4.

Visual Analysis

Chapter 4
Visual analysis is the first step of spatial analysis.
Visual Analysis
It greatly helps us find interesting or important patterns
in the distribution of spatial objects, scalar and vector
fields, relationship between spatial distributions, etc..

4. Visual Analysis

You can visualize spatial objects in various ways. One

method is to visualize them in their original form.

However, conversion of spatial objects into a different

form often makes it easy to detect spatial patterns in
complicated spatial phenomena, which is the main subject
of this chapter.

http://www.ph.ucla.edu/epi/snow.html
Figure: Dr. John Snow’s Map of London (cases of cholera, 1854)

4. Visual Analysis 4. Visual Analysis

• References - visualization

1. Monmonier, M. (1993): Mapping It Out : Expository 5. Robinson, A. H., Morrison, A. L., Muehrcke, P. C.,
Cartography for the Humanities and Social Sciences , Kimerling, A. J., and Guptill, S. C. (1995): Elements of
University of Chicago Press. Cartography, John Wiley.
2. MacEachren, A. M. (1994): Some Truth with Maps: A 6. Kraak, M.- J. and Ormeling, F. (1996): Cartography:
Primer on Symbolization and Design, Association of Visualization of Spatial Data, Addison-Wesley.
American Geographers. 7. Slocum, T. A. (1998): Thematic Cartography and
3. MacEachren, A. M. and Taylor, D. R. F. (1994): Visualization, Prentice-Hall.
Visualization in Modern Cartography, Pergamon. 8. Campbell, J. (2000): Map Use & Analysis, McGraw
4. MacEachren, A. M. (1995): How Maps Work, Hill.
Guildford Press.

1
 4. Visual Analysis

4.1 Visualization of point distributions

Point is the most fundamental spatial object in GIS. There

are many ways of visualizing point distribution.

Figure: Point distribution

4. Visual Analysis

4.1.1 Quadrat method

Quadrat method converts point data into raster data by

1. overlaying a square lattice, and

2. count the number of points in each cell.

Quadrat method is thus a kind of data aggregation.

Figure: Quadrat method

4. Visual Analysis 4. Visual Analysis

Limitations of quadrat method 4.1.2 Kernel method

Though quite simple and easy to perform, quadrat Kernel method overcomes the first limitation of quadrat
method has two limitations. method. Kernel method generates smooth surfaces by

1. The distribution represented by raster data is discrete, 1. putting small bumps called kernels on the points, and
so it does not look nice. 2. visualizing the accumulated small bumps.
2. The result depends on cell size. Users may have to try
various cell sizes until they obtain a good result.

2
 4. Visual Analysis 4. Visual Analysis

• References - kernel method

Kernel method was originally developed for estimating 1. Silverman, B. W. (1986): Density Estimation, Chapman
the probability density function from a set of observed & Hall.
data (sample data). 2. Scott, D. W. (1992): Multivariate Density Estimation,
John Wiley.

4. Visual Analysis 4. Visual Analysis

Kernel functions

z i: Location of point i 1. Gaussian (normal) kernel

k(x, t): Kernel function at x generated from a point at t
1 ⎡ 1 2⎤
k ( x, t ) = exp ⎢ − 2 x − t ⎥
2πσ 2
⎣ 2σ ⎦

4. Visual Analysis

2. Epanechnikov kernel (a part of paraboroid)

⎧3
( )
2
⎪ 1− x − t if x − t ≤ 1
2 2

k ( x, t ) = ⎨ π
⎪⎩ 0 otherwise

Figure: Gaussian kernel

3
 4. Visual Analysis

3. Triangle kernel (quadrangular pyramid)

Though this kernel function has a simple shape, its

mathematical representation is rather complicated.

Figure: Epanechnikov kernel

4. Visual Analysis

A point distribution is converted into a surface function

by

∑ k ( x, z ) i
f ( x) = i

∑ ∫ k ( t, z ) dt
i
i

Figure: Triangle kernel

4. Visual Analysis

The result depends on the shape of the kernel function,

especially the steepness of its slope.

If we use a gentle kernel, we obtain a smooth surface,

which is suitable for understanding the global structure of
point distribution.

A steep kernel yields a rough surface, which is suitable for

looking at local variation of point distribution.

Figure: Kernel method

4
 4. Visual Analysis

This property is quite similar to the effect of cell size in

quadrat method.

If we use large cells, we have a smooth, though not

continuous, surface. Large cells correspond to gentle
kernel. Similarly, small cells yield rough surface as steep
kernel does.

Figure: Rough surface obtained by a steep kernel

Figure: Smooth surface obtained by a gentle kernel Figure: Surface can become smoother

Figure: Customer distribution of a shopping street Figure: Expected customer distribution of a new supermarket

5
 4. Visual Analysis

Quadrat and kernel methods are often used to hide the

exact locations of points. In crime maps, for instance, the
distribution of crime density is shown instead of the
distribution of crime occurrences. Census data are
provided in raster format in Japan, in order to keep
individuals' privacy.

Figure: Crime occurrences (snatch) in Shinjuku

Figure: Crime occurrences (snatch) in Bunkyo-ku Figure: Crime occurrences (snatch) in Kyoto

4. Visual Analysis

4.1.3 Visualization of attributes of point objects

Quadrat and kernel methods are used for visualizing

‘spatial information’ about point objects.

Point objects, on the other hand, also have ‘aspatial

information’, that is, attribute data.

Figure: Population distribution in Kanto area

6
 4. Visual Analysis

To visualize attribute data of point objects, we use point

symbols. Point symbols have four properties:

1. shape,
2. size,
3. color, and
4. orientation.
Shape Size Color Orientation
Difference in attribute data is represented by that in the
properties of point symbols. Figure: Four properties of point symbols

4. Visual Analysis 4. Visual Analysis

Two relationships of symbol size

Size and color are used for visualizing numerical variables. In principle, point symbols get larger with an increase of
Larger symbols usually correspond to (relatively) larger attribute value.
values of an attribute variable.
There are two quantitative relationships between symbol
Shape is used for categorical variables. Difference in size and attribute value that are frequently used in GIS.
shape is associated with qualitative difference of spatial
objects.

Orientation is used for wind direction, flows, etc..

4. Visual Analysis

-10,000
1. Proportional method
-25,000
The size (area) of point symbols is proportional to the
attribute value. -50,000

-100,000

Figure: Proportional method

7
 4. Visual Analysis

-10,000
2. Perceptual method -25,000

Symbol size is determined so as to fit the human -50,000

perception of symbol size.

Perceptual method exaggerates the difference in symbol

size than proportional method. For instance, symbol size -100,000
is proportional to the square of attribute value.

Figure: Perceptual method

4. Visual Analysis 4. Visual Analysis

4.2 Visualization of spatial tessellation

In commercial GIS, perceptual method is used more Spatial tessellation is a set of spatial units that extensively
frequently than proportional method. cover a bounded region. In this sense it is a special case of
a set of polygons.

Typical examples include census tracts, administrative

units, and school districts.

4. Visual Analysis 4. Visual Analysis

Spatial tessellation is a very important spatial structure in ‘Spatial information’ about spatial tessellation is
GIS and spatial analysis, because a lot of detailed data visualized by drawing boundaries of spatial units.
including census data are aggregated by spatial units to
ensure confidentiality when they are opened to the public. Attribute data include numerical variables, say,
population density and average income, and categorical
variables such as land use category.

8
 4. Visual Analysis

Choropleth map

To visualize attribute data of spatial tessellations, a map

called ‘choropleth map’ is used. Choropleth map is a map
showing attribute data of a spatial tessellation by colors
and textures.

Categorical variable is directly visualized by colors and

textures: different colors indicate different categories.
Numerical variable is first categorized into several classes,
and then visualized by a progression of colors and
textures.
Figure: Tessellations with numerical and categorical variables

4. Visual Analysis 4. Visual Analysis

4.2.1 Classification schemes

To make a choropleth map of a numerical variable, we There are four schemes for categorizing numerical
determine variables.

1. classification scheme, 1. Equal interval scheme

2. class number, 2. Quantile scheme
3. class boundaries, and 3. Nonuniform scheme
4. colors and textures. 4. Irregular interval scheme

4. Visual Analysis

Equal interval scheme

84 149 74 162 91

Equal interval scheme categorizes a numerical variable by

an equal interval value.

If we specify interval value, GIS calculates boundary

values, classifies spatial units into categories, and
visualizes the categories by different colors or textures.

0 20 40 60 80 100

Figure: Equal Interval

9
 4. Visual Analysis

Quantile scheme
100 100 100 100 100

Quantile scheme categorizes a numerical variable so that

every category has the same number of spatial units.

In this scheme we give the number of categories instead of

interval value. GIS then calculates boundary values,
classifies spatial units into categories, and visualizes the
categories by different colors or textures.

0 22 32 58 78 100

Figure: Quantile

4. Visual Analysis
Arithmetic
0 33.3 60..0 80.0 93.3 100
progression
Nonequal interval scheme 33.3 26.0 20.0 13.3 6.7

03.2 9.7 22.6 48.4 100

3.2 6.5 12.9 25.8 51.6
Geometric
When we are interested in a certain range of attribute progression
value, we want to use finer categories in the range. In this 0 51.6 77.4 90.393.3
100
case we use nonequal interval scheme.
51.6 25.8 12.9 6.53.2

Typical examples include monotonically increasing 0 10.7 26.7 47.5 72.5 100
(decreasing) interval scheme, in which the interval Irregularly
10.7 16.0 20.8 25.0 27.5
monotonically increases (decreases) with attribute value. increasing
progression
0 9.3 20.4 34.9 56.6 100

9.3 11.1 14.5 21.7 43.4

Figure: Nonequal interval

4. Visual Analysis
Breakpoints
Irregular interval scheme

Frequency distribution of attribute value often shows

‘breakpoints’, where the frequency suddenly drops.

In such a case, we can obtain a natural classification of

attribute variable if we take the breakpoints as
boundaries of intervals.

0 34 56 78 100

Figure: Irregular interval

10
 4. Visual Analysis
Breakpoints
4.2.2 Class number

In theory, we can use any number of categories in

classification of numerical variables. You may think that
the more categories you use, the better map you obtain.

In practice, however, we can discriminate only a limited

number of colors used for visualizing categories. It is not
always useful to increase class number.

0 16 34 44 56 78 94100

Figure: Irregular interval

4. Visual Analysis

We use from five to seven classes in choropleth maps of

the single hue progression.

Figure: Effect of class number

4. Visual Analysis

Bi-polar progression
Partial spectral
hue progression
Blended
hue progression
If we want a finer classification of the variable, we should
use two or more hues, or we should combine several Value progression
progressions together. Full spectral
progression

Two-variable
color progression

Figure: Color progressions

11
 4. Visual Analysis

6.2.3 Class boundary

Class boundary greatly affects the appearance of

choropleth maps that represent numerical variables.

We should be careful when classification scheme involves

subjective choice of class boundaries, as seen in nonequal
and irregular interval schemes.

Figure: Effect of class boundary

4. Visual Analysis

4.2.4 Effect of the size of spatial units

In choropleth map, map readers tend to pay attention to

large polygons while they often overlook small polygons.

Attribute data of large polygons are more influential than

those of small polygons in the perception of map readers.
Large polygons are emphasized as a result of visualization.
This often leads to misunderstanding of the spatial
distribution of attribute values.

Figure: Effect of the size of spatial units

4. Visual Analysis

To avoid this perceptual illusion, we often perform areal

interpolation. We convert spatial data reported by
irregular spatial units into those based on regular lattice,
where all the spatial units are identical.

Figure: Spatial data based on a square lattice

12
 4. Visual Analysis

Kernel smoothing is also useful to avoid the perceptual

illusion. The result is visualized by isopleth (isarithm)
map or bird’s eye view. 80
100 60
60 80
40

20

Figure: Isopleth map

Figure: Isopleth map with color Figure: Bird’s eye view

4. Visual Analysis

4.3 Cartogram

Cartogram is a cartographic technique that transforms a

map to be ‘uniform’ with respect to the distribution of
spatial objects. Spatial units containing many objects are
represented by large polygons, while those of fewer
objects are small polygons.

Figure: Bird’s eye view

13
 4. Visual Analysis

A typical example is found in the analysis of the

distribution of disease cases. If we find a cluster of cases,
we think that the disease is epidemic. However, it may not
be true because we can find clusters in regions where
many people live.

Cartogram rescales the map of case distribution

according to the population count to make the population
distribution uniform. Cartogram helps map readers’
understanding of a spatial distribution considered to be
closely related to another non-uniform spatial distribution. Figure: Census tracts of San Francisco City/County (1980)

Figure: Original map Figure: Map of uniform population distribution

Figure: Hypothetical cases of a disease Figure: The cases under the uniform population distribution

14
 4. Visual Analysis 4. Visual Analysis

4.4 Visualization of higher-dimensional objects

Cartogram is useful Visualization of two-dimensional spatial objects is relatively

straightforward because computer displays and papers are
1. when we analyze the distribution of spatial objects both two-dimensional spatial objects.
that seems affected by other non-uniform distributions,
say, population distribution, terrain elevation, etc., On the other hand, visualization of higher-dimensional
and spatial objects is far more difficult and is a recent topic in
2. when we determine the location of urban facilities that GIS research.
should be uniformly located with respect to other
nonuniform distributions: schools, post offices, etc.. 1. Visualization of three-dimensional spatial objects
2. Visualization of four-dimentional (spatiotemporal) objects

4. Visual Analysis

• References - modern visualization

1. Cromley, R. G. (1992): Digital Cartography, Prentice-

Hall.
2. MacEachren, A. M. and Taylor, D. R. F. (eds) (1994):
Visualization in Modern Cartography, Elsevier.
3. Slocum, T. A. (1998): Thematic Cartography and
Visualization, Prentice-Hall.

Figure: Three-dimensional point distribution

Figure: Three-dimensional scalar field Figure: Spatiotemporal scalar field

15
 4. Visual Analysis

4.5 GeoVISTA

GeoVISTA Center at the Pennsylvania State University

Its goals are to support research in GIS, with an emphasis

on geographic visualization (geovisualization), and to
facilitate application of advances in GIS and related
technologies within science more broadly as well as within
business, government, and education.

http://www.geovista.psu.edu

Figure: Spatiotemporal scalar field

4. Visual Analysis

4.6 Visualization and sound

Human beings have five senses -- sight, smell, hearing,

taste, and touch. This implies that we can use other than
sight in order to communicate spatial information.

J. B. Krygier (1994): “Sound and Geographic

Visualization”, in MacEachren, A. and Taylor, D.R.F. (eds.),
Visualization in Modern Cartography, Pergamon, pp. 149-
166.

http://www.owu.edu/~jbkrygie/krygier_html/krysound.html
Figure: Communication of spatial information by sound

4. Visual Analysis 4. Visual Analysis

Homework Q.4.1 Homework Q.4.2

Census survey is conducted in many countries all over the Cartogram is a unique method of adjusting the
world. The survey is done for all the individuals, but the differences of a numerical property such as population
data are aggregated either spatially or aspatially when density among different regions. Find several examples of
open to public in order to keep the confidentiality of cartogram applications and discuss whether they are
individuals. The data are provided in various formats successful or not. When does cartogram works and when
such as tables, figures, maps, spatial data, etc. Report how does it fail?
the census (demographic) data are open to public as
spatial data in various countries, say, USA, UK, France,
Germany, Netherlands, China, and Japan.

16