20
2. ALTERNATIVE VIEWS OF GEOGRAPHICAL SPACE
To begin, four alternative views of geographical space are described and contrasted. Current GIS present only a
subset of these views to the user (Goodchild, 1992), with poorly defined relationships between the views; if indeed
they are defined at all.
2.1 FIELD MODEL OF SPACE
In the field model it is assumed that a region (surface) is composed of an infinite number of points, each with a
vector of real variables describing the values of some real-world phenomenon at that point. The model assumes that
the variables change continuously over the surface. In practice, it is not possible to measure values at all points, so a
finite set is chosen. There are often restrictions on the measurements taken, in terms of dynamic range, quantisation
and domains- a field point may theoretically possess values for all phenomena of interest, but a surveyed point will
usually have only a subset of these. Thus a sample within the field model is (at all but the finest scales) a point
measurement gathering data across a finite number of concurrent surfaces. For example, a weather station may
gather rainfall, barometric pressure and humidity values. To form a rainfall surface, data from samples must be
interpolated across the region, using one of many specifically designed techniques (e.g. Cressie, 1993).
2.2 IMAGE MODEL OF SPACE
class
by p
purp
A cL
Max
label
resol
men:
feati
conr
that
extn
of a
intrc
wou
Othe
absti
of a
edge
meai
The image model may be regarded as a specialisation of the more general field model. An image model is given by
restricting the number of sample points, and imposing a grid pattern to define the (square) region described by each
data vector within the field model 2 .
Remote Sensing Systems (RSS) typically operate on an image view of space, where features are not explicitly
defined. The user operates on source data values (for example pixel reflectance values) and typically may proceed
as far as classifying these data. With respect to GIS tasks, data in image form is much harder to work with since its
content is unspecified. Its use explicitly demands an in-depth understanding of the sensor and the sensed properties
of the various landuse classes within the scene, and hence is not a suitable task for the non-expert.
The image view represents remotely-sensed data in its purest form, since no summarising or generalising has been
carried out 3 . The data is also highly flexible, since it may have many different types of interpretation imposed upon
it.
2.3 THEMATIC MODEL OF SPACE
The thematic model is based on classes. It is more abstract in nature than the image model, since each geometric
object has some meaning or interpretation imposed upon it- this interpretation in fact replaces the original data
value(s). However, individual geographic features are not explicitly recognised. The thematic model regards the
extents of some arbitrary region as a series of overlays or thematic maps. Each overlay represents a monothetic
2 The values captured at the sensor are usually an average reflectance over an approximately circular area, the diameter of which
is defined by the point spread function of the sensing device. The science of signal processing retains the data as a set of
(approximately circular) samples on a surface (e.g. Lim, 1990); whereas the approach taken by image processing, and used here,
is to project the sample values onto a grid. In practice the sensed area is either only a part of the pixel (e.g. Landsat TM), or
alternatively, sensed areas overlap into adjacent pixels (e.g. Landsat MSS). It is recognised that the physics of the sensors that
provide the data used here is more complex than is modelled by simply imposing the data onto a grid (e.g. Barrett and Curtis,
1992, Ch. 2, 5, 9), but since gridded data is all that is available, it is assumed here that the pre-processing of this data before it is
made publicly available accounts for such complexities
3 Image data is often processed in some way before it becomes publicly available, for example to correct for distortion, to re
project or to sample. The comment refers more to the fact that no attempt has been made to uncover the content of the image.
2.4
Cun
labei
prop
class
origi
coar
to w
are i
class
137)
whai
This
anal
func
Whi
utilil
appe
com
take
supr
As t
that