Landscape classification using GIS and national digital databases
Degree GrantorUniversity of Canterbury
Degree NameDoctor of Philosophy
This study considers whether visual landscape character can be classified using GIS. Landscape classification is needed to give landscape researchers and planners a frame of reference for communicating and comparing their research. Such classification is difficult because of the complex nature of landscapes and because it must be explicit. Classification needs to be based on theory, but there is a distinct lack of landscape theory. It is argued that to effectively develop landscape theory a classification is required and that a classification evolves with theory. GIS provides a suitable platform to facilitate this evolution. A set of criteria is established to which a landscape classification should adhere. To be useful for evaluative and cognitive research, a landscape classification needs to distinguish the important characteristics that affect landscape. These characteristics are identified from what little landscape theory exists: a landscape classification needs to incorporate landform, vegetation, naturalness, and water; the classes should be based on the public's perception; the classes should be general and involve compositions; and the classes should incorporate movement and exploration. Besides these criteria, more general criteria that have been used on other land based classifications also apply, particularly the need for a classification to be repeatable. GIS and national digital databases can incorporate these criteria in a landscape classification and this is demonstrated on a transect of the South Island of New Zealand, using mainly a 1:250,000 topographical database and a vegetation database. Difficulties associated with these databases are discussed. A three-phase landscape classification process is developed: 1) Selection of attributes, 2) Definition and classification of the attributes to six levels of generalisation, and 3) Creation of landscape classes from compositions of the attributes. The sensitivity of the process to different operational definitions is considered, and it was significant in some cases. An important analysis function that enables GIS to classify landscapes is the focal neighbourhood function. This in effect analyses the study area from many different points. Once a landscape classification is developed, it can be used with GIS for description, mapping, and inventory purposes. Uniqueness and variety of landscapes can also be determined. A range of observer perspectives can be recognized in the classification by using an application of fuzzy set theory that incorporates entropy. Automating landscape classification requires developing appropriate operational definitions that balance the human concept model of landscapes, the characteristics of national digital databases, and GIS capabilities. Operational definitions can be formulated using four abstractions: classification, generalisation, association, and aggregation, and then represented using GIS analysis techniques. Classifying landscapes automatically is an exercise in generalisation, as there is a considerable amount of information to consider. The challenge is to produce a meaningful generalised classification, rather than a very detailed classification. Expressing association is also important because landscapes are a composition of different landscape components. Focal neighbourhood functions enable the spatial influence of different components to be expressed and from this landscape compositions can be identified. The national digital databases used in this study do not contain conceptualised information on morphological landforms. Height contour databases are available from which it is possible to classify landforms and a substantial part of this study investigates this. Hammond's manual landform classification was automated and applied to the study area. Some problems were identified and a modified process was subsequently developed.