Dasymetric mapping for population and sociodemographic data redistribution - Urban Remote Sensing: Monitoring, Synthesis and Modeling in the Urban Environment - page 186

refined to include only those areas that contain the phenomenon
of interest (e.g., populated areas). This spatial refinement is
achieved through the use of ancillary data, which represent
a separate phenomenon that is considered to be related to
the distribution of the mapped phenomenon itself (Flowerdew,
Green and Kehris, 1991). The result is not a continuously varying
surface, nor a space-filling surface, but instead a discretized set of
spatially refined areas (either polygons or pixels, depending upon
the geospatial data model used) onto which the cartographic
representations for the phenomenon are assigned. The set of
spatially-refined areas are estimates of the spatial distribution of
the phenomenon, which in fact may be unknown.
Finally, the term ''areal interpolation'' refers to the process of
transforming data from one set of source zones (e.g., ZIP codes)
to a set of spatially incompatible target zones (e.g., census tracts)
(Goodchild and Lam, 1980). Therefore, the spatial redistribu-
tion of population and other sociodemographic data, as we will
focus our discussion in this chapter, involves two transformation
processes: dasymetric mapping, which involves data disaggrega-
tion, to in turn enable areal interpolation, which involves data
reaggregation to a set of desired areal units.
While we focus on methodological aspects of areal interpola-
tion, there are numerous examples of its use in applied settings,
for example: examining the economic impact of water usage in
California (Goodchild, Anselin and Deichmann, 1993), analyz-
ing spatially-aggregated crime data (Poulsen and Kennedy, 2004),
revealing urban sprawl (Lo, 2004), understanding spatial dynam-
ics of asthma (Maantay, Maroko and Herrmann, 2007; Maantay,
Maroko and Porter-Morgan, 2008), mapping urban risks for
flood hazards (Maantay and Maroko, 2009), environmental jus-
tice/equity analysis (Mennis, 2002; Boone, 2008); generating
burden of disease estimates (Hay et al ., 2005), refining spatial
accessibility measures for healthcare (Langford and Higgs, 2006),
catastrophic loss estimation (Chen et al ., 2004), and developing
a National Agriculture Land Use Dataset (Comber, Proctor and
Anthony, 2008).
Various methods for areal interpretation have been devel-
oped. These can be distinguished by whether or not the method
involves the use of ancillary data to estimate the relationship
between population and a related variable (e.g., land use). Areal
interpolation methods that do not rely upon ancillary data can
be further distinguished as either area-based methods or point-
based methods (Wu, Qiu and Wang, 2005). Area-based methods
include polygon overlay (Lam, 1983), also known as areal weight-
ing (Goodchild and Lam, 1980; Howenstine, 1993; Sadahiro,
2000) and pychnophylactic interpolation (Tobler, 1979; Rase,
2001). Point-based methods include kriging (Kyriakidis, 2004)
and kernel-based methods (Bracken and Martin, 1989; Martin,
1989, 1996).

(Flowerdew, Green and Kehris, 1991; Flowerdew and Green,
1992, 1994; Eicher and Brewer, 2001).
Several types of ancillary data exist and have been evaluated
for their utility and accuracy in areal interpolation, including:
expert knowledge of the study area (Wright, 1936; Mennis and
Hultgren, 2006), street network data (Xie, 1995; Mrozinski and
Cromley, 1999; Chen et al ., 2004; Reibel and Bufalino, 2005),
zoning/parcel/cadastral data (Maantay, Maroko and Herrmann,
2007; Maantay, Maroko and Porter-Morgan, 2008; Maantay
and Maroko, 2009), scanned raster maps (Langford, 2007), and
remotely sensed data (Langford, Maguire and Unwin, 1991;
Langford and Unwin, 1994, Eicher and Brewer, 2001; Harvey,
2002; Holt, Lo and Hodler, 2004; Liu, Kyriakidis and Goodchild,
2008; Bhaduri et al . 2007; Briggs et al ., 2007; Kressler and
Steinnocher, 2008).
Further distinctions can be made based on how the relation-
ship between the ancillary data and the variable of interest is
estimated. These can be either apriori or empirically specified;
the former is based on expert knowledge and/or assumptions
(which can be arbitrary) and the latter is determined through
statistical techniques. Examples of apriori specification include
Wright (1936), Langford and Unwin (1994), Eicher and Brewer
(2001) and Holt, Lo and Hodler (2004). Statistical methods
include regression techniques (Flowerdew and Green, 1989,
1992; Flowerdew, Green and Kehris, 1991; Langford, Maguire
and Unwin, 1991; Goodchild, Anselin and Deichmann, 1993;
Harvey, 2002; Yuan, Smith and Limp, 1997; Langford, 2006;
Lo, 2008). The latter three papers accounted for the possibility
of spatial non-stationarity in the relationships between ancillary
data and population, and Lo (2008) addressed this through the
application of geographically weighted regression (Fothering-
ham, Brusndon and Charlton, 2002) to a case study of data for
metropolitan Atlanta, Georgia.

14.2.2 Dasymetric mapping

The focus of this chapter is the use of dasymetric mapping for
areal interpolation. As previously defined, dasymetric mapping
is the process of transforming data from one spatial aggrega-
tion (typically a choropleth map) into a map that is a more
accurate depiction of the magnitude and spatial extent of the
data. This process involves the use of additional information
about the distribution of the data, so as to make the resulting
redistribution more meaningful. We first discuss the origins of
dasymetric mapping, which has been somewhat misunderstood
until recently, followed by methodological variations to the dasy-
metric approach, including binary dasymetric and three-variable
dasymetric mapping. We then discuss different approaches to
incorporating intelligence into the process, through different
forms of ancillary data.

14.2.1 Ancillary data

Ancillary data are a necessary component for areal interpolation
through dasymetric mapping as they are the basis on which
the unknown statistical surface of the phenomenon of interest
is modeled. The key assumption in the use of ancillary data is
that their distribution is related to the variation in the statistical
surface (Eicher and Brewer, 2001). The use of ancillary data,
regardless of the type or source of the data, is an attempt to add
'intelligence' to the process of understanding this relationship

14.2.3 Origins

The seminal article on dasymetric mapping for most English-
speaking audiences is J.K. Wright's (1936) description of a
dasymetric method for redistributing population densities for
Cape Cod, Massachusetts. Wright relied on information on
population distributions from a combination of an almost (at