Cellular automata and agent base models for urban studies: from pixels to cells to hexa-dpi's - Urban Remote Sensing: Monitoring, Synthesis and Modeling in the Urban Environment - page 312

processes on decision-making, and their dynamic socioeconomic
environmental linkages (Matthews et al ., 2007).
In the development of ABM models, the focus is on the
moment of taking a decision, the moment of moving from one
place to another; in other words the moment of the ''command''
that will lead to a decision plays an important role (in these
micro-simulation ''command and control'' models where each
agent has its own unit of command and control) - that is to say
that, for instance, in the decision tree the ''nodes'' that allow
the ramification of the decision tree will be pivotal in order to
enable a new change as a result of that decision/command. As a
consequence most of these ABM should rely on the simulation
capabilities of GAs and representations of phenomena in terms of
explanations of what is at the basis of decisions and options (i.e.,
why do we decide to choose a specific option). Therefore ABM-
GAs can play an explanatory/experimental role in formalizing the
decision-making processes that leads to specific spatial actions.
For the purpose of this chapter, we will keep addressing
ABM-GA, as we feel that this will be one of the most important
modelling approaches to optimize the decision trees of each
individual agent (or groups of agents).
ABM-GAs are constituted of: (i) agents that do not have
the constraints of neighborhood effects; (ii) behavioral roles
among agents and the environment itself; (iii) independence
from central command/control, but able to act if action at
a distance is required; (iv) states of agents tend to represent
behavioral forms.
The most basic model environment of an ABM-GA will have
a set of attributes per agent (or group of agents), (one)a set
of decision trees and trigger points that will allow to set the
context for a new movement (upgrade of the spatial/temporal
environment) in time/space (Fig. 22.4).
These models are characterized by behaving as populations of
simple agents that interact locally with each other and their envi-
ronment and produce swarm intelligence, that is to say, machine
resulting intelligence, or emergent behaviour in machines not
directly resulting from code development (Fig. 22.4). This emer-
gent behaviour in GA, ABM and also CA are one of the most
interesting developments of these new theories, models and data
structures, pointing out to what is called 'the intelligence in the
machine', 'artificial intelligence', or 'artificial life' as described in
thenextpointsofthischapter.
While today research starts to integrate both approaches
of CA and GA, ABM during several decades these were stud-
ied apart. While CA has a spatial explicit representation that
makes it very apt to model urban and environmental systems
(i.e., land parcels, transportation infrastructures), ABM-GAs are
important for their behavioural roles that are very apt to model
individual agents and their behaviour (i.e., households, vehicles
and pedestrians).
From the 1950s to the 1980s the basic elements of complexity
analysis both in the natural and social sciences were drafted.
Some of them were at that time still conjectures that are now
being explored with more detail. During the 1990s it started to
become clear that these two approaches of modeling with CA
or with ABMs were the two most used approaches to work with
complexity in a quantitative formulation.
Nevertheless, researchers would tend to separate both
approaches. CA and ABM researchers tended to be in opposing
fields, defending one or another approach. These were almost
hermetic areas of research where little information would
cross-fertilize the development of new methodologies. For some

of us it started to be clear that without this cross-fertilization it
would be difficult to select the best methodological option (CA,
ABM, GA) or to develop hybridmethodologies that included CA,
ABM and GA and fully integrate aspatial and spatial structures.

22.5 Cells and agents in a
computer's `` artificial life´´

Artificial life is the study of man-made systems that exhibit behav-
ioral characteristics of natural living systems. CA and intelligent
agents (ABMs that rely in technological developments such as
the previously described GAs) are typical artificial life techniques,
and in the past decades they have been increasingly implemented
in urban studies (Wu and Silva, 2010).
It is now commonly accepted that CA are very capable
of simulating complex spatial phenomena. CA models include
the following capabilities: (1) representation of complexity and
dynamics in systems (Silva, 2010a, b); (2) spatial integration
(Batty, Xie andSun, 1999; Piyathamrongchai andBatty 2007; Silva
and Clarke 2002, 2005) and self-organizing mapping (Castilla
and Blas et al . 2008); 3. Extensibility and adaptability (Dragicevic
2007; Kocabas and Dragicevic 2007; Jenerette and Wu 2001); (4)
both simplicity and complexity: (Torrens and O'Sullivan 2000
(5) visibility: through the lattice structure (cells in CA model
correspond to the grids in a raster image), and the link to
geographic data makes CA models highly visual: CA are popular
as land change models (Clarke and Gaydos, 1998; Batty, Xie and
Sun, Straatman and Engelen, 2004; Silva, Wileden and Ahern,
2008).
Ligtenberg et al . (2004) described an ABM to explore differ-
ent ways in which decisions regarding land use could be made.
Ligmann and Jankowski (2007) focused on using ABM for spa-
tially explicit modeling of real-world policy scenarios. Benenson,
Martens and Birfir, (2007) simulated urban parking policy sce-
narios and analyzed their impacts from the user and public policy
perspective. The simulation of social-economic interactions in
urban systems is another important application of ABM. Milner-
Gulland et al . (2006) described the use of an ABM to investigate
the trade-offs in allocation of wealth by households. Ettema et al .
(2007) presented a multi-agent based urban model (PUMA)
concerning land conversion problems.
The phenomena of land change needs to include spatial and
aspatial dynamics. During the past, the stationary transition
probabilities limited the model's ability to reflect feedbacks to
global changes (action at a distance) to influence transitions at
the cellular level. Some exceptions include, SLEUTH, by enabling
and disabling boom and bust phases (Clarke et al ., 1998, Silva
and Clarke, 2002), and CVCA, by allowing localized changes
accordingly to proximity, area issues and globally by allowing
changes to the landscape shape index (Silva, Wileden and Ahern
2008). Nevertheless, if CA are incorporated with ABM, it expands
its potentiality by integrating two important components: cellular
models are used to describe spatial dynamics and ABM represent
a-spatial, social interactions (Wu and Silva, 2010).
Recently, integrated approaches using hybrid modelling sys-
tems are providing promising solutions for urban land studies.
Agent-based cellular automata (AB-CA) is one such hybrid sys-
tems, that integrates agent basedmodelling and cellular automata.
This is particularly important when one needs to include both