Environmental Engineering Reference
In-Depth Information
24.1
Introduction
or individual-based models) are increasingly used to represent
decision-making in urban land-change context (Parker, Manson
and Janssen, 2003; Xie, Batty and Zhao, 2007). ABMs have been
generally employed as a computational model for simulating the
actions and interactions of autonomous individuals in a net-
work. Though what agents actually are remains a topic of debate,
they are generally regarded to be goal-driven, autonomous, and
adaptive. In the context of urban and geographic studies, these
agents are 'geo', because they interact with scale-dependent geo-
graphic environment in given contexts and have the ability to
reason spatially (Yu and Peuquet, 2009). Therefore, these fea-
tures of ABM are salient to supplement some weaknesses in
the commonly used urban models. For instance, the ABM can
integrate natural environments (the agents' physical space) with
policy making rules (the agents' intelligence), combine bottom-
up actions with global interactions, and simulate processes of
urban growth that are locally determined but moderated by
higher-level macro economy.
Urban landscape changes are usually outcomes of urban growth.
It has been a challenge for both researchers and practition-
ers to capture the dynamics of urban growth and the causal
relationships between urban landscape changes and underlying
socioeconomic driving forces. Rigorous efforts have beenmade in
developing computational and mathematical models to explore,
explain, and predict urban growth during the past half a century.
However, urban modeling has been a controversial field since its
reception. The critics claim that conventional urban modeling
is either based on aggregated numerical analysis, or general-
ized behavior approach. These models are either averaging out
the characteristics of study units of interest, or assuming that
urban drivers behave in a rational way (Ligmann-Zielinska and
Jankowski, 2007). True urban complexity may be lost in the pro-
cess of generalization forced by mathematical tractability (Axtell,
2003). Moreover, the differential equation-based approach may
unnecessarily make the model mathematically complicated and
therefore the potential user perceives it as a black box (Lee, 1973).
Urban modeling was regarded as a misunderstanding of the role
of modeling in real-world planning (Lee, 1973).
Only very recently have the conceptual and mathematical
foundations for substantive inquiry into urban dynamics been
made possible due to the growing understanding of open sys-
tems structures and human decision processes. The applications
of systems theory to urban dynamics have been made possible
by fundamental advances in the theories of nonlinear systems,
including dissipative structures, synergetics, chaos and bifurca-
tion in the physical sciences. In fact many of the originators
of these new ways of articulating how complex systems work,
have seen cities as being a natural and relevant focus for their
work. Prigogine's work on dissipative structures, for example, has
been applied to urban and regional systems by Allen (1997) while
Haken's work on self-organization has been implemented for city
systems by Portugali (2000) and Weidlich (2000). Ideas about
how life can be created artificially have guided many of these
developments and in this context, highly disaggregate dynamic
models based on cellular automata (CA) have become popular
as the metaphor for a complex system (Batty and Xie, 1994; Xie,
1996). CA models articulate a concern that systems are driven
from the bottomup, in which local rules generate global patterns,
and are good indicators that there is no hidden hand in the form
of top-down control. Again cities are excellent exemplars of these
kinds of systems (Holland, 1975). More recently CA has been
elevated to the status of a 'new science', articulated as the basis
for taking a new look at a wide range of applications to scientific
inquiry (Wolfram, 2002).
In fact, CA focuses on physical processes of urban systems
and simulates land use changes through rules usually acting upon
immediate neighboring cells or at best some wider set of cells in
which the notion of a restricted neighborhood for spatial influ-
ence is central (Batty, 1998; Batty, Xie and Sun, 1999; Clarke and
Gaydos, 1998; Li andYeh, 1998;White andEngelen, 1993;Wuand
Webster, 1998; Wu, 2002; Xie, 1996). Though many innovative
ideas such as genetic algorithms, neural network methods, and
stochastic calibration for determining weights and parameters
have been proposed and successfully integrated, such CA mod-
els are essentially heuristic and simplistic. Hence, agent-based
models (ABMs) and simulations hold a promise to remedy these
modeling weaknesses. ABMs (also termed multi-agent systems
24.2
Design, construction,
calibration, and validation
of ABM
From the implementation point of view, ABM involves four
important steps: design, construction, calibration, and valida-
tion, which are four key elements of an ABM. Moreover, ABMs
are primarily applied as a tool-box for computer simulation
at present. Computer simulation (or computational modeling)
is one of three ''symbol systems'' available to social scientists
for developing models in addition to the familiar verbal argu-
mentation and mathematics (Ostrom, 1988). The underlying
motivation of using computer simulation in developing mod-
els is analogous to the use of more familiar statistical methods
in modeling. In either case, there is a driving issue that the
researchers want to understand and investigate, which is called
the ''research question''. A model exploring this research ques-
tion is often built through a theoretically motivated process of
abstraction (Gilbert and Terna, 2000). However, the approaches
that are deployed to assess behaviors of various models are
significantly different between statistical models and computer
simulations. If the model is a statistical equation, it is run
through a statistical analysis program such as SPSS. If the model
is a computer simulation, its behavior is assessed by ''running''
a computational program iteratively to evaluate the effect of
different input parameters (Gilbert and Terna, 2000). It is the
task of model design to parameterize the behavior of a computer
simulation program.
In the urban modeling context, the design task of an ABM
is to enable ''agents'' to represent complex behavior of interact-
ing entities such as households, businesses, planners, developers,
or decision-makers in a given (metropolitan) region. Model
researchers have to provide theoretical and methodological
foundations as well as practical solutions to the identified
decision-making task that drives urban land change or urban
development policy. Thus the heart of urban ABM design lies in
how to abstracting and computerizing a decision-making prob-
lem when it is identified. Agents are designed to be capable of
making decisions concerning an identified research question,
Search WWH ::
Custom Search