321
Different modeling techniques have been used to model the local species
diversity- spectral heterogeneity relationship, ranging from simple univariate models (Gould 2000), to multivariate statistics (Feilhauer and Schmidtlein 2009), to
neural networks (Foody and Cutler 2003) and generalized additive models (GAMs,
Parviainen et al. 2009). A number of different measures of spectral heterogeneity
have been proposed and used to assess ecological heterogeneity and thus species
diversity (Cavender-Bares et al., Chap. 2). Many of these are related to the variability in a spectral space of different pixel values, such as the variance or texture in a
neighborhood of the spectral response (Gillespie 2005) or the distance from the
spectral centroid, which may be represented as the mean of spectral values in a
multidimensional system whose axes are represented by each image band or by
principal components where noise related to band collinearity has been removed
(Rocchini 2007). Moreover, in addition to the use of common vegetation indices
such as the Normalized Difference Vegetation Index (NDVI), some studies have
demonstrated an increase in the strength of the relationship when using additional
spectral information (e.g., Landsat bands 5 and 7 in the shortwave infrared (SWIR)
(Rocchini 2007) and (Nagendra et al. 2010)).
Beta Diversity
While alpha diversity is related to local variability, species turnover (beta diversity)
is a crucial parameter when trying to identify high-biodiversity areas (Baselga
2013). In fact, for a given level of local species richness, high beta diversity leads to
high global diversity of the area. This is one of the basic rules underpinning the
concept of irreplaceability of protected areas (e.g., Wegmann et al. 2014).
In some cases spatial distance/dispersal ability might not be the only driver of
species turnover, which seems to be more strictly related to environmental conditions. Hence, models have been built to relate species and spectral turnover to
explain their potential relationship and its causes (Rocchini et al. 2018b). In some
cases, spatial distance accounted only for a small fraction of variance in species
similarity, while environmental variation is expected to account for a much larger
one. When using spatial distances, distance decay does not necessarily account for
environmental heterogeneity (Palmer and Michael 2005), especially in heavily fragmented landscapes. Thus, the use of spectral distances for summarizing beta diversity patterns may be more reliable because this method explicitly takes environmental
heterogeneity into account instead of mere spatial distances among sites. Therefore,
it is expected that the higher the spectral distance among sites, the higher their difference in terms of environmental niches, potentially leading to higher beta diversity.
A straightforward method for measuring beta diversity is to calculate the differences between pairs of plots in terms of their species composition using one of the
many (dis)similarity coefficients proposed in the ecological literature (e.g., Legendre
and Legendre 1998) and assess the spectral turnover variability derived remotely
from the variation in species composition among sites. This has been mainly related
to spectral distance decay models in which species similarity decays once spectral
13 A Range of Earth Observation Techniques for Assessing Plant Diversity
Different modeling techniques have been used to model the local species
diversity- spectral heterogeneity relationship, ranging from simple univariate models (Gould 2000), to multivariate statistics (Feilhauer and Schmidtlein 2009), to
neural networks (Foody and Cutler 2003) and generalized additive models (GAMs,
Parviainen et al. 2009). A number of different measures of spectral heterogeneity
have been proposed and used to assess ecological heterogeneity and thus species
diversity (Cavender-Bares et al., Chap. 2). Many of these are related to the variability in a spectral space of different pixel values, such as the variance or texture in a
neighborhood of the spectral response (Gillespie 2005) or the distance from the
spectral centroid, which may be represented as the mean of spectral values in a
multidimensional system whose axes are represented by each image band or by
principal components where noise related to band collinearity has been removed
(Rocchini 2007). Moreover, in addition to the use of common vegetation indices
such as the Normalized Difference Vegetation Index (NDVI), some studies have
demonstrated an increase in the strength of the relationship when using additional
spectral information (e.g., Landsat bands 5 and 7 in the shortwave infrared (SWIR)
(Rocchini 2007) and (Nagendra et al. 2010)).
Beta Diversity
While alpha diversity is related to local variability, species turnover (beta diversity)
is a crucial parameter when trying to identify high-biodiversity areas (Baselga
2013). In fact, for a given level of local species richness, high beta diversity leads to
high global diversity of the area. This is one of the basic rules underpinning the
concept of irreplaceability of protected areas (e.g., Wegmann et al. 2014).
In some cases spatial distance/dispersal ability might not be the only driver of
species turnover, which seems to be more strictly related to environmental conditions. Hence, models have been built to relate species and spectral turnover to
explain their potential relationship and its causes (Rocchini et al. 2018b). In some
cases, spatial distance accounted only for a small fraction of variance in species
similarity, while environmental variation is expected to account for a much larger
one. When using spatial distances, distance decay does not necessarily account for
environmental heterogeneity (Palmer and Michael 2005), especially in heavily fragmented landscapes. Thus, the use of spectral distances for summarizing beta diversity patterns may be more reliable because this method explicitly takes environmental
heterogeneity into account instead of mere spatial distances among sites. Therefore,
it is expected that the higher the spectral distance among sites, the higher their difference in terms of environmental niches, potentially leading to higher beta diversity.
A straightforward method for measuring beta diversity is to calculate the differences between pairs of plots in terms of their species composition using one of the
many (dis)similarity coefficients proposed in the ecological literature (e.g., Legendre
and Legendre 1998) and assess the spectral turnover variability derived remotely
from the variation in species composition among sites. This has been mainly related
to spectral distance decay models in which species similarity decays once spectral
13 A Range of Earth Observation Techniques for Assessing Plant Diversity