number of wavelength bands that can be used for accurate land cover classification and
inventory.
15.3 HYPERSPECTRAL REMOTE SENSING IN LAND COVER
EXTRACTION
15.3.1 Introduction
Remote sensing data have been an attractive source in the determination of land cover
thematic mapping, providing valuable information in delineating the extent of land
cover classes, as well as in performing temporal land cover change analysis at various
scales (Kavtzoglou and Colkesen, 2009). Multispectral remote sensing data have been
widely used for land use/cover mapping at diverse spatial scales (Cihlar, 2000; Carrao
et al., 2008). The multispectral remote sensors produce images in few relatively broad
spectral bands for every unit—such as pixel—on digital imagery. The repetitive
availability, relatively low cost per unit area, and high spatial resolution (e.g., less than
1 m from IKONOS, QuickBird, and WorldView-2 sensors) make the multispectral
information of potentially high interest to scientists and decision makers, thus being
exploited for land cover mapping and enhancing change detection applications (Xie
et al., 2008). However, the coarse spectral resolution is an important limitation of such
data for detailed land cover analysis. Even though the spatial resolution of some
multispectral sensor systems is very high, spectral differences of land cover such as
vegetation recorded with multispectral sensors are often very small (Avery and Berlin,
1992). Vegetation cover has naturally high spectral variability driven by the biochemical complexity, diversity, and phenology. At low canopy cover, where spectral
dominance of soil background occurs, analysis of vegetated landscapes becomes even
more difficult (Curran, 2001). Therefore, vegetation cover analysis on multispectral
imagery is performed at a community level or on land cover classes of broader
physical meaning (e.g., deciduous vs. evergreen vegetation) (Thenkabail et al., 2004b).
Unlike the multispectral sensors, hyperspectral remote sensors have been available
for about two decades or so. They provide spectral data for each measurement unit in
numerous continuous spectral bands. As the very narrow spectral bands characterize
the target reflectance better, this abundant spectral information can be used to more
effectively detect and identify the variability of different land covers on the basis of
the spectral signatures than is possible with broader band multispectral sensors
(Abdel-Rahman et al., 2010; Xie et al., 2008). Differences in the spectral, spatial,
and temporal characteristics between several multi- and hyperspectral imaging
systems are illustrated in Table 15.1.
The remainder of this section provides an overview of the techniques employed in
deriving information on the spatial distribution of land cover from hyperspectral
remote sensing imagery, providing results from selected case studies conducted
recently. In this framework, the importance of the observation scale in land cover
extraction using remote sensing data is also discussed. It should be noted that apart
from the techniques developed for performing land cover mapping from hyperspectral
302
HYPERSPECTRAL REMOTE SENSING WITH EMPHASIS ON LAND COVER MAPPING
inventory.
15.3 HYPERSPECTRAL REMOTE SENSING IN LAND COVER
EXTRACTION
15.3.1 Introduction
Remote sensing data have been an attractive source in the determination of land cover
thematic mapping, providing valuable information in delineating the extent of land
cover classes, as well as in performing temporal land cover change analysis at various
scales (Kavtzoglou and Colkesen, 2009). Multispectral remote sensing data have been
widely used for land use/cover mapping at diverse spatial scales (Cihlar, 2000; Carrao
et al., 2008). The multispectral remote sensors produce images in few relatively broad
spectral bands for every unit—such as pixel—on digital imagery. The repetitive
availability, relatively low cost per unit area, and high spatial resolution (e.g., less than
1 m from IKONOS, QuickBird, and WorldView-2 sensors) make the multispectral
information of potentially high interest to scientists and decision makers, thus being
exploited for land cover mapping and enhancing change detection applications (Xie
et al., 2008). However, the coarse spectral resolution is an important limitation of such
data for detailed land cover analysis. Even though the spatial resolution of some
multispectral sensor systems is very high, spectral differences of land cover such as
vegetation recorded with multispectral sensors are often very small (Avery and Berlin,
1992). Vegetation cover has naturally high spectral variability driven by the biochemical complexity, diversity, and phenology. At low canopy cover, where spectral
dominance of soil background occurs, analysis of vegetated landscapes becomes even
more difficult (Curran, 2001). Therefore, vegetation cover analysis on multispectral
imagery is performed at a community level or on land cover classes of broader
physical meaning (e.g., deciduous vs. evergreen vegetation) (Thenkabail et al., 2004b).
Unlike the multispectral sensors, hyperspectral remote sensors have been available
for about two decades or so. They provide spectral data for each measurement unit in
numerous continuous spectral bands. As the very narrow spectral bands characterize
the target reflectance better, this abundant spectral information can be used to more
effectively detect and identify the variability of different land covers on the basis of
the spectral signatures than is possible with broader band multispectral sensors
(Abdel-Rahman et al., 2010; Xie et al., 2008). Differences in the spectral, spatial,
and temporal characteristics between several multi- and hyperspectral imaging
systems are illustrated in Table 15.1.
The remainder of this section provides an overview of the techniques employed in
deriving information on the spatial distribution of land cover from hyperspectral
remote sensing imagery, providing results from selected case studies conducted
recently. In this framework, the importance of the observation scale in land cover
extraction using remote sensing data is also discussed. It should be noted that apart
from the techniques developed for performing land cover mapping from hyperspectral
302
HYPERSPECTRAL REMOTE SENSING WITH EMPHASIS ON LAND COVER MAPPING