Causes And Consequences Of Map Generalization

Elsa Joao

• Publisher: CRC Press
• Year: 2020
• ISBN:
  9781000162813
• Edition: 1
• Topic: Ciencia de la computación
• Subtopic: Procesamiento óptico de datos

This text describes late-1990s understanding of map generalisation in the context of paper maps and GIS. Its particular value should be in helping to further automate and measure the process of map generalisation.; The research has concentrated on quantifying generalisation effects and on analysing how these effects of generalisation locked into the maps were measured. Elsa Joao's book covers the background to the problems of map generlasation; the methodology developed by the author to investigate the consequences of the map generalisation; a detailed description of results, and a conclusion that draws together consequences for the broader applications to GIS.

CHAPTER ONE
Scale and generalisation in GIS

Generalisation is an inherent characteristic of all geographical data. All maps, whether digital or analogue, are generalised representations of reality; and the more generalised a map is, the more remote from reality it becomes. Generalisation denotes a process by which the 'presence of phenomena or events in a referent space are essentially reduced and/or modified in terms of their size, shape and numbers within map space' (Balodis, 1988, p. 71). The end product of this generalisation process is a derived data set with less complex and usually more desirable properties than those of the original data set.

However, the generalisation process results in a dilemma. On the one hand, it is necessary to generalise in order: to improve the display quality of a map at a scale smaller than the one from which it was compiled; to allow analysis with different degrees of detail; and to reduce data storage requirements. On the other hand, generalisation also causes unintended transformations of the data (such as changes in measured lengths and areas) that can alter the topology of geographical phenomena and affect subsequent statistical or geometrical calculations. Normally, the user would wish to minimise the unforeseen and unqualified effects of generalisation. It is the study of these generalisation effects that is the subject of this book.

The manner in which generalisation affects the statistical and geometric properties of spatial data is fundamental to the use of Geographical Information Systems (GIS). These are computer-based systems that are increasingly being used to store and manipulate geographical information. The main advantage of GIS over paper maps is their ability to carry out sophisticated and extensive analysis, and their success in doing so depends on keeping control of unintended data transformations. If the results from combining two or more data sets are spurious because of the poor quality of the data ~ rather than of the situation in the real world that the data are supposed to represent then all subsequent interpretations and actions based upon them are liable to be flawed. The oft-quoted advantage of GIS as a means of data integration becomes problematic when data compiled from different sources or scales fail to match in a logical fashion (Rhind and Clark, 1988), leading, for instance, to the improbable situation of soils supposedly located in the sea, or river channels located along the side of a ravine. Unfortunately, most generalisation errors are more difficult to spot.

The research in this book measures the magnitude of generalisation effects already 'locked into' paper maps and compares these with transformations by GiS-based generalisation algorithms. By carrying out analyses using different source scale data, it is possible to quantify just how important these generalisation effects can be. On the basis of this information, ways of improving the generalisation process within a GIS are proposed. Although this work has relevance to conventional cartographic techniques, it is expected to be more pertinent to GIS and geographical database use. The ultimate objective of this work is to provide the key by which future GIS users will be able to generalise geographical information according to their own need and criteria of quality but to carry this out via an automated procedure, so that the user's possibly limited skills are not a constraint - and to quantify the effects of generalisation on the data and subsequent analyses. Finally, in this book attention is drawn to the best ways of applying generalisation within a GIS, by controlling the unintended consequences of generalisation.

A multitude of factors are involved when generalising from one spatial data set to another data set with less detail:

• ■ The purpose for which the map is going to be used.
• ■ The geographical region that needs to be mapped.
• ■ The original and the final map scale.
• ■ The particular individual undertaking the generalisation.
• ■ The taste and knowledge of the user or client.

All of these elements will shape the form of the final product, and will determine the type and the amount of generalisation effects 'locked into' the maps. One of the most important factors though, in determining which features are maintained and how much they are modified, is the scale transformation.

1.1 Scale and resolution

Almost always, maps take up less physical space than the area that they represent. This reduction in size is reflected by the scale of the map, and is usually represented as a ratio or a graphic scale. The ratio indicates to the map user the number of metric units on the ground that are represented by a unit in the space of the map model. Scale is considered to be the single most crucial mathematical feature of any map and, because of its importance, scale is often used as a primary means of categorising maps from large-scale maps of 1: 10 000 to atlases of 1: 1 million.

Often associated with scale is the term resolution: 'The resolution of a data set defines the smallest object or feature which is included or is discernible in the data' (Goodchild, 1991, p. 113). Although scale and resolution have distinctive meanings, they are closely linked because for each map scale there is a lower limit to the size of an object that can be usefully shown on a map. For topographic maps and dark printing colours, the minimum size of a point that is still discernible to the human eye has a diameter of 0.25 mm (Swiss Society of Cartography, 1987). On the basis of this minimum resolution, the minimum size of the objects that can be shown true to scale on a map will vary for different scales. For example, at the scale 1: 10 000 the smallest object would be 2.5 m long, while at the scale of 1: 250000 the minimum size increases to 62.5 m.

'It is impractical or functionally impossible to collect data using a one-to-one correspondence between cartographic entities and objects: resulting maps would be a replication of the real world, not a model of it' (Cromley, 1992a, p. 132). The choice of the scale at which data should be collected needs to be determined at the outset as it must reflect the objectives of the map-maker, the precision of the instruments used in the original survey, and the type of spatial processes being modelled and analysed. The resolution of the instruments used sets the limit, because it is not possible to measure features more precisely than the equipment allows.

When the scale of a map is decreased, there is less physical space in which to represent the geographical features of a region. As the process continues, the features will need to be exaggerated in size in order to be distinguishable at a smaller scale. As geographical features 'fight' for representation in the reduced map space, some features will need to be eliminated, and those remaining may be further simplified, smoothed, displaced, aggregated or enhanced. In the extreme case, the map loses its geometric properties and becomes a caricature. Choice of scale, therefore, 'sets a limit on the information that can be included in the map and on the degree of reality with which it can be delineated' (Robinson and Sale, 1969, p. 15). Thus objects such as houses could be reproduced readily at 1: 1250, 1: 2500 or even 1:10000 scale, but the constraints of human drawing and perception ensure that a number of the houses have to be grouped together into blocks, or eventually urban areas, for depiction at smaller scales.

With paper maps, different scales have always been represented by physically distinct maps. However, with the advent of GIS, to an extent, the representations have become independent of scale (Frank, 1991). With GIS it is possible to 'zoom in and out' of the map, producing a continuous series of scales as desired. However, the scale at which the original data were digitised is still a limiting factor for the amount of features, shapes, level of detail, and so on, that the user can see. When the user 'zooms in' beyond this scale, more detail does not miraculously appear; eventually, all that remains is a blank screen. Conversely, 'zooming out' from a map can result in too much detail being presented, and finally the collapse of all features into a single point. More importantly, the generalisation effects contained within the manually generalised paper map are 'locked into' the digital versions of the maps. As digital maps are usually converted from traditional paper maps, at the very least, their intrinsic generalisation is determined by the manual cartographers who prepared the original versions of those maps. Therefore, the source scale of any digital map remains a crucial defining characteristic and one which will ultimately affect the map's accuracy, precision and quality control.

1.2 The cartographer, the map and the region

Most generalisation is carried out manually by trained cartographers, their work and research being driven primarily by the need to show features at a map scale much smaller than that at which the information was originally presented. The cartographers' experience, natural intuition, possible knowledge of the area and taste all influence how they generalise a map; in other words, their selection of how to draw what, with what priority and detail, and which attributes to display as text. Manual generalisation is accomplished using basic cartographic rules and techniques, and the cartographer's ability to view the map as a whole and for what it represents.

Because individual cartographers use their own knowledge and judgement to carry out generalisation, the practice of manual generalisation is often described as subjective (Keates, 1989). The subjectivity of the process is exacerbated by a further aspect that cartographers have frequently evoked - the need to produce an end result that has some aesthetic qualities (Robinson, 1989). Consider, for instance, the views of Eduard Imhof (1982, p. 86), former President of the International Cartographic Association, on producing the perfect map:

. . . the greatest possible accuracy, with respect to the scale of the map; clear expression of metric information; good characterisation in the forms; the most naturalistic forms and colours; the greatest possible clarity of meaning and good legibility, simplicity and clarity of graphic expression; and finally, summarising all these qualities, a beauty peculiar to the map itself.

Generalisation as art has been proposed since the beginning of the twentieth century. According to Eckert (1908, pp. 346 7), 'As long as the scale allows the objects in nature to be represented in their true proportion on the map, technical skill alone is necessary. Where this possibility ends, the art of the cartographer begins. With generalisation, art enters into the making of maps'.

The way in which cartographers work will reflect the cartographic traditions of their country or the specific rules of the mapping agency. The map purpose will also affect the underlying classification of which features are the more important. Obviously, a road map will give predominance to roads and will be less concerned with the way in which rivers are portrayed. The magnitude and type of the generalisation process also changes with the geography of the region. The concentration of features and the type of spatial data influence how generalisation affects each feature. The cartographer, for example, is likely to want to retain an isolated building in the countryside, while the same sized building might be removed if located in a town. A region with a large concentration of rivers might proportionally have more rivers eliminated than a region with only a few rivers present. Also, the more densely that features are concentrated together, the more they need to be generalised. The classical example of this is the case in which a river, a railway and a road pass through a narrow valley. In this case, one or more of these features needs to be displaced at the smaller scale, in order that all features can be displayed without overlapping each other. This displacement would not be necessary if all of the features were situated far apart.

Map generalisation as done by manual cartographers requires adaptability to all of these different circumstances, from the characteristics of the particular geographical region that needs to be mapped to the differing purposes for which the maps are going to be used. Because of the multitude of factors that need to be taken into account in order to generalise, it would seem that a computer-based system (such as a GIS) could potentially offer a good solution to the problem. A computer-based system would also have the added benefit of an iterative capacity the ability to redo a process until an acceptable solution is found. However, although there is potential for GIS to improve cert...