1. Introduction
Large numbers of visitors to natural World Heritage sites may threaten their conservation and is thus a major issue for their management (Fei et al., Reference Fei, Xiong, Fei, Zhang and Zhang2023). Prominent examples are the Galápagos Islands, the Historic Sanctuary of Machu Picchu and the Iguazu National Park, which are formally identified as being threatened by visitation in different UNESCO State of Conservation reports and the Conservation Outlook assessment reports. Besides this, the availability of measures of visitation and threats to natural sites and their possible comparability is scarce, leading researchers to rely on different kinds of online information. The sites mentioned above also receive an extensive number of TripAdvisor reviews, for instance. However, the formal reporting of threats does not always coincide with a presumptive one identified based on information from TripAdvisor. One example is the Teide National Park, where no formal tourism threat is identified despite many visitors. Thus, this raises the question of what kind of indicators best reflect a threat and whether data based on TripAdvisor reviews is useful in this context.
The aim of this study is to explore if threats to natural World Heritage sites (WHSs) can be identified based on data from an online travel platform. By doing so, this digitally sourced measure of interest in natural and mixed WHSs is contrasted with the number and severity of threats as formally recognised by UNESCO (State of Conservation database) and IUCN (Conservation Outlook Assessment) when typical site characteristics are accounted for. TripAdvisor reviews are used as the digital measure and the characteristics provided by UNESCO include size, year of inscription, kind of site (marine and coastal, forests, religious and sacred, mountains, glaciers, rivers and islands) as well as a distinction between mixed and entirely natural sites. By use of random effects ordered probit as well as count data models, the empirical approach establishes if the locations that experience the largest interest are also the ones most vulnerable to visitor flows.
Tourism to natural WHSs may also render positive effects (Chopra and Adhikari, Reference Chopra and Adhikari2004; Kido and Seidl, Reference Kido and Seidl2008; Taylor et al., Reference Taylor, Hardner and Stewart2009), although these aspects are not considered in the present study. Nor are other kinds of threats to natural WHSs addressed, such as illegal activities (Johannesen, Reference Johannesen2005; Veillon, Reference Veillon2014), wars and civil unrest (Levin et al., Reference Levin, Ali, Crandall and Kark2019).
A growing number of analyses use data from social media or mobile devices to measure the digital footprint of persons visiting protected areas or nature-based sites (Levin et al., Reference Levin, Kark and Crandall2015; Sessions et al., Reference Sessions, Wood, Rabotyagov and Fisher2016; Tenkanen et al., Reference Tenkanen, Di Minin, Heikinheimo, Hausmann, Herbst, Kajala and Toivonen2017; Ghermandi and Sinclair, Reference Ghermandi and Sinclair2019; Correia et al., Reference Correia, Ladle, Jarić, Malhado, Mittermeier, Roll, Soriano-Redondo, Veríssimo, Fink, Hausmann, Guedes-Santos, Vardi and Di Minin2021; Ghermandi, Reference Ghermandi2022; Liang et al., Reference Liang, Yin, Pan, Lin, Miller, Taff and Chi2022; Ghermandi et al., Reference Ghermandi, Langemeyer, Van Berkel, Calcagni, Depietri, Vigl and Wood2023). This approach is increasingly regarded as a powerful ex-post tool for assessing and tracking spatial trends in human-nature relationships and understanding conservation debates (Souza et al., Reference Souza, Rodrigues, Correia, Normande, Costa, Guedes-Santos, Malhado, Carvalho and Ladle2021; Ghermandi, Reference Ghermandi2022). Examples are Wikipedia page views (Guedes-Santos et al., Reference Guedes-Santos, Correia, Jepson and Ladle2021; Mittermeier et al., Reference Mittermeier, Correia, Grenyer, Toivonen and Roll2021; Falk and Hagsten, Reference Falk and Hagsten2022; Owuor et al., Reference Owuor, Hochmair and Paulus2023), Instagram posts (Tenkanen et al., Reference Tenkanen, Di Minin, Heikinheimo, Hausmann, Herbst, Kajala and Toivonen2017; Falk and Hagsten, Reference Falk and Hagsten2022), Twitter (Tenkanen et al., Reference Tenkanen, Di Minin, Heikinheimo, Hausmann, Herbst, Kajala and Toivonen2017; Hausmann et al., Reference Hausmann, Toivonen, Fink, Heikinheimo, Tenkanen, Butchart, Brooks and Di Minin2019) and Flickr photographs (Levin et al., Reference Levin, Kark and Crandall2015, Reference Levin, Ali, Crandall and Kark2019; Hausmann et al., Reference Hausmann, Toivonen, Fink, Heikinheimo, Tenkanen, Butchart, Brooks and Di Minin2019; Ghermandi et al., Reference Ghermandi, Camacho-Valdez and Trejo-Espinosa2020).
Several studies show that there is a high correlation between social media or other internet-based data on the one hand and actual visitor numbers on the other (Sessions et al., Reference Sessions, Wood, Rabotyagov and Fisher2016; Toivonen et al., Reference Toivonen, Heikinheimo, Fink, Hausmann, Hiippala, Järv, Tenkanen and Di Minin2019; Sinclair et al., Reference Sinclair, Mayer, Woltering and Ghermandi2020; Owuor et al., Reference Owuor, Hochmair and Paulus2023 as well as Ghermandi, Reference Ghermandi2022, for a review of the literature). Other research demonstrates that TripAdvisor reviews, specifically, are a good proxy of the number of visitors to the destination (Borowiecki et al., Reference Borowiecki, Pedersen and Mitchell2024).
There are, however, few examples in the literature that contrast threats from tourism to natural WHSs with the digitally sourced measure number of TripAdvisor reviews. Unlike Instagram, for instance, TripAdvisor provides detailed information about attractions and sub-areas within a site.
This study contributes to the discussion on how interest in natural WHSs can be measured in several ways. First, it contrasts tourism threats from two different official sources with a digital measure of interest in the site, when its characteristics and location are controlled for. The approach is influenced by Correia et al. (Reference Correia, Ladle, Jarić, Malhado, Mittermeier, Roll, Soriano-Redondo, Veríssimo, Fink, Hausmann, Guedes-Santos, Vardi and Di Minin2021), who emphasise the importance of multiple indicators for the monitoring of protected areas. In addition, ordered probit models with random effects capture the degree of tourism threats, something that is rarely applied in literature because of data deficits. The present study also contributes to the literature on the acceptable load limit since one of the determinants used is size of the site (McCool and Lime, Reference McCool and Lime2001; Selcuk et al., Reference Selcuk, Karakas, Cizel and Ipekci Cetin2023).
TripAdvisor reviews or ratings are increasingly used to analyse the behaviour of tourists and their interest in different attractions (Taecharungroj and Mathayomchan, Reference Taecharungroj and Mathayomchan2019; Yang et al., Reference Yang, Zhang and Fu2022; Borowiecki et al., Reference Borowiecki, Pedersen and Mitchell2024). It is the largest and most popular online travel platform in the world, where reviews can also be posted (Taecharungroj and Mathayomchan, Reference Taecharungroj and Mathayomchan2019). A typical advantage of the platform is the detailed information on the attractions within a destination (Taecharungroj and Mathayomchan, Reference Taecharungroj and Mathayomchan2019; Van der Zee et al., Reference Van der Zee, Camatti, Bertocchi and Shomali2024).
The study is structured as follows. Section 2 introduces the conceptual background while section 3 presents the empirical model. This is followed by an introduction to the database and descriptive statistics in section 4. The empirical results are reported in section 5, followed by discussions in section 6 and conclusions in section 7.
2. Conceptual background
In this section, typical research on threats to natural WHSs is highlighted. This includes discussions of traditional as well as digital data sources for the identification of threats. The section concludes with a display of the determinants used for the empirical analysis.
2.1 Typical research on threats to natural World Heritage sites
Although numerous studies analyse visitor pressure on natural WHSs, their approaches and measurements are disparate (table 1). The least common denominator appears to be a focus on iconic sites (see Mandić (Reference Mandić2023) for Plitvice Lakes National Park; Larson and Poudyal (Reference Larson and Poudyal2012) and Schlauderaff et al. (Reference Schlauderaff, Press, Huston, Su and Tsai2022) for Machu Picchu; and De Groot (Reference De Groot1983) and Kenchington (Reference Kenchington1989) for the Galápagos Islands). Certainly, sites without physical boundaries offer particular challenges to researchers in their analyses. This also means that non-traditional measurement of visitation is a possible path forward.
Studies on tourism threats to natural WHSs
Note: Selected literature from journals listed in Scopus.
Case studies of natural WHSs are common, and they document negative impacts in the form of mass tourism, noise, traffic congestion, crowds, visitor pressure and disturbance of the flora and fauna (table 1). Excessive tourism may degrade a site as documented in the case of Machu Picchu (Schlauderaff et al., Reference Schlauderaff, Press, Huston, Su and Tsai2022). Another example is the presence of waste at beaches near resorts in Ta Long Bay (Huong et al., Reference Huong, Lan and Le2024). In the vicinity of the WHS Mount Sanqingshan in China, the inhabitants are faced with a conversion of their agricultural land and forests into tourism businesses (Su et al., Reference Su, Wall and Xu2016).
Relatively few studies employ a large representative number of natural WHSs to analyse presumptive threats from visitors. Exceptions include Falk and Hagsten (Reference Falk and Hagsten2023), who investigate tourism threats to sites in Europe and North America. Another is Selcuk et al. (Reference Selcuk, Karakas, Cizel and Ipekci Cetin2023), who uses a large set of visitation pressure indicators on 24 natural WHSs. Based on the UNESCO State of Conservation database, Birendra (Reference Birendra2021) studies threats for 16 natural WHSs in danger of losing their inscription, where the tourism threat is measured as a dummy variable. None of the above studies employ TripAdvisor as a source for the attractiveness of natural WHSs, nor do they include a comparison with official threat data.
2.2 Data sources for identification of threats to natural World Heritage sites
Many studies on tourism threats to natural WHSs are based on data from official sources. There are three main official sources with comparable measures of threats. The first is the UNESCO periodic report (Falk and Hagsten, Reference Falk and Hagsten2023) and the second is the State of Conservation database (Veillon, Reference Veillon2014). Both databases have their advantages and disadvantages. The State of Conservation database is longitudinal, starting in 1979, although threats are only measured as dummy variables. In contrast to this, the periodic report is more detailed where the level of the threat is scaled. Yet, these data do not allow dynamic analyses since they appear in waves over several years that also differ across continents. A third data source is the Conservation Outlook Assessment, which contains more detailed measurements of the tourism threat, categorised by kind (Osipova et al., Reference Osipova, Emslie-Smith, Osti, Murai, Åberg and Shadie2020).
Besides official data on tourism threats, digital indicators based on different online sources such as travel or social media platforms are becoming increasingly popular for the identification of tourism pressure on sites (table 2). Flickr photos is one of the most popular platforms used as a source of information for protected areas worldwide (Levin et al., Reference Levin, Kark and Crandall2015), natural WHSs in the United States (Sessions et al., Reference Sessions, Wood, Rabotyagov and Fisher2016) and the Dolomites (Egarter Vigl et al., Reference Egarter Vigl, Marsoner, Giombini, Pecher, Simion, Stemle, Tasser and Depellegrin2021). Falk and Hagsten (Reference Falk and Hagsten2022), on the other hand, introduce both an ex-ante (Wikipedia page views) and an ex-post (Instagram hashtags) digital measure. TripAdvisor reviews and ratings are applied to European destinations, including those with a WHS (Van der Zee et al., Reference Van der Zee, Camatti, Bertocchi and Shomali2024).
Typical research on natural WHSs based on online data
Note: Selected literature from Scopus-listed journals.
A major concern with data from social media and other online platforms is their quality and representativeness (Correia et al., Reference Correia, Ladle, Jarić, Malhado, Mittermeier, Roll, Soriano-Redondo, Veríssimo, Fink, Hausmann, Guedes-Santos, Vardi and Di Minin2021). It is possible, for instance, that reviews on TripAdvisor are commonly posted by a younger audience (Carter, Reference Carter2016). Another presumptive problem with TripAdvisor is the number of fake reviews. A study of hotels based on TripAdvisor reviews shows that the proportion of fake reviews is around 20 per cent (Schuckert et al., Reference Schuckert, Liu and Law2016). With respect to this, TripAdvisor is making significant efforts to remove fake reviews (Lee et al., Reference Lee, Song, Li, Lee and Yang2022). However, fake reviews are less likely for natural WHSs since the management cannot be expected to have an interest in them. Even if there should be exaggerated reviews for natural WHSs, this is a lesser problem for the present study since the focus is on the number of reviews rather than on the ratings.
Besides the possible selection bias arising from which individuals may use different social media or online platforms, is the coverage and representativeness across countries and continents. The main concern in this respect is China, where other platforms are used (see Wang et al. (Reference Wang, Liu, Sun and Zhao2024) based on Sina Weibo or Gan et al. (Reference Gan, Liao, Kang, Xu, Fu, Cao and Lan2024) based on the online travel platform Ctrip). Ctrip is regarded as one of the largest online tourism service providers in China (Gan et al., Reference Gan, Liao, Kang, Xu, Fu, Cao and Lan2024), while access to TripAdvisor is difficult for domestic travellers. Yet, TripAdvisor is likely to represent the foreign visitors to China.
2.3 Determinants of digital measures of interest in natural World Heritage sites
Regardless of the data source or measure, it is particularly important not to interpret the interest in the natural WHSs until their characteristics and location are controlled for. Thus, included in this group of important factors is the year of inscription, since well-known sites attract larger numbers of visitors and are possibly more at risk (Correia et al., Reference Correia, Jepson, Malhado and Ladle2018; Bragagnolo et al., Reference Bragagnolo, Correia, Gamarra, Lessa, Jepson, Malhado and Ladle2021; Guedes-Santos et al., Reference Guedes-Santos, Correia, Jepson and Ladle2021; Falk and Hagsten, Reference Falk and Hagsten2022). These sites may also have more natural attractions and exceptional characteristics as early inscribed obvious candidates (Bragagnolo et al., Reference Bragagnolo, Correia, Gamarra, Lessa, Jepson, Malhado and Ladle2021; Guedes-Santos et al., Reference Guedes-Santos, Correia, Jepson and Ladle2021).
Location of the site is an important factor relating both to the number of TripAdvisor reviews and the identified threat from tourism. Where the site is located may indicate if there are decent policies and tools available to address presumptive threats. Yet another variable of importance is the size of the site (Falk and Hagsten, Reference Falk and Hagsten2022, Reference Falk and Hagsten2023). Natural WHSs vary markedly in size, and some of them stretch over several countries. Larger sites can host more visitors and might have several entrance points. If the number of visitors is set in relation to the size of the protected area or the number of residents, this can be interpreted as a measure of an acceptable load limit for the site (McCool and Lime, Reference McCool and Lime2001). Several studies expand the definition of protected area resilience in terms of acceptance of impacts on natural resources and people, measured by selected biophysical resources and social conditions rather than visitor numbers, although the present study does not stretch so far (Prato, Reference Prato2001; Salerno et al., Reference Salerno, Viviano, Manfredi, Caroli, Thakuri and Tartari2013).
Kind of site is another characteristic that potentially relates to the interest of visitors. Certain sites, such as glaciers, are difficult to access while others like religious and sacred sites are not. The analyses are based on the assumption that the interest in, as well as the threats to, the natural WHSs is related to similar variables, although not necessarily to the same degree.
Comprehensive econometric analyses that include the characteristics of the site in the digital measure are rare. Exceptions include Guedes-Santos et al. (Reference Guedes-Santos, Correia, Jepson and Ladle2021), Falk and Hagsten (Reference Falk and Hagsten2022) and Hausmann et al. (Reference Hausmann, Toivonen, Fink, Heikinheimo, Tenkanen, Butchart, Brooks and Di Minin2019) who use count data models and quantile regression methods to study factors influencing the interest in WHSs or other protected areas based on online data.
3. Empirical approach
The specification is inspired by previous studies (Hausmann et al., Reference Hausmann, Toivonen, Fink, Heikinheimo, Tenkanen, Butchart, Brooks and Di Minin2019; Guedes-Santos et al., Reference Guedes-Santos, Correia, Jepson and Ladle2021; Falk and Hagsten, Reference Falk and Hagsten2022) and models the interest in natural WHSs and the number of tourism threats since inscription as a function of site characteristics and location:
Subscript i = 1,…,264 denotes the natural (or mixed) WHSs, ln() is the natural logarithm, ${\varepsilon _i}$
reflects the error term and ${\beta _0}$
is the constant. The dependent variable Y is either measured as the number of TripAdvisor reviews of the natural WHS, including main attractions (points of interest & landmarks, nature & wildlife areas, geologic formations, volcanos, national parks, caverns & caves, bodies of water, hiking trails, visitor centres) or as the number of threats from tourism, visitation or recreation since inscription. Size represents the area of the site measured in hectares and $Year\_inscription$
indicates the year of inscription. The variable Mixed is a dummy for mixed cultural and natural sites and Countrygroup offers a set of dummy variables for country-specific factors such as different levels of economic development (Africa, Arab states, Australia & New Zealand, China, North America, Other Asia and Latin America with Europe as the reference category). There is also a dummy variable specifying natural WHSs that cross national borders, Transboundary.
Eight per cent of the observations lack TripAdvisor reviews, implying that the dataset has a typical Poisson distribution. In such cases, the OLS estimator is not appropriate. Instead, the Poisson regression model is more suitable (Cameron and Trivedi, Reference Cameron and Trivedi2013). The second dependent variable, the number of tourism threats, on the other hand, is a non-negative count variable. This means that a standard count data model can be applied. Both the Poisson and negative binomial models are employed whereas the former is a special case of the more general negative binomial regression model (Cameron and Trivedi, Reference Cameron and Trivedi2013).
The second step of the analysis models the intensity of threats using a random effects ordered probit model (Greene, Reference Greene2018). In this case, the underlying dependent variable takes on four categories: (0) No threat, (1) Very low threat, (2) Low threat and (3) High or Very High threat. The latter two categories are combined into one class because of the small number of cases in the highest category. With the same explanatory variables as in equation (1), the specification can be written as follows:
Subscript i denotes the site and t = 2014, 2017 and 2020. The dependent variable, $THR_{it}^\ast$
, reflects the likelihood of falling into one of the threat intensity classes in the current period, ${X_i}$
is the (1xk) vector of independent variables described in connection with equation (1) and ${\theta _i}$
is the (kx1) vector of coefficients for the same variables. The random effects are represented by ${v_i}$
, the time effects by ${\gamma _t}$
, and the error term by ${\varepsilon _{it}}$
. Subsequently, the observed threat intensity is defined as follows:
where ${\mu _i}$
are the cutoff points to be estimated. The random effects ordered probit model is estimated using the maximum likelihood and the Gauss-Hermite quadrature algorithm (Roodman, Reference Roodman2011). Based on the estimates, the marginal effects for the four categories can be calculated. Standard errors are clustered at the site level to control for common factors within each site over time.
4. Data and stylised facts
4.1 Data sources
Data for the analyses originates from four sources: TripAdvisor, the UNESCO State of Conservation database, the IUCN Conservation Outlook Assessment database, and the UNESCO World Heritage list. The number of TripAdvisor reviews was collected manually from the TripAdvisor homepage in February 2024. Both English and national language names of the sites are considered. Sites in Latin America are often named in Spanish or Portuguese. Large natural world heritage parks such as the Dolomites or the Canadian Rocky Mountains National Park consist of several sub-areas, which have been identified on the basis of the map material on the UNESCO website and are all included. The number of TripAdvisor reviews for each location is made up of the sum of the reviews of various attractions within each site, where available. The category ‘things to do’ and the setting ‘places sorted by traveller favourites’ are used for an identification of this. In some cases, information for up to 30 scenic attractions or locations within the area of the site is used. The attractions identified on TripAdvisor include:
• ancient ruins, speciality museums, visitor centres
• beaches, islands
• bodies of water, hot springs & geysers, waterfalls
• caverns & caves, geologic formations
• forests, gardens, hiking trails, lookouts, mountains, nature & wildlife areas, points of interest & landmarks, scenic drives, scenic walking areas, valleys.
Accommodations, restaurants, shops and tour operators are excluded as this would otherwise lead to double counting. The data collection method results in 1.0 million TripAdvisor reviews for 264 natural and mixed WHSs.
There are 21 natural WHSs that are not listed on TripAdvisor. This group encompasses mainly remote and uninhabited islands such as Lagoons of New Caledonia, Henderson Island, Heard and McDonald Islands, New Zealand Sub-Antarctic Islands, French Austral Lands and Seas, or other natural WHSs which are closed to visitors. The latter include Río Abiseo National Park in Peru, where public visits are strictly limited and controlled due to the sensitivity of the protected area (UNESCO World Heritage Center, 2024). A group of areas is not included in Tripadvisor (Aldabra Atoll, St Kilda, Surtsey, Gough and Inaccessible islands, Macquarie Island and Papahānaumokuākea), all of which only allow seasonal visits by scientists (Thorsell and Sigaty, Reference Thorsell and Sigaty2001). Some parks such as the Chiribiquete National Park – ‘The Maloca of the Jaguar’ – do not allow visitors because they are a potential threat to the indigenous people (IUCN, 2020). For one particular natural WHS, the number of TripAdvisor reviews could not be calculated (Ancient and Primeval Beech Forests of the Carpathians and Other Region) since it encompasses 93 locations that are difficult to track.
The number of TripAdvisor reviews is compared with the incidence of tourism threats based on two official databases. The first is the IUCN Conservation Outlook Assessment database that provides information on the degree of threats. This database encompasses information on threats for the full universe of natural and mixed WHSs as assessed by their experts (IUCN Conservation Outlook Assessment, n.d.; Osipova et al., Reference Osipova, Emslie-Smith, Osti, Murai, Åberg and Shadie2020). For the purposes of this study, information on the assessment of threats in the current period is used for three available points in time: 2014, 2017 and 2020. The severity of threats appears on a five-grade scale: no threat, very low threat, low threat, high threat, and very high threat. In addition, there is the category of ‘data deficient’. However, this is not regarded as a major problem as only eight respondents report a lack of information on tourism threats. In such cases, information on the most severe threat is kept for the analysis. In the Conservation Outlook Assessment database, there is information on kinds of tourism threats. These threats are mainly related to a large number of visitors, given the size of the site, or to the related infrastructure (table 3, upper panel).
Tourism threats to natural WHSs based on the State of Conservation and Conservation Outlook Assessment databases (selected sites)
Notes: Conservation Outlook Assessment: selected sites with information on kind of tourism threat for the period 2014–2020. State of Conservation database: selected sites with information on kind of tourism threat based over the years 1979–2023.
Source: Conservation Outlook Assessment, State of Conservation database.
The other official database on threats employed is the UNESCO State of Conservation, which contains information on tourism threats since inscription of the site, starting in 1979. In this case, information on threats is only available in a binary format. As tourist threats are a rare event, accounting for 2 per cent of the 6,800 annual observations, the cumulative number of tourist threats since inscription is used. Tourism threats to natural and mixed WHSs are commonly related to the number of visitors (table 3, lower panel). Therefore, data on the popularity of the site, based on the number of TripAdvisor reviews, could serve as an alternative indicator of the presumptive tourism pressure.
The final source employed for the analysis is the World Heritage database (UNESCO World Heritage List), which contains information on site, year of inscription, presence of a mixed or full natural site, location and whether the site extends beyond the national boundaries. Kind of site (marine & coastal, forests, religious & sacred, mountains, glaciers, rivers and islands) is sourced directly from the description of the WHSs by a text mining exercise.
4.2 Descriptive statistics
The average accumulated number of TripAdvisor reviews by site is 3,733 in February 2024 (table A2, online appendix). Each site is threatened by tourism on average 1.4 times since inscription while the degree of tourism threat (measured categorically from 0 to 3 for the period 2014 to 2020) is 1.5. In 28 per cent of the cases, the threat is considered high or very high. The severity of threats from tourism is slightly declining in 2020.
Among the natural WHSs, the Canadian Rocky Mountain Parks, the Galápagos Islands and the Iguaçu National Park receive the highest numbers of reviews. Based on the classification in the Conservation Outlook Assessment database, 12 of the 25 most popular sites are critically endangered by tourism or recreation pressures during the period 2014 to 2020 and one is critically threatened (table 4). The Galápagos Islands is identified as threatened by tourism 24 times since inscription based on the State of Conservation database, while the Iguaçu National Park and the Historic Sanctuary of Machu Picchu encounter 16 threats each.
Number of TripAdvisor reviews and tourism threats (top 25)
Notes: CAD denotes Canadian dollars, USD US dollars, EUR EURO and AUD Australian dollars. TPA denotes Taxa De Preservação Ambiental (Environmental Preservation Tax).
Source: TripAdvisor, State of Conservation, Conservation Outlook Assessment and website of the natural WHSs.
When adjusted by size of the site, the ranking based on TripAdvisor changes. This means that the relatively most reviewed sites are the Giant's Causeway and the Causeway Coast (6,669 TripAdvisor reviews per km2), the Hierapolis-Pamukkale (1,449 TripAdvisor reviews per km2) as well as the Göreme National Park and the Rock Sites of Cappadocia. Sites with the relatively highest number of TripAdvisor reviews are generally the smallest ones, with an area of less than 3 km2 (table A1, online appendix). This includes Vallée de Mai Nature Reserve (15,626 TripAdvisor reviews per km2) and Meteora (2,051 TripAdvisor reviews per km2). There are also a certain group of islands with many reviews given their size (Isole Eolie – Aeolian Islands) with 762 TripAdvisor reviews per km2.
Sites with the highest number of reviews coincide well with the analysis of geo-tagged Flickr photos by Levin et al. (Reference Levin, Kark and Crandall2015), who find that Yosemite National Park, the Grand Canyon, and Yellowstone National Park belong to the most photographed protected areas. According to Thorsell and Sigaty (Reference Thorsell and Sigaty2001), the Canadian Rocky Mountain Parks, Grand Canyon National Park, Yosemite National Park and Great Smokey National Park are also among the most visited natural Heritage sites with between 4 and 9 million visitors per year. While two of the most threatened sites (Galápagos Islands and Machu Picchu) have high entrance fees, there are other threatened sites with low (Ha Long Bay – Cat Ba Archipelago) or no (Giant's Causeway and Causeway Coast) fees (table 4).
Natural WHSs in North America, Australia and New Zealand as well as in Europe experience the highest level of interest and those in the Arab region the lowest (figure 1; figure A3, online appendix). In contrast, the cumulated number of tourism threats is the highest in the Arab region and Asia (excluding China), with two threats on average (figure 2; figure A4, online appendix). When the Conservation Outlook database is used, sites in Europe (Italy, Portugal, Croatia), Latin America (Argentina) and Asia (Nepal) have a higher degree of threats from tourism (figure A5, online appendix).
Distribution of average number of TripAdvisor reviews by region.
Source: TripAdvisor, UNESCO and own calculations.
Distribution of average number of tourism threats by region.
Source: State of Conservation and own calculations.
There is also a large variation by kind of site. Religious sites and those with sacred characteristics exhibit the largest number of reviews, while the number of tourism threats is highest for forests (including rainforests) as well as marine and coastal sites (figures A1 and A2, online appendix).
5. Empirical results
Estimations show that the digitally sourced measure of interest in natural WHSs only coincides to a small degree with the institutional identification of threats (tables 5–7). Aspects of importance are year of inscription, certain kinds of sites, and continents, although they appear with contradictory signs. Sites in North America, for instance, with the largest number of TripAdvisor reviews given size and year of inscription, have among the smallest numbers of and less severe threats.
Factors of importance for number of TripAdvisor reviews (Poisson estimation)
Notes: Standard errors are in parentheses. The dependent variable is the number of TripAdvisor reviews for the year 2024. Estimates are based on the Poisson model with the maximum likelihood technique. The STATA command poisson is used. Standard errors are adjusted for 108 clusters at the country level. $dy/dx$
denotes the marginal effects.
Source: TripAdvisor, UNESCO and own calculations.
Factors of importance for the number of threats from tourism (count data model estimations)
Notes: The dependent variable is the accumulated number of tourism threats. Estimates are based on the Poisson regression and negative binomial regression model. The STATA commands Poisson and nbreg are used. Standard errors are adjusted for 109 clusters at the country level. The Likelihood Ratio test of alpha equal to 0 cannot be rejected at the 1 per cent level, indicating that the negative binomial regression model is more appropriate than the Poisson model.
Source: State of Conservation, UNESCO and own calculations.
Factors of importance for the degree of threats from tourism (ordered probit estimates with random effects, marginal effects)
Notes: The underlying dependent variable is the degree of tourism threats. Estimates are based on a random effects ordered probit model (see table A3, online appendix). The STATA command cmp with the option || id is used. Naylor-Smith adaptive quadrature is used as an algorithm and the number of draws is 101. Standard errors are adjusted for 109 clusters at the country level. The reference group for kind of sites is other sites.
Source: Conservation Outlook Assessment, UNESCO and own calculations.
The Poisson estimations also demonstrate that location is the most important variable for the number of TripAdvisor reviews, although year of inscription and, to a lesser extent, kind of site are also significant (table 5). Natural WHSs in North America and Europe are most commonly reviewed on TripAdvisor, while the Arab region, China and Australia receive far less attention in this context. In 2024, sites in the Arab region received 8,790 fewer TripAdvisor reviews on average than the reference category Europe (given size and year of inscription and other control variables). In contrast, the sites in North America received 4,143 more TripAdvisor reviews than the reference category Europe (table 5).
The year of inscription on the UNESCO list is also a significant determining factor for the number of TripAdvisor online reviews (p-value < 0.01). A natural WHS which is one year older receives on average 159 more TripAdvisor reviews, given its characteristics and location. Kind of sites is only partly significant. Rivers receive much less attraction on TripAdvisor while there is little variation across religious and sacred sites, mountains, glaciers and islands. Size of the area is not significant at conventional levels. Sites which stretch over several borders have a lower number of reviews than those within a single country. The marginal effect of the transboundary coefficient is −4,273 indicating that these sites receive on average 4,273 fewer TripAdvisor reviews given their characteristics and location compared to single country sites.
While natural WHSs in North America have the highest number of TripAdvisor reviews, they are the least threatened ones, based on information in the UNESCO State of Conservation database (table 6). Results of the negative binomial regression model show that the expected count of a tourism threat for the locations in Australia and New Zealand as well as in North America is less than one tenth and one fifth, respectively, compared with the reference category Europe (calculated as exp(−2.59) = 0.07 and exp(−1.49) = 0.23). This means that the sites in these countries have a threat count of 0.11 and 0.34, respectively, as compared to those in Europe with 1.5 on average (cumulated since inscription).
As indicated by the Wald test, there is limited variation in the number of tourism threats across kinds of sites. Year of inscription as a UNESCO site coincides with the TripAdvisor variable in that it renders highly significant results. A natural WHS one year younger as inscribed is expected to have an 8 or 9 per cent lower number of tourism threats. Just like for the results of the TripAdvisor reviews estimations, neither the size of the site nor its category (mixed) is relevant (table 6).
Results based on the random effects ordered probit model for the years 2014, 2017 and 2020 demonstrate that the severity of the threat from tourism depends significantly on size, year of inscription, kind, location and whether it is a mixed or solely natural WHS (table 7). The extent of the threat is largest for religious and sacred sites, glaciers and mountains. In contrast, the degree of threat is lowest for natural WHSs characterised by forests or rivers. Location is highly significant with sites in North America having the lowest level of threats identified. Sites in Africa and the Arab region also have significantly lower levels of threats than those in the reference category Europe. Larger sites have a higher probability of very low or no threats, while they are lower for mixed sites.
With respect to magnitude, the largest marginal effects can be found for religious and sacred sites as well as forests. The former sites have an 18 percentage points larger probability of being categorised with a high or very high threat in relation to other sites, while forests have a 17-percentage point lower probability.
Glaciers have an 8-percentage point larger probability of being identified as the most severe threat. Across locations, sites in North America are assessed with a lower degree of high or very high threats (8 percentage points lower probability than for sites in Europe).
6. Discussion
As specified in the introductory section, the aim of this study is to explore if threats to natural WHSs can be identified based on data from an online travel platform. By doing so, this digitally sourced measure of interest in natural and mixed WHSs is contrasted with the number and severity of threats as formally identified by UNESCO (State of Conservation database) and IUCN (Conservation Outlook Assessment) when typical site characteristics are accounted for. An important reason for looking at alternative data sources for this kind of analysis is the unbounded characteristics of natural WHSs; that is, they are commonly either not fenced off or are simply too large for simple monitoring. This also implies that threats from tourism identified by use of formal data sources may be overly generic and do not necessarily take sub-sections or sub-areas into account, which, for instance, the online platform TripAdvisor does.
Estimation results indicate that the TripAdvisor variable and the number of threats based on the UNESCO State of Conservation database clearly coincide in what kinds of variables are significant. These are mainly years since inscription, and to a certain extent, kind of site as well as continent. However, there are also contradictions, where, for instance, sites on the North American continent receive a higher number of reviews even when their characteristics are controlled for, while the opposite result is achieved based on the number of threats. This means, on the one hand, that a large number of visitors (Tripadvisor reviews) does not automatically pose a threat to a site. On the other hand, it also indicates that TripAdvisor cannot exactly identify pure threats to a site, although it can well be used to identify which sites are the most popular given their characteristics. The discrepancy may also appear because sites in the United States are used to many visitors and thus also have successful management to deal with this.
In the case when TripAdvisor reviews are adjusted for the size of the site, smaller sites such as the Isole Eolie (Aeolian Islands), Hierapolis-Pamukkale, Meteora, Giant's Causeway and Causeway Coast, Vallée de Mai Nature Reserve are rated relatively most attractive. This stands in contrast to the absolute number where the hotspots are Canadian Rocky Mountain Parks, Galápagos Islands and Iguaçu National Park. Mitigation measures like the distribution of the visitors within the national park are thus only possible for larger sites. Smaller sites may need stricter management.
An interesting additional layer of information to the study appears when the variable for the degree of threats is analysed. Most characteristics are of importance, although with less emphasis on continent. Presumably, this reveals that continent may lay behind the risk for a threat, but once it appears, the severity of it is related to the local context, which could of course include the ability and resources of the management.
As can be noticed, the TripAdvisor data have advantages over the threat data from the UNESCO sources. First, they are available in a timely manner and provide a quantitative measure of the number of reviews which can easily be related to the size of the site or other indicators, while the official threat indicators are based on dummy or categorical variables which do not have a quantitative interpretation. This source also provides more detailed information about the site, given that it commonly includes the different sub-areas. These areas are not identified by UNESCO, making the official data somewhat imprecise in that, for instance, the natural WHS of the Dolomites in practice consists of more than ten separate parks.
Results from this analysis are difficult to benchmark against previous literature since very few studies explicitly compare threats from visitors with digitally sourced indicators of visitation. Nevertheless, the importance of experience as inscribed is consistent with Guedes-Santos et al. (Reference Guedes-Santos, Correia, Jepson and Ladle2021), who report similar relationships for protected areas and with Falk and Hagsten (Reference Falk and Hagsten2022), solely based on alternative digital measures of interest in natural WHSs (number of Instagram posts and Wikipedia page views). There is also earlier evidence that sites in North America receive the largest number of ratings of interest when different social media or digital measures are used (Falk and Hagsten, Reference Falk and Hagsten2022).
Yang et al. (Reference Yang, Zhang and Fu2022) conclude that even if TripAdvisor data may not represent all domestic visitors in China, it is a reliable source for international guests. Thus, this means that the results for China are only representative if the domestic tourists follow the same behaviour patterns as the international ones, something that is presently unknown. Meanwhile, the use of the TripAdvisor app is banned in China (Gillette and Boyd, Reference Gillette and Boyd2024) but allowed for foreigners with a subscription to a foreign internet provider.
Methodologically, the study shows the importance of carefully choosing an estimation method that fits the characteristics of the dataset. Some sites do not have visitors at all and thus a linear regression model is not suitable. Tourism threat indicators on the other hand are available in the form of count data, dummy variables or ordered categorical variables. Another methodological finding is that the measurement of a tourism threat based on a dummy variable (0/1) is too simplistic. The attempt of IUCN to scale the threat based on several categories, for instance, is thus more nuanced. The periodic report of UNESCO also contains a categorisation of degree of threat, but is only available every ten years.
Although the threat and review estimations give similar but not identical results, the overall insight is that multiple indicators should be used and contrasted with each other as well as with management strategies to reach a full picture of the vulnerability of a natural WHS.
7. Conclusion
In this study, a novel measure of interest in natural or mixed WHSs sourced from an online platform is contrasted with the degree and number of threats as formally identified by the UNESCO (State of Conservation) and IUCN (Conservation Outlook Assessment), when typical site characteristics are accounted for. TripAdvisor reviews are used as the digital measure while the site characteristics of importance originate from the UNESCO World Heritage database, including size, year of inscription, kind (marine & coastal, forests, religious & sacred, mountains, glaciers, rivers and islands) as well as mixed or entirely natural.
Empirical results based on the random effects ordered probit as well as count data models on the linked datasets reveal that a large interest in a site does not necessarily follow the pattern of threats by visitors. North America has the largest number of TripAdvisor reviews, while the sites on this continent are less often assessed to have many or severe threats.
Thus, there is a clear distinction in the significance and importance of the control variables for the three measures. Location is an important aspect of both the number of TripAdvisor reviews and the number of threats. The highest number of TripAdvisor reviews can be observed for natural WHSs in North America and Europe. This may be partly explained by the absence of alternative platforms in use in these regions, as compared with China, for instance.
Another variable that stands out is the year of inscription as a UNESCO natural WHS. One year older as inscribed leads to another 160 TripAdvisor reviews on average, given characteristics such as size and location. While there are no major differences in the number of TripAdvisor reviews across kinds of sites, the Conservation Outlook Assessment data reveal that religious sites and glaciers are the most threatened ones. As opposed to the number of reviews and number of threats estimations, the results based on the severity of threats demonstrate that most characteristics of a site are of importance for this aspect, except possibly the variable continent.
Theoretically, the study highlights the importance of using more than one indicator and one estimator for analyses of tourism flows or threats to natural WHSs. It is also of importance not to interpret the indicators in isolation without controlling for important aspects such as year of inscription, kind and location. The present study is the first to use the number of TripAdvisor reviews as a measure of visitation to natural WHSs. This gives more detailed information about the sites than any of the formal datasets presently available.
Several implications can be drawn based on the findings. Firstly, the number of TripAdvisor reviews of a site, given its characteristics, is a variable that clearly indicates its popularity, although this does not automatically mean that it is also threatened, as can be seen from the results for North America. Thus, these kinds of popularity measures need to be benchmarked against threats controlled for similar aspects. The Conservation Outlook Assessment database is very useful because of its detailedness, but it is unfortunately no longer updated. It is imperative that it be re-opened.
Several limitations should be noted. First, the analysis of TripAdvisor reviews is cross-sectional. Future research work should study the variation over time using panel data models. Another limitation is that the TripAdvisor variable most likely is only representative of international visitors to China. Data from online travel platforms tend to suffer from a selection bias to some degree. Investigating the degree of selection bias could be a topic for future study. Additional ideas for future work include new explanatory variables, such as accessibility and distance to major population centres, as well as additional site characteristics. Yet another idea is to study the management plans and policy measures such as entrance fees and visitor restrictions to combat identified threats. Many sites employ various measures to manage visitor flows, such as pricing, restrictions, monitoring and limits.
Supplementary material
The supplementary material for this article can be found at https://doi.org/10.1017/S1355770X25000142.
Competing interest
None.
Open access
