511
growing season), and (iii) vegetation physical structure, including overall plant
form and leaf shape and longevity. At the leaf level, the interaction between a ray of
light and foliar tissue is generally consistent among taxa, with biochemistry (e.g.,
pigments), water content, and leaf structure as the primary drivers of absorption and
scattering (Jacquemoud et al. 2009). Scaled up to the canopy—i.e., as viewed by a
satellite—the interactions change due to both the structure and ecology of vegetation. Variations in leaf area and layering, leaf orientation distribution and clumping,
multiple scattering, and canopy heterogeneity add complexity to the signal that varies based on gross vegetation structure (Jacquemoud et al. 2009). However, ecological differences among vegetation structural types may be as significant: In some
cases, the structural variability itself may be important to characterizing ecological
dimensions of biodiversity (Townsend et al. 2013) or may confound the detection of
variables driving diversity (Knyazikhin et al. 2013). Most significantly, however,
the interpretation of variability (diversity) in retrieved traits may differ among ecosystem types.
Because of vegetation structure, the algorithms to map components of biodiversity (e.g., functional traits; see Serbin and Townsend, Chap. 3) differ. For example,
a key ecological trait to characterize ecosystems is leaf mass per area (LMA), which
is frequently used as a basis to estimate photosynthetic capacity and is also representative of environmental or evolutionary trade-offs between leaf construction
costs and carbon uptake, i.e., longevity (thick leaves or needles, high LMA) vs. fast
growth (thin leaves, high leaf area, low LMA) (Díaz et al. 2016; Reich et al. 1992;
Shipley et al. 2006; Wright et al. 2004).
Differences in algorithms among functional types are illustrated by the retrieval
algorithms to predict LMA and nitrogen in grasslands vs. forests (Fig. 19.2), which
exhibit considerable differences due to variations in bidirectional reflectance.
Standardized partial least squares regression (PLSR) coefficients indicate the relative contribution of a wavelength to prediction of a trait, and PLSR is one among
multiple methods for mapping traits from imaging spectroscopy (discussed in more
detail in Serbin and Townsend, Chap. 3). PLSR coefficients are especially useful for
diagnostic purposes because they can be interpreted with respect to known foliar
features. Figure 19.1 illustrates the differences in PLSR retrievals between ecosystem types but also that combined equations are possible. Also, LMA and nitrogen
are generally negatively correlated among ecosystems worldwide (Wright et al.
2004), and as a consequence their PLSR coefficients are inverse of each other at
some wavelengths (e.g., the grassland and forest model at 1730 nm).
The difference in structure among ecosystem types is driven in large part by
several components of vegetation variation within a pixel. This has implications for
both the derivation of biodiversity parameters and their interpretation. Leaf area
index and canopy layering greatly influence the detection of vegetation characteristics. Elements of the canopy not directly exposed to the sun will be greatly reduced
in significance for the calculation of any canopy characteristic. In forests, this means
that a rich understory is often obscured, in contrast to a grassland in which many
more components of the canopy are in the sunlit portion of the canopy exposed to a
sensor. In addition to having technical consequences for the detection of functional
or taxonomic diversity of a whole canopy, this also means that diversity metrics of
19 Prospects and Pitfalls for Spectroscopic Remote Sensing of Biodiversity…
growing season), and (iii) vegetation physical structure, including overall plant
form and leaf shape and longevity. At the leaf level, the interaction between a ray of
light and foliar tissue is generally consistent among taxa, with biochemistry (e.g.,
pigments), water content, and leaf structure as the primary drivers of absorption and
scattering (Jacquemoud et al. 2009). Scaled up to the canopy—i.e., as viewed by a
satellite—the interactions change due to both the structure and ecology of vegetation. Variations in leaf area and layering, leaf orientation distribution and clumping,
multiple scattering, and canopy heterogeneity add complexity to the signal that varies based on gross vegetation structure (Jacquemoud et al. 2009). However, ecological differences among vegetation structural types may be as significant: In some
cases, the structural variability itself may be important to characterizing ecological
dimensions of biodiversity (Townsend et al. 2013) or may confound the detection of
variables driving diversity (Knyazikhin et al. 2013). Most significantly, however,
the interpretation of variability (diversity) in retrieved traits may differ among ecosystem types.
Because of vegetation structure, the algorithms to map components of biodiversity (e.g., functional traits; see Serbin and Townsend, Chap. 3) differ. For example,
a key ecological trait to characterize ecosystems is leaf mass per area (LMA), which
is frequently used as a basis to estimate photosynthetic capacity and is also representative of environmental or evolutionary trade-offs between leaf construction
costs and carbon uptake, i.e., longevity (thick leaves or needles, high LMA) vs. fast
growth (thin leaves, high leaf area, low LMA) (Díaz et al. 2016; Reich et al. 1992;
Shipley et al. 2006; Wright et al. 2004).
Differences in algorithms among functional types are illustrated by the retrieval
algorithms to predict LMA and nitrogen in grasslands vs. forests (Fig. 19.2), which
exhibit considerable differences due to variations in bidirectional reflectance.
Standardized partial least squares regression (PLSR) coefficients indicate the relative contribution of a wavelength to prediction of a trait, and PLSR is one among
multiple methods for mapping traits from imaging spectroscopy (discussed in more
detail in Serbin and Townsend, Chap. 3). PLSR coefficients are especially useful for
diagnostic purposes because they can be interpreted with respect to known foliar
features. Figure 19.1 illustrates the differences in PLSR retrievals between ecosystem types but also that combined equations are possible. Also, LMA and nitrogen
are generally negatively correlated among ecosystems worldwide (Wright et al.
2004), and as a consequence their PLSR coefficients are inverse of each other at
some wavelengths (e.g., the grassland and forest model at 1730 nm).
The difference in structure among ecosystem types is driven in large part by
several components of vegetation variation within a pixel. This has implications for
both the derivation of biodiversity parameters and their interpretation. Leaf area
index and canopy layering greatly influence the detection of vegetation characteristics. Elements of the canopy not directly exposed to the sun will be greatly reduced
in significance for the calculation of any canopy characteristic. In forests, this means
that a rich understory is often obscured, in contrast to a grassland in which many
more components of the canopy are in the sunlit portion of the canopy exposed to a
sensor. In addition to having technical consequences for the detection of functional
or taxonomic diversity of a whole canopy, this also means that diversity metrics of
19 Prospects and Pitfalls for Spectroscopic Remote Sensing of Biodiversity…