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

Experiences gained through in person (in-situ) interactions with ecosystems provide cultural ecosystem services. These services are difficult to assess because they are non-material, vary spatially and have strong perceptual characteristics. Data obtained from social media can provide spatially-explicit information regarding some in-situ cultural ecosystem services by serving as a proxy for visitation. These data can identify environmental characteristics (natural, human and built capital) correlated with visitation and, therefore, the types of places used for in-situ environmental interactions. A range of spatial models can be applied in this way that vary in complexity and can provide information for ecosystem service assessments. We deployed four models (global regression, local regression, maximum entropy and the InVEST recreation model) to the same case-study area, County Galway, Ireland, to compare spatial models. A total of 6,752 photo-user-days (PUD) (a visitation metric) were obtained from Flickr. Data describing natural, human and built capital were collected from national databases. Results showed a blend of capital types correlated with PUD suggesting that local context, including biophysical traits and accessibility, are relevant for in-situ cultural ecosystem service flows. Average trends included distance to the coast and elevation as negatively correlated with PUD, while the presence of major roads and recreational sites, population density and habitat diversity were positively correlated. Evidence of local relationships, especially town distance, were detected using geographic weighted regression. Predicted hotspots for visitation included urban areas in the east of the region and rural, coastal areas with major roads in the west. We conclude by presenting a guide for researchers and practitioners developing cultural ecosystem service spatial models using data from social media that considers data coverage, landscape heterogeneity, computational resources, statistical expertise and environmental context.