CHAPTER 1

The use of manual methods for classifying the important characteristics of visual landscape has been well documented (Countryside Commission, 1970). They have been driven by the need for landscape evaluation.

A landscape character classification is fundamental to landscape research because it provides an important frame of reference for researchers to communicate and compare their work. Landscape research is needed not only to understand landscapes but also for landuse planning. In particular, planners need to know how development can be incorporated within the landscape so that it does not unduly compromise the perceptual quality of the landscape. Despite this need, manual landscape classification has had very little success because of technical and cost issues. The classification of landscapes is a complex problem that has yet to be sufficiently resolved because of the complex nature of landscapes. The principal research question that will be investigated in this thesis is whether Geographical Information Systems (GIS) and national digital databases (NDDB) can be used to classify landscape character. This study will focus on the classification problem, rather than on issues of landscape evaluation. 

It appears viable from past work (such as Duffield and Coppock, 1971, Dikau et al., 1991, and Lay, 1991) that recent developments in GIS can partly solve this landscape classification problem. This will be investigated by exploring different options with the use of GIS tools, and by developing an automated process that classifies landscapes. This process will be demonstrated on a transect of the South Island of New Zealand that has a wide range of landscapes. A set of criteria will be established for assessing the validity of a landscape classification. This will consider the important characteristics of landscapes, as well as general classification principles. This thesis shows how GIS and NDDB can revolutionize landscape modelling, and explores some interesting theoretical issues.

Inadequate information on the visual landscape is now a major concern in New Zealand as impacts on the landscape are one of the most controversial environmental issues resulting from development initiatives (Jackman, 1988). This is particularly the case with respect to two of New Zealand's main growth industries - tourism (Collier, 1991), and forestry (Kilvert and Hartsough, 1993). It can be argued that landscape perception needs to be integrated with other landuses to maximize the total value to society. The value of the landscape can be easily compromised by different landuses. In New Zealand, the booming tourism industry, although dependent on the landscape, is actually changing it through the construction of hotels, gondolas, roads, and other infrastructure. If this is not carefully planned, it could diminish the landscape resource that it is dependent on. Commercial forestry is another example of humans altering the landscape on a large scale. Although the scale of indigenous logging in New Zealand has substantially diminished in the last decade, exotic plantation forestry is expanding. The establishment of exotic plantations changes the character of the landscape. This may be having significant consequences on the landscape and its associated values. The Marlborough Sounds is an example where this is happening, and the Mackenzie basin is an example of where it could happen if proposed forestry plans are accepted (Boffa Miskell, 1993). Research and monitoring are required. Whether landscape values are significantly compromised by different landuses depends on the landuse in question, the landscape context, the spatial context, and the observers of the landscape. Some landscapes are more sensitive to development than others due to their proximity to tourist circuits or urban recreational areas, or because they are regarded as natural. This sensitivity to development depends on whose perspective, for instance the developer or the conservationist. Because of all these considerations, research on landscape values is complex, yet essential. Leopold (1969) argues that quantitative data on landscapes are required in order to empower their protection from conflicting landuses. Often landuses that conflict with landscape values are proposed by developers who employ strong quantitative arguments, while the value of landscapes has been dependent on emotional pleas from environmentalists. Leopold's view in 1969 was that environmentalists should begin to support their arguments with numbers. This view is still valid today. The Resource Management Act 1991 makes it a statutory requirement for regional councils to monitor and provide information on New Zealand's significant landscapes, and makes provisions for their protection. Thus, resource managers, developers, and conservationists require landscape information. The utilization of Geographical Information Systems (GIS) and national digital databases appears to offer an effective method for providing parts of this information.

In the last ten years there has been a dramatic change in the utility and power of Geographical Information Systems (GIS), because of advancements in computer hardware, as well as improvements in the GIS software. Closely linked with this advancing GIS technology is the increase in the amount of digital data available to be analysed. This is often referred to as the "fire hose" of data (Maguire, 1991). Significant improvements in automated data capturing devices, such as satellite scanners, airborne scanners, Global Positioning Systems (GPS), and office scanners and digitisers, have dramatically increased the amount of digital data available for describing and monitoring the environment. An inventory of available digital databases in New Zealand was compiled by the Department of Statistics (1992). This inventory reveals the significant amount of data available for reporting on the state of the environment. The challenge is to analyse and present this data so that it becomes useful information for decision makers. GIS can play an important role in this.

Perhaps the most significant databases now available are the topographical databases developed by mapping agencies all over the world. Because of advances in GIS and automated cartography, standard topographic maps are now being produced in digital format. This means that topographical maps, covering extensive regions, can now be analysed using GIS. In fact, the whole world can be analysed using global databases, such as the Digital Chart of the World (Environmental Systems Research Institute, 1993). Complex spatial queries over extensive areas can now be implemented automatically with a computer, as in Dikau et al.'s (1991) attempt to classify the landforms of the state of New Mexico. From reviewing such works and from personal experience with GIS, any measurement that can be derived manually from assessing a map can now be derived automatically. Moreover, because GIS can do billions of spatial measurements in short periods, there are some parameters that a GIS can obtain quantitatively from a map that would be impossible to obtain manually because of practical constraints. Considering the importance of maps and the spatial analysis of maps to geography, such technology ought to be a powerful tool for landscape classification. This thesis develops and demonstrates this tool.

Landscape evaluation is an important end use for a landscape classification. Classification is important for the implementation of public preference surveys that ascertain landscape quality, because it provides a frame of reference that enables different research initiatives to be communicated and compared. A landscape classification can also be used for assessing landscape variety and uniqueness, which will be demonstrated using GIS once a classification has been devised. In fact, landscape classification is important to all forms of landscape research because it helps organise our understanding of landscapes and provides a means for communicating about different types of landscapes (Countryside Commission for Scotland, 1970). The basic rationale for this study is to compare the amenity values of scenery against other resource considerations. Landscape research is necessary for improving resource inventories, making carrying capacity decisions, and assessing environmental impacts.

The classification of the landscape is a particularly difficult spatial analysis problem. Landscape is defined as the appearance of the land (Swaffield, 1991). On the one hand, the landscape is a generalisation of the environment because only the larger objects are perceived. However, it also includes the composition of objects, and this makes landscape considerably diverse and complex (Jackson, 1984, and Robinson et al., 1976). This is further complicated by the fact that different observers view the landscape differently (Bourassa, 1991). In addition, classification must be based on explicit definitions (Rhind and Hudson, 1980). Even though landscapes are heterogeneous in nature, it is necessary to identify homogeneity in order to classify them. This is in common with all resources. People identify homogeneity to make sense of reality, and to describe and communicate realities. Evidence of people's cognitive landscape classification is demonstrated by common words, such as "coastal", "mountainous", or "flat", which are, in effect, describing landscape classes.

To attempt to define landscape classes explicitly to a level of sophistication that incorporates the important characteristics of landscapes requires sophisticated quantitative definitions that are too difficult to implement manually. Quantitative manual methods instead have used simple definitions that do not capture the important attributes of the landscape. For instance, the Manchester evaluation method attempted to classify landforms by counting the number of contours in a one kilometre grid cell (Penning-Rowsell and Searle, 1977). It will be shown that the landform features important for landscape classification cannot be accurately defined in this way. More commonly, landscape classification practitioners have avoided quantitative definitions, and instead used more intuitive approaches, as in the Auckland Regional Authority (ARA) landscape study (ARA, 1982). The intuitive approach suffers because it cannot be repeated by different practitioners, making it difficult to compare landscapes in different regions. Considering that the main purpose of landscape classification is to provide a frame of reference for communication and for describing and comparing landscapes, this is a severe limitation. In comparison, GIS approaches are totally explicit and repeatable.
 
 

With GIS, it appears that sophisticated quantitative definitions of important landscape characteristics can be implemented and applied to extensive areas. GIS has been used for analysing related phenomena such as cliffs and farms (Barbanente et al., 1992), visibility (Miller, et al., 1994), wilderness (Lesslie et al., 1988, and Kliskey and Kearsley, 1993), and extracting terrain information (Lay, 1991, Cowen, 1993, Dikau, 1989, Tang, 1992, and Weibel and DeLotto, 1988). These works are useful, not only because the features studied are an important part of landscape, but also because the techniques and structural frameworks that they use can be applied to landscapes. However, when landscapes are classified as a whole, generalisation becomes a complex issue. Past research that has concentrated on individual components has not had to develop classes that are overall impressions of many different components, therefore many issues remain unresolved. Automated cartography literature on generalisation (Shea, 1991) and semantic data models (Nyerges, 1991) also provides useful frameworks that can be incorporated in an automated landscape classification. Since automated landscape classification is relatively new, dating from the release of commercial GIS in the late 1980s, it is necessary to bring together many fields of study that have some expertise in different aspects of automation. Mitchell (1993), and the Countryside Commission (1988) have commented on GIS as a possible future direction for landscape research, but there does not appear to have been any research initiative that tackles the application of GIS to landscape classification directly and fully.

The information in NDDB that can be used for landscape classification includes roads, railways, transmission lines, rivers, lakes, coastlines, and contours, which are all available from topographic databases. Also obtainable are vegetation classes from specialized vegetation databases, and population information from census databases (Supermap2). If GIS and NDDB prove to be valuable tools for landscape classification, then this could have important implications for the development and use of NDDB. The amount of information (in different layers) within NDDB has generally been kept to a level that can be adequately displayed at the scale mapping agencies publish their hard copy maps as these have often been the primary source of information. Yet, GIS can analyse information that is much more detailed. The data models used for NDDB have mostly been in vector format, but GIS can also use raster format, which is perhaps better for some spatial analyses within a GIS. It appears that significant improvements can be made to NDDB to realise the full potential of automated spatial analysis.

Chapter 2 presents the research problem of this project. This includes the meaning and complex nature of landscapes, a consideration of why landscapes need to be researched and classified, a brief outline of landscape research, and a list of criteria that a landscape classification should comply with.

Chapter 3 frames automated landscape classification as an operational definition problem. In a GIS context, Lay (1991) identifies three factors that need to be balanced with operational definitions: the human concept model (for landscapes this is discussed in chapter 2), characteristics of the digital databases, and GIS capabilities. A brief overview of GIS capabilities is given, followed by a detailed description of focal neighbourhood functions as these are important for landscape classification. Appropriate databases are then discussed and identified. This discussion on operational definitions incorporates theoretical input from automated cartography literature. Although this has a different objective to landscape classification, both are concerned with automated abstraction of structural geographical meaning. Nyerges (1991a and 1991b) identifies four important types of abstraction: classification, association, generalisation, and aggregation. To classify landscapes, these abstractions need to be represented using GIS functions. National digital databases contain geographical meaning, though the objects within them can be further formulated to identify even more complex geographical meaning, such as landscape classes. With landscape classification the most difficult abstractions to represent are generalisation and association, and these are discussed in detail. Chapter 3 also describes the method of investigation, discusses validity, and introduces the study area.

The process for classifying landscapes is subdivided to simplify the task. Vegetation, naturalness, and water are classified in chapter 4. Many characteristics of these three landscape attributes are already conceptualised in existing databases. In chapter 5, landform is classified. This is a complex process because a contour coverage must be conceptualised. Chapter 5 also introduces an application of fuzzy set theory. Chapter 6 combines the vegetation, naturalness, water, and landform classifications to produce a landscape classification. The validity of this resulting landscape classification is then discussed using criteria established earlier in the thesis. Conclusions then follow in the last chapter.