The results provide a global overview of CES types in PAs, highlighting their sheer diversity across countries. Unexpectedly, a large proportion of photos in the dataset appeared to describe activities of PAs in mountainous regions. This supports previous findings that PAs are generally “high and far”, which means that they tend to be further from urban regions and roads due to their higher altitude38. However, contrary to the expectation that these “high and far” PAs are inaccessible, we identified such PAs to instead be rich in their provision of CES types, as noted by the many photos of people engaging in alpine sports.
We also observe other distinct trends in CES types among spatially related PAs. There is a greater proportion of photos with Biotic CES type in equatorial countries, with the opposite trend being observed at temperate countries. These trends align well with expected traits of physical environments in PAs. For instance, the many sites of important historical value being located within PAs in Europe could help explain the finding of a larger proportion of photos with Humans CES type being detected. In addition, the location of most of the biodiversity hotspots in the world being in equatorial countries39 correlates well with the finding of higher proportions of photos with Biotic CES type in these regions.
Although PAs vary widely in their physical environments, we find that these differences may have a smaller than expected influence on the dominant CES type in each PA. Instead, PAs within each country tend to have a high degree of overlap in CES types, even if they differ in biome types. Such observations were found from the mixed-effects models using biome type as fixed effects and country of PAs as random effects to predict the CES types in PAs. As seen in Supplementary Table S1, the conditional R2 was much larger than the marginal R2 for all three CES types, thereby indicating that country-level differences are more important than the biome type of PAs in explaining variation in CES types across PAs. Such findings resemble a previous study that aimed to identify CES activities from social media data among countries in Europe by van Zanten et al.40. In it, they found that country differences were able to explain more variance in CES as compared to geographical predictors, such as accessibility and terrain.
Such observations could be explained by the process with which PAs get designated. Different land use policies for all PAs within each political boundaries for ease of administration occur. Subsequently, this then results in country-specific land use management guidelines having a strong influence on the provision of available CES types in PAs41,42. Such differing guidelines across countries can then cause PAs in different countries that share similar biome types to diverge in their CES types. For instance, due to Norway having a more liberal conservation policy than Canada, this meant that a wider range of human activities (such as camping and fishing) were allowed in Norway, as opposed to in Canada43. As such, it is important to also consider the potential contribution of its country’s land use management guidelines when attempting to explain the existence of certain CES in a PA. Another potential explanation is the different preferences and values towards the use of PAs held across countries. For instance, in a public ecosystem services mapping survey, Norwegians mapped more values related to hunting/fishing and gathering than Polish respondents focusing on scenery, biodiversity and water quality in the ecosystem services they attribute to protected places42. Such differences in CES can also vary depending on the cultural values ascribed to nature by different countries. For instance, although Chinese visitors appear to share many similarities with Western visitors in the CES they attribute to protected places, they appear to differ in their expectations with regards to wildlife interactions and birdwatching44. Consequently, it is thus pertinent to additionally consider the cultural component of CES and how they differ across countries.
Additionally, another notable observation is that most photos taken in PAs tend to be from IUCN Category II (National Park), with the number of photos in PAs of IUCN Categories Ia (Strict Nature Reserve) and Ib (Wilderness Area) being comparatively fewer (Fig. 3). As expected, the PA categories which have stricter protections against human intervention also face less footfall from visitors due to tighter restrictions on their entry. However, there is still a notable number of photos found in these stricter IUCN PA categories, thereby indicating that the human footprint in these areas may be higher than expected. Similar findings were observed by other studies of global PAs45,46. Nonetheless, Leberger et al.47 noted that the stricter IUCN categories still experienced lower forest loss than other categories, despite the higher than expected human pressure exerted on them. Beyond this observation, another finding from our dataset is that there is a significantly large number of photos from PAs with unclear IUCN designation, although not having an IUCN classification, these PAs have been designated as official PAs by their respective countries’ governments and demonstrate an important role in the provision of cultural ES. As cautioned by others48,49, some of the IUCN categories, especially V (Protected Landscape or Seascape) and VI (Protected Area with Sustainable Use of Natural Resources), tend to be ambiguous in their definitions, and also tend to be less strictly monitored on their condition. Such constraints subsequently limit our capacity to elucidate the patters of activities performed in these categories.
More prominently, in this study, the major CES activities observed in PAs tended to align with the intended activities that the countries are using to attract visitors to those PAs. For instance, both the Komodo National Park in Indonesia and the Moremi Game Reserve in Botswana have high proportions of photos of Biotic CES type taken in them. This supports previous findings where activities identified on social media were found to agree strongly with the actual activity preferences of visitors to PA as previously surveyed50,51. For example, Hausmann50 noted that both the surveys and social media data collected within Kruger National Park in South Africa identified that the most favoured subject in photos taken by visitors are the charismatic, large-bodied mammals present within the national park.
Consequently, such high levels of concordance observed between movement patterns of individuals on social media and physical surveys52 underscores the possibility of social media as an alternative tool for local governments for monitoring trends in human activities in PAs. By tracking dominant CES in PAs, countries can hence uncover the precise activities undertaken in those areas. Among the different activities that could be undertaken in PAs, there are some which are more damaging to the environment, such as mountain biking instead of hiking53. As such, if there are any notable changes in CES types towards those with greater environmental impact, constant monitoring of social media posts would enable countries to react more rapidly to such marked increases in land use degradation by enacting measures such as targeted temporary closure of those areas. Social media mining thus has utility for guiding country-level land-use management to ensure that PAs remain conserved.
There were three main categories of CES observed in this study: Biotic, Abiotic, and Humans. Unlike the CES categories proposed by CICES (Supplementary Tables S2 and S3), the addition of Humans CES type in the proposed definitions in this paper appears to have been previously overlooked. To incorporate this proposed third dimension of CES, we suggest that CES definitions could be reframed by considering the relational values that people receive from their interactions with nature14. Such a modification to Humans CES categories is also supported by studies of CES in PAs. Egarter Vigl et al.11, using a survey of photos by visitors to The Dolomites in Italy, identified four main categories of CES: “aesthetic value, outdoor recreation, cultural heritage, symbolic species”. The “outdoor recreation” and “cultural heritage” are closely linked to Humans CES instead. Similar findings were also present in Cardoso et al.10, where an analysis of social media posts revealed that users value “aesthetics appreciation” in Peneda-Gerês in Portugal, while they value “cultural heritage and spiritual enrichment” in Sierra Nevada in Spain.
Another finding is that CES definitions made from objective descriptions of visible objects in photos are more relevant for CES identification from social media. Such an incongruity between proposed and observable activities was also present in a similar study by Richards and Tunçer27. They found that photos could only be classified according to their dominant object, which fell into the categories of “Transport, Plants, Animals, Food, People, Sports, Landscape and miscellaneous”. This limited applicability of CES definitions was also recognized as one limitation of the current CICES framework35. However, making CES definitions more concrete potentially deemphasizes the emotional and cognitive evaluations that attract people to nature and misses the underlying motivations for humans to take photos in nature; there is thus a danger of losing important information about CES types27. To circumvent this subjectivity in photo interpretation, other forms of information could also be concurrently collected. By using a hybrid method of combining both photo metadata and text data, estimation of the intent of people who visit a protected place may be more accurate54.
Our study had several limitations. Firstly, the most notable limitation is that the entire dataset was sourced from Flickr alone, with no corresponding fieldwork studies done to corroborate any findings. This consequently has implications on the ability of the users in the sampled dataset to be a robust representation of their respective country’s populations. In general, users of Flickr tend to be skewed towards a certain kind of demographic, that being younger than the average visitor age to PAs27. Additionally, as the photos of animals uploaded onto social media platforms tend to be taken by specialized equipment, there is also a gap in availability of the necessary digital devices required to obtain good quality photos, with such infrastructure being less available in developing countries55. As such, there is need to correct for such biases in the data collected before generalizing any findings from social media towards actual visitor habits in PAs. One such possible method could be to consider a twin approach of combining both big data analytical methods with conventional fieldwork approaches. Such fieldwork can involve conducting interviews with people visiting the PAs to validate the CES activities identified from social media56.
Secondly, in our Flickr dataset, although there are a high number of photos available for developed, English speaking, countries, the number of photos in less developed and/or non-English speaking countries is much fewer. Most prominently, there are a number of countries with low number of photos counts (Supplementary Figs. S14, S16 and S17). For instance, there were only 85 photos found in PAs located within Barbados, in comparison to the countries with the highest (Nicaragua, 1076 photos) and lowest (Anguilla, 9 photos) photo counts in North America (M = 380, SD = 286). This glaring disparity in available photos across countries on Flickr highlights the shortcoming of using social media for sourcing photos: there is unequal international coverage. As highlighted in Supplementary Fig. S3, there appear to be distinct clusters of countries with significantly more photos than other regions in the world. For instance, European countries (M = 591, SD = 376) such as France have disproportionately more photos (1579 photos) than African countries (M = 262, SD = 336) such as the Gambia (118 photos). Additionally, as the background profile of the users may not be clear at times, it is uncertain as to whether most visitors are indeed representative of the local population, or that they tend to be overseas visitors making short trips abroad; this is particularly pertinent when considering the photos on social media taken in developing countries57. To better capture the activities of the local population, future studies may hence wish to consider using more than one social media platform, with a particular emphasis on regional social media sites. For instance, the recent study by Wang et al.58 mined social media posts on Weibo in China to uncover CES in Xiamen, which would give a more accurate insight into the activities of locals.
Lastly, another limitation of this study is that we chose the number of clusters and matched them with the CICES classification manually. Despite this approach, most of the clusters successfully captured common activities that could describe the labeled objects in each cluster. In addition, the resolution of the CES types identified in PAs through machine learning and social media is sufficiently high in identifying dominant CES types between neighboring PAs. These findings are supported by past studies9. As such, with social media and machine learning having utility in sampling from a wider spatial range without losing much accuracy18, this combination is viable for addressing existing gaps in manually collected data.
In conclusion, we observed clear trends in CES types in PAs across countries and continents. While the Abiotic CES type dominates most PAs worldwide, there are distinct clusters for other CES types, such as equatorial countries for the Biotic CES types, and European countries for the Humans CES type. This study demonstrates that pairing social media with machine learning is a viable approach for large scale socio-ecological data analyses. Additionally, despite their diversity of physical environments, PAs within each country were found to have similar CES types. These CES types were also found to match similar nature-based recreational activities as marketed by these countries. Ultimately, with such concordance in CES types having been uncovered, this study has highlighted the usefulness of social media for monitoring global distributions of land use patterns in PAs.
