Presence of agriculture in photos of Norwegian landscapes uploaded to Flickr

ABSTRACT

The aim of the article is to assess whether agricultural landscapes play a role in the perception of Norway held by tourists and residents. An additional aim is to analyse whether information accompanying images on social media indicate that the photographers have acknowledged the agricultural landscape. The authors used geotagged images uploaded to the image-sharing platform Flickr in their analyses. They selected photos from within the agricultural landscapes, inspected them, and categorized them according to extent and content. Additionally, they analysed the accompanying hashtags. The findings revealed that a large proportion of the photos contained agricultural landscapes, and thus confirmed the importance of the agricultural landscape for visual perception of and access to Norwegian landscapes. In addition, the lack of agricultural-related hashtags strengthened the authors’ suspicions that this might not have been widely recognized by the photographers. Thus, while agricultural landscapes commonly are considered primarily as landscapes of food production, the authors conclude that these landscapes also fulfil other functions and that their contribution to the perception of Norway is important. Additionally, many of the landscape elements seen and analysed in the sample of photos are elements that play a role in providing cultural ecosystem services.

KEYWORDS:

• agricultural landscape
• cultural ecosystem services
• hashtags
• landscape photographs
• social media

Introduction

While we may think of agricultural landscapes as landscapes focused on food production, they in fact also fulfil many additional functions. Agricultural landscapes may be habitats for a wide range of species (Artsdatabanken 2021), hold a significant proportion of cultural heritage (Olsson & Rønningen 1999), and be ‘everyday landscapes’ (sensu European Landscape Convention) (Council of Europe n.d.) for many people.

Agriculture, defined as fully cultivated land, represents only c.3.5% of land use in Norway. However, this is only one part of the agricultural landscape. We argue that the neighbouring area to the agricultural fields also needs to be considered part of the agricultural landscape. Also, through outfield grazing and harvesting of resources in the outfields, the actual influence of agriculture on the Norwegian landscape is much more extensive while still representing a minor proportion of the Norwegian land area, compared to forested and mountainous land. In addition, the geography of agriculture in Norway is worth mentioning. This 3.5% of fully cultivated land is distributed across the approximately 13 degrees of latitude that Norway spans, and thus is important to a much larger part of the country than the proportion itself indicates.

Cultural ecosystem services are defined by IPBES as ‘the nonmaterial benefits people obtain from ecosystems through spiritual enrichment, cognitive development, reflection, recreation, and aesthetic experiences’ (Bongaarts 2019). We anticipated that the agricultural landscape, being the ‘everyday landscape’ of a large proportion of the Norwegian population (Council of Europe n.d.), would also be important in terms of delivering cultural ecosystem services. We therefore expected that this could be inferred from images, as people could be expected to capture, for example, recreational, aesthetic, or spiritual motifs.

The function of the ‘everyday landscape’ is linked to historical settlement patterns; people settled based on the availability of food and other resources, transport possibilities, and where the climate was agreeable. As a result, the agricultural landscapes in Norway are closely linked to settlements and infrastructure. This becomes clearly apparent if we analyse land use in, for example, a 1 km wide buffer zone around cities and urban areas. In such zones, the proportion of cultivated agricultural land is not 3% but 20% (Aune-Lundberg & Ulfeng 2020). Thus, it can be hypothesized that cultivated agricultural land plays an important role in how Norwegian landscapes are experienced and perceived by tourists or residents. The link between agricultural landscapes and both settlements and infrastructure also implies that a wide range of landscape elements are intertwined with the areas used for agricultural production. There are buildings of different kinds, water in the form of ponds, creeks, lakes, and rivers, as well as the sea, and areas used for sports and recreation, such as football, golf, and horses used for recreation and leisure activities. Basically, most landscape elements linked to human presence can be found in agricultural landscapes in Norway.

Travel is one of the fastest growing industries globally, and Norway is no exception in this regard. Typically, Norway is marketed as a landscape of fjords and mountains, waterfalls, and steep cliffs. Agricultural landscapes appear to receive less attention from the Norwegian travel industry. However, given the three-dimensionality of Norway, many roads are built where the terrain is less steep. Typically, such terrain is along the coast and follows the flatter valley bottoms. It is also where a large proportion of agricultural land is located. Furthermore, given the wide geographical distribution of agriculture in Norway, and the fact that the agricultural landscape is possibly more accessible than other types of landscape in Norway, both visually and physically, we were curious to know whether this could be assessed as being important in recreation, tourism, and more general landscape perception.

Landscape photos can be seen as a source of data that possibly contain information that is absent in texts (Joo & Steinert-Threlkeld 2019). Such extra information presents novel opportunities for the assessment of landscape experienced by a large group of people. Moreover, Richards & Friess (2015) argue that content analysis can help us to understand the cultural ecosystem services represented in photographs uploaded to social media.

The research question we developed based on our above-described curiosity was whether agricultural landscapes appeared in photographs of Norway that had been uploaded to social media, and, if so, the extent to which the agricultural landscapes could be seen as important to the landscape experiences captured in the images. To do this we analysed (1) whether photos were taken in an agricultural landscape, (2) to what extent agricultural landscape elements appeared in photos, and (3) whether people used hashtags related to those agricultural landscape elements.

Methods

Data source and data access

We considered several social media platforms as image data sources. We decided to use Flickr in our study, given the positive findings from other studies that used Flickr (e.g. Oteros-Rozas et al. 2018; Komossa et al. 2020; 2021; Wartmann et al. 2021; Hartmann et al. 2022). Flickr is a service for hosting and sharing images with a focus on social interaction between users. According to da Mota & Pickering (2020), Flickr had more than 90 million registered users in 2018. Even though Flickr has been running since 2004, we did not want to access photos from an earlier period. Rather, we wanted our sample to be as recent as possible to avoid having to include temporal changes in land cover, technology, or people’s preferences.

We used the R package photosearcher (version 1.0) to access metadata for the selected photographs (Fox et al. 2020). The photosearcher package facilitates requests to the Flickr API, with the functionality of selecting photographs based on criteria such as spatial location, keywords, and minimum and maximum date the photograph was taken. Knowing the steep decline in the number of active farmers in Norway since the early 2000s (Statistisk sentralbyrå n.d.), and given our focus on agricultural landscape, we decided to limit our time period to 2016–2020 in order to minimize the possibility of landscape change. Regarding spatial location, we used Norway as our bounding box. Our sample was thus limited to photos taken in Norway without consideration of the nationality or country of residence of the people who had taken the photos. Upon finding that downloading multiple years at the same time gave unpredictable and unexplainable differences in the number of photos, we decided to download images for one year at a time. When repeating the download on consecutive days there were only small differences in the number of images per year and in total. In addition to the criteria that the photographs were geotagged (longitude, latitude), we fetched data for ID (id), owner (owner), whether the image was available to the public (ispublic), the date of the image (datetaken), the web address to access the image (url_c), the license connected to the image (license_name), and a list of hashtags connected to the image (hashtags). The URLs enabled us look at the images in a web browser.

Data filtering

The accessed image metadata were further filtered to keep only data from each unique image location (e.g. Dunkel 2015). For our purpose, we found it useful to capture images from different locations. People who are traveling may cover quite some distance and move through differing landscapes during the course of one day. We thus found it useful to accept multiple images by one user taken on one day, but we restricted the geographical locations to one image per point in space. If two or more photographs were taken at the exact same geographical coordinates, we kept the first image and excluded the additional images, without making any assessment of the image content. For the resulting selection, we recorded whether their geographical location placed them in the agricultural landscape. We used the same definition of agricultural landscape as the Norwegian monitoring programme for agricultural landscapes (Engan & Bentzen 2017), which defines an agricultural landscape as areas within a buffer of 100 m outside agricultural land. To identify agricultural land, we used the categories ‘fully cultivated’, ‘surface cultivated’, and ‘infield pastures’ in the Norwegian land resource map AR5 (Ahlstrøm et al. 2019). Using analyses available in GIS (geographic information systems), we identified the images georeferenced as having been taken from a location inside the agricultural landscape. Only these images, located within the 100 m buffer, were further assessed for image characteristics.

Spatial patterns of uploaded images

A prerequisite was that all downloaded images were geotagged. This enabled us to not only divide photos between agricultural and non-agricultural landscapes but also to investigate the location of images, related to other images or land cover. Overlaying the photos with the national land resource map (AR5) enabled us to assign the point from which the photo was taken to a land cover/land use category (for a description, see Ahlstrøm et al. 2019). In addition, we were interested in identifying whether there were any locations with a particularly high density of images. We thus applied a Kernel density analysis with a 50 km search radius.

Image characteristics

When annotating the images, we defined a list of 18 image attributes to be assessed within each image (Table 1). Attributes were selected to reduce subjectivity during the annotation process. Whether an image attribute was present or not in an image was registered by manual inspection of each image by looking at the image using the URL contained in the metadata (Figs. 1 and 2). The scores were 1 (present) or 0 (absent). To ensure agreement on annotations, we first assessed a number of photos together. Later, if the person responsible for the annotation had any doubts, we met and discussed the doubts specifically to reduce interannotator differences. After inspection, we decided to exclude images representing indoor environments, images taken from the air (drone, airplane), and images solely of the moon. Images taken from the air appeared to demonstrate the technology more than to represent landscape images. For example, many such images showed almost the exact same landscape. Also, the very different perspectives made them less comparable with the other images.

Fig. 1. Example images representing the different image attributes (Photos: Wenche Dramstad, NIBIO, upper row from left to right 2021, 2018, 2021, lower row from left to right 2017, 2018, 2019)

Fig. 2. A photo example illustrating identification of attributes in an example image (Photo: Wenche Dramstad, NIBIO, 2014)

Table 1. Predefined image attributes identified during inspection of the photos

Photograph attributes
Outside	Tree covered	Active farming	Buildings
Visible agriculture	Agricultural background	Living nature	Farm building
Snow covered	Mood or art	Leisure activities	Water
Distance view	Landscape	Objects	Aerial or drone
People	Mountains	–	–

Information from hashtags

To enable analysis of hashtags, we started by combining all tags for all photos into one table. This made it possible to count occurrences of each tag. We decided to continue our analysis of the 100 most frequently occurring tags within each of the four samples of image types: (1) all images, (2) outside agricultural landscape, (3) inside the agricultural landscape, and (4) ‘visible agriculture’. When there were ties in rank, we kept all tags. The tags ranking in the top 100 were then aggregated. Aggregating, for example, field, farm, cornfield, and grass to a category named agriculture enabled us to see the proportion and distribution of tags and to analyse tags at a higher thematic level. An example of an image with different tags is shown in Fig. 3. Based on our research questions, we were particularly interested in the images containing visible agriculture. Thus, we decided to investigate the tags of those images further. To assess the match between the content of the photo and tag (i.e. to what extent images containing visible agriculture or a church were tagged with ‘agriculture’ or ‘church’) we recorded the number of tags in the two categories agriculture and church for each image.

Fig. 3. Example of tag categories and aggregation of tags in the data illustrated on an example photo (Photo: Wenche Dramstad, NIBIO, 2010)

Results

Image sample

The total sample of photos uploaded to Flickr for the five years 2016–2020, geotagged, and falling within the geographical boundary of Norway, was 16,499. They were uploaded by 1982 unique users (IDs). Excluding multiple images taken from the exact same geographical coordinates reduced our sample to 11,575 photos. When we delimited our sample to those taken on agricultural land or within a 100 m buffer surrounding agricultural land, the number of photos left in our sample was 2280. The number of unique IDs was 734. After recording image attributes by manual inspection, the sample was further reduced to 1867 images and the number of unique IDs reduced to 645.

Based on the number of photos uploaded per year, we found a continuing decline in the number of photos uploaded to Flickr in Norway. While a total of 4398 photos had been taken in 2016, the corresponding number for 2020 was only 1908 (i.e. the number had more than halved) (Fig. 4). There was a characteristic pattern to the curve displaying unique user IDs and the number of images they had uploaded. While a large number of users had uploaded either only one image or a few images, a few users had uploaded a large proportion of the total number of images. In our study, looking at the entire country, we found one user ID represented by more than 2000 photos. A total of 18 users had 100 photos or more, while 1525 users were represented by 5 photos or fewer. Within our agricultural landscape sample, the pattern was similar. Among the 734 unique users, 2 were represented by more than 100 photos. The user with the highest number of photos was the same as the user having the highest number of photos in our total sample. That user had uploaded 171 photos taken within the agricultural landscape.

Fig. 4. Number of photographs taken within Norway and uploaded to Flickr between 2016 and 2020 (total 16,499)

Spatial pattern

A first inspection demonstrated that the total sample of unique photo locations gave a good geographical coverage of Norway and that there seemed to be a high number of photos in or near urban areas (Fig. 5, left-hand side). As Norway spans a large number of latitudes, this was not necessarily a given result. From our kernel density analysis, we decided to keep the top 10 density regions (Fig. 5, right-hand side). In addition to the larger cities, such as Oslo, Tromsø, Bergen, Trondheim, and Stavanger, locations with particularly high visitation rates such as Lofoten, Aurlandsfjorden, Geiranger, Arendal, and Ulsteinvik were prominent. The assessment also demonstrated how one user could contribute a high number of images from a particular location. For example, in our sample we had a total of 458 images uploaded by 31 users in total in Arendal. However, one ‘super user’ uploaded 365 of the 458 images (79.7%). By comparison, the highest number of uploaded photos by any user in Aurlandsfjorden was 78, out of a total 404 images (123 users in total).

Fig. 5. Location of the 11,575 unique photo points (left-hand side) and top 10 photo locations, based on kernel density with a 50,000 m search radius and 5000 m cell size (right-hand side)

Image distribution and land cover/land use

When looking at the proportion of photos taken in the different land cover/land use categories, we found that a large proportion of photos was taken in built-up areas (Fig. 6a). The second largest proportion was taken in open areas. When we took the extent of the different land cover types into consideration, most photos still fell into the category ‘built-up areas’, with photos in the ‘infrastructure’ category as the second most frequent (Fig. 6b).

Fig. 6. a) Proportion of photos and area per land cover type in Norway; b) Photos per km² for the different land cover types in Norway (photos apparently taken at sea are excluded from both graphs)

When mapping the distribution of photos by the different categories, we did not see any apparent differences in the geographical distribution of photos in the four categories depicted in Fig. 7. The exceptions are mountainous areas, which lack some locations in central Norway, and agriculture, which lacks some locations in the central and northern Norway. However, both of the aforementioned distributions follow the distribution of the land cover extent. As shown in Fig. 7, images taken ‘on water’, such as along scenic routes, represent a large proportion of the image material.

Fig. 7. Photos taken in the different land cover types (categories): (a) agriculture, (b) forest, (c) mountains, and (d) water

Images’ characteristics

The total area of Norway (excluding Svalbard) is 323,808 ha and the total area of agricultural landscape therein is 29,776 ha. This implies that according to our definition the agricultural landscape covers 9% of Norway. Regarding unique photo locations, we found 2280 photos had been taken within the agricultural landscape and 11,575 taken in total, meaning that 19.7% of the photos had a location within the agricultural landscape.

The two attributes recorded most frequently in the 1867 manually inspected images were ‘landscape’ (1303) and ‘distance view’ (1315) (Table 1). The results further showed that the top two landscape features in the photos were ‘buildings’ (897) and ‘water’ (858). Agriculture was visible in 30.3% of the photos. Of the 1867 photos, we saw buildings in 48.0% and water in 46%.

A close look at the buildings in the photos revealed that 26.2% of all the buildings could be categorized as a building related to agricultural practices. Agricultural buildings in Norway are usually easily recognizable from their shape, size, and colour. Mountains were a common feature in the photos (38.0%) too, also in combination with water. Of photos showing both water and mountains (28.4%), 41.2% also showed agriculture. In total, 11.9% of the 1867 photos showed water, mountains, and visible agriculture.

When categorizing the content of the photos, we assessed that 18.2% were intended to visualize a certain emotion, mood, or art. A similar proportion of photos contained what we categorized as ‘living nature’ (17.3%). For the two categories ‘mood or art’ and ‘living nature’, more than 20% contained also ‘visible agriculture’ (22.9% and 24.8% respectively). We classified 15.0% of the images as visualizing some type of leisure activity. Also, these photos commonly included ‘visible agriculture’ (22.1%). Only 5.3% of all the photos were categorized as ‘tree covered’.

Most of the photographs we inspected manually included large tracts of land and we then categorized them as ‘landscape’ photos (69.8%). Photographs in that category contrasted with photos that typically were a close-up view of a plant, an animal, a monument, or a specific object of interest. Within the landscape photo category, water was the most commonly appearing landscape feature (59.5%), followed by buildings and mountains, both of which occurred in more than half of the photos. We also had several photos that included both water and mountain (39.7%). Agricultural land was seen in 40.3% of the landscape photos, and farming activities (‘active farming’) were visible in 90.3% of them. Of the total number of images showing visible agriculture, 90% showed also active agriculture.

Image hashtags

The most common tag in all groups of photos was ‘Norway’ (Table 2), commonly followed by a slightly more specific location categorized as ‘place’ (e.g. county, city, place name). Within the agricultural landscape, ‘nature’ and ‘other’ were the two next most commonly occurring tags (10.0% and 10.2% respectively). The use of tags related to nature was more frequent for images taken within the agricultural landscape, compared to images taken outside the agricultural landscape. Two further categories of tags for which we found a higher proportion of images taken in the agricultural landscape rather than outside the agricultural landscape were ‘church’ and ‘landscape’. Of all the photos taken within the agricultural landscape, 5.9% were tagged with water-related tags. The group had the highest proportion of water tags. Of the 100 most used tags in images identified as ‘visible agriculture’, the use of tags within the category ‘agriculture’ constituted just 4.6%. The numbers of photos with tags and unique tags for each group are shown in Table 3.

Table 2. Distribution of tags within each group of photos (%)

Category	All photos	Outside agriculture	Inside agriculture	Visible agriculture
Norway	38.5	38.3	35.8	32.9
Place	25.1	25.9	23.0	20.6
Nature	6.8	5.7	10.0	7.7
Agriculture	0.0	0.0	0.8	4.6
Church	0.4	0.0	2.0	3.3
Other	12.3	13.5	10.2	9.5
Weather/season	5.9	5.3	9.0	11.2
Year	2.9	3.1	1.2	1.1
Urban	1.7	2.7	0.0	0.0
Landscape	1.6	1.5	2.1	3.3
Water	4.8	3.9	5.9	5.8
Totals	100.0	100.0	100.0	100.0

Table 3. Numbers of photos with hashtags and unique tags for each group

Photos/Tags	All photos	Outside agriculture	Inside agriculture	Visible agriculture
Photos (no.)	11,576	9295	1867	565
Tagged photos	7057	5676	1158	320
Tags (no.)	60,749	48,278	11,996	2940
Unique tags	12,620	10,672	4056	1318

We were particularly interested in whether photographs we knew contained agricultural areas were tagged with ‘agriculture’. Of the 565 photos classified as containing ‘Visible agriculture’, 323 had one or multiple tags of some form. Of these, 16.1% had one or more agriculture-related tags (e.g. field, farmland, farm building, grass, cornfield, farm, cow). For comparison, we wanted to assess tags related to churches and graveyards, which often are visible elements in agricultural landscapes in Norway. Of the 323 images with tags, 31 images contained identifiable churches or graveyards, as assessed through the manual image inspection. Of these 31 images, 21 had either church-related or graveyard-related tags (67.7%).

Discussion

Spatial patterns and image characteristics

We were a little surprised to find such a broad geographical distribution of Norway in the photos. It is well-established in the literature that more photos are taken in larger cities and tourist hotspots (Cielsielski & Stereńczak 2021) and our results provided further support in this respect. While urban areas cover only 0.9% of the total Norwegian land area, more than one-third (37.4%) of photos were taken in urban landscapes. In addition, tourist hotspots such as Lofoten were prominent (Fig. 5). As can be expected, more photos have been taken in landscapes where people tend to move through. In our results, a typical example is the photos that appear to have been taken at sea, and which on closer inspection were found to have been taken from the more popular cruise ship routes. This may be related to a particularly high interest in the Norwegian landscape among cruise ship travellers, as this form of travel often is advertised as a way to see the Norwegian landscape.

Furthermore, given Norway’s topography and geography, major roads and railways tend to be located in agricultural landscapes, while large parts of the country are more inaccessible. We thus expected a large proportion of the images to have been taken within what we defined as the agricultural landscape. This was proved to be the case, since 19.7% of the photos were taken within agricultural landscapes, which only cover 9% of the total land area. Angradi et al. (2018) used photos uploaded to social media in a study of ecosystem service provision in the Great Lakes area of North America. They reflected on the use of photos as follows:

Although we cannot know the photographer’s exact motivation behind each photograph, we reasoned that the act of taking a photograph reflects the photographer’s individual preference for, or choice of, the depicted subject matter among all the other possible subject matter. (Angradi et al. 2018, 340)

Our study was not designed to explicitly emphasize the provisioning of ecosystem services. However, we realize that many of the landscape elements we saw and analysed in our sample of photos are elements that also play a role in providing cultural ecosystem services. For example, the occurrence of churches in many of the photos can be linked to elements with symbolic meaning, and flowers, animals, and water are clearly linked to enjoyment of the beauty of nature, as described by Haines-Young & Potschin (2018) and discussed by, for example, Havinga et al. (2021).

Water is a landscape element that has long since been identified as attractive to humans (Purcell & Lamb 1998; Yamashita 2002; Nasar& Li 2004). It was thus no surprise that water was a frequently found photo motif in our visual inspection of image contents. In total, 46% of the photos taken from within the agricultural landscape contained water features. The visible water in the photos included water in many different forms, including waterfalls, lakes, rivers, fjords, and ocean views. This could be seen as reflecting how common water as a landscape feature is in Norwegian landscapes.

The small number of photos depicting forest interiors was a surprise. Forests are important as recreation landscapes in Norway and cover c.32.2% of the Norwegian land area. However, only 10.9% of photos were taken in forested areas. This finding may relate to who had uploaded photos. If our analysed photos had mainly been taken by tourists, those tourists might not have been aware of the Norwegian allemannsretten (lit. everyone’s right), which means that the general public have the right to roam on or otherwise access privately owned land, lakes, and rivers for recreational purposes and exercise. Moreover, tourists may not know where there are paths and trails, or they may perceive forested areas as more threatening to be in (e.g. due to not being familiar with the terrain, fear of getting lost, or knowledge of other potential dangers). It may also be seen as more difficult to take good quality photos (i.e. photos that one wants to share) within forested areas, due to light and visibility conditions. Still, our results are in line with those of Wartmann et al. (2021), who hypothesize that forests are considered less photogenic.

Image content and hashtags

Our results clearly documented how people take photos that include agriculture and agricultural landscapes but refrain from tagging them as such. Another interesting result was how tags related to ‘nature’ were more frequently found on photos taken within agricultural landscapes than elsewhere when we analysed the most common tags. This might have been because people experience animals or flowers (i.e. something they would tag as nature) in the agricultural landscape. We suspect, however, that people also may mix nature and agricultural landscapes in Norway. Previous studies of landscape perception in Norway (e.g. Bryn et al. 2013) have indicated that this may be the case. It may also be that people use tags mainly to represent the main focus of their uploaded photo. For example, a photo of a snow-covered mountain on the other side of a lake and with an open agricultural field in the foreground would be tagged with mountain and lake, but not agriculture.

Agriculture in Norway is small-scale and heterogenous, as well as dispersed throughout the entire country. This contributed to our expectation that many photographs would have elements of agricultural activity in them. It would have been very interesting to elucidate whether the photographers were aware of the agricultural landscape, even if merely as a backdrop. It would also be interesting to assess how important agriculture is in providing visual access to other landscape elements in Norway. One such element may be water. We found that ‘water’ was a common tag in agricultural landscapes. We interpret this as a demonstration of the importance of agricultural land in providing visual access to water, in addition to the omnipresence of water in Norwegian landscapes.

One element commonly occurring within the agricultural landscape in addition to water is churches. As a comparison, it was thus interesting to assess the extent of tags that included mentions of churches or similarly relevant tags on photos with a visible church. The results clearly showed a difference in the tagging. When the photos included a church, the majority of those who had uploaded the photos had used ‘church’ as a tag. This was significantly different from those who uploaded photos including, for example, farm buildings. We suspect this may be related to the possibility that the church might have been the primary motif, not agricultural land. Again, the agricultural landscape and agricultural landscape elements constituted the backdrop and were not seen as important enough to be tagged.

The results of our analysis of the visual content of the photos are in line with the findings made by Angradi et al. (2018), who express how interpretation and classification will be partly subjective, even if based on a detailed classification scheme and categorization rules. We did require, for example, that a farm building should be clearly visible and undoubtedly a farm building for it to be classified as such. Still, some classifications were more demanding and to a larger extent best left to judgment. For example, assigning the attribute ‘landscape’ to a photo could present some difficult decisions. Despite our criteria that an extended area should be included in the photo, there were difficult cases. In general, our approach was to be as restrictive as possible and to try and assess what could be the main objective of the photo. For instance, this implied that photos containing only a small area, such as a flowerbed in a garden, a close-up view of a pet, or a water fountain would not be assigned the ‘landscape’ attribute. Nevertheless, the choice of attributes and their definition and use would need further method development.

Using Flickr as a source of data

The representativeness of our photos is an interesting, yet unanswered question. Similar questions have been discussed by other authors conducting analyses using Flickr images. For instance, it has been stated that Flickr users are more like professional photographers than amateur photographers (Tenkanen et al. 2017), tend to be younger, wealthier, and highly educated (da Mota & Pickering 2020), and represent a small proportion of the actual visitors to a given location (da Mota & Pickering 2020). The characteristics may be a rather fitting description of many tourists visiting Norway, despite the proportion of such visitors being unknown. We used our data to assess unique user IDs and found that some users had uploaded a large number of photos. We would not necessarily state that the photos in general appeared more professional, although there are undoubtedly examples of professional photographers who might have been aiming to sell their photos. However, we do suspect the photographers to have been dominated by tourists, rather than typical local inhabitants, as also suggested by Hartmann et al. (2022). This would also affect what the motifs in the photos conveyed. The same applies to accessibility: while mountainous areas in Norway represent a large proportion of the total land area, accessibility to these areas is probably an important reason why they scored so low when we compared the numbers of photos taken in them. The opposite applies to what we categorized as infrastructure, of which a high proportion of photos was taken.

We suspect that where people come from affects how they tag their photos (i.e. the content of the tags). For instance, we consider it unlikely that Norwegians tend to tag their photos with ‘Norway’. A Norwegian photographer may use comparatively more local names as tags, such as the name of a mountain or lake. The content of tags can thus be expected to be a means of communication to the desired viewer group.

Flickr promotes itself as the ‘the best online photo management and sharing application’ (Flickr n.d.). Still, we found a reduction in the number of photos uploaded between 2016 and 2020 in Norway. This might have related to how Flickr changed its policy regarding free uploads in 2019 (Flickr 2018). It is also possible that the onset of the COVID-19 pandemic had an effect on the number of photos taken and uploaded in 2020. We also believe that the majority of the users were foreign residents, and we suspect that Instagram is gaining momentum when it comes to photo sharing. This suggestion is in line with findings by Angradi et al. (2018) who report that Instagram was used more for everyday activities and active recreation than for other reasons, while Flickr and Panoramio had more photos of passive recreation, such as sightseeing. Unfortunately, Instagram photos are no longer accessible for research dependent on the downloading of photos and their accompanying data. If they had been, it would have been possible for us to conduct a study comparing the two platforms. Then, for instance, we might have found more photos taken in forested landscapes, representing either an everyday dog walking route or a jogging path. Another concern that has been raised in the literature regarding Flickr users is gender bias (Angradi et al. 2018). As we do not know the gender ratios among the photographers in our study, we cannot discuss gender further. However, the question of who uses different applications is an interesting topic.

Antoniou et al. (2010) discuss photo sharing as a source of spatial information, and their conclusion is similar to that of Bubalo et al. (2019), who point out that it is difficult to ensure a representative sample selection when using crowdsourced data. However, in common with a statement by Bubalo et al. (2019), we would argue that using photos has large potential when studying landscape perception and preference.

Conclusions

While agricultural land accounts for only a small proportion of the total Norwegian land area, the geographical distribution of agriculture in Norway is wide. Also, agriculture is spatially correlated with other human activities such as travel and settlements. One consequence of this situation is that what can be described as agricultural landscapes contain a wide range of landscape elements and that agriculture is important for landscape visibility. This was clearly apparent in our results. It was also clear that agriculture may not be considered ‘significant’ in the landscape, as represented by tags used on photographs. In fact, more photos containing visible agriculture were tagged as ‘nature’ than as ‘agriculture’.

Compared to agriculture, forest covers an extensive area in Norway and is commonly seen as important for recreation and outdoor life. Despite this, only a small number of photos were taken within forests. This finding might have related to the application analysed in our study but is worth further assessment if possible. Another point worthy of further study is the characteristic of the users and the representativeness of their photos. While we anticipated that the act of taking a photo and uploading it would indicate landscape appreciation, this too could be questioned. Still, we argue that we have documented the importance of agricultural land in experiences of the Norwegian landscape, even though it may not be recognized as such.

Acknowledgements

The project on which this article is based was funded by the Research Council of Norway (grant IDs 194 051 and 342631/L10), and the Norwegian Agriculture Agency (Landbruksdirektoratet) project ‘VERDIBALANSE - Naturmangfold og kulturverdier i kulturlandskapet - Mot en mer balansert og bærekraftig forvaltning’).

Disclosure statement

No potential conflict of interest was reported by the authors.