Geoscience Reference
In-Depth Information
C9
C8
C7
C6
C5
C4
C3
C2
C1
62.55%
78.65%
86.56%
89.42%
99.63%
100%
70.83%
100%
98.04%
100
0
10
20
30
40
50
60
70
80
90
Percentage
FIGURE 11. 5
System performance against nine validation criteria. (From Demetriou, D. et al.,
Environ.
Plann. B
, 39(4), 609, 2012b.)
been accounted for by the ES. However, they did not appear to have much effect on the overall
performance of the ES.
Although the system clearly performs well when compared to the human expert, the biggest gain
is in time. This problem would take an expert around 30 days to solve based on a survey carried
out in which 10 land consolidation experts were asked to evaluate this case study. In contrast, the
ES took 6 min to produce a single solution. In addition, it is possible to evaluate many different
alternatives, which would not be possible using the regular manual approach. Finally, as shown in
Demetriou et al. (2012a), the system produced better solutions than that of the human experts due
to its objectivity, although more experimentation and additional case studies are needed to test this
hypothesis further.
11.9 FUTURE DIRECTIONS
Although ES have become an accepted technology in many fields such as medicine and finance,
where the benefits of these systems have been clearly revealed, this is less apparent in the geospatial
sciences and GC. It is clear that the integration of GIS with ES has been used for solving a variety
of spatial problems since the 1980s with a large number of studies appearing since the last edition
of GC in 2000. However, many of the results of these studies have revealed some disappointment
in the solutions. Moreover, there are a number of outstanding issues that have hindered the develop-
ment and transfer of these tools into operational planning practice. For example, there are a number
of different ways to build a spatial ES, but there is currently a lack of appropriate and user-friendly
mechanisms (e.g. ES shells) embedded directly within proprietary GIS that would facilitate the
specification and incorporation of specific problem knowledge into a system. In addition, the lack
of self-learning capabilities in ES is an inherent weakness that limits the ability of ES to adequately
address the dynamic aspects of many real-time spatially relevant problems. Thus, the ability to add
intelligence to both GIS and ES through the addition of other AI techniques to develop self-training
ES, for example, is only possible through completely bespoke solutions. Although very substan-
tial progress in interoperability and shared ontologies (Goodchild 2009) has been made within the
GIScience and GC communities, more attention should be shifted towards developing flexible inte-
gration tools within GIS for embedding AI techniques and capturing knowledge more generally. If
these limitations can be addressed, then the next generation of spatial ES and intelligent GIS may
really help solve more complex and ill-structured spatial problems in the future.
