Geoscience Reference
In-Depth Information
9.1.6 M
ulti
-a
gent
S
ySteMS
aS
h
uMan
-d
riVen
c
ellular
a
utoMata
CA were adopted by GC soon after their introduction in the field of computer science in the early
1950s (von Neumann, 1951). As described by Batty and Longley (2014), geographic features are rep-
resented in CA by the cells of a grid. The CA cells are characterised by
states
. To describe CA dynam-
ics,
transition rules
define, for every cell
c
of the CA, its state at a time step t + 1 as a function of (1) the
state of c and (2) the states of the cells that are the neighbours of c at a time step t. Thus, the definition
of a CA consists of the definition of cell states, cell neighbourhood and transition rules (Batty and
Longley, 2014). Note that according to the terminology of AB modelling, CA cells are objects.
Geographic applications of CA are primarily concerned with land-use/land-cover dynamics
(Batty and Longley, 2014). These applications are based on essentially wider than von Neumann's
specification of the CA:
•
A square grid of cells is often substituted by any partition of a 2D space into polygons,
including the real-world coverage of land parcels.
•
The state of a cell at t + 1 is dependent upon the state of cells that are far beyond the neigh-
bourhood of the cell, for example, via dependence of the cell state on the distance to a
highway or to a city centre.
•
Global constraints are imposed on the location and the number of cell state changes during
the time step. For example, the amount of the land cells that can change their state from
open land
to
dwellings
is limited to the amount permitted in the development plan.
The aforementioned generalisations do not change the basic assumptions of CA that cells are
pas-
sive
land units that can change their state in response to the states of their close and distant neigh-
bours. This passiveness makes CA inherently insufficient for describing the dynamics of human
systems. In reality, the developers and landowners, private or public, decide on whether to change
a land parcel's use and construction (Figure 9.1). These decisions cannot be described as dependent
only on the state of the other parcels and objects - there are always subjective factors that influence
a decision of a developer or landowner.
CA cannot directly reflect landowners and developers' behaviour; instead, the modeller has to
interpret their behaviour in terms of transition rules. Formally, this is easy in case of
perfectly ratio-
nal
behaviour. That is, the complete information on all system objects and agents, including rules
Neighbour cells, my dear,
these people force me to
change my state...
To build?
To sell?
Built-up area
Vegetated area
Agriculture
Open spaces
FIGURE 9.1
From a standard to a human-driven CA.
