Geoscience Reference
In-Depth Information
selected to optimise the models' fi t to the observed data using genetic algorithms.
Thus, Jenerette and Wu's predictions of urban change in the region combine statisti-
cal extrapolation with simulation modelling. Jenerette and Wu (2001) deemed the
performance of the coarse-scale model over the period 1975-95 to be satisfactory.
However, the fi ne-scale model did not perform as well, which Jenerette and Wu
attributed to a mismatch in the scales at play in the system and in the model.
Jenerette and Wu (2001) also experienced problems with the models' temporal
(re)scaling. Estimation of the transitions between different land uses was based on
observed data separated by a 20-year interval. These data had to be downscaled
to annual transitions, but this downscaling failed when urbanisation was 'non-
accretive' (i.e., occurred in entirely new parts of the landscape).
Agent-based models (ABMs) explicitly simulate interactions between autono-
mous goal-seeking entities, especially, in the case of LUCC, in some sort of dynamic
landscape. Over the last decade ABMs have received increasing attention as tools
for exploring human-environmental interactions and change (e.g., see Parker et al.,
2003). One reason that they have been so eagerly adopted is dissatisfaction with
the analytical rational-choice models traditionally used by economists. It has even
been argued that bottom-up modelling (of which ABMs are a conspicuous compo-
nent) represents a new 'generative' approach to social (Epstein, 1999) and landscape
sciences (Brown et al., 2006).
An interesting use of ABMs of LUCC, in its broadest sense, is the reconstruction
of human-environment interactions. One of the best known of such applications is
the 'Artifi cal Anasazi' model. The Anasazi were a Puebloan (meso-American) group
who occupied parts of the south-west of the USA. The Anasazi developed a rich
culture in and around Long House Valley (NE Arizona) from about 1800 BC. before
a rapid collapse triggered abandonment of these sites c.1300 AD. Detailed recon-
structions of palæoecological and palæoclimatic conditions, based on dendrochro-
nology and analysis of Packrat middens, have enabled estimates of annual maize
production and hydrological dynamics, which have been used to parameterise the
model. ABMs of this social system have been developed covering the period 300-
1300 AD; in these models, the individual households are the agents (Dean et al.,
2000; Axtell et al., 2002; Gumerman et al., 2003). The 'Artifi cal Anasazi' ABM
follows the fate of individual families in the valley with households fi ssioning (as
female agents age and marry) and moving in the landscape in response to water
availability and food production. Early versions of the 'Artifi cial Anasazi' model
(Dean et al., 2000) included few differences between individual actors and limited
heterogeneity in the physical environment. Although this version of the model
showed qualitative similarities to reconstructed population and settlement dynam-
ics, quantitatively it was very different in that it predicted much larger populations
and individual settlements than seems likely from the archæological record. More
recent versions of the model (Axtell et al., 2002; Gumerman et al., 2003) incorpo-
rating more spatial heterogeneity in the landscape and variation in individual agent's
characteristics provide a closer fi t to the available data. In a spatial sense, the model
now mirrors the known (from the archæological record) location of settlements,
and it also mirrors, with one crucial exception, the expansion and rapid collapse of
the population, in the face of deteriorating environmental conditions, in particular
drought and changes in the water table (fi gure 20.2). The crucial exception is that
the 'Artifi cial Anasazi' model predicts continued occupancy of the valley after it is
believed that Long House Valley was completely abandoned. Thus, the modelling
