Understanding patterns of human-nature interactions is crucial for sustainable land use planning and nature conservation (Venter et al., 2016). However, spatially and temporally accurate data on threats affecting biodiversity persistence are lacking (Joppa et al., 2016), and datasets needed to inform conservation decision-making are limited and often biased (Di Minin & Toivonen, 2015).
Spatial data generated by non-experts have recently become a valuable resource both in academia and the society in addition to more traditional data produced by scientists and other authorities (See et al., 2016;Goodchild, 2007). Geographic information generated by crowds, such as geotagged photos and other location-based social media data provide diverse information about human activities across the globe. These user-generated data, as opposed to official data sources (such as census data, visitor counts and surveys), may provide complementary information about the values, observations and activities of different groups of people especially in regions where official data is collected rarely.
Social media, in general, refers to computer-based applications used for networking and sharing digital content. Here, social media data refers to the spatial attributes (location), temporal attributes (time), and relevant content (text, photos, and video) generated by users of different social media platforms (such as Flickr, Instagram and Twitter). Publicly shared content in different social media services can be often accessed in large quantities through Application Programming Interfaces (APIs).
Social media data is used in various fields of science, also increasingly in environmental sciences, to explore spatial and temporal patterns of human activities. However, issues of data quality and data ownership might limit the use and trustworthiness of these data sources in systematic analyses and decision-making support (Sui & Goodchild, 2011) and not many studies, especially in the context of environmental studies, have aimed to validate observed patterns or to systematically asses evident gaps in the data, for example related to spatial coverage of the available data in different platforms.
This work aims to underpin the potential and limitations of using user-generated geographic information in environmental studies. Furthermore, this work aims to understand how usergenerated geographic information can complement traditional data sources in the study of human-nature interactions. Focusing on natural and semi-natural environments such as national parks and urban green areas at different spatial scales, the aim is to map and analyse nature recreation and human pressure using publicly shared social media posts in combination with other available data. The general goals of this wor are to 1) extract meaningful patterns related to humannature interactions in green areas in order to inform Understanding human activities in green areas with social media data conservation-related decision making at different spatial scales, while 2) understanding the biases in location-based social media data. The work includes the development of automated workflows for data processing and analysis, comparisons between user-generated data and official datasets, and accounting for gaps, inaccuracies and bias in the user-generated data at different spatial scales. The main objectives and related questions are the following:
• Analyzing the potential and limitations of different social media data for studying human activities in green areas: What kind of information do we get from social media, what biases are included in the data and how is it useful for studying human activities in green areas? How does social media data compare with other information sources from focus areas?
Understanding the park visitors: Whose views and observations are presented in user-generated content from national parks? Where do the visitors come from and how do they move within and between green areas?
Mapping conservation opportunities and threats: Can we characterize national parks and national park visitors based on social media data? Can we identify human pressure on the environment from social media data? What tradeoffs between nature recreation and nature conservation can we discover on a regional/global scale?
Data and tools related to the information age (Castells, 2000) and the big data revolution (Mayer-Schönberger & Cukier, 2013; Kitchin, 2014) have opened up new possibilities for geographic knowledge discovery (Mennis & Guo, 2009;Crampton et al., 2013). Before, recreational use patterns and preferences related to green spaces have been studied using surveys (Tyrväinen, Mäkinen & Schipperijn, 2007), activity diaries (Mytton et al., 2012), GPS tracking (Korpilo, Virtanen & Lehvävirta, 2017) and public participatory GIS (PPGIS) (Brown, Schebella & Weber, 2014;Laatikainen et al., 2015). However, these methods are often costly to implement (Kwan, 2013) and often limited to a specific case study area (Ives et al., 2017). Recently, large amounts of geographic 'big data', such as location-based social media data, have become available for capturing information about people's movement and activities in unprecedented volumes. This "location-based story telling" (Sui & Goodchild, 2011) in various online platforms such as Facebook, Twitter and Instagram, has fundamentally transformed the notion of geographic information in recent years.
Location-based social media data is often discussed in the context of Volunteered Geographic Information (VGI). The concept VGI, coined by Goodchild in 2007, is widely used to describe geographic datasets generated by non-experts. Vast amounts of spatial data are continuously created in collaborative projects such as the OpenStreetMap, social networks such as Twitter and other location-aware platforms on the web which host user-generated content. However, the term VGI does not fully capture the nature of more spontaneously generated data (See et al., 2016) such as tweets and Flickr photos which have originally been shared for other purposes than mapping and research. These passively shared data evidently require special consideration related to the ethical use of data, and representativeness of the results. Thus, there is a need to further position social media among other sources of user-geographic information and authorative data sources.
Social media data has been used in many application fields of geography to study spatial phenomena, especially in the urban context. The study of population dynamics in cities (Longley & Adnan, 2016;Steiger et al., 2015), spatial diffusion (Crampton et al., 2013) and humanitarian response applications (Crooks et al., 2013) are only a few examples of existing research from the fields of geography and geographic information science.
However, examples in environmental studies, especially in conservation science are still limited (Di Minin, Tenkanen & Toivonen, 2015). Studies focusing more on human-nature interactions include the quantification of visitation rates (Wood et al., 2013;Levin, N., Kark, S. and Crandall, 2015), assessment of cultural ecosystem services and people's interests (Richards & Friess, 2015;Roberge, 2014), and the extraction of species data (Barve, 2014;Stafford et al., 2010) from social media. Also, only a few studies have used social media data to understand human-nature interactions in urban environments. These include methodological development for studying cultural ecosystem services based on social media content analysis with a case study from urban mangroves in Singapore (Richards & Friess, 2015;Thiagarajah et al., 2015) and tourism crowding (including parks) based in check-in data from Shanghai (Shi, Zhao & Chen, 2017).
Existing studies are often limited to only a single social media platform, and lack comparisons and validation against other data sources. Most studies using social media data for environmental studies rely solely on one platform (mostly Flickr). Flickr might be the most suitable platform when looking at biodiversity features, while Instagram or other available data sources might reflect better the activities present in the area of interest (Hausmann et al., 2017b). Studies with more in-depth and advanced content analysis have thus far been limited to smaller study sites (for example Richards & Tunçer, 2017) and there is great potential to scale up such analysis to continental and even global scales.
A critical approach is needed when using social media as a source of information (Boyd & Crawford, 2012). Firstly, ethical issues related to using people's personal data need to be considered even when using publicly available content. Secondly, it can be difficult to assess how representative the captured social media users are of the population in question (Longley & Adnan, 2016). Furthermore, the data is often biased both spatially and temporally in relation to infrastructure, mobile phone coverage, and popular events, and potentially towards certain socioeconomic classes. This work aims to provide a deeper understanding of these issues in the context of conservation-related questions.
The study sites of this research consists of green area networks at different scales; urban green areas from the city of Helsinki, Finland, individual protected areas from Finland and South Africa and the global protected area network and key biodiversity areas worldwide. National parks with regular visitor monitoring schemes provide a test environment for comparing social media user counts and content to official statistics.
Main material for the study consist of openly shared locationbased social media posts from different social media platforms including (but not restricted to) Instagram, Twitter, and Flickr. Data is retrieved from platform APIs using existing packages and custom-made scripts in the Python Programming Language.
Social media data are used in conjunction with, and contrasted to official visitor statistics, surveys and other available data from the focus areas. Official visitor statistics and visitor survey data from national park authorities are used together with social media content from different platforms. Global analysis from projected areas and key biodiversity areas is done using additional data from the International Union for the Conservation of Nature (IUCN), and Birdlife International.
The first part of the study is focused on the potential and limitations of social media data for environmental studies through analyzing the different components of the data in terms of precision, accuracy and fit-for-purpose. Accuracy and precision of spatial and temporal information are assessed trough methods of data exploration and comparisons to ancillary data sets. Manual and automated content analysis of texts and images is used to explore if the content shared from green areas is thematically meaningful for green area management and conservation, and to produce further analyses of the spatial and temporal patterns of observed content categories (for example related to a specific activity or species within a national park). Results are compared with existing information about park visitation from case study sites with existing reference data. For example, activities revealed from social media photo content are compared with surveyed activities in a case study from a Finnish National Park.
Understanding bias in social media data includes the analysis of age-and language groups as well as the differentiation between data generated by locals and visitors. Questions related to visitors' social media usage are conducted in selected national parks in Finland in order to find out the proportion of national park visitors who are active in social media and to link this information to demographic background variables. Furthermore, different platforms are compared in terms of their information content. Based on earlier findings from Kruger National park in South Africa, Flickr contains more information related to biodiversity features, whereas Instagram posts portray more often human activities (Hausmann et al., 2017a). In this work, such differences in shared content will be explored further in different spatial and temporal contexts.
The last sections of the work apply the earlier findings for answering questions related to conservation opportunities and threats in the global protected area network and key biodiversity areas using a combination of user-generated and official data sources.
The work has potential to bridge the gap between recent advances in social media data analytics and information needs in conservation science. The work will likely reveal new information about spatial and temporal patterns of human activities in green areas worldwide, especially from areas with no systematic visitor monitoring in place.