Environmental Engineering Reference
In-Depth Information
13.1
Introduction
of bands 1, 2, and 3 of SPOT HRV imagery to obtain popula-
tion density information in Hong Kong, China. He discovered
that strong negative correlations exist between population den-
sity and the radiances of band 3 (0.79-89 μm) and band 1
(0
.
500
.
59 μm), and a positive correlation exists between popula-
tion and the radiances of band 2 (0.6-0
.
68 μm). In Lo's study,
although urban biophysical parameters were not utilized, it is
clear that the radiances of bands 1 and 3 in SPOT HRV imagery
are closely associated with concentration of vegetation, which
has a negative relationship with population density. Moreover,
the radiance of band 2 is highly related to urban built-up areas;
therefore, it has a positive correlation with population density.
In addition to the radiances in individual bands, Harvey (2002)
utilized radiance transformations, including radiance squares,
cross-product of radiances in different bands, radiance ratio
from different bands, difference-sum-ratio, and other transfor-
mations. In another study, Sutton
et al
. (1997) utilized light
energy extracted from the Defense Meteorological Satellite Pro-
gram Operational Linescan System (DMSP-OLS) imagery for
population estimation. The light energy represents the intensity
of urban land uses, with higher energy existing in commercial
and residential areas, and lower energy in agricultural areas. In
summary, in these implicit estimations, although urban physi-
cal parameters are not directly utilized, specific urban physical
environments are represented by radiances and their transfor-
mations. The second category of regressionmodels utilizes urban
biophysical and land use information extracted from remote
sensing imagery for population estimation. For example, Lo
(1995, 2003) utilized high and low urban land use areas to esti-
mate zonal population counts. Chen (2002) used three levels of
residential density for projecting population density in Sydney,
Australia. Li and Weng (2005) also applied land use types as
independent variables for developing separate regression mod-
els for different residential regions in Indianapolis, Indiana. A
detailed review of existing methods based on remote sensing can
be found in Wu, Qiu and Wang, (2005).
Although very high-spatial resolution satellites (IKONOSwith
1 m and QuickBird with 0.65 m in their respective panchromatic
bands) have been available for around a decade now (IKONOS
launched in 1999 and QuickBird launched in 2001), studies
taking advantages of this fine spatial resolution for small-area
population estimations are still scarce and generally limited to
the application of traditional visual interpretation for housing
units count (e.g., Yagoub, 2006). Likewise, airborne light detec-
tion and ranging (lidar) devices have allowed rapid access to
vertical information of urban structures, but the integration of
this new level of information for population estimation has not
been fully investigated. Recently, it was suggested that the vol-
ume information provided by lidar could serve to best improve
small-area population estimations (Wu, Wang and Qiu, 2008).
Whether or not lidar measurements coupled with automated
techniques for building extraction and land use classification can
lead to improved small-area population estimation is still an
unresolved matter.
This study sought to address the following questions: Can lidar
and high-spatial resolution imagery be employed for automat-
ing and refining small-area population estimation? If so, what
types and levels of information extracted from these sensors are
mostly effective? These questions were addressed through a com-
parative study of seven linear models that drew upon inputs from
building count, building area and/or building volume, at two dif-
ferent land use levels. At the finer land use level, buildings were
Small-area population estimates are essential for understanding
and responding to many social, political, economic, and envi-
ronmental problems (Liu, 2003). The size and distribution of the
population often are key determinants for resource allocation
for state and local governments (Smith, Nogle and Cody, 2002).
Population estimates are critical in decisions about when and
where to build public facilities such as schools, libraries, sewage
treatment plants, hospitals, and transportation infrastructure.
For example, in public transit route design, population density is
considered a primary indicator of the number of potential daily
trips originating in an area (Benn, 1995), with population den-
sities below approximately 4000 persons per square mile found
to generate low demand for public transit (Downs, 1992; Trans-
portation Research Board, 1996, 1997). Population estimates are
also often used by the private sector for customer profile analysis,
market area delineation, and site location identification (Martin
and Williams, 1992; Plane and Rogerson, 1994). In addition,
population estimates are also extensively utilized as denomi-
nators in generating many diagnostic indicators for studies of
environmental and socioeconomic conditions and trends. Based
on these indicators, such as unemployment rates, mortality and
morbidity rates, etc., billions of dollars in public funds are allo-
cated every year. Lastly, population information is an important
input in many urban and regional models, such as land use
and transportation interaction models, urban sprawl analysis,
environment equity studies, and policy impact analysis (Rees,
Norman and Brown, 2004). In short, the generation of accurate
and timely population estimates is crucial (Smith, Nogle and
Cody, 2002). Although small-area population estimates are of
great significance and have many applications, detailed and accu-
rate population and socioeconomic information is only available
for one date per decade through the national census. There-
fore, the need for frequent intercensal updates, particularly in
rapid growth areas, is critically apparent for public and private
sector planning.
Remote sensing has long been used to estimate population,
particularly for large areas. The earliest application of remote
sensing for population estimates involved manually counting
the number of houses using aerial photos (Lo, 1986a, b). A
survey is then conducted to estimate the average number of
persons per house. The product of the number of houses and
average household size then produces the total population of
the study area. Although this method is relatively accurate, it
involves manual interpretation of aerialphotos.Thisisquitetime
consuming and labor intensive, which prohibits its application
in large urban areas. This method therefore is rarely used by state
and local agencies for small area population estimates. Moreover,
accurate ''persons per household'' information, which is required
for a variety of dwelling types, is difficult to obtain (Lo, 1989,
Watkins and Morrow-Jones, 1985).
Automatic approaches with satellite remote sensing imagery,
therefore, have been proposed for estimating population density
(Lo, 1995). These approaches can be classified into: (1) implicit
estimation, in which spectral radiance/reflectance information
and their transformations are utilized for population estima-
tion, and (2) explicit estimation, which utilizes urban physical
parameters extracted from remote sensing imagery for popula-
tion generation (Cowen and Jensen, 1998, Jensen and Cowen,
1999). For implicit estimation, Lo (1995) utilized the radiances
Search WWH ::
Custom Search