Environmental Engineering Reference
In-Depth Information
example, population cycles with a wavelength exceeding
the average lifetime of individual agents can emerge in
the presence of spice trade, compared to the extinction
under equivalent conditions without trade (Figure 18.3c).
This modelling approach allows the investigation of the
realism of different economic and policy models and
ideas of sustainability within the landscape.
While Sugarscape considers artificial landscapes to
investigate general patterns, more recently attempts have
been made to simulate real landscapes, both historical
(e.g. Axtell et al ., 2002; Wainwright, 2008; Janssen, 2009)
and contemporary (e.g. An et al ., 2005; Castella et al .,
2005; Manson, 2005; Millington et al ., 2008). An et al .
(2005) took an agent-based approach to explore how the
interactions of household dynamics and energy demands
affect panda habitat in a landscape in China. Their model
provides a particularly good example of how an ABM
can provide a framework to represent nonlinear interac-
tions, cross-scale (spatial and temporal) data, feedbacks,
and time lags between different subsystems of a broader
coupled human-environment system. For example, a fuel-
wood subsystem model (An et al ., 2001) links household
fuelwood demand to demographic and socioeconomic
factors but is unable to link this demand to its impact
on forest growth in the landscape because it is essentially
aspatial. An electricity demand submodel is similarly lim-
ited because while it is able to estimate the probability
of a household switching to electricity (from using fuel-
wood) using a set of socioeconomic, demographic, and
geographic predictors (An et al ., 2002), being aspatial it
cannot identify the impact of this change to specific forest
locations. Furthermore, in both these submodels house-
hold demographics are assumed to be exogenous driving
forces. The ABM developed by An et al . (2005) integrates
these submodels to track the life history of individual
people and represents household demographics explicitly
(in-part by integrating a household formation model,
An et al ., 2003). By situating this ABM within a spatially
explicit representation of a heterogeneous forest land-
scape (complete with forest-growth model and human
fuelwood spatial search algorithm) the spatial limitations
of the submodels are overcome and interactions between
the different subsystems can be explored. For example,
using the model An et al . showed that, counter-intuitively,
the maximum 'buffer distance' around their homes in
which locals are allowed to collect fuelwood will have
minimum impact on panda habitat when either small
or large but will have greatest impact for intermediate
distances. This outcome may be because when the buffer
distance is small impacts on panda habitat are limited
because locals are forced to switch from fuelwood to
electricity and when the buffer distance is high, fuelwood
collection takes places across a large area allowing forest
time to regenerate between collection episodes. However,
when buffer distance is intermediate, fuelwood demand
is satisfied by cutting all available wood in a region of
intermediate area, likely going beyond its carrying capac-
ity (its ability to regenerate) and causing greater habitat
loss than for other buffer distances.
One of the benefits of using ABMs to represent real-
world landscapes (and the actions and interactions they
contain) is their potential to improve our understanding
(not least by forcing us to think about holistic con-
ceptualizations of controls on these landscapes) and
support analyses of the potential outcomes of hetero-
geneous decision-making. However, for this potential to
be realized, credible agent representations of real-world
actors must be produced. At the very least, these agent
representations should attempt to improve upon the
homogenizing limitations of the assumptions of neo-
classical economics. When developing rules of agent
relationships and responses, Bousquet and Le Page (2004)
suggest that three key issues must be addressed:
Decision-making: what mechanisms do agents use to
make decisions? How are agents' perceptions, actions
and responses linked?
Control: how are agents related and synchronized?
Communication: what information may be passed from
on agent to another? How is it passed?
To assist the development of agent behaviours empir-
ical methods such as sample surveys and interviews
with the actors being represented, participant observa-
tion, and field and laboratory experiments are useful
(Robinson et al ., 2007). For example, in the develop-
ment of the Chinese landscape ABM described above, An
et al . (2002) used data from interviews with stated choice
methods (e.g. Louviere et al ., 2000) to estimate willing-
ness to switch between fuelwood and electricity. Possibly
one of the best ways to develop an agent-based model
is the use of participatory modelling approaches that
involve the actors being represented throughout the mod-
elling process. 'Companion modelling', as it has become
known, allows actors to communicate with agent-based
modellers and contribute to the development of agent
behaviours by participating in r ole-playing games (e.g.
Castella et al ., 2005) or by interacting with simulation
agents themselves (e.g. Nguyen-Duc and Drogoul, 2007).
Although this approach has been promoted as an ideal
Search WWH ::




Custom Search