Environmental Engineering Reference
In-Depth Information
17.1 Introduction
Grafton WI and Mankato, MN, USA. The remainder of this
chapter is as follows. Section 17.2 reviews the pixel- and object-
based models for impervious surface estimation. Case studies of
these two groups of models are given in Sections 17.3 and 17.4.
In particular, several pixel-based models, including a spectral
mixture analysis, a regression tree model, and an integrated
approach, are detailed in Section 17.3, and an object-oriented
model is given in Section 17.4. Further discussion and conclusions
are detailed in Section 17.5.
Impervious surfaces are defined as those surfaces that water can-
not infiltrate. In urban areas, building rooftops, streets, highways,
parking lots, and sidewalks are the typical impervious surfaces.
While compacted soil or gravel, high clay content soils, and
extraction areas are also classifiedasimpervioussurfaces,trans-
portation elements have been determined to contribute the most
to total impervious surface area (Schueler, 1994). Urbanization
in the form of converting rural land uses to urban land uses is
directly associated with the increase of urban imperviousness.
Urban imperviousness directly affects water quality, the amount
of runoff to streams and lakes, aquatic habitats, and the aesthet-
ics of landscapes (Dougherty et al ., 2004). The spatial structure
of urban thermal patterns and urban heat balances are also
associated with urban surface characteristics. Therefore, accurate
measurement of impervious surface area provides an essential
indicator of environmental quality and valuable input to urban
planning and management activities (Schueler, 1994). Urban
imperviousness has been utilized to assess adverse influences of
urbanization on urban climate, air and water quality, and nat-
ural habitat (Schueler, 1994; Dougherty et al ., 2004; Yuan and
Bauer, 2007), and to quantify urban development and population
growth (Xian and Crane, 2005; Yang and Liu, 2005; Yang, 2006;
Wu and Murray, 2007; Morton and Yuan, 2009).
The most accurate methods for impervious mapping are tra-
ditional ground surveys and aerial photographic interpretation
and digitizing. These methods, however, are time consuming and
labor intensive. Alternatively, the decreasing costs and increasing
availability of multispectral digital imagery have led to more and
more successful programs of impervious surfaces mapping by
automatic processing of digital remote sensing data. For regional
scale studies, remote sensing of impervious surface has focused on
subpixel analysis using moderate resolution Landsat and SPOT
satellite data since they have the advantages of relatively large
coverage, multiple spectral bands, and comparatively low cost
(Adams et al ., 1995; Yang et al ., 2003; Wu and Murray, 2003;
Wu, 2004; Lu and Weng, 2004; Bauer, Loeffelholz, and Wil-
son, 2007). The major subpixel level approaches to estimate
percent impervious surface area include spectral mixture analy-
sis, regression analysis, regression tree, artificial neural networks
and expert systems. Detailed properties and comparison of major
methods can be found in Yuan, Wu, and Bauer (2008) andWeng
and Hu (2008).
On the other hand, for local studies, higher resolution satellite
imagery is preferred because the 30-m resolution of the Landsat
and the 20-m resolution of the SPOT data are generally not
sufficient to discriminate individual features (e.g., buildings,
streets, trees) within the urban mosaic (Small, 2003). Use of
high resolution satellite images for impervious surface mapping
has attracted more and more attention since the launches of
4-m IKONOS and 2.4-m Quickbird sensors in 1999 and 2001
respectively. In the literature, two groups of models, pixel-based
and object-based approaches, have been applied to estimating
both medium- and high-resolution impervious surface areas.
Pixel based models consider each individual pixel as the unit
of analysis, while object-based approaches employ pre-identified
objectsastheanalysisunit.
The objective of this chapter is to review popular pixel-
and object-based models for impervious surface estimation, and
comparethesetwogroupsofmodelsthroughcasestudiesin
17.2 Impervious surface
estimation
17.2.1 Pixel-basedmodels
For estimating impervious surface areas, many pixel-based
models have been developed in the literature. These models,
especially subpixel analysis methods, have been widely applied
to medium-resolution imagery (e.g., Landsat and SPOT data).
For high-resolution imagery, pixel size may be comparable to or
smaller than the size of urban objects, and therefore most pixels
in the imagery can be considered as pure pixels. While there
are usually less mixed pixels in high-resolution satellite imagery
such as 4-m IKONOS and 2.4-m Quickbird data, the mixed
pixel problem still exists. In particular, Wu (2009) estimated the
proportion of pure impervious surface pixels in Grafton, WI, and
found that approximately 40-50% of pixels containing imper-
vious surfaces are mixed pixels. Therefore, mixed pixel problem
may still be one of the major factors that affect the accuracy
of impervious surface estimation in high resolution imagery,
given the heterogeneous characteristics of urban environment.
Thus, this section reviews several pixel-based methods applied to
bothmedium and high resolution remote sensing imagery. These
models include regression modeling, spectral mixture analysis,
regression tree, and artificial neural network (ANN).
17.2.1.1 Regressionmodeling
Regression modeling estimates impervious surface areas through
constructing the relationship between the percentage of imper-
vioussurfaceareas(%ISA)andthespectraland/orspatial
information of an individual pixel. Specially, greenness index,
the second component of the Tasseled Cap (TC) transformation
or the Normalized Difference Vegetation Index (NDVI) has been
applied to both medium and high resolution remote sensing
imagery (Heinert, 2002; Gillies et al ., 2003; Sawaya et al . 2003;
Yuan et al ., 2005; Bauer, Loffelholz and Wilson, 2007). In urban-
ized areas with little amount of soil, the amount of impervious
surface area is significantly and inversely correlated to the amount
of green vegetation, represented by the TC Greenness or NDVI.
With the constructed relationship, high resolution impervious
surfaces have been derived with reasonable accuracy (Sawaya
et al ., 2003).
17.2.1.2 Spectral mixture analysis
Spectral mixture analysis (SMA) is utilized for calculating land
cover fractions within a pixel and involves modeling a mixed
Search WWH ::




Custom Search