PE&RS May 2016 - page 338

green vegetation period from April until November (Liu
et al.,
2014). The Tianjin area has a relatively low vegetation cover-
age compared with the Beijing area. An old, well-developed,
central urban area in Tianjin was selected as the second study
area (Figure 1). The area includes most urban segments (com-
mercial, industrial, and residential) with many buildings and
roads, as well as small green patches and water bodies.
Two WorldView-2 images, acquired on 09 September 2014
(summer) and 05 February 2015 (winter, without snow), were
used in this study (Table 1). The image preprocessing proce-
dure (e.g., image fusion and co-registration) for the Tianjin
area was the same as that used for the Beijing images.
A pan-sharpened multispectral image subset of 5,400
×
4,600
pixels of the Tianjin area was used (Plate 2). The land cover
types are similar to those in the Beijing area, with a higher pro-
portion of impervious surface mainly occupied by residential
quarters with less vegetation and, especially, tree cover.
Methods
An object-based classification method was adopted to extract
impervious surface from both summer and winter images,
as many studies have shown that the object-based classifica-
tion methods are more appropriate for
VHR
image processing
and analysis than the pixel-based methods (Cablk and Minor,
2003; Yuan and Bauer, 2006; Li
et al.,
2011).
Prior to object-based classification, image segmentation was
performed to generate multilevel segmentation results, which
are required in the proposed method. The proposed object-
based impervious surface extraction method includes three se-
quential steps. First, a general object-based land cover classifi-
cation was conducted using a coarse level segmentation result
(explained later in this section), where shadow was considered
as a class. Second, the shaded areas identified in the first step
were further classified into two classes; shaded impervious
surface and shaded pervious surface using a fine level segmen-
tation result, since a significant proportion of
VHR
imagery over
urban area is affected by shadows. In most existing studies,
the shaded areas were often left unclassified or simply classi-
fied as shadow (Shackelford and Davis, 2003), resulting in a
significant loss of land cover information (Zhou
et al.,
2009).
Finally, the classification results from the two previous steps
were combined, and all classes were then aggregated into two
classes: impervious surface and pervious surface. The full pro-
cedure of the proposed method is shown in Figure 2.
Class Definition and Selection
Determination of an appropriate classification scheme is an
important step in a land cover classification task (Jensen,
2004). Similar to many other urban areas, the land cover
classes in the two study areas selected generally include im-
pervious surface, tree, grass, water, bare land, and shadow. The
impervious surface often shows significant spectral variation.
Different types of impervious surface (e.g., road, rooftop, and
playground) have different spectral signatures (e.g., dark or
bright) (Lu
et al.,
2011). On the other hand, the impervious sur-
face is also spectrally similar to other land cover classes, such
as bare land, water, and shadow (e.g., Ji and Jensen, 1999; Lu
et
al.,
2011). Thus, to properly represent different spectral signa-
tures of all impervious surface materials and to produce more
accurate classification results, several subclasses of impervious
surface were selected in the general land cover classification,
including rooftop, road (including walkway), and other imper-
vious surface (e.g., playground and parking lot).
Most land cover types in the study areas that appeared
in the winter images were the same as those in the summer
images. However, the main difference in land cover type be-
tween summer and winter seasons was observed in the areas
covered by deciduous trees. Most deciduous trees are planted
along streets and sidewalks, while some deciduous trees are
planted with grass, bushes, or evergreen trees in garden or
other green spaces. In summer, the areas covered by decidu-
ous trees are shown as vegetation in a remotely sensed view
which would all be classified as vegetation using
VHR
satellite
images. However, in winter deciduous trees drop their leaves
and are dormant, leaving only non-photosynthetic vegetation
(
NPV
), such as tree trunks, stems, and branches. Thus, land
cover types underneath the deciduous tree canopies, e.g., im-
pervious surface and vegetation (e.g., grass and bush), are ex-
posed or mixed with
NPV
and its shadow in the winter image.
These exposed areas in winter images should be classified
into different classes accordingly (e.g., the exposed impervi-
ous surface would be classified as impervious surface), rather
than vegetation as identified by using summer images.
Plate 3 shows a typical scenario of a deciduous tree
covered zone in the study area. Plate 3A-1 shows a parterre
area which is mainly covered by dense foliage of deciduous
trees in summer. In winter, evergreen trees, bushes, and grass
underneath deciduous trees in the area are exposed and some
of them are mixed with
NPV
and its shadow (Plate 3B-1). As
NPV
is very sparse and its shadow only occupies a very small
section, the mixed area mainly shows the spectral signature
of the vegetation types underneath (Plate 3B-1), and thus the
area is still considered as vegetation, e.g., evergreen trees
mixed with
NPV
and shadow are all considered as the class
tree. On the other hand, although the grass in winter is not as
vigorous as in summer, and is potentially withered and yel-
low, it still shows a distinct spectral signature that maintains
its separability from impervious surface (Everitt
et al.,
2007).
Thus, grass mixed with
NPV
and shadow in the winter image
was also categorized as the class grass.
Figure 2. The procedure of impervious surface extraction pro-
posed in this study.
338
May 2016
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
299...,328,329,330,331,332,333,334,335,336,337 339,340,341,342,343,344,345,346,347,348,...390
Powered by FlippingBook