Mapping Population Distribution in the Urban Environment: The Cadastral-Based Expert Dasymetric System (CEDS)

Abstract

This paper discusses the importance of determining an accurate depiction of total population and specific sub-population distribution for urban areas in order to develop an improved "denominator," which would enable the calculation of more correct rates in GIS analyses involving public health, crime, and urban environmental planning. Rather than using data aggregated by arbitrary administrative boundaries such as census tracts, we use dasymetric mapping, an areal interpolation method using ancillary information to delineate areas of homogeneous values. We review previous dasymetric mapping techniques (which often use remotely sensed land-cover data) and contrast them with our technique, Cadastral-based Expert Dasymetric System (CEDS), which is particularly suitable for urban areas. The CEDS method uses specific cadastral data, land-use filters, modeling by expert system routines, and validation against various census enumeration units and other data. The CEDS dasymetric mapping technique is presented through a case study of asthma hospitalizations in the Bronx, New York City, in relation to proximity buffers constructed around major sources of air pollution. The case study shows the impact that a more accurate estimation of population distribution has on a current environmental justice and health disparities research project, and the potential of CEDS for other GIS applications.

Full text

Cartography and Geographic Information Science, Vol. 34, No. 2, 2007, pp. 77-102
Mapping Population Distribution in the Urban
Environment: The Cadastral-based Expert
Dasymetric System (CEDS)
Juliana Astrud Maantay, Andrew R. Maroko,
and Christopher Herrmann
ABSTRACT: This paper discusses the importance of determining an accurate depiction of
total population and specific sub-population distribution for urban areas in order to develop
an improved “denominator,” which would enable the calculation of more correct rates in GIS
analyses involving public health, crime, and urban environmental planning. Rather than
using data aggregated by arbitrary administrative boundaries such as census tracts, we use
dasymetric mapping, an areal interpolation method using ancillary information to delineate
areas of homogeneous values. We review previous dasymetric mapping techniques (which
often use remotely sensed land-cover data) and contrast them with our technique, Cadastral-
based Expert Dasymetric System (CEDS), which is particularly suitable for urban areas. The
CEDS method uses specific cadastral data, land-use filters, modeling by expert system routines,
and validation against various census enumeration units and other data. The CEDS dasymet-
ric mapping technique is presented through a case study of asthma hospitalizations in the
Bronx, New York City, in relation to proximity buffers constructed around major sources of
air pollution. The case study shows the impact that a more accurate estimation of population
distribution has on a current environmental justice and health disparities research project,
and the potential of CEDS for other GIS applications.
KEYWORDS: Dasymetric, geographic information systems, cadastral maps, areal
interpolation, areal weighting, expert systems, thematic maps, asthma, New York City
What is Dasymetric Mapping?
Dasymetric mapping refers to a pro-
cess of disaggregating spatial data to a
finer unit of analysis, using additional
(or “ancillary”) data to help refine locations of
population or other phenomena (Mennis 2003).
This disaggregation process will result in areas
of homogeneity that take into account (and
more closely resemble) the actual phenomena
being modeled, rather than areal units based
on administrative or other arbitrary boundaries.
Although it is generally used to get better results
for actual locations of population, dasymetric
mapping theoretically can be used to disaggre-
gate any quantitative variable that is aggregated
by geographic units, such as administrative divi-
sions including census enumeration units, ZIP
codes, counties, and police precincts; or environ-
mental districts, including watersheds, wetlands,
or flood plains.
The locations of census tract boundaries, ZIP
code postal zones, or any other administrative
boundaries, do not necessarily relate to the
underlying phenomena, having been created
arbitrarily or to suit other governmental purposes.
Population totals within a given areal zone are
assumed to be distributed evenly throughout
the zone, when in fact population distribution
is generally much more heterogeneous (Wu
Juliana Astrud Maantay, Associate Professor, Urban Environmental Geography, Director of GISc Program, Lehman College, City University
of New York, Environmental, Geographic, and Geological Sciences Department, 250 Bedford Park Blvd. West, NY 10468. Ph: (718) 960-
8574; Fax: (718) 960-8584. E-mail: <juliana.maantay@lehman,cuny.edu>; <maantay@aol.com>. City University of New York Graduate
Center, Earth and Environmental Sciences Ph.D. Program. Lehman College, City University of New York, Department of Health Sciences, Public
Health Graduate Program. Andrew R. Maroko, Lehman College, City University of New York, Department of Environmental, Geographic, and
Geological Sciences; and Geographic Information Sciences Program. City University of New York Graduate Center, Earth and Environmental
Sciences Ph.D. Program. Christopher Herrmann, John Jay College of Criminal Justice, City University of New York; and New York City Police
Department.

78 Cartography and Geographic Information Science
et al. 2005). This creates errors when trying to
establish accurate rates for GIS analyses pertaining
to health studies, crime patterns, hazard/risk
assessment, land-use planning, or environmental
impacts, among others, that rely on a smaller unit
of analysis than the original zones. Examples
of this are impact buffers that intersect the
census enumeration unit, or a different set of
zones altogether that do not coincide with the
original set (e.g., overlaying data from units with
non-coincident boundaries and/or overlapping
spatial units such as census tracts and police
precincts or health districts).
Although dasymetric mapping has been in use
since at least the early 1800s, it has never achieved
the ubiquity of other types of thematic mapping,
and thus the means of producing dasymetric maps
have never been standardized and codified the
way other types of thematic mapping techniques
have been (Eicher and Brewer 2001; Slocum
1999). Therefore, dasymetric methods remain
highly subjective, with inconsistent criteria. The
reason for this relative lack of popularity and
the paucity of standard methodology surely lies
at least partially in the difficulty inherent in
constructing dasymetric maps, and until recently,
the difficulties in obtaining the necessary data,
as well as access to the computer power required
to generate them.
The dasymetric method we have developed
uses census data in conjunction with cadastral
(tax lot) data in order to create a more precise
picture of where people actually live. Using data
aggregated by census enumeration units assumes
that population is distributed homogeneously
throughout the unit, which is rarely the case in
reality. This assumption of homogeneity results
in incorrect denominators (counts of the total
population affected by the phenomena being
investigated) being used to calculate rates
(disease, crime, impacted populations, and so
forth), which in turn results in either an under-
or overestimation of the risk or its occurrence.
The proposed Cadastral-based Expert
Dasymetric System (CEDS) leads to a better
estimation of population (and potentially
of specific sub-populations), and thus to a
more complete understanding of the spatial
distribution and patterns of disease, crime,
hazard, exposure, and other issues. Following
our review below of some of the most frequently-
used dasymetric methods, we will describe the
CEDS method, and then present it through an
example of mapping population distribution
in New York City, by comparing choropleth
mapping, areal interpolation, filtered areal
weighting, and CEDS. We then further illustrate
CEDS through a case study showing how the
CEDS method improves an environmental
health justice analysis of asthma in the Bronx.
Historical Background of Dasymetric
Mapping
Many early cartographic endeavors were con-
cerned predominantly with producing maps
intended for navigational and exploration
purposes; these required furthering our abili-
ties to observe and measure the physical world
with increasing levels of precision (Hall 1994).
Technical advancements in instrument design
and geometric theory made these more precise
maps possible, and they generally portrayed
tangible aspects of the physical world, such as
areal sizes of geographic units, topography, tem-
peratures, and sea depths (Dorling and Fairbairn
1997). Maps depicting social, cultural, or eco-
nomic aspects of the world are termed thematic
maps—those showing a particular “theme,” such
as poverty levels, disease rates, or the flow of
migration. Thematic maps (also called statistical
maps, if depicting a quantitative data theme) are
generally of more recent vintage (Dent 1999).
One of the earliest known examples of a the-
matic map is the mathematician Baron Pierre
Charles Dupin’s 1826 unclassed choropleth map
showing illiteracy levels in each of the admin-
istrative departements comprising France, where
the areal units were shaded in greytones, with
the darker tones indicating a higher illiteracy
rate (Robinson 1982: p. 232).
Although no real typology of thematic maps had
been developed at that time, most of the major
types of statistical graphics and thematic maps as
we know them today originated in the first half of
the 19
th
century as a means to visualize quantitative
information. As national governmental powers
grew and consolidated in this time period, the
need arose for a more detailed view of the popu-
lation and associated data related to population,
such as numbers about health, crime, education,
poverty, and economics (Koch 2005). Statistical
mapping met this need, and for the first time,
the types of data needed to produce these maps
were collected and made available.
Milestones in dasymetric mapping would have
to include Scrope’s 1833 classed population den-
sity map of the world, which used a rudimentary
dasymetric technique (Scrope 1833). However, the
Russian geographer, Semenov-Tyan-Shansky (1827-

Vol. 34, No. 2 79
1914), who studied under von Humboldt and Ritter
in Berlin and advanced the use of statistical map-
ping, has often been credited with inventing the
dasymetric map (Bielecka 2005). The American
geographer, John Kirtland Wright (1891-1969),
who was perhaps the first person to publish a paper
on dasymetric mapping in an English-language
journal, stated that dasymetric means “density
measuring.” His 1936 paper is generally consid-
ered the seminal paper on dasymetric mapping,
in which he extolled the virtues of the dasymetric
map over the choropleth map (Wright 1936). He
also coined the term “choropleth” (value-by-area)
map, although choropleth maps were in use since
at least the early nineteenth century.
Today, the need for visualization of population
data is even more necessary, not just for descrip-
tive purposes—to show the geographic extent and
density of populations—but also for spatial analyti-
cal and predictive modeling purposes, in order to
inform risk assessments and public policy formation
on many urban issues (Gregory 2000; Moon and
Farmer 2001; Poulsen and Kennedy 2004; Sleeter,
2004). The more traditional thematic mapping
techniques may not be sufficient to display and
analyze these data. Choropleth mapping, one of
the most widely used thematic map techniques
today, has many benefits, but it is lacking in a few
important ways. The choropleth method is familiar,
and easily comprehended and interpreted by the
map reader, and it is comparatively straightforward
to compute. For instance, population density for
a given enumeration unit can be normalized by
dividing the total population by the areal measure-
ment of the unit. However, drawbacks include the
Modifiable Areal Unit Problem (MAUP), which
describes the phenomenon that, by modifying
areal boundaries and/or the level of data aggre-
gation, the results of the spatial analysis will be
substantially different (Openshaw 1984).
Choropleth maps also have a propensity to gen-
eralize the high and low values within a given enu-
meration unit, removing the spatial heterogeneity
in the data values. Additionally, choropleth maps
depict abrupt changes at the boundaries of enu-
meration units, which are based on the existence of
artificially defined boundaries, and not boundar-
ies defined by the reality of the data. Dasymetric
maps can be subject to abrupt boundary changes
as well, but “these transitions are a better reflec-
tion of the true underlying geography of the area
than the transitions in choroplethic maps, which
are artifacts partially attributable to the arbitrary
delineation of areal boundaries. This limitation of
dasymetric mapping is offset by the technique’s
better visualization of population patterns, due
to the high degree of spatial disaggregation that
can be achieved” (Holt et al. 2004, p. 104).
Methods and Data Used in Dasymetric
Mapping
Transferring data from one set of geographic
zones or districts to another set of non-coinci-
dent zones is often necessary in spatial analysis.
For instance, we might have data on the number
of people living within a certain census tract
but need to estimate the number of people in
a smaller area within the tract, or an area that
includes only part of that tract and part of other
tracts. We may be interested in population or
other data at a watershed level and only have
population data available at the census enu-
meration units. “In any one study, several dif-
ferent types of data may be collected at differing
scales and resolutions, at different spatial loca-
tions, and in different dimensions” (Gotway and
Young 2002, p. 632).
A typical example of this is the problem encoun-
tered when conducting spatial analysis on histori-
cal census data from various time periods, with
each temporally different attribute data set using
different spatial data as well, because the tract
boundaries used to aggregate the attribute data may
change with each census period (Gregory 2000).
How can one determine the number of people
living in only a portion of an area for which data
have been aggregated, or in an area for which
the zones containing the data of interest do not
coincide among various data layers?
Several methods of disaggregating population data
are discussed below: weighted areal interpolation;
filtered weighted areal interpolation; the use of
land use/land cover as ancillary data for filtering;
three-class and limiting variable methods; “image
texture” method; statistical approaches, such as
regression-based methods; heuristic sampling; kernel
density surface using weighted census centroids;
the use of other types of ancillary data sets, such as
street-weighted interpolation; and the proposed
CEDS method.
Areal Interpolation
A common method for calculating disaggre-
gated population values is areal interpolation.
This is defined as “the transfer of data from one
set (source units) to a second set (target units)
of overlapping, non-hierarchical, areal units
(Langford et al. 1991, p. 56). Areal interpola-
tion is closely related to dasymetric mapping

80 Cartography and Geographic Information Science
of population densities (Holt et al. 2004). The
main difference between areal interpolation
and dasymetric mapping is that with the later
approach, the data are not re-aggregated into a
desired enumeration unit as they are with areal
interpolation (Eicher and Brewer 2001).
A simple method of areal interpolation is to
weight the variable’s values by a ratio derived from
the relative areal measurements of the two types
of zones (source and target) (Goodchild and Lam
1980). Areal weighting is based on the assumption
that population (or another variable) is distributed
homogeneously throughout the “source” zone (the
original unit of data aggregation). The amount of
population estimated to be in the intersecting zone
(or “target” zone) is assumed to be proportional
to the amount of area in the source zone versus
the target zone. The ratio of area of source zone
to target zone is then applied to population in
the source zone to yield the population total in
the target zone
In a study of areal interpolation for socioeco-
nomic data, Goodchild et al. (1993) looked at
a typical problem of spatial analysis using non-
coincident areal units, namely the 58 counties of
California (the source zones) and the state’s 12
major hydrological basins (the target zones). The
boundaries of the two sets of spatial units were, for
the most part, incompatible. The socioeconomic
data are available on the county level, but data
connected with water issues are collected based
on the hydrologic basin units that correspond to
major watershed boundaries. In order to conduct
a major economic impact study of water usage
and policy, variables such as employment, income,
and population had to be transferred from the
county spatial units to the hydrological regions.
Goodchild et al. (1993) used direct areal weighting
to accomplish this, assuming that densities in the
source zones (the counties) were uniform. When
later comparing the results of the areal weight-
ing method with other methods using statistical
approaches, they found that areal weighting had
a much higher mean percentage error than did
the other methods.
One of the major sources of error in the areal
weighting method is that, typically, population is
not distributed evenly throughout a geographic
unit. Many things may make this so: large areas of
the zone may be uninhabitable due to the existence
of parks, water bodies, and industrial area; or, the
zone may be comprised of very different housing
types—one part of the zone may have high-rise
housing projects, while another part contains lower-
density single family homes. Therefore, having a
better way of disaggregating the data—rather than
assuming homogeneity—should help give more
accurate population totals for the target areas.
Filtered Areal Weighting (Binary Method)
Many previous dasymetric mapping studies have
used areal weighting as a starting point, adding
the additional step of filtering the data using an
ancillary data set (Flowerdew and Green 1989;
Goodchild and Lam 1980; Holt et al. 2004). The
ancillary data very often consist of land-use or
land-cover data that indicate where the uninhab-
itable areas are, then exclude these areas and re-
distribute the population in the remaining areas.
The simplest of these methods has been termed
the “binary” method (Eicher and Brewer 2001),
which uses remotely sensed data or land-cover/
land-use polygon data as a filter or mask to elim-
inate the areas deemed to be uninhabitable. It
is considered binary because land is designated
as either inhabited or uninhabited. Examples of
uninhabited land are parks, water bodies, cem-
eteries, industrial parks, and so forth.
In a study of burglaries in central Massachusetts,
Poulsen and Kennedy (2004) wanted to show the
distribution of residential burglaries as a rate, with
the number of housing units as the denominator
and the number of residential burglaries as the
numerator. Using the housing unit data in the
census led to misleading rates, since the housing
units were not distributed evenly throughout the
census blocks. Using the residential land use
and zoning layer as a mask, the non-residential
areas were removed from the equation, and rates
could be based on the potential burglary targets
(the housing units) and the number of burglaries
in the municipality. “Land-use data are classified
by assigning a value of 1 to residential cells
and 0 to all other cells to create a mask. This
mask is used to isolate residential areas within
the municipality. The source and target zones
are multiplied by the mask to remove cells that
do not fall into residential areas (Poulsen and
Kennedy 2004, p. 255).
Although results of the filtered areal weighting
are generally an improvement over simple areal
weighting, there are still considerable deficiencies
in this method. For instance, all residential areas
do not have the same density of housing units or
population, but filtered areal weighting assumes
that all residential areas are homogeneous with
respect to density. Additionally, non-residentially
zoned areas often have population, too, which is
totally eliminated in the purely binary approach.
Some of the methods discussed below offer

Vol. 34, No. 2 81
further refinements to filtered areal weighting,
taking it from a binary model to a more nuanced
approach, which results in a more realistic
depiction of densities typically encountered in
the real world.
Land Use/Land Cover as Ancillary Data
Although land-use polygon (vector) data sets
have occasionally been used as the filtering
layer (e.g., Bielecka 2005; Poulsen and Kennedy
2004), most dasymetric studies have used satel-
lite data (raster data format) to determine the
locations of uninhabited areas, and/or to classify
inhabited areas by population density (Holloway
et al. 1999; Langford and Unwin 1994; Mennis
2003; Sleeter 2004). For instance, using Digital
Terrain Elevation Data (DTED) for global cover-
age of roads and slopes and USGS’ Global Land
Cover Characteristics derived from Advanced
Very High Resolution Radiometry (AVHRR)
satellite imagery as indicators of population
distribution, researchers have created a global
population database known as LandScan, with
a spatial resolution of under one kilometer.
“LandScan…is the finest global population data
(<1 km resolution) ever produced and is several
orders of magnitude more spatially refined than
some of the previously available global popu-
lation datasets,” (Bhaduri et al. 2002, p. 34).
While this broad-brush approach may be neces-
sary and sufficient when working with very small
scales, such as continents, countries, states, or
large regions, it is less satisfactory when working
with metropolitan areas, counties, local commu-
nities, or other relatively large-scale areas.
In highly urbanized areas, land-cover data
derived from satellites may not yield precise
enough results to get a true picture of population
density, due to limitations in available pixel
resolution and intra-pixel heterogeneity of urban
areas (Forster 1985). Additionally, using the
extent of impervious surfaces or other physical
or morphological variables as interpreted
from satellite data as a proxy for degree of
urbanization or urban development can result in
misclassification for population density classes.
For instance, industrial and commercial areas
usually have large extents of impervious surfaces
and thus are classified as highly urbanized. Using
this interpretation of satellite imagery, these
areas are counted as areas of high population
density, which they usually are not.
Where higher-resolution satellite data
are available, such as in the United States,
LandScan is being developed at a 90-meter
resolution (Bhaduri et al. 2002). But even at
this resolution, densely settled urban areas may
have too much within-pixel heterogeneity to pin-
point population distribution with the accuracy
necessary for fine-grained analysis. For instance,
90 meters is approximately the size of a New York
City street block (200 linear feet per block face),
and within one block there can be several very
different land uses, as well as various densities of
residential dwellings, from single-family homes
to multi-family apartment buildings, all of which
would have very different population densities
but would show as one value in the image.
Recent advances in remote sensing have yielded
very high spatial resolution images, with pixels
representing one meter or less on the ground.
But even with these data, difficulties of correctly
assigning population to land-use/land-cover
classes abound. Although one can interpret the
images as to the locations of impervious surfaces,
and perhaps even show building footprints,
the image does not easily reveal the height of
the buildings or how many dwelling units are
contained within each structure, or indeed,
whether the structure is residential, commercial,
or industrial. For the purposes of many health,
crime, or environmental analyses, this ambiguity
makes even high-resolution data insufficient.
In their excellent review of methods to estimate
population using GIS and remote sensing, Wu et
al. (2005) discuss the method of estimating the
population of an area by multiplying the total
number of dwelling units with the number of
persons normally living in a dwelling unit. They
say that the number of dwelling units in an area
may be estimated from high-resolution remote
sensing images. However, this would really
only work when the environment is comprised
of single family homes, each constituting
one dwelling unit; it is unlikely to be accurate
when dealing with mid- or high-rise residential
buildings having many dwelling units per floor,
which constitute the majority of residences in
many dense urban areas. Although with the
increased use of LIDAR it may be feasible to
estimate building height, it is still unlikely that
this will result in an accurate accounting of the
number of dwelling units per floor.
The main drawbacks of using remotely sensed
images for the purposes of constructing an
accurate population map are summed up by
Moon and Farmer (2001): “[s]atellite imagery is
also relatively expensive, requires significant data
storage, processing, and computational capacity,
and suffers from weaknesses in classification

82 Cartography and Geographic Information Science
routines used to separate residential areas from
nonresidential areas” (Moon and Farmer 2001,
p. 42).
Three-Class and Limiting Variable Methods
Further refinements of the binary method
include the three-class (or class percent) method
and the limiting variable method (Eicher
and Brewer 2001). In the three-class method,
percentages are applied to each of the three (or
more) major land-use categories for that area,
representing the percentage of population (or
another variable) that is likely to be contained
within that land use, per district. For instance, if
the three major land-use categories are said to
be urban, agricultural/woodland/exurban, and
forested, the percentages might be 70 percent,
20 percent, and 10 percent, respectively. In
this case, we would expect that within a given
geographical unit, such as a county, 70 percent
of the population would be allocated to the
polygons (or grid cells, if using raster data)
determined to be in the “urban” category, 20
percent of the population would be allocated to
the “agricultural” polygons, and so forth. These
percentage numbers will vary depending on the
location of the area of interest, and are subject
to perceived local conditions and arbitrariness of
the analyst.
The assigning of the percentages is fairly subjective
and not based on statistical evidence. Furthermore,
“a major weakness of the three-class method is that
it does not account for the area of each particular
land use within each county. A worst-case scenario
would result for a county that had only one or two
small urban polygons [or grid cells, if using raster-
based data]. These polygons would still receive
70 percent of the [county-wide] data… causing
the urban areas in that county to have unusually
high densities and the other land-use areas to
have lower densities” (Eicher and Brewer 2001, p.
130). The assumption behind this method is that
“each land-use class has a characteristic population
density. The problem with this approach is that
although the difference between land-use classes
is recognized, the differences within a land-use
class are ignored. Not all residential areas have
the same population density, as evidenced by the
contrast between detached housing and multiple-
unit housing” (Liu et al. 2006, p. 188).
The limiting variable method expands upon
the three-class method by setting threshold
density limits for population assigned to the
various categories of land-use polygons (or grid
cells). The data are distributed within each unit
by simple areal weighting, and, subsequently,
the limiting thresholds are applied. Upper limits
on densities are established using data from
geographic units having only one class of land
use within their boundaries. The data for all
such units are ranked, and a certain percentile is
selected to be the limiting threshold for that class.
If any land-use polygon exceeds the established
threshold for its class of land use, the excessive
population is “removed” and redistributed to the
other land-use polygons within that geographic
unit. A problem with this method (as well with
as the three-class method) is that although it is
assumed that there are significant differences
in density between classes of land use there are
likely also to be significant differences within
any given land-use class, and this method does
not address those intra-land-use class differences.
Additionally, if the sample size of the mono-class
geographic units used to determine the threshold
number is small, and “the within-class population
density variance is high, the low number of samples
may yield a non-representative estimate of [the
population density for that class], and hence an
inaccurate dasymetric map” (Mennis and Hultgren
2005, p. 7).
“Image Texture” Method
Another method using very high spatial reso-
lution satellite images, such as the Ikonos one
meter imagery, estimates population based on
image “texture.” Rather than using categorical
land use as a surrogate for population density,
as in the methods discussed above, this method
examines the correlation between census popu-
lation density and image texture. Spatial units
called “Homogeneous Urban Patches” (HUP)
are obtained by “using texture-based image
segmentation which maximizes between-patch
differences while minimizing within-patch dif-
ferences” (Liu et al. 2006, p. 188).
Unlike other spatial units used in population
estimation from remotely sensed images, such as
the kernel window or the pixel, HUPs consist of
realistic, irregularly shaped units, each having a
similar within-unit characterization. Spatial met-
rics are used to characterize the texture of each
HUP, taking into account such factors as variety
and abundance of patch types within each HUP;
the spatial arrangement, position, or orientation
of patches within each HUP; patch density, con-
nectivity, and contagion of patches; and degree
of fragmentation—similar to analyses conducted
in landscape ecology. Of the nine spatial metrics
examined in the study by Liu et al. (2006), only

Vol. 34, No. 2 83
three showed significant correlation: percentage
of built-up area, percentage of vegetation in the
area, and the patch density of the built-up area.
Although some correlation between image texture
and population density was demonstrated, it is
not high enough to provide reliable estimates of
population distribution. “This research shows that
remote sensing images can indeed help to estimate
population density; however, the correlation may
not be strong enough for empirical applications”
(Liu et al. 2006, p. 195).
Statistical Approaches—Regression-based Analyses
Other researchers have used more statistical
approaches to dasymetric mapping, such as
area-to-point spatial interpolation (Kyriakidis
2004), as well as regression-based techniques to
correlate population density classes with land-
use/land-cover data (Bielecka 2005; Flowerdew
et al. 1991; Flowerdew and Green 1992, 1994;
Goodchild et al. 1993; Langford et al. 1991).
Bielecka recently (2005) produced a dasymetric
map of northeastern Poland by using the “modi-
fying areal weighted method” which assumes
that the ratio between the population density
of two land-cover categories is the same for any
given commune (local administrative division).
The ancillary data was the CORINE land cover
database, a polygon (vector) spatial file derived
from visual interpretation of satellite images and
composed of 31 classes of land cover in Poland.
A regression model was used to find the relation-
ship between land-cover classes and population
density in each commune. This resulted in six
categories of population density –land use, and
coefficients were developed weighting popula-
tion to the land-cover category.1 Comparing
the modeled population and the population as
measured by statistical data shows that they are
roughly in agreement. However, the coefficients
appear to be too high for urban communes and
too low for rural areas. In addition to this poten-
tial accuracy issue, regression analyses usually
conducted in order to estimate zonal weights
are relatively complex compared to traditional
dasymetric methods.
Heuristic Sampling Method
In another recent study of a five-county area
in Pennsylvania, Mennis (2003) uses dasymet-
ric mapping, areal weighting, and empirical
sampling based on satellite data to generate a
surface-based representation of population dis-
tribution not reliant on pre-existing areal unit
aggregation. This heuristic sampling approach
addresses some of the shortcomings of the pre-
vious methods; it takes the three-class method
to a higher level of accuracy, while reducing
some of the inherent subjectivity of that method.
The population of each block group was distrib-
uted to each grid cell in the population surface
based on two factors: 1) the relative difference
in population densities among the three urban-
ization classes (low, high, and non-urban), and
2) the percentage of total area of each block
group occupied by each of the three urbaniza-
tion classes (Mennis 2003). Population density
values for each urbanization class were sampled
empirically, which mitigated the subjectivity of
the assignment of a percentage of population
to a given ancillary data class (land use or land
cover). Area-based weighting addressed the dif-
ferences in area among ancillary data classes
within a given areal unit. The sampling pro-
cess selected all block groups that were entirely
contained within each urbanization class, found
their total population and area, and calculated
their aggregated population density. This frac-
tion established for each class was further modi-
fied by the amount of area occupied by each class
per block group, using an area ratio derived by
areal weighting.
Even though Mennis’ study represents an improve-
ment over many previous dasymetric mapping
methods, the results for dense urban areas are not
sufficiently detailed for many urban planning and
analysis purposes. The results for the suburban and
rural areas are more detailed and accurate than
vector maps made from census block group data;
however, his results for the highly urbanized core
areas do not differ significantly from the vector
block group maps that he started with, because
“considerable intra-block group variation exists
in urbanization class and, thus, population den-
sity” (Mennis 2003). To sum up, intra-block group
variation is not revealed for the most urbanized
population density class by using this methodol-
ogy and data.
Kernel Density Surface from Population-weighted
Census Centroids
A number of related surface-generating meth-
ods and refinements of these methods are used
to model a population surface by kernel density
estimation (Bracken and Martin 1989; Martin
et al. 2000; Martin 2006). Rather than mapping
1 The coefficients are computed in an iterative way.

84 Cartography and Geographic Information Science
population by zones, such as a census enumera-
tion district (or census tract in the U.S.), this
method entails creating a surface of population
density by interpolating population based on the
population-weighted enumeration district (ED)
centroid, as are available in the United Kingdom
census data. A kernel window is moved over the
cells containing these centroids, and the popula-
tion count at each centroid is distributed to the
cells in the kernel window by a distance-decay
model. An early iteration of this method had the
drawback that the surface population counts did
not always add up to the “correct” counts in the
census zonal data (Bracken and Martin 1989).
This was later revised so that the pycnophylac-
tic (volume-preserving) quality was maintained,
and the kernel window distributed the counts
from each centroid only to cells within the cen-
troid’s original zonal boundaries (Martin 2006).
Cell sizes ranging from 50 meters to 200 meters
were used, although the 200-meter cell was
deemed too crude a resolution for most urban
EDs. Even the 50-meter cell is often larger than
the smallest of the EDs and, therefore, does not
adequately represent the nuances of popula-
tion distribution within the densest parts of the
city, typically composed primarily of small EDs.
However, this density surface model reproduces
the essential form of population distribution
in the city and indicates major non-populated
open spaces such as cemeteries, industrial areas,
and commercial districts.
Use of Other Types of Ancillary Data—
Street-weighted Interpolation
Analyses of urban issues often require very
detailed data on population location in order
to yield meaningful results. Previous methods
relied primarily on land-use/land-cover data,
which has not proved to be as fine-grained as
necessary for many urban analysis purposes. In
another recent study, Reibel and Bufalino (2005)
used the street network data (TIGER files from
the U.S. Census) to derive weights for interpo-
lation for population and housing unit counts
for incompatible zone systems in Los Angeles
County, California. Building on previous simi-
lar work by Ong and Houston (2003) and Xie
(1996), they used the street and road grid as a
proxy for approximate population and hous-
ing unit density surfaces for census tracts in the
county. They then conducted an error analysis,
comparing the results of the street-weighting
method with traditional areal weighting, and
found that the street-weighting method offers
a 70 percent improvement over areal-weight-
ing for the estimation of the housing unit count
variable and a 20 percent improvement for the
estimation of the total population count variable.
They note, however, that “the street-weighting
method appears to reduce errors most com-
pared with the area-weighting method in those
areas where the lack of population is reflected in
the lack of roads and least in those areas (such as
industrial areas) with a more developed, but non-
residential transportation infrastructure” (Reibel
and Bufalino 2005, p. 136).
By using only vector data sets, this method
obviates the need for the analyst to be familiar
with processing and interpreting remotely sensed
images and integrating raster and vector datasets,
making dasymetric mapping more accessible to
demographers, urban planners, environmental
managers, and others for whom vector data sets are
typically a more familiar part of their GIS experi-
ence than satellite data. This is a real advantage
to this method, since heretofore areal weighting
methodology was the most frequently used method
for those using only vector data sets, and the street-
weighting method is a clear improvement over that
approach. However, the street-weighting method
appears least reliable in the very areas of most
concern to urban analysts—the densely settled,
heterogeneous urban core areas.
With the Vienna, Austria, metropolitan area as
a study area, Weichselbaum et al. (2005) also used
street infrastructure as ancillary data, along with
very high resolution Earth Observation satellite
imagery, to create a “building block” spatial data
set. “The road network data are used for building
the geometric framework for population allocation
by converting the linear road segments to polygons
that make up the building blocks. Such building
blocks constitute meaningful reference units for
the disaggregated population, as they reflect the
real-world housing patterns and population dis-
tributions” (Weichselbaum et al. 2005).
The population density classes for these build-
ing block polygons were derived from satellite
images and sampling of representative densities
in homogeneous census tracts, similar to Mennis’
(2003) empirical sampling technique, but using five
land-use/land-cover categories and four residential
population density classes. When comparing the
modeled disaggregated population to the original
census grid data, the average error found was
12.44 percent. This compares very favorably to
the relative error of 102 percent obtained without
using the ancillary land-use data. However, even an
error margin of 12 percent can obfuscate analyti-

Vol. 34, No. 2 85
cal findings when dealing with very fine-grained
and heterogeneous urban areas.
Cadastral-based Expert Dasymetric System (CEDS)
Our method of using cadastral data as the
ancillary data appears to be an innovative and
progressive approach to dasymetric mapping.
Cadastral data are used in recording prop-
erty boundaries, property ownership, property
valuation, and, of course, for the all-important
purpose of property tax collection. The type of
cadastral-level data used in our CEDS method
is commonly available for most urbanized areas
in the United States, western Europe, and other
more developed areas. The data are usually
organized by township, municipality, or county,
and, less often, by metropolitan region.
However, in many parts of the world, census and
cadastral data may not be readily available, current,
or accurate. Baudot (2001) makes the point that
for urban areas in less-developed countries, very
often there are no census property tax records or
city planning data on population to work with, and
even when such data are collected, the exponential
growth rates of these cities makes the census data
obsolete almost immediately. This is why satellite
data are most often used for dasymetric map-
ping—they are available for almost all parts of
the world and are very current. However, “urban
environments are often considered too complex
to be analyzed by satellite remote sensing, and,
indeed, the spatial resolution of current satellite
sensors means that they are not particularly well
suited to the task” (Baudot 2001, p. 266).
In urban areas where census and cadastral
data are available, the CEDS method will
be an improvement. For instance, munici-
palities where property tax records are
linked to a digital spatial database (e.g.,
most cities and larger towns in the United
States), the cadastral data required by the
CEDS method will be available. Although
these data may not be available to the
general public for free, they still tend to
be less expensive for the end user than
high-resolution remotely sensed images
for the equivalent spatial extent.
Comparison of Areal Weighting,
Filtered Areal Weighting, and
CEDS Dasymetric Mapping: A
Hypothetical Example
The following diagrams illustrate how
the CEDS method of dasymetric map-
ping can be beneficial to health, environmental,
crime, risk assessment, and urban planning anal-
yses. The diagrams in Figure 1 contrast standard
areal weighting interpolation and filtered areal
weighting dasymetric (binary) techniques with
the cadastral expert dasymetric system (CEDS)
method.
The CEDS method differs from other forms
of dasymetric mapping because it does not use
areal weighting or the binary (filtered areal weight-
ing or “punch-out”) method alone. The ancillary
data used are not remotely sensed land cover/land
use, interpreted to estimated population density
classes, but rather very detailed cadastral data, more
appropriate to estimating population distribution
in hyper-heterogeneous urban areas. The CEDS
method also uses an expert system to determine
which variable in the cadastral dataset to use as
the ancillary data, calculating which ancillary data
fits the data best. In this way, each source record
within the area of interest can be customized as
to the method of disaggregation, which, when
validated, yield results that best fit the data.
Using the CEDS method, the modeled population
data always preserve the pycnophylactic property,
meaning that the estimated (modeled) value of
the tract, when re-aggregated, equals the original
value of the tract (Tobler 1979). Preservation of the
pycnophylactic property is not always achievable
with previously used dasymetric methods based
on population density classes derived from land-
use/land-cover data.
Figure 1. Methodological differences and potential improvement of
population estimation of the CEDS method (c), over both filtered areal
weighting (b), and simple areal weighting (a).

86 Cartography and Geographic Information Science
CEDS Methodology and Analysis
The strength of the Cadastral-based Expert
Dasymetric System (CEDS) methodology is
the ability to disaggregate data at high spatial
resolution in a densely populated urban envi-
ronment such as New York City (NYC). Much
of our previous work has involved analyses of
hyper-heterogeneous large-scale geographies
for health, environmental, and crime studies
in the New York City metropolitan area (Clarke
and Maantay 2006; Herrmann and Maroko
2006; Maantay 2001, 2002, 2005); Maantay
and Strelnick 2003; Maroko and Maantay, 2006,
unpublished study). We have found that these
sorts of analyses could be greatly improved by
the use of more fine-grained spatial data in
place of a coarse aggregation of available popu-
lation data.
This CEDS method was designed to disaggregate
the total population counts from the census block
group level (5,733 in NYC) to the tax lot level
(847,153 in NYC) using cadastral data. Census
block groups, rather than the smaller census blocks,
were used in order to avoid data suppression of
subpopulations in the latter. The accuracy of these
data will ultimately become important when the
CEDS method is applied to subpopulations (e.g.,
Hispanic, non-Hispanic white, non-Hispanic black,
Asian, etc.). If the selected subpopulations do not
follow the same trends as does the total population
and benefit from individualized expert systems, it
will be necessary to use the coarser census block
group aggregation. However, if the subpopulations
do mirror the behavior of the total population,
the disaggregation will be performed on census
block data to further constrain the calculations
and improve accuracy.
The technique uses residential area (RA) and
number of residential units (RU) as proxies for
population distribution. In other words, it is
assumed that where there are more potential living
accommodations there will be higher populations.
As such, the population in each block group was
disaggregated (or redistributed) among the tax
lots based on either RA or RU. The proxy unit
(RA or RU) used in the disaggregation was indi-
vidually determined by an expert system for each
geographic unit. The results were then validated
against census data and compared to commonly
used dasymetric techniques to assess predictive
accuracy and possible improvement over other
methods.
The CEDS disaggregation of census populations
can be compartmentalized into three fundamental
steps: 1) data preparation, 2) dasymetric calcula-
tions, and 3) expert system implementation. The
discussion of these steps is followed by an evalu-
ation of the results.
Data Preparation
Two datasets were used for this process: the
2000 census data (United States Bureau of the
Census 2001a, 2001b) and LotInfo (LotInfo,
LLC 2001). Decennial census data for New York
City was downloaded via www.census.gov. The
total population data (SF1, table P001), which
are aggregated in a hierarchical fashion (see
Table 1), were obtained at the census tract and
census block group levels, with each tract con-
taining multiple block groups. LotInfo, a prod-
uct of LotInfo, LLC, which combines spatial
data from the New York City Department of City
Planning (DCP) and attribute information from
the Real Property Attribute Data (RPAD) data-
base provided by the New York City Department
of Finance (DOF), contains exhaustive data at
the tax lot level in New York City (e.g., zoning,
ownership, building attributes, residential area,
and residential units). Although this study was
done in New York City, similar data are often
available from planning departments of metro-
politan areas or urbanized counties.
The goal of the data preparation was to minimize
error-inducing anomalies and discrepancies. The
data were refined by editing geographic identi-
fiers and deleting fields that were not needed
so that they would be readily comparable and
more efficiently manipulated. Included in this
refinement was the exclusion of the Riker’s Island
census tract and complimentary block groups and
tax lots. Riker’s Island, with one of the highest
populations of any tract in the city, is a prison,
and as such there is inconsistency in the way this
area is handled between the census and the lot
data. For this reason, results would have been
drastically skewed with its inclusion.
Table 1. Comparison of spatial unit counts in New York
City.
Areal Units in New York City
Unit Number Average Number
of Units per Tract
Census Tracts 2,217 1.00
Census Block Groups 5,733 2.59
Tax Lots 847,153 382.12

Vol. 34, No. 2 87
was aggregated up to the block group and tract
levels (see Figure 2). This table was then used
to generate a tax lot-level spatial data layer
with RU and ARA values aggregated at the tax
lot, block group, and tract levels, as well as the
census population data at the block group and
tract levels. It is important to note that the data
which are aggregated to larger areal units are
identical for each lot within any given areal unit.
In other words, if tax lots “1”, “2”, and “3” are
all within tract “A,” they will all share the same
tract-level information.
Several dasymetrically derived populations
were then calculated. The general equation is
solved by multiplying the census population with
the ratio of population proxy units thus:
POPl = POPc * Ul / Uc (2)
where:
POPl = dasymetrically derived lot-level
population;
POPc = census population (block group or
tract);
Ul = the number of proxy units at the tax lot
level (RU or ARA); and
Uc = the number of proxy units at the
census level (RU or ARA per block
group or tract).
Values were calculated from the block group
and tract census populations using both RU and
ARA as the proxy units. The process resulted in
four dasymetrically derived population values for
each tax lot (tract ARA, tract RU, block group ARA,
and block group RU).
Expert System Implementation
This expert system was designed to determine
which proxy unit—number of residential units
(RU) or adjusted residential area (ARA)—more
As was already mentioned, residential area (RA)
and number of residential units (RU) are important
attributes in the CEDS process. Within the lot-level
data, the RU variable did not require additional
processing; however, there were many instances
of missing data values for the RA variable in the
original RPAD data from the Department of Finance.
As such, a new variable, adjusted residential area
(ARA), was created. The adjusted variable is identi-
cal to RA in many cases, however when the value
for RA is zero and the number of residential units
(RU) does not equal zero (i.e., there are residential
units but no value for residential area), ARA is
defined as the total building area multiplied by
the ratio of the number of residential units and
the total number of units. The ARA variable can
be written as follows:
ARA = M * (BA * RU / TU) + RA
IF RA = 0 AND RU <> 0, THEN M = 1,
ELSE M=0 (1)
where:
ARA = adjusted residential area within tax lot;
BA = building area (residential and commer-
cial) within tax lot;
RU = number of residential units within tax lot;
TU = total number of units (residential and
commercial) within tax lot;
RA = Residential area; and
M = Binary variable designating ancillary data
for ARA.
Dasymetric Calculations
Using the GIS capabilities of ARCGIS 9.1 (ESRI
2005) and the LotInfo data set, the total amounts
of RU and ARA were calculated for each census
tract and census block group in New York City
and saved in tabular form. In other words, the
RU and ARA information, at the tax lot-level,
Figure 2. Aggregation of cadastral data (adjusted residential area (ARA) in this case) to the census block group and
census tract levels. The results are: block group A has 2,000 ft2, block group B has 15,000 ft2, and the entire tract has
17,000 ft2 of adjusted residential area.

88 Cartography and Geographic Information Science
accurately predicts the population distribution on
a tract-by-tract basis. This was accomplished by
re-aggregating the tax lot level population figures
that were derived from the census tract data back
to the block group level; the result was an estimated
block group population. In other words, tract data
were disaggregated down to the tax lot and then
re-aggregated up to the block group. It was neces-
sary to use the tract-level data as a starting point
so that there would be a smaller unit of aggrega-
tion (block group) available in the census data with
which to compare the estimated values. Although
the census data are available by census block, a unit
smaller than the block group, much of the data
is suppressed due to small numbers and privacy
issues, particularly when dealing with sub-popula-
tions. The absolute value of the difference between
census populations and estimated populations can
be written as follows:
POPdiff = | POPBG – POPest | (3)
where:
POPdiff = the difference between census and
estimated populations per block group;
POPBG = census block group population; and
POPest = estimated population (RU or ARA)
derived from the census tract (not
block group).
By comparing the estimated population to the
census population for both the RU- and ARA-based
techniques, it can be assumed that the process which
resulted in estimates more similar to the census
block group values (i.e., smaller POPdiff values) more
accurately redistributed the data. After re-joining
the POPdiff data with the LotInfo data, the expert
system would then select the superior proxy unit as
the disaggregation technique for each block group.
This can be described as follows:
IF RU_POPdiff <= ARA_POPdiff, THEN POPl =
POPRU_BG, ELSE POPl = POPARA_ BG (4)
where:
RU_POPdiff = the absolute difference between
the census block group population and the
estimated block group population derived
from the census tract population based upon
number of residential units;
ARA_POPdiff = the absolute difference between
the census block group population and the
estimated block group population derived
from the census tract population based upon
residential area;
POPl = the final estimated tax lot population
dasymetrically derived from the census block
group population (not the census tract);
POPRU_BG = the estimated tax lot population
dasymetrically derived from the census block
group population (not the census tract)
based on number of residential units; and
POPARA_BG = the estimated tax lot population
dasymetrically derived from the census block
group population (not the census tract)
based on the adjusted residential area.
In essence, it is the performance of the tract-level
disaggregation that defines the proxy units used
for each block group disaggregation, ultimately
resulting in a final dasymetrically derived value
individually tailored for each block group.
Evaluation and Discussion of
Results for CEDS
The evaluation of the results focuses on compar-
isons among census data, a commonly used dis-
aggregation technique referred to as the ‘filtered
areal weighting’ method in this paper, and dasy-
metrically derived data based on adjusted resi-
dential area (ARA), number of residential units
(RU), and the Cadastral-based Expert System
(CEDS). As was noted in the expert system
implementation section above, it is not possible
to empirically verify the derived tax lot popula-
tion numbers since the U.S. Census Bureau data
does not provide information at such a fine reso-
lution. For this reason, statistics were run on the
block group level re-aggregations of the derived
tax lot populations calculated from the tract
level census data (i.e., tract population disaggre-
gated down to the tax lot and re-aggregated up
to block group).
Modified Expert System
The expert system used in the evaluation of the
results (Equation 5), as opposed to the original
expert system used in CEDS (Equation 4), can
be described as:
IF RU_POPdiff <= ARA_POPdiff, THEN POPlot =
POPRU_TR, ELSE POPlot = POPARA_TR (5)
where:
RU_POPdiff = the absolute difference between
the census block group population and the
estimated block group population derived
from the census tract population based upon
number of residential units;
ARA_POPdiff = the absolute difference between
the census block group population and the
estimated block group population derived

Vol. 34, No. 2 89
from the census tract population based upon
residential area;
POPlot = the final estimated tax lot population
dasymetrically derived from the census tract
population (not the census block group)
based on the best performing proxy unit;
POPRU_TR = the estimated tax lot population
dasymetrically derived from the census tract
population (not the census block group)
based on number of residential units; and
POPARA_TR = the estimated tax lot population
dasymetrically derived from the census tract
population (not the census block group)
based on the adjusted residential area.
As can be seen, this is very similar to the origi-
nal CEDS, except that in this process the tax lot
population derived from the census block group
data was not utilized at all. The expert system,
in terms of this evaluation, is based solely on the
tract-based populations. In other words, the final
dasymetric results are not being tested here. The
final tax lot level populations are based on block
group data, which would prove tautological if re-
aggregated and compared with census block group
population. It is for this reason that the CEDS was
modified for this analysis to accommodate only
the census tract-derived tax lot populations in
order to avoid artificially inflated results.
Comparison with Filtered Areal
Weighting
The filtered areal weighting (binary) method
was used in order to compare the accuracy of
CEDS against a commonly used disaggregation
technique, essentially acting as a control vari-
able. The filtered areal weighting methodology
is comparatively simple, using a combination of
“cookie cutter” overlay and areal weighting pro-
cesses.
Census tract, census block group, TIGER land-
mark, and TIGER water body geographic files were
downloaded from the U.S. Census Bureau’s web
site. The landmark and water body data layers
were then combined and processed to make an
“open spaces” layer where there is known to be no
residential population (e.g., parks, airports, cem-
eteries, water bodies, golf courses, and national
recreation areas). The open spaces layer acted
as a “cookie cutter” on the tract and block group
boundaries, resulting in the tracts and block groups
being geographically modified to exclude the open
space regions. Note that the data as provided by
the Census Bureau is somewhat coarse. The results
from the filtered areal weighting may be improved
if the open spaces layer was created using a more
comprehensive data set at a finer resolution.
Area of the census polygons (as calculated within
ArcGIS 9.1) and total population (from census SF1,
table P001) attribute data were added to the tract
and block group boundary layers. Areal weighting
was then utilized to complete the filtered areal
weighting process by equating the estimated block
group population to the census tract population
multiplied by the ratio of block group area and
tract area, as modified by the binary filtering. It
is important to note that this weighting technique
makes the assumption that the population is uni-
formly distributed within each census tract, rather
than using additional ancillary data to redistribute
the population in a heterogeneous manner. It can
be written as follows:
POPFAW = POPTR * AREABG / AREATR (6)
where:
POPFAW = estimated block group population
from filtered areal weighting;
POPTR = census tract population;
AREABG = modified census block group area
(open spaces excluded); and
AREATR = modified census tract area (open
spaces excluded).
Comparison of CEDS, Filtered Areal
Weighting, and Dasymetrically Derived
Populations
In order to assess the accuracy and validity of the
dasymetrically derived populations (as obtained
by filtered areal weighting, ARA alone, RU alone,
and CEDS), the results were compared to census
block group populations. This can be done
very simply by comparing the estimated block
group populations to the census block group
populations. The absolute values for the differ-
ence between each block group population were
summed, divided by the entire population in
New York City, and converted to a percentage
(see Figure 3). This very simple analysis suggests
that CEDS, with only 6.37 percent difference,
outperformed RU (8.69 percent), ARA (9.44
percent), and filtered areal weighting (21.91
percent).
For a more comprehensive analysis, linear
regressions similar to Qiang Cai’s approach in
“Age-sex population estimation for small areas”
(Cai et al. 2006) were performed, except that
all block groups were used rather than selected
block group pairs. The estimated block group
populations from the four disaggregation

90 Cartography and Geographic Information Science
methods were regressed against the block
group population data from the U. S.
Census Bureau to evaluate their relative
effectiveness in New York City as a whole
and separated by borough. This analysis
involved linear regression, with the
regression line forced through the origin.
The R2, standard errors, and regression
coefficients were then compared and are
summarized in Figure 4.
As expected, the regression coeffi-
cients for all of the methodologies were
approximately ‘1’, with the CEDS method
producing the closest value (.996) and
the filtered areal weighting producing
the most dissimilar (.978). An examina-
tion of the differences in R
2
values shows
that the expert system produced more
highly correlated results (R2 = .991) than did the
ARA (.983), RU (.986), or filtered areal weighting
(.924). The standard errors also imply that the
CEDS methodology (std. error = 164) outper-
formed the other three (std. error = 481, 339,
and 210 for filtered areal weighting, ARA, and
RU, respectively). That CEDS produced better
results than ARA or RU is not unexpected since
Figure 3. Percent absolute difference between census block group
population and estimated block group populations in New York City for
the different methods.
Figure 4. Simple linear regressions for NYC showing R2, standard errors, and regression coefficients of block group populations
estimated by filtered areal weighting (a), ARA (b), RU (c), and CEDS (d) versus census block group populations.

Vol. 34, No. 2 91
CEDS selects the better performing proxy unit on
a tract-by-tract basis. What is more substantive is
the contrast between the filtered areal weighting
method (serving more or less as a control) and
the expert dasymetric system. This is seen most
intuitively by examining the wider spread of data
points in the filtered areal weighting scatterplot
(Figure 4(a)) as compared to the CEDS scatterplot
(Figure 4(d)). When regression analyses were per-
formed on a borough-by-borough basis, the results
were similar, although some spatial variation can
be seen (see Figures 5 and 6).
Figure 5. Linear regression R2 of block group populations estimated by each of the four disaggregation methods versus
census block group populations.
Figure 6. Standard errors for linear regressions of block group populations estimated by each of the four disaggregation
methods versus census block group populations.

92 Cartography and Geographic Information Science
Even though filtered areal weighting resulted
in acceptable R2, standard error, and parameter
estimates for these densely settled urban areas, the
dasymetric technique used in this study is clearly
superior. It is also important to note that what is
being compared in this section of the analysis is
Figure 7. Visual comparison of CEDS-derived population, CEDS-derived population density by tax lot, and traditional
choroplethic population density by census block group.

Vol. 34, No. 2 93
not the end-product of the dasymetric process,
rather a validation of its efficacy at a compara-
tively coarse spatial aggregation. The result of
the CEDS methodology is tax lot-level rather than
block group-level population data, an areal unit
that has approximately 150-times finer resolution.
See Figure 7 for a comparison of CEDS-derived
population, CEDS-derived population density by
tax lot, and traditional choroplethic population
density by census block group.
Asthma and Air Pollution Case
Study Using CEDS
The asthma and air pollution case study is dis-
cussed here in order to illustrate, on a concrete
example, the value of the CEDS method for a
particular type of analysis, in this case, an envi-
ronmental health justice study. A consortium of
Bronx-based researchers has been investigating
the association between asthma hospitalizations
and outdoor air pollution in the Bronx, one of
the five boroughs of New York City (Maantay
and Strelnick 2003). The purpose of that initial
study was to determine if there is a spatial cor-
respondence between the locations of land uses
that contribute to poor air quality and the loca-
tions of people who have been hospitalized for
asthma in the Bronx. Asthma is extremely prev-
alent in the Bronx, affecting people of all ages
and diminishing their quality of life. In some
cases, asthma can cause death; the asthma death
rate in the Bronx (6 per 100,000) is double that
of New York City. Children in the Bronx are
especially affected by asthma, and the asthma
hospitalization rate for children is 70 percent
higher in the Bronx than in New York City as
a whole, and 700 percent higher in the Bronx
than for the rest of New York State (excluding
New York City).2
The most reliable and complete data for
asthma currently available for New York City is
the hospitalization database created by the New
York State Department of Health. Although this
data set only includes asthma hospitalizations,
and not all cases of asthma prevalence, it does
include the most severe and dire cases. Because
we were able to obtain patient record level data,
we were able to geo-code addresses as points of
latitude and longitude, thus permitting the kind
of fine-grained spatial analysis that would not
have been possible with aggregated health data,
the most commonly health data available due to
issues of patient confidentiality.
Air quality in the Bronx is adversely impacted
by the concentration of Toxic Release Inventory
(TRI) facilities and other major stationary
point sources (SPS) of air pollution, limited
access highways (LAH), and major truck routes
(MTR). The locations of these four categories
of environmentally burdensome land uses were
plotted and then buffered at distances reflecting
standard guidelines for fate and transport of
airborne pollutants. We analyzed each buffer
type in relation to the home addresses of
persons hospitalized for asthma, using five years
of hospitalization data (Figure 8). See Maantay
(2007) for a complete description of methodology
of the initial project.
The analysis found that people living within the
buffers were much more likely to be hospitalized
for asthma than those living outside the buffers
(up to 60 percent more likely), and the correlation
between asthma hospitalization rates and proximity
to major air pollution sources remains significant
even when controlling for race, ethnicity, and pov-
erty status. However, the risks vary depending on
the pollution source type (Maantay 2007). Living
within TRI and major stationary point source buf-
fers poses a much higher risk than living within
the limited access highway and major truck route
buffers, according to the proximity and odds ratio
analyses. People within the highway and truck
route buffers for the most part do not appear to
have an increased risk of asthma hospitalization,
based on the results of the initial study.
These unexpectedly neutral findings for the
truck routes and highways might be due to an
artifact of how the population numbers within
the buffers were calculated. The areal weighting
algorithm used to estimate population within the
buffered areas assumed population is spread evenly
throughout the census block group. However, these
highway buffer areas may, in fact, be less densely
populated than the remainder of the block group,
for various reasons including building clearances
and urban renewal at the time the highways were
constructed. If the population near the highways
is actually less than that estimated by the areal
weighting script, then the denominator used to
calculate rates would be too high, making the
asthma hospitalization rates lower than they actu-
ally are within these buffers.
2 Findings reported by the New York City Department of Health in Asthma Facts, a report based on 2000 data collected
by the state (New York City Department of Health 2003).

94 Cartography and Geographic Information Science
Figure 8. Asthma hospitalization proximity analysis using distance buffers as proxies for pollution exposure.

Vol. 34, No. 2 95
One way to test this possible explanation would
be to utilize finer-resolution population data as
the denominator when calculating asthma hos-
pitalization rates.
CEDS Asthma Study Methodology
In order to assess potential improvement in
rate calculations, we decided to re-examine the
association between asthma rates in the Bronx
and proximity to limited access highways (LAH)
using the CEDS methodology. The LAH option
was re-examined due to its questionable results
in the initial study. Hospitalizations for asthma
(by home address), LAH locations, census popu-
lations, and an “open spaces” layer constitute
the data used in this analysis.
Roadways were downloaded from
the U.S. Census Bureau’s website (U.S.
Census Bureau 2001(b), and those that
fit the criteria of limited access highways
were selected. The LAH layer was then
buffered by 150 meters, as in the initial
study. This newly created buffer layer
(LAH buffer) served to define proximity
to the LAH (Figure 9).
Five years of asthma hospitalization
data were provided by the Statewide
Planning and Research Cooperative
System (New York State Department
of Health 2003), together with each
patient’s home address in latitude and
longitude. The hospitalizations were
geo-coded (plotted) and then overlaid
with the LAH buffer. The cases which
fell inside the buffer were separated
from those that were beyond the 150
meter buffer. The results served as the
numerators for rate calculations.
The denominator of the rate
equation is the susceptible population.
To determine the rate outside the
buffer, the cases beyond 150 m of
the LAH must be divided by the
corresponding population. Naturally,
the same is true for the rate inside the
buffer. Census block group information
(SF1, table P001) was attached to
the census block group boundary
file. Complications arose due to the
fact that the LAH buffer does not
coincide with the census block group boundaries,
therefore the population information had to be
disaggregated in order to determine the rates.
This disaggregation was done following two
methodologies—filtered areal weighting (as in
the initial study) and Cadastral-based Expert
Dasymetric System (CEDS).
Filtered areal weighting, as was described above,
uses areas with no likely residential population
(open spaces) as a “cookie cutter” to remove unin-
habited areas from the areal weighting calcula-
tions. TIGER landmark and TIGER water body
geographic files were combined and processed to
make the open spaces (uninhabited) layer which
included such land uses as parks, airports, cem-
eteries, water bodies, golf courses, and national
recreation areas. These areas were removed from
the census block group layer to obtain geographi-
cally modified block groups which exclude the
aforementioned open space regions.
Areal weighting was then utilized to complete
the filtered areal weighting process. This involved
calculating the estimated inside/outside buffer
block group population obtained by multiplying
Figure 9. A buffer of 150 meters around limited-access highways in
the Bronx.

96 Cartography and Geographic Information Science
the original census block group population by the
ratio of the buffer area (inside or outside) and fil-
tered block group area (excluding open spaces). It
is important to note that this weighting technique
makes the assumption that the population is uniformly
distributed within each filtered census block group.
The process can be written as follows:
POP_INFAW = POPBG * AREAIB / AREABG
POP_OUTFAW = POPBG * AREAOB / AREABG (7)
where:
POP_INFAW = estimated population inside the
LAH buffer from filtered areal
weighting;
POP_OUTFAW = estimated population outside
the LAH buffer from filtered
areal weighting;
POPBG = census block group population;
AREAIB = filtered census block group
area inside the LAH buffer
(open spaces excluded);
AREAOB = filtered census block group
area outside the LAH buffer
(open spaces excluded); and
AREABG = filtered census block group area
(open spaces excluded).
The CEDS methodology, as
described in detail elsewhere in
this paper, was used to disag-
gregate the census block group
population to the tax-lot level.
A combination of number of
residential units and residential
area was used as ancillary data to
redistribute the population. Pop-
ulations inside and outside the
buffer were calculated by select-
ing the tax lots whose centroids
fall inside and outside the 150
m LAH buffers and summing
the associated populations. With
this method, population is not
assumed to be homogeneous;
instead, cadastral information
regarding residential dwellings
is used to redistribute the data
preferentially. The equations can
be written as follows:
POP_INCEDS = Σ LOTPOPi * M
POP_OUTCEDS = Σ LOTPOPi * |M-1| (8)
where:
POP_INCEDS = estimated population inside the
LAH buffer from CEDS;
POP_OUTCEDS = estimated population outside
the LAH buffer from CEDS;
LOTPOPi = CEDS-derived tax lot
population for tax lot i; and
M = 1 if LOTi has its centroid within
the LAH buffer, else it has a
value of 0.
CEDS Asthma Study Results and
Discussion
Since it can be difficult to visualize the difference
in methodologies on small-scale maps (i.e., the
entire Bronx), three block groups were selected
in the South Bronx to illustrate the methods and
respective results more explicitly. Filtered areal
weighting and CEDS techniques were performed
on the selected block groups containing a popu-
lation of 2,166. Two sets of data were obtained
for this area, which show a dramatic distinction
between the methods even though they may
not necessarily be representative of the entire
Bronx. The filtered areal weighting resulted in
an estimation of 1,017 people (~47 percent) of
the three selected block groups residing within
150 meters of an LAH and 1,149 people (~53
percent) lived outside of the buffer. The CEDS
method estimated that only 57 people (~3 per-
cent) were within the LAH buffer and 2,109
(~97 percent) were outside of the 150 meter
threshold (see Figures 10 and 11). This some-
what stark example demonstrates the utility of
the CEDS methodology and the usefulness of
Figure 10. Population estimation inside and outside buffers, using filtered areal
weighting versus CEDS Method, for three selected block groups, as shown in
Figure 11.

Vol. 34, No. 2 97
having a more realistic understanding of popu-
lation distribution.
The phenomenon that is apparent in the three-
block group example above can also be seen in the
entire Bronx. Between 1995 and 1999 (the study
time frame) there was an average of 8,623 asthma
hospitalizations in the Bronx per year. Inside the
Bronx-wide LAH buffer there were 950 hospital-
izations per year, and outside the buffer there were
7,672 hospitalizations per year. Rates were calcu-
lated by dividing the number of asthma hospitaliza-
tions by the susceptible population, as follows:
RATEBX = CASESBX / POPBX
Figure 11. Visual representation of the comparison of filtered areal weighting and CEDS method.

98 Cartography and Geographic Information Science
RATEIN = CASESIN / POPIN
RATEOUT = CASESOUT / POPOUT (9)
where:
RATE = rate of asthma hospitalization per year;
CASES = five-year average asthma
hospitalization;
POP = population;
BX = Bronx;
IN = inside the 150 m LAH buffer; and
OUT = outside the 150 m LAH buffer.
Using Equation (9), the asthma hospitalization
rate for the entire Bronx was approximately 6.53
per 1,000 people per year. The results with the
filtered areal weighting methodology were some-
what counterintuitive. Inside the buffer, the rate of
susceptible populations was 6.20 per 1,000 people
per year, whereas outside the buffer, the rate was
6.58 per 1,000 people per year, i.e., greater than
the inside rate. If exposure to certain outdoor air
pollution increases asthma hospitalizations, and
close proximity to limited access highways increases
exposure to this pollution, one would assume that
living in close proximity to an LAH would show
an elevated rate of asthma hospitalizations when
compared with those living beyond this threshold.
The filtered areal weighting methodology, however,
claims that the opposite is true. Although there are
many other sources of outdoor air pollution in the
Bronx, and also many other variables which may
increase asthma hospitalization rates, the results
were nonetheless unexpected.
When using the CEDS method, however, the
results were quite different. Rates inside the buffer
were found to be 7.03 per 1,000 people per year
and outside the buffer 6.49 per 1,000 people per
year (see Figure 12a). An examination of the stan-
dardized incidence ratios (SR) by filtered areal
weighting shows that residing within the LAH
buffer is protective, with a 5 percent lower chance
of being hospitalized for asthma. However when
the CEDS-derived data are used as the denomi-
nator, the SR shows a 7.5 percent higher chance
for being hospitalized for asthma when residing
within the LAH buffer (see Figure 12b). These
CEDS-derived results, based on the more precise
Figure 12a. Estimated asthma hospitalization rates inside and outside buffers for the entire Bronx, with filtered areal
weighting versus CEDS method.

Vol. 34, No. 2 99
location information of the population, seem to
bolster the hypothesis that exposure to pollutants
released from the limited access highways in the
Bronx elevate asthma hospitalization rates.
The reason for this inconsistency in asthma
rates is clearly a product of the difference in
methodology for population estimation. The
filtered areal weighting method estimates that
11.60 percent of the population resides within
150 meters of the LAH, whereas the CEDS
method estimates only 10.25 percent of the
population in the same area. With a smaller
population (denominator), the rates, given equal
number of hospitalizations (numerator), will be
higher. The inverse is true for the results outside
the buffer (88.40 percent and 89.75 percent for
filtered areal weighting and CEDS, respectively).
Dasymetric Mapping—Where Do
We Go From Here?
Based on the application of the CEDS methodol-
ogy to New York City population data and given
the case study example of asthma hospitalization
rates in the Bronx, we have demonstrated that
the Cadastral-based Expert Dasymetric System
can improve research and analyses that utilize
population distribution information, while also
creating more realistic models of real-world con-
ditions. We are currently exploring techniques
for modifying the CEDS methodology to esti-
mate the spatial distribution of sub-populations,
such as those characterized by race/ethnicity, age
cohort, gender, or family structure. By employ-
ing different ancillary data sets, socio-economic
variables may also be mapped dasymetrically
with the CEDS method.
We have established the usefulness of the CEDS
method for any analyses employing population-based
rates (such as public health and epidemiological
research, crime mapping, and risk assessment),
but the CEDS method is not limited to improving
the development of rates alone. This method will
be useful in many disparate fields and serve many
purposes. For instance, one can improve emergency
management operations and implementation by
providing more precise information about actual
positions of susceptible populations, thereby increas-
ing the quality of functions such as evacuation route
planning, optimal site selection for emergency
shelter locations, and critical rescue and recovery
Figure 12b. Estimated asthma hospitalization rates inside and outside buffers for the entire Bronx, represented by
standardized ratios, with filtered areal weighting versus CEDS method.

100 Cartography and Geographic Information Science
prioritization for first responders. Obviously, this
can be extended to police operations, criminal
justice, fire and ambulance services, utility provid-
ers, and any other crucial public support systems
dependant upon population information.
Additionally, the knowledge of accurate popula-
tion distribution can be extremely valuable in the
sphere of urban planning. The understanding of
the locational characteristics of target populations
would allow for more equitable resource alloca-
tion in areas such as community infrastructure
development, provision of open space and recre-
ational opportunities, transportation access, and
necessary environmental facilities.
As the morphology of cities becomes increasingly
complex, the need continues to grow for immediate
and well informed decision-making with regard
to both catastrophic and everyday events. We
anticipate that advances in dasymetric mapping,
such as the CEDS method, will help us to “perfect
the denominator” and better our understanding
of the human urban project.
ACKNOWLEDGEMENTS
This research was partially supported by grant
number 2 R25 ES01185-05 from the National
Institute of Environmental Health Sciences of
the National Institutes of Health. The National
Oceanic and Atmospheric Administration’s
Cooperative Remote Sensing Science and
Technology Center (NOAA-CREST) also pro-
vided critical support for this project under
NOAA grant number NA17AE162. The state-
ments contained within this paper are not the
opinions of the funding agency or the U.S.
government, but reflect the authors’ opinions.
Thanks are also due to the member organiza-
tions of the South Bronx Environmental Justice
Partnership, who understood the relevance of
this project to environmental health justice and
gave their unstinting encouragement and assis-
tance in the effort.
REFERENCES
Baudot, Y. 2001. A method for the geographical
analysis of the population of fast-growing cities in
the Third World. In: Donnay, J.-P., Barnsley, M., and
Longley, P. (eds), Remote sensing and urban analysis.
London, U.K.: Taylor and Francis. pp. 249-68.
Bhaduri, B., E. Bright, P. Coleman, and J. Dobson.
2002. LandScan: Locating people is what matters.
GeoInformatics April/May: 34-35, 37.
Bielecka, E. 2005. A dasymetric population density map
of Poland. In: Proceedings of the 22nd International
Cartographic Conference, July 9-15, A Coruna, Spain.
Bracken, I., and D. Martin. 1989. The generation of
spatial population distributions from Census centroid
data. Environmental and Planning A, 21:537-43.
Cai, Q., B. Bhaduri, P. Coleman, G. Rushton, and E.
Bright. 2006. Estimating small-area populations
by age and sex using spatial interpolation and
statistical inference methods. Transactions in GIS
10(4): 577-98.
Clarke, M., and J. A. Maantay. 2006. Optimizing
recycling in all of New York City’s neighborhoods:
Using GIS to develop the REAP index for improved
Recycling Education, Awareness, and Participation.
Resources, Conservation, and Recycling 46 (2): 128-48.
Dent, B.D. 1999. Cartography: Thematic map design.
New York, New York: McGraw Hill.
Dorling, D., and D. Fairbairn. 1997. Mapping: Ways of
representing the world. Prentice Hall, Upper Saddle
River, New Jersey: Prentice Hall.
Eicher, C., and C. Brewer. 2001. Dasymetric mapping
and areal interpolation: Implementation and
evaluation. Cartography and Geographic Information
Science 28:125-38.
ESRI (Environmental Systems Research Institute).
2005. ArcGIS 9.1. Redlands, California, USA.
Flowerdew, R., and M. Green. 1994. Areal
interpolation and types of data. In: Fotheringham,
A.S. and Rogerson, P.A., (eds), Spatial analysis and
GIS. London, U.K.: Taylor and Francis. pp. 121-45.
Flowerdew, R., and M. Green. 1992. Developments
in areal interpolation methods and GIS. Annals of
Regional Science 26: 67-78.
Flowerdew, R., and M. Green. 1989. Statistical
methods for inference between incompatible zonal
systems. In: Goodchild, M.F., and S. Gopal (eds),
Accuracy of spatial databases. London, U.K.: Taylor
and Francis. pp. 239-48.
Flowerdew, R., M. Green, and E. Kehris. 1991.
Using areal interpolation method in geographic
information systems. Papers in Regional Science 70:
303-15.
Forster, B.C. 1985. An examination of some problems
and solutions in monitoring urban areas from
satellite platforms. International Journal of Remote
Sensing 6(1): 139-51.
Goodchild, M., and N.S-N. Lam. 1980. Areal
interpolation: A variant of the traditional spatial
problem. Geo-Processing 1: 297-312.
Goodchild, M., L. Anselin, and U. Deichmann.
1993. A framework for the areal interpolation of
socioeconomic data. Environment and Planning A
25: 383-97.
Gotway, C., and L. Young. 2002. Combining
incompatible spatial data. Journal of the American
Statistical Association 97: 632-48.
Gregory, I. N. 2000. An evaluation of the accuracy
of the areal interpolation of data for the analysis
of long-term change in England and Wales. In:
Proceedings of the 5th International Conference on
GeoComputation, University of Greenwich, United
Kingdom, 23 - 25 August 2000.

Vol. 34, No. 2 101
Hall, S. 1994. Mapping the next millenium: The discovery
of new geographies. New York, New York: Random
House.
Herrmann, C., and A.R. Maroko. 2006. Crime pattern
analysis: Exploring Bronx auto thefts using GIS.
In: Maantay, J.A., and J. Ziegler, GIS for the Urban
Environment. Redlands, California: ESRI Press.
Holloway, S.R., J. Schumacher, and R.L. Redmond.
1999. People and place: dasymetric mapping using
Arc/Info. Pages 283-291 in S. Morain, ed. GIS
Solutions in Natural Resource Management. OnWord
Press, Santa Fe, New Mexico.
Holt, J. B., C. P. Lo, and T. W. Hodler. 2004.
Dasymetric estimation of population density and
areal interpolation of census data. Cartography and
Geographic Information Science 31:103-21.
Koch, T. 2005. Cartographies of disease: Maps, mapping,
and medicine. Redlands, California: ESRI Press.
Kyriakidis, P. 2004. A geostatistical framework for area
to point spatial interpolation. Geographical Analysis
36(3):259-89.
Langford, M., D.J. Maguire, and D. Unwin. 1991. The
areal interpolation problem: Estimating population
using remote sensing in a GIS framework. In: Masser,
I., and M. Blakemore (eds), Handling geographic
information: Methodology and potential applications.
London, U.K.: Longman.
Langford, M., and D. Unwin. 1994. Generating
and mapping population density surfaces within
a geographical information system. Cartographic
Journal 31: 21-6.
Liu, X., K. Clarke, and M. Herold. 2006. Population
density and image texture: A comparison study.
Photogrammetric Engineering and Remote Sensing
72(2):187-96.
LotInfo, LLC. 2001. LotInfo. SpaceTrack, Inc. 304
Park Ave, 11th Floor New York, NY 10010.
Maantay, J.A., 2007/2005. Asthma and air pollution in
the Bronx: Methodological and data considerations
in using GIS for environmental justice and health
research. Health and Place [Special Issue: Linking
Environmental Justice, Population Health, and
Geographical Information Science] 23: 32-56. (E-
published, November 24, 2005).
Maantay, J.A. 2002. Industrial zoning changes in New
York City and environmental justice: A case study in
“expulsive” zoning. Projections: the Planning Journal
of Massachusetts Institute of Technology (MIT). (Special
Issue: Planning for Environmental Justice). pp. 63-
108.
Maantay, J.A. 2001. Zoning, equity, and public health.
American Journal of Public Health 91(7):1033–41.
Maantay, J.A., and A. H. Strelnick. 2003. Mapping
asthma hot spots: The geography of asthma and
air pollution in the Bronx. Journal of Urban Health,
Bulletin of the New York Academy of Medicine Vol. 80,
Supplement 2, Oxford University Press.
Maroko, A.R., and J.A. Maantay. Residential racial
segregation and health disparities in New York
City: Mapping TB and AIDS in conjunction with
segregation indices. Unpublished research study,
Urban GISc Lab, Lehman College.
Martin, D. 2006. An assessment of surface and zonal
models of population. International Journal of
Geographical Information Systems 10(8): 973-89.
Martin, D., M. Langford, and N.J. Tate. 2000.
Refining population surface models: Experiments
with Northern Ireland census data. Transactions in
GIS 4(4): 343-60.
Mennis, J. 2003. Generating surface models of
population using dasymetric mapping. The
Professional Geographer 55(1): 31-42.
Mennis, J., and T. Hultgren. 2005. Dasymetric
mapping for disaggregating coarse resolution
population data. In: Proceedings of the 22nd Annual
International Cartographic Conference, July 9-16, 2005,
A Coruna, Spain.
Moon, Z.K., and F. L. Farmer. 2001. Population density
surface: A new approach to an old problem. Society
and Natural Resources 14: 39-49.
New York City Department of Health. 2003. Asthma
facts, 2nd ed. New York, New York. [http://www.
nyc.gov/html/doh/pdf/asthma/facts.
pdf.]
New York State Department of Health, Statewide
Planning and Research Cooperative System. 2003.
Technical Documentation. Albany, New York.
Ong, P., and D. Houston. 2003. Draft Socioeconomic
Report for 2003 Air Quality Management Plan. South
Coast Air Quality Management District, Diamond
Bar, California.
Openshaw, S. 1984. Ecological fallacies and the
analysis of areal census data. Environmental and
Planning A 16: 17-31.
Poulsen, E., and L.W. Kennedy. 2004. Using dasymetric
mapping for spatially aggregated crime data. Journal
of Quantitative Criminology 20(3): 243-62.
Reibel, M., and M.E. Bufalino. 2005. Street-weighted
interpolation techniques for demographic
count estimation in incompatible zone systems.
Environmental and Planning A 37: 127-39.
Robinson, A.H. 1982. Early thematic mapping in the
history of cartography. Chicago, Illinois: University
of Chicago Press.
Scrope, G.P. 1833. Principles of Political Economy,
Deduced from the Natural Laws of Social Welfare, and
Applied to the Present State of Britain. London, U.K.:
Longmans.
Sleeter, R. 2004. Dasymetric mapping techniques for
the San Francisco Bay region, California. In: Urban
and Regional Information Systems Association Annual
Conference Proceedings, Reno, Nevada, November 7-
10, 2004.
Slocum, T.A. 1999. Thematic cartography and visualization.
Upper Saddle River, New Jersey: Prentice Hall.
Tobler, W. 1979. Smooth pycnophylactic interpolation
for geographical regions. Journal of the American
Statistical Association 74: 519-36.
United States Bureau of the Census. 2001(a). Census 2000
Summary File 1, New York State. Washington, D.C.

102 Cartography and Geographic Information Science
A message from the publisher of CaGIS
CaGIS online
Cartography and Geographic Information Science (CaGIS) has been published online since
2003, and soon, all CaGIS volumes published since 1974 will be available online. In order
to access individual articles, you need to register for access.
o to: hp://www.ingentaconnect.com/content/acsm/cagis
Need to register? Sign up here.”
Select the option for personal registration.
 
• Choose a username and password (username/password must be at least 5 characters).
Register.”
You can then set up your “Personal Subscriptions” thus:
 
• Enter in your publisher subscription number, check next to the journal title, and then press
Add.”


com/register/personal and follow instructions. You may need your MO member ID to
successfully complete the registration.
Online CaGIS is provided to members of the Cartography and Geographic Information
Society at no additional cost!! Members also have the option to opt for online only … mark
your renewal forms accordingly, if you want to receive only the online version of the journal.
If you are not a member of CaGISapply for membership
at hp://www.cartogis.org/membership.
We hope that you have a very pleasant experience browsing your online CaGIS
help with registration or accessing CaGIS
happy to provide further instructions.
CaGIS publisher
www.cartogis.org
United States Bureau of the Census. 2001(b). TIGER
Files. Geography Division, Cartographic Products
Management Branch. Washington, D.C.
Weichselbaum, J., F. Petrini-Monteferri, M. Papathoma,
W. Wagner, and N. Hackner. 2005. Sharpening census
information in GIS to meet real-world conditions:
The case for Earth observation. WIT Transactions on
Ecology and the Environment, Sustainable Development
and Planning II 1: 143-52.
Wright, J.K. 1936. A method of mapping densities
of population with Cape Cod as an example.
Geographical Review 26: 103-10.
Wu, S., X. Qiu, and L. Wang. 2005. Population
estimation methods in GIS and remote sensing: A
review. GIScience and Remote Sensing 42(1): 80-96.
Xie, Y. 1996. The overlaid network algorithms for areal
interpolation problem. Computers, Environment, and
Urban Systems 19: 287-306.