parametric one in order to determine the similarity and their usefulness in potential
mapping. The Mann–Whitney U-test is a common nonparametric test used for further
pairwise comparisons in detailed discriminative spectral analysis.
15.2.3.3 Multivariate Statistical Techniques In contrast to the univariate, the
multivariate statistical techniques provide an optimal linear combination of more
dependent variables that satisfy specific statistical criteria as an explanation of more
independent variables. Consequently, such techniques are used as both dimension
reduction and explanatory tools.
Principal-component analysis (PCA) is a nonparametric multivariate data analysis
technique that takes a cloud of data points analyzed and rotates it such that the
maximum variability is visible. The outputs are uncorrelated explanatory variables, as
linear combinations of the original variables, called principal components, and
usually the first principal component accounts for the maximum variability in the
data, and each following component accounts for as much of the remaining variability
as possible (Castro-Esau et al., 2004).
Canonical discriminant analysis is another dimension–reduction multivariate
technique which explains which variables (here wavelengths) discriminate the
best between classes (vegetation groups). It summarizes the between-class variation
in much the same way that PCA summarizes the total variation (Dimitrakopoulos,
2001). Cluster analysis, or clustering, is the procedure of an assignment of a set of
observations into subsets (called clusters) so that observations in the same cluster are
similar in some sense, that is, in a way that the degree of association between two
objects is maximal if they belong to the same group and minimal otherwise. As given
above, cluster analysis can be used to determine how the data are organized without
providing an explanation or interpretation, that is, it is an unsupervised approach
(Holden and LeDrew, 1998).
Other statistical techniques and quantitative methods coupled on the spectral data
obtained at the field scale attempt not only to determine spectral regions where
different covers are most likely to be statistically discriminated but also to measure
how well different covers can be separated. Such separability measures look at either
the distance between the class means (e.g., the Euclidean distance) or both the
differences between the class means and the distribution of the values around those
means (e.g., Jeffreys–Matusita or Bhattacharyya distance) at one or at more wavelengths at a time.
15.2.3.4 Field Spectroradiometry Applications Related to Land Cover Mapping
Hyperspectral discrimination of different land covers at the field scale is a topic that
has been the subject of research over the last two decades (Clark et al., 2005;
Cochrane, 2000; Lewis, 2001; Manevski et al., 2011, 2012; van Aardt and Wynne,
2001). On the one hand, laboratory spectroradiometry omits the natural field
conditions, such as the variation of the sun’s energy, or the effect of canopy
architecture, just to mention few, thus resulting in investigation of prerequisites,
rather than applicability to the future investment of hyperspectral sensors onboard
(Vaiphasa et al., 2005). On the other hand, field spectroradiometry studies deal with
298
HYPERSPECTRAL REMOTE SENSING WITH EMPHASIS ON LAND COVER MAPPING
mapping. The Mann–Whitney U-test is a common nonparametric test used for further
pairwise comparisons in detailed discriminative spectral analysis.
15.2.3.3 Multivariate Statistical Techniques In contrast to the univariate, the
multivariate statistical techniques provide an optimal linear combination of more
dependent variables that satisfy specific statistical criteria as an explanation of more
independent variables. Consequently, such techniques are used as both dimension
reduction and explanatory tools.
Principal-component analysis (PCA) is a nonparametric multivariate data analysis
technique that takes a cloud of data points analyzed and rotates it such that the
maximum variability is visible. The outputs are uncorrelated explanatory variables, as
linear combinations of the original variables, called principal components, and
usually the first principal component accounts for the maximum variability in the
data, and each following component accounts for as much of the remaining variability
as possible (Castro-Esau et al., 2004).
Canonical discriminant analysis is another dimension–reduction multivariate
technique which explains which variables (here wavelengths) discriminate the
best between classes (vegetation groups). It summarizes the between-class variation
in much the same way that PCA summarizes the total variation (Dimitrakopoulos,
2001). Cluster analysis, or clustering, is the procedure of an assignment of a set of
observations into subsets (called clusters) so that observations in the same cluster are
similar in some sense, that is, in a way that the degree of association between two
objects is maximal if they belong to the same group and minimal otherwise. As given
above, cluster analysis can be used to determine how the data are organized without
providing an explanation or interpretation, that is, it is an unsupervised approach
(Holden and LeDrew, 1998).
Other statistical techniques and quantitative methods coupled on the spectral data
obtained at the field scale attempt not only to determine spectral regions where
different covers are most likely to be statistically discriminated but also to measure
how well different covers can be separated. Such separability measures look at either
the distance between the class means (e.g., the Euclidean distance) or both the
differences between the class means and the distribution of the values around those
means (e.g., Jeffreys–Matusita or Bhattacharyya distance) at one or at more wavelengths at a time.
15.2.3.4 Field Spectroradiometry Applications Related to Land Cover Mapping
Hyperspectral discrimination of different land covers at the field scale is a topic that
has been the subject of research over the last two decades (Clark et al., 2005;
Cochrane, 2000; Lewis, 2001; Manevski et al., 2011, 2012; van Aardt and Wynne,
2001). On the one hand, laboratory spectroradiometry omits the natural field
conditions, such as the variation of the sun’s energy, or the effect of canopy
architecture, just to mention few, thus resulting in investigation of prerequisites,
rather than applicability to the future investment of hyperspectral sensors onboard
(Vaiphasa et al., 2005). On the other hand, field spectroradiometry studies deal with
298
HYPERSPECTRAL REMOTE SENSING WITH EMPHASIS ON LAND COVER MAPPING