Environmental Engineering Reference
In-Depth Information
degraded to the level of the next-best dasymetric method (Fisher
and Langford, 1996). Binary dasymetric mapping also can be
more accurate than regression-based models of areal interpo-
lation; Langford (2003) attributes this to the fact that binary
dasymetric maps are calibrated to each areal unit of the source
data, whereas regression models are typically fitted at the global
level. An additional property of binary dasymetric maps (and of
three-class maps as well) is the preservation of volume for each
areal unit; this property was noted by Lam (1983) to hold true
for area-based areal interpolation methods in general. In other
words, the sum of population counts (dasymetrically derived)
within areal units remains the same as for the original source
zones prior to dasymetric mapping. This property, also known
as the pycnophylactic property (Tobler, 1979), is important in
areal interpolation tasks because of the desire to avoid statistically
creating or eliminating population, which would introduce error
in the resulting population or sociodemographic data estimates.
Despite its outward simplicity, binary dasymetric mapping is
equally applicable at a variety of spatial scales (Langford, 2003),
which makes it a flexible approach, adaptable to a variety of
application settings.
he sampled the population of all block groups that are com-
pletely contained within their respective LULC classes, on a
county-by-county basis to account for spatial heterogeneity of
densities across his study area of metropolitan Philadelphia,
Pennsylvania. He used the population density fractions to assign
the percentage of each block group's population to each of three
LULC classes in each county. These percentages were further
adjusted by an area-weighted fraction to account for the differ-
ential areas of each LULC within each block group. An extension
to Mennis' 2003 work is the ''Intelligent Dasymetric Method'',
or ''IDM'' (Mennis and Hultgren, 2006). They combine ana-
lysts' expert judgment along with a flexible empirical sampling
approach in order to quantitatively derive population densities
for individual LULC classes. They apportion population to target
zones based on a ratio of these empirically-derived densities. In
their tests, they found the IDM method to be more accurate than
a simple areal weighting approach and at least as accurate as a
binary dasymetric approach.
14.2.4.3
Dasymetricmapping with vector
ancillary data
14.2.4.2
Three-class (or N-class)
dasymetric
In contrast to the previously described dasymetric methods that
rely upon ancillary data from remote sensors, other researchers
have used either vector data or a combination of vector, raster,
and census data to delineate dasymetric mapping units. In two
examples of using vector data for dasymetric mapping, Xie (1995)
and Reibel and Bufalino (2005) use street network data derived
from TIGER/Line files produced by the US Census Bureau. Xie
illustrates three techniques to apportion population from source
zones to the street network segments, including segment length,
street segment class, and the number of houses presumed to
be contained within the address ranges of each street segment.
This set of methodologies is similar to areal weighting with the
main difference being that population is assumed to be evenly
distributed along a one-dimensional line as opposed to a two-
dimensional area. Xie's research indicated that using a weighted
approach - namely the use of street segment classes - provided
the most accurate results for population distribution. Reibel and
Bufalino (2005) replicated the street-weight count approach orig-
inally proposed by Xie (1995) and conducted an error analysis,
in which they obtained smaller errors than from using an area-
weighting technique. Reibel and Bufalino chose their vector-only
methodology as a way to promote the use of areal interpolation
in applied situations. They state that ''dasymetric weighting using
remote sensing has been almost completely restricted to compu-
tational experiments by geographers and allied spatial scientists''
(Reibel and Bufalino, 2005, p. 129), and it is their contention
that using remote sensing data requires specialized raster data
processing skills and raster GIS software.
Maantay, Maroko and Herrmann (2007), use cadastral data
for New York City, and incorporate an expert system (the
Cadastral-based Expert Dasymetric Systems, or CEDS) to deter-
mine the cadastral variable (either number of residential units or
the adjusted residential areas for each tax lot) that fits the data
best. They critique the use of remote sensing data for ancillary
data on four points: first, they suggest that remote sensing data
do not adequately capture the precision needed to adequately
model population density in highly urbanized areas, due to
limitations to satellite imagery spatial resolution and intrapixel
heterogeneity (citing a report by Forster from 1985, prior to
Several researchers have noted the utility of binary dasymet-
ric mapping yet have suggested avenues for improving on the
method. Flowerdew, Green and Kehris (1991) may have been
the first to suggest that the binary method is crude and some-
what limited. They suggest that additional information about
the distribution of the variable of interest may be helpful (they
used clay versus limestone soil belts and fitted the data using an
expectation-maximization (EM) algorithm). Such an extension
to the binary method is an approach that apportions population
to specific LULC categories, such as residential, urban, forested,
and agricultural (Holloway, Schumacher and Redmond, 1996;
Eicher and Brewer, 2001) or high-density urban, low-density
urban, and nonurban (Mennis, 2003); this has been termed the
''three-class method'' (Eicher and Brewer, 2001, p. 129). Con-
ceptually, there is no limitation to the number of classes to which
population may be apportioned; therefore a more-appropriate
name for this method in general may be the ''N-class dasymet-
ric'' method. In Eicher and Brewer's example for Southwestern
Pennsylvania, they utilized landcover data provided by the US
Geological Survey, and assigned 70% of each county's population
to urban land-use polygons, 20% to agricultural and woodland
polygons, and the remaining 10% to forest polygons.
The three-class (or N-class) method is made feasible by
improvements in classification accuracy and calibration of pop-
ulation densities for multiple residential classes. The challenge in
this method is determining the appropriate assignment of den-
sity values per class (Langford, 2003); a difficulty that is avoided
in binary dasymetric mapping because 100% of population is
assigned to just one class and 0% to the other class. A second
weakness, identified by Eicher and Brewer (2001) as the major
weakness of their approach, is that ''the three-class method does
not account for the area of each particular LULC within each
county'' (Eicher and Brewer, 2001, p. 130).
Recent research has addressed both of these methodological
concerns. Mennis (2003) introduces a technique for estimating
the population density fraction of each LULC type, in which
Search WWH ::
Custom Search