331
the spatial distribution of biodiversity. For example, within the context of developing understanding of the spatial distribution of sensitive arctic shore habitat and
biodiversity areas in the event of oil spills, Banks et al. (2014a) used decomposition
parameters in a comparison of three unsupervised polarimetric classifiers for their
potential in mapping multiple classes of substrate (nonvegetated), tundra vegetation, and wetland type. Similarly, Varghese et al. (2016) classified water, settlements, agriculture, shrub/scrub, and three forest density classes in a comparison of
parameters derived from six decomposition techniques. In an analogous manner,
specific classes within an ecotype can be mapped. For example, mapping the diversity of given classes within a wetland complex (e.g., Touzi and Deschamps 2007;
Gosselin et al. 2013; Dingle Robertson et al. 2015; Hong et al. 2015; Dubeau et al.
2017) can aid identification of the variety of habitat conditions available and potential biodiversity. In general, radar data have not been found to provide consistently
better overall classification accuracy than optical data. However, since radar data are
typically complementary and not highly correlated with optical data, they can provide additional information for certain vegetation classes that can be distinguished
by structure in cases where optically derived spectral reflectance and vegetation
indices are similar. Thus, combining radar and optical imagery has often been
shown to improve the accuracy of such classes over either data type alone (e.g.,
Bergen et al. 2007; Wang et al. 2009; Bwangoy et al. 2010; Banks et al. 2014b).
An alternative approach to thematic classification is estimation of vegetation
structure parameters that can serve as indicators of potential habitat diversity or
biodiversity, for example, the average or spatial heterogeneity of AGB, LAI, vegetation height, and stem and branch parameters. Luckman et al. (1997), Lucas et al.
(2006), and Le Toan et al. (2004), among others, have reported that the backscatterAGB relationship typically saturates in the range of 100–150 t/ha. However, AGB
spatial variability can be mapped in environments with lower vegetation density
(e.g., Häme et al. 2013), and efforts to produce suitable models with a higher saturation threshold by improving data information content are common. For example,
use of the following has proven beneficial: cross-polarized (HV) data rather than
co-polarized; longer wavelengths that penetrate deeper into the canopy (Santos
et al. 2003); ratios such as VV/HH (Manninen et al. 2009 for LAI); shorter-tolonger wavelength ratios such as C−/L-bands (Foody et al. 1997); averaging of
multitemporal data sets to reduce moisture/rain effects (Englhart et al. 2011); and
integrating optical and radar data (Vaglio et al. 2017). Imhoff et al. (1997) modeled
canopy parameters using steep incidence angle C-, L-, and P-band airborne radar
and found strong correlations for C-HV and LAI, L-VV and branch surface area or
volume, and P-VV with bole surface area or volume; these relationships were then
used to map broad avian habitat classes. Bergen et al. (2009) combined biomass
estimates from C- and L-band backscatter with Landsat vegetation classification,
thereby improving habitat mapping for three bird species over use of vegetation
type alone.
InSAR has been used to estimate canopy height and height variance, which can
be an indicator of vegetation type, structural complexity, and age diversity. Canopy
height is most commonly estimated from the difference between scattering phase
13 A Range of Earth Observation Techniques for Assessing Plant Diversity
the spatial distribution of biodiversity. For example, within the context of developing understanding of the spatial distribution of sensitive arctic shore habitat and
biodiversity areas in the event of oil spills, Banks et al. (2014a) used decomposition
parameters in a comparison of three unsupervised polarimetric classifiers for their
potential in mapping multiple classes of substrate (nonvegetated), tundra vegetation, and wetland type. Similarly, Varghese et al. (2016) classified water, settlements, agriculture, shrub/scrub, and three forest density classes in a comparison of
parameters derived from six decomposition techniques. In an analogous manner,
specific classes within an ecotype can be mapped. For example, mapping the diversity of given classes within a wetland complex (e.g., Touzi and Deschamps 2007;
Gosselin et al. 2013; Dingle Robertson et al. 2015; Hong et al. 2015; Dubeau et al.
2017) can aid identification of the variety of habitat conditions available and potential biodiversity. In general, radar data have not been found to provide consistently
better overall classification accuracy than optical data. However, since radar data are
typically complementary and not highly correlated with optical data, they can provide additional information for certain vegetation classes that can be distinguished
by structure in cases where optically derived spectral reflectance and vegetation
indices are similar. Thus, combining radar and optical imagery has often been
shown to improve the accuracy of such classes over either data type alone (e.g.,
Bergen et al. 2007; Wang et al. 2009; Bwangoy et al. 2010; Banks et al. 2014b).
An alternative approach to thematic classification is estimation of vegetation
structure parameters that can serve as indicators of potential habitat diversity or
biodiversity, for example, the average or spatial heterogeneity of AGB, LAI, vegetation height, and stem and branch parameters. Luckman et al. (1997), Lucas et al.
(2006), and Le Toan et al. (2004), among others, have reported that the backscatterAGB relationship typically saturates in the range of 100–150 t/ha. However, AGB
spatial variability can be mapped in environments with lower vegetation density
(e.g., Häme et al. 2013), and efforts to produce suitable models with a higher saturation threshold by improving data information content are common. For example,
use of the following has proven beneficial: cross-polarized (HV) data rather than
co-polarized; longer wavelengths that penetrate deeper into the canopy (Santos
et al. 2003); ratios such as VV/HH (Manninen et al. 2009 for LAI); shorter-tolonger wavelength ratios such as C−/L-bands (Foody et al. 1997); averaging of
multitemporal data sets to reduce moisture/rain effects (Englhart et al. 2011); and
integrating optical and radar data (Vaglio et al. 2017). Imhoff et al. (1997) modeled
canopy parameters using steep incidence angle C-, L-, and P-band airborne radar
and found strong correlations for C-HV and LAI, L-VV and branch surface area or
volume, and P-VV with bole surface area or volume; these relationships were then
used to map broad avian habitat classes. Bergen et al. (2009) combined biomass
estimates from C- and L-band backscatter with Landsat vegetation classification,
thereby improving habitat mapping for three bird species over use of vegetation
type alone.
InSAR has been used to estimate canopy height and height variance, which can
be an indicator of vegetation type, structural complexity, and age diversity. Canopy
height is most commonly estimated from the difference between scattering phase
13 A Range of Earth Observation Techniques for Assessing Plant Diversity