1 Introduction

Integrating biophysical and social aspects of ecosystems has become a cornerstone in conservation and sustainability mechanisms in protected areas worldwide (Daily, 2000). In these areas, local communities depend on and co-exist with protected landscapes, playing an important role in the management of ecosystems and their resources (Venter et al., 2014). In this context, the UNESCO “Biosphere Reserve” regime was launched to help reconciling conservation goals and the sustainable use of natural resources (Van Cuong et al., 2017), placing cultural benefits from ecosystems, generally known as cultural ecosystem services (CES), as a topic of interest in mountain socio-ecological systems (Schirpke et al., 2016, 2020).

CES offer an opportunity to explore how people interact with ecosystems, capturing multiple values from nature pertaining, for instance, to spiritual and inspirational enrichment, cognitive development, recreational engagement, or aesthetic fulfilment (Chan et al., 2012). Among others, CES opportunities promote tourism revenues, shape human heritage and traditions, and sustain public support for conservation investments (Wood et al., 2018). Nevertheless, under unsustainable management options, the over-exploration of CES may bring undesirable effects, for example by promoting human pressure and impacts on strictly protected biodiversity values (Buckley et al., 2016). Therefore, knowing where and how CES are shaped inside Biosphere Reserves is key to promote conservation policy, management, and communication.

Large digital data shared online have been increasingly used by researchers to support the identification and monitoring of CES at several scales (Richards & Friess, 2015). From this “digital conservation” perspective (Arts et al., 2015), the use of social media data to infer on CES is receiving particular attention (e.g. Egarter Vigl et al., 2021). The content analysis of publicly shared social media data, such as photographs, has been helpful to e.g., infer on aspects of nature appreciation (Vaz et al., 2020), monitor visitors’ movements (Tenkanen et al., 2017), or identify visitors’ preferences in protected areas (Hausmann et al., 2018).

Earth observation (EO) technology has also emerged as a promising tool to capture information on CES opportunities (Braun et al., 2018; Van Berkel et al., 2018; Vaz et al., 2019). The use of ancillary data from Geographic Information Systems (GIS) and satellite information can be particularly useful to describe and analyse the biophysical context and nature attributes that support CES opportunities, for instance by informing on the location of accessible cultural features (e.g., hiking trails or monuments) or by inferring on landscape visual-sensory characteristics (e.g., colour diversity and complexity) which are attractive to people (Tveit et al., 2006). Particularly when combined with social media analysis, EO can constitute a promising tool for CES monitoring in mountain protected areas (Vaz et al., 2019, 2020).

Nevertheless, despite the potential of EO and social media for addressing CES in mountain landscapes, their combined application is still far from being completely explored. Following our previous research on the subject (Vaz et al., 2019, 2020), this chapter uses publicly available data from the two digital sources to infer on CES opportunities in Biosphere Reserves. Specifically, we aim to: (a) identify which attributes contributing to CES prevail in the Biosphere Reserves, considering social media users’ information; and (b) understand how those CES attributes relate with different biophysical and landscape settings, captured through GIS and satellite EO data. Our approach is tested in mountain landscapes from two contrasting Biosphere Reserves in the Iberian Peninsula (Southwestern Europe): Peneda-Gerês (Northern Portugal) and Sierra Nevada (Southern Spain).

2 Methods

2.1 Test Areas

Our approach focused on two mountain protected areas: Peneda-Gerês (part of the Gerês-Xurés Biosphere Reserve, in Northern Portugal) and the Biosphere Reserve Sierra Nevada (Southern Spain; Fig. 8.1). Elevation in Peneda-Gerês (950 km2 area) ranges between 100 and 1548 m. The climate is Warm-Summer Mediterranean (following the Köppen-Geiger climate classification). The mean annual temperature ranges between 13 and 15 °C and the total mean annual rainfall is 2000 mm. Elevation in Sierra Nevada (1722 km2 area) goes from 860 to 3482 m. The climate is mostly Hot-Summer Mediterranean, with mean annual temperature of 0 (above 3000 m) and 12–16 °C (below 1500 m), and total mean annual rainfall is 600 mm (reaching more than 1500 mm as snow during winter and above the 2000 m elevation). Both mountain areas hold several protection regimes (from Natural and National Parks to broader Natura 2000 Special Protection Areas), showing remarkable biodiversity values in the wider Mediterranean hotspot. The socio-economy in both reserves is grounded on recreational and touristic activities and traditional land use practices devoted to local agro-pastoral and farming revenues.

Fig. 8.1
figure 1

Location of the test areas in the Iberian Peninsula (Southern Europe): Peneda-Gerês, in Portugal, and Sierra Nevada, in Spain

2.2 Empirical Approach

For the two test areas, we followed a three-step approach (Fig. 8.1A–C). First (A), we compiled a georeferenced dataset of in-field photographs from social media and classified the possible CES opportunities displayed in each photograph. Then (B), we compiled a set of EO variables, including satellite information and ancillary GIS data, expressing attributes pertaining to topography, accessibility, points of leisure interest, vegetation functioning and landscape visual-sensory attributes, as potential predictors of the prevalence of those CES opportunities. Finally (C), we applied a Multi Model Inference (MMI) to evaluate the explanatory power of the selected predictors on CES (inferred from the social media content).

2.3 Social Media Data

We evaluated the content of social media photographs from Flickr and Wikiloc platforms, to identify nature attributes underlying CES opportunities, here understood “as the characteristics of elements of nature that provide opportunities for people to derive cultural goods or benefits” (Haines-Young & Potschin, 2018). Photographs were collected using the Application Programming Interface (API) from each social media platform together with Python tools, for the years 2015 to 2018 (see Vaz et al., 2020 for details on data collection). We considered this time period to coincide with the time from which satellite data from Sentinel 2 became available for both test areas (section: Earth observation data). For each photograph, we registered its spatial location (latitude and longitude) and manually classified its main content in one of the following categories: (1) Landscape and nature, i.e. ‘wide-open’ shots of nature, often with a visible horizon, representing people’s enjoyment of landscape aesthetics (Richards & Friess, 2015); (2) Flora and fauna, i.e. close-up shots of animals or plants, broadly aligning with the CES of species appreciation (Goodness et al., 2016); (3) Recreation and sports, representing people engaging in recreational activities such as skiing or cycling (Richards & Friess, 2015); (4) Cultural heritage, dominated by cultural structures, e.g. historic monuments, relating to cultural heritage and spiritual enrichment (Blicharska et al., 2017); and (5) Rural tourism, i.e., human activities relating to social enjoyment, such as gastronomic enjoyment (Riechers et al., 2016). Given the experience of the team on the topic (e.g. Vaz et al., 2019, 2020; Ros-Candeira et al., 2020; Moreno-Llorca et al., 2020), and to minimise classification biases, the manual classification of the photographs was done by one author (ASV), being then independently verified by other two authors (RML and ASC). We excluded photographs with irrelevant subjects (e.g., advertisements, pamphlets, or drawings). Photographs protected by users’ privacy were neither downloaded nor analysed. The final set included 1644 and 761 georeferenced photographs for Peneda-Gerês and Sierra Nevada, respectively.

2.4 Earth Observation Data

We considered an initial set of candidate variables, derived from the most updated and freely available satellite platforms and ancillary GIS data for modelling the distribution of CES opportunities in each test area (Table 8.1). GIS data included information on: (1) topography (i.e., elevation and slope), (2) visual (viewshed dimension) and physical (presence and distance to rivers, roads and trails) accessibility, and (3) points of leisure interest, namely presence and distance to touristic lakes, ski resorts, public recreational facilities, and villages (e.g. Schirpke et al., 2016; Yoshimura & Hiura, 2017). Satellite data from Sentinel-2a/b L1C images (available from the year 2015 to 2018) was considered to obtain information on: (4) vegetation functioning, i.e. through the Normalized Difference Vegetation Index, and (5) landscape visual-sensory attributes, including the amount and diversity of landscape colours using the visual RGB spectrum (e.g. Braun et al., 2018; Van Berkel et al., 2018). Following Vaz et al. (2019), visual-sensory attributes were computed separately for each meteorological season of the year: Winter (December–February), Spring (March–May), Summer (June–August) and Autumn (September–November). Based on a series of spatial autocorrelation tests with increasing moving-window sizes, a grid cell size of 500 × 500 m spatial resolution was considered suitable for subsequent analyses (see Vaz et al., 2020 for details).

Table 8.1 Types of predictors considered in each competing model (M) for both test areas with a brief methodological description

2.5 Multi Model Inference

The number of photographs of each CES category in each grid cell was used as response variable in a Multi Model Inference (MMI) framework (Burnham & Anderson, 2002). Six competing models (M) were used to test the hypotheses that CES were mostly explained by: M1 - topography; M2 - accessibility; M3 - points of leisure interest; M4 - vegetation functioning, and M5 – landscape visual-sensory attributes. Generalized Linear Models (GLM) with Poisson distributions (for count data) were fitted separately for each CES category (Burnham & Anderson, 2002) and implemented in the R software. The maximum number of predictors per model was set to four and only predictors with a pairwise Spearman value lower than 0.6 and Variance Inflation Factor (VIF) lower than 5 were considered, to avoid multicollinearity issues (Fox & Weisberg, 2018). A total of 27 and 22 predictors were considered for Peneda-Gerês and Sierra Nevada, respectively. For model comparison, we calculated the Akaike Information Criterion difference (ΔAICc), as ΔAICc = AICc initial - AICc minimum (where AICc initial is the second order AICc of the competing model and AICc minimum is the second order AIC of the best model in the set). We calculated the weight (wi) of each competing model, that represents the proportion of evidence from a competing model in relation to the total evidence from all models (ranging between 0 and 1). We also computed the Nagelkerke deviance D2, corresponding to the difference between the residual deviance of each competing model against the deviance of the null model (M6), as a goodness-of-fit measure of each competing model. Only models with a D2 value higher than 0.10 were considered (following Dormann et al., 2018).

3 Results and Discussion

From our results, the categories of CES attributes “landscape and nature” and “fauna and flora” were largely found in the content of social media photographs in both test areas, being congruent with the natural values that typically dominate protected areas (Hausmann et al., 2018; Richards & Friess, 2015). The category “recreation and sports” was also amongst the most frequently identified categories in Sierra Nevada, a pattern in agreement with the offer of high mountain-related activities (such as skiing). “Cultural heritage” and “rural tourism” were the least frequent in both areas, despite holding popular cultural traditions and festivities as well as rural villages of touristic importance (Fig. 8.2).

Fig. 8.2
figure 2

Proportion of social media photographs assigned to each cultural ecosystem service in (a) Peneda-Gerês and (b) Sierra Nevada showing the prevalence of photographs capturing “landscape and nature” appreciation, “fauna and flora” and “recreation and sports”. “Cultural heritage” and “rural tourism” were the least represented CES opportunities

In Peneda-Gerês, “landscape and nature”, “fauna and flora” and “recreation and sports” were primarily explained by landscape visual-sensory attributes expressing the richness of colour diversity, particularly during the meteorological seasons with higher spectral contract, i.e., Autumn and Spring (Fig. 8.3, Table 8.2). This trend highlights the relation between landscape colour seasonality and landscape appreciation (Tveit et al., 2006), as previously suggested in Vaz et al. (2019). The easiness of visual and physical accessibility (namely through the viewshed dimension and distance to rivers) and the distance to public leisure facilities were also of high predictive relevance for landscape appreciation and recreational engagement. The importance of these variables in the creation of CES opportunities has already been suggested for other mountain landscapes (Schirpke et al., 2016; Tenerelli et al., 2016; Vaz et al., 2019). CES pertaining to “cultural and heritage” and “rural tourism” were found to be primarily explained by the existence of human infrastructures, particularly in more accessible areas (e.g., closer to trails; Table 8.2).

Fig. 8.3
figure 3

Multi Model Inference (MMI) results: Akaike weights (wi) and explained adjusted deviance (D2) for each competing model. A gray shading (D2 > 0.10) is used in the figure to highlight the model with the highest explanatory power (dark gray) and those that follow (light gray), for each category of nature attributes underlying cultural ecosystem services (inferred from the content of social media photographs)

Table 8.2 Summary of results from each competing model holding explanatory power (highlighted in Fig. 8.3) for the distribution of social media photographs in Peneda-Gerês and Sierra Nevada. Models are presented from the best to the least fit hypothesis, based on the Akaike Information Criterion difference (ΔAICc) (see Fig. 8.3)

In Sierra Nevada, CES opportunities were mostly driven by their proximity to points of leisure, including lakes, leisure infrastructures, and ski facilities, alongside with accessible features (i.e. trails). Nevertheless, landscape visual-sensory attributes, and particularly NDVI heterogeneity and richness of visible spectral colours, were also found to determine CES incidence. Landscape visual-sensory attributes during Spring were of particular relevance for explaining the cultural appreciation of “fauna and flora”, which may well reflect the time in which animals (e.g., through active physiological behaviours) and plants (e.g., through colourful leaves and flowers) become more evident for the visitors (Table 8.2).

Overall, differences found in the predictors of “landscape and nature” and “recreation and sports” between both areas suggest that, in Sierra Nevada, visitors may rely largely on the presence of leisure infrastructures for their nature-based activities. For instance, skiing is extremely popular in high-altitudinal areas of Sierra Nevada and can only take place in the protected area due to sport facilities. Contrastingly, visitor preferences in Peneda-Gerês may be more determined by the visual-sensory characteristics of the landscape, without necessarily being constrained by human infrastructures. The occurrence of infrastructures was also prominent for “rural tourism” in Sierra Nevada, converging with the prevalence of rural villages of touristic importance in the region (e.g., Alpujarras), in contrast to Peneda-Gerês, where “rural tourism” category was more associated to accessibility. Despite these differences, the distribution of CES attributes associated to “fauna and flora” was mostly explained by visual-sensory properties in both areas, which may indicate the pursuit for wildlife appreciation in more diversified and natural landscape mosaics. Similarly, “cultural heritage” attributes were also largely explained by the same predictors in both protected areas, and much associated to the existence of public facilities and local villages.

The combined results for Peneda-Gerês and Sierra Nevada obtained in this study advance our understanding of the potentialities of using EO and social media data to identify opportunities for CES in mountain protected areas. Considering also previous research on these and other protected areas (Vaz et al., 2019, 2020), this study reinforces that the combination of satellite-derived metrics on vegetation functioning and colour diversity with GIS data, expressing accessibility efforts and the location of leisure facilities, can be used to inform cultural benefits in mountain protected areas, and thereby support the timely monitoring of human-nature interactions over wide spatial scales. Inevitably, some methodological considerations should be highlighted. For instance, in a MMI framework, the predictive power of a model is evaluated against the power of the other competing models, not necessarily meaning that the whole variation in CES is explained by that model. Also, although we can infer on the prevailing CES opportunities from the content of social media photographs, extrapolating which nature elements are indeed most preferred and selected by people should be done considering psychological (e.g., perceptions) and social (e.g., values) variables, that inevitably rely on the use of complementary socio-cultural approaches, such as participatory mapping or questionnaire-based surveys.

The combination of online interactions (i.e. through social media) and EO data used in this chapter showed to be promising to assess and monitor ecosystem features that underly CES opportunities (e.g. wide-view landscapes), yet the demand for those features and the way people enjoy them (e.g. “beautiful landscape”) will need further elucidations. Creating community science initiatives and encouraging citizen participation (inside and outside social media) to advance knowledge about nature’s cultural benefits to people in the targeted protected areas would constitute a further step to develop CES monitoring systems and to adopt more adequate management decisions. Considering that the EO data used in this chapter can be spatially projected (in a map) and timely updated, it can aid local managers in the identification of areas with synergies between biodiversity conservation and cultural/touristic values (Turnhout et al., 2013). It can also help to identify potential conflicts between tourism and strictly protected zones in Biosphere Reserves (Van Cuong et al., 2017), and thereby guide awareness campaigns or even target reinforcement actions to restrict accessibility and safeguard wider natural values in the mountain Biosphere Reserves.