Social media reveals where nature delivers cultural benefits in Ireland

A new comparison of spatial models has used geotagged photographs to map where people encounter ecosystems in County Galway, Ireland, offering a practical way to identify the places where landscapes generate cultural benefits such as recreation, social connection, spiritual enrichment and appreciation of biodiversity. Researchers analysed 25,170 geotagged Flickr photographs uploaded between 2010 and 2019, converting them into a visitation indicator designed to avoid exaggerating the contribution of prolific photographers. The study found that a mixture of natural conditions, accessibility and human activity shapes where cultural ecosystem services may flow, while demonstrating that a single statistical model can conceal important local differences.

The team tested four approaches to the same landscape: global regression, geographic weighted regression, maximum entropy modelling and the InVEST recreation model. These tools are often used independently, yet few studies have compared them on identical data and for the same research question. The Galway analysis suggests that model choice is not a purely technical decision. Data coverage, spatial resolution, statistical expertise, computing resources, environmental heterogeneity and whether the aim is to estimate average relationships or predict visitor suitability can all change the conclusions of an assessment. The results therefore function as both a local ecological finding and a guide for researchers planning large-scale analyses of people’s interactions with nature.

To measure visitation, the researchers calculated photo-user-days, or PUD. A PUD records the number of users uploading at least one photograph in a particular place on a given day, rather than simply counting every image. This reduces the influence of highly active contributors while still allowing one person to contribute to several locations visited during a day. For the landscape analysis, photographs were assigned to 2-square-kilometre grid cells after a 200-metre buffer was applied around GPS coordinates to accommodate expected positional error. The resulting dataset contained 6,762 PUDs, with individual cells ranging from none to 413 and an average of 5.65. Hotspots occurred along the western coast and south-east of Galway City, whereas low-activity coldspots were concentrated farther east.

Before modelling, the researchers tested whether social media activity could serve as a credible proxy for real-world visitation. They compared PUD values at 38 attractions with official visitor statistics supplied by Ireland’s national tourism body, excluding indoor venues because they were unlikely to provide opportunities for in-person ecosystem experiences. The comparison produced a Pearson correlation of 0.7 and a statistically significant regression relationship, providing support for using Flickr activity as a visitation indicator in Ireland. The result does not mean every photograph represents a unique tourist or that social media users represent the population as a whole. Instead, it shows that the geographic pattern of documented visits contains useful information about where people encounter cultural services in this landscape.

A non-spatial logistic regression initially produced an average relationship for the entire county. Greater distance from the coast and higher elevation were associated with a lower likelihood of a photograph record, whereas major roads, population density, habitat diversity, recreational sites and other accessibility indicators tended to increase that likelihood. The model achieved an area under the receiver operating characteristic curve, or AUC, of 0.839. However, spatial autocorrelation in its residuals showed that the model performed differently in different parts of the study area, violating an important assumption of independence. A mixed model incorporating spatial random effects produced a better fit and removed the detected autocorrelation. Its lower Akaike information criterion, 1,518 compared with 1,629, made it the preferred global model.

The most striking result emerged when geographic weighted regression allowed environmental relationships to change from place to place. Its AUC reached 0.909, and the model identified a complete reversal in the role of distance from towns. In eastern Galway, areas close to towns were more likely to contain visitor records, whereas remote western areas were more likely to be visited even when far from urban centres. Recreational sites were significant in three distinct hotspots but not throughout the county. Habitat diversity and geological heritage likewise mattered in some locations rather than everywhere. The contrast may reflect the region’s varied geography: rugged mountains, wetlands and remote Atlantic coastline in the west differ from the more agricultural landscapes surrounding eastern towns. An average regional coefficient would have concealed this pattern entirely, showing why local relationships should be examined before global trends are applied to specific places.

Maximum entropy modelling produced a different kind of output because it treated the task as prediction of habitat-like suitability rather than direct estimation of a statistical relationship. The method compares locations with known photograph records against background environmental conditions, making it suitable where apparent absences may simply reflect incomplete data. With environmental layers converted to 100-metre rasters, the model produced clusters of high predicted visitation along roads, near Galway City and around parts of the coast, while identifying wetlands with weak connectivity and low population as less suitable. Its mean AUC was 0.8. Jackknife analysis indicated that elevation, distance from the coast, major roads and recreational sites contributed especially important information, whereas variables such as river length, slope and geological heritage had limited predictive power. Such rankings can help managers decide which environmental information to collect or which areas to monitor, although predicted suitability is not a direct measure of cultural benefit or visitor welfare.

The InVEST recreation model offered the most accessible workflow, with a self-contained interface and online documentation, but it also imposed greater constraints. It used an archived Flickr dataset covering 2005 to 2017, analysed annual PUD values with ordinary least squares regression and allowed only a limited set of preconfigured environmental indicators. It also did not automatically provide the same model optimisation and collinearity diagnostics used in the custom regressions. Consequently, its results could not be compared statistically with the other models. Nevertheless, it highlighted several of the same influential characteristics, including elevation, coastal distance, roads and recreational facilities, while requiring less advanced coding than the other approaches. In practice, this tool is attractive for users who need a rapid, accessible assessment and possess adequate spatial data, but its fixed archive and restricted customisation can make it less suitable for current, question-specific analysis.

The comparison exposes several limitations that matter as much as the maps. Flickr contributors form a self-selecting group, and the platform may underrepresent people who do not upload photographs, including those with less access to technology or different reasons for visiting a landscape. The study also used location as a proxy for the potential delivery of a cultural service without asking contributors whether they experienced recreation, inspiration, social interaction or spiritual benefit. Photographing an ecosystem may capture scenery, but it may also document an event whose main purpose was unrelated to nature. GPS errors, uneven platform coverage and the tendency to photograph picturesque places can introduce further distortion. The researchers therefore recommend combining social media with official visitor statistics, surveys, choice-based experiments and participatory methods rather than treating digital footprints as a complete measure of social value.

Even with those caveats, the study offers a strong framework for making cultural ecosystem services visible in planning and natural capital accounting. It identifies rural coastal areas with major roads, moderate elevations and diverse habitats as potential zones of concentrated visitor activity, alongside the urban hotspot around Galway City. High visitation can support tourism and local economies, but it can also increase disturbance, waste and pressure on ecosystems, so the same maps may help managers target protection as well as promotion. The central lesson is that cultural services emerge from a combination of natural, human and built capital rather than from scenery alone. Roads and towns influence access, while elevation, coastlines and habitat diversity shape experience. Because those interactions change across space, researchers should inspect residual autocorrelation, test for local relationships, match indicators to spatial scale and justify model choices transparently. In an era of rapidly growing digital datasets, such discipline can transform everyday photographs into useful evidence of where ecosystems matter to people, without pretending that a map can tell the whole story.

Subject of Research: Spatial modelling of cultural ecosystem services using geotagged social media data

Article Title: Spatial analysis of cultural ecosystem services using data from social media: A guide to model selection for research and practice

Article References: Neill, A., O'Donoghue, C., & Stout, J. (2023). Spatial analysis of cultural ecosystem services using data from social media: A guide to model selection for research and practice. One Ecosystem, 8, Article e95685. https://doi.org/10.3897/oneeco.8.e95685

Image Credits: AI Generated

DOI: 10.3897/oneeco.8.e95685

Keywords: cultural ecosystem services, social media data, Flickr, spatial modelling, visitor patterns, geographic weighted regression, maximum entropy modelling, ecosystem assessment, County Galway, recreation, Spatial, analysis