Geoscience Reference
In-Depth Information
The problems with a high temporal and spatial resolution are many. Besides potentially low
performance of the model, agents should be precisely located in space, their decisions should be
precisely located in time and the modeller has to decide which agents and when would become
aware of other agents' decisions. At the same time, the problems entailed by high spatial and
temporal resolution encourage the modeller to be precise when formulating agent behaviour and
interactions.
A low temporal resolution has its own problems, the most important being the problem of
parallel versus sequential updating. Let several interacting agents make decisions during the
single (and extended in time) time step: What should be the order of implementing these deci-
sions?
Parallel (synchronous)
updating means that each agent decides what to do depending on
the system state at the end of the previous time step, without any knowledge of what happened at
the current time step.
Sequential (asynchronous)
updating, which has many forms, assumes that
during the same time step, agents make decisions in order and an agent may know the decisions
of other agents.
A good example of parallel versus sequential updating can be illustrated with the game of life,
and I like its presentation on http://www.math.com/students/wonders/life/life.html best of all. As
one can see at that and many other sites, the standard Game of Life employs parallel updating and
produces gliders and other nice creatures. However, all of them disappear when parallel updating is
substituted by sequential (Blok and Bergersen, 1999). The importance of model updating requires
a much longer discussion than is possible here. However, I conclude this section by claiming that
sequential updating in which agents and objects are chosen randomly at each time step is the sim-
plest and often a sufficient choice for AB model updating.
9.2.5 S
Patial
r
eSolution
of
the
aB M
odel
The decision about the spatial resolution of the model depends on what types of geographic
features are used to represent the agents, the behavioural rules of the agents and the temporal
resolution of the model. The AB modeller has to decide on the meaning of the statement
two
agents or objects are at the same location
. For the PARKAGENT model, a driver agent must
distinguish between the states of the parking places, and, thus, 5 m is an inherent spatial reso-
lution for this model. For the hypothetical example of Mali horticultural dynamics, the spatial
resolution is determined by the size of the farmer's field, which varies between 0.5 and 1.0 ha.
That is, either we should implement a model over a polygon layer of Kita agriculture fields or, if
such a map is unavailable, consider agricultural space as consisting of 50 × 50-100 × 100 m cells
each representing a field.
As already mentioned earlier, the inherent spatial and/or temporal resolutions of the modelled
phenomenon can be too high and result in a low performance in the model. My advice here is to
apply the model to a smaller area, with a minimum possible number of agents and objects, and
investigate the model dynamics at an inherent and at a lower spatio-temporal resolution. Comparing
these dynamics, the modeller could decide whether and when the model for the larger area can be
considered at the lower spatial and temporal resolution. Note that the difference between the model
dynamics in the case of parallel and sequential updating can be of fundamental importance for such
a comparison (Benenson, 2007).
9.2.6 r
ecognition
of
c
ollectiVe
P
roPertieS
and
e
Mergence
An investigation of the system dynamics, that is, the changes that affect the entire system, is an
ultimate goal of every model of a complex system. Collective patterns, spatial and non-spatial, can
emerge, evolve and disappear. Excellent topics on complex systems theory explain these processes
and present intriguing examples of collective dynamics (Flake, 1998). To recognise collective spa-
tial patterns, any method of spatial data mining may be applied; see the 2009 special issue of
