Understanding human preferences for landscape, and particularly to what degree preferences differ between populations is of importance both from a basic research perspective and from a practical landscape and environmental management perspective. A high consensus in preference across populations would enable general theoretical models of landscape appraisal and justify the application of general models of landscape management across cultures.

Since the mid-1970s, several well-established and oft-cited paradigms in environmental psychology have advocated a universal consensus in human preferences for certain types of natural environments. Such paradigms include the Prospect-Refuge theory (Appleton, 1975), the Savannah Theory (Orians, 1980), the Biophilia Theory (Fromm, 1964; Wilson, 1984; Ulrich, 1993), and The Preference Matrix (Kaplan and Kaplan, 1978, 1982, 1989). At the heart of these theories lies the idea that human adaptions during evolution have led to the development of innate preferences for particular environments; landscapes with physical characteristics that support psychological dimensions like understanding and exploring the environment, and feeling safe in it. Many of these theories make a clear connection between preference and visibility, stating for example that high preference will occur for landscapes where you can see without being seen (Appleton, 1975), landscapes with open expanses with clusters of trees (Orians, 1980), or landscapes which are visually understandable and offer possibilities for exploration (Kaplan and Kaplan, 1989). These hypotheses are one reason why landscape preference research is dominated by work on the visual modality but this practice is further enforced by the convenience of testing with visual stimuli, as compared to the provision of multi-sensoric stimuli which demands a more complex experimental setup.

The vast majority of empirical landscape preference studies also suffer from a sampling bias toward urban Western populations. Although a growing number of studies target the general public or particular stakeholder groups such as residents or recreationists (Soliva et al., 2010; Ode Sang and Tveit, 2013; Junge et al., 2015), university students are overrepresented as sample populations. Some examples of work using students as respondents in this area of work are the research by Strumse on visual preference for agrarian landscapes in Norway (Strumse, 1994, 1996), studies by Herzog (Herzog, 1984, 1989) and others conducted in the context of Kaplan and Kaplan's Landscape Preference Matrix, and many recent studies in China and other Asian countries (Zhao et al., 2013a; Wang et al., 2016). There is a growing awareness within the behavioral sciences that such populations are not necessarily the most representative ones for generalizing about humans. Indeed, they form outliers in a range of fundamental psychological and behavioral domains (Henrich et al., 2010)¹. At least in the context of the more general question of pan-human consensus in preference for particular gross properties of landscapes, the inclusion of much more diverse samples of respondents will be necessary in order to reach reliable conclusions.

The degree of visual openness of the landscape is central to some of the key concepts in the paradigms of environmental psychology cited above (Tveit et al., 2006). In some form or other, degree of openness has consequently been a frequent variable in the empirical studies of landscape preference, and, generally, moderate to high openness has been preferred, at least by the North American, European and East Asian sample populations on which such studies have been typically carried out (Ulrich, 1993). Studies also draw attention to the fact that, in cultures where environments are manipulated by humans for aesthetic pleasure, i.e., parks and gardens, such environments are often half-open landscapes with water features and resemble a savannah-type landscape (Orians, 1980, 1986). Two empirical photo-based studies of different biomes give some support to the idea of an innate preference for savannah over tropical rainforest, desert, temperate deciduous forest and coniferous forest (Balling and Falk, 1982; Falk and Balling, 2010). The results suggest this preference for savannah is present in childhood but then declines with age and experience (as a result of familiarity and enculturation). A third similar study of different biomes (desert, tundra, grassland, coniferous forest, deciduous forest and tropical forest) found tundra and coniferous forest to be the most favored biomes, and hence not supporting the Savannah Theory (Han, 2007). However, theories and studies like those above consistently investigate general preferences for a landscape scene as a whole. Openness is then implicitly recognized as one structural component making up the scene. However, the significance of openness is likely to be a more complex issue when studied in a specific context where openness might have a particular functional importance or be commonly associated with the type of landscape. For instance, in a study focusing on visual scale of agricultural landscapes in Norway, Tveit (2009) found that both degree of open land in the landscape as well as size of landscape rooms were predictors of preference for students in landscape professions but not for the general public. This indicates that expertise or special interests as well as the typology of landscape might influence the preferred degree of openness.

On a basic structural level landscape openness is a product of topography and vegetation, where vegetation in itself plays an important role in the landscape preference studies as its presence or absence often defines what is labeled natural or urban built or human-influenced landscapes in the study design. In comparisons of general preferences for natural vs. built environments studies have consistently found natural environments to be preferred over built ones (Kaplan and Kaplan, 1989). It could be argued, however, that both the context of comparing natural to built environments and the limited range of settings studied are major shortcomings. Studies of variations within a built or a natural scene type are rarer and the majority of studies are focused on everyday environments in urban settings, limiting the available data for more wild or pristine natural environments. To people in industrialized societies the category of “nature” includes not only wilderness but also for instance agricultural land and golf courses (Ulrich, 1993). Studies have shown that European, North American and Japanese respondents tend to think of the environment as natural if it is dominated by features like vegetation, mountains and water as opposed to environments that contain buildings and urban objects like cars and signs (Ulrich, 1993).

The basic research interest in understanding human responses to landscape is only one of the incentives for landscape preference research. Equally important is the need for general models and systematic assessments of human experiences of landscape in environmental management (Daniel, 2001). In planning and management of environments it is important to take into account the opinions and needs of various stakeholders. High consensus and generalizability of results would of course be desirable in practice. Although there is a long-standing debate about the consensus assumption (Van Den Berg et al., 1998) and its evolutionary basis (Joye and Van Den Berg, 2011), not many empirical studies have made a serious attempt at investigating it on the more cross-cultural level.

There are only very few studies that include respondents from non-Western and non-urbanized populations as well as from more non-Western landscape settings. Rare exceptions of photograph-based quantitative preference studies in non-Western contexts are the studies by Falk and Balling (2010) and Sonnenfeld (1967). With the explicit aim of extending their previous research on innate preferences for savannah, mostly based on Western populations, to a different cultural and environmental context, Falk and Balling (2010) used three different samples from River State, Nigeria. One of the samples consisted of students from a technical college who lived spread out in the area of River State (delta and upland area). The other two samples were secondary school children from an area with mature tropical rainforest and from an area with mangrove forest. The majority of the respondents in these two samples had no experience of landscape types other than the one they lived in. As in the previous study (Balling and Falk, 1982), the results showed that savannah was preferred as the place to live over tropical rainforest, desert, temperate deciduous forest, and coniferous forest.

In the study by Sonnenfeld (1967) students and teachers of Inuit populations in three villages in Alaska were compared with various non-indigenous professional groups living and working in the Arctic as well as a sample of students in Delaware. The study revealed differences both between and within indigenous Arctic and non-indigenous Arctic populations. Non-indigenous had higher preference than natives for rugged landscapes and more vegetation. Furthermore, the Delaware students ranked significantly higher on vegetation than the other non-indigenous groups working in the Arctic, but lower on topography. As expected there was also variation between the indigenous Inuit groups, where for instance the inland caribou hunting group ranked highest on topography and vegetation in comparison to the other two Inuit groups who were from villages on the coast. Quantitative studies with indigenous populations like the one by Sonnenfeld are extremely rare. Landscape preference and perception of indigenous populations is to some extent studied within natural resource management, where there has been a particular concern with how landscape changes through management practices for production and conservation have an impact on indigenous people living in affected areas (Lewis and Sheppard, 2005, 2006; Lewis, 2008, 2010). These are however typically context-dependent case studies aimed at understanding the perceptions of a specific area by the local inhabitants.

As the above overview makes clear, there is a lack of quantitative studies of how preferences of general landscape structure and content might be similar or different across a broad sample of geographically, culturally and linguistically diverse populations. The study presented in this paper was designed to test the assumption of universal consensus in preferences for natural landscapes, and particularly for moderate to high openness in the landscape, by including culturally diverse populations from a wide range of geographical settings. Our sample populations are represented by small-scale and lesser-known indigenous communities with a close and largely traditional relationship to their environment. The sample also includes two university student populations (one of which is involved in landscape studies and one which is not) in order for us to compare how more conventional sample populations respond to the stimulus across experts and non-experts. In addition to this sampling of an unprecedented cross-section of humanity, our approach breaks new ground in that it represents an interdisciplinary collaboration between landscape studies and field linguistics. Thus, language experts with established field sites provide practical and communicative access to the diverse sample and enable a culturally and linguistically attuned experimental protocol in each of our sample populations.

We developed an in-situ choice experiment for landscape preference using computer generated visualizations. Within landscape preference research the use of ranking or rating of images on a Likert scale has been the dominant method for establishing preference (Kaplan and Kaplan, 1989). However, it has been suggested that choice experiments more closely correspond to real world behavior and are therefore more suitable for analysing landscape preference (Arnberger and Eder, 2011). The usefulness of choice experiments is likely to be greater when comparing landscapes with large similarities since it forces a ranking between landscapes, which the Likert scale does not. While choice experiments are still rather rare within landscape preference research, they have been more frequently used to gain a monetary evaluation of different type of policies and management strategies related to landscape preference (e.g., Nielsen et al., 2007; Dachary-Bernard and Rambonilaza, 2012; Grammatikopoulou et al., 2012).

First, looking at the effects of the factors of topography and density of vegetation on preference we found a difference between the non-Western sample groups and the two Western (Swedish) groups. For the five non-Western groups — Jahai, Lokono, Makalero, Makasae and Wayuu — there were no significant interactions between topography and density of vegetation (see Figure 3) and therefore it is sufficient to indicate the differences within the main effects of topography and density. In contrast, for the two Swedish samples there were interaction effects between topography and density of vegetation and the interpretation is more complicated (see Figure 4). With a significant interaction it is not possible to separate the analysis into an analysis of the main effects only and for this reason the letters in the figures show a comparison of all nine pictures simultaneously.

Second, examining which images were most preferred, we again note a difference between the non-Western and the Western (Swedish) groups. For the Jahai, Lokono, Makalero, Makasae, and Wayuu, the general pattern is that the most preferred images are those in the bottom left corner of the matrix (see Figures 1, 3), i.e., flatter landscapes with high density of vegetation. The preferences of the two Swedish samples instead show the highest preference for the image in the center of the matrix, (see Figures 1, 4), i.e., the image with mid-level of topography and mid-level of vegetation density. This preference is particularly pronounced in the sample consisting of students of landscape architecture.

Third, considering effects of familiarity on preferences, we once again note a difference between the sample populations, which fall into three groups (see Table 2). For Lokono, Makasae and Jahai preferences coincide to quite a large degree with the landscape in which they grew up; 50–60% of respondents prefer the image also chosen as most resembling the landscape in which they grew up. This is different from the Swedish respondents, who show little tendency (< 10%) to pick the landscape they grew up in as the preferred one. Makalero and Wayuu make up a third group with a slightly higher tendency, around 15%.

The main objective of this study is to test to what degree preferences for natural landscapes differ across human populations, and if a manifestly cross-cultural sample reveals a universal preference for moderate to high openness predicted by theories in the field and previous empirical research. Of particular interest is also the degree to which the responses from the commonly used Western urbanized student-based samples correspond to those of other populations. The study used culturally neutral representations of landscapes which varied only in two gross structural components, topography and density of vegetation. Given the theoretical claims, a high consensus between populations is expected for such stimuli.

The results show that the populations fall into two groups. The preference of the two Western (Swedish) samples, whose preferences were influenced by interactions between topography and vegetation density. On the other hand no such interaction effects were present for the other five populations. For the non-Western, population main effects of topography and vegetation density were found. This indicates that strategies for how judgements are made, at least when using photographically controlled representations of landscape, might differ between Western urbanized schooled respondents and the indigenous populations who still have a close relationship to the land. The Swedish groups displayed interaction effects between topography and vegetation which suggests that the respondents from this group were not separating these variables when they perceived and evaluated the landscape. In contrast, the other groups considered the variables separately in their evaluation and therefore appear to respond more directly to topography and density as separate properties of the landscape.

The result could also reflect that the Swedish student respondents are more affected and tuned to the test situations. This would suggest that they try to understand the parameters of the task and respond accordingly rather than responding and choosing images based on preference.

Furthermore, the pattern of preference for openness differed between the non-Western and the Swedish samples. For the two Swedish samples, their overall highest preference was for the image with medium density and medium variation of topography. This tendency was particularly strong in the students of landscape architecture, who can be considered to be experts trained in landscape evaluation. The students of landscape architecture thus demonstrate the preference for a moderate openness asserted by theory (Appleton, 1975; Orians, 1980; Kaplan and Kaplan, 1989) and previous empirical work (Balling and Falk, 1982; Ulrich, 1993; Tveit, 2009; Falk and Balling, 2010). The five non-Western samples, on the other hand, favored the highest level of vegetation density, i.e., a low degree of openness.

Hence the indigenous samples favored more forested settings. It has been discussed that alternatively to the Savannah Theory humans evolved in closed forested settings and some limited empirical evidence showing preferences for forested biomes have been put forward as support (Han, 2007).

Although both Swedish student samples had the highest preference for the center image of the matrix, i.e., a landscape with medium level of both variables, there was also some difference between the student groups. The landscape architects showed high preference for all the mid topography images, while the humanities students did not select the one image with mid topography and low density of vegetation to the same degree. Another image for which there was a notable difference between the student groups is the landscape with high density of vegetation and high topography, which was preferred by the humanities students but not by the landscape architect students. The humanities students appear to be somewhat more in favor of higher vegetation density than the landscape architect students. Herzog et al. (2000) reported a similar result, with landscape architect students having higher preference for open scene categories than a group called regular students. Also, in a study by Tveit (2009), openness was a predictor of preference for students from landscape oriented fields, but not so for respondents from the general public. However, in the study by Herzog et al. (2000) regular students had the lowest preference for the vegetation category compared to the other student groups (landscape architects and aboriginal students), which is contrary to our finding with humanities students tending to favor more vegetation in the images.

It should be noted that inconsistencies between studies could partly be due to the difference in how the content of the stimuli is controlled and labeled. A factor like openness or vegetation can either be completely controlled with visualizations like in our study, a priori calculated based on coverage in a photograph (Tveit, 2009; Zhao et al., 2013b), based on ratings by judges (Han, 2007) or derived from the respondents' preference ratings (Herzog et al., 2000). High control of stimuli content would be preferable in future research to enable comparisons between studies and replications.

Based on the above possible differences in judgement strategies and in degree of openness favored by the groups one might speculate that the respondents in the two Swedish student groups are also making judgments more guided by the range of stimuli included and tend to opt for the most balanced stimuli in relation to those factors that vary. In other words, we may observe an effect of schooling and familiarity with surveys, which with expert knowledge becomes even more pronounced. The other indigenous population groups generally preferred the lower left hand corner of the matrix, i.e., flatter and more densely vegetated landscapes. The non-significant interaction also shows that the judgements are made more distinctly based on differences in topography and density of vegetation independently.

Familiarity or attachment to the type of landscape where one has grown up is a factor commonly considered as a predictor of landscape preferences. In this study the respondents were asked to choose the one of the nine stimuli that they thought resembled best the place where they grew up, offering us a possibility to test if the topographical variation and vegetation density of their landscape of origin was affecting current preferences. Results again showed that the populations differed in this respect. Some of the non-Western groups (Lokono, Makasae and Jahai) showed quite a strong tendency to prefer their landscape of origin (50–60%, see Table 2), while again the two Swedish samples were clearly different with a low percentage (< 10%) preferring their landscape of origin. This echoes the results of Sonnenfeld (1967), which showed that groups which were free from subsistence concerns had a stronger preference for exotic landscapes. It is possible that the students could be more likely to favor landscapes that are novel and interesting while preferences in some of the indigenous groups, still living in and off the land, could be more guided by experiences of how the environment can be utilized.

Our results bear evidence to clear differences between Western student samples and indigenous communities in the way landscapes are evaluated, what landscapes contain, and the impact of previous experiences. The non-Western groups responded more directly to topography and density of vegetation as specific properties of the landscape when evaluating preference. In contrast, the Western student groups, did not separate the properties out to the same extent but showed interaction effects between topography and vegetation. This could be because this sample might evaluate the landscape more as a whole. Alternatively, it might reflect a greater understanding of the type of survey and the parameters tested and a production of responses accordingly.

The study did show that there were internal similarities between the two Western samples on the one hand, and between the five non-Western samples on the other. To some extent this supports the idea of consensus in preference, not universally but within those categories respectively.

Our study shows that strong universal consensus with preferences for moderate to high openness—asserted by current theory mostly on the basis of studies drawing on Western and/or urbanized sample populations—can be contested if the empirical venture is expanded to include diverse and small-scale indigenous communities living in non-urban environments. Thus, claims of innate preferences, driven by evolution, for semi-open savannah-style environments are not supported by our cross-cultural sample. Our results rather point to strong consensus in preference for densely forested environments for the non-Western samples. However, in order for any generalizations to be made in this regard, additional preference studies from an even broader sample of communities in a wider range of ecologies will be necessary.

The research program of which this study was part was approved by the Regional Ethical Review Board in Lund and determined by the Board not to fall under the Ethical Review Act due to the non-sensitive and non-invasive nature of the research. All subjects in the literate sample populations gave written informed consent in accordance with the Declaration of Helsinki. All subjects in the non-literate sample populations gave informed consent according to a protocol of oral approval developed in full accordance with the standards and guidelines of the European Commission's Seventh Framework Program as well as the CODEX of the Swedish Research Council, and the recommendations of the Code of Conduct of the Volkswagen Foundation's Documentation of Endangered Languages program and the Code of Ethics of the International Society of Ethnobiology.

CH co-desinged the survey in collaboration with ÅOS. Took part in the analysis and intepretation of the results and the writing of the paper. ÅOS co-designed the survey in collaboration with CH. Developed the imagery. Took part in the analysis of the results and the writing of the paper. J-EE developed and carried out the statistical analysis of the survey. Contributed to the writing of method and result section. FA, KR, JH, and NB carried out the survey in field and provided guidance for local adaptation of the questions and intepretation of results for local areas. Contributed to the writing of the paper, particularly regarding description of local groups. NB co-ordinated the project and the carrying out of the survey.