This paper describes a GIS process used to develop the New Zealand Landscape Classification (NZLC). The classification provides a frame of reference for communication and an inventory of scenery. The classification does not provide information on quality but provides information on character, such as landform, vegetation, naturalness, and water. It is hoped that researchers will use this classification to link landscape character with the public’s perception of quality. GIS is an ideal platform for the development of a landscape classification because it provides an explicit framework that enables criteria to be questioned and the classification to evolve with improved understanding.

Landscape in this paper is defined as the "appearance of the land" (Swaffield 1991), or the "environment perceived" (Appleton 1980) or the "spectacle presented by the countryside" (Countryside Commission 1970). In most countries, landscape contributes considerably to people’s quality of life and is a major part of the tourism industry in many cases. In New Zealand, for example, the landscape is a multibillion dollar tourism resource that has considerable quality of life value for ordinary New Zealanders. Given the importance of landscape, it is not surprising that any human-induced change to the landscape, such as the construction of hotels, infrastructure or plantation forestry, can be very controversial. This is particularly the case in wilderness areas and planners face the dilemma of how to provide facilities for tourists without compromising the landscapes that the tourists have come to see.

When considering the landscape impacts of a particular development, planners need information on the compromises that are being made to landscape quality. Landscape quality depends on whose perceptions are being considered – beauty is in the eyes of the beholder (Dearden 1989). Landscape researchers need a methodology for obtaining landscape quality information that can be validated by independent researchers and applied to a number of locations. One such approach is known as psychophysical analysis.

The psychophysical approach is based on the work of classical psychophysics, which sought to establish quantitative relationships between physical features of environmental stimuli and human perceptual responses (Daniel and Vining 1983). Landscape quality is determined by the public, not by experts. This approach typically involves two stages. The first is the classification of the landscape into reasonable homogeneous classes, which are usually based on aspects of landform, landcover, naturalness and water. The second stage is the evaluation of quality of the different classes within the classification. This is done with a public preference survey, with the classification acting as a frame of reference. With public preference surveys, typically, samples of the public are questioned about how they rank each landscape class for its visual quality. The survey could require a sample of the public to visit landscape sites, but usually photographs of the classes are used, and these are ranked. The psychophysical approach might identify a consensus within society, but it could also show that different groups in society value landscapes differently.

The psychophysical approach requires a landscape character classification as a frame of reference for the research. To manage and research landscapes effectively, it is essential that landscapes are classified so that planners have an inventory and a frame of reference for communication and research. Many of the major sciences use classifications because they enable researchers to communicate effectively and precisely and, therefore, to compare results and develop a consensus of knowledge. For example, consider the benefits of plant classification for botanists. Despite this, landscape researchers do not have a well-established landscape classification and, as a result, their research cannot be effectively communicated and compared. Consequently, a consensus of understanding on landscapes has been slow to develop. It is critical that a landscape classification be developed so that landscape research can advance our understanding of the nature of landscapes and ascertain the values associated with different landscapes. A landscape classification can be used for monitoring landscape change, calculating uniqueness and determining variety. Although landscape evaluation is an important end use for a landscape character classification, evaluation or the assessment of landscape quality will not be directly considered in this paper. A landscape character classification does not identify quality, but because it is used for evaluation, it must differentiate classes that are different in perceived quality.

This paper describes a landscape classification developed for New Zealand. It also examines some applications of the classification.

To develop a useful landscape classification, it is necessary to have a set of criteria. Two sets of criteria are used here: a general set of classification criteria that is common to all land-based classifications, and a set of specific criteria developed for landscape classification.

General Criteria

The general criteria were adapted from Rhind and Hudson (1980), who used them as criteria for a landuse classification. These are as follows:

Specific Criteria

For a classification to be useful for science, it must have a specific purpose and be based on theory relating to that purpose (Haines-Young and Petch, 1986). As previously mentioned, a landscape character classification is used for subsequent evaluation research. A landscape classification for this purpose must therefore distinguish characteristics that affect quality. So what is the theory on the characteristics that affect quality? The specific criteria that a classification should follow are listed below; they are designed to capture important landscape characteristics that affect quality. A discussion on the theory behind these criteria follows.

The landscape attributes have been identified from content category research. The Countryside Commission (1988) summarised this research, and the above attributes comply with it, as well as with other research (Amedeo, Pitt and Zube 1989). A problem with content category research is that it is specific to particular locations. However, the above attributes have been used by landscape researchers worldwide, and common language used to describe landscapes often relate to these attributes. In this paper, landscape attributes refer to landform, vegetation, naturalness or water, while landscape components refer to objects that are part of these attributes (mountains, forest, lakes, cities).

Criterion 2

The landscape is derived from an interaction between aspects of the environment (landscape components), and human perception processes. Landscape perception depends not only on these physical landscape components, which may be extremely diverse, interrelated and complex, but also on the values, past experience and social-cultural conditioning of the observers (Dearden 1989). Jones (1991) describes this complexity as the "elusive reality of landscape". He adds that, in the past, the lack of recognition that landscapes are a physical reality and a social or cultural construct has led to an "academic battlefield", with different disciplines and schools concentrating on either the physical landscapes or on the observer. It is necessary for a landscape classification to be based on a particular type of observer. Since it is the evaluation of the public's perception of landscape quality that is important to planners, the public's perception of landscape character should be considered in a landscape classification, rather than the perceptions of specialised scientists (Zube, 1984).

The difficult questions that arise are, what is the public, and what is the nature of its perception? The notion of a general public eliminates extreme perceptions, yet the general public is not a homogeneous group. A classification should be based on the lowest common denominator within this group. For example, not all people perceive the difference between exotic and indigenous vegetation, but many members of the public do; therefore, the distinction between exotic and indigenous vegetation is necessary. This lowest common dominator level is not always obvious, and a classification should be hierarchical (a general criterion that has been specified) to cope with a range of different perceptions.
 

Criterion 3

People’s landscape impressions cannot possibly contain all the information that exists in reality (Zube, Sell and Taylor 1982). The landscape is therefore a generalisation of reality. The notion that landscapes are an impression from a distance gives an indication of this generalisation. Landscape does not include small objects, such as snails, frogs and forest understory.

Landscapes cannot be simply classified by dividing the land into areas that reflect boundaries between different landscape attributes, such as landform and vegetation, because it is these boundaries that are important characteristics of landscapes (Jackson 1984). The interaction of components is sometimes more important than the components themselves (Arthur, Daniel and Boster 1977). A landscape classification, therefore, needs to incorporate this juxtaposition of boundaries between different components of the landscape in order for the classification to distinguish landscape quality. This consideration makes landscape classification particularly difficult, but if it is not included, the classification will not be useful for evaluation research. In one respect, landscape is a generalisation of the environment, because it includes only the large visual entities. However, when combinations of these generalised entities are considered, an unlimited number of different combinations become apparent.
 

Criterion 4

The last specific criterion (taken from Zube, Sell and Taylor 1982) suggests that from a point a landscape does not just include components that can be directly seen from that point, but also incorporates components that may have been viewed during exploration of the neighbouring region.

Other Possible Criteria

There are many aspects of the landscape that vary considerably, such as the weather, polluted air and the sun. It is not necessary to include these in a classification because they are often held constant in landscape research when different landscapes are compared.

The list of specific criteria may appear oversimplified considering the complex nature of landscapes. Many questions are left unanswered- what is the exact nature of the landscape components and their relationship with each other, and what is the exact nature of the observer? If more specific criteria were stated, their validity would be questionable because of the lack of landscape theory - "theoretical vacuum" (Appleton, 1975). More research is required. However, a landscape classification is needed so that researchers can communicate their results. The dilemma is that it is necessary to understand the nature of landscapes to classify. Yet, to understand the nature of landscapes, it is necessary to classify. A classification must be developed with the information available and reassessed as more information becomes available. It is in this way that classifications evolve. Currently, our understanding of landscapes is superficial and needs to be based on many assumptions.

Despite these oversimplified specific criteria, if a classification can meet all of them and the general criteria, a significant advancement has been made. To date, this has not been done successfully with manual processes. Complex definitions that incorporate the important characteristics of landscapes have been too tedious and costly to implement manually. Instead, intuition has been used to decide the appropriate classes for different areas. This has not been repeatable and does not provide a frame of reference for different researchers, which in effect makes the classification of little value for science. When manual classifications have used explicit definitions, these have often been simple so that they can be easily implemented. The result is that the classification does not incorporate the important characteristics of landscapes.

The landscape classification was developed using a Geographical Information System (GIS) and national topographic databases. The classification is based on the notion that the landscape character of an area can be determined from a range of hard-copy topographical maps. It is possible to identify what the landforms of an area are like using contour information. It is also possible to determine whether an area is developed and contains lots of roads, buildings and utilities. A map will also show whether an area is by a lake, river or the coast. It is also possible to consult more specialised maps such as vegetation and population maps, and these will influence the type of landscape perceived for an area. A good map-reader can develop an accurate impression of what the landscape will be like without visiting the site. Many of the latest topographical maps now exist as digital layers within a GIS, and it is possible to use GIS to automatically analyse these digital maps.

An important function within GIS that enables landscapes to be identified from topographical data sets is the focal neighbourhood function (Tomlin, 1990). A GIS can represent topographical data sets in a raster data model, whereby an area is divided into square cells and each cell is assigned a number that reflects the presents of a topographical feature. With a focal neighbourhood function, each cell within the specified coverage becomes, in turn, the center for processing. When a cell is being processed, the cell values within the specified neighbourhood of that central cell are included in the processing. The result of this processing is then assigned to the cell of a new grid coverage with the same position as the central cell, and the next cell in the first grid coverage becomes the center of the processing. This continues for all the cells available for processing. The neighbourhood can be of any shape, which can be directly specified in some GIS software or can be custom designed. Focal functions are particularly powerful for landscape classification. In effect, they assess the land from thousands of different points and can be used to capture the characteristics of the surrounding location of each point. For this reason, it is more suitable for landscape classification than the conventional areal analysis units. The spatial influence of many different landscape components can be assessed, and this can be used to identify landscape combinations. Duffield and Coppock (1975) used a focal function for identifying recreational landscapes - a commendable effort considering this was pre-commercial GIS.

The classification process analyzes New Zealand from about one million different points. From each point, it calculates the influence of a large range of different landscape components (different vegetation types, landforms, infrastructure and water components) in the surrounding area (3,000 m radius), with spatial distributions obtained from national topographical databases. Using this influence information, landscape classes are defined. The classification also has three levels of generalisation. For level one, 7,461 classes are identified and a polygon outline of this is shown in map 1. Map 2 and Map 3 show the polygon outlines of level 2 and 3, which have 1,732 and 339 classes respectively. The classification is constructed from the unique combination of landform, vegetation, naturalness, and influence of water, thesefore, the best way to comprehend the information available in this digital classification is to view the landform map, vegetation map, naturalness map, and water map separately. All the information seen in these maps is available in the digital classification. Tables 1, 2, 3 and 4 below show the classes included, as well as how the different levels of generalisation were derived.
 

Table 1 Generalisation of Landform Classes

Table 2 Generalisation of vegetation classes

Table 3 Generalisation of naturalness classes

Table 4 Generalisation of Water Classes

As previously stated, the main purpose of a landscape classification is to provide a frame of reference for public preference studies (psychophysical approach) that assess the quality of different landscapes. A landscape classification can also be used to map variety by calculating, for each point, the number of unique classes in the surrounding neighbourhood (map 8). Landscape variety has been used to assess landscape quality, although the validity of this has not been proven (Arthur, Daniel, and Boster 1977). In the past, landscape practitioners used a more intuitive approach, rather than quantitative techniques for assessing variety.

The classification can also be used to obtain statistical information because it provides an inventory of the resource. The digital representation of the landscape classification contains both the Regional Council and Department of Conservation conservancy boundaries, so statistics can be obtained on these regions. The benefit of the New Zealand-wide classification is that regional statistics can be put into a national perspective. Table 5 shows an example, using the Canterbury Regional Council boundary, of the type of statistics that can be generated. A selection is made using the level 2 classification for classes that are coastal. These selected classes are then listed showing the name of the class, the frequency of this class in the region (number of discrete areas), the total area of this class in hectares in the region and the percentage of the total area in New Zealand that exist in the region. Similar statistics can be easily produced for any other region in New Zealand.
 

These statistics can be used to map unique landscapes. Map 9 shows landscapes that are unique to the Canterbury region. This is based on an analysis of all the different types of landscapes in New Zealand. Uniqueness is dependent on scale and the level of generalisation - two parameters that are rarely specified by landscape researchers. The total area of each class can easily be determined with GIS for a given study area and level of generalisation, and this describes explicitly how unique the classes are. Why the study area and the level of generalisation were appropriate would have to be substantiated, but that is another matter. With GIS, uniqueness of an area can be calculated for a range of different scales and for a range of different levels of generalisation. Uniqueness of a particular landscape type does not necessarily mean that it is a valued landscape, as some highly modified landscapes can be quite unique. However, if a landscape is perceived as high quality and it is also unique, it becomes particularly important.

This landscape classification of New Zealand meets the previously mentioned criteria better than any other landscape classification developed to date. The fact that the classification process is totally explicit, yet sophisticated enough to incorporate the complex nature of landscape, makes this classification stand out from other classifications. The definitions and classes used can now be questioned, and the classification can evolve as our understanding of landscapes improves. The use of GIS has allowed consistent classification across the whole of New Zealand.

This classification, combined with public preference research, will enable planners to determine nationally significant landscapes as required under New Zealand law (Resource Management Act). The classification of New Zealand's landscapes enables development of sound management practices, which are more likely to withstand the scrutiny of planning law than current practises.

The impacts of different modifications to the landscape depend on the context of the landscape. A new hotel in an area that already has a few hotels might not make too much difference to the quality of the landscape. The vegetation and the landform also affect the impact, as hilly and forested areas may hide the visual impact of buildings. It is for this reason that the landscape classification incorporates vegetation and landform. A landscape classification that just considers degrees of modification would not be that useful.

The approaches used for developing the New Zealand landscape classification could be applied to other regions of the world if suitable data sets were available. Some modifications may be required to incorporate issues that may be important to a region; however, this would not pose a problem to a process that can be easily altered and reapplied. The author is currently working on a Welsh Landscape classification with the Wales Countryside Commission and the MaCaulay Landuse Institute in Scotland

Perceptual research into landscape quality would probably reinforce the value of unmodified landscapes such as wilderness areas. Everybody would agree that no wilderness area in the world is safeguarded against some form of development. Often the most contentious landscape planning issues are associated with landscapes on the fringe of wilderness status. The contention is often compounded by the lack of understanding over landscape impacts and public perceptions. Given that landscapes are so valuable to people and that there is so much contention, more research on landscapes is needed. This research should utilise a landscape classification to ensure that the research can be compared with other research and applied to other areas. A landscape classification improves communication and provides a sound framework for research.

Amedeo, D. Pitt, D.G. and Zube, E.H. (1989) Landscape feature classification as a determinant of perceived scenic value. Landscape Journal, Vol. 8, No. 1, 36-50.

Appleton, J. (1975) Landscape evaluation: the theoretical vacuum. Transactions of the Institute of British Geographers, 66: 120-123.

Appleton, J. (1980) Landscape in the arts and the sciences. University of Hull, U.K.

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

Countryside Commission for Scotland, (1970) A planning classification of Scottish landscape resources. Countryside Commission. Perth, Scotland.

Countryside Commission for Scotland, (1988) A review of recent practice and research in landscape assessment. Countryside Commission. Perth, Scotland.

Daniel, T.C. and Vining, J. 1983 Methodological issues in the assessment of landscape quality. In: Altman, I. and Wohlwill, J.F. 1983 Behaviour and natural environments. New York; Plenum Press.

Dearden, P. (1989) Societal landscape preferences: A pyramid of influences. In: Dearden, P. and Sadler, B. (1989) Landscape evaluation: Approaches and applications. Western Geographical Series. Vol. 25.

Duffield, B.S. and Coppock, J.T. (1975) The delineation of recreational landscapes: the role of a computer-based information system. Transactions Institute British Geographers, 66(1), 141-148.

Jackson, J.B. (1984) Discovering the vernacular landscape. Yale University Press, New Haven.

Jones, M. (1991) The elusive reality of landscape. Concepts and approaches in landscape research. Norsk Geografisk Tidsskrift, 45, 229-244.

Haines-Young, R.H. and Petch, P.R. (1986) Physical geography: Its nature and methods. Paul Chapman Publishing, London.

Rhind, D. and Hudson, R. (1980) Landuse. Methuen, London.

Swaffied, S.R. (1991) Roles and meanings of "Landscape". Ph.D. thesis, Lincoln University.

Tomlin, C.D. (1990) Geographic information systems and cartographic modelling. Prentice-Hall, New Jersey.

Zube, E. Sell, J. and Taylor, J. (1982) Landscape perception: Research, application and theory. Landscape Planning 9: 1-33.

Zube, E.H. (1984) Environmental evaluation: Perception and public policy. Cambridge University Press.