- Email: [email protected]
Predicting the visual impact of onshore wind farms via landscape indices: A method for objectivizing planning and decision processes
Applied Energy xxx (xxxx) xxx–xxx
Contents lists available at ScienceDirect
Applied Energy journal homepage: www.elsevier.com/locate/apenergy
Predicting the visual impact of onshore wind farms via landscape indices: A method for objectivizing planning and decision processes ⁎
Petr Sklenickaa, , Jan Zouhara,b a b
Czech University of Life Sciences Prague, Faculty of Environmental Sciences, Prague, Czech Republic University of Economics, Prague, Department of Econometrics, Faculty of Informatics and Statistics, Prague, Czech Republic
H I G H L I G H T S tested landscape indices to predict the visual impact of onshore wind farms. • We respondents from four countries evaluated images of 32 landscapes. • 400 of 12 landscape indices describing relief and land cover were significant. • 5Weoutpresent method for predicting the visual impact of onshore wind farms. • The methodahelps to objectivize planning and decision processes. •
A R T I C L E I N F O
A B S T R A C T
Keywords: Renewable energy Wind energy Visual assessment Landscape metrics Perception Landscape aesthetics
Visual impact is one of the main factors influencing the acceptance of wind farms by the public and by the authorities. It therefore often sets the environmental and social limits of energy policy and energy use. However, the assessment of visual impacts is subjective, as is often pointed out by critics of the evaluation process. The study presented here for the first time uses accurately and objectively measurable landscape indices to directly predict the visual impact of onshore wind turbines. The method also for the first time evaluates map-based landscape indices in a panoramic simulation, and this provides a better match of visual preferences with landscape indices than the cartographic projection used until now. 400 respondents from four Central European countries (Austria, Germany, Poland and Czechia) provided an evaluation of their scenic perception of 32 different landscapes, in each case with and without wind turbines. At the same time, we analysed 12 indices characterizing the principal landscape components (relief, land cover and landscape pattern) on the basis of the 32 landscape photographs. These were further tested as predictors of visual impact. The most prominent predictors of visual impact were the Percentage of Industrial Area (including Commercial, Logistic and Mining Areas), Percentage of Forest Cover, Density of Technical Infrastructure, Number of Elevation Landmarks, and Elevation Variation. None of the three landscape pattern indices was statistically significant. On the basis of a regression model that is able to predict the potential visual impact in large areas of four Central European countries (over 830,000 km2), we present the general principles of an objectivized method for predicting the visual impact of onshore wind farms. The method makes an automatic assessment of the visual impact in large areas of entire regions or countries via a GIS analysis of Sentinel data and DEM data. This forms a good basis for both preventive evaluation and causal evaluation, and provides significant support for objectivizing the planning and decision process in order to mitigate negative environmental and social impacts of the use of wind energy.
1. Introduction
used visual impact assessments that are limited to the visibility of WTs. They have aimed to define visual thresholds [2] or a maximum visible distance [3], without taking into account the qualities of the impacted landscape. With the easy availability of GIS data and techniques, an assessment of this kind is nowadays not very difficult to make. Methods for assessing the visibility and the qualitative attributes of
The visual impacts of wind turbines (WTs) are usually a decisive element in the decision to reject or permit their construction [1]. A need has therefore arisen to establish a method that can be used for making a visual impact assessment of WTs. Some recent methods have
⁎
Corresponding author at: Czech University of Life Sciences Prague, Faculty of Environmental Sciences, Kamycka 129, Prague 165 21, Czech Republic. E-mail addresses: [email protected] (P. Sklenicka), [email protected] (J. Zouhar).
https://doi.org/10.1016/j.apenergy.2017.11.027 Received 12 July 2017; Received in revised form 29 September 2017; Accepted 2 November 2017 0306-2619/ © 2017 Elsevier Ltd. All rights reserved.
Please cite this article as: Sklenicka, P., Applied Energy (2017), http://dx.doi.org/10.1016/j.apenergy.2017.11.027
Applied Energy xxx (xxxx) xxx–xxx
P. Sklenicka, J. Zouhar
basis for a large-scale and objectivized assessment of landscape visual qualities. The use of accurately measurable landscape indices cannot take into account all landscape factors relevant to scenic perception or aesthetic evaluation. There are aspects, phenomena and landscape features that are not measurable, but that affect the scenic perception of a landscape. To point out just a few examples, we can mention here the distinctiveness of a landscape or, in particular, the ‘genius loci’, which is probably the best-known phenomenon in this sense [22]. In our study, we investigate whether significant landscape indices can adequately explain the variability of the visual impact of WTs. The aim is to evaluate significant predictors of the visual impact of WTs from a set of indices describing landscape relief, landscape cover and landscape pattern. We present a method for objectivized prediction of the impact of onshore wind farms based on these significant variables.
landscapes, and how they are visually impacted by objects such as wind turbines, face issues that are typical for any aesthetic evaluation. These methods may be classified into: (i) expert approaches using subjective evaluations made by appropriately educated and experienced assessors [4,5], (ii) approaches based on landscape classifications, which generalize the impact of WTs for individual landscape types [6–8], (iii) approaches using specific multi-criterion indicators of visual impact [9], including the so-called Spanish Method [10], and (iv) approaches using exactly measurable map-based indicators [11–13]. These methods have until now been used for evaluating the visual qualities of WTs, rather than for making a visual impact assessment of WTs. Assessments of visual impacts for planning or decision-making purposes are highly subjective. The outcome therefore depends to a large extent on the assessor, his/her attitudes, experience and other relevant characteristics, and this is frequently criticized [14]. This drawback is due to the high degree of subjectivity that is inherent in the scenic perception of landscape. Lothian [15] discussed objectivist and subjectivist paradigms as two contrasting views of landscapes. According to Zube [16] and Daniel [17], the aesthetic quality of a landscape is a joint product of particular visual features of the landscape (objective component of the assessment) interacting with relevant psychological – perceptual, cognitive and emotional – processes in the human observer (subjective component). However, the subjective component of aesthetic quality evaluation is often a matter of contention between investors and their opponents in the WT approval process. In the interests of reaching a justifiable conclusion, the subjective element in assessments of landscapes and in the assessment of the influence of WTs on a landscape therefore needs to be reduced as far as possible. In other words, the evaluation process needs to be objectivized. There are several ways to objectivize a visual impact assessment of WTs: (1) consensus among multiple experts, (2) evaluations performed by a recognized authority, (3) evaluations performed by an expert whose competence has been proved by specialized examinations, (4) the use of a rigorous and transparent methodology, (5) a sociological survey by representatives of the public, or (6) an analysis of precisely measurable visible landscape indices followed by a statistical evaluation. A combination of options (5) and (6) forms the principle of the method presented in this paper. Past experience has shown that one of the decisive determinants of the visual impact of WTs on the landscape is the quality of the landscape itself. Studies published so far have highlighted the significance of landscape type [6], perceived naturalness and wildness [7], and landscape aesthetic value [8]. Classifying all these categories, of course, often involves a subjective element. The use of these categories in predicting the impact of WTs on a landscape therefore further increases the subjectivity of the entire evaluation process. However, the use of partial, objectively measurable landscape indices (metrics) can provide a way to limit the subjective element in the evaluation process. This method not only enhances the objectivity of the evaluation and decision making, but also, thanks to the detailed scale of values of the individual indices, enhances the objectivity of the assessment itself. In addition, the method provides repeatability, and therefore makes it easier to audit the results and conclusions. However, insufficient exploration of these evaluation methods remains the principle impediment to their implementation. Landscape indices are most commonly used for landscape assessment in the landscape ecological context [18,19]. However, there is a lack of studies where landscape indices are tested as indicators of visual characteristics or of landscape quality [20]. In this sense, Dramstad et al. [21] made a pioneering study testing the relationships between visual landscape preferences and map-based indicators of landscape pattern. Palmer [11], Svobodova et al. [12] and Frank et al. [13] successfully used spatial metrics to predict the scenic perception of landscapes. All three studies used GIS tools to evaluate map data for predicting the visual perception of landscapes. They therefore provide the
2. Methods The assessment methodology can be divided into five steps: (1) take photographs of the landscape and visualize the WTs, (2) evaluate public visual preferences, (3) analyse the landscape indices, (4) make a statistical analysis of predictors of visual impact, and (5) predict the visual impact of WTs on the landscape. The study areas were located on the territories of four Central European countries – Germany, Austria, Poland and Czechia. In each of these countries, 4 study areas 50 × 50 km in size were defined, all of them in locations where wind farms are currently present (Fig. 1). The study areas were selected with regard to the presence of different types and qualities of the surveyed landscapes, in such a way that the tested indices are represented in a wide range of values.
2.1. Landscape photography and visualization of WTs In each of the 16 study areas, ground landscape photographs without WTs were taken. The photographs were taken between the beginning of June and the end of August 2015, on days with clear weather conditions, using a digital camera with a focal length of 50 mm and a tripod set to a height of approx. 170 cm (an “adult man's eye view”). The photographs were composed in such a way that the sky occupied approximately the upper quarter of the height of the image. Placing landscape features of significant interest according to the rules of the Golden Section or the Rule of Thirds was avoided, as this type of composition may have a significant effect on the observer’s perception [23]. A total of 185 pictures of different landscapes were taken, from which 32 photos of landscapes were selected for the final set. For each of these 32 photographs, Adobe Photoshop was used to digitally add a wind farm with a total of 10 WTs. The WTs were of the Vestas V90 type (hub height 105 m, rotor diameter 90 m). This is one of the most common types set up in Central Europe in recent years. The positions of the blades were rotated differently, in order to obtain a realistic photomontage. The apparent distance of each wind farm from the observer was in all cases from 3 to 4 km, which corresponds to the medium visibility range [8]. The resulting 32 pairs of landscapes (with and without the wind farm, a total of 64 images) were printed in colour with dimensions of 280 × 190 mm. The landscape photos were numbered, and were assigned to 16 sets of 16 photos, using a random number method. Each landscape was represented in 4 sets. Each set consisted of photos of 8 landscapes with and without WTs in order to allow a calculation to be made of the differences in preferences for each landscape, with and without WTs, for each respondent, and thus to obtain the dependent variable – Visual Impact. The sequence of photos in the set was random, the only condition being that photos of the same landscape with and without WTs would not immediately follow each other. 2
Applied Energy xxx (xxxx) xxx–xxx
P. Sklenicka, J. Zouhar
Fig. 1. Locations of the 16 study areas within 4 Central European countries (Austria, Germany, Poland and Czechia).
measurable landscape indices were analysed. We therefore evaluated these indices directly from the photographs, not from maps or orthophotos as in the case of cartographic projection indices [12]. The only exception was the Elevation Variation assessment, which was analysed from a 3D model of the landscape, because it is not feasible to evaluate this parameter from a photograph. We used this method for the first time in our earlier study [24]. Landscape indices were calculated on the basis of an analysis of 32 landscape photos, using Arc-GIS 9.3 software. The Elevation Variation, a parameter describing the relief, was calculated from the Digital Elevation Model (DEM). Each landscape was evaluated using 12 indices in a way that took into account all three basic components: relief, land cover and landscape pattern. A detailed overview of the indices is given in Table 2. The set of tested indices included indices relevant to these three basic components from the point of view of visual impact. In our study, Elevation Variation refers to the height variation of the landscape relief, while Ridge Density characterizes the perceived depth of view. The third relief variable is Elevation Landmarks. All three indices are important determinants of landscape beauty and indicators of visual scale [25,26], and at the same time they affect the range of WT visibility [27]. From the land cover characteristics, we selected indices that, in our opinion and according to the literature, have the greatest effect on the visual perception of landscapes. Three of the six variables representing the land cover component of the landscape (Forests, Water Area and Non-wood Vegetation) are indicators of the naturalness of the landscape. If they are represented to a high degree in a particular landscape, they also indicate higher visual preferences [13,28]. By contrast, the three remaining variables of the land cover component
2.2. Evaluation of visual preferences of the public The evaluation of the visual preferences was performed by a total of 400 respondents aged 15+. There were 100 respondents from each of the 4 countries involved, with 25 respondents from each of the 16 study areas. The respondents were asked to indicate landscape preferences through rating series of 16 colour photographs on a 20-point scale ranging from −10 (dislike very much) to 10 (like very much), according to the perceived attractiveness of each image. In addition, the questionnaire included general questions about the basic characteristics of the respondents, including gender, age, education, income, and the zip code of their current place of residence. The survey was undertaken in situ in 16 study areas, during the months of April 2016 and October 2016. We used the roaming method to contact respondents, i.e., driving and walking through the study areas and asking people to fill in questionnaires. Respondents were evenly approached in various landscape types, in cities and also in the countryside. The respondents were selected in a way aimed at acquiring a balanced sample of respondents from the point of view of sociodemographic characteristics. As is shown in Table 1, these basic sociodemographic characteristics of the group of respondents from each of the countries represented in the study reflected the total populations of the participating countries. The respondents usually spent approx. 15 min completing the questionnaire. 2.3. Analysis of landscape indices In each of the 32 photographs of the evaluated landscapes, exactly 3
Applied Energy xxx (xxxx) xxx–xxx
P. Sklenicka, J. Zouhar
twice among the pictures – with and without WTs. For each observation, we quantified the visual impact of the WTs as Visual Impact = 5 (score without WT – score with WTs). We multiplied the difference in beauty scores by 5 to facilitate interpretation: as the available range of beauty score values was 20 (Likert scale from −10 to 10), the magnitude of the Visual Impact is a measure of the effect of WTs, expressed as a percentage of this range. Mixed-effects linear regression was applied to assess the effect of independent variables on Visual Impact. As repeated observations for both landscapes (photographs) and respondents were available, twoway (crossed) random effects were included in the model specification to account for possible correlations in the disturbances. To express this in a more specific way, for photograph p assessed by respondent r, the Visual Impact was modelled as
Table 1 Comparison of the basic demographic characteristics of the sample of respondents with the entire population in the respective country. The percentages of age groups are calculated for the population aged 15 and over (100%); the urban population parameter denotes the proportion of the total population in the country. Country
Basic demographic profiles Characteristics
Study
Country
Germany
Gender ratio (m/f) Age 15–24 [%] Age 25–54 [%] Age 55–64 [%] Age 65 and over [%] Urban population [%]
1.1 15 61 20 4 69
0.8 12 47 16 25 75
Austria
Gender ratio (m/f) Age 15–24 [%] Age 25–54 [%] Age 55–64 [%] Age 65 and over [%] Urban population [%]
1.0 14 56 19 11 65
0.8 13 50 15 22 66
Poland
Gender ratio (m/f) Age 15–24 [%] Age 25–54 [%] Age 55–64 [%] Age 65 and over [%] Urban population [%]
1.0 14 59 18 9 60
1.0 13 51 17 19 61
Czechia
Gender ratio (m/f) Age 15–24 [%] Age 25–54 [%] Age 55–64 [%] Age 65 and over [%] Urban population [%]
1.2 10 53 21 16 70
1.0 12 51 16 21 73
Visual Impactrp = α + βTxp + γTzr + up + vr + εpr ,
(1)
where xp is a (column) vector of the characteristics of the photograph (landscape indices); zr is a vector of the characteristics of the respondent; up and vr are the photograph-specific random effects and the respondent-specific random effects; εpr is an observation-specific random error; α, β, and γ are the estimated coefficients (bold symbols indicate column vectors). Regressions were estimated via the maximum likelihood (assuming independence and normality of the random effects) in Stata 14 (StataCorp, College Station, TX). As the number of covariates was relatively large, there was a reason for concern about multicollinearity. We therefore obtained the variance inflation factors (VIFs) for all covariates. In the model that included all variables, the maximum VIF was 5.01, a value below the usual threshold of 10 [34], but considered alarming by many [35]. Multicollinearity was not an issue in the case of respondent characteristics, since (i) there was more variation in the respondents than in the photographs (400 respondents versus 32 photographs), and (ii) hardly any correlation was found among the characteristics of the respondents (the VIF factor calculated for respondent characteristics alone was 1.54). To mitigate the risk of false negatives among the photograph characteristics due to multicollinearity, we adopted two different approaches. First, we ran a series of hierarchic regressions, where the landscape characteristics entered the model in a gradual fashion. Indices related to landscape relief (Elevation Variation, Ridge Density, Elevation Landmarks) were included first; next, we added covariates related to the landscape pattern (Patch Density, Landscape Shape Index, Patch Richness); in the final stage, we included the set of indices describing land cover (Forests, Non-wood Vegetation, Water Area, Developed Area, Industrial Area, Infrastructure Density). Second, as a robustness check, we used a model selection procedure
(Developed Area, Industrial Area and Infrastructure Density) represent the rate of anthropogenic influence, which, especially in the case of Industrial Area and Infrastructure Density, usually indicates a significant decrease in visual preferences [29–31]. Patch Density, along with the Landscape Shape Index, describes the spatial heterogeneity of a landscape pattern; the vast majority of relevant studies have confirmed higher preferences for more heterogeneous landscapes [21,32]. This information is further supplemented by the Patch Richness variable, which adds information on the patch type diversity of the landscape, which has also repeatedly been reported directly to determine the visual quality of a landscape [33]. 2.4. The statistical analysis of visual impact predictors As was mentioned above, we obtained beauty scores for 8 landscapes from each of the 400 respondents, with each landscape occurring
Table 2 List of landscape indices (explanatory variables) used in the present study representing three basic landscape components, and their descriptive statistics. Landscape index
Description
Mean
SD
Min
Max
Relief Elevation variation Ridge density Elevation landmarks
S.D. of elevation values [km] Length of visible ridges per unit area [km·km–2] Number of elevation landmarks in a view [No.]
101.1 5.03 0.72
52.8 1.29 0.96
24 3.20 0
245 7.80 3
Landscape pattern Patch density Landscape shape index Patch richness
Number of landscape patches per unit area [No.·km–2] A standardized measure of patch compactness that adjusts for the size of the patch The number of different patch types present within the view [No.]
3.07 3.54 9.06
1.54 0.96 2.35
0.84 2.13 5
6.35 6.17 13
Land cover Forests Non-wood vegetation Water area Developed area Industrial area Infrastructure density
Percentage of forest cover in a view [%] Percentage of non-wood vegetation area in a view [%] Percentage of water area in a view [%] Percentage of developed area in a view [%] Percentage of area of an industrial (or commercial, logistic, mining) character in a view [%] Length of visible features of the technical infrastructure (roads, rails and power lines) per unit area [km·km–2]
28.1 6.47 4.72 9.03 2.59 4.92
20.0 4.91 6.44 10.6 5.59 2.83
0 0 0 0 0 0.60
75 18 29 32 21 12.1
4
Applied Energy xxx (xxxx) xxx–xxx
P. Sklenicka, J. Zouhar
Table 3 Regression analysis of the visual impact of onshore WTs: effects of respondent and landscape characteristics on the difference in beauty scores. Model 1
Landscape indices Relief − Elevation variation − Ridge density − Elevation landmarks Landscape pattern − Patch density − Landscape shape index − Patch richness Land cover − Forests − Non-wood vegetation − Water area − Developed area − Industrial area − Infrastructure density Respondent characteristics Country (Poland = Ref.) − Austria − Czechia − Germany Education (Elementary = Ref.) − Secondary − Tertiary Other demographic char. − Age − Female − High income − Urban residence − 10+ km from WT Adjusted R2 AICc Max. VIF Mean VIF p(Country) p(Education)
Model 2
Model 3
Coef.
SE
Coef.
SE
Coef.
SE
0.166*** 1.819 6.099**
(0.0354) (1.649) (2.253)
0.162*** −0.622 4.948*
(0.0312) (1.750) (2.195)
0.0540† 0.126 4.107*
(0.0326) (1.608) (1.869)
2.622 0.504 0.957
(1.702) (1.689) (1.152)
1.225 −0.387 0.0946
(1.251) (1.224) (0.986)
0.200* −0.267 0.103 −0.0969 −1.217*** −1.146*
(0.0977) (0.420) (0.310) (0.120) (0.267) (0.457)
−1.871*** −0.164 −2.712***
(0.517) (0.523) (0.518)
−1.865*** −0.164 −2.713***
(0.517) (0.523) (0.518)
−1.870*** −0.163 −2.710***
(0.517) (0.523) (0.518)
−0.230 −0.621
(0.508) (0.513)
−0.230 −0.618
(0.508) (0.513)
−0.224 −0.620
(0.507) (0.513)
0.00339 −0.515 0.232 0.246 −0.210
(0.0124) (0.366) (0.422) (0.378) (0.391)
0.00348 −0.518 0.229 0.249 −0.207
(0.0124) (0.367) (0.422) (0.378) (0.391)
0.00335 −0.515 0.228 0.246 −0.212
(0.0124) (0.366) (0.422) (0.378) (0.391)
0.458 24163.094 1.914 1.372 < 0.0001 0.425
0.520 24161.297 3.104 1.623 < 0.0001 0.430
0.628 24147.295 5.007 2.246 < 0.0001 0.424
Notes: (i) R2 was obtained as the squared correlation between actual and fitted values (no random effects have been included in the fitted values); the usual degrees-of-freedom adjustment was applied. (ii) The last two rows show the p-value of a Wald test for the joint significance of the indicated variables. † p < .10. * p < .05. ** p < .01. *** p < .001.
Elevation Variation, Elevation Landmarks, Forests, Industrial Area and Infrastructure Density. The significance of individual variables appears to be stable across all models, with the exception of Elevation Variation: the significance of this variable was marginal (p = .098) in Model 3, where land cover characteristics were included, and the estimated coefficient changed substantially. This is most likely due to multicollinearity issues. Fig. 2 shows the results of the RVI analysis, and tells a story consistent with the main regression analysis: the same variables as were significant in Table 3 are identified as the most important predictors on the basis of the Akaike-weights inspection. The RVI of a variable can be loosely interpreted as the probability that the best model contains this variable; Elevation Variation is the last variable to score above 0.5 for the RVI criterion.
based on the Akaike information criterion with small-sample correction (AICc). Again, we estimated Eq. (1) with different subsets of the landscape characteristics included in the equation; this time, however, we explored all possible subsets, making a total of 212 = 4096 different model specifications. For each model, we calculated the Akaike weight [36], and then we obtained the Relative Variable Importance (RVI) for each landscape index by summing the Akaike weights across all models that included the given index; see [37] for further details on this procedure. 3. Results 3.1. Identifying important predictors The results of the hierarchical mixed-effect regressions are presented in Table 3. Respondents’ characteristics, which entered all models in the same fashion, were largely insignificant, with the sole exception of country of residence. Poland and Czechia assessed the presence of a WT most negatively, while the effect was rated approx. 1.9% less negatively by the Austrian respondents, and approx. 2.7% less negatively by the German respondents, in terms of the grading scale. Five landscape characteristics seem to have had a significant effect:
3.2. Predictive power of the model The five landscape characteristics listed above seem to be the most important predictors of the visual impact of a WT, and could serve as the key variables in a predictive model. The adjusted R2 for a model that contained only the five most important landscape characteristics (along with the respondent characteristics) was 0.619, which is almost 5
Applied Energy xxx (xxxx) xxx–xxx
P. Sklenicka, J. Zouhar
4. Discussion
Industrial area Forests
4.1. Significant predictors of the visual impact of WTs
Elevation landmarks Infrastructure density
The five most significant predictors of the visual impact of WTs on a landscape comprise two indices describing the relief and three indices characterizing the land cover. These five landscape indices explain approximately 80% of the variability of the average visual impact of WTs on a landscape. According to the relative importance of the landscape indices (based on Akaike weights), we can list the five significant predictors in descending order, from the most important to the least important: Industrial Area, Forests, Elevation Landmarks, Infrastructure Density and Elevation Variation. None of the landscape pattern indices was found to be a statistically significant predictor of the visual impact of WTs. In contrast to studies that have demonstrated the effect of visible ridges on the scenic perception of a landscape [25,26], these relief characteristics were not found to be significant in our study. Two other relief-describing indices were however revealed as significant in our study, namely Elevation Landmarks and Elevation Variation. The Elevation Landmarks parameter usually refers to obvious reference points in the landscape used to locate or navigate in unknown terrains (or even to explore Mars [38]). This is exactly the role that provides the key to understanding their meaning in predicting the visual impact of WTs. According to a classical work by the renowned town planner and urban designer Kevin Lynch [39], landmarks rank among the five principal elements of the mental map of an urban environment. However, the role of landmarks is often neglected in present-day efforts to quantify landscape indices affecting scenic perception. WT construction introduces new and visually significant artificial landmarks into a landscape. In most cases, WTs suppress the dominant visual impression of elevation landmarks, such as distinct hills, mountain ridges, river valleys, etc. They often actually assume the role of reference points, replacing the elevation landmarks. This probably explains the important role that this parameter plays in predicting the visual impact of WTs. At the same time, it is also a likely reason why the respondents in our study gave the lowest ratings to WTs in landscapes with distinct elevation landmarks. There are two possible ways to determine elevation landmarks for purposes similar to those in our study. The first utilizes a separate “landmark GIS layer”, prepared on the basis of a field survey (using a subjective element) or, even better, using an automated DEM analysis, e.g. via an algorithm for locating topologically prominent points [40]. Similarly as in the studies by DeLucio and Múgica [25] and by Germino et al. [26], Elevation Variation was confirmed to be a significant predictor of a visual landscape evaluation. In our study, we have also proved that it is a significant predictor of the visual impact of WTs. According to our results, respondents significantly penalized the introduction of WTs into landscapes with a diverse relief. Conversely, did not much mind the construction of WTs in flat landscapes or in landscapes with a low relief. The relief of landscapes accentuates their natural character, and visually enhances the smaller proportions (scale) of the landscape. The introduction of WTs into high-relief landscapes is therefore much more problematic. WTs are more contrasting and more dominant there, and this is probably the reason for the respondents' assessments. The percentage of forest cover in the photograph of the landscape was the second most important predictor of the visual impact of the WTs. The tendency of respondents to reduce the scenic beauty score significantly when evaluating WTs in landscapes with a comparatively high proportion of forests corresponds with the results of earlier studies, which have shown a similar tendency in relatively natural landscapes [13,28]. A higher proportion of forest in a landscape picture basically indicates a greater naturalness, without any necessity to classify landscape types. It thus introduces another subjective element into the assessment. Unlike in the work of Buhyoff et al. [41], who used very
Elevational variation Patch density Water area Developed area Landscape shape index Ridge density Patch richness Non−wood vegetation 0
.2
.4
.6
.8
1
Relative variable importance (sum of Akaike weights)
Fig. 2. Relative importance of the variables (landscape indices), based on Akaike weights.
as much as the most saturated model (Model 3 in Table 3, adj. R2 = 0.628), indicating that little predictive power is sacrificed by focusing on these five characteristics alone. Nevertheless, it should be noted that the R2 reported in Table 3 is not entirely representative of the prediction target: it measures the fraction of the explained variation in the visual impact across the whole sample of 4200 landscape-respondent observations. In prediction endeavours, the goal is not to predict the visual impact on an individual respondent; what we aim to find is the average impact on a particular landscape. For this reason, we decided to obtain a version of predictive R2 in the following manner. For each photograph, we obtained the sample mean for Visual Impact. Then we predicted the Visual Impact for each photograph, using the concept of leave-one-out cross-validation: the prediction for a given photograph came from a model that used data only on the remaining 31 photographs (and only the top five predictors). Finally, we calculated predictive R2 as the squared correlation between the actual values and the predicted values. The resulting value is 0.802, which indicates a fairly good fit, which is also documented in Fig. 3.
45° 60
Predicted visual impact
40
20
0
−20 −20
0
20
40
60
Average visual impact Fig. 3. Predictive power of a model based on the five most important landscape characteristics: actual values vs. leave-one-out predictions for all 32 photographs.
6
Applied Energy xxx (xxxx) xxx–xxx
P. Sklenicka, J. Zouhar
from urban development, etc. Many of these indicators may be significant for predicting the visual impact of WTs, but an analysis would make it impossible to automate the evaluation process, and the assessment of larger landscapes would be time-consuming. Previously published methods evaluate landscape indices in a cartographic projection, i.e., directly from maps, orthophotos, Corine, etc. [21]. However, our method uses indices directly from a photograph for the landscape regression analysis. Similarly, indices that are evaluated from maps or orthophotos along with DEM for the purposes of a subsequent spatial prediction of the visual impact of WTs are obtained from panoramic simulation. This panoramic simulation can be obtained after an appropriate transformation into a panoramic projection in the GIS environment. In our opinion, this approach gives us a better chance of successfully linking visual preferences with measurable landscape indices, because we measure the landscape indices (with the exception of Elevation Variation) in the same panoramic simulation in which the respondents assess the visual preferences. Therefore, only landscape attributes are evaluated that are visible in a given view and, in addition, they are in exactly the same proportions and shapes as during ordinary ground observation. By contrast, if we attempt to link visual preferences with landscape attributes measured directly from maps or from orthophotos [21], there is a risk of bringing otherwise obscured parts of the landscape into the evaluation. In addition, the proportions and shapes observed during ground observation can be different. In order to predict the visual impact on the basis of a regression model calculated from significant predictors, it is necessary to define a network of observer locations (viewpoints) in the area of interest from which the relevant part of the landscape will be evaluated. These viewpoints can be defined in a regular network or, alternatively, irregularly in places or lines that provide good views. The network density must be sufficient to allow for overlaps of individual viewsheds. The area is then transformed from a cartographic projection into a panoramic simulation, as if observed from the individual viewpoints [44], and the visual impact of the WTs is computed using a regression function. As a result, each viewshed is assigned a value, which is derived from the model, corresponding to the percentage reduction of the scenic beauty score after considering the WT structures. Where there are overlaps of individual viewsheds, a lower score is always awarded. The resulting values can be interpolated to create a final map of the predicted visual impact of WTs on the landscape (for example, in the form of isolines of visual impact, Fig. 5). This map of the predicted visual impact of WTs on a landscape forms an objective basis for planning new wind farms. Investors generally welcome similar backgrounds, because they can direct their efforts and finances to areas with a lower probability of reducing the visual quality of the landscape. In such areas, considerably less resistance to the construction of WTs can be expected from citizens and authorities [5,45,46]. Data of this type is also welcomed by municipalities and regional administrations, which can use it as an analytical layer in their land-use planning or as a tool for decision-making for individual investment projects [47]. The method presented here can be used for predicting visual impacts, e.g., for land-use planning purposes, landscape protection schemes, zoning and conservation plans, prospective plans of individual investors (for a preventive assessment of the visual impact). The result of this procedure is a planning background that is useful for directing investment plans into territories with the lowest possible impact on visual landscape values or, alternatively, for preventive delimitation of landscape zones where the construction of WTs will not be permitted due to excessive conflicts with landscape aesthetics. If specific projects of investors are under evaluation (a causal assessment of visual impact), exact parameters, locations and numbers of WTs can be specified in the assessment process. This makes the evaluation more accurate and more objective. The huge contribution of the method is that it can make an automatic assessment of very large areas of regions, countries or even
detailed quantification, we only measured a percentage of the forest in view in our study. The reason for using this simple parameter lies in our effort to define indices that can be further automatically analysed in GIS applications and used for an automated evaluation of large areas. The spatial characteristics of the forests were also taken into account in our study when analysing the landscape pattern indices, but they did not turn out to be significant. The last two significant landscape indices were Percentage of Industrial Area and Density of Technical Infrastructure. Both of these prediction indicators indicate the rate of highly visible and relatively harsh anthropogenic encroachment into a landscape, in ways that are inconsistent with its natural character. Industrial, commercial, logistic, mining, transport or electrification buildings in the landscape have a distinctly technical character. The presence of such elements in a landscape significantly reduces the aesthetic impression of the landscape and reduces its ecosystem services [42]. If elements of this type are already present in the landscape, we can talk about already-initiated disruption of an originally harmonious landscape, or, in other words, about “technization” of the landscape. The respondents were much more forgiving in their assessment of the introduction of WTs into such a landscape, primarily because the WTs would not be the first major breach of the visual balance of the natural and cultural components of the landscape [8]. The only demographic characteristic among the significant predictors of a respondent’s evaluation of the visual impact of WTs was the nationality of the respondents. The German and Austrian respondents were in general more forgiving than their Czech and Polish counterparts. However, the differences between countries were very small, and the results in our opinion rather confirm than disprove a similar perception of citizens throughout Central Europe. Variations in the perception of individual aspects of the implementation of WTs in different countries have previously been described, for example, by Ladenburg and Dubgaard [43]. We believe that these differences are mainly associated with the experience of the population in relation to the current implementation of WTs in their country. WTs are received more critically in countries where WTs are more often likely to be located in an aesthetically valuable landscape. This is especially the case at higher elevations with relatively natural landscapes, and where the wind conditions at low elevations are not suitable for wind farms (e.g., in Czechia). By contrast, Germany is an example of a country where WTs are very often constructed in areas with comparatively low aesthetic qualities, e.g. along motorways, in and around industrial areas, etc. 4.2. A landscape indices-based method for predicting the visual impact of WTs The landscape indices-based method (Fig. 4) uses landscape indices to reduce the subjective element in the evaluation process. This approach has already been used in other works [11,13], but in the context of a general assessment of scenic perception. In this study, we have verified the use of this method for evaluating the visual impact of WTs, i.e., for predicting the potential change in the rating of a landscape after the construction of WTs on it. This type of prediction is crucial when making an environmental impact assessment of WTs. The method consists of the five steps introduced in the Methods section above. The landscape indices used in this method are accurately and objectively measurable variables that can be evaluated on relatively detailed scales, and can therefore increase the accuracy of the evaluation. They represent three basic landscape components (Relief, Land Cover, and Landscape Pattern), without the use of similar indices that are inverted (e.g., Mean Patch Size vs. Patch Density). The method uses only indices that are automatically measurable in a GIS environment, which makes them easily attainable for large regions. Such easy attainability is not available with indicators that depend the use of a subjective assessment by the evaluator. Examples of inappropriate indicators are the existence of a composed landscape, differentiation of rural development 7
Applied Energy xxx (xxxx) xxx–xxx
P. Sklenicka, J. Zouhar
Fig. 4. Scheme of the method for objectivizing the visual impact assessment of onshore wind farms.
based on landscape indices. It reduces the subjective element in the evaluation by directly linking significant, accurately and objectively measurable landscape indices to the visual preferences of the public. This allows the user to skip the step of classifying landscape types, which is a stage in which the application of a subjective element usually cannot be avoided. At the same time, the method makes the evaluation more sophisticated, more accurate and more controllable. The objectivization of this method has been further supported by performing a sociological survey on a representative (demographically stratified) sample of the respondents, each of whom filled in a questionnaire on the impact of WTs on their visual preferences. This further reduces the subjective element in comparison with an expert assessment. The method presented here for the first time ever uses map-based landscape indices in a panoramic simulation to predict the visual impact of WTs. This provides a better match of visual preferences with the analysis of landscape indices than the cartographic projections used until now. The method of objectivized prediction of the impact of onshore wind farms makes it possible to analyse automatically the visual impact of WTs both in small areas of interest and in a large region or a
whole continents. To achieve valuable results, we recommend the use of high spatial resolution Sentinel data together with DEM data for GIS analysis.
5. Conclusions Using an extensive group of respondents from four Central European countries, 5 out of 12 tested landscape indices characterizing relief (Elevation Variation, Elevation Landmarks) and land cover (Forests, Industrial Area and Infrastructure Density) were shown to be significant predictors of the visual impact of onshore wind farms. These five landscape characteristics serve as the key variables in a model that has been computed to predict the visual impact of onshore wind farms in the territory of four Central European countries (Germany, Austria, Poland and Czechia), covering an area of more than 830,000 km2. In a recent form, the model is ready for use in any part of these countries to support the planning and decision process. We have presented a method for objectivizing visual impact assessments of these wind farms. The method generates a prediction 8
Applied Energy xxx (xxxx) xxx–xxx
P. Sklenicka, J. Zouhar
Fig. 5. Isolines of visual impact of onshore wind farms. A detail of the study area in proximity with an important landmark of Mount Rip (459 m above sea level; Czechia) and with the highway D8. The visual impact is expressed as a percentage of the range of beauty score scale, i.e. −100 would mark a landscape with the highest possible beauty score without a wind farm, and the lowest possible score with a wind farm. Positive values would imply that a wind farm makes the landscape more attractive.
9
Applied Energy xxx (xxxx) xxx–xxx
P. Sklenicka, J. Zouhar
whole country, using e.g. Sentinel data products. Our method forms a suitable basis for both preventive and causal forms of visual impact assessments of WTs, and provides significant support for objectivizing the planning and decision process.
landscape indicators: a review. Ecol Ind 2013;28:100–6. [21] Dramstad WE, Tveit MS, Fjellstad WJ, Fry GL. Relationships between visual landscape preferences and map-based indicators of landscape structure. Landscape Urban Plan 2006;78:465–74. [22] Norberg-Schulz C. Genius loci: towards a phenomenology of architecture. Rizzoli; 1980. [23] Svobodova K, Sklenicka P, Molnarova K, Vojar J. Does the composition of landscape photographs affect visual preferences? The rule of the Golden Section and the position of the horizon. J Environ Psychol 2014;38:143–52. [24] Svobodova K, Sklenicka P, Vojar J. How does the representation rate of features in a landscape affect visual preferences? A case study from a post-mining landscape. Int J Mining Reclam Environ 2015;29:266–76. [25] DeLucio J, Múgica M. Landscape preferences and behaviour of visitors to Spanish national parks. Lands Urban Plan 1994;29:145–60. [26] Germino MJ, Reiners WA, Blasko BJ, McLeod D, Bastian CT. Estimating visual properties of Rocky Mountain landscapes using GIS. Landsc Urban Plan 2001;53:71–83. [27] Klouček T, Lagner O, Šímová P. How does data accuracy influence the reliability of digital viewshed models? A case study with wind turbines. Appl Geogr 2015;64:46–54. [28] Van Berkel DB, Verburg PH. Spatial quantification and valuation of cultural ecosystem services in an agricultural landscape. Ecol Ind 2014;37:163–74. [29] Nagle JC. Cell phone towers as visual pollution. Notre Dame J Law Ethics Publ Policy 2009;23:536–68. [30] Blumentrath C, Tveit MS. Visual characteristics of roads: a literature review of people’s perception and Norwegian design practice. Transp Res A-Pol 2014;59:58–71. [31] Zaunbrecher BS, Linzenich A, Ziefle M. A mast is a mast is a mast…? Comparison of preferences for location-scenarios of electricity pylons and wind power plants using conjoint analysis. Energy Policy 2017;105:429–39. [32] Sklenicka P, Lhota T. Landscape heterogeneity—a quantitative criterion for landscape reconstruction. Landsc Urban Plan 2002;58:147–56. [33] de Val GDLF, Atauri JA, de Lucio JV. Relationship between landscape visual attributes and spatial pattern indices: a test study in Mediterranean-climate landscapes. Landsc Urban Plan 2006;77:393–407. [34] Hair Jr JF, Anderson RE, Tatham RL, Black WC. Multivariate Data Analysis (3rd ed). New York: Macmillan; 1995. [35] Rogerson PA. Statistical methods for geography. London: Sage; 2001. [36] Burnham KP, Anderson DR. Model, selection and multimodel inference: A practical information-theoretic approach. Springer Science & Business Media; 2002. [37] Arnold TW. Uninformative parameters and model selection using Akaike’s Information Criterion. J Wildlife Manage 2010;74:1175–8. [38] Olson CF. Selecting landmarks for localization in natural terrain. Auton Robot 2002;12:201–10. [39] Lynch K. The image of the city. MIT Press; 1960. [40] Kweon IS, Kanade T. Extracting topographic terrain features from elevation maps. CVGIP: Image Understand 1994;59:171–82. [41] Buhyoff GJ, Gauthier LJ, Wellman JD. Predicting scenic quality for urban forests using vegetation measurements. Forest Sci 1984;30:71–82. [42] Plieninger T, Bieling C, Ohnesorge B, Schaich H, Schleyer C, Wolff F. Exploring futures of ecosystem services in cultural landscapes through participatory scenario development in the Swabian Alb, Germany. Ecol Soc 2013;18:39. [43] Ladenburg J, Dubgaard A. Willingness to pay for reduced visual disamenities from offshore wind farms in Denmark. Energy Policy 2007;35:4059–71. [44] Bishop ID. Determination of thresholds of visual impact: the case of wind turbines. Environ Plann B 2002;29:707–18. [45] Wolsink M. Wind power and the NIMBY-myth: institutional capacity and the limited significance of public support. Renew Energy 2000;21:49–64. [46] Tsoutsos T, Tsitoura I, Kokologos D, Kalaitzakis K. 2015 Sustainable siting process in large wind farms case study in Crete. Renew Energ 2015;75:474–80. [47] Savino MM, Manzini R, Della Selva V, Accorsi R. A new model for environmental and economic evaluation of renewable energy systems: the case of wind turbines. Appl Energy 2017;189:739–52.
Acknowledgements This study was supported by grant No. LH11069 from the Ministry of Education, Youth and Sports of Czechia. The authors owe special thanks to Robin Healey for his useful advice. References [1] Molnarova K, Sklenicka P, Stiborek J, Svobodova K, Salek M, Brabec E. Visual preferences for wind turbines: location, numbers and respondent characteristics. Appl Energy 2012;92:269–78. [2] Shang H, Bishop ID. Visual thresholds for detection, recognition and visual impact in landscape settings. J Environ Psychol 2000;20:125–40. [3] Wróżyński R, Sojka M, Pyszny K. The application of GIS and 3D graphic software to visual impact assessment of wind turbines. Renew Energy 2016;96:625–35. [4] Höfer T, Sunak Y, Siddique H, Madlener R. Wind farm siting using a spatial analytic hierarchy process approach: a case study of the Städteregion Aachen. Appl Energy 2016;163:222–43. [5] Tsoutsos T, Tsouchlaraki A, Tsiropoulos M, Serpetsidakis M. Visual impact evaluation of a wind park in a Greek Island. Appl Energy 2009;86:546–53. [6] Wolsink M. Planning of renewables schemes: Deliberative and fair decision-making on landscape issues instead of reproachful accusations of non-cooperation. Energy Policy 2007;35:2692–704. [7] Warren CR, Lumsden C, O'Dowd S, Birnie RV. ‘Green on green’: public perceptions of wind power in Scotland and Ireland. J Environ Plan Manag 2005;48:853–75. [8] Betakova V, Vojar J, Sklenicka P. Wind turbines location: how many and how far? Appl Energy 2015;151:23–31. [9] Maslov N, Claramunt C, Wang T, Tang T. Method to estimate the visual impact of an offshore wind farm. Appl Energy 2017. [in press]. [10] Manchado C, Gomez-Jauregui V, Otero C. A review on the Spanish Method of visual impact assessment of wind farms: SPM2. Renew Sust Energ Rev 2015;49:756–67. [11] Palmer JF. Using spatial metrics to predict scenic perception in a changing landscape: Dennis, Massachusetts. Landsc Urban Plan 2004;69:201–18. [12] Svobodova K, Sklenicka P, Molnarova K, Salek M. Visual preferences for physical attributes of mining and post-mining landscapes with respect to the sociodemographic characteristics of respondents. Ecol Eng 2012;43:34–44. [13] Frank S, Fürst C, Koschke L, Witt A, Makeschin F. Assessment of landscape aesthetics—validation of a landscape metrics-based assessment by visual estimation of the scenic beauty. Ecol Ind 2013;32:222–31. [14] Betakova V, Vojar J, Sklenicka P. How education orientation affects attitudes toward wind energy and wind farms: implications for the planning process. Energy Sustain Soc 2016;6:31. [15] Lothian A. Landscape and the philosophy of aesthetics: is landscape quality inherent in the landscape or in the eye of the beholder? Landsc Urban Plan 1999;44:177–98. [16] Zube EH. Cross-disciplinary and intermode agreement on the description and evaluation of landscape. Environ Behav 1974;6:69–89. [17] Daniel TC. Whither scenic beauty? Visual landscape quality assessment in the 21st century. Landsc Urban Plan 2001;54:267–81. [18] Šímová P, Gdulová K. Landscape indices behavior: A review of scale effects. Appl Geogr 2012;34:385–94. [19] Sklenicka P, Šímová P, Hrdinová K, Salek M. Changing rural landscapes along the border of Austria and the Czech Republic between 1952 and 2009: roles of political, socioeconomic and environmental factors. Appl Geogr 2014;47:89–98. [20] Uuemaa E, Mander Ü, Marja R. Trends in the use of landscape spatial metrics as
10
Contents lists available at ScienceDirect
Applied Energy journal homepage: www.elsevier.com/locate/apenergy
Predicting the visual impact of onshore wind farms via landscape indices: A method for objectivizing planning and decision processes ⁎
Petr Sklenickaa, , Jan Zouhara,b a b
Czech University of Life Sciences Prague, Faculty of Environmental Sciences, Prague, Czech Republic University of Economics, Prague, Department of Econometrics, Faculty of Informatics and Statistics, Prague, Czech Republic
H I G H L I G H T S tested landscape indices to predict the visual impact of onshore wind farms. • We respondents from four countries evaluated images of 32 landscapes. • 400 of 12 landscape indices describing relief and land cover were significant. • 5Weoutpresent method for predicting the visual impact of onshore wind farms. • The methodahelps to objectivize planning and decision processes. •
A R T I C L E I N F O
A B S T R A C T
Keywords: Renewable energy Wind energy Visual assessment Landscape metrics Perception Landscape aesthetics
Visual impact is one of the main factors influencing the acceptance of wind farms by the public and by the authorities. It therefore often sets the environmental and social limits of energy policy and energy use. However, the assessment of visual impacts is subjective, as is often pointed out by critics of the evaluation process. The study presented here for the first time uses accurately and objectively measurable landscape indices to directly predict the visual impact of onshore wind turbines. The method also for the first time evaluates map-based landscape indices in a panoramic simulation, and this provides a better match of visual preferences with landscape indices than the cartographic projection used until now. 400 respondents from four Central European countries (Austria, Germany, Poland and Czechia) provided an evaluation of their scenic perception of 32 different landscapes, in each case with and without wind turbines. At the same time, we analysed 12 indices characterizing the principal landscape components (relief, land cover and landscape pattern) on the basis of the 32 landscape photographs. These were further tested as predictors of visual impact. The most prominent predictors of visual impact were the Percentage of Industrial Area (including Commercial, Logistic and Mining Areas), Percentage of Forest Cover, Density of Technical Infrastructure, Number of Elevation Landmarks, and Elevation Variation. None of the three landscape pattern indices was statistically significant. On the basis of a regression model that is able to predict the potential visual impact in large areas of four Central European countries (over 830,000 km2), we present the general principles of an objectivized method for predicting the visual impact of onshore wind farms. The method makes an automatic assessment of the visual impact in large areas of entire regions or countries via a GIS analysis of Sentinel data and DEM data. This forms a good basis for both preventive evaluation and causal evaluation, and provides significant support for objectivizing the planning and decision process in order to mitigate negative environmental and social impacts of the use of wind energy.
1. Introduction
used visual impact assessments that are limited to the visibility of WTs. They have aimed to define visual thresholds [2] or a maximum visible distance [3], without taking into account the qualities of the impacted landscape. With the easy availability of GIS data and techniques, an assessment of this kind is nowadays not very difficult to make. Methods for assessing the visibility and the qualitative attributes of
The visual impacts of wind turbines (WTs) are usually a decisive element in the decision to reject or permit their construction [1]. A need has therefore arisen to establish a method that can be used for making a visual impact assessment of WTs. Some recent methods have
⁎
Corresponding author at: Czech University of Life Sciences Prague, Faculty of Environmental Sciences, Kamycka 129, Prague 165 21, Czech Republic. E-mail addresses: [email protected] (P. Sklenicka), [email protected] (J. Zouhar).
https://doi.org/10.1016/j.apenergy.2017.11.027 Received 12 July 2017; Received in revised form 29 September 2017; Accepted 2 November 2017 0306-2619/ © 2017 Elsevier Ltd. All rights reserved.
Please cite this article as: Sklenicka, P., Applied Energy (2017), http://dx.doi.org/10.1016/j.apenergy.2017.11.027
Applied Energy xxx (xxxx) xxx–xxx
P. Sklenicka, J. Zouhar
basis for a large-scale and objectivized assessment of landscape visual qualities. The use of accurately measurable landscape indices cannot take into account all landscape factors relevant to scenic perception or aesthetic evaluation. There are aspects, phenomena and landscape features that are not measurable, but that affect the scenic perception of a landscape. To point out just a few examples, we can mention here the distinctiveness of a landscape or, in particular, the ‘genius loci’, which is probably the best-known phenomenon in this sense [22]. In our study, we investigate whether significant landscape indices can adequately explain the variability of the visual impact of WTs. The aim is to evaluate significant predictors of the visual impact of WTs from a set of indices describing landscape relief, landscape cover and landscape pattern. We present a method for objectivized prediction of the impact of onshore wind farms based on these significant variables.
landscapes, and how they are visually impacted by objects such as wind turbines, face issues that are typical for any aesthetic evaluation. These methods may be classified into: (i) expert approaches using subjective evaluations made by appropriately educated and experienced assessors [4,5], (ii) approaches based on landscape classifications, which generalize the impact of WTs for individual landscape types [6–8], (iii) approaches using specific multi-criterion indicators of visual impact [9], including the so-called Spanish Method [10], and (iv) approaches using exactly measurable map-based indicators [11–13]. These methods have until now been used for evaluating the visual qualities of WTs, rather than for making a visual impact assessment of WTs. Assessments of visual impacts for planning or decision-making purposes are highly subjective. The outcome therefore depends to a large extent on the assessor, his/her attitudes, experience and other relevant characteristics, and this is frequently criticized [14]. This drawback is due to the high degree of subjectivity that is inherent in the scenic perception of landscape. Lothian [15] discussed objectivist and subjectivist paradigms as two contrasting views of landscapes. According to Zube [16] and Daniel [17], the aesthetic quality of a landscape is a joint product of particular visual features of the landscape (objective component of the assessment) interacting with relevant psychological – perceptual, cognitive and emotional – processes in the human observer (subjective component). However, the subjective component of aesthetic quality evaluation is often a matter of contention between investors and their opponents in the WT approval process. In the interests of reaching a justifiable conclusion, the subjective element in assessments of landscapes and in the assessment of the influence of WTs on a landscape therefore needs to be reduced as far as possible. In other words, the evaluation process needs to be objectivized. There are several ways to objectivize a visual impact assessment of WTs: (1) consensus among multiple experts, (2) evaluations performed by a recognized authority, (3) evaluations performed by an expert whose competence has been proved by specialized examinations, (4) the use of a rigorous and transparent methodology, (5) a sociological survey by representatives of the public, or (6) an analysis of precisely measurable visible landscape indices followed by a statistical evaluation. A combination of options (5) and (6) forms the principle of the method presented in this paper. Past experience has shown that one of the decisive determinants of the visual impact of WTs on the landscape is the quality of the landscape itself. Studies published so far have highlighted the significance of landscape type [6], perceived naturalness and wildness [7], and landscape aesthetic value [8]. Classifying all these categories, of course, often involves a subjective element. The use of these categories in predicting the impact of WTs on a landscape therefore further increases the subjectivity of the entire evaluation process. However, the use of partial, objectively measurable landscape indices (metrics) can provide a way to limit the subjective element in the evaluation process. This method not only enhances the objectivity of the evaluation and decision making, but also, thanks to the detailed scale of values of the individual indices, enhances the objectivity of the assessment itself. In addition, the method provides repeatability, and therefore makes it easier to audit the results and conclusions. However, insufficient exploration of these evaluation methods remains the principle impediment to their implementation. Landscape indices are most commonly used for landscape assessment in the landscape ecological context [18,19]. However, there is a lack of studies where landscape indices are tested as indicators of visual characteristics or of landscape quality [20]. In this sense, Dramstad et al. [21] made a pioneering study testing the relationships between visual landscape preferences and map-based indicators of landscape pattern. Palmer [11], Svobodova et al. [12] and Frank et al. [13] successfully used spatial metrics to predict the scenic perception of landscapes. All three studies used GIS tools to evaluate map data for predicting the visual perception of landscapes. They therefore provide the
2. Methods The assessment methodology can be divided into five steps: (1) take photographs of the landscape and visualize the WTs, (2) evaluate public visual preferences, (3) analyse the landscape indices, (4) make a statistical analysis of predictors of visual impact, and (5) predict the visual impact of WTs on the landscape. The study areas were located on the territories of four Central European countries – Germany, Austria, Poland and Czechia. In each of these countries, 4 study areas 50 × 50 km in size were defined, all of them in locations where wind farms are currently present (Fig. 1). The study areas were selected with regard to the presence of different types and qualities of the surveyed landscapes, in such a way that the tested indices are represented in a wide range of values.
2.1. Landscape photography and visualization of WTs In each of the 16 study areas, ground landscape photographs without WTs were taken. The photographs were taken between the beginning of June and the end of August 2015, on days with clear weather conditions, using a digital camera with a focal length of 50 mm and a tripod set to a height of approx. 170 cm (an “adult man's eye view”). The photographs were composed in such a way that the sky occupied approximately the upper quarter of the height of the image. Placing landscape features of significant interest according to the rules of the Golden Section or the Rule of Thirds was avoided, as this type of composition may have a significant effect on the observer’s perception [23]. A total of 185 pictures of different landscapes were taken, from which 32 photos of landscapes were selected for the final set. For each of these 32 photographs, Adobe Photoshop was used to digitally add a wind farm with a total of 10 WTs. The WTs were of the Vestas V90 type (hub height 105 m, rotor diameter 90 m). This is one of the most common types set up in Central Europe in recent years. The positions of the blades were rotated differently, in order to obtain a realistic photomontage. The apparent distance of each wind farm from the observer was in all cases from 3 to 4 km, which corresponds to the medium visibility range [8]. The resulting 32 pairs of landscapes (with and without the wind farm, a total of 64 images) were printed in colour with dimensions of 280 × 190 mm. The landscape photos were numbered, and were assigned to 16 sets of 16 photos, using a random number method. Each landscape was represented in 4 sets. Each set consisted of photos of 8 landscapes with and without WTs in order to allow a calculation to be made of the differences in preferences for each landscape, with and without WTs, for each respondent, and thus to obtain the dependent variable – Visual Impact. The sequence of photos in the set was random, the only condition being that photos of the same landscape with and without WTs would not immediately follow each other. 2
Applied Energy xxx (xxxx) xxx–xxx
P. Sklenicka, J. Zouhar
Fig. 1. Locations of the 16 study areas within 4 Central European countries (Austria, Germany, Poland and Czechia).
measurable landscape indices were analysed. We therefore evaluated these indices directly from the photographs, not from maps or orthophotos as in the case of cartographic projection indices [12]. The only exception was the Elevation Variation assessment, which was analysed from a 3D model of the landscape, because it is not feasible to evaluate this parameter from a photograph. We used this method for the first time in our earlier study [24]. Landscape indices were calculated on the basis of an analysis of 32 landscape photos, using Arc-GIS 9.3 software. The Elevation Variation, a parameter describing the relief, was calculated from the Digital Elevation Model (DEM). Each landscape was evaluated using 12 indices in a way that took into account all three basic components: relief, land cover and landscape pattern. A detailed overview of the indices is given in Table 2. The set of tested indices included indices relevant to these three basic components from the point of view of visual impact. In our study, Elevation Variation refers to the height variation of the landscape relief, while Ridge Density characterizes the perceived depth of view. The third relief variable is Elevation Landmarks. All three indices are important determinants of landscape beauty and indicators of visual scale [25,26], and at the same time they affect the range of WT visibility [27]. From the land cover characteristics, we selected indices that, in our opinion and according to the literature, have the greatest effect on the visual perception of landscapes. Three of the six variables representing the land cover component of the landscape (Forests, Water Area and Non-wood Vegetation) are indicators of the naturalness of the landscape. If they are represented to a high degree in a particular landscape, they also indicate higher visual preferences [13,28]. By contrast, the three remaining variables of the land cover component
2.2. Evaluation of visual preferences of the public The evaluation of the visual preferences was performed by a total of 400 respondents aged 15+. There were 100 respondents from each of the 4 countries involved, with 25 respondents from each of the 16 study areas. The respondents were asked to indicate landscape preferences through rating series of 16 colour photographs on a 20-point scale ranging from −10 (dislike very much) to 10 (like very much), according to the perceived attractiveness of each image. In addition, the questionnaire included general questions about the basic characteristics of the respondents, including gender, age, education, income, and the zip code of their current place of residence. The survey was undertaken in situ in 16 study areas, during the months of April 2016 and October 2016. We used the roaming method to contact respondents, i.e., driving and walking through the study areas and asking people to fill in questionnaires. Respondents were evenly approached in various landscape types, in cities and also in the countryside. The respondents were selected in a way aimed at acquiring a balanced sample of respondents from the point of view of sociodemographic characteristics. As is shown in Table 1, these basic sociodemographic characteristics of the group of respondents from each of the countries represented in the study reflected the total populations of the participating countries. The respondents usually spent approx. 15 min completing the questionnaire. 2.3. Analysis of landscape indices In each of the 32 photographs of the evaluated landscapes, exactly 3
Applied Energy xxx (xxxx) xxx–xxx
P. Sklenicka, J. Zouhar
twice among the pictures – with and without WTs. For each observation, we quantified the visual impact of the WTs as Visual Impact = 5 (score without WT – score with WTs). We multiplied the difference in beauty scores by 5 to facilitate interpretation: as the available range of beauty score values was 20 (Likert scale from −10 to 10), the magnitude of the Visual Impact is a measure of the effect of WTs, expressed as a percentage of this range. Mixed-effects linear regression was applied to assess the effect of independent variables on Visual Impact. As repeated observations for both landscapes (photographs) and respondents were available, twoway (crossed) random effects were included in the model specification to account for possible correlations in the disturbances. To express this in a more specific way, for photograph p assessed by respondent r, the Visual Impact was modelled as
Table 1 Comparison of the basic demographic characteristics of the sample of respondents with the entire population in the respective country. The percentages of age groups are calculated for the population aged 15 and over (100%); the urban population parameter denotes the proportion of the total population in the country. Country
Basic demographic profiles Characteristics
Study
Country
Germany
Gender ratio (m/f) Age 15–24 [%] Age 25–54 [%] Age 55–64 [%] Age 65 and over [%] Urban population [%]
1.1 15 61 20 4 69
0.8 12 47 16 25 75
Austria
Gender ratio (m/f) Age 15–24 [%] Age 25–54 [%] Age 55–64 [%] Age 65 and over [%] Urban population [%]
1.0 14 56 19 11 65
0.8 13 50 15 22 66
Poland
Gender ratio (m/f) Age 15–24 [%] Age 25–54 [%] Age 55–64 [%] Age 65 and over [%] Urban population [%]
1.0 14 59 18 9 60
1.0 13 51 17 19 61
Czechia
Gender ratio (m/f) Age 15–24 [%] Age 25–54 [%] Age 55–64 [%] Age 65 and over [%] Urban population [%]
1.2 10 53 21 16 70
1.0 12 51 16 21 73
Visual Impactrp = α + βTxp + γTzr + up + vr + εpr ,
(1)
where xp is a (column) vector of the characteristics of the photograph (landscape indices); zr is a vector of the characteristics of the respondent; up and vr are the photograph-specific random effects and the respondent-specific random effects; εpr is an observation-specific random error; α, β, and γ are the estimated coefficients (bold symbols indicate column vectors). Regressions were estimated via the maximum likelihood (assuming independence and normality of the random effects) in Stata 14 (StataCorp, College Station, TX). As the number of covariates was relatively large, there was a reason for concern about multicollinearity. We therefore obtained the variance inflation factors (VIFs) for all covariates. In the model that included all variables, the maximum VIF was 5.01, a value below the usual threshold of 10 [34], but considered alarming by many [35]. Multicollinearity was not an issue in the case of respondent characteristics, since (i) there was more variation in the respondents than in the photographs (400 respondents versus 32 photographs), and (ii) hardly any correlation was found among the characteristics of the respondents (the VIF factor calculated for respondent characteristics alone was 1.54). To mitigate the risk of false negatives among the photograph characteristics due to multicollinearity, we adopted two different approaches. First, we ran a series of hierarchic regressions, where the landscape characteristics entered the model in a gradual fashion. Indices related to landscape relief (Elevation Variation, Ridge Density, Elevation Landmarks) were included first; next, we added covariates related to the landscape pattern (Patch Density, Landscape Shape Index, Patch Richness); in the final stage, we included the set of indices describing land cover (Forests, Non-wood Vegetation, Water Area, Developed Area, Industrial Area, Infrastructure Density). Second, as a robustness check, we used a model selection procedure
(Developed Area, Industrial Area and Infrastructure Density) represent the rate of anthropogenic influence, which, especially in the case of Industrial Area and Infrastructure Density, usually indicates a significant decrease in visual preferences [29–31]. Patch Density, along with the Landscape Shape Index, describes the spatial heterogeneity of a landscape pattern; the vast majority of relevant studies have confirmed higher preferences for more heterogeneous landscapes [21,32]. This information is further supplemented by the Patch Richness variable, which adds information on the patch type diversity of the landscape, which has also repeatedly been reported directly to determine the visual quality of a landscape [33]. 2.4. The statistical analysis of visual impact predictors As was mentioned above, we obtained beauty scores for 8 landscapes from each of the 400 respondents, with each landscape occurring
Table 2 List of landscape indices (explanatory variables) used in the present study representing three basic landscape components, and their descriptive statistics. Landscape index
Description
Mean
SD
Min
Max
Relief Elevation variation Ridge density Elevation landmarks
S.D. of elevation values [km] Length of visible ridges per unit area [km·km–2] Number of elevation landmarks in a view [No.]
101.1 5.03 0.72
52.8 1.29 0.96
24 3.20 0
245 7.80 3
Landscape pattern Patch density Landscape shape index Patch richness
Number of landscape patches per unit area [No.·km–2] A standardized measure of patch compactness that adjusts for the size of the patch The number of different patch types present within the view [No.]
3.07 3.54 9.06
1.54 0.96 2.35
0.84 2.13 5
6.35 6.17 13
Land cover Forests Non-wood vegetation Water area Developed area Industrial area Infrastructure density
Percentage of forest cover in a view [%] Percentage of non-wood vegetation area in a view [%] Percentage of water area in a view [%] Percentage of developed area in a view [%] Percentage of area of an industrial (or commercial, logistic, mining) character in a view [%] Length of visible features of the technical infrastructure (roads, rails and power lines) per unit area [km·km–2]
28.1 6.47 4.72 9.03 2.59 4.92
20.0 4.91 6.44 10.6 5.59 2.83
0 0 0 0 0 0.60
75 18 29 32 21 12.1
4
Applied Energy xxx (xxxx) xxx–xxx
P. Sklenicka, J. Zouhar
Table 3 Regression analysis of the visual impact of onshore WTs: effects of respondent and landscape characteristics on the difference in beauty scores. Model 1
Landscape indices Relief − Elevation variation − Ridge density − Elevation landmarks Landscape pattern − Patch density − Landscape shape index − Patch richness Land cover − Forests − Non-wood vegetation − Water area − Developed area − Industrial area − Infrastructure density Respondent characteristics Country (Poland = Ref.) − Austria − Czechia − Germany Education (Elementary = Ref.) − Secondary − Tertiary Other demographic char. − Age − Female − High income − Urban residence − 10+ km from WT Adjusted R2 AICc Max. VIF Mean VIF p(Country) p(Education)
Model 2
Model 3
Coef.
SE
Coef.
SE
Coef.
SE
0.166*** 1.819 6.099**
(0.0354) (1.649) (2.253)
0.162*** −0.622 4.948*
(0.0312) (1.750) (2.195)
0.0540† 0.126 4.107*
(0.0326) (1.608) (1.869)
2.622 0.504 0.957
(1.702) (1.689) (1.152)
1.225 −0.387 0.0946
(1.251) (1.224) (0.986)
0.200* −0.267 0.103 −0.0969 −1.217*** −1.146*
(0.0977) (0.420) (0.310) (0.120) (0.267) (0.457)
−1.871*** −0.164 −2.712***
(0.517) (0.523) (0.518)
−1.865*** −0.164 −2.713***
(0.517) (0.523) (0.518)
−1.870*** −0.163 −2.710***
(0.517) (0.523) (0.518)
−0.230 −0.621
(0.508) (0.513)
−0.230 −0.618
(0.508) (0.513)
−0.224 −0.620
(0.507) (0.513)
0.00339 −0.515 0.232 0.246 −0.210
(0.0124) (0.366) (0.422) (0.378) (0.391)
0.00348 −0.518 0.229 0.249 −0.207
(0.0124) (0.367) (0.422) (0.378) (0.391)
0.00335 −0.515 0.228 0.246 −0.212
(0.0124) (0.366) (0.422) (0.378) (0.391)
0.458 24163.094 1.914 1.372 < 0.0001 0.425
0.520 24161.297 3.104 1.623 < 0.0001 0.430
0.628 24147.295 5.007 2.246 < 0.0001 0.424
Notes: (i) R2 was obtained as the squared correlation between actual and fitted values (no random effects have been included in the fitted values); the usual degrees-of-freedom adjustment was applied. (ii) The last two rows show the p-value of a Wald test for the joint significance of the indicated variables. † p < .10. * p < .05. ** p < .01. *** p < .001.
Elevation Variation, Elevation Landmarks, Forests, Industrial Area and Infrastructure Density. The significance of individual variables appears to be stable across all models, with the exception of Elevation Variation: the significance of this variable was marginal (p = .098) in Model 3, where land cover characteristics were included, and the estimated coefficient changed substantially. This is most likely due to multicollinearity issues. Fig. 2 shows the results of the RVI analysis, and tells a story consistent with the main regression analysis: the same variables as were significant in Table 3 are identified as the most important predictors on the basis of the Akaike-weights inspection. The RVI of a variable can be loosely interpreted as the probability that the best model contains this variable; Elevation Variation is the last variable to score above 0.5 for the RVI criterion.
based on the Akaike information criterion with small-sample correction (AICc). Again, we estimated Eq. (1) with different subsets of the landscape characteristics included in the equation; this time, however, we explored all possible subsets, making a total of 212 = 4096 different model specifications. For each model, we calculated the Akaike weight [36], and then we obtained the Relative Variable Importance (RVI) for each landscape index by summing the Akaike weights across all models that included the given index; see [37] for further details on this procedure. 3. Results 3.1. Identifying important predictors The results of the hierarchical mixed-effect regressions are presented in Table 3. Respondents’ characteristics, which entered all models in the same fashion, were largely insignificant, with the sole exception of country of residence. Poland and Czechia assessed the presence of a WT most negatively, while the effect was rated approx. 1.9% less negatively by the Austrian respondents, and approx. 2.7% less negatively by the German respondents, in terms of the grading scale. Five landscape characteristics seem to have had a significant effect:
3.2. Predictive power of the model The five landscape characteristics listed above seem to be the most important predictors of the visual impact of a WT, and could serve as the key variables in a predictive model. The adjusted R2 for a model that contained only the five most important landscape characteristics (along with the respondent characteristics) was 0.619, which is almost 5
Applied Energy xxx (xxxx) xxx–xxx
P. Sklenicka, J. Zouhar
4. Discussion
Industrial area Forests
4.1. Significant predictors of the visual impact of WTs
Elevation landmarks Infrastructure density
The five most significant predictors of the visual impact of WTs on a landscape comprise two indices describing the relief and three indices characterizing the land cover. These five landscape indices explain approximately 80% of the variability of the average visual impact of WTs on a landscape. According to the relative importance of the landscape indices (based on Akaike weights), we can list the five significant predictors in descending order, from the most important to the least important: Industrial Area, Forests, Elevation Landmarks, Infrastructure Density and Elevation Variation. None of the landscape pattern indices was found to be a statistically significant predictor of the visual impact of WTs. In contrast to studies that have demonstrated the effect of visible ridges on the scenic perception of a landscape [25,26], these relief characteristics were not found to be significant in our study. Two other relief-describing indices were however revealed as significant in our study, namely Elevation Landmarks and Elevation Variation. The Elevation Landmarks parameter usually refers to obvious reference points in the landscape used to locate or navigate in unknown terrains (or even to explore Mars [38]). This is exactly the role that provides the key to understanding their meaning in predicting the visual impact of WTs. According to a classical work by the renowned town planner and urban designer Kevin Lynch [39], landmarks rank among the five principal elements of the mental map of an urban environment. However, the role of landmarks is often neglected in present-day efforts to quantify landscape indices affecting scenic perception. WT construction introduces new and visually significant artificial landmarks into a landscape. In most cases, WTs suppress the dominant visual impression of elevation landmarks, such as distinct hills, mountain ridges, river valleys, etc. They often actually assume the role of reference points, replacing the elevation landmarks. This probably explains the important role that this parameter plays in predicting the visual impact of WTs. At the same time, it is also a likely reason why the respondents in our study gave the lowest ratings to WTs in landscapes with distinct elevation landmarks. There are two possible ways to determine elevation landmarks for purposes similar to those in our study. The first utilizes a separate “landmark GIS layer”, prepared on the basis of a field survey (using a subjective element) or, even better, using an automated DEM analysis, e.g. via an algorithm for locating topologically prominent points [40]. Similarly as in the studies by DeLucio and Múgica [25] and by Germino et al. [26], Elevation Variation was confirmed to be a significant predictor of a visual landscape evaluation. In our study, we have also proved that it is a significant predictor of the visual impact of WTs. According to our results, respondents significantly penalized the introduction of WTs into landscapes with a diverse relief. Conversely, did not much mind the construction of WTs in flat landscapes or in landscapes with a low relief. The relief of landscapes accentuates their natural character, and visually enhances the smaller proportions (scale) of the landscape. The introduction of WTs into high-relief landscapes is therefore much more problematic. WTs are more contrasting and more dominant there, and this is probably the reason for the respondents' assessments. The percentage of forest cover in the photograph of the landscape was the second most important predictor of the visual impact of the WTs. The tendency of respondents to reduce the scenic beauty score significantly when evaluating WTs in landscapes with a comparatively high proportion of forests corresponds with the results of earlier studies, which have shown a similar tendency in relatively natural landscapes [13,28]. A higher proportion of forest in a landscape picture basically indicates a greater naturalness, without any necessity to classify landscape types. It thus introduces another subjective element into the assessment. Unlike in the work of Buhyoff et al. [41], who used very
Elevational variation Patch density Water area Developed area Landscape shape index Ridge density Patch richness Non−wood vegetation 0
.2
.4
.6
.8
1
Relative variable importance (sum of Akaike weights)
Fig. 2. Relative importance of the variables (landscape indices), based on Akaike weights.
as much as the most saturated model (Model 3 in Table 3, adj. R2 = 0.628), indicating that little predictive power is sacrificed by focusing on these five characteristics alone. Nevertheless, it should be noted that the R2 reported in Table 3 is not entirely representative of the prediction target: it measures the fraction of the explained variation in the visual impact across the whole sample of 4200 landscape-respondent observations. In prediction endeavours, the goal is not to predict the visual impact on an individual respondent; what we aim to find is the average impact on a particular landscape. For this reason, we decided to obtain a version of predictive R2 in the following manner. For each photograph, we obtained the sample mean for Visual Impact. Then we predicted the Visual Impact for each photograph, using the concept of leave-one-out cross-validation: the prediction for a given photograph came from a model that used data only on the remaining 31 photographs (and only the top five predictors). Finally, we calculated predictive R2 as the squared correlation between the actual values and the predicted values. The resulting value is 0.802, which indicates a fairly good fit, which is also documented in Fig. 3.
45° 60
Predicted visual impact
40
20
0
−20 −20
0
20
40
60
Average visual impact Fig. 3. Predictive power of a model based on the five most important landscape characteristics: actual values vs. leave-one-out predictions for all 32 photographs.
6
Applied Energy xxx (xxxx) xxx–xxx
P. Sklenicka, J. Zouhar
from urban development, etc. Many of these indicators may be significant for predicting the visual impact of WTs, but an analysis would make it impossible to automate the evaluation process, and the assessment of larger landscapes would be time-consuming. Previously published methods evaluate landscape indices in a cartographic projection, i.e., directly from maps, orthophotos, Corine, etc. [21]. However, our method uses indices directly from a photograph for the landscape regression analysis. Similarly, indices that are evaluated from maps or orthophotos along with DEM for the purposes of a subsequent spatial prediction of the visual impact of WTs are obtained from panoramic simulation. This panoramic simulation can be obtained after an appropriate transformation into a panoramic projection in the GIS environment. In our opinion, this approach gives us a better chance of successfully linking visual preferences with measurable landscape indices, because we measure the landscape indices (with the exception of Elevation Variation) in the same panoramic simulation in which the respondents assess the visual preferences. Therefore, only landscape attributes are evaluated that are visible in a given view and, in addition, they are in exactly the same proportions and shapes as during ordinary ground observation. By contrast, if we attempt to link visual preferences with landscape attributes measured directly from maps or from orthophotos [21], there is a risk of bringing otherwise obscured parts of the landscape into the evaluation. In addition, the proportions and shapes observed during ground observation can be different. In order to predict the visual impact on the basis of a regression model calculated from significant predictors, it is necessary to define a network of observer locations (viewpoints) in the area of interest from which the relevant part of the landscape will be evaluated. These viewpoints can be defined in a regular network or, alternatively, irregularly in places or lines that provide good views. The network density must be sufficient to allow for overlaps of individual viewsheds. The area is then transformed from a cartographic projection into a panoramic simulation, as if observed from the individual viewpoints [44], and the visual impact of the WTs is computed using a regression function. As a result, each viewshed is assigned a value, which is derived from the model, corresponding to the percentage reduction of the scenic beauty score after considering the WT structures. Where there are overlaps of individual viewsheds, a lower score is always awarded. The resulting values can be interpolated to create a final map of the predicted visual impact of WTs on the landscape (for example, in the form of isolines of visual impact, Fig. 5). This map of the predicted visual impact of WTs on a landscape forms an objective basis for planning new wind farms. Investors generally welcome similar backgrounds, because they can direct their efforts and finances to areas with a lower probability of reducing the visual quality of the landscape. In such areas, considerably less resistance to the construction of WTs can be expected from citizens and authorities [5,45,46]. Data of this type is also welcomed by municipalities and regional administrations, which can use it as an analytical layer in their land-use planning or as a tool for decision-making for individual investment projects [47]. The method presented here can be used for predicting visual impacts, e.g., for land-use planning purposes, landscape protection schemes, zoning and conservation plans, prospective plans of individual investors (for a preventive assessment of the visual impact). The result of this procedure is a planning background that is useful for directing investment plans into territories with the lowest possible impact on visual landscape values or, alternatively, for preventive delimitation of landscape zones where the construction of WTs will not be permitted due to excessive conflicts with landscape aesthetics. If specific projects of investors are under evaluation (a causal assessment of visual impact), exact parameters, locations and numbers of WTs can be specified in the assessment process. This makes the evaluation more accurate and more objective. The huge contribution of the method is that it can make an automatic assessment of very large areas of regions, countries or even
detailed quantification, we only measured a percentage of the forest in view in our study. The reason for using this simple parameter lies in our effort to define indices that can be further automatically analysed in GIS applications and used for an automated evaluation of large areas. The spatial characteristics of the forests were also taken into account in our study when analysing the landscape pattern indices, but they did not turn out to be significant. The last two significant landscape indices were Percentage of Industrial Area and Density of Technical Infrastructure. Both of these prediction indicators indicate the rate of highly visible and relatively harsh anthropogenic encroachment into a landscape, in ways that are inconsistent with its natural character. Industrial, commercial, logistic, mining, transport or electrification buildings in the landscape have a distinctly technical character. The presence of such elements in a landscape significantly reduces the aesthetic impression of the landscape and reduces its ecosystem services [42]. If elements of this type are already present in the landscape, we can talk about already-initiated disruption of an originally harmonious landscape, or, in other words, about “technization” of the landscape. The respondents were much more forgiving in their assessment of the introduction of WTs into such a landscape, primarily because the WTs would not be the first major breach of the visual balance of the natural and cultural components of the landscape [8]. The only demographic characteristic among the significant predictors of a respondent’s evaluation of the visual impact of WTs was the nationality of the respondents. The German and Austrian respondents were in general more forgiving than their Czech and Polish counterparts. However, the differences between countries were very small, and the results in our opinion rather confirm than disprove a similar perception of citizens throughout Central Europe. Variations in the perception of individual aspects of the implementation of WTs in different countries have previously been described, for example, by Ladenburg and Dubgaard [43]. We believe that these differences are mainly associated with the experience of the population in relation to the current implementation of WTs in their country. WTs are received more critically in countries where WTs are more often likely to be located in an aesthetically valuable landscape. This is especially the case at higher elevations with relatively natural landscapes, and where the wind conditions at low elevations are not suitable for wind farms (e.g., in Czechia). By contrast, Germany is an example of a country where WTs are very often constructed in areas with comparatively low aesthetic qualities, e.g. along motorways, in and around industrial areas, etc. 4.2. A landscape indices-based method for predicting the visual impact of WTs The landscape indices-based method (Fig. 4) uses landscape indices to reduce the subjective element in the evaluation process. This approach has already been used in other works [11,13], but in the context of a general assessment of scenic perception. In this study, we have verified the use of this method for evaluating the visual impact of WTs, i.e., for predicting the potential change in the rating of a landscape after the construction of WTs on it. This type of prediction is crucial when making an environmental impact assessment of WTs. The method consists of the five steps introduced in the Methods section above. The landscape indices used in this method are accurately and objectively measurable variables that can be evaluated on relatively detailed scales, and can therefore increase the accuracy of the evaluation. They represent three basic landscape components (Relief, Land Cover, and Landscape Pattern), without the use of similar indices that are inverted (e.g., Mean Patch Size vs. Patch Density). The method uses only indices that are automatically measurable in a GIS environment, which makes them easily attainable for large regions. Such easy attainability is not available with indicators that depend the use of a subjective assessment by the evaluator. Examples of inappropriate indicators are the existence of a composed landscape, differentiation of rural development 7
Applied Energy xxx (xxxx) xxx–xxx
P. Sklenicka, J. Zouhar
Fig. 4. Scheme of the method for objectivizing the visual impact assessment of onshore wind farms.
based on landscape indices. It reduces the subjective element in the evaluation by directly linking significant, accurately and objectively measurable landscape indices to the visual preferences of the public. This allows the user to skip the step of classifying landscape types, which is a stage in which the application of a subjective element usually cannot be avoided. At the same time, the method makes the evaluation more sophisticated, more accurate and more controllable. The objectivization of this method has been further supported by performing a sociological survey on a representative (demographically stratified) sample of the respondents, each of whom filled in a questionnaire on the impact of WTs on their visual preferences. This further reduces the subjective element in comparison with an expert assessment. The method presented here for the first time ever uses map-based landscape indices in a panoramic simulation to predict the visual impact of WTs. This provides a better match of visual preferences with the analysis of landscape indices than the cartographic projections used until now. The method of objectivized prediction of the impact of onshore wind farms makes it possible to analyse automatically the visual impact of WTs both in small areas of interest and in a large region or a
whole continents. To achieve valuable results, we recommend the use of high spatial resolution Sentinel data together with DEM data for GIS analysis.
5. Conclusions Using an extensive group of respondents from four Central European countries, 5 out of 12 tested landscape indices characterizing relief (Elevation Variation, Elevation Landmarks) and land cover (Forests, Industrial Area and Infrastructure Density) were shown to be significant predictors of the visual impact of onshore wind farms. These five landscape characteristics serve as the key variables in a model that has been computed to predict the visual impact of onshore wind farms in the territory of four Central European countries (Germany, Austria, Poland and Czechia), covering an area of more than 830,000 km2. In a recent form, the model is ready for use in any part of these countries to support the planning and decision process. We have presented a method for objectivizing visual impact assessments of these wind farms. The method generates a prediction 8
Applied Energy xxx (xxxx) xxx–xxx
P. Sklenicka, J. Zouhar
Fig. 5. Isolines of visual impact of onshore wind farms. A detail of the study area in proximity with an important landmark of Mount Rip (459 m above sea level; Czechia) and with the highway D8. The visual impact is expressed as a percentage of the range of beauty score scale, i.e. −100 would mark a landscape with the highest possible beauty score without a wind farm, and the lowest possible score with a wind farm. Positive values would imply that a wind farm makes the landscape more attractive.
9
Applied Energy xxx (xxxx) xxx–xxx
P. Sklenicka, J. Zouhar
whole country, using e.g. Sentinel data products. Our method forms a suitable basis for both preventive and causal forms of visual impact assessments of WTs, and provides significant support for objectivizing the planning and decision process.
landscape indicators: a review. Ecol Ind 2013;28:100–6. [21] Dramstad WE, Tveit MS, Fjellstad WJ, Fry GL. Relationships between visual landscape preferences and map-based indicators of landscape structure. Landscape Urban Plan 2006;78:465–74. [22] Norberg-Schulz C. Genius loci: towards a phenomenology of architecture. Rizzoli; 1980. [23] Svobodova K, Sklenicka P, Molnarova K, Vojar J. Does the composition of landscape photographs affect visual preferences? The rule of the Golden Section and the position of the horizon. J Environ Psychol 2014;38:143–52. [24] Svobodova K, Sklenicka P, Vojar J. How does the representation rate of features in a landscape affect visual preferences? A case study from a post-mining landscape. Int J Mining Reclam Environ 2015;29:266–76. [25] DeLucio J, Múgica M. Landscape preferences and behaviour of visitors to Spanish national parks. Lands Urban Plan 1994;29:145–60. [26] Germino MJ, Reiners WA, Blasko BJ, McLeod D, Bastian CT. Estimating visual properties of Rocky Mountain landscapes using GIS. Landsc Urban Plan 2001;53:71–83. [27] Klouček T, Lagner O, Šímová P. How does data accuracy influence the reliability of digital viewshed models? A case study with wind turbines. Appl Geogr 2015;64:46–54. [28] Van Berkel DB, Verburg PH. Spatial quantification and valuation of cultural ecosystem services in an agricultural landscape. Ecol Ind 2014;37:163–74. [29] Nagle JC. Cell phone towers as visual pollution. Notre Dame J Law Ethics Publ Policy 2009;23:536–68. [30] Blumentrath C, Tveit MS. Visual characteristics of roads: a literature review of people’s perception and Norwegian design practice. Transp Res A-Pol 2014;59:58–71. [31] Zaunbrecher BS, Linzenich A, Ziefle M. A mast is a mast is a mast…? Comparison of preferences for location-scenarios of electricity pylons and wind power plants using conjoint analysis. Energy Policy 2017;105:429–39. [32] Sklenicka P, Lhota T. Landscape heterogeneity—a quantitative criterion for landscape reconstruction. Landsc Urban Plan 2002;58:147–56. [33] de Val GDLF, Atauri JA, de Lucio JV. Relationship between landscape visual attributes and spatial pattern indices: a test study in Mediterranean-climate landscapes. Landsc Urban Plan 2006;77:393–407. [34] Hair Jr JF, Anderson RE, Tatham RL, Black WC. Multivariate Data Analysis (3rd ed). New York: Macmillan; 1995. [35] Rogerson PA. Statistical methods for geography. London: Sage; 2001. [36] Burnham KP, Anderson DR. Model, selection and multimodel inference: A practical information-theoretic approach. Springer Science & Business Media; 2002. [37] Arnold TW. Uninformative parameters and model selection using Akaike’s Information Criterion. J Wildlife Manage 2010;74:1175–8. [38] Olson CF. Selecting landmarks for localization in natural terrain. Auton Robot 2002;12:201–10. [39] Lynch K. The image of the city. MIT Press; 1960. [40] Kweon IS, Kanade T. Extracting topographic terrain features from elevation maps. CVGIP: Image Understand 1994;59:171–82. [41] Buhyoff GJ, Gauthier LJ, Wellman JD. Predicting scenic quality for urban forests using vegetation measurements. Forest Sci 1984;30:71–82. [42] Plieninger T, Bieling C, Ohnesorge B, Schaich H, Schleyer C, Wolff F. Exploring futures of ecosystem services in cultural landscapes through participatory scenario development in the Swabian Alb, Germany. Ecol Soc 2013;18:39. [43] Ladenburg J, Dubgaard A. Willingness to pay for reduced visual disamenities from offshore wind farms in Denmark. Energy Policy 2007;35:4059–71. [44] Bishop ID. Determination of thresholds of visual impact: the case of wind turbines. Environ Plann B 2002;29:707–18. [45] Wolsink M. Wind power and the NIMBY-myth: institutional capacity and the limited significance of public support. Renew Energy 2000;21:49–64. [46] Tsoutsos T, Tsitoura I, Kokologos D, Kalaitzakis K. 2015 Sustainable siting process in large wind farms case study in Crete. Renew Energ 2015;75:474–80. [47] Savino MM, Manzini R, Della Selva V, Accorsi R. A new model for environmental and economic evaluation of renewable energy systems: the case of wind turbines. Appl Energy 2017;189:739–52.
Acknowledgements This study was supported by grant No. LH11069 from the Ministry of Education, Youth and Sports of Czechia. The authors owe special thanks to Robin Healey for his useful advice. References [1] Molnarova K, Sklenicka P, Stiborek J, Svobodova K, Salek M, Brabec E. Visual preferences for wind turbines: location, numbers and respondent characteristics. Appl Energy 2012;92:269–78. [2] Shang H, Bishop ID. Visual thresholds for detection, recognition and visual impact in landscape settings. J Environ Psychol 2000;20:125–40. [3] Wróżyński R, Sojka M, Pyszny K. The application of GIS and 3D graphic software to visual impact assessment of wind turbines. Renew Energy 2016;96:625–35. [4] Höfer T, Sunak Y, Siddique H, Madlener R. Wind farm siting using a spatial analytic hierarchy process approach: a case study of the Städteregion Aachen. Appl Energy 2016;163:222–43. [5] Tsoutsos T, Tsouchlaraki A, Tsiropoulos M, Serpetsidakis M. Visual impact evaluation of a wind park in a Greek Island. Appl Energy 2009;86:546–53. [6] Wolsink M. Planning of renewables schemes: Deliberative and fair decision-making on landscape issues instead of reproachful accusations of non-cooperation. Energy Policy 2007;35:2692–704. [7] Warren CR, Lumsden C, O'Dowd S, Birnie RV. ‘Green on green’: public perceptions of wind power in Scotland and Ireland. J Environ Plan Manag 2005;48:853–75. [8] Betakova V, Vojar J, Sklenicka P. Wind turbines location: how many and how far? Appl Energy 2015;151:23–31. [9] Maslov N, Claramunt C, Wang T, Tang T. Method to estimate the visual impact of an offshore wind farm. Appl Energy 2017. [in press]. [10] Manchado C, Gomez-Jauregui V, Otero C. A review on the Spanish Method of visual impact assessment of wind farms: SPM2. Renew Sust Energ Rev 2015;49:756–67. [11] Palmer JF. Using spatial metrics to predict scenic perception in a changing landscape: Dennis, Massachusetts. Landsc Urban Plan 2004;69:201–18. [12] Svobodova K, Sklenicka P, Molnarova K, Salek M. Visual preferences for physical attributes of mining and post-mining landscapes with respect to the sociodemographic characteristics of respondents. Ecol Eng 2012;43:34–44. [13] Frank S, Fürst C, Koschke L, Witt A, Makeschin F. Assessment of landscape aesthetics—validation of a landscape metrics-based assessment by visual estimation of the scenic beauty. Ecol Ind 2013;32:222–31. [14] Betakova V, Vojar J, Sklenicka P. How education orientation affects attitudes toward wind energy and wind farms: implications for the planning process. Energy Sustain Soc 2016;6:31. [15] Lothian A. Landscape and the philosophy of aesthetics: is landscape quality inherent in the landscape or in the eye of the beholder? Landsc Urban Plan 1999;44:177–98. [16] Zube EH. Cross-disciplinary and intermode agreement on the description and evaluation of landscape. Environ Behav 1974;6:69–89. [17] Daniel TC. Whither scenic beauty? Visual landscape quality assessment in the 21st century. Landsc Urban Plan 2001;54:267–81. [18] Šímová P, Gdulová K. Landscape indices behavior: A review of scale effects. Appl Geogr 2012;34:385–94. [19] Sklenicka P, Šímová P, Hrdinová K, Salek M. Changing rural landscapes along the border of Austria and the Czech Republic between 1952 and 2009: roles of political, socioeconomic and environmental factors. Appl Geogr 2014;47:89–98. [20] Uuemaa E, Mander Ü, Marja R. Trends in the use of landscape spatial metrics as
10