing protected area sites can include health and wellbeing, as well as cognitive and cultural benefits (Thompson et al., 2012; Kabisch et al., 2015). In economic terms, the consumer surplus that accrues to recreational users of protected areas can be large. Neher et al. (2013) undertook a meta-analysis of travel cost studies from 58 protected areas in the U.S.A to estimate the recreational value of the US protected area network at ~$31 billion per annum (in $USD 2016). Our own travel cost study estimated the recreational value of the protected area network in south-eastern Australia at $3.1 billion per annum (equivalent to $USD 2.85 billion), and reported per-hectare values that exceeded those of other land-uses like agriculture and forestry (Heagney et al., 2016). Economic expenditure from protected area visitors can also generate economic

* Corresponding author at: NSW Office of Environment and Heritage, PO Box A290, Sydney South, NSW 1232, Australia.

E-mail address: Elizabeth.heagney@environment.nsw.gov.au (E.C. Heagney).

benefits across the broader community (Driml and Common, 1995; Fortin and Gagnon, 1999; Orr, 2011; Selby et al., 2011) -including in some small regional or remote communities that may have otherwise limited opportunity for economic growth and development (Sims, 2010; Ferraro et al., 2011; Heagney et al., 2015). A growing body of literature identifies the provision of economic benefits as a powerful means of engendering local community support for protected areas and other biodiversity conservation initiatives (Armsworth et al., 2007). As such, the ability to demonstrate and quantify the economic benefits associated with recreation in protected areas is likely to become an increasingly important mechanism for securing local and political support for ongoing conservation of protected area sites into the future.

Despite the potential for protected area visitation to generate economic benefits, the mere provision of a protected area may not be sufficient to automatically generate recreation demand, nor the associated economic benefits it provides. Stevens et al. (2014) report declining rates of visitation to the 58 major nature-based parks managed by the U.S. National Parks Service, including well known parks like Yellowstone, Yosemite and the Grand Canyon, in the 13 years to 2010. They attribute the decline to increasing cost of travel relative to income. Balmford et al. (2009) report

https://doi.org/10.1016/j.ecoser.2017.10.007 2212-0416/® 2017 The Authors. Published by Elsevier B.V.

This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).

E.C. Heagney et al. /Ecosystem Services xxx (2017) xxx-xxx

falling per capita protected area visitation rates in 6 of 20 countries studied over the period 1992-2006. They also linked falling visitation rates to wealth, with declining visitation mainly observed in high income OECD countries, including Australia. It follows that realising, and optimising, recreation benefits may require that protected areas are designed to align with visitor preferences. Moreover it may require that protected areas are re-designed to re-align with changing visitor preferences and budgetary constraints as countries' economies change and develop through time.

Travel cost modelling has been used to estimate the economic value of recreation in protected areas since the late 1940s. It was first proposed by Hotelling in a letter to the Director of the U.S. National Parks Service in 1947 (Arrow and Lehmann, 2005). It has since been used to estimate the recreational value of a large number of protected area sites around the globe, and to justify the ongoing management and protection of those sites (e.g. Beal, 1995; Gurluk and Rehber, 2008; Saraj et al., 2009). However, studies of individual protected area sites are of limited use for protected area managers (Pendleton, 1999) who seek to balance conservation and recreation across protected area networks that can typically encompass tens or even hundreds of individual sites. Single site studies cannot provide insights into which site attributes are the key drivers of recreation demand, much less identify complexities arising from interactions between those site attributes and contextual factors like site remoteness and population distributions (e.g. Lansdell and Gangadharan, 2003). This means they cannot be used to assess the likely impact of a change in any site attribute on overall site quality or associated recreational value (Pendleton, 1999). It also means the findings from one site-specific study cannot easily be applied to other protected area sites. The economic literature consistently supports the notion that benefit transfer is best undertaken through the transfer of a utility function, rather than an overall of per-hectare dollar figure, and even then, it recommends that careful consideration should be given to contextual factors like the socio-demographics characteristics of the surrounding population (Loomis, 1992; Brouwer, 2000; Plummer, 2009).

In order to address this information short-fall, researchers often use random utility travel cost modelling to investigate patterns of recreational use within a single protected area, or across multiple protected area sites. Random utility travel cost studies consider site-level visitation to be an aggregate outcome arising in response to a range of underlying site attributes; they seek to disaggregate total site value to determine the contribution made by each attribute (Brown and Mendelsohn, 1984) and characterise consumers' recreational preferences (Beal, 1995; Von Haefen and Phaneuf, 2003). A small number of multi-site travel cost studies have been undertaken to assess the relative contribution of various site attributes to recreation demand in protected areas. For example, Amoako-Tuffour and Martinez-Espineira (2012) assessed the influence of camping and other accommodation options, as well as the potential to hike or visit fjords on visitation rates to Gros Morne National Park in Canada; Font (2000) assessed the influence of a range of attributes, including accessibility, tree cover, area of recreational infrastructure, and the presence of cafes and other facilities, on tourism demand for protected areas in Mallorca, Spain. Looking to the broader travel cost literature, a number of studies have estimated the recreational value of forest sites and the relative contribution made by various site attributes, including length of walking trail, elevation, quality of view, the presence or absence of specific ecosystem types or water features, to recreational site demand (Englin and Shonkwiler, 1995; Englin et al., 2006).

Although they represent a considerable improvement on single site studies, we believe most of the multi-site random utility travel cost studies of protected areas undertaken to date still encounter a number of issues that limit their ability to identify and generalise

visitor preferences. First, such studies typically assess visitation at only a small number of sites. This means they are only able to discriminate the influence of a small number of site attributes (a maximum of n - 1 if n is the number of sites included in the study). In contrast, the literature identifies a very broad range of attributes that might influence visitation to protected areas, including both site-specific attributes (as identified in the preceding paragraph) and additional contextual factors like remoteness, the size of the surrounding local population (Balmford et al., 2015), and the availability of substitute recreation sites within the surrounding region (Hanink and White, 1999; Henrickson and Johnson, 2013; Cho et al., 2014). A study that includes only a small number of sites must necessarily omit some of these key attributes, and in doing so it carries a high risk that the associated model will be mis-specified.

Second, most travel cost studies use on-site data collection processes. Traditionally this has meant that park visitors are surveyed at selected protected area sites, but more recently 'on-site' survey techniques have also included the interrogation of visitor logs or licensing databases (Neher et al., 2013; Stevens et al., 2014). From our review of the literature we conclude that on-site surveys are typically undertaken at high-profile, high-visitation parks (e.g. Font, 2000; Amoako-Tuffour and Martinez-Espineira, 2012; Neher et al., 2013; Stevens et al., 2014). Undertaking travel cost studies in high profile parks can be a valid and sensible research decision for a range of reasons. Surveying at high visitation parks enables researchers to encounter visitors in sufficient numbers for statistical analysis within a reasonable time and/or budget constraint, or there may be more complete visitor logs and other datasets available for these sites. Moreover, quantifying the value of recreation at high-profile, high-visitation parks is important for ensuring that the recreational values of those sites are appreciated and accounted for during site management. But an ongoing systematic bias towards travel cost studies that are undertaken almost exclusively at high-profile high-visitation parks is problematic. High profile parks are likely to share a range of characteristics that distinguish them from lower visitation parks, confounding attempts to discern which park attributes are responsible for generating visitor demand. This limits the potential for informed management decisions that optimise recreational services from protected areas at both site- and network-scales.

Our study is the first to use random utility travel cost methods to explore and characterise recreational preferences across a large protected area network. Our study aims to identify the role of various site attributes, including natural values, 9 types of built infrastructure, and contextual factors, in generating recreational demand. We propose and apply a new statistical correction (a random zero-inflation technique) to ensure that information about the attributes of very low (zero) visitation protected area sites are adequately incorporated into travel cost modelling. We use the results of our analysis to suggest how strategic protected area management might help to optimise recreational opportunities at both site- and network- scales.

2. Methods

2.1. Study site

This study analyses travel patterns across a network of 728 protected areas in the state of New South Wales (NSW) in southeastern Australia. The NSW protected area network spans the length and breadth of the state's 800km2 land area - from densely populated urban areas to very remote locations (Fig. 1). The protected areas of NSW are very varied with respect to their size, conservation status (each has been assigned an 1UCN conservation category) and natural features. Each protected area site also

Fig. 1. Map of the NSW protected area network. Shaded areas show the 728 individual sites that are currently protected within the NSW National Parks Estate.

contains a unique suite of tourism and recreational infrastructure, including walking tracks, lookouts and features, camping and day-use areas, accommodation options, visitor centres, carparks and more. The large number of protected area sites within NSW, combined with a high degree of variability in both population density and protected area characteristics across the state, make NSW an ideal setting to explore both site- and contextual- drivers of recreation demand.

All NSW protected areas, including the most strictly protected 'wilderness' areas allow public access for recreation purposes. Most general recreation activities, like bushwalking, swimming, kayaking, fishing and camping are unrestricted; selected activities like mountain biking and horse riding are allowed only in designated areas. In our study we define 'recreational use' to encompass any site visit, irrespective of whether park visitors participated in specific recreational activities like bush-walking or water sports.

2.2. Data collection and travel cost modelling

This study draws on data from a random stratified phone survey of more than 60,000 respondents across three states in southeastern Australia (NSW, Queensland (QLD) and Victoria). The survey was conducted every second year from 2008 to 2014. Surveys were conducted in 'waves' throughout each relevant year, with ~1,250 households surveyed every 4 weeks. Survey respondents were asked about their rates of visitation to any and all of the 728 protected areas across NSW in the preceding 4-week period. This strategy ensured that survey results encompassed visitation rates during both high- and low- visitor seasons, and were representative of year-round visitation. The survey mentioned the actual date four weeks prior to the date of interview in the questionnaire to minimise the effects of telescoping - i.e. the tendency for respondents to report visits that were made prior to the period of interest. The survey was also stratified by postcode to ensure good spatial representation across the relevant Australian states.

Travel costs were estimated based on shortest road-based driving distance using a conservative time + driving + accommodation cost formula. We estimated respondents' driving time using a

Google Maps API and calculated associated opportunity cost of time at the minimum NSW wage rate (which is approximately 1/3 the average wage rate).This approach is consistent with a range of recent travel cost studies, as reported in Amoako-Tuffour and Martinez-Espineira (2012). Driving costs were estimated at 58c per km, mid-way between costs that would be incurred using a fuel-only model (e.g. Beal, 1995) and a full-vehicle cost model (e.g. Fleming and Cook, 2008; Cho et al., 2014). For any trip that exceeded 4hrs driving duration (round trip) we assumed an overnight stay with average length of stay and cost of accommodation figures sourced from the Australian Bureau of Statistics (2015) and Tourism Research Australia (2014) respectively. We attributed 45% of the costs associated with each multi-day trip to visiting the protected area, based on responses from a sub-sample of 800 respondents undertaken during ongoing survey work by Morgan in 2016. We analysed this dataset using a logit ordered-choice random utility travel cost model using NLOGIT econometric software.

2.3. Accounting for site attributes and contextual factors

We have modelled protected area visitation as a function of a broad range of site attributes, including natural values, built infrastructure and contextual factors. Site attributes were selected from a review of the travel cost literature and in consultation with staff from the NSW National Parks and Wildlife Service. The final set of site attributes included in econometric modelling is provided in Table 1. We included diminishing returns from increasing quantities of infrastructure within a specified park or region by applying a quadratic function, as described in Heagney et al., 2016).

In analysing the role of contextual factors, we sought to investigate the role of protected area remoteness in determining visitors' preferences for specific site attributes, as well as the influence of recreational substitutes within the surrounding region. These effects were captured using two sets of model interaction terms, as summarised in Fig. 2.

Our first set of interaction terms was designed to capture attribute-specific interactions with remoteness. For example, selected site attributes may be more or less attractive depending

Table 1a

Mean median and maximum size, conservation status and remoteness of protected areas sites within the NSW protected area network.

Park attribute Measured as Data source Mean Median Max

Size ha NSW NPWS 9501 801 690,000

Remoteness ARIA++score* University of Adelaide (2015) 2.9 2.5 14.1

Conservation status IUCN conservation category (1-5)" NSW NPWS 1.6 1 5

National Park Values Presence/absence of 1) Aboriginal Heritage, 2) Historic Heritage, 3) Caves & Landforms, 4) Marine & Estuarine Values, 5) River and

Wetlands and 6) Natural values (mostly those associated with biodiversity - like threatened species, habitats or communities). Data on the presence/absence of natural values was provided by NSW NPWS from their 'State of the Parks' database.

* Higher score indicates a higher level of remoteness. See University of Adelaide (2015) for details. ** Higher score indicates lower conservation status or priority.

Table 1b

Mean, median and maximum infrastructure quantities currently present in individual sites within the NSW protected area network All infrastructure data was provided by the NSW National Parks and Wildlife Service from their asset management database.

In-park infrastructure Description Units Mean Median Max

Paths & walking tracks Length of paths and walking tracks (sealed or unsealed) km 3 0 191

Lookouts & other features Number of look-outs, elevated walkways, bridges No. 2 0 224

Roads Length of road within park (2WD or 4WD accessible) km 43 7 1967

Built retail outlets Building with designated purpose 'retail' No. 0 0 59

Built accommodation Building with designated purpose 'accommodation' No. 1 0 203

Day-use areas Formally designated day-use areas with signage no 1 0 51

Camping areas Formally designated camping areas with signage no 1 0 50

Parking areas Formally designated parking areas with signage 100s sq. m 8 0 509

Amenity blocks Number of amenity blocks within park No. 2 0 159

on the availability of other recreational alternatives in metropolitan versus rural or remote settings, and visitors may have different preferences for recreational experiences in metropolitan versus rural or remote locations.

The first set of interaction terms took the form attribute* remoteness for the nine types of built infrastructure and four of the natural value attributes included in our model. We excluded interaction terms for marine or estuarine and natural values due to spatial autocorrelation with our remoteness term (i.e. coastal areas in NSW all share relatively high population densities and are classified as having consistently low remoteness scores; they also score disproportionately on the 'natural values' scale used). Remoteness has been estimated using the Accessibility/Remoteness Index for Australia (ARIA++), which is a spatially lagged indicator that accounts for local population size and distance to larger population centres (University of Adelaide, 2015). Remoteness (using the ARIA++ index) can be considered a proxy for the

extent of complementary tourism infrastructure like roads, and retail and accommodation services. For the purposes of this study we consider metropolitan parks to be those located near major metropolitan centres or the urban fringe (ARIA++ = 0-0.7), and remote parks to be those located in more remote parts of the landscape (ARIA++ = 5.1-15).

In this context the included interaction terms can account for potential substitution (or complementarity) with other activities available in the surrounding local area. A similar, spatially lagged variable has been used to encapsulate substitution effects in Hanink and White (1999). Note that in our study these attribute-specific interaction terms were modelled in addition to the standalone ARIA++ term included in our model, which captured any effects of the remoteness of the park itself.

The second set of interaction terms sought to capture the influence of recreational infrastructure situated in alternate protected areas in close proximity to the protected area of interest. It

Fig. 2. Schematic diagram showing modelled relationships between natural site values, site remoteness and infrastructure. Note: Black arrows show the primary influence of natural values, remoteness, infrastructure and visitor demographics on park visitation; grey arrows show the potential for secondary effects of site remoteness on visitation via interaction with visitor preferences for specific natural values or built infrastructure types; white block arrows indicate potential for substitution (or complementarity) effects between sites of for specific types of built infrastructure.

E.C. Heagney et al. /Ecosystem Services xxx (2017) xxx-xxx

included interaction terms taking the form infrastructure (park) / infrastructure (NPWS area) for the eight of the nine types of inpark recreational infrastructure presented in Table 1. (amenities were excluded). 'NPWS area' is one of 50 management areas delineated by the National Parks and Wildlife Service across NSW. Each area contains an average of approximately 14 protected area sites that, due to their proximity to one another, are likely to share key ecosystem characteristics. As such they represent potential substitute sites within a reasonable travel distance from one another. A similar approach has been used by Hanink and White (1999), who included a spatial interaction term 'S' taking the form 1 to 0 depending on the presence or absence of alternative national parks within the surrounding local area. That study found a statistically significant negative effect of the interaction term indicating substitution between parks within the same region. However, these interaction terms may also return positive coefficients indicating complementarity between parks, as observed in Cho et al. (2014) in their analysis of travel patterns to five forest sites in the USA.

2.4. Avoiding bias from high-profile, high-visitation parks

By asking survey respondents about their recent visitation to any and all of the 728 sites within the NSW protected area network, our methodology has attempted to avoid bias associated with sampling only at high-profile, high-visitation parks. But a non-trivial proportion of parks within the network (274 of the 728 sites, approximately 38% of protected areas in the network) recorded zero visitation during our large scale survey, meaning that none of our 60,000 respondents had visited those parks in the four weeks prior to being surveyed. So in some ways this represents another case of restricted sampling at only high-profile, high visitation-sites, albeit a much less extreme case relative to other multi-site studies, where we could be considered to have sampled all but the lowest visitation sites.

In order to further minimise the problems associated with selective site sampling we have employed a statistical zero inflation approach that, to our knowledge, has not previously been applied to protected area travel cost studies. Note that a number of 'zero-inflation' processes exist within the travel cost literature. Zero inflation is most commonly applied when researchers use 'hurdle models' to discern whether visitors versus non-visitors differ with respect to underlying demographic or socio-economic characteristics - information that is subsequently used to inform population-weighted estimates of visitation and/or value (Font, 2000). A small number of travel cost studies have also employed zero inflation in 'double hurdle' models, where data about non-visitors is retained in second-stage modelling of the quantity of trips demanded (e.g. Flynn and Francis, 2009; Anderson, 2010, Timah, 2011), or to impute travel times and distances associated with non-chosen alternatives (Washington et al., 2014). Empirical comparisons suggest that this type of zero-inflation can provide more robust estimates of individual coefficients than models based only on positive value count data (Flynn and Francis, 2009).

Our approach applies zero inflation somewhat differently, on a site basis. By randomly assigning non-visitors to a very low (zero) visitation site, and recording a 'zero' count for the respondent's (hypothetical) trip we are able to capture data relating to attribute levels at very low visitation sites and incorporate them into our econometric model. This approach seeks to discern whether high- versus low- (zero) visitation sites differ with respect to underlying attributes or contextual factors - information that is subsequently used to calculate their relative contribution to recreation demand (i.e. model coefficients).

Zero inflation, applied in this context, ensures that information about the attributes of non-selected protected area sites is adequately incorporated into econometric modelling, and allows

better discrimination of the key attributes driving differences in recreation demand at high- versus low visitation sites. Similar random zero inflation techniques are common in the ecological literature (Warton, 2005).

2.5. Identifying drivers of recreation demand

Our study sought to compare and contrast the relative contribution of key protected area attributes to overall levels of recreation demand, as well as the implications arising from contextual factors. We used an ordered choice logit model with a repeated measures format to account for respondents who reported visiting multiple protected area sites in the preceding 4 week period. Accordingly, our random utility travel cost format took the simplified form:

P (No. of visits) =ftravel cost + site attributes (park size, conservation status, remoteness, natural values, built infrastructure) + interaction effects (site attributes * remoteness, infrastructure (park) * infrastructure (NPWS area))

+ respondent demographics + error) (1)

where P (No. of visits) is the probability that a respondent will make 0,1,2,3,4 or 5 visits to a specific protected area site over a given 4 week period. Full details of our econometric modelling and rationale for model selection are provided in Heagney et al. (2016). We used model coefficients to investigate the degree to which different site attributes influence overall rates of protected area site visitation across NSW. We applied model coefficients obtained from our econometric model to the attribute levels observed in each of the 728 protected area sites examined in our study in order to determine the average contribution of each attribute to overall utility and demand at the site-level.

2.6. Model verification

We undertook model verification by comparing modelled visitation estimates with independent data contained in the NSW Office of Environment and Heritage 'State of the Parks' database. The database includes visitation estimates for all national parks and reserve across NSW, but most are subjective reports of visitation estimated by different individuals (on-ground staff) at different locations. In order to avoid the subjective bias that this process would inevitably entail, we have restricted our comparative analysis to a smaller number of sites (n=31) for which visitor estimates were obtained using traffic counters.

3. Results

3.1. Summary statistics and model diagnostics

Independent third party analysis of the household survey we have used in our travel cost modelling estimated total visitation to the NSW National Parks Estate at 39.2 million visitors for 2014 (Morgan, 2015). Of the households surveyed, approximately 11% had visited one or more NSW protected areas in the preceding 4 week period. Visits were reported at 443 out of the 728 protected area sites that were included in the survey, but were spatially concentrated, with 75% of visitation reported at a smaller subset of approximately 50 sites (Fig. 3). The high number of sites that were not visited by our selected households (n = 274, representing ~38% of sites) highlights the importance of applying zero-inflation techniques to 1) extract information about the attributes of non-selected sites and incorporate them into our understanding

E.C. Heagney et al. /Ecosystem Services xxx (2017) xxx-xxx

Number of protected

Fig. 3. Cumulative visitation to 443 protected area sites for which visits were recorded during household surveys.

of visitor preferences and 2) estimate visitation rates at lower visitation sites (those not visited by the survey population) using modelled preference data.

Our final model had estimation based on n = 63,578 and K = 58. Model diagnostics indicated a good model fit, with McFadden pseudo R2 = 0.54, which is equivalent to ordinary least squares R2 of ~0.85-0.9 (McFadden, 1973). Comparative analysis of modelled visitation estimates and independent traffic counter data relating to 31 national parks and reserves showed good correlation between modelled and measured estimates of visitor numbers (R2 = 0.74-0.9; Fig. 4).

3.2. Identifying key drivers of recreation demand

Our use of a very large dataset based on a survey of more than 60,000 individuals has provided us with considerable statistical capacity, and enabled us to identify the underlying drivers of recreation demand in NSW protected areas. We were able to isolate the effect of a variety of site attributes, including natural values, infrastructure and contextual factors on overall site utility, with 32 out of 38 model parameters relating to park attributes returning coefficients that were significantly different to zero (Table 2). We estimated the current contribution made by the selected park attributes to overall park-level demand by applying the model coefficients presented in Table 2 to the existing configuration of the NSW protected area network. The average, maximum and min-

R2 = 0.92 *

•fc 4 **• —

200 400 600

Visitors (modelled, thousands)

Fig. 4. Correlation between modelled and measured (traffic counter) estimates of visitation at 31 national parks and reserves across NSW. *Some of the goodness of fit obtained in our correlation analysis arises from the outlying point at the top right of the graph. When we remove this outlier and repeat our analysis we retain good correlation between modelled and measured visitation estimates, with R2 = 0.74.

imum relative contribution made by remoteness, natural values and built infrastructure are provided in Fig. 5.

Our results indicate that built infrastructure has a much greater influence on overall site demand than the other attributes tested in this study. The relative contribution made by built infrastructure to overall site-level demand was, on average, tenfold that of either site remoteness or natural site values - which includes caves and landforms, rivers, wetlands and other natural values (Fig. 5). Within the current network configuration, built infrastructure has a positive effect on demand generation at most protected area sites (the grey shaded area in Fig. 5 represents mean relative contribution to demand ± one standard deviation - or approximately 70% of cases assuming a normal distribution).

Given the dominant influence of built infrastructure on recreation demand (Fig. 5), we have also investigated the relative contribution of specific built infrastructure types to overall site level demand. We have included attribute level multiplied by the model coefficient or the basic infrastructure term (e.g. 'paths') as well as all relevant interaction terms (e.g. paths (park) / paths (area) and paths / remoteness) in order to estimate the relative contribution of each infrastructure type within the current configuration of the NSW protected area network. We found that any type of built infrastructure can make a positive contribution to recreation demand in the right setting (Fig. 6). A small number of built infrastructure types made a positive contribution to recreation demand across the entire NSW protected area network irrespective of contextual factors. These were roads, parking facilities and look-outs -all of which are central pieces of infrastructure responsible for allowing, and improving, access to protected area sites and attractions within them. Other types of infrastructure, like amenities, day-use areas, retail outlets and paths & walking tracks had a predominantly positive effect, but could also bring disutility, depending on the setting in which they are located and/or the quantities provided across the state (see Section 3.3).

Modelling returned negative coefficients for a number of park values (historic heritage, rivers and wetlands, and biodiversity values) and built infrastructure types (built accommodation, day-use areas and camping areas; Table 2a). The literature offers a number of possible explanations for negative coefficients in travel cost studies. The first possibility is that the model has been mis-specified, either through the omission of key variables, or through the inclusion of correlated or co-linear variables (Pendleton and Mendelsohn, 2000). We have tested whether these the negative coefficients observed in our study are likely to arise from correlations amongst descriptor variables using a comparative modelling approach following Pendleton et al. (1998). We compared coefficients for the relevant infrastructure types (built accommodation,

Table 2

Model coefficients (and standard error) for all included model variables. Asterisks indicate level of significance: *p < .1, **p < .05, ***p < .01.

A - Primary effects of park attributes

Effect type Dependent variable Units Coefficient Std Error

Constant -3.034** 0.156

National Park attributes Conservation status IUCN cat. 0.074** 0.024

Remoteness ARIA++ -0.159** 0.031

Size log (ha) -0.091* 0.045

National Park values Aboriginal heritage presence/absence 0.493** 0.074

Historic heritage -0.177 0.095

Caves and other landforms 0.234* 0.091

Marine or estuarine 0.165 0.099

River or wetland -0.719** 0.100

Biodiversity values -0.585** 0.092

In-park infrastructure Paths & walking tracks log km 0.723** 0.045

Lookouts & other features P No. 0.036 0.033

Roads log km 0.259** 0.034

Built retail outlets P No. 0.444** 0.073

Built accommodation P No. -0.167** 0.038

Day-use areas P No. -0.483** 0.054

Camping areas P No. -0.308** 0.060

Parking areas log 100s m2 0.609** 0.050

Amenity blocks P No. 0.046 0.050

B - Contextual effects

Effect type Dependent variable Units Coefficient Std Error

Values in a spatial context Aboriginal heritage * ARIA++ presence/absence -0.007 0.026

Historic value * ARIA++ -0.151** 0.033

Caves & landforms * ARIA++ -0.098** 0.031

River values * ARIA++ 0.273** 0.034

Infrastructure interaction Paths & walking tracks Infrastructure at park of -0.044** 0.004

terms (in-park Lookouts & other features interest (transformed as 0.019** 0.004

infrastructure) Roads above) * (infrastructure in -0.0002 0.001

Built retail outlets relevant NPWS area) /1000 -0.655** 0.107

Built accommodation 0.032** 0.005

Day-use areas 0.275** 0.020

Camping areas -0.054** 0.020

Parking areas -0.055** 0.015

Infrastructure interaction Paths & walking tracks Infrastructure at park of -0.125** 0.014

terms (non-park Lookouts & other features interest (transformed as 0.017 0.015

infrastructure & Roads above) * ARIA++ at park 0.019* 0.008

services) Built retail outlets centroid 0.121** 0.028

Built accommodation -0.033* 0.015

Day-use areas 0.119** 0.025

Camping areas 0.101** 0.019

Parking areas -0.016 0.012

C - Travel cost and demographic effects

Effect type Dependent variable Units Coefficient Std Error

Travel cost $, 100's -0.403** 0.013

Respondent home NSW metro dummy coding 1.118** 0.104

region NSW regional 1.348** 0.107

ACT 0.856** 0.152

VIC metro -0.109 0.125

VIC regional 0.146 0.161

QLD metro -0.078 0.135

QLD regional - -

Respondent Sex dummy coding -0.233** 0.041

characteristics Age_18-24 -0.101 0.072

Age_25-34 0.097 0.060

Age_35-49 0.141** 0.054

Age_50 + - -

Children in household No. -0.005 0.021

Survey year 2008 dummy coding 0.652** 0.058

2010 -0.199** 0.063

2012 -0.214** 0.061

2014 - -

Visitation thresholds Mu(01) 2.88*** 0.047

Mu(02) 3.95*** 0.060

Mu(03) 4.42*** 0.067

E.C. Heagney et al. /Ecosystem Services xxx (2017) xxx-xxx

Fig. 5. Relative contribution to demand made by natural park values, built infrastructure and site remoteness (primary influence only - as per schematic diagram in Fig. 2). Relative contributions have been estimated based on average attribute levels observed across the NSW protected area network (Table 1).

Fig. 6. Relative demand for built infrastructure in response to contextual factors A) remoteness and B) on-park substitutes. Relative contributions have been estimated based on average attribute levels observed across the NSW protected area network (Table 1).

day-use areas and camping areas) when we a) omitted individual infrastructure types from the model one at a time, and b) omitted interaction terms relating to regional infrastructure loads and offpark substitutes. The results of this step-wise testing procedure (Table 3) indicate that the negative coefficients relating to these three infrastructure types are unlikely to have arisen as a result of mis-specification.

We adopt an alternate explanation: that the negative coefficients relating to built accommodation, day-use areas and camping areas reflect some level of disutility arising from these infrastructure types. The literature suggests that negative travel cost coefficients may indicate disutility that is limited to only a subset of users (Train, 1998), or it may indicate disutility at high levels of provision - termed 'oversatiation' by Pendleton and Mendelsohn

E.C. Heagney et al./Ecosystem Services xxx (2017) xxx-xxx

Table 3

Coefficients arising from alternate model runs with omitted model terms.

Original 1 2

10 11 12

Term omitted from modelling Coefficient relating to

Historic value River or wetland Natural values Accommodation Day-use areas Camping areas

Paths & walking tracks Lookouts & other features Roads

Built retail outlets

Built accommodation

Day-use areas

Camping areas

Parking areas

Amenity blocks

Park region interaction terms

Remoteness interaction terms

Both interaction terms

Average coefficient value (standard error)

% of model runs with negative sign

v0.177 0.105 -0.218 0.012 -0.164 -0.301 -0.122 -0.363 0.143 -0.206 -0.261 -0.268 -0.195

-0.155

-0.719

-0.734

-0.661

-0.814

-0.494

-0.736

-0.661

-0.878

-0.826

-0.714

-0.559

-0.867

-0.752

-0.585 -0.501 -0.362 -0.512 -0.417 -0.676 -0.423 -0.544 -0.606 -0.604 -0.558 -0.517 -0.46

-0.520

-0.167 -0.098 -0.147 -0.235 -0.165

-0.103

-0.297

-0.137

-0.021

-0.088

-0.014

-0.134

-0.483 -0.361 -0.408 -0.459 -0.256 -0.363

-0.417

-0.465

-0.234

-0.303

-0.049

-0.352

-0.308 -0.175 -0.213 -0.433 -0.188 -0.155 -0.25

-0.288

-0.194

-0.197

(2000). In our study, the observed negative infrastructure coefficients were moderated by positive interaction terms: built accommodation and day-use areas had higher levels of utility when they were part of a regional network of on-park infrastructure (positive infrastructure (park) /infrastructure (region) interaction terms in Table 2c); day-use and camping areas showed increasing utility in more remote locations (positive infrastructure / remoteness terms in Table 2b). This suggests that all infrastructure types can provide utility in the right setting. But it also indicates that there may be some mis-match between visitor preferences and current configuration of the NSW protected area network. We suggest that coefficients relating to specific infrastructure types should only be used to inform management planning when key contextual factors (regional infrastructure load and site remoteness) are also taken into account.

Similar arguments apply to the negative coefficients observed for protected area 'value' attributes. For rivers and wetlands we observed increasing utility at more remote locations, indicating a context specific effect on utility, as described for built infrastructure types above. But we consider that two additional factors are likely to be contributing to the negative coefficients observed for historic heritage and biodiversity values. First, parks values were assigned by national parks rangers or other conservation professionals, and, in many cases, they relate to areas or aspects of a protected area that tourists will never see (e.g. rare, threatened or endangered species, protected cultural sites). In this context it makes sense that tourists are opting for more accessible values like marine and estuarine values (which include beaches) and caves and landforms. Second, sites with very high biodiversity values or cultural values tend to be assigned a high IUCN status (in our study biodiversity values and historic heritage values are both correlated with IUCN status, p < .005 in each case). More restrictive conservation management at these sites (which may include limits on recreational infrastructure provision) may deter some visitors.

3.3. Investigating the role of contextual factors

We found a significant effect of site-level remoteness on recreation demand, with highest levels of visitation occurring at the least remote (metropolitan) protected areas, and lowest levels of visitation occurring at the most remote sites (Table 2). Most of the site attributes included in our study also displayed a significant interaction with protected area remoteness - i.e. the size of their contribution to recreation demand was significantly different in metropolitan versus rural or remote sites. This finding supports

the idea that recreational park users are seeking different protected area experiences in different settings. The relative influence of the 9 types of built infrastructure included in our model in generating recreation demand in metropolitan and remote protected areas is summarised in Fig. 7.

Our modelling identified a significant effect of substitution with on-park infrastructure in surrounding protected areas for 7 out of 8 built infrastructure types tested. Four of these cases indicated significant negative substitution effects, whereby the total recreation demand for a specified region is likely to be shared across a multiple sites with similar infrastructure types (paths and walking tracks, built retail outlets, camping areas and parking areas). The remaining three were complementarity effects, whereby tourism demand for a specified region is enhanced by the availability of similar infrastructure across multiple protected area sites (lookouts and features, day-use areas and built accommodation) (Table 2, Fig. 6b).

4. Discussion

4.1. Drivers of recreation demand - the role natural values, built infrastructure and site remoteness

Of all the site attributes modelled in our study, built recreational infrastructure was the primary contributor to demand generation at protected area sites. In the context of the NSW protected area network, built infrastructure was the primary driver of recreation demand, with the largest and most consistent contribution made by roads and parking, key pieces of infrastructure that enable site access. The importance of built infrastructure, compared to the smaller influence of natural site values like biodiversity, rivers, wetlands, caves and other landforms, may seem surprising at first, given the notion that the opportunity to access biodiversity, landscapes and other natural wonders is touted as one of the key goals and/or benefits associated with establishing and maintaining protected areas. However, our model indicates that the provision of any protected area generates some 'base level' recreation demand (as captured in the constant term of our model). This base level demand may be attributed, in part, to visitors' expectation that these sites will contain some special or important natural value above and beyond those found in other natural settings in order to warrant their protection; under that assumption it is only the variation in visitation in response to natural values that is small. Moreover, if we assume that all protected area sites contain natural

Metro parks

Parking areas Built retail outlets Roads

Paths & walking tracks Amenities Lookouts & other features Built accommodation Camping areas Day-use areas

Built retail outlets Roads Camping areas Day-use areas Parking areas Lookouts & other features Amenities Built accommodation Paths & walking tracks

Remote parks

Fig. 7. Ranking and relative demand for built infrastructure types in metropolitan versus remote protected areas. We have highlighted relative demand for paths and walking tracks (hatched) and day-use areas (cross-hatched) to demonstrate the role of contextal factors; walking tracks are ranked 4/9 in metropolitan parks but 9/9 in remote parks; day use areas are ranked 9/9 in metropolitan parks but 4/9 in remote parks.

values, then it is the ability to access those sites that is important in a recreational context. Our findings regarding the high value visitors assign to site accessibility are consistent with other studies into the recreational value of protected areas. For example, a travel cost study of 16 protected area sites in Mallorca found that, out of 10 site attributes tested, provision of parking areas had the largest positive correlation with overall site visitation (Font, 2000); a contingent valuation survey undertaken in the Phu Kradueng National Park in Thailand found that visitors' willingness to pay to access the site more than doubled in response to proposed road improvements and other site upgrades (Boontho, 2008).

Our analysis also identified protected area remoteness as a significant driver of recreation demand, with higher levels of visitation observed at less remote sites. A similar finding is reported by Balmford et al., (2015), who identified protected area size, remoteness and natural attractiveness, along with the size of the surrounding local population, as key determinants of visitation at the global scale. Our observed relationship between remoteness and visitation also mirrors findings from a number of small scale travel cost studies; proximity to population or urban centres has been identified as a key (and usually primary) driver of recreation demand for protected areas and other forest or parkland sites (Arnberger, 2006; Zandersen and Tol, 2009; Kim et al., 2010; Lankia et al., 2015). Lansdell and Gangadharan (2003) have termed this type of demand generation 'proximity power':

''meaning that a park situated in a highly populated area is likely to

have a high consumer surplus due to its convenient location, thus

attracting large number of visitors from nearby"

In the definition presented above, Lansdell and Gangadharan (2003) refer to a site-scale effect of location, but our analysis indicates that only a small proportion of recreation demand arises directly in response to the remoteness of the protected area site itself. Most of the variation in recreation demand observed in our study was captured through the influence of remoteness on demand for specific types of infrastructure (Fig. 6). These findings support the idea that recreational users are seeking different protected area experiences in different settings.

4.2. Why do people visit parks?

In the broadest sense, our findings indicate that people have different reasons or motivations for visiting protected areas in different settings. Our results identify some general trends in recreation demand in response to specific site attributes and contextual factors. These allow us to provide more tangible guidance for managers to optimise the provision of recreational opportunities from protected areas at both the site- and network- scales.

We observed high demand for walking tracks in urban settings (Figs. 6 and 7), suggesting that protected areas in metropolitan centres can be important for exercise or other physical activity. Support for this hypothesis comes from Arnberger (2006), who found a greater proportion of visitors engaged in 'routine' activities like dog-walking and jogging in inner-urban (compared with periurban) forests. Similarly, Sugiyama et al. (2010) found that adults living near attractive urban parks in Western Australia were more likely to engage in higher levels of walking exercise. Paths and

walking tracks may also be popular because they provide access to natural values and nature-based experiences in the interior of protected areas; such access is likely to be of relatively greater importance in a metropolitan setting given that modern high-density cities show declining per-capita access to public greenspaces (Fuller and Gaston, 2009). Both these possibilities are supported by recent research into health and wellbeing benefits provided by greenspaces in urban settings. For example, a study by Thompson et al. (2012) found that both the area of greenspace provided and independently reported levels of physical activity at those sites improved the overall wellbeing of recreational greenspace users in Dundee in Scotland. A review of urban greenspace literature undertaken by Kabisch et al. (2015) found that of 30 studies into health and wellbeing, most reported a benefit associated with accessing greenspace experiences or exercising in urban parklands.

In more remote settings across NSW, visitor demand was associated with facilities like day-use areas, retail and camping, perhaps reflecting a lack of alternative facilities in the surrounding area. Indeed, most types of infrastructure that were included in our modelling (with the exception of paths & walking tracks which are discussed in the previous paragraph) made a greater contribution to tourism demand in protected areas in increasingly remote locations. A study of recreation of Korean forest sites made similar recommendations regarding visitors' preference for camping areas and picnic tables at remote forest sites (compared with an identified need for amenities and walking tracks at less remote locations) (Kim et al., 2010). This provides some suggestion that trends observed in our study may be transferable to other park networks.

4.3. Implications for management - Optimising recreation potential of protected area sites

The significant interaction between built infrastructure and remoteness reported from our study suggests that a one-size fits all approach is unlikely to be suitable for management of protected areas across NSW, particularly when it comes to the provision of built infrastructure. A similar conclusion was made by Cho et al. (2014) in their travel cost study of 5 national forest sites across the USA. That study recommends investment in different types of built infrastructure at different sites on the basis of variation in elasticity of demand observed through travel cost modelling. But there is very limited capacity to generalise these findings to other sites, as recommendations about infrastructure preferences were reported on a site by site basis for a small number of sites (n = 5) without any investigation of the role of underlying site attributes or contextual factors. This is not an isolated problem. A recent review of the international literature relating to recreational use of urban greenspace reports that there are very few comparative multiple site studies Kabisch et al. (2015). Given that our study draws on trends in visitation across a very large number of diverse sites, we have been able to make some more general observations regarding the types of infrastructure that are likely to make a substantive contribution to recreation demand for protected area sites within a specific context (as presented in Fig. 7). Our observations are supported by studies undertaken elsewhere in the world (e.g. Arnberger, 2006; Kim et al., 2010) but additional large scale random utility travel cost studies across a much broader range of socio-economic contexts will be required to further develop and refine the trends and relationships observed in our study.

Managing and optimising recreational opportunities from protected areas requires planning at regional- and network- scales. The key influence of contextual factors, like remoteness, in generating recreation demand means that a substantial proportion of the recreation potential of a protected area is determined at the time of site acquisition. Protected area management agencies that include the provision of recreational experiences as one of their key man-

agement objectives should therefore assess the 'recreational potential' of any proposed protected area as part of site acquisition planning processes. Understanding recreational potential of a particular site from the initial planning phases provides an opportunity to optimise both recreational experiences and the associated economic benefits they can provide to the surrounding local community. Consideration should be given to both 'raw site potential' (provided by existing site factors) and 'improved site potential' (visitation demand that might be generated at the site with additional investment in appropriate infrastructure types). The latter should obviously be considered in tandem with current and likely future budgetary capacity to provide appropriate built infrastructure improvements at the site under consideration.

Our findings also suggest that there is opportunity for regional planning to develop protected area 'hubs' that focus on providing infrastructure with strong complementarity effects (e.g. lookouts, day-use areas and built accommodation) in order to leverage the regional-scale increase in demand that these infrastructure types can provide. Similar potential for complementarity effects are reported by Cho et al. (2014) and Henrickson and Johnson (2013) - although these relate to site-level complementarity. Our model coefficients relating to 1UCN status and other natural values (negative coefficient for 'natural values' and positive coefficient for '¡UCN status' - Table 2) indicated a visitor preferences for sites of lower conservation value; this suggests that, with sufficient planning, such hubs could be developed without compromising the conservation objectives within protected areas.

An additional important management application, beyond recreation management and optimisation, relates to the identification of proposed or existing protected area sites with very low recreation potential. Obviously, establishing protected areas on such sites is entirely valid if they help meet conservation targets. But it is important in these instances that protected areas are not sold' to the surrounding local community on the basis of any tourism potential. Over-selling the economic benefits of protected areas and related conservation strategies has been reported from a number of studies (Fabricius et al., 2001; Minang and van Noordwijk, 2013) and may ultimately lead to community disappointment (Naidoo and Ricketts, 2006), disengagement, or even hostility towards protected area initiatives. Moreover, if management agencies seek to ensure that protected areas contribute to the economic wellbeing of the surrounding local community, as consistent with UNESCO (1996) and 1UCN (1998) guidelines, then they should be aware of cases where these will need to rely on some economic stimulus outside the tourism and recreation sector.

4.4. Implications for recreational travel cost studies

1dentifying and characterising substitutes for protected areas has been an enduring challenge in travel cost literature (Gurluk and Rehber, 2008), and as a result substitution effects are often omitted from studies that seek to value any individual protected area, an approach that can bias resultant estimates of consumer surplus (Liston-Heyes and Heyes, 1999). Where investigations have included substitution and complementarity effects, they have generally been modelled in fairly coarse terms. For example, Henrickson and Johnson (2013) have used 'No. of parks within 1000 miles' in their analysis of demand for spatially complementary protected areas; they also cite five other papers that have employed this approach. Blaine et al. (2015) report that studies of networks of protected areas are somewhat different to site-based studies in that visitation rates to individual parks can reasonably be assumed to encompass substitution effects from surrounding parks or other opportunities. While this is true to some degree, our work suggests that more complex modelling is required to adequately address substitution or complementarity

E.C. Heagney et al. /Ecosystem Services xxx (2017) xxx-xxx

effects relating to both on- and off-park substitutes. Moreover, our results suggest that substitution and complementarity effects vary substantially for different types of recreational infrastructure on offer and the availability of substitutes should be modelled at this (rather than the broader site) level.

Our results highlight two additional areas of concern relating to traditional travel cost studies. First, our study has identified a large number of site attributes that have a significant influence on the value of recreation in protected areas; 32 of the 38 attributes (including interaction terms) tested were statistically significant drivers of tourism within the NSW protected area network. It follows that random utility or hedonic travel costs studies that include only a small number of sites, and are hence restricted to investigating the influence of a relatively small number of site attributes on recreation demand (usually many fewer than 32) are likely to omit important variables and yield mis-specified models. Second, we observed major differences in recreational preferences at metropolitan versus remote protected area sites. These differences in recreational preferences mean that benefit transfer from a single site study, or a multi-site study that restricts sampling to high-profile, high-visitation parks, could lead to very poor outcomes for recreational planning for remote protected areas, even when benefit transfer is undertaken on the basis of a utility function. We encourage additional large scale random utility travel cost studies that test the influence of a large number of protected area attributes across a more diverse range of contexts in order to help refine the trends and relationships reported in our study, and to facilitate more robust benefit transfer to inform the management of recreational values at other protected area sites.

We recognise that there may be logistical and budgetary restrictions to undertaking large scale travel cost studies of all sites across an entire protected area network. Such an approach will not even always be desirable where there is particular research or management interest at a smaller number of specific sites. In smaller studies some consideration should be given to random zero-inflation as a statistical correction that can help provide better discrimination of the attributes that make the greatest contribution to recreation demand. The novel zero inflation technique proposed and applied in this study, whereby non-visitors are randomly assigned to a very low (zero) visitation site and a 'zero' count for the respondent's (hypothetical) trip is recorded against that site - provides new potential to capture data relating to attribute levels of very low visitation sites and incorporate them into econometric modelling without the need for detailed and costly on-site data collection. We envisage that travel cost studies of individual sites might use a further adapted zero inflation technique that randomly assign hypothetical visitors drawn randomly from a representative population distribution to sites where visitation is known (or reported by local management staff) to be very low. This is akin to the use of zero inflation in the ecological literature as described by Warton (2005).

5. Conclusions

Our study represents the only random utility travel cost study of a large protected area network undertaken to date. Our results point to a key role for built infrastructure in generating tourism demand for protected areas. But only a small number of key infrastructure types, those that are necessary for enabling site access (roads & parking areas), have the capacity to generate demand across the full range of contextual factors observed in our study. For all other infrastructure types, we found a significant effect of contextual factors, indicating that context-specific planning is necessary if the recreational value of protected area sites is to be optimised. Our study suggests that protected area planning processes

should consider recreation potential during the earliest stages of site-acquisition planning, and ensuring that due consideration is given to site-, regional-, and network- scale effects. Our results are supported by those of smaller studies across a range of jurisdictions, including Vienna (Arnberger, 2006) and Korea (Kim et al., 2010), suggesting that the insights from our study might be transferable to other protected area networks around the globe. We encourage similar research across a wider range of contexts so that the general guidelines for optimising recreational opportunities provided by protected areas identified in our study can be validated and refined.

Although this paper highlights the value of recreation in protected areas, it is important to note that protected areas offer a range of other ecosystem services, including carbon sequestration and water filtration services (Dudley and Stolton, 2010; Palomo et al., 2013). They also provide a number of passive or non-use values, including existence and bequest values (Turner et al., 1994). A recent study by Haefele et al. (2016) found that recreational values comprised less than half the total economic value of the U.S. National Parks System; non-use values associated with protected areas and associated conservation programs comprised more than half of the total economic value. It follows that any management strategy that seeks to optimise the recreational values of protected area sites should by wary of the potential for adverse impacts on natural site values and conservation objectives.

In reality, it is difficult to separate conservation values from infrastructure provision in protected areas, as the two may be (deliberately) linked through a number of park management strategies. Infrastructure provision and associated visitor access can have a range of adverse impacts on protected ecosystems (Esteves et al., 2011). Infrastructure investment within a protected area network may be concentrated at less ecologically or culturally sensitive locations in order to limit access and associated impacts. Alternatively, infrastructure may be concentrated at ecologically or culturally sensitive sites, or very high visitation sites, in order to reduce the impacts of inappropriate or unregulated recreational access (e.g. walking tracks to reduce trampling effects, formal parking areas, tracks to manage and direct access routes, waste facilities to avoid inappropriate disposal). The degree to which these different management strategies have been implemented within the NSW protected area network, and the degree to which they have impacted both conservation goals and visitors' recreational experiences, warrants further investigation. Our future modelling will focus on if and how the new insights into visitor preferences presented in this current paper can be used to help balance conservation and recreation at both site- and landscape- scales.

References

Amoako-Tuffour, J., Martinez-Espineira, R., 2012. Leisure and the net opportunity cost of travel time in recreation demand analysis: An application to Gros Morne national park. J. Appl. Econ. 15, 25-49.

Anderson, D.M., 2010. Estimating the economic value of ice climbing in Hyalite Canyon: an application of travel cost count data models that account for excess zeros. J. Environ. Manage. 91, 1012-1020.

Armsworth, P., Chan, K., Daily, G., Ehrlich, P., Kremen, C., Ricketts, T., Sanjayan, M., 2007. Ecosystem-service science and the way forward for conservation. Conserv. Biol. 21, 1383-1384.

Arnberger, A., 2006. Recreation use of urban forests: an inter-area comparison. Urban For. Urban Greening 4,135-144.

Arrow, K., Lehmann, E., 2005. Harold Hotelling 1895-1973. Biographical Memoirs 87, 3-11.

Australian Bureau of Statistics (ABS), 2015. Catalogue 8635.0 - Tourist Accommodation, Australia, 2014-15. http://www.abs.gov.au/ausstats/abs®. nsf/mf/8635.0 accessed 27/07/2016.

Balmford, A., Beresford, J., Green, J., Naidoo, R., Walpole, M., Manica, A., 2009. A global perspective on trends in nature-based tourism. PLoS Biol 7, e1000144.

Balmford, A., Green, J.M., Anderson, M., Beresford, J., Huang, C., Naidoo, R., Walpole, M., Manica, A., 2015. Walk on the wild side: estimating the global magnitude of visits to protected areas. PLoS Biol 13, e1002074.

Beal, D.J., 1995. A travel cost analysis of the value of Carnarvon Gorge National Park for recreational use. Rev. Marketing Agric. Econ. 63, 292-303.

Blaine, T.W., Lichtkoppler, F.R., Bader, T.J., Hartman, T.J., Lucente, J.E., 2015. An examination of sources of sensitivity of consumer surplus estimates in travel cost models. J. Environ. Manage. 151, 427-436.

Boontho, C., 2008. An economic analysis of phu kradueng national park. Pages 337341. In: Ardil, C. (Ed.), Proceedings of World Academy of Science. Engineering and Technology.

Brouwer, R., 2000. Environmental value transfer: state of the art and future prospects. Ecol. Econ. 32,137-152.

Brown Jr., Mendelsohn, G.R., 1984. The hedonic travel cost method. Rev. Econ. Stat., 427-433

Cho, S.-H., Bowker, J.M., English, D.B.K., Roberts, R.K., Kim, T., 2014. Effects of travel cost and participation in recreational activities on national forest visits. For. Policy Econ. 40, 21-30.

Driml, S., Common, M., 1995. Economic and financial benefits of tourism in major protected areas. Aust. J. Environ. Manage. 2,19-29.

Dudley, N., Stolton, S., 2010. Arguments for protected areas: multiple benefits for conservation and use. Routledge.

Englin, J.E., McDonald, J.M., Moeltner, K., 2006. Valuing ancient forest ecosystems: an analysis of backcountry hiking in Jasper National Park. Ecol. Econ. 57, 665678.

Englin, J., Shonkwiler, J.S., 1995. Estimating social welfare using count data models: an application to long-run recreation demand under conditions of endogenous stratification and truncation. Rev. Econ. Stat., 104-112

Esteves, C.F., de Barros Ferraz, S.F., de Barros, Micchi, Ferraz, K.M.P., Galetti, M., Theobald, D.M., 2011. Human accessibility modelling applied to protected areas management. Natureza Conservacao 9, 232-239.

Fabricius, C., Koch, E., Magome, H., 2001. Towards strengthening collaborative ecosystem management: lessons from environmental conflict and political change in southern Africa. J. R. Soc. N. Z. 31, 831-844.

Ferraro, P.J., Hanauer, M.M., Sims, K.R., 2011. Conditions associated with protected area success in conservation and poverty reduction. Proc. Natl. Acad. Sci. 108, 13913-13918.

Fleming, C.M., Cook, A., 2008. The recreational value of Lake McKenzie, Fraser Island: an application of the travel cost method. Tourism Manage. 29, 11971205.

Flynn, M., Francis, L.A., 2009. More flexible glms zero-inflated models and hybrid models. Casualty Actuarial Soc. 2009,148-224.

Font, A.R., 2000. Mass tourism and the demand for protected natural areas: A travel cost approach. J. Environ. Econ. Manage. 39, 97-116.

Fortin, M.-J., Gagnon, C., 1999. An assessment of social impacts of national parks on communities in Quebec, Canada. Environ. Conserv. 26, 200-211.

Fuller, R.A., Gaston, K.J., 2009. The scaling of green space coverage in European cities. Biol. Lett. 5, 352-355.

Gurluk, S., Rehber, E., 2008. A travel cost study to estimate recreational value for a bird refuge at Lake Manyas, Turkey. J. Environ. Manage. 88,1350-1360.

Haefele, M., Fort Collins, C., Loomis, J., Bilmes, L.J., 2016. Total economic valuation of the national park service lands and programs.

Hanink, D.M., White, K., 1999. Distance effects in the demand for wildland recreational services: the case of national parks in the United States. Environ. Plann. A 31, 477-492.

Heagney, E., Kovac, M., Fountain, J., Conner, N., 2015. Socio-economic benefits from protected areas in southeastern Australia. Conserv. Biol. 29,1647-1657.

Heagney, E.C., Rose, J.M., Ardeshiri, A., Kovac, M., 2016. The economic value of tourism and recreation across a large protected area network. Technical paper. Economic and Strategic Research Branch, NSW Office of Environment and Heritage, Sydney.

Henrickson, K.E., Johnson, E.H., 2013. The demand for spatially complementary national parks. Land Economics 89, 330-345.

IUCN (International Union for Conservation of Nature), 1998. Economic values of protected areas: guidelines for protected area managers. IUCN, Gland, Switzerland.

Kabisch, N., Qureshi, S., Haase, D., 2015. Human-environment interactions in urban green spaces—A systematic review of contemporary issues and prospects for future research. Environ. Impact Assess. Rev. 50, 25-34.

Kim, B.-S., Noh, J., Jong-Chan, P., 2010. A Study on the Users' attribute of national recreation forest in and around seoul national capital area - Sanem and doota national recreation forest. J. Korean Inst. For. Recreation 14, 98-107.

Lankia, T., Kopperoinen, L., Pouta, E., Neuvonen, M., 2015. Valuing recreational ecosystem service flow in Finland. J. Outdoor Recreation Tourism-Res. Plann. Manage. 10,14-28.

Lansdell, N., Gangadharan, L., 2003. Comparing travel cost models and the precision of their consumer surplus estimates: Albert Park and Maroondah Reservoir. Aust. Econ. Papers 42, 399-417.

Liston-Heyes, C., Heyes, A., 1999. Recreational benefits from the Dartmoor National Park. J. Environ. Manage. 55, 69-80.

Loomis, J.B., 1992. The evolution of a more rigorous approach to benefit transfer: Benefit function transfer. Water Resour. Res. 28, 701-705.

McFadden, D., 1973. Conditional logit analysis of qualitative choice behavior.

Minang, P.A., van Noordwijk, M., 2013. Design challenges for achieving reduced emissions from deforestation and forest degradation through conservation: leveraging multiple paradigms at the tropical forest margins. Land Use Policy 31, 61-70.

Morgan, Roy, 2015. Annual Visits to PWG Managed Parks in New South Wales. Roy Morgan Research Ltd., Sydney.

Naidoo, R., Ricketts, T.H., 2006. Mapping the economic costs and benefits of conservation. PLoS Biol 4, e360.

Neher, C., Duffield, J., Patterson, D., 2013. Valuation of national park system visitation: the efficient use of count data models, meta-analysis, and secondary visitor survey data. Environ. Manage. 52, 683-698.

Orr, S.K., 2011. The private sector on public land: Policy implications of a SWOT analysis of Banff National Park. J. Nat. Resour. Policy Res. 3, 341-354.

Palomo, I., Martin-Lopez, B., Potschin, M., Haines-Young, R., Montes, C., 2013. National Parks, buffer zones and surrounding lands: mapping ecosystem service flows. Ecosyst. Serv. 4,104-116.

Pendleton, L., 1999. Reconsidering the hedonic vs. RUM debate in the valuation of recreational environmental amenities. Resour. Energy Econ. 21,167-189.

Pendleton, L., Mendelsohn, R., 2000. Estimating recreation preferences using hedonic travel cost and random utility models. Environ. Resour. Econ. 17, 89108.

Pendleton, L., Sohngen, B., Mendelsohn, R., Holmes, T., 1998. Measuring environmental quality in the southern Appalachian Mountains. For. Sci. 44, 603-609.

Plummer, M.L., 2009. Assessing benefit transfer for the valuation of ecosystem services. Front. Ecol. Environ. 7, 38-45.

Saraj, S.B., Yachkaschi, A., Oladi, J., Teimouri, F.S., Latifi, H., 2009. The recreational valuation of a natural forest park using travel cost method in Iran. Iforest-Biogeosci. For. 2, 85-92.

Selby, A., Petajisto, L., Huhtala, M., 2011. The realisation of tourism business opportunities adjacent to three national parks in southern Finland: entrepreneurs and local decision-makers matter. For. Policy Econ. 13, 446-455.

Sims, K.R., 2010. Conservation and development: Evidence from Thai protected areas. J. Environ. Econ. Manage. 60, 94-114.

Stevens, T.H., More, T.A., Markowski-Lindsay, M., 2014. Declining national park visitation: An economic analysis. J. Leisure Res. 46, 153.

Sugiyama, T., Francis, J., Middleton, N.J., Owen, N., Giles-Corti, B., 2010. Associations between recreational walking and attractiveness, size, and proximity of neighborhood open spaces. Am. J. Public Health 100,1752-1757.

Thompson, C.W., Roe, J., Aspinall, P., Mitchell, R., Clow, A., Miller, D., 2012. More green space is linked to less stress in deprived communities: evidence from salivary cortisol patterns. Landscape Urban Plann. 105, 221-229.

Timah, P.N., 2011. Non-market valuation of beach recreation using the travel cost method (TCM) in the context of the developing world.

Tourism Research Australia, 2014. Travel by Australians December 2014 quarterly results of the National Visitor Survey. http://tra.gov.au/documents/nvs/Travel_ by_Australians_December_2014_results_of_the_National_Visitor_Survey.html# OTP_Overnighttrips (accessed November 2016).

Train, K.E., 1998. Recreation Demand Models with Taste Differences Over People. Land Economics 74, 230-239.

Turner, R.K., Pearce, D., Bateman, I., 1994. Environmental economics: an elementary introduction. Harvester Wheatsheaf.

UNESCO, 1996. Biosphere reserves: The Seville strategy & the statutory framework of the world network. UNESCO Man and the Biosphere Programme, Paris.

University of Adelaide (2015) Australian Population and Migration research Centre. ARIA ++ (Accessibility/Remoteness Index of Australia). http://www.adelaide. edu.au/apmrc/research/projects/category/about_aria.html Accessed 27/27/ 2016.

Von Haefen, R.H., Phaneuf, D.J., 2003. Estimating preferences for outdoor recreation: a comparison of continuous and count data demand system frameworks. J. Environ. Econ. Manage. 45, 612-630.

Warton, D.I., 2005. Many zeros does not mean zero inflation: comparing the goodness-of-fit of parametric models to multivariate abundance data. Environmetrics 16, 275-289.

Washington, S. et al., 2014. Bayesian imputation of non-chosen attribute values in revealed preference surveys.'' .J. Adv. Transp. 48 (1), 48-65.

Zandersen, M., Tol, R.S., 2009. A meta-analysis of forest recreation values in Europe. J. For. Econ. 15, 109-130.