Landscape Evaluation

Introduction

Definitions in landscape evaluation

Subjectivity versus objectivity

Landscape evaluation methods

Descriptive inventories

Formal aesthetic models

Ecological models

Examples in brief

Public preference models

Psychological models

Phenomenological models

Consensus

Examples in brief

Quantitative holistic methods

Psychophysical models

Surrogate component models

Visual Management Systems

Examples in brief

Methodological problems and error in models

Economics of landscape evaluation

Methods of scenic resource financial valuation

Contingent Valuation Technique and Willingness-to-Pay

Landscape components

The theory and problems behind landscape characteristics

Terrain and hydrological components

Indices of attractiveness

Landform elements

Landscape themes

Landscape qualities

Landscape dimensions and preference predictors

Use of models to predict preference

Theoretical models and some results from them

Abello, Bernaldez and Galiano's model

Calvin, Dearinger and Curtin's model

Patsfall, Fiemer, Buhyoff and Wellman's model

Visual Resource Management System (VRMS)

Bishop and Hulse's model

Steinitz's model and experiment

Explanation of Visual Assessment (EVA)

Policy capturing

Hammitt, Patterson and Noe's model

The Shafer model

Summary of landscape components used in predictive models

Summary of explanatory results of the predictive models

References

Introduction

Landscape should be recognised as a resource and therefore a variable to be considered in land use decisions (Dearden, 1985). When evaluating landscapes one should use an interdisciplinary approach, should communicate with other evaluators and, importantly, recognise the academic respectability of the elementary (Appleton, 1975).

"Just as the Brisbane wicket after rain used to be said to reduce all batsmen to an equal level of incompetence, so this absence of aesthetic theory brings the professional down to the same plane as the man in the street. It is true that the theory underlying the judging of a fine wine or a good piece of sculpture is probably as obscure as that which underlies the evaluation of landscape. In those arts, however, we still have some faith - possibly misplaced - in the ability of the expert to recognise excellence, however defined (Appleton, 1975). "

A structured method of landscape assessment, linking description, classification, analysis and evaluation, will provide an integrated framework within which land use management and advice decisions can be debated (Cooper and Murray, 1992). One of the biggest problems in developing quantitative assessment methods for scenic impacts is that of measuring the contributions of specific landscape elements to overall preference (Buhyoff and Riesenmann, 1979).

Unwin (1975) describes three phases of landscape evaluation. First there must be landscape measurement, an inventory of what actually exists in the landscape. Secondly, the landscape value should be measured, this will be an investigation and measurement of value judgements or preferences in the visual landscape. Finally, there is the landscape evaluation, an assessment of the quality of the objective visual landscape in terms of individual or societal preferences for different landscape types (Unwin, 1975).

Definitions in landscape evaluation

Before the subject of landscape evaluation can be reviewed, it is necessary to define several key words. In my opinion the term "total landscape" refers to the less tangible properties of the landscape as well as the more obvious visual properties and should not be confused with "landscape" which refers to the visual properties only. Unfortunately, this differentiation is not apparent in the literature.

Landscape

Hull and Revell (1989) define landscape and scenes as:

" The outdoor environment, natural or built, which can be directly perceived by a person visiting and using that environment. A scene is the subset of a landscape which is viewed from one location (vantage point) looking in one direction ..."

The term landscape clearly focuses upon the visual properties or characteristics of the environment, these include natural and man-made elements and physical and biological resources which could be identified visually; thus non-visual biological functions, cultural/historical values, wildlife and endangered species, wilderness value, opportunities for recreation activities and a large array of tastes, smells and feelings are not included (Daniel and Vining, 1983; Amir and Gidalizon, 1990).

Landscape quality

Often landscape quality is defined as including a wide range of environmental / ecological, sociocultural and psychological factors. According to Jacques (1980) the distinction between `value' and `quality' is meaningless, since both terms refer to the comparison of the landscape in front of your eyes to an idealised landscape in one's mind.

Visual impact

Visual impact on landscape quality is concerned with physical changes introduced to a site by a new development activity (Amir and Gidalizon, 1990).

Objective definitions

Visual quality - a phrase synonymous with beauty, but intended to convey an impression of objectivity; landscape evaluation - ascertaining of a single, often numerical, measure of visual quality, more appropriately would be " landscape quality survey"; judgement - the presumed ability by the design professions to evaluate `visual quality', as distinct from value (Jacques, 1980).

Subjective definitions

Landscape value - a personal and subjective assessment of aesthetic satisfaction derived from a landscape type; landscape appraisal - the study of the effect of landscape changes upon landscape value; preference - the liking of one landscape type better than another (Jacques, 1980).

Subjectivity versus objectivity

There is a fundamental theoretical divergence of opinion over the question of whether landscapes have an intrinsic or objective beauty which may in some way be measurable or comparable, or whether scenic beauty is a value that can only be subjectively attributed to an area or specific landscape (Shuttleworth, 1980b). While physical geographers have devised ways of measuring landscape parameters to reflect visual quality; human geographers have probed individual and societal attitudes toward landscape (Dearden, 1985).

Orland et al (1995) have described qualitative approaches as those which focus upon evaluating the complexity of landscape using the judgements of panels of human subjects, and quantitative approaches as those which measure physical characteristics of the visual field directly.

On the physical/objective side, Buhyoff and Riesenmann (1979) have presented evidence that certain landscape dimensions can be used successfully to prepare an evaluation, and that aesthetic impact can be measured from specific landscape dimensions. There is an increasing interest in the use of mapped data and geographic information systems (GISs) to assess visual landscape variables using reproducible methods over wide area (Bishop and Hulse, 1994). Recent research efforts have shown that the public's scenic preferences can be assessed objectively and quantitatively (Dearden, 1980). This research has also demonstrated that public perceptions can be related to and, in fact, predicted from environmental attributes of a more tangible nature (Buhyoff et al, 1994).

The assessment and/or quantification of scenic quality is mandatory for proper consideration of the aesthetic consequence of management actions (Buhyoff et al, 1994). The Belgian experience with landscape evaluation, especially in rural reallotment projects, indicates, and international literature from a great number of disciplines or research field confirms, the necessity to speak of scenic or visual resource management (Tips, 1984).

Landscape evaluation methods

Numerous techniques of landscape evaluation have been devised in recent years (Crofts and Cooke, 1974). They form a spectrum in which the extremes are represented on the one hand by techniques based unequivocally on the subjective assessments of landscape quality by individuals or groups (e.g. Shafer et al, 1969) and on the other by techniques using physical attributes of landscape as surrogates for personal perception (e.g. Linton, 1968; Land Use Consultants, 1971).

The various models can be subdivided several ways. Arthur et al (1977) splits them into descriptive inventories and public preference models, both categories being further split into nonquantitative and quantitative methods. Briggs and France (1980) use direct and indirect methods to subdivide the models, but state that the former has not been used to any great extent, while indirect methods are further classed into classificatory and nonclassificatory methods. Crofts (1975) describe two sorts of techniques - preference and surrogate component techniques. Daniel and Vining (1983) split the methods into ecological, formal aesthetic, psychophysical, psychological and phenomenological models. For this review the methods will be split into descriptive inventories, public preference methods (after Arthur et al, 1977) and a third category of quantitative holistic techniques.

Descriptive inventories include ecological and formal aesthetic models, methods which are mostly applied by experts in an objective manner. Public preference models, such as psychological and phenomenological, are often undertaken using questionnaires, and are unavoidably linked to the problems of consensus among the public. Quantitative holistic techniques use a mixture of subjective and objective methods and include psychophysical and surrogate component models.

It is important to examine the reliability and validity of landscape evaluation models and to identify any assumptions central to the models. Internal and external validity are of concern in the development of any landscape visual assessment system. External validity reflects, in part, how well the system-generated assessments correspond to other, known measures of visual quality. Internal validity reflects how well the system's internal logic withstands testing and violation of assumptions (Buhyoff et al, 1995).

Descriptive inventories

Descriptive inventories comprise the largest category of techniques for assessing scenic resources; they include both quantitative and qualitative methods of evaluating landscapes by analysing and describing their components (Arthur et al, 1977). Descriptive inventories can be divided into two methods - classificatory and non-classificatory methods.Classificatory methods are those which first attempt to classify the survey units on the basis of their overall similarity, and then to grade or evaluate the resulting clusters - formal aesthetic models are an example of this method. Non-classificatory methods, such as ecological models, attempt to identify the relationships between selected landscape components and environmental quality, then use these relationships to predict landscape quality (Briggs and France, 1980).

Scenic elements (such as landform and visual effects), vegetative patterns and so forth, are first identified then described or rated (Arthur et al, 1977). The ratings are primarily based on traditional values within the landscape architecture profession (Palmer, 1983). Although these surrogate methods' of landscape evaluation can provide general assessments of landscape quality and a landscape inventory based on subjectively-selected but objectively-applied criteria, the objectivity of their application, and their precise, often quantitative, results disguise their underlying subjectivity (Crofts and Cooke, 1974).

The descriptive inventory approach contains several assumptions. One is that the value of a landscape can be explained in terms of the values of its components. Another is that scenic beauty is embedded in the landscape components, that it is a physical attribute of the landscape; however, scenic beauty depends on the observer as well as that which is being observed (Arthur et al, 1977).

Formal aesthetic models

The basic theory of the formal aesthetic model is that aesthetic values are inherent in the abstract features of the landscape i.e. aesthetic quality resides in the formal properties of the landscape. These properties are defined as basic forms, lines, colours and textures and their interrelationships (Daniel and Vining, 1983). In this model landscapes are first analysed into their formal abstract properties. The relationships between these elements are then inspected to classify each area in terms of variety, unity, integrity or other complex formal characteristics. Due to the formal training required for this, the method is almost always applied by an expert, usually a landscape architect (Daniel and Vining, 1983).

Because the landscape-quality assessment results in ordered categories, and not in cardinal or interval measures, it is difficult to relate these assessments to economic or trade-off types of valuation processes. Thus, valuing landscape quality relative to other social values is rather restricted (Daniel and Vining, 1983). The models have been found to be seriously deficient with regard to the fundamental criteria of sensitivity and reliability (Daniel and Vining, 1983).

Ecological models

Within the ecological model, the environmental features that are relevant to landscape quality are primarily biological or ecological. The landscape is characterised in terms of species of plants and animals present, ecological zones, successional stage or other indicators of ecological processes. Humans are characterised as users of the landscape, their contribution is in the form of negative aesthetic impacts (Daniel and Vining, 1983).

Ecological models tend to be designed for specific areas and are therefore difficult to apply to landscapes in general; they are also more sensitive in distinguishing between natural and human-influenced environments than in making distinctions within either of those classes. If the alternatives for land management are to manipulate or not manipulate the environment, the ecological models will almost invariably indicate against any manipulation (Daniel and Vining, 1983).

A major underlying assumption of the ecological model is that landscape quality is directly related to naturalness, or ecosystem integrity. The validity of this model depends upon the assumption that "natural" areas undisturbed by humans are highest in landscape quality. Reliability depends on the consistency and accuracy of the individual applying the method as the assessments are usually carried out by an "ecological expert" (Daniel and Vining, 1983).

Examples in brief

An example of a formal aesthetic model is the Visual Management System (VMS) developed by the USDA Forest Service. It has the purpose of evaluating scenic resources within a land-management framework and assumes that scenic quality is directly related to landscape diversity or variety (Daniel and Vining, 1983). VMS uses character classification (such as gorges, mountains, foothills and plateaus), variety classification (form, line, colour and texture) and sensitivity level (referring to the relative importance of the landscape as a visual or recreational resource). In some cases VMSs are more quantitative holistic techniques that descriptive inventories.

Leopold's "uniqueness ratio" illustrates a landscape assessment methodology based primarily on ecological measures of the landscape.The uniqueness of a given landscape is defined by multiple physical, biological, and human-use dimensions that reflect the implicit assumption that aesthetic value is primarily a function of ecological criteria (Daniel and Vining, 1983).

Cooper and Murray (1992) used local patterns of land class distribution and land class clusters to divide a region into geographically distinct landscape units. High-elevation and upland areas were differentiated from lowlands and unit boundaries were then drawn in relation to selected physiographic and landform features such as watershed boundaries and specified juxtapositions of land classes (Cooper and Murray, 1992).

Public preference models

The recent upsurge in public interest in preserving the beauty of public lands has resulted in development of scenic assessment based on public input (Arthur et al, 1977), indeed, it can be argued logically that the best source of data upon such a subjective issue as landscape quality is the general public. Although planners may claim that it is their duty to guide public taste in these matters, the visual attractiveness of the landscape is ultimately a product of the aggregated opinions of all the individuals concerned with that landscape (Briggs and France, 1980).

The visual quality (or value) of a landscape is rated on the basis of an observer's individual preference of the whole landscape. Those techniques that are based on subjective assessments of scenery and attempt to encompass the diverse and changing perceptions of individuals are likely to be most successful. The essence of the preference approach is the judgement of the landscape in totality, as opposed to the measurement techniques, which rely on the definition of factors to explain variation in landscape quality (Dunn, 1976).

Questionnaires or verbal surveys are the most commonly used nonquantitative method for sampling scenic preference of various groups. They are a valuable source of quick information but accuracy can be sacrificed for speed. They are useful for determining preferences for extremely divergent categories of landscape (Arthur et al, 1977). Alternative to questionnaires, one can provide visual stimuli for evaluation, such as photographs (e.g.Shuttleworth, 1980a; Wade, 1982) or one can use other stimuli, such as sound (Anderson et al, 1983). Although perceptions still vary, the variation is less than with verbal descriptions .

There are various difficulties when carrying out such evaluations. Past studies show that the personality of the observer, and their location affect what they observe, as does the duration of observation, the socio-economic profile of the observers, the type of physical characteristics of the landscape and the dynamics of its components and its complexity (Amir and Gidalizon, 1990). Two concerns are noted by Hull and Stewart (1992) - a concern about the threat to the ecological validity of photo-based assessments caused by differences between on-site and photo-based contexts and a concern that the individual rater, rather than the group average, is the more appropriate unit of analysis for tests of validity of photo-based assessments. The techniques have other problems - their psychological basis is at best uncertain; the validity of their quantitative or semi-quantitative results is invariably questionable; and in order to be representative of society's views, they require extensive, time-consuming surveys (Crofts and Cooke, 1974).

Psychological models

The psychological approach has been used in many studies where dimensional analyses of people's preferences for different landscapes are performed. These studies have demonstrated that various psychological constructs such as complexity, mystery, legibility and coherence are important predictors of human landscape preferences (Buhyoff et al, 1994). The psychological model refers to the feelings and perceptions of people who inhabit, visit, or view the landscape. A high-quality landscape evokes positive feelings, such as security, relaxation, warmth, cheerfulness or happiness; a low-quality landscape is associated with stress, fear, insecurity, constraint, gloom, or other negative feelings (Daniel and Vining, 1983).

Because psychological methods use multiple observers and yield one or more quantitative scale values for each assessed landscape, their reliability and sensitivity can be precisely determined. This is an important advantage, since users of these assessments can know the degree of precision and to prove confidence in the landscape values produced. The methods base landscape assessments on the reactions and judgements of the people who experience and/or use the landscapes. In this regard there is an important element of validity inherent in the method (Daniel and Vining, 1983).

Without clear relationships to objectively determine environmental features, the psychological methods leave landscape assessment in a correlational feedback loop; psychological reactions to the landscape are explained only in terms of other psychological reactions:-

"From a practical perspective, this leaves the landscape manager with both feet firmly planted in midair (Daniel and Vining, 1983)."

Phenomenological models

The phenomenological model places even greater emphasis on individual subjective feelings, expectations, and interpretations. Landscape perception is conceptualised as an intimate encounter between a person and the environment (Daniel and Vining, 1983). The principal method of assessment is the detailed personal interview or verbal questionnaire. Phenomenological models tend not to be used to rank landscapes in terms of scenic beauty.

Phenomenological approaches have largely sacrificed reliability in favour of achieving high levels of sensitivity; by emphasising very particular personal, experiential and emotional factors, the visual properties of the landscape become only very tenuously associated with landscape experience (Daniel and Vining, 1983).

This model represents the extreme of subjective determination of relevant landscape features. It fails to establish systematic relationships between psychological responses and landscape features. However, by emphasising the unique role of individual experiences, intentions, and expectations, the phenomenological model serves to point out the importance of the human context in which landscapes are encountered (Daniel and Vining, 1983).

Consensus

Most landscape techniques proceed on the assumption that there is a broad consensus within our society upon what is considered to be of high landscape value. This assumption is linked to another: that "visual quality" is an intrinsic property of landscape and can be stated objectively (Jacques, 1980). The issue of observer consensus is a major topic in landscape perception and preference.

Examples in brief

A series of studies by Kaplan and Kaplan illustrated the psychological model of landscape assessment. A basic method in these studies is to identify relevant psychological variables on photographs of landscapes. Preference ratings and ratings on the landscape dimensions are then obtained from naive observers (Daniel and Vining, 1983).

The essence of Fines' technique is the classification by field observers of subjective responses to the attractiveness of views according to a single, comprehensive and predetermined scale (Crofts and Cooke, 1974).

Most literature on phenomenological methods is devoted to studies of developed landscapes or to perception of environmental hazards. There are only a few specific studies seeking to assess natural landscapes (Daniel and Vining, 1983).

Quantitative holistic methods

Quantitative holistic methodologies combine two approaches: quantitative public preference surveys and landscape features inventories. Measures of landscape quality should be systematically related to physical / biological and social features of the environment so that accurate predictions of the implications of environmental change can be made (Arthur et al, 1977).

Models, such as that of Shafer et al (1969) represent a compromise between techniques which assess the effects of landscape elements on overall preference by summing evaluations of individual dimensions (descriptive methods) and techniques which emphasise the interactions of landscape elements by evaluating the scenic quality of the entire image (preference models); this compromise creates the quantitative holistic models such as the psychophysical and surrogate component models (Buhyoff and Riesenman, 1979; Arthur et al, 1977).

These predictive models have tended to be more a tool for research than for impact assessment. Their orientation is to predict scenic quality based on the presence of quantifiable landscape attributes (Palmer, 1983). Psychophysical modelling uses measurements of physical landscape features to predict people's preferences for the overall visual quality of the landscapes. Classical psychophysics sought to establish precise quantitative relationships between physical features of environmental stimuli and human perceptual responses (Daniel and Vining, 1983).

Traditional psychophysical models, while not "classifying" landscapes, are developed to make predictions of scenic preference or visual quality from variables which are often selected for their predictive, rather than "genuine" explanatory ability (Buhyoff et al, 1994). Surrogate component techniques are based on the identification of physical landscape components which can be compared with preference ratings. Visual management systems aim to be able to both predict and explain scenic preference; their essential purpose is the prediction and assessment of impacts resulting from potential management alternatives (Bishop and Hulse, 1994).

Psychophysical models

Psychophysical methods of landscape assessment seek to determine mathematical relationships between the physical characteristics of the landscape and the perceptual judgements of human observers (Daniel and Vining, 1983). The relationships of interest are those between physical features of the environment (e.g. topography, vegetation, water etc.) and psychological responses (typically judgements of preference, aesthetic value or scenic beauty). Landscape features such as land cover, land use, forest stand structure, and arrangement are measured and then statistically related to scenic quality judgements. Models such as paired comparisons, Likert scales, and sorting and ranking scales are a means to evaluate scenes quantitatively (Arthur et al, 1977); multiple linear regression has recently been the most commonly used techniques to determine these relationships (Buhyoff et al, 1994).

Of all landscape assessments, these methods have been subjected to the most rigorous and extensive evaluation. They have been shown to be very sensitive to subtle landscape variations and psychophysical functions have proven very robust to changes in landscapes and in observers (Daniel and Vining, 1983). Relying on ordinal or interval scales of measurement, psychophysical methods have consistently been able to provide different landscape-quality assessments for landscapes that vary only subtly. However, they require the full range of scenes to be selected to represent all of the physical characteristics used as predictors of scenic beauty (Hull and Revell, 1989). They also provide good assessments of public perceptions of the relative scenic quality differences between landscapes (Buhyoff et al, 1994) based on the assumption that the aesthetic judgements of public panels provide an appropriate measure of landscape quality (Daniel and Vining, 1983).

However, the models can be expensive and time consuming to develop and are restricted to a particular landscape type and to a specified viewer population and perspective; in the short term they are not highly efficient (Daniel and Vining, 1983). The very structure of these models is often a limiting factor in their explanatory value and wide generalisation (Buhyoff et al, 1994).

Psychophysical assessments are useful in many management contexts - features such as quantitative precision, objectivity, and a basis in public perception and judgement are important. The assessments are not based on one expert's opinion, but reflect a measured consensus among observers representative of the public that views landscapes and is affected by management actions (Daniel and Vining, 1983).

Surrogate component models

The basis of component techniques is the identification and measurement of those physical components of the landscape which are regarded as surrogates of scenic quality. The individual components are isolated, their identification and measurement discussed and their combined utility within existing techniques evaluated. Because component ratings are compared to overall preference ratings in these models, the contribution of particular components to scenic beauty can be measured in terms of explained variance (Arthur et al, 1977).

These components can be assigned to three groups in relation to their assumed importance in determining scenic quality. The major components comprise the landscape skeleton as expressed by macro relief (measured by terrain types), relative relief and water presence (measured by drainage density). To these can be added the minor but permanent components which are the variations of the macro forms at smaller scales. They are the overall variations such as surface texture and ruggedness, particular features such as the irregularity of two-dimensional outlines and three-dimensional forms, and the singularities such as isolated features. Finally, there are the transitory components with regard to the characteristics of water bodies and surface textures (Crofts, 1975).

Visual Management Systems

Another approach to the evaluation/assessment of visual resources is the design-based classification/assessment. Visual management systems (VMS) are straightforward systems that use intuitive constructs and easily observable physical landscape attributes to arrive at landscape classification decisions (Buhyoff et al, 1994). Because expert, or knowledge-based, computer systems are capable of carrying out reasoning and analysis functions in narrowly defined subject areas at proficiency levels approaching that of humane experts, they have characteristics that can be used to develop not just a method of predicting visual landscape quality but also a system that explains why certain levels of quality exist. In fact, the specification of knowledge may well be the most important contribution of a scenic quality assessment or prediction system (Buhyoff et al, 1994).

Examples in brief

The prolific work of Shafer and colleagues (Arthur et al, 1977) is an example of quantitative holistic methodologies. They have measured areas, perimeters, and tones of the differentiated landscape zones of photographs and related them to preference rankings (e.g. Shafer et al, 1969; Shafer and Tooby, 1973; Brush and Shafer, 1975). Shafer's studies illustrate a sound and systematic approach to relating components to preferences (Arthur et al, 1977).

The Scenic Beauty Estimation (SBE) method requires that landscapes be observed and judges by panels of persons representative of targeted populations. To develop models using this system, a number of different landscapes must be assessed and their physical characteristics evaluated (Daniel and Vining, 1983). This can be done using colour photographs or slides (Arthur, 1977) or on-site at the landscapes (Schroeder and Daniel, 1981).

Psychophysical models have been developed for landscape vistas by relating measured characteristics of the vistas to scaled landscape preference (Daniel and Vining, 1983). These have often used colour slides of forest or panoramic views and gained preference ratings using paired-comparison formats (Buhyoff and Wellman, 1980; Buhyoff and Riesenmann, 1979).

In Eleftheriadis and Tsalikidis's (1990) model coastal landscape quality was expressed in terms of scenic beauty preferences of the resource users, and these preferences were related to quantitative measures of land use designations, and of forest stand and site characteristics. Carls (1974) used the landscape zones of Shafer et al (1969) together with a people zone, and low and high development zones to look at the effects of people and man-induced conditions of preferences for outdoor recreation landscapes.

Component models and models which predict public preference for scenic beauty are discussed in greater detail later in the review.

Methodological problems and error in models

All these types of models are complicated by methodological problems that can affect interpretation of results. One such problem is whether numerical ratings of landscape beauty represent people's preferences for the landscapes, their judgements of scenic beauty of the landscapes, or both. There is considerable support for the argument that scenic beauty judgements differ from scenic preferences (Arthur et al, 1977); according to Jacques (1980) public preferences tend to give a measure of `value', `quality' is discerned through judgement. When asked to indicate their preference for various landscapes, observers tend to apply criteria for use of those areas (recreation, residence etc) rather than for inherent beauty (Arthur et al, 1977).

There are some persistent errors in the evaluation of landscape, as identified by Hamill (1985). Examples of the following seven types of error have been found in the literature: incorrect use of numbers derived from place in a classification; incorrect use of numbers to stand for words; use of spurious numbers in simple mathematical operations; use of bad data in complex mathematical and statistical operations; use of data that does not satisfy requirements of the model; use of numbers to support, derive, or demonstrate meaningless, spurious or useless concepts; and use of concepts without adequate operational definitions.

Economics of landscape evaluation

How do preference values determined from models relate to economic values of the same landscapes. Results of an exploratory study (Brush and Shafer, 1975) suggest that a consumer's evaluation of real estate that overlooks a given natural scene correlates highly with the scene's predicted preference scores. It should be possible to develop an equation that ties scenic preference values to economic land values. The results of such research should be useful in benefit-cost and environmental impact analyses of the effect of proposed man-made changes in natural environments.

Traditional economic analyses have generally failed to account for unmarketed (nonpecuniary) resources, such as aesthetics. The effect of excluding nonpecuniaries from trade-off (economic) decisions is that they have entered the system as if they were free. Recognition of this problem has, in part, motivated attempts to evaluate scenic resources. If applied to aesthetic resources, redefinition would require putting a price on scenic beauty or charging for its "use". However, putting a price on aesthetic resources is probably not feasible for several reasons. First, aesthetic experiences are difficult to define, second, there is the problem of placing charges on aesthetic experiences (Arthur et al,1977).

Methods of scenic resource financial valuation

Several methods have been used to obtain values for scenic resources. "Willingness-to-pay" values (how much will a commuter pay to preserve the trees), revealed demand (does he take a different highway?) and opinion tallies (does he complain to his MP?) are only a few of the methods used (Arthur et al, 1977). Opinion tallies, often result in undervaluation. Revealed demand is complicated by the necessity of identifying all of the variables acting on the situation.

The hedonic price method (HPM) is a less subjective way of scoring landscape components; components of landscape are valued against people's willingness to pay to live in particular types of landscape, defined as comprising of different bundles of components (Willis and Garod, 1993). HPM is a process of constrained maximisation in which systems of equations involving both prices and quantities for the composite commodity and its attributes are constructed and then solved (Price, 1994).

The travel cost method (TCM) uses a sample of visitors to a site which embodies desired environmental attributes and asks them factual questions about the origin of their journey to the site, their mode of transport and perhaps about other costs incurred and their own socio-demographic characteristics (Bergin and Price, 1994).

Contingent Valuation Technique and Willingness-to-Pay

Contingent valuation techniques (CVT) in landscape evaluation are seen as a natural evolution from landscape evaluation methods based on the scoring of landscape components and other public preference techniques such as landscape ranking. By valuing landscape as an entity, CVT avoids many of the problems, such as those of separability and collinearity, often associated with travel cost and hedonic price methods of landscape valuation (Willis and Garod, 1993).

Willingness-to-pay (WTP) studies can assist in valuing today's landscape, they also attempt to value the benefits which residents and visitors might derive from alternative landscapes which could arise at some time in the future (Willis and Garod, 1993).

WTP's linearity means that it is a linear or other predetermined function of the quantity of the feature in the landscape. However, evidence suggests that the impact of a landscape feature does not increase in proportion to its size (Willis and Garod, 1993). Unfortunately, WTP values are often too high (Arthur et al, 1977) and WTP to gain a commodity is generally less than willingness to accept compensation for losing it (Price, 1994).

Landscape components

The psychophysical and surrogate component techniques of landscape evaluation require the landscape to be divided. This can be done in many ways, similar to those found in the models previously discussed, from simple methods to more abstract definitions. Examples include landform elements (Gardiner, 1974; Land Use Consultants, 1971), landscape patterns or themes (Hammitt et al, 1994; Linton, 1968), landscape character (Crofts, 1975), landscape qualities (Palmer, 1983; Morisawa, 1971), dimensions (Propst and Buhyoff, 1980) and landscape preference predictors (Hammitt et al, 1994; Brush and Shafer, 1975). These examples are detailed and any important points in their respective models noted.

The theory and problems behind landscape characteristics

There are questions regarding the selection of components, the relationships between components, and the relationship between the components and scenic quality as perceived by individuals and groups (Crofts, 1975). The assumption that the visual landscape can be reduced to constituent components, that the visual quality of each component can be measured in isolation and that, when added together, these components represent the total landscape, is a major weakness in component based models (Dearden, 1980).

"Without some theoretical grounds, why should we assume that beauty resides in mountains, in woods, in streams, and not in some unexamined relationship between them (Appleton, 1975)."

In many studies reference is made to `edge', `edge tracts' or `edge categories' all of which terms seem to refer to boundaries or zones of contact between contrasting landscape features. Suppose we were to discover some theoretical grounds for believing that these phenomena of `edges' and `skylines' are of outstanding importance in the aesthetics of landscape. Outdoor recreationalists commonly prefer to use edge environments e.g. the lake edge, the river edge, the cliffs edge, the edge of forests. Landscape architects also find edge and borders a strongly preferred design feature in visual landscapes (Hammitt et al, 1994). If edges are of more or equal importance than the landscape elements they contain, are any of the models discussed truly describing landscapes?

However detailed the search for relevant factors is, there will inevitably be a proportion of the landscape that cannot be explained by the assembled factors alone. This proportion will consist of the subtleties of landscape, such as interaction between elements, and properties of the landscape such as colour, form, shade and lighting (Dunn, 1976).

Methods based on intrinsic landscape factors are proclaimed objective, to avoid public scrutiny of certain projects and achieve "democratic" legitimacy (Tips, 1984), yet the components, regarded as surrogates of scenic quality, are subjectively selected; this selection must be subjective, but it must also be shown to be more than just the opinion of the professional judge (Crofts, 1975; Dearden,1980). The techniques also fail to take account of both the quality of the scenery at a point and the quality of a view from that point in all directions. Lastly, some measurement methods contravene the theories of levels of measurement by using nominal or ordinal scales of measurement and then employing standard arithmetic procedures, such as multiplication and addition. In these circumstances the methods become invalid (Dearden, 1980).

Terrain and hydrological components

Crofts (1975) identified three of the most important terrain components: macro form, relief and landform types; other important components include minor and ephemeral components and hydrological components. The elements of macro form include the categories of the geomorphologist, such as mountains, uplands, valleys and coast (Linton, 1968; Land Use Consultants, 1971). Relief is used in some classifications as a single factor alongside biological, hydrological and human components (Crofts, 1975). Other methods use relief as an absolute measurement, or use variants such as available relief, relative relief or relief as a measure of grandeur (Linton, 1968; Crofts, 1975). Landform types, classified genetically as individual or groups of landforms in geomorphological mapping schemes, represent an amalgamation of virtually all geomorphological components (Crofts, 1975). Landform is the most permanent of all landscape features, as it is the most difficult to alter (Brush, 1981), with exception of volcanoes and the chisel (Mount Rushmore can no longer be classed as natural) and therefore should be a stable basis for landscape classification.

Minor terrain components include visually significant irregularities, such as abruptness of accidentation (Linton, 1968) and contour distinction (e.g. rock outcrops and isolated landforms). Micro features of the landscape, such as surface texture are rarely included in classifications, they are more important in smaller areas; they are perhaps best suited to site evaluations, along with biological textural components (Crofts, 1975). Singularities in landform include isolated hill masses, waterfalls and other unique forms, such as found in glaciated areas (Linton, 1968); these components have an immediate visual impact be focusing the attention of observers. Ephemeral components add detail to the landscape, and can be measured in terms of presence/absence or in terms of the rates of physical change (Crofts, 1975).Hydrological components refer to either water bodies or to river valleys and basins. The primary component is the presence/absence of water, a water body acts as a focal point and can be regarded as a singularity (Crofts, 1975). Of visual importance is the contrast in water surface character - such as discharge, flow variability and velocity. The physical parameters of a linear water body - river width and depth, bed slope, bank erosion and deposition are also important factors. Water has always been a great geomorphological agent, which models the landscape in both physical and economic aspects (Ramos and Aguilo, 1988).

Indices of attractiveness

Landform elements

Land Use Consultants (1971) developed a technique for use in evaluating Scottish landscapes using two series of physical landscapes: relief classes defined in terms of high, normal and low relief per unit area; and landform types such as valley, lowland, plateau, edge and coast. These are then amalgamated and allotted to previously defined landscape tracts. The method omits major landforms, such as mountains and uplands but includes the negative aspects of landform.

In the Leopold method (Crofts, 1975) components of valley and river character are identified to obtain a comparative assessment of the scenic quality of particular sites along certain rivers. Valley character is derived by comparing the width of the valley floor with the height of adjacent mountains (a measure of grandeur); river character is assessed in terms of the width, depth, size and presence and frequency of rapids. A plus point of the technique is the introduction of the concept of scenic uniqueness, however, the technique lacks wide applicability.

Linton (1968) divided up the Scottish landscape into six "landform landscapes" - lowland, hill country bold hills, mountains, plateau uplands and low uplands. Linton never defined the components rigorously but used personal judgement, the landscapes as defined vary greatly in scale and hence visual impact is equally variable. Linton also used the highly subjective assumption that attractiveness increases with an increase in steepness of slopes and boldness of landforms (Crofts, 1975).

Kaplan et al (1989) describe a method using physical attributes which are divided into landform and landcover. Landform elements are slope/relief (the prominence of the landform), edge contrast (contrast between adjacent landforms) and spatial diversity (variety of space created by landform). Landcover elements include naturalism (absence of direct human influence), compatibility (fit between adjacent landcover types), height contrast (height variation among adjacent elements) and variety (diversity of landcover types or patterns within a type).

Morisawa (1971) classified landscape based on relief and water-appearance components, together with seven landscape qualities. Gardiner (1974) argued that relative relief, the presence of water and slope characteristics are the basic landform elements contributing to scenic quality; by using the drainage basin as the fundamental unit of scenic quality, the method omits a large area of landscape from consideration (Crofts, 1975). The approach of Warsynska attempts to derive mathematical notations of scenic beauty in order to evaluate the scenic attraction of areas for tourism. The method uses coefficients of relief attractiveness, surface water attractiveness and forest cover attractiveness. From these three coefficients an overall coefficient of attractiveness is derived (Crofts, 1975).

The Norwegian Institute of Land Inventory (NIJOS, 1995) has developed a landscape classification system based on three levels of regionalisation of the landscape. The following landscape components are systematically described and evaluated: terrain type; geological characteristics; vegetation; water structure; cultivated land; and human population distribution. Maps for evaluation of vulnerability and evaluation of perceptual value can be made from this.

Landscape themes

Nine forest and pastoral landscape patterns or themes were used in the model of Hammitt et al (1994). These were: stream/river; pond/lake; several-ridged; rolling plateau; valley development; farm valley; ridge and valley; one-ridged; and unmaintained (Hammitt et al, 1994).

Linton (1968) divided the landscape into seven land-use landscapes: urbanised and industrialised landscapes; continuous forest; treeless farmland; moorland; varied forest and moorland landscapes; richly varied farmed landscapes; and wild landscapes. Kaplan et al (1989) also described land cover types. These were agriculture, cut grassland, weedy field, scrubland, forests and woodlawn.

Landscape qualities

The Bureau of Land Management in the USA simplified aesthetic criteria into a procedure for making professional appraisals of four qualities (form, line, colour and texture) inherent to three landscape components - land/water, vegetation and structures (Palmer, 1983). Craik (1972) emphasised components of visual analysis, such as vertical enclosure, texture, and focal view.

Morisawa (1971) classified landscape based on relief and water-appearance components, together with vista, colour, vegetation, serenity, naturalness, access and pollution. The method has two major problems: the quantitative assessments assume the components are of equal importance; and that no other factors are relevant to the scenic assessment of the site (Crofts, 1975).

Buhyoff et al (1994) in their EVA model use a visual composition component which assesses the effects of landscape characteristics such as complexity and the vividness of patterns in the landscape and a spatial organisation component which includes accessibility, mystery, enclosure, scale, image refuge, prospect and contemplation.

The informational variables of Kaplan et al (1989) describe variables that could be used with psychological models. Coherence is orderly, with repeated elements and regions; complexity has richness, is intricate and has a number of different elements; legibility is concerned with finding one's way there are back, as well as with distinctiveness; mystery is the promise of new but related information. The last variable is a key element in the informational model. Mystery emphasises an inferential processes and points to the importance of a search for information. It has turned out to be a remarkably reliable and effective predictor, consistently outperforming complexity (Kaplan et al, 1989). Kaplan et al (1989) also describe perception based variables, which are openness, smoothness and locomotion.

Landscape dimensions and preference predictors

Brush and Shafer (1975) describe a landscape-preference model which uses perimeter and area measurements of certain landscape features. The landscape zones used are the immediate, where individual leaves of trees and shrubs are discernible, the intermediate, where only the forms of trees and shrubs are discernible, and the distant zone, where the forms of individual trees cannot be distinguished. The model uses measurements of the area or perimeter of major vegetation, such as trees and shrubs, nonvegetation, such as exposed ground, snowfields and grasses, and water, including streams, lakes and waterfalls to relate to preference rankings (Brush and Shafer, 1975).

Propst and Buhyoff (1980) describe policy capturing, a potential methodology for evaluating landscape preference. A multi-variate linear model simulating each judge's decision is computed by calculating the regression of landscape preferences on ten dimensions thought to influence such preference. This method uses ten important dimensions: foreground vegetation; mountains; man-changes area; visible distant landforms; green colours; blue colours; unobstructed expanse of view; clouds; and undisturbed forest (Propst and Buhyoff, 1980).

The regression model of Hammitt et al (1994) uses visual preference estimates for nine forest and pastoral landscape patterns or themes. Seven variables were used to predict scenic preference. The equation used is based on six significant predictors: area of sky; area of largest ridge (background or very distant); linear perimeter of ridge line (background/very distant); area of moving water (e.g. streams, rivers); obstructing vegetation squared; and area of rolling plateau (background/very distant) (Hammitt et al, 1994).

Use of models to predict preference

It is recognised that there are a few permanent landscape characteristics which are prime contributors to scenic quality - terrain, water, ground cover, and human artefacts (Crofts, 1975). Where psychophysical models have been derived, public judgements do seem to distinguish landscapes on the basis of features that are intuitively appropriate. For example, rushing water, large trees, grassy meadows and jagged mountains have all been found to be positive aesthetic features by the criterion of public judgements. Downed wood and slash, dense stands of small trees, and recently killed trees have been found to be negative aesthetic features (Daniel and Vining, 1983).

Landscape measurement (component models) is the first stage of the landscape evaluation process, after which follow landscape preference or value measurement and the evaluation of landscape in terms of individual and societal preferences for different landscapes as measured by components (Crofts, 1975).

Predictive modelling requires sensitive and reliable assessments of the predictors and the response which, ideally, are measurements of interval or better quality (Bishop and Hulse, 1994). The computational capabilities of a geographic information system (GIS), together with prediction equations based on assessment of video panoramas of locations affected by landscape change, could enable a more objective and cost-effective visual assessment and prediction procedure to be developed (Bishop and Hulse, 1994).

However complicated the technique might be, public preference for natural environments is itself a complex phenomenon. Not to use mathematics (and computer technology) to examine this phenomenon would be like trying to fell a tree with a chain saw without turning on the power switch (Brush and Shafer, 1975). While linear models have performed quite well, preliminary experiments suggest that second-degree terms (squares and products) in a polynomial regression may somewhat improve the precision of the models. Nonlinear transformations of the predictors may also be appropriate (Schroeder and Daniel, 1981); however, it has been shown that more complex nonlinear models perform only slightly better than the simple linear ones, and therefore may not be appropriate for predicting landscape preference from field data (Schroeder and Brown, 1983).

Theoretical models and some results from them

Some of the models used over the last thirty years to predict scenic preference are discussed below. The models of Hammitt et al (1992) and Shafer et al (1969) have produced results which go some way to explaining landscape preferences.

Abello, Bernaldez and Galiano's model

In this study preference ranking was compared with landscape characteristics. Multiple linear regression between rank and landscape characteristics identified four preference-determining characteristics: fertility / plant vigour / healthy biomass - less plant vigour; wintery / defoliated landscape with increased plant structure legibility - no defoliation; barren soil - covered grassy soil; recurrent patterns / rhythm - no recurrent pattern. These predicted average preference, absorbing 85% of the total variance. The results from this model showed that fertility / plant vigour was a key feature deciding the preference between two scenes. However, this visual characteristic may be challenged by a strong expression of pattern / rhythm / recurrent texture of the landscape elements (Abello et al, 1986).

Calvin, Dearinger and Curtin's model

This model identified three important factors, namely natural scenic beauty, natural force and natural starkness. The first two of these factors accounted for 85% of the variance among preference for scenes. Natural scenic beauty represents a variation between scenes described as colourful, beautiful, natural and primitive to ones described as drab, ugly, artificial and civilised. Natural force represents a distinction between turbulent, loud, rugged and complex scenes and tranquil, hushed, delicate and simple scenes. Natural starkness is of doubtful significance, but represents a distinction between scenes judged as warm and fertile to those judges as cold and barren (Calvin et al, 1972).

Patsfall, Fiemer, Buhyoff and Wellman's model

Two studies were conducted to examine distance classes of vegetation (foreground, middleground and background) and scene composition (presence of vegetation in left, centre or right section of the image) as predictors of perceived scenic beauty. For each landscape image areal measures of vegetation in each distance class and for each vertical section were taken and used as predictors. Among the most important contributors to scenic beauty were amount of centre middleground vegetation, and centre background vegetation. Left foreground vegetation and right foreground vegetation were found to have significant and opposing regression weight signs - negative for the left and positive for the right (Patsfall et al, 1984).

Visual Resource Management System (VRMS)

Hadrian et al (1988) used a GIS to develop a partial VRMS for application to a specific visual problem. They created a GIS containing topography, zoning, built forms, vegetation density and road orientation as map layers. This was used with a predictive model to map the distribution of visual impacts for any given electricity transmission line route design. The model as developed took account of: distance between observer and towers; complexity of vegetation between observers and towers; attractiveness of tower environments; attractiveness, orientation, nature of use and rate of observer environment; and the masking effects of terrain and buildings (Bishop and Hulse, 1994).

Bishop and Hulse's model

A high level of prediction of scenic beauty values was achieved using five variables computed using a GIS database. These were amount of foreground river, amount of high slope in the foreground, amount of orchard land use in the foreground, amount of forest in all distance ranges and range of relief in visible cells without vegetative screening (Bishop and Hulse, 1994).

The coefficients for nearby rivers, high foreground slope and high relief were positive. This result fits well with established theory of landscape preference. The positive influence of water has been clearly shown by many researchers. Relative relief has also been found to contribute positively to scenic beauty estimates. Although rough ground textures have been negatively correlated with preference, a high slope at the view position generally is indicative of a significant vantage point offering views and a range of positive visual characteristics such as legibility and complexity. Less clear is the meaning of negative coefficients on nearby orchard land use and extent of visible forest (Bishop and Hulse, 1994).

The coefficient for the proportion of all cells containing visible forest was apparently negative because although scenes with near vegetation and quite low visibility sometimes scored high SBEs, scenes in which there was some higher level of visibility which was dominated by forests in the middle and background zones were not so well liked. In other words, the presence of this negative predictor appears to suggest that people need to see the texture of trees to appreciate them properly; when the trees become an amorphous dark mass the effect is contrary. It is possible that the choice of video, with its comparatively low resolution, as the display medium influenced this result (Bishop and Hulse, 1994).

Steinitz's model and experiment

Steinitz (1990) compared five existing alternative explanatory theories and one new model, which was developed from the existing models. A visitor survey looked at patterns of use and characteristics of park users as well as at user visual preferences. The six models were tested for their ability to have predicted the pattern of responses to the survey - results showed the existing models, including that of Shafer et al (1969) to be far less accurate than the new model.

The survey used diverse views of developed landscape as well as photographs which videographically simulated the effects of landscape change. Agreement among the survey participants showed a definite pattern, with higher disagreement as the scores got higher and lower (Steinitz, 1990).

A model of ecological integrity was also developed as part of this study, to identify the roles of various landscape elements in maintaining a high diversity of wildlife habitats. Habitat diversity was defined as "the ability of any landscape elements to be important in the potential habitat of successively more species". The patterns of visual preference and ecological integrity were each mapped using a GIS and then compared (Steinitz, 1990).

Explanation of Visual Assessment (EVA)

Explanation of visual assessment (EVA) consists of five separate components: four landscape attribute domain evaluations and a final visual quality rating component that integrates and analyses the results of the four domains evaluations into a final, comprehensive visual quality assessment (Buhyoff et al, 1994). The attribute domains are evaluation of the visual effects of man-made features, evaluation of the visual effect of natural features, evaluation of visual composition of the view and evaluation of the spatial configuration of the landscape.

The man-made evaluation component assesses the effects, both positive and negative, of alterations and structures on visual quality. Natural features evaluation component assesses features such as forests, trees, mountains, natural openings and so on. A visual composition component assesses the effects of landscape characteristics such as complexity and the vividness of patterns in the landscape. Spatial organisation effects such as accessibility, mystery, enclosure, scale, image refuge, prospect and contemplation are assessed. The final visual quality rating component analyses the output of the four landscape/perception domain systems and derives a final, overall landscape evaluation.

After the four components are executed, one of nine possible results is assigned to each component, indicating the contribution of that component to the overall visual quality rating. These results include six ratings - outstanding, high, moderately high, moderate, moderately low and low - that indicate what the visual quality of the landscape would be if there were no other effect on visual quality by any of the other components. There are also two results that indicate that the component did not have a significant enough effect to determine visual quality by itself, but that it does have a modifying effect, either a slight positive or a slight negative effect. Lastly, the result can indicate that the component has "no effect" on visual quality for the landscape being rated.

EVA provides detailed explanations for visual assessments which are maintained throughout the process, whereas a traditional VMS or psychophysical model loses information as a result of generalisation (Buhyoff et al, 1994). Evaluations preformed by EVA are more subjective than those performed by statistical models. EVA appears to have quite good internal validity and good reliability when the user is familiar with the concepts utilised by the system and is trained in the system's use (Buhyoff et al, 1995).

Policy capturing

Policy capturing, a potential methodology for evaluating landscape preference. A mutli-variate linear model simulating each judge's decision was computed by calculating the regression of landscape preferences on ten dimensions thought to influence such preference. An equation which captures, or defines, the policy of each judge was generated (Propst and Buhyoff, 1980). The objectives of the research are to determine if Policy Capturing (PC) is a viable procedure for modelling human judgement of scenic beauty and to identify and determine the relative importance of those landscape features which explain variations in judgements of scenic beauty.

This study employed ten important dimensions: foreground vegetation; mountains; man-changed area; visible distant landforms; green colours; blue colours; unobstructed expanse of view; clouds; and undisturbed forest.

The method identified groups of homogeneous raters. For example on group put heavy reliance on visible distant forms and foreground vegetation, while another attached more importance to undisturbed forest, man-changed area, and the proportion of blue colour. Possible reasons for the failure of the model to capture more of the predictable variance in preference can be summarised as follows: raters were using cues in a non-linear fashion; the wrong independent variables were used; slides are not valid representations of actual landscapes; cue intercorrelations deflated values obtained; and simple rating scales are subject to too many biases. Rating scales are notoriously prone to such errors as the halo effect and errors of central tendency.

For individuals, amount of clear sky, mountains, man-changed area, undisturbed forest and clouds were the dimensions most often given the highest weights. However, there was no clear pattern in terms of which dimensions accounted for most of the variance in preference (Propst and Buhyoff, 1980). One value of PC is that it allows the inference of an individual judge's policy by requesting an overall evaluation of a total stimulus (e.g. a landscape) rather than requiring evaluations of the elements making up that stimulus.

Hammitt, Patterson and Noe's model

The method uses visual preference estimates for nine forest and pastoral landscape patterns or themes. These were: stream/river; pond/lake; several-ridged; rolling plateau; valley development; farm valley; ridge and valley; one-ridged; and unmaintained. A seven-variable regression model was used to predict scenic preference (Hammitt et al, 1994). The equation used is based on six significant predictors: area of sky; area of largest ridge (Background or very distant); linear perimeter of ridge line (background/very distant); area of moving water (e.g. streams, rivers); obstructing vegetation squared; and area of rolling plateau (background/very distant).

This model explains 76% of the variation in scenic preference. Sharp peaked mountains accounted for the largest amount of variance in visual preference for Colorado vistas. Long unobstructed scenic views of dense forests leading toward a backdrop of sharp peaked mountains were the most preferred vistas (Hammitt et al, 1994). Predictors with a positive influence on scenic preference perceptions were linear length of ridge line, area of rolling plateau and area of moving water. Features detracting from scenic preference perceptions were area of largest ridge, area of sky, and area of obstructing vegetation. All of these factors have the potential to impede vision of the total landscape.

However, area of sky may be a surrogate for other variables. Specifically, the area of sky in photographs corresponds to the absence of other aesthetically values landscape features (e.g. ridges, rolling plateau, water, and possibly other variables not identified in the model). That is, as area of sky increases, the area of potentially relevant landscape features decreases (Hammitt et al, 1994). It is suggested that scenic value increases as the number and area of ridges increase. However, a single dominating ridge lowers rather than enhances scenic preference (Hammitt et al, 1994). A theoretical explanation for the negative influence of one-ridges vistas is that they lack the visual complexity and content involvement of multi-ridged scenes, plus they lack inter-lacing ridge lines and resulting ravines that lead the viewer to visually explore multi-ridge scenes.

Moving water is also important as a predictor, however, the factor analysis showed that scenes with moving water were the most preferred of all vistas (Hammitt et al, 1994). The area of obstructing vegetation squared was important. This predictor suggests that a negative curvilinear relationship exists between amount of obstructing vegetation and scenic preference perceptions (Hammitt et al, 1994). The final predictor was area of rolling plateau in the background / very distant portions of the scene. Rolling plateau had a positive effect on scenic beauty perceptions (Hammitt et al, 1994).

An alternative model explains 71% of the variance in scenic preference. The model contains five variables also found in the `best' model: area of moving water, linear perimeter of ridge line, area of rolling plateau, area of obstructing vegetation and area of obstructing vegetation squared. Considered together the two variables associated with obstructing vegetation again suggest that vegetation which partially obscures the view detracts from scenic preference. The two new predictors were area of forest bordering fields and streams in the fore/mid ground and the same landscape feature in the background/very distant portions of the scene (Hammitt et al, 1994).

Based on variables common to both models, it is concluded that forest vistas of highest visual preference are determined by: area of moving water, represented by forested rivers and streams; linear perimeter of ridge line, represented by multi-ridged mountain viewsheds; area of rolling plateau terrain; area of obstructing vegetation. The two new predictors in equation 2 involving area of forest border along fields and streams have a positive influence, particularly amount of forest border/edge in the fore and mid-ground views (Hammitt et al, 1994).

Factor analysis of visual preference estimate values reduced the data set to 9 vista landscape themes. Mountain forest landscapes with water, either flowing or stationary, ranked highest in scenic preference, followed by mountain vistas with multi-ridges in the background. Least preferred were vista landscapes consisting of views with only one mountain ridge present, and vistas in which the foreground woody vegetation obscured a portion of the vista (Hammitt et al, 1994). Forest border or edge in the fore/mid ground, moving water, and amount of ridge line (multi-ridges) were positive predictors, while obstructing vegetation in the foreground of unmaintained vistas detracted the most from scenic preference.

The Shafer method

The Shafer method (Shafer et al, 1969; Brush and Shafer, 1975; Shafer and Brush, 1977) is a landscape-preference model which uses perimeter and area measurements of certain landscape features. The landscape zones used are: the immediate, where individual leaves of trees and shrubs, soil texture, stones and rocks are discernible; the intermediate, where only the forms of trees and shrubs are discernible and the outlines of rocks and prominent features of snow covered or bare land are distinguishable; and the distant zone, where the forms of individual trees cannot be distinguished and no details of soil, rocks, grasses or snow can be recognised (Brush and Shafer, 1975; Shafer et al, 1969). The model uses measurements of the area or perimeter of major vegetation, such as trees and shrubs, nonvegetation, such as exposed ground, snowfields and grasses, and water, including streams, lakes and waterfalls.

The details of the ten zones used in the original study by Shafer et al (1969) are: sky zone, sky and clouds only; immediate-vegetation zone; intermediate-vegetation zone; distant-vegetation zone; immediate non-vegetation zone; intermediate non-vegetation zone; distant non-vegetation zone; stream zone, including only water and rocks in a stream; waterfall zone, including only water and rocks in a waterfall; an lake zone, including water and rocks in a lake.

The model predicts quite accurately how people will rank (or score) natural landscapes although it does not predict landscape appeal directly, as it predicts the appeal of a photograph (Brush and Shafer, 1975; Shafer et al, 1969). The model's terms include those features that are important in a landscape's aesthetic appeal. The three landscape elements measured - vegetation, nonvegetation and water - are also the gross features in natural landscapes that man is capable of altering to an appreciable degree. Through area and perimeter measurements, the model also uses the relative proportions of landscape features in the landscape. The use of three zones in the model takes into consideration the textural variation of vegetation and nonvegetation. The sense of depth in a view, as established by textural gradients and overlapping land forms, is generally recognised as a major factor in scenic preference (Shafer and Brush, 1977).

The following items had positive effects on the aesthetic appeal of landscape: perimeter of immediate vegetation; perimeter of intermediate non-vegetation; perimeter of distant vegetation multiplied by area of water; area of intermediate vegetation multiplied by area of distant non-vegetation; and area of intermediate vegetation multiplied by area of water. The negative effects items were: perimeter of immediate vegetation squared; area of water squared; perimeter of immediate vegetation multiplied by distant vegetation; perimeter of immediate vegetation multiplied by area of intermediate vegetation; and perimeter of intermediate vegetation multiplied by area of distant non-vegetation (Shafer et al, 1969).

Perimeter measurements stress the prominent edges between the forest canopy and open ground or water, edges that separate masses of contrasting texture and tone. Studies have shown that viewer's attention focuses at points along such edges (Brush and Shafer, 1975). Water, when combined in a scene with forest cover, strongly enhances scenic quality, yet if it occupies too much of a scene, water detracts from the scenic quality. This suggests that without the contrast of dark vertical masses of trees in the distance, the presence of water can diminish the scenic quality of a scene (Shafer and Brush, 1977).

Factors that may have influenced model results include weather conditions and photographic composition. Third order terms may have explained more variation, unfortunately the computer used did not have the storage space to cope with these additional terms (Shafer et al, 1969). The study was repeated in Scotland in 1972; the landscape preference equation developed from the original study in 1969 would have predicted quite accurately the preferences of Scottish people, although it was developed from data collected in the United States (Shafer and Tooby, 1973).

A principal advantage of the regression model is the use of second-order terms that describe interrelated or interlocking elements of the landscape. The model also recognises the importance of framing and of water. Two of the strongest criticisms of the Shafer model are the lack of any theoretical foundation and the failure to account for individual differences (Propst and Buhyoff, 1980).

Summary of landscape components used in predictive models

Two distinct patterns are apparent in the predictors used for preference models - the abstract, more subjective properties and the more objective, measurement based properties of landscapes. The former appear to be used to explain measured preference, and are rarely used to create a model to predict preference for a new landscape. The latter are used to create models which can be generically applied, such as the model of Shafer et al (1969) which was developed in the United States and was later applied to Scotland (Shafer and Tooby, 1973).

The model of Abello et al (1986) and Calvin et al (1972) used preference rankings to determine relatively abstract properties of landscapes - the studies would be difficult to apply. EVA (Buhyoff et al, 1994) is an expert-system and is as such objective, however, it still uses semi-tangible concepts (mystery, complexity, vividness of patterns) which have been defined so as to be quantifiable.

Many of the other models use areas and perimeters of landscape zones, in three distance classes: immediate, intermediate and distant (Shafer et al, 1969) or foreground, middleground and background (Patsfall et al, 1984); although using distance classes does not always increase prediction of scenic preference (Hammitt et al, 1994). Some use all three in a very systematic fashion (Shafer et al, 1969; Patsfall et al, 1984) while others use only some in conjunction with other descriptors (Bishop and Hulse, 1994; Hammitt et al, 1994; Propst and Buhyoff, 1980).

Vegetation presence within the zones is used to delineate the landscape - Patsfall et al (1984) used presence in the left, centre and right of the distance classes while Shafer et al (1969) used vegetation, non-vegetation and water in the distance classes. The more complex zonations of Hammitt et al (1994), Bishop and Hulse (1994) and Propst and Buhyoff (1980) use other components of the landscape such as sky or cloud (Propst and Buhyoff, 1980; Hammitt et al, 1994), ridge lines or range of relief (Bishop and Hulse, 1994; Hammitt et al, 1994), green and blue colours (Propst and Buhyoff, 1980), area of obstructing vegetation (Hammitt et al, 1994) and area of unobstructed view (Propst and Buhyoff, 1980).

Summary of explanatory results of the predictive models

Vegetation is important in predicting scenic quality (Abello et al, 1986); too much in the foreground can obstruct vision and detract from visual quality (Hammitt et al, 1994; Shafer et al, 1969). Fore and mid-ground vegetation has been shown to be a major predictor of visual quality (Hammitt et al, 1994). It has also been shown to have different effects on the left and right of a scene (Patsfall et al, 1984). Water is also very important, in the form of rivers (Bishop and Hulse, 1994), moving water (Hammitt et al, 1994) and waterfalls (Brush and Shafer, 1975), these are all aspects of natural force, a predictive factor identified by Calvin et al (1972). Again, too much water can detract from a scene (Brush and Shafer, 1975).

Similarly, a large area of sky or a large ridge is a negative predictor of scenic quality (Hammitt et al, 1994), possibly because it impedes vision of the landscape and decreases the area of potentially interesting or complex landscape. This is also highlighted in the visual composition and spatial organisation components of the EVA system (Buhyoff et al,1994). The importance of edges of landscape components (Shafer and Brush, 1977) is due to contrast of texture and tone, this can be looked at as the level of complexity or diversity in the landscape.

Such landscape components as mountains, undisturbed forest in the middle and background and length of unobstructed view are all important positive predictors (Propst and Buhyoff, 1980; Patsfall et al, 1984; Hammitt et al, 1994) whereas human development and changes are negative predictors of scenic preference (Propst and Buhyoff, 1980); this relates to the natural beauty predictor in the model of Calvin et al (1972).

In many ways, it is the semi-tangible landscape properties that make sense of the predictive power of a wide range of explicit, measurable landscape components.

References

Abello, R.P., Bernaldez, F.G. and Galiano, E.F. (1986) Consensus and contrast components in landscape preference. Environment and Behaviour, 18, 155-176.

Amir, S. and Gidalizon, E. (1990) Expert based Method for the Evaluation of Visual Absorption Capacity of the Landscape. Journal of Environmental Management, 30, 251-163.

Anderson, L.M., Mulligan, B.E., Goodman, L.S. and Regen, H.Z. (1983) Effects of sounds on preferences for outdoor settings. Environment and Behaviour, 15, 539-566.

Appleton, J. (1975) Landscape evaluation: the theoretical vacuum. Transactions of the Institute of British Geographers, no 66, 120-123.

Arthur, L.M. (1977) Predicting scenic beauty of forest environments: some empirical tests. Forest Science, 23, 151-160.

Arthur, L.M., Daniel, T.C. and Boster, R.S. (1977) Scenic assessment: an overview. Landscape Planning, 4, 109-129.

Bergin, J. and Price, C. (1994) The travel cost method and landscape quality. Landscape Research, 19, 21-23.

Bishop, I.D and Hulse, D.W. (1994) Prediction of scenic beauty using mapped data and geographic information systems. Landscape and Urban Planning, 30, 59-70.

Briggs, D.J. and France, J. (1980) Landscape Evaluation: A comparative study. Journal of Environmental Management, 10, 263-275.

Brush, R.O. (1981) Landform and scenic preference: a research note. Landscape Planning, 8, 301-306.

Brush, R.O. and Shafer, E.L. (1975) Application of a Landscape-Preference Model to Land Management. In Landscape Assessment: Values, Perceptions and Resources (eds. Zube, E.H., Brush, R.O. and Fabos, J.G.), p168-181, Halstead Press, Pennsylvania.

Buhyoff, G.J. and Riesenmann, M.F. (1979) Experimental manipulation of dimensionality in landscape preference judgements: a quantitative validation. Leisure Sciences, 2, 221-238.

Buhyoff, G.J. and Wellman, J.D. (1980) The specification of a non-linear psychophysical function for visual landscape dimensions. Journal of Leisure Research, 12, 257-262.

Buhyoff, G.J., Miller, P.A., Roach, J.W., Zhou, D. and Fuller, L.G. (1994) An AI Methodology for Landscape Visual Assessments. AI Applications, 8, 1 - 13.

Buhyoff, G.J., Miller, P.A., Hull, R.B. and Schlagel, D.H. (1995) Another look at expert visual assessment: validity and reliability. AI Applications, 9, 112-120.

Calvin, J.S., Dearinger, J.A. and Curtin, M.E. (1972) An attempt at assessing preferences for natural landscapes. Environment and Behaviour, 4, 447-470.

Carls, E.G. (1974) The effects of people and man-induced conditions on preferences for outdoor recreational landscapes. Journal of Leisure Research, 6, 113-124.

Cooper, A and Murray, R. (1992) A structured method of landscape assessment and countryside management. Applied Geography, 12, 319-338.

Craik, K.H. (1972) Psychological factors in landscape appraisal. Environment and Behaviour, 4, 255-266.

Crofts, R.S. (1975) The landscape component approach to landscape evaluation. Transactions of the Institute of British Geographers, no 66, 124-129.

Crofts, R.S. and Cooke, R.U. (1974) Landscape Evaluation: A comparison of techniques. Occasional Papers, no 25, Department of Geography, University College London.

Daniel, T.C. and Vining, J. (1983) Methodological Issues in the Assessment of Landscape Quality. In Behaviour and the Natural Environment (eds. Altman, I. and Wohwill, J.),Chapter 2, 39-83, Plenum Press.

Dearden, P. (1980) A Statistical Technique for the Evaluation of the Visual Quality of the Landscape for Land-use Planning Purposes. Journal of Environmental Management, 10, 51 - 68.

Dearden, P. (1985) Philosophy, theory, and method in landscape evaluation. Canadian Geographer, 29, 263-265.

Dunn, M.C. (1976) Landscape with photographs: testing the preference approach to landscape evaluation. Journal of Environmental Management, 4, 15-26.

Eleftheriadis, N. and Tsalikidis, I. (1990) Coastal pine forest landscapes: modelling scenic beauty for forest management. Journal of Environmental Management, 30, 47-62.

Hadrian, D.R., Bishop, I.D. and Mitcheltree, R. (1988) Automated mapping of visual impacts in utility corridors. Landscape and Urban Planning, 16, 261-282.

Hamill, L. (1985) On the persistence of error in scholarly communication: the case of landscape aesthetics. Canadian Geographer, 29, 270-273.

Hammitt, W.E., Patterson, M.E. and Noe, F.P. (1994) Identifying and predicting visual preference of southern Appalachian forest recreation vistas. Landscape and Urban Planning, 29, 171-183.

Hull, R.B. and Revell, G.R.B. (1989) Issues in sampling landscapes for visual quality assessments. Landscape and Urban Planning, 17, 323-330.

Hull, R.B. and Stewart, W.P. (1992) Validity of photo-based scenic beauty judgements. Journal of Environmental Psychology, 12, 101-114.

Jacques, D.L. (1980) Landscape Appraisal: The Case for a Subjective Theory. Journal of Environmental Management, 10, 107-113.

Kaplan, R., Kaplan, S. and Brown, T. (1989) Environmental preference: a comparison of four domains of predictors. Environment and Behaviour, 21, 509-530.

Land Use Consultants (1971) A Planning Classification of Scottish Landscape Resources. Countryside Commission for Scotland, Occasional paper no1, Vol 2.

Linton, D.L. (1968) The assessment of scenery as a Natural Resource. Scottish Geographical Magazine, 84, 219 - 238.

NIJOS (1995) Landscape Inventory.

Orland, B., Weidemann, E., Larsen, L. and Radja, P. (1995) Exploring the relationship between visual complexity and perceived beauty. Imaging Systems Laboratory, Department of Landscape Architecture, University of Illinois at Urbana-Champaign. Internet page: http://imlab9.landarch.uiuc.edu/projects/compleximages/complexity.html.

Palmer, J.F. (1983) Visual Quality and Visual Impact Assessment. In Social Impact Assessment Methods (eds. Finsterbusch, K., Llewellyn, L.G. and Wolf, C.P.), Chap 13, p268-283, Sage Publications.

Patsfall, M.R., Feimer, N.R., Buhyoff, G.J. and Wellman, J.D. (1984) The prediction of scenic beauty from landscape content and composition. Journal of Environmental Psychology, 4, 7-26.

Price, C. (1994) Appendix: Literature Review. Landscape Research, 19, 38-42.

Propst, D.B. and Buhyoff, G.J. (1980) Policy capturing and landscape preference quantification. Journal of Environmental Management, 11, 45-59.

Ramos, A. and Aguilo, M. (1988) The landscape of water: introduction. Landscape and Urban Planning, 16, 1-11.

Schroeder, H.W. and Brown, T.C. (1983) Alternative functional forms for an inventory-based landscape perception model. Journal of Leisure Research, 15, 156-163.

Schroeder, H. and Daniel, T.C. (1981) Progress in Predicting the Perceived Scenic Beauty of Forest Landscapes. Forest Science, 27, 71 - 80.

Shafer, E.L. and Brush, R.O. (1977) How to measure preferences for photographs of natural landscapes. Landscape Planning, 4, 237-256.

Shafer, E.L. and Tooby, M. (1973) Landscape preferences: an international replication. Journal of Leisure Research, 5, 60-65.

Shafer, E.L., Hamilton, J.F. and Schmidt, E.A. (1969) Natural landscape preferences: a predictive model. Journal of Leisure Research, 1, 1-19.

Shuttleworth, S. (1980a) The Use of Photographs as an Environmental Presentation Medium in Landscape Studies. Journal of Environmental Management, 11, 61-76.

Shuttleworth, S. (1980b) The Evaluation of Landscape Quality. Landscape Research, 5, 14 - 20.

Steinitz, C. (1990) Toward a sustainable landscape with high visual preference and high ecological integrity: the Loop Road in Acadia National Park, USA. Landscape and Urban Planning, 19, 213-250.

Tips, W.E.J. (1984) A Review of Landscape Evaluation in Belgium and Some Implications for Future Research. Journal of Environmental Management, 18, 57 - 71.

Unwin, K.I. (1975) The relationship of observer and landscape in landscape evaluation. In Transactions of the Institue of British Geographers, no.66, 130-133.

Wade, G. (1982) The relationship between landscape preference and looking time: a methodological investigation. Journal of Leisure Research, 14, 217-222.

Willis, K.G. and Garod, G.D. (1993) Valuing Landscape: a Contingent Valuation Approach. Journal of Environmental Management, 37, 1-22.