metrie
*orum,
Week
Ortho-
t, Mu-
Ire for
ves of
), Part
forma-
». 107-
rischen
on and
S-Con-
. Auto-
To be
\ustria,
MEASUREMENT AND MODELLING OF URBAN REMOTELY-SENSED DATA
Victor Mesev
ESRC Research Fellow, Department of Geography, University of Bristol, University Road, Bristol, BS8 1SS, United Kingdom
Commission IV, Working Group 1
KEY WORDS: Classification, Modeling, Urban, Integration
ABSTRACT
This paper lies at the interface between remote sensing, GIS, and spatial analysis. By developing an integrated framework, this
interface can be used to generate detailed and frequently updated measurements and classifications of urban land coverages, as well
as extensive measurements and models of urban change. Specifically, the classification is a Bayesian-modified maximum-likelihood
decision rule, with a priori probabilities determined by GIS data, and the models of urban change are fractal-based measurements of
residential development and fractal-based measurements of density profiles. All of which, when integrated and applied to the
settlement of Norwich, in the United Kingdom, reveal insightful patterns within urban morphologies and detail repercussions on
urban density arising from the residential configuration and effects of physical and planning constraints. The data is a single
Landsat 5 (TM) image, together with population and housing data from the UK Population Censuses of 1991. This work is an
important contribution to the advancement of integrated data handling and analysis systems, as well as providing a means to
examining and understanding the complex arrangement of urban structures and processes.
BACKGROUND
The importance of monitoring urban areas is indisputable.
With nearly 90% of the European population now living in
areas designated as built-up urban’ there is a tremendous need
for vital information on understanding how cities expand,
contract and develop. Specifically, there is a need to monitor
and analyse population shifts, employment restructuring, and
the layout of residential morphologies. The formulation of
national monitoring, management, and planning policies needs
to be based on precise and accurate source data. Effective
mapping of the structure of urban areas is an essential baseline
component to the assessment of the general structure and
sustainability of settlements. In conjunction, a statistical
framework is vital in revealing more objective measurements of
urban configuration, as well as comparisons of how these urban
structures change both across space and through time.
Remote sensing has long been recognized as an important
technology for reproducing ‘snap-shot’ observations of the
Earth’s surface and atmosphere. In the observation of urban
areas, satellite sensor data have allowed extensive areal
coverage at consistent and readily updateable intervals. Given
its rapid retrieval and global availability, satellite remote
sensing is an ideal means for producing measurements from
which to monitor various aspects of urban dynamism,
particularly at the regional scale (examples in Lo, 1986; and the
recent GISDATA Specialist Meeting ). However, it is also
common knowledge that because of the complex heterogeneous
nature of urban surfaces, once the spatial resolution of satellite
images begin to approach a more local scale, more and more
pixels become invariably spectrally mixed (Forster, 1985). In
this paper, some of the problems of urban remote sensing will
be addressed by exclusive reference to the growing debate on
Official Journal of the European Communities, C138/52,
Paragraph 5.5 (1993)
Conference on *Remote Sensing and urban analysis" given by
the European Science Foundation, Strasbourg, France, 11 June,
1995
557
GIS/Remote sensing integration, in particular the role of GIS in
image classification. The methodology will hinge upon the
ability of GIS to handle extraneous, non-spectral data, which
are then used to determine and vary the a priori probabilities of
the standard maximum-likelihood (ML) classifier (begun by
Strahler, 1980). This essentially involves the use of GIS data to
first stratify urban images according to some spatial and
contextual rules, and then determine the area estimates of urban
classes within each stratum. Area estimates are then normalised
and directly inserted into the ML classifier as prior
probabilities, producing accuracy levels above classifications
simply based on the standard equal prior probability
assumption. In work elsewhere, favourable results have also
been generated from area estimates which have been used as
part of an iterative process for updating ML a posteriori
probabilities (Mesev et al, 1996).
Most research treat the classified image as the end product (the
spectral result) and neglect the wealth of information available
on the spatial form of classified images. Along with the
probabilistic modification of the ML classifier, this paper will
also assess the abilities of fractal geometry to measure and
summarise the highly irregular spatial patterns of urban land
cover/use produced by image classification. In a similar vein to
De Cola's work in 1989, fractal geometry will be used to
characterise the spatial properties of classified multi-
dimensional feature space. However, unlike De Cola, the
derived fractal dimensions will further be used for comparative
analyses which are designed to evaluate how form and density
of urban land use vary within settlements. Furthermore, the
assumption of classified urban classes as being fractal will also
allow these classes to be represented by cumulative and density
profiles generated from functions based on fractal-modified
inverse power relations (see Batty and Kim, 1992; Mesev et al,
1995 for full description). These urban profiles will provide a
means with which incremental urban development is precisely
monitored and will lead to an insight on which urban processes
may be in operation. It is hoped that remotely-sensed data will
rejuvenate the role of urban density functions in measuring and
prescribing changes in urban development (Zielinski, 1979).
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B4. Vienna 1996
