mapping (Weng
et al.,
2009) in temperate regions. However,
a few recent studies conducted in subtropical regions have
presented different results. Sung and Li (2011) found that
in subtropical regions, the use of a winter image produced a
more accurate mapping result, because there were still spe-
cies with dense foliage in the winter season. Another study
found that the spring image was more effective for impervious
surface estimation than the summer image in a subtropical
region (Zhang
et al.,
2009).
Accuracy is not the only concern of impervious surface
mapping. Mapping using summer images may significantly
underestimate the impervious surface area (Cablk and Minor,
2003; Lee and Lathrop, 2006; van der Linden and Hostert,
2009; Sung and Li, 2011). This is because the impervious
surfaces in urban areas are often obscured by tree canopies,
which are classified as vegetation in
VHR
imagery (Cablk and
Minor, 2003; Lee and Lathrop, 2006; van der Linden and
Hostert, 2009; Zhang
et al.,
2012). For example, van der Lin-
den and Hostert (2009) used airborne hyperspectral HyMap
data of the metropolitan area of Berlin, Germany, acquired
on 20 June 2005, to extract impervious surface. They found
that more than 30 percent of the street area was obscured by
tree canopies and was identified as vegetation by image clas-
sification. In contrast, because the deciduous trees drop their
leaves in winter, and the ground features can be seen more
distinctly, winter images can have maximum impervious ex-
posure. Zhou
et al
. (2012) therefore used multiple Landsat
TM
and
ETM+
images of the winter season in impervious surface
mapping.
As a result, the trade-off between the greater spectral dif-
ferentiation between tree canopies and impervious surface in
summer and the underestimation of the impervious surface
area under tree canopies should be considered together when
selecting an appropriate image acquisition date for impervi-
ous surface extraction (Sung and Li, 2011). However, most
existing studies (Wu and Yuan, 2007; Weng and Hu, 2008; Hu
and Weng, 2009; Hu and Weng, 2010) only focus on the accu-
racy of impervious surface mapping. Furthermore, all exist-
ing studies on the selection of the image acquisition date for
impervious surface mapping were conducted using medium
resolution images (i.e., Landsat
TM
/
ETM+
and
ASTER
) (Wu and
Yuan, 2007; Weng and Hu, 2008; Hu and Weng, 2009; Weng
et
al.,
2009; Sung and Li, 2011).
The overall objective of this study was to quantitatively
compare impervious surface extractions from summer and
winter
VHR
images over urban areas in the temperate region
and to investigate the seasonal sensitivity of impervious sur-
face extraction. The specific objectives were: (a) to develop an
appropriate method to map urban impervious surface using
VHR
imagery; (b) to compare the results of impervious sur-
face extraction from
VHR
images acquired in the summer and
winter seasons; and (c) to analyze the effect of the image ac-
quisition season on the impervious surface extraction and to
recommend a suitable season of image acquisition for urban
impervious surface extraction using
VHR
images.
Study Areas and Data
Two megacity urban regions from Northern China (Beijing
and Tianjin) were selected as study areas (Figure 1). These
two cities have experienced a rapid urbanization in the past
three decades, resulting in increased infrastructure and hous-
ing construction, and urban expansion. The cities are located
in the temperate region, with slightly different climates. The
Beijing area has a temperate continental monsoon climate
with four distinct seasons characterized by hot and humid
summers (due to the East Asian monsoon), and cold, windy,
and dry winters (reflecting the influence of the vast Siberian
anticyclone) (Wu
et al.,
2006). Influenced by the inland Bohai
Sea, the Tianjin area possesses a semi-moist continental mon-
soon climate, with monsoon prevailing all year round. It also
has four clearly divided seasons, including a hot and humid
summer, and a cold and dry winter with little snow.
Beijing Area and Data
According to the phenological indicator for the Beijing area
(Zhong
et al.,
2008; Zhong
et al.,
2012), summer usually
occurs from early-May to late-September, and winter from
late-October to early-March. The common evergreen species
include Chinese juniper (
Sabina chinensis
), Chinese Red
Pine (
Pinus tabuliformis
), and Common Boxwood (
Buxus
sinica
). The deciduous species include pagoda tree (
Sophora
japonica
), Salix babylonica, and Chinese Rose (
Rosa chinen-
sis
) (Chen
et al.,
1998). The evergreen trees account for nearly
one-third of the trees planted in the urban area of Beijing
(Chen
et al.,
1998). There is dense foliage of all trees in the
summer season. However, although the deciduous trees drop
their leaves and are dormant in winter, there are still some ev-
ergreen plants in parks, gardens, and other green spaces. Land
cover types in the area include tree, grass, water, impervious
surface (e.g., road, rooftop, and playground), and bare land
(very few, mainly in sites under construction) (Han, 2005).
Two WorldView-2 images acquired on 14 September 2012
(summer) and 01 February 2011 (winter, without snow), were
used in this study (Table 1). The WorldView-2 data include
eight multispectral bands with a 2 m resolution, namely
coastal, blue, green, yellow, red, red edge,
NIR1
, and
NIR2
, and
a panchromatic band with 0.5 m resolution. Because of its
improved spatial and higher spectral resolution, WorldView-2
images provide higher potential for distinction of urban land
cover features with similar spectral properties and better
classification accuracy compared with other
VHR
images with
only four multispectral bands such as Ikonos, QuickBird and
GeoEye-1 (DigitalGlobe, 2010; Novack
et al.,
2011; Pu and
Landry, 2012).
Figure 1. The location map of Beijing and Tianjin study areas (the
grey areas show the urban areas and the black points show the
locations of images used).
336
May 2016
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
