encountered in many medium- to high-resolution sensors, is one of the main tasks of
the spectral libraries. Efforts are therefore being made toward the creation of extensive
spectral libraries of land cover for different environments (wetlands, deserts, etc.) and
land cover types (vegetation, soils, minerals) and across a broad range of bioclimatic,
edaphic, and disturbance conditions (Hueni et al., 2009; Milton et al., 2009).
A few publicly available spectral libraries exist and are available today, such as the
recent SPECCHIO online spectral database system from the Remote Sensing
Laboratories (RSL) of the University of Zurich (http://www.specchio.ch/). The latter
includes rich metadata sets enclosed within the reference spectra and spectral
campaign data in order to ensure longevity and shareablity of spectral data between
research groups (Hueni et al., 2009).
The U.S. Geological Survey (USGS) offers a comprehensive spectral library
(http://speclab.cr.usgs.gov/) and integrates the ASTER spectral library, the Johns
Hopkins University (JHU) Spectral Library, and the Jet Propulsion Laboratory (JPL)
Spectral Library (Baldridge et al., 2009). This library offers a comprehensive choice
of land cover such as vegetation and man-made materials, measured with different
spectrometers, such as the Beckman 5270 (200–3000 nm), the ASD portable field
spectrometer (350–2500 nm), the Nicolet Fourier transform infrared (FTIR) interferometer spectrometer (1300–15000 nm), and the NASA Airborne Visible/Infra-Red
Imaging Spectrometer (AVIRIS) (400–2500 nm).
Other small-scale spectral libraries are also developed, such as the Vegetation
Spectral Library (VSL) of the Systems Ecology Laboratory at the University of Texas
at El Paso (UTEP) in cooperation with colleagues at the University of Alberta (http://
spectrallibrary.utep.edu/). Metadata regarding type of target and measurement site
and images are also enclosed. The spectral library is open to contributions of new
users to upload their spectral libraries toward building up an extensive knowledge
base worldwide.
15.2.3 Statistical Approaches in Field Spectroradiometry for Vegetation
Discrimination
15.2.3.1 Basic Assumptions The natural variation of vegetation reflectance results
in its distribution, which usually does not cluster around the mean in a bell shape if
graphically presented; that is, it does not have a normal, or Gaussian, distribution
(Manevski et al., 2012). Other studies report normality in the spectral reflectance data
distribution (Mutanga and Skidmore, 2007; Vaiphasa et al., 2005). As a consequence,
statistical techniques coupled with such data can be grouped into parametric and
nonparametric tests. Parametric tests assume normal data distribution and homogeneous variance as a measure of the amount of variation in the data being compared.
They are considered robust because violations of the assumptions, especially of the
data distribution, do not significantly affect the tests (Robson, 1994). On the other
hand, when the distribution is obviously not bell shaped, neither is the variance
homogeneous, alternative approaches, such as nonparametric techniques, are recommended. In both cases, the resulting p-value demonstrates the probability of making
an error in conluding a difference between the groups being compared when none
296
HYPERSPECTRAL REMOTE SENSING WITH EMPHASIS ON LAND COVER MAPPING
the spectral libraries. Efforts are therefore being made toward the creation of extensive
spectral libraries of land cover for different environments (wetlands, deserts, etc.) and
land cover types (vegetation, soils, minerals) and across a broad range of bioclimatic,
edaphic, and disturbance conditions (Hueni et al., 2009; Milton et al., 2009).
A few publicly available spectral libraries exist and are available today, such as the
recent SPECCHIO online spectral database system from the Remote Sensing
Laboratories (RSL) of the University of Zurich (http://www.specchio.ch/). The latter
includes rich metadata sets enclosed within the reference spectra and spectral
campaign data in order to ensure longevity and shareablity of spectral data between
research groups (Hueni et al., 2009).
The U.S. Geological Survey (USGS) offers a comprehensive spectral library
(http://speclab.cr.usgs.gov/) and integrates the ASTER spectral library, the Johns
Hopkins University (JHU) Spectral Library, and the Jet Propulsion Laboratory (JPL)
Spectral Library (Baldridge et al., 2009). This library offers a comprehensive choice
of land cover such as vegetation and man-made materials, measured with different
spectrometers, such as the Beckman 5270 (200–3000 nm), the ASD portable field
spectrometer (350–2500 nm), the Nicolet Fourier transform infrared (FTIR) interferometer spectrometer (1300–15000 nm), and the NASA Airborne Visible/Infra-Red
Imaging Spectrometer (AVIRIS) (400–2500 nm).
Other small-scale spectral libraries are also developed, such as the Vegetation
Spectral Library (VSL) of the Systems Ecology Laboratory at the University of Texas
at El Paso (UTEP) in cooperation with colleagues at the University of Alberta (http://
spectrallibrary.utep.edu/). Metadata regarding type of target and measurement site
and images are also enclosed. The spectral library is open to contributions of new
users to upload their spectral libraries toward building up an extensive knowledge
base worldwide.
15.2.3 Statistical Approaches in Field Spectroradiometry for Vegetation
Discrimination
15.2.3.1 Basic Assumptions The natural variation of vegetation reflectance results
in its distribution, which usually does not cluster around the mean in a bell shape if
graphically presented; that is, it does not have a normal, or Gaussian, distribution
(Manevski et al., 2012). Other studies report normality in the spectral reflectance data
distribution (Mutanga and Skidmore, 2007; Vaiphasa et al., 2005). As a consequence,
statistical techniques coupled with such data can be grouped into parametric and
nonparametric tests. Parametric tests assume normal data distribution and homogeneous variance as a measure of the amount of variation in the data being compared.
They are considered robust because violations of the assumptions, especially of the
data distribution, do not significantly affect the tests (Robson, 1994). On the other
hand, when the distribution is obviously not bell shaped, neither is the variance
homogeneous, alternative approaches, such as nonparametric techniques, are recommended. In both cases, the resulting p-value demonstrates the probability of making
an error in conluding a difference between the groups being compared when none
296
HYPERSPECTRAL REMOTE SENSING WITH EMPHASIS ON LAND COVER MAPPING