Using digital sources to explore landscape in Switzerland

In this presentation, we focus on the potential uses of new data sources (Purves & Derungs 2015), particularly in the form of implicitly and explicitly georeferenced digital text sources and social media, as a route to gathering bottom-up information on values, behaviours and preferences with respect perceived natural landscapes (c.f. Kienast et al. 2015). Explicitly georeferenced sources are social media data, for example Flickr images, where coordinates are stored within metadata associated with content. Implicitly georeferenced text sources include toponyms or other information on which basis coordinates can be assigned through the use of methods to identify references to location and disambiguate these appropriately.

We treat these data as bottom-up since they are generated by a large number of individuals, with differing motivations and behaviours, generated at different times and through varying mechanisms.

Through the prism of Swiss protected areas, we explore three complementary dimensions of such data:

1. Spatial patterns of visitation after normalisation for accessibility
A key problem using data generated bottom-up is normalisation for biases (for example, what is the underlying distribution of locations which are visited, and thus photographed (e.g. Crandall et al. 2009)). Here, we generate a simple model of accessibility and explore visitation as a function of accessibility based on social media data.

2. Patterns of use as expressed through individual contributors
Social media data can be explored on many levels. For instance, by linking individual contributions, it is possible to generate patterns of behaviours at the level of individuals, and categorise these to identify particular classes of behaviour (c.f. Giradin et al. 2008). We explore the origins of visitors according to social media data (i.e. are these locals), and the range of protected areas they visit as well as variation in behaviour over time.

3. Semantics of values and behaviours as expressed in these sources
In particular when richer text sources are analysed, it is possible to have access to more detailed information about the semantics of perceived landscapes and values associated with them through the use of natural language processing methods (c.f. Derungs & Purves 2016). We will present results illustrating not only what elements of the landscape are mentioned, but also which collocates are used to describe such elements particularly with respect to perceived properties of locations (e.g. “spectacular views” and “tedious paths”)

We will both illustrate the types of methods which can be applied to these data sources and discuss potential limitations of the analytical approaches we take. We will also compare our results to existing landscape monitoring approaches in Switzerland (Kienast et al. 2015), which, in the case of culturally determined perception, are primarily based on more traditional data sources such as questionnaires, focus groups and interviews.