The Use of the Landscape Metaphor in Understanding Population Data
This paper discusses an approach to exploring population data using 2-1/2 dimensional surfaces and is illustrated using 1991 data for London.
For more information see:
Wood, J.D., Fisher, P.F., Dykes, J.A., Unwin, D.J. and Stynes, K. (1999) The use of the landscape metaphor in understanding population data. Environment and Planning B: Planning and Design 26, pp.281-295
Introduction
Recently, continuous surface representations of population density have been shown to have advantages in the visualisation and analysis of population distributions (Langford and Unwin, 1994; Martin and Bracken, 1991). Here we explore both the visualisation potential of such surfaces and the use of a landscape 'metaphor' as a way of understanding the data.
We argue that a relief metaphor, together with concepts and techniques developed for the analysis of real terrain surfaces, provides interesting analytical and visual insights into population distributions. Ideas are illustrated using population density data for Greater London. These data are at a 200m resolution provided by the spreading and aggregation technique developed by Martin and Bracken (1991).
Visualising Density Surfaces
One of the problems of visualising population density is that it varies greatly over small distances. This manifests itself as `spikes' in the surface which when viewed from above have little expression (Figure 1), but when viewed obliquely, can produce a confusing spikey surface (Figure 2).
Figure 1 - Greater London Population Density - Overhead View "at night".
Reduced image or Full image
Figure 2 - Greater London Population Density - Oblique View.
Reduced image or Full image
Animation 1 - Transformation between overhead and oblique views of Greater London Population Density.
MPEG animation.
It becomes necessary to transform the data in some way to produce effective terrain-like surfaces. This may be done by transforming the density data into alternative measures of concentration (eg linear density measure), and/or by passing some form of smoothing convolution filter over the surface.
Figure 3 shows an alternative measure of density - linear density, which is approximately equal to the square root of areal density. The effect of this transformation is to normalise a very positively skewed distribution (see Figure 5).
Figure 4 shows the effect of smoothing using quadratic generalisation with a kernel of 1.25km and distance decay exponent of -1.0. The Thames is overlain in order to provide some context. Animation 2 shows the effect of the degree of smoothing on the resultant surface.
Figure 3 - Linear density, unsmoothed.
Reduced image or Full image
Figure 4 - Population density, smoothed using 1.25km kernel.
Reduced image or Full image
Figure 5 - Frequency histograms of alternative density measures.
Reduced image or Full image
Animation 2 - Effect of quadratic smoothing on density surface.
MPEG animation.
r.param.scale - GRASS 4.x source code for quadratic interpolation (written in C).
Source code.
Morphometric Parameters
The first and second derivatives of a continuous surface can be regarded as morphometric parameters. Figure 6 shows the `slope' map of the London area which emphasises the concentric pattern of population distribution. Note the relationship between high slope values and constraints on development (river valleys and greenbelt in particular). Figure 7 shows the second derivative - `curvature' which tends to emphasise radial patterns. In particular, note the relationship with the rail network and railway stations.
Figure 6 - The rate of change of population density - "slope".
Reduced image or Full image
Figure 7 - The second derivative of density - "curvature".
Reduced image or Full image
Morphometric Features
Morphometric parameters can be combined to classify the elements of a surface into one of six types: pits, channels, passes, ridges, peaks and planes. These feature types correspond with features of urban morphology, in particular they help define local neighbourhoods and the boundaries between them. Figure 8 shows the relationship between these features and existing district boundaries. Note how the notion of London as a collection of 'villages' emerges in this visualisation.
Figure 8 - Morphometric features of the population surface.
Reduced image or Full image
Geomorphological Features
The metaphor of the population density landscape can be taken a step further by measuring some of the properties that are traditionally associated with geomorphology and hydrology. Figure 9 shows the `inverted drainage basins' of the surface. These correspond to local neighbourhoods and can be compared with existing district boundaries.
Figure 9 - "Inverted drainage basins" - population spheres of influence.
Reduced image or Full image
Other Surfaces
The convenience of the raster processing used to generate these images means that any other census variables may be examined in a similar way. Figure 10 shows the proportion of people not working due to illness. Looking east along the Thames, the draped slope map highlights areas of rapid change. Figure 11 shows the same techniques applied to US census data at an entirely different scale. Here the midwest counties of the United States are shown with the surface representing the proportion of the population holding college degrees. As well as the major urban centres, many of the university towns within the region are identifiable.
Figure 10 - Illness slope map (looking east along the Thames).
Reduced image or Full image
Figure 11 - Proportion of population with college degrees - midwest United States.
Reduced image or Full image. Contextual information.
Discussion
This brief overview aims to illustrate how visualising the distribution of people as a landscape can tell us something about the urban landscape they occupy. In the case of London, three important factors in its development have been revealed through this metaphor. Constraints on development both physical and regulatory have been shown with the slope map of London. The development of the city along transport routes (railway lines in particular) has been revealed by the curvature map. The amalgamation of villages into the urban structure has been shown with the feature map and inverted drainage basins.
For a more detailed analysis and discussion see full paper in Environment and Planning B.
Acknowledgements
Population, illness and retiement data were all taken from the 1991 Census, Crown Copyright ESRC/JISC purchase. The original ungeneralised surface data used in this work were generated by David Martin and the late Ian Bracken and were obtained from the University of Manchester Computing Centre (Manchester Computing). Rail, district boundary and river lines were provided by and are copyright Bartholemew UK Ltd. Railway station data were kindly provided by John Shepherd, SE Regional Research Laboratory.
This work was carried out as part of Project Argus - Visualisation Toolkit for the Spatial Sciences.
High quality figures are also available for download.