67
vegetation) to characterize spatial and temporal patterns in plant functional (e.g.,
pigments) and structural (e.g., LAI) properties. In RTM inversion, the leaf-scale
model is often the focus, where the goal is to invert the canopy and leaf models
jointly to extract estimated foliar traits based on observed canopy reflectance (e.g.,
Colombo et al. 2008). Many other studies have focused on retrieving canopy-scale
parameters, such as LAI (e.g., Darvishzadeh et al. 2008; Banskota et al. 2015).
Early approaches leveraged RTM inversions that focused on numerical optimization
techniques to minimize the difference between modeled and observed reflectance
across similar wavelengths (e.g., Jacquemoud et al. 1995). Other methods have utilized look-up table (LUT) inversion (e.g., Weiss et al. 2000) where a range of simulated canopy reflectance patterns are generated in advanced by varying leaf and
canopy inputs across predetermined values. These simulated spectra are then compared to observations where either a single or select number of closely matching
modeled spectra, and their associated inputs, are selected as the solution to the
inversion. Bayesian RTM inversion methods have also been utilized (e.g.,
Shiklomanov et al. 2016) as a means to retrieve leaf and canopy properties as joint
posterior probability distributions through iterative sampling of the input parameter
space. The use of RTMs ranges from retrieval of vegetation functional and structural
traits to the characterization of landscape functional diversity (Kattenborn et al.
2017; Kattenborn et al. 2019).
3.3 Important Considerations, Caveats, and Future
Opportunities
3.3.1 Field Sampling and Scaling Considerations
There are several important considerations and best practices when developing
algorithms for the remote estimation of plant traits (see Schweiger, Chap. 15). We
will only briefly touch on these here. A key first step is to consider the scope of the
research and area of interest, focusing specifically on considerations such as local
climate conditions, terrain, vegetation, and canopy access. Specifically, the spatial
locations, site, and canopy access (e.g., is it possible to reach canopy foliage?);
vegetation composition and canopy architecture; timing of collection; and methods
for sample retrieval are key to identify prior to field campaigns in order to maximize
the utility of the field samples for conversion of RS signatures to accurate trait maps.
Furthermore, it may be important to consider what approach may be best to characterize the vegetation canopy architecture and/or composition to facilitate scaling of
each trait to the pixel or plot scale (e.g., using basal area, LAI). This may strongly
depend on the dominant vegetation types, where more open canopies may require a
different approach to a closed canopy, or on the spatial resolution of the imagery.
Observational data range is a primary consideration (see Schweiger, Chap. 15), and
sample locations should be chosen to cover the range of canopy types and vegetation communities that will fall within the RS observations. The timing of the field
3 Scaling Functional Traits from Leaves to Canopies
vegetation) to characterize spatial and temporal patterns in plant functional (e.g.,
pigments) and structural (e.g., LAI) properties. In RTM inversion, the leaf-scale
model is often the focus, where the goal is to invert the canopy and leaf models
jointly to extract estimated foliar traits based on observed canopy reflectance (e.g.,
Colombo et al. 2008). Many other studies have focused on retrieving canopy-scale
parameters, such as LAI (e.g., Darvishzadeh et al. 2008; Banskota et al. 2015).
Early approaches leveraged RTM inversions that focused on numerical optimization
techniques to minimize the difference between modeled and observed reflectance
across similar wavelengths (e.g., Jacquemoud et al. 1995). Other methods have utilized look-up table (LUT) inversion (e.g., Weiss et al. 2000) where a range of simulated canopy reflectance patterns are generated in advanced by varying leaf and
canopy inputs across predetermined values. These simulated spectra are then compared to observations where either a single or select number of closely matching
modeled spectra, and their associated inputs, are selected as the solution to the
inversion. Bayesian RTM inversion methods have also been utilized (e.g.,
Shiklomanov et al. 2016) as a means to retrieve leaf and canopy properties as joint
posterior probability distributions through iterative sampling of the input parameter
space. The use of RTMs ranges from retrieval of vegetation functional and structural
traits to the characterization of landscape functional diversity (Kattenborn et al.
2017; Kattenborn et al. 2019).
3.3 Important Considerations, Caveats, and Future
Opportunities
3.3.1 Field Sampling and Scaling Considerations
There are several important considerations and best practices when developing
algorithms for the remote estimation of plant traits (see Schweiger, Chap. 15). We
will only briefly touch on these here. A key first step is to consider the scope of the
research and area of interest, focusing specifically on considerations such as local
climate conditions, terrain, vegetation, and canopy access. Specifically, the spatial
locations, site, and canopy access (e.g., is it possible to reach canopy foliage?);
vegetation composition and canopy architecture; timing of collection; and methods
for sample retrieval are key to identify prior to field campaigns in order to maximize
the utility of the field samples for conversion of RS signatures to accurate trait maps.
Furthermore, it may be important to consider what approach may be best to characterize the vegetation canopy architecture and/or composition to facilitate scaling of
each trait to the pixel or plot scale (e.g., using basal area, LAI). This may strongly
depend on the dominant vegetation types, where more open canopies may require a
different approach to a closed canopy, or on the spatial resolution of the imagery.
Observational data range is a primary consideration (see Schweiger, Chap. 15), and
sample locations should be chosen to cover the range of canopy types and vegetation communities that will fall within the RS observations. The timing of the field
3 Scaling Functional Traits from Leaves to Canopies