Location and GIS - Location Science - page 503

and analytical capabilities of GIS have inspired location analysts to extend classical
location science models by taking into account geographical information. A typical
example is given by Suárez-Vega et al. ( 2011 ) who developed a multicriteria com-
petitive model for locating a hypermarket on a network. They extended traditional
Huff-based competitive location models, originally proposed by Huff ( 1964 ), by
employing GIS procedures to consider geographical aspects such as distance to the
main roads, land-use, slope of the terrain and distance to the distribution centers.
Another important contribution of GIS concerns the nature of the entities involved
in location models. In conventional location models the facilities to be located and
the customers are usually represented by points. This is clearly not sufficient in many
problem settings since the customers as well as the facilities may be described by
general objects other than points, such as lines, polygons or even other irregular
shapes. Using the computational geometry techniques implemented within GIS,
the adjacency and contiguity between such objects may be determined and their
distances to other objects may be calculated. As a result, various models have been
developed where the facilities to be located are described by lines, bands (corridors)
or polygons. Perhaps the most indicative example is the corridor location model
and its variations. This is the problem of identifying one or more paths across
a landscape such that some criteria are met. These paths may represent power
transmission lines such that their cost is minimized, roads through a region such
that visibility of a beautiful landscape is maximized, routes for a military unit that
minimize the view or the probability of being detected, etc. Although this problem
does not fall exclusively within the domain of GIS, its detailed study requires the
capabilities of GIS. Hence, nearly all the approaches to solving this problem employ
GIS at some stage. Some of the earlier approaches to the corridor location problem
are reviewed by Church ( 2002 ) whereas more recent ones can be found in Gonçalves
( 2010 ).
A crucial issue in location science, especially when large amounts of data are
available, is to determine the level of aggregation or the scale where the elements
of a particular problem will be represented. For instance, if population data is
available at census block level, should this detailed data be directly used in a
location model or should it be aggregated at block group or census tract level?
Including all the data may imply significant effort and cost for data collection
and require considerable computation time, thus making the problem intractable.
On the other hand, aggregating the data reduces the amount of computation work
required but introduces errors in the analysis, as originally shown by Goodchild
( 1979 ). Shortly afterwards Openswaw and Taylor ( 1981 ) remarked that the results
of spatial analysis may vary depending on the representation scheme adopted. This
effect, known as the Modifiable Areal Unit Problem (MAUP), can be divided into
two components, the aggregation level used and the type of unit utilized. Clearly,
any model suffering from this effect is problematic and should be used with caution.
As far as location models are concerned, Murray ( 2005 ) showed that the Set
Covering Location Problem is susceptible to MAUP and, then, he developed a GIS-
based alternative formulation that can be applied to points, lines, polygons or other
objects. Moreover, Tong and Murray ( 2008 ) and Alexandris and Giannikos ( 2010 )