Environmental Engineering Reference
In-Depth Information
14.1 Introduction
ancillary data for dasymetric mapping, with emphasis on both
population mapping and sociodemographic data redistribution.
We illustrate some of these dasymetric mapping approaches with
examples from Gwinnett County, Georgia, which is part of the
Atlanta metropolitan area. While areal interpolation, along with
other statistical techniques, is often used for population estima-
tion, population estimation per se is not the focus of this chapter.
Readers interested in that particular application are referred to
the chapter in this text by Wang.
Statistical data are commonly required for social science research
and practice. These data originate from a variety of sources,
such as government agencies, non-governmental organizations,
and the private sector. Because of variations in data collection
purposes and jurisdictions, statistical datasets are compiled for
different administrative units. Quite often, these administrative
units differ in spatial extent. Furthermore, concerns over individ-
uals' rights to privacy have given rise to data release restrictions
aimed at preserving confidentiality; this often results in the aggre-
gation of individual data to arbitrary administrative units prior
to data dissemination. For survey data, limitations in sample
sizes can lead the data collection agency to aggregate responses to
geographic levels that do not meaningfully coincide with spatial
variations in the data, but rather are chosen for the statistical
stability of the resulting data estimates. The choices of areal unit
to which statistical data are aggregated vary by agency and pur-
pose of the dataset. For example, some health-related datasets
aggregate to the level of respondents' ZIP code. Other datasets
are aggregated to the county-level. Census data are available at
various aggregations that correspond to a complex hierarchy of
US Census geography. These constraints in data collection, aggre-
gation, and reporting create great difficulties and pose significant
methodological challenges to data users.
An additional complication that is related to the issue of data
spatial incompatibility is the possibility of temporal changes in
administrative boundaries. Data from a particular agency may
be reported consistently for the same administrative boundaries,
yet these boundaries may be revised over time. For example,
changes in census geography such as block groups and census
tracts (Howenstine, 1993), ZIP codes (or their spatial corollaries,
ZIP code Tabulation Areas), US Congressional Districts, or
census enumeration and dissemination areas in other countries
(e.g., Canada) may require the use of methods designed to
account for temporally-induced spatial mismatches (Martin,
Dorling and Mitchell, 2002; Gregory, 2002; Schuurman et al .,
2006; Schroeder, 2007).
Gotway and Young (2002) present a detailed discussion of the
problems associated with combining data from multiple spatially
incompatible datasets. They describe the Change of Support
Problem, or COSP, in which the area or volume associated with
the data, as well as the shape and orientation of the spatial units
for which the data are collected, can affect statistical associations
based on those data. The Modifiable Areal Unit Problem or
MAUP (Openshaw, 1984) is a special case of the COSP, in which
varying scales and spatial aggregations of units may result in a
change in statistical analysis results for areal data.
Where it is necessary to use data that are reported in incom-
patible spatial units, a commonly used approach is to employ
areal interpolation techniques. Areal interpolation is the process
of transforming a dataset that is collected and/or reported in one
set of areal units to another set of areal units (Goodchild and Lam,
1980; Lam, 1983; Flowerdew and Green, 1994). For example, data
reported at the ZIP code level can be apportioned to census tracts
through a variety of areal interpolation techniques. These tech-
niques, which will be described further below, vary in terms of
methodological approach, complexity, and accuracy. The focus
of this chapter is to describe in detail one of these areal interpo-
lation approaches: dasymetric mapping. We will also discuss the
specific role of remotely sensed satellite imagery as a source of
14.2 Dasymetric maps,
dasymetric mapping, and
areal interpolation
There exists the potential for confusion between the terms and
concepts of dasymetric maps (as a product), dasymetric map-
ping (as a process), and areal interpolation. A dasymetric map
is an area-based cartographic tool that enables representation
of a statistical phenomenon, such as population density (Lang-
ford, 2003). Dasymetric maps are meant to convey both the
magnitude of a statistical surface as well as the spatial extent of
the phenomenon being mapped. Slocum et al . (2009) describe
dasymetric maps as an alternative to choropleth maps:
Like the choropleth map, a dasymetric map displays stan-
dardized data using areal symbols, but the bounds of the
symbols do not necessarily match the bounds of enumera-
tion units (e.g., a single enumeration unit might have a full
range of gray tones representing differing levels of popula-
tion density)
(Slocum et al., 2009, p. 271).
In contrast, the term ''dasymetric mapping'' represents the pro-
cess of transforming data that are aggregated arbitrarily (usually
due to the constraints and requirements of data collection) into a
dasymetric map that is a more accurate depiction of the statistical
distribution of the data. Mennis (2009) convincingly argues that
the dasymetric mapping process usually does not yield true dasy-
metric maps, because the dasymetric mapping process results in
mapped distributions that are estimated from a disaggregation of
choropleth maps units.
To better illustrate this principle, consider a choropleth map,
in which the entire land surface is divided into a space-filling tes-
sellation and each area is assigned a value of the phenomenon and
color-shaded accordingly. It is implicitly assumed in choroplethic
mapping that the phenomenon of interest is homogeneously dis-
tributed throughout each particular mapping unit and that the
distribution of the phenomenon takes places over the entire
mapped area. It is also assumed that sharp breaks in data values
occur at the boundaries of mapping units. In the population
mapping example, this would translate to the assumption there
are people residing in all areas and conversely there are no
unpopulated areas. Crampton (2004) provides a critical historical
overview of choropleth and dasymetric maps and argues in favor
of the use of dasymetric maps on both a conceptual/theoretical
basis as well for reasons of geostatistical soundness.
A dasymetric map can be considered a variation of the
choropleth map in the sense that the mapping units are spatially
Search WWH ::




Custom Search