The urban sprawl dynamics: does a neural network understand the spatial logic better than a cellular automata?

1. Introduction

Land use dynamics and fragmentation of settlements is a crucial question for planning.
In the general framework of sustainability objectives, the policies to control a suitable
process of urbanisation involve more and more deep knowledge on complex criteria of
location chosen by the different agents. Planners realize that is crucial to understand and
provide the best possible explanation for the observed spatial distribution of urban
activities.

Principles and technologies of Artificial Intelligence (AI) in general, and of NN and CA
in particular, offers the potentiality to increase the knowledge in urban dynamics by
multiplying the information capacity of the GIS and by offering a new approach to
territorial modelling. Most geocomputation currently deals with models on spatio-
temporal dynamics in urban land-use and morphogenesis.

Among them some applications, mainly based on Cellular Automata, have opened more
promising directions for the goal: Clarke, Hoppen and Gaydos (1997) modelled the
historical development of San Francisco area; (Batty, Xie and Sun 1999; Wu 1998) built
several urban models and in particular a model on the residential development in the
fringe of Buffalo; Portugali, Benenson and Omer (1994; 1997) have focused their
research on models of socio-spatial segregation; the many contributions of Engelen,
Ulje and White (White and Engelen 2000) have produced several CA based models
with integration of several economic theories.

Cellular Automata appear to be the most attractive and favoured technique for
implementing high resolution models of spatial dynamics for a number of reasons:

• They are inherently spatial; their definition on a raster of cells, and on
neighbouring relationships are crucial;

• They are simple and computationally efficient;

• They are dynamic and can then represent a wide range of situations and
processes;

It is worthwhile to note that, in most of the models carried out until now, CA are based
on explicit spatial rules which allow to simulate different dynamic behaviours on the
base of a ”trial and error” procedure.

But this condition, the explicit and exogenous formulation of assumptions, represents
the greatest limit of this approach, since it reduces the variability of the different
territorial contexts on the base of few theoretical principia (spatial interaction, diffusion
processes and so on), inhibiting the discovery and arise of new features in urban

<<   1   2   3   4   5   >>

More intriguing information