SENSITIVITY ANALYSIS OF A GIS-BASED CELLULAR AUTOMATA MODEL
V. Kocabas“ *, S, Dragicevic“
a Simon Fraser University, Department of Geography, Spatial Analysis Modeling Laboratory
8888 University Dr., Burnaby, BC, Canada, V5A1S6 — verdak@sfu.ca, suzanad@sfu.ca
KEY WORDS: Sensitivity Analysis, Cellular Automata, Modeling, Urban Growth, Spatial Metrics.
ABSTRACT:
Urban growth is dynamic and complex spatial process that has severe environ
results in the transformations of forested areas or high quality agricultural lands
urban growth and land-use change processes are complex due to the difficulties to re
and human environments. One of the models increasingly applied to ur
models are approximations of the real world, they contain inherent errors due to
parameters and model misspecification thereby generating uncertainties in the resu
(SA) of a GIS-based CA urban growth model. The impacts of changing CA
ddressed. The cross-classification, KAPPA statistic and spatial metrics
uncertainties through the sensitivity analysis
neighbourhood size and type on the model outcome were a
were used as measures of sensitivity analysis in order to understand the CA model beh
ights for improving the capabilities of current CA models to create more realistic output scenarios.
research can provide better ins
1. INTRODUCTION
Urban growth modeling can assist urban and regional planners
to foresee impacts of their actions and policies (Wegener,
1994). To date, various urban growth models are developed due
to the simple vs. complex. aspatial vs. spatial views of urban
phenomena. For more then a decade, research focus has been on
models using cellular automata (CA) theory as the approach
which is capable to address the spatial complexity of the urban
change process (Allen, 1997). CA models are receiving more
attention due to the capability for handling spatial and temporal
dimensions, using bottom-up approach, relying on geospatial
data and capacity to couple with raster-based geographic
information system (GIS) as well as with other approaches such
as agent-based or multi-criteria evaluation (Batty, 1998; Wu,
1998: O'Sullivan and Torrens, 2000).
The main advantages of CA are in their simplicity, easy
integration with raster GIS, and adaptability to various urban
growth situations. CA models can generate complex patterns
through the use of simple rules (Wolfram, 2002). In particular,
it is possible to realisticallv represent spatial complexity and
dynamics of urban growth change by choosing the
configurations of basic elements of CA models such as cell
states, cell size, neighborhood size and type, transition rules and
temporal increments (Torrens, 2000; White and Engelen, 2000;
Yeh and Li, 2001). In most of CA urban growth models the
effects of varying different basic elements of CA are not yet
fully addressed in the research literature. This study introduces
an approach to sensitivity analysis of CA model in order to
analyze the model responses and behavior with respect to the
change of its elements more particularly neighborhood size and
type. This study provides the assessment of CA model
sensitivity versus its elements’ variations, the consistency of
model outcomes and the locational differences at specific land
use types.
* Corresponding author.
mental and social impacts. High population growth
into urban land-use. The study and modeling of the
present the interactions between the physical
ban research is based on cellular automata (CA) theory. Since
the digital data input and are sensitive on model
Its. The objective of this study is to explore these
avior and its limitations. The results from this
2. SENSITIVITY ANALYSIS
Sensitivity Analysis (SA) addresses the relationship of
information flowing in and out of the model and deals with the
sources of variation influencing the model outputs (Saltelli et
al., 2000). [t measures the change in the model output relative to
a change in one or more of the input parameter values. In
modeling practices, SA is a prerequisite since it determines the
reliability of the model through assessing the uncertainties in
the simulation results. It can also be considered as a resource
optimization process since data gathering is the most important
and expensive part of GIS. SA can be used also to test sub-
models of the actual model and to determine the dependency of
model outcomes on input parameters (Crosetto et al., 2000).
In current CA applications, SA is often used to help understand
the behaviour of a model as well as the coherence between a
model and the real world. The most common approach is based
on variations of basic spatial and temporal CA elements, which
represent input parameters in order to assess the outcome
differences (White et al., 1997; Barredo et al., 2003). However,
some recent studies have addressed the issue of errors and
uncertainties related to CA models (Yeh and Li, 2003) and
provided the analysis of CA model behavior with respect to
changing model components (Clarke, 2003). The appropriate
choice of transition rules were considered as the key component
of the CA model (Childress et al.. 1996). White et al. (1997)
changed the transition rules in the CA model and compared the
model results and simulation outputs visually and cell-by-cell
with the actual land use. This approach can be regarded in
current CA literature as a calibration procedure since model and
simulation outputs were compared with the real data.
Variations of different cell size and cellular configurations were
explored by Chen and Mynett (2003) and Jenerette and Wu
(2001). Moreover, Liu and Andersson (2004) examined the
effécts of temporal dynamics on the behavior of a CA-based
86
Inte
ID e
ENN lS
C0 — 0 o 9 0B ped TA AN
m = m
zn, ÓmI
ED r= —
IN
