International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B4. Istanbul 2004 
  
  
3.2 Varying the CA elements 
in CA models, the transition of the cell is based on the 
neighbourhood adopted in the CA simulation, since it affects 
the cell state change. Finding the areas of influence on the state 
of the cell is important for realistically modeling the urban land 
use change. Therefore in this study the variations of 
neighborhood size and type were examined to measure the 
sensitivity of these CA elements. 
  
  
  
  
  
  
Figure 1. Size and type of neighbourhood a. rectangular 
neighborhood; b. circular neighbourhood; c. small rectangular 
neighborhood (inner: 2 cells, outer: 4 cells); d. large rectangular 
neighborhood (inner: 4 cells, outer: 8 cells); e. small circular 
neighborhood (inner: 2 cells, outer: 4 cells): f. large circular 
neighborhood (inner: 4 cells, outer: 8 cells) 
3.2.1 Neighborhood type 
The commonly used neighborhoods in the case of two- 
dimensional raster-based CA models are Von Neumann, 
rectangular — Moore, and circular neighbourhood. In this study 
the rectangular and circular neighborhood types (Figure ! a. b). 
were chosen in order to evaluate the effects of neighbourhood 
type on simulation results. The impacts of these neighborhood 
sizes on the CA model outcome were addressed for 50m, 100m, 
250m. and 500m spatial resolutions. 
88 
3.2.2 Neighborhood size 
In the literature. both smaller and larger neighborhood sizes 
have been applied to the models of urban growth (Figure 1) 
(Clarke and Gaydos, 1998, Wu 1998, White and Engelen, 1993 
and 1997). However, no particular validation on what is the 
appropriate neighborhood size (e.g. four or six cells in radius) 
has been made in these urban model applications. 
Furthermore, due to the bifractal characteristics of the cities, as 
stated by White et al. (1997), White and Engelen (1993), and 
Batty and Longley (1994), an urban area can be divided into 
two zones. Inner zone corresponds to the area where the urban 
growth is considered finished or slow dynamics of 
transformations are expected. Otherwise, growth is considered 
still ongoing or faster in the outer zone (White et al. 1997; 
Barredo et al., 2003). In consideration of this fractal structure, 
in this study two zones are defined in a neighborhood: inner and 
outer on which development of the cell depends. Both the 
rectangular (Figure | c, d) and circular. (Figure «le, :id) 
neighborhood types were defined based on these two zones. 
Two different sizes were specified to represent small and large 
neighborhood size. The size of the small neighborhood is 2 cells 
and of the large is 4 cells surrounding the central cell. The outer 
size is also defined to contain 4 and 8 cells for small and large 
neighborhood sizes, respectively. Therefore, small and large 
neighborhood effects were compared with each other. The 
impacts of these neighborhood sizes on the model outcome 
were addressed for 50m, 100m, 250m, and 500m spatial 
resolutions. 
4. RESULTS AND DISCUSSION 
GIS-based urban CA simulations for San Diego Bay area were 
produced for temporal interval of 10 years and by using one- 
vear time increment. The results were obtained by varying the 
neighbourhood size and type for different spatial resolutions 
and the outcomes were compared using the integrated 
approaches of sensitivity analysis. 
The simulation results of different neighbourhood tvpes were 
produced. It was observed from the results that the increase of 
the cell size cause a decrease in KAPPA - from 0.93 for 50m to 
0.66 for 500m. This indicates that variation of neighborhood 
type does affect the discordance of the obtained map outputs. 
Figure 2 represents the cross-tabulation map of circular and 
rectangular neighbourhood simulations for 250m cell size. The 
cross-tabulation map indicates the emergence of commercial 
land-use areas when both rectangular and circular 
neighborhoods are used. The class area graph (Figure 3a) 
depicts the smaller discordances in generated surfaces for 
commercial land-use type but bigger discordance for housing 
land-use type. The graph of FD (Figure 3b) does not suggest 
any major variation. 
The pattern of discordance in class area spatial metric graph 
was detected to have a tendency to increase in the simulation 
results at 500m cell size. However, no significant discordances 
were observed from the class area spatial metric graphs for 50m 
and 100m cell sizes. For all other spatial resolutions, the similar 
values of the FD were obtained. 
In