Geoscience Reference
In-Depth Information
TABLE 7.9
Goodness of Fit Statistics for the Initial SEM
2004/2005
2005/2006
Layer (NCCOA Group)
SRMSE
R
2
AED
SRMSE
R
2
AED
Overall SEM performance
All years
3.09
0.69
0.02
3.05
0.69
0.02
Performance of individual layers
3, countryside
3.40
0.76
0.29
3.48
0.77
0.32
6, typical traits
2.98
0.71
0.21
2.94
0.72
0.22
4, prospering suburbs
2.70
0.79
0.17
2.65
0.79
0.17
1, blue collar communities
2.55
0.74
0.06
2.51
0.75
0.06
2, city living
4.13
0.51
0.02
4.02
0.53
0.00
5, constrained by circumstance
3.08
0.66
0.17
3.06
0.65
0.15
7, multicultural
3.88
0.45
0.40
3.81
0.48
0.41
Source:
Harland, K. and Stillwell, J., Commuting to school: A new spatial interaction
modelling framework, in
Technologies for Migration and Population Analysis:
Spatial Interaction Data Applications
, eds. Stillwell, J.C.H., Duke-Williams, O.,
and Dennett, A., IGI Global Snippet, 2010.
first preference. Table 7.9 shows the results from the initial run of the SEM for 2004/2005 and
2005/2006. The overall results for the SEM outputs are good with SRMSE values of 3.09 and 3.05,
respectively, and
R
2
of 0.69 and AED 0.02 for both years. This indicates a good overall model fit.
Each layer in the SEM shows an increase in performance against the calibration statistics (shown
previously in Table 7.8), except for the first layer to execute, NCCOA group three, and the last layer
to execute, NCCOA group seven. These two layers show degraded performance in all goodness of
fit statistics with respect to the calibration statistics.
The research presented here has demonstrated that GAs can be successfully applied to the
exploration and construction of optimal SI model equations. This is particularly advantageous for
calibrating complex models that contain multiple parameters and have a potentially large solu-
tion space. Following the previous work of Diplock (1998), Diplock and Openshaw (1996) and
Openshaw (1998), we recommend that an SI model equation and constraint equation can be, and
indeed should be, considered as two distinct stages. Initial exploration has shown that the incorpora-
tion of the calibrated layers improved the overall fit of the model.
7.9 DISCUSSION AND CONCLUSIONS
The aim of this chapter has been to examine the use of EAs for the optimisation of geographi-
cal model parameters and the development of geographical models. A brief overview of the core
building blocks of EAs and the mechanisms by which they operate was provided followed by
two case studies, which demonstrated that GAs can be used to derive optimal parameters for an
ABM of a spatially driven retail market and to breed SI model equations in the area of education.
The purpose of presenting these applications was twofold: first, to show how to construct a GA
and second, to show how readily this approach can be linked to other techniques, in this case, an
ABM and an SI model.
The research domain of EAs is considerable and continually growing; however, uptake within
geography still remains slow in comparison to other disciplines. The use of EAs other than GAs
is also still at an early stage. The original chapter was written at a time when the scope for the
