This study of recreation in two geographically distinct regions of the United States indicates that social media can predict visitation at recreation sites on public land. We conclude that social media data can be applied with moderate success to estimate visitation at sites that are unmonitored or otherwise lack on-site counts, even in new regions. A basic visitation model that relies solely on generic predictors (e.g., weather, holidays, and seasonality) is only modestly successful due to regional differences in visitor behavior (Model 1). Performance is improved by including relationships between visitation and social media (Flickr, Instagram, and Twitter), even when these relationships are transferred from a different region (WWA, Model 2). These results are consistent with prior research findings that social media counts are correlated with on-site visitor counts from public lands8,9,24,32, and extend earlier findings by showing the potential for statistical models to estimate absolute numbers of visitors at unmonitored sites with parameters derived from social media. This is evidence of patterns in how visitors use and share social media. Furthermore, it suggests that measurable variables associated with social media use could support transferable models for accurately estimating visitation to public lands across large geographies.
We expected to detect regional differences in social media use given regional distinctions in climate, land management, population density, mobile phone signal coverage, and the demographics of visitors. Contrary to our expectations, we observed that the rate of posting to social media about recreation visits is similar across sites in NNM and WWA, and both regions displayed positive correlations between each of these social media data sources and observed visitation (Fig. 3). Furthermore, a model parameterized with social media use in WWA (Model 2) explains 45% of the variation in visitation across all 13 sites in NNM and 79% of the variability in visitation at the subset of sites that had social media posts. Re-parameterizing Model 2 with a portion of NNM visitation data does not improve its performance (Model 3) until these data are considered at the site level (Model 5). There is a noisy but consistent relationship between a destination’s popularity with visitors and its popularity on social media, regardless of whether the site is in NNM or WWA. We interpret this as evidence that visitors are equally likely to share their recreation experiences in NNM or WWA, despite the regional differences in the types of recreation opportunities and the people who are visiting.
Although there are consistent relationships between social media use and visitation at a regional scale, we see large site-to-site variability in how visitors use social media within both regions. Beyond simple differences in numbers of posts by site, the proportion of people who post about their visits varies by destination, ranging from 7% of visitors to Kasha–Katuwe Tent Rocks National Monument posting on Instagram to zero recorded social media user-days at several trails in the Valles Caldera National Preserve. As a result, models with random effects that allow sites to have unique relationships between social media, weather, and visitation (Models 4–5) perform substantially better than models that assume social media use is consistently related to visitation for all types of sites (Models 1–3). This is especially true at the most and least visited sites (Figs. 4, 5), where visitors may be sharing social media differently than they do at moderately visited sites, and responding differently to other conditions such as weather or holidays. These results indicate that while data on social media use are helpful for predicting visitation with moderate certainty in an otherwise unknown region (Model 1 vs. 2), their utility for estimating visitation is less clear when local data on the effects of environmental and institutional conditions such as weather and holidays are available to parameterize site-specific models.
A primary goal of this study is to test approaches for estimating visitation over relatively small areas in order to explore the limits of the data and methods. We find that six of the 13 sites in NNM—representing individual trails or groups of trails within a larger park—lack social media during the study period. Model 2, which depends on social media data to estimate visitation at unmonitored sites, consequently performs relatively poorly at these six sites (Fig. 4). Generally, sites that have sparse social media data tend to receive few visitors, but there are exceptions. Alcove House in Bandelier National Monument, for instance, is a very highly visited site that lacks Instagram images in our study because the site does not appear as a prescribed location for Instagram users. Our informal observation is that many visitors instead share untagged photographs of Alcove House or assign their images to other relevant place names such as Bandelier National Monument. Clearly, there are thresholds to where and how social media can be leveraged for visitor estimation. Our research suggests that future studies and visitation models could be improved by accounting for the popularity of sites on social media in the study design12. Visitation models that include predictors derived from social media (Models 2, 3, and 5, here) will likely out-perform alternative models for estimating use at popular sites or when longer time series are available. At locations with low or no posting activity, where social media contributes less to visitor estimates, it could be more useful to collect on-site data such as vehicle and pedestrian counts. Further research is necessary to understand what combination of on-site, social media, environmental, and other data is most valuable at different spatial and temporal scales.
These observations suggest that variability in correlations between social media and on-site visitor counts seen here and in previous studies8,12,20 is derived from local factors influencing visitors’ day-to-day decisions about whether and how to share a destination on social media. Choices about whether to post to social media are likely influenced by the characteristics of the local site—perhaps its topography, amenities, predominant activity, or unique natural features—and the ways that people relate to these features33. The characteristics of the visitors and the relative contributions of the natural versus the social experience in the motivation for the trip may play a role. For example, if a given type of site attracts visitors wishing to “unplug” and have a nature-based outing (a calm forest glade, say), there may be fewer posts per visit than for a site attracting visitors who desire a social experience within a natural setting (a famous scenic overlook, for instance). Another possibility is that the prevailing popularity of certain destinations on social media creates a positive feedback, whereby new visitors feel compelled to share content about their visit in response to the posts of others or the local hashtags that may make it easier or more enticing to post. Variability could also arise from the recent trend towards discouraging visitors from posting geolocated content and attracting attention to less popular or back-country sites that are not equipped to sustain higher use, although this is probably of minor importance, currently.
This is the first study to our knowledge that develops and tests models for estimating absolute numbers of visitors at unmonitored recreation sites or times using multiple social media data sources with differential effects. Building on earlier research exploring relationships of park visitation with numbers of posts to multiple social media platforms9,12,20,27, the present study tests whether models with a mixture of predictors to represent varying effects of three online platforms can estimate visitation in novel situations. We find that each social media data source contributes information that explains a statistically significant portion of the variability in visitation and improves the accuracy of the estimate. This is the case not only for Instagram, which captures 3–4% of visitor-days at our research sites in WWA and NNM, but also for Flickr and Twitter, with relatively small amounts of content shared (< 1% of visitor-days). Additionally, while the user-days of content shared on each platform is correlated with actual use at our research sites (Fig. 3), posting frequencies for the three social media platforms are only weakly correlated with each other. Perhaps this is because each platform represents different groups of users who are participating in different recreational activities or differ in their propensity for sharing content about certain types of experiences on public lands. This indicates to us that visitation models could be improved by adding predictors derived from social media platforms beyond these three (historically) popular social media platforms that we chose to test. Social media from platforms that are oriented towards specific types of activities or audiences—such as anglers or bicyclists—who may be under-represented on Flickr, Instagram, and Twitter could be especially valuable34,35.
While our understanding of when and how social media are effective for understanding visitor use continues to improve, there are still many unanswered questions and potential issues to consider. The future use of similar approaches will require researchers and practitioners to continually reassess and validate the underlying patterns and assumptions, following best practices for model-building and lessons learned from previous research, such as the failure of Google Flu Trends36. Social media users, and those who choose to share geolocated content from parks, are not a representative sample of visitors in most locations22. Over time, visitors will likely change how they use social media, including their willingness to geotag and share content, and which online platforms or media are vogue. Furthermore, social media platforms are continually being redesigned in ways that affect the type of content that is available (e.g., images versus text), how it is shared or promoted, and whether researchers and users have access to the information. In 2016, for example, Instagram stopped sharing the geocoordinates that are uploaded by its users29. Our study relies entirely on counts of photographs that were assigned by users to prescribed places. These realities must factor into plans for using social media as data now and into the future.
Practical guidance and implications for managers
The quantity, apparent fine spatial resolution, and “live” nature of social media data make it an enticing information source for recreation practitioners, who often desire knowledge of recreation use and visitor characteristics at fine spatial and temporal resolutions. Traditional approaches rely on visitor intercept surveys and physical technology, such as tube or infrared counters. Based on the results described here and in other research11,12,17, we conclude that social media data and related visitation models hold promise for estimating recreation use, especially for large recreation site complexes or collections of recreation resources (e.g., multiple trails within a wilderness area) that are otherwise difficult to monitor. Additionally, social media and other volunteered geographic information provide an opportunity for managers to demonstrate that their recreation monitoring approach is current, practical, and up-to-date with technology.
Previous research has shown that social media data are appropriate for depicting the relative magnitudes of recreation use across a given group of recreation sites. This allows managers to determine when one specific location receives more use than another location6,8,11. For estimating the absolute number of visits to a recreation resource, our results suggest that visitation models are currently most effective when some recreation count data are gathered on-site. Our visitation models were able to explain 91% of the variability in actual use at sites in NNM when we included on-site count data, compared with up to 79% when we included only social media, precipitation, and calendar data. For practitioners, this suggests that social media data do not fully substitute for traditional on-site counts. Clearly, social media provide the greatest capacity for estimating recreation use at sites where visitors share at least some content. Social media data may also be most beneficial for filling in spatial and temporal gaps in traditional recreation monitoring programs, to capture unique events or other situations that might cause visitation to deviate from the long-term trend. In addition, these two approaches could be paired to strategically inform one another. For instance, social media data might be used to help stratify individual sites according to relative use (e.g., high, medium, low) for traditional recreation monitoring programs that employ stratified on-site sampling schemes37,38.
Considerable technical sophistication is required to gather data from social media platforms and other sources, to aggregate and query data, and to apply statistical models to estimate visitation. This study relies on a mixture of recreation data sources spanning a range of large areas and time periods. Storing, aggregating, and properly querying data required us to develop a data ontology for tracking the location and provenance of various types of visitor counts. Further, multiple types of on-site counts as well as numbers and locations of social media posts were stored with a relational model that could be queried and processed to generate visitation estimates by site and day. Recreation managers likely lack the time, access to computing and storage, and quantitative skills necessary to perform these tasks. This highlights the need for continued partnerships among scientists, practitioners, and data providers to develop tools that practitioners can use to apply new approaches for recreation monitoring programs.
Next steps for researchers
This study extends our knowledge of how social media can improve estimates of recreation use on public lands. Yet there is a clear need for additional research in other landscapes where visitor demographics and recreation resources differ from those in our two study regions, including urban parks, dispersed-use wilderness, or shoreline areas, and in regions outside of the US. Future research could also apply our current understanding of measuring recreation use with social media data to address questions that currently face recreation managers and policymakers. How, for instance, does recreation use change around sites that are suddenly closed because of management actions or natural disturbance, such as wildfire or flooding? During this study in NNM, multiple sites were closed to the public at various times because of heightened fire risk, active fires, and flooding. It would be interesting to apply the new visitation modeling techniques to investigate if use was redistributed to other substitute sites or whether people ignored certain closures. Alternatively, well-enforced closures might present opportunities to investigate the amount of social media that is falsely or mistakenly posted from locations at times when people are not actually present. Another promising future research topic is the potential feedback loops (positive or negative) between the popularity of social media posts for individual recreation sites and recreation use at those sites, and how managers might anticipate, react to, or mitigate that feedback.
The social media environment is characterized by dynamic popularity of online platforms and evolving data availability and cost. The research community working in this area may benefit from additional research to improve our understanding of how to identify and incorporate new data platforms in existing processes and tools, and address data gaps across time or space caused by changes in data availability. Future research could also consider the possibility that visitors may share the same content on multiple social media platforms. While this study focused on simply counting social media posts to estimate recreation use, other information included in posts clearly offers opportunities for understanding visitation, recreation behavior, and visitor characteristics. The content (words and images) and other metadata associated with social media (such as the user’s profile) may provide additional insights into visitor numbers, experiences, activities, and satisfaction during the recreation visit39,40, as well as the characteristics of the visitors themselves, though research is needed in order to understand this potential. Layered on top of these research opportunities is an ongoing need for appropriate protocols to protect individual privacy and maintain ethical standards.
Source: Ecology - nature.com
