A place in history: a guide to using GIS in historical research

CHAPTER 2: THE WORLD AS VIEWED THROUGH A GIS

2.4 Raster systems

The raster data model sub-divides space into square pixels or other regular tessellations to provide a continuous representation of the study area rather than subdividing it into discrete points, lines, or polygons. For example, to represent relief each pixel might have its height as an attribute; to represent land use each pixel would have a land use class attached to it, and so on. Although less likely to be of use to historians, satellite images are a more complex form of raster data. On the image the earth's surface is sub-divided into pixels with each pixel storing information about the amount and type of light being reflected by that part of the earth's surface, for example, how much green, how much blue, how much red, and how much infra-red.

Figure 2.4: The raster data model
The source map of the study area shows three landuse classes: agricultural, urban and forested. The raster data model sub-divides the study area into pixels with each pixel being allocated to the landuse that covers the highest proportion of its area. Each pixel is given the number 1, 2 or 3 representing agriculture, urban or forest respectively.

The simplest raster file formats are two-dimensional arrays in which each value corresponds to a pixel on the grid that the array is modelling. The header information provides data on, for example, pixel size and the location of the bottom left-hand corner of the grid, while the remainder of the file simply consists of the values for each cell. Figure 2.4 shows an example of this. The original map shows three types of landuse: agriculture, urban and forest. The raster sub-divides the study area into pixels with each pixel being allocated a numeric value representing landuse class. The pixel is allocated to the type of land-use that covers most of its area. Obviously the choice of pixel size is very important to the model. Too large a size will lead to a poor representation of the features, too small a size will lead to unwieldy file sizes. More sophisticated methods, such as run length encoding and quadtrees for example, can be used to compress the file sizes, but the basic model remains the same. For more information on the details of different raster structures see, for example, chapter 3 of Heywood et al ( 1998).

In general, raster data are more suited to environmental applications while vector data are more suited to human activity. Raster systems model complex spatial patterns with limited attributes, such as land-use patterns very well, while the vector data model is better for more clearly defined space with complex attributes such as census data. There are exceptions to this: Martin (1996b) uses derived raster surfaces to model 1981 and 1991 census data and compare change between the two, claiming that the raster model provides a more realistic model of the underlying population distribution. A good example of a raster system in a historical context is provided by Bartley and Campbell (1997). They examined the Inquisitions Post-Mortem of the 14th century and used these to create a raster GIS of medieval land-use that, they claim, is potentially the most detailed survey possible until the 19th century tithe surveys. A raster system was used because it provides a complete coverage of the land area, it provides a more realistic representation of land-use than polygons, and because it handles the inaccuracies of the sources better than polygons.