69
sciences and not discussed in detail here. On the other hand, efforts to provide product uncertainties do exist. Serbin et al. (2014) and Singh et al. (2015) illustrate how
to incorporate data and modeling uncertainties at the leaf and canopy scales in the
mapping of plant functional traits. This approach captures the uncertainties stemming from the leaf-level estimation of traits (Serbin et al. 2014) and the modeling of
plot-level spectra and trait values (Singh et al. 2015) using a similar PLSR and
uncertainty analysis approach. The result is an ensemble of PLSR models to apply
to new RS data providing mean and error metrics for every pixel in the image.
However, even approaches such as these fail to incorporate and propagate the uncertainties stemming from the atmospheric correction workflow given the challenge of
extract the information needed to enable this on a pixel-by-pixel or even a scene-byscene basis. Future work will be required to focus on capturing this information and
providing it to the end-user who conducts the trait mapping efforts.
Uncertainty in RTM approaches have generally been derived based on inversion approaches applied to imagery. For example, as described in Sect. 3.2.2.2, a
commonly used approach to the inversion of RTM simulations for the RS of functional traits is the use of LUTs. Some LUT approaches provide results based on
the “best fit” of the model inversion results to the RS observations. However, this
only provides an assessment of error where field measurements can be used to
evaluate the retrieved values. Given the challenge of equifinality in RTM
approaches, later efforts have used an ensemble of best fit results to provide a
mean and distribution of values that provide a good fit of modeled reflectance to
observed (e.g., Weiss et al. 2000; Banskota et al. 2015). Using this approach
allows for the description of pixel-level uncertainty based on the best fit ensembles; however, these need to be combined with an accuracy assessment to get a
true uncertainty of the functional trait retrievals. More recent approaches have
leveraged Bayesian inversion approaches that provide output that is not a point
estimate for each parameter but rather the joint probability distribution that
includes estimates of parameter uncertainties and covariance structure
(Shiklomanov et al. 2016). Regardless of the approach, the key is that the derived
products provide a reasonable assessment of trait uncertainty across the spatial
and temporal domain (where appropriate).
3.3.3 Current and Future Opportunities in the Use of Remote
Sensing to Characterize Functional Traits
and Biodiversity
The ability to map foliar functional traits from imaging spectroscopy greatly
expands the potential for understanding patterns of vegetation function and functional diversity both locally and broadly across biomes, especially in comparison to
the challenges of fully characterizing spatial and temporal (across seasons and
between years) variation using field data (e.g., the TRY database). With forthcoming
spaceborne sensors (see Schimel et al., Chap. 19) and continental-scale experiments
3 Scaling Functional Traits from Leaves to Canopies
sciences and not discussed in detail here. On the other hand, efforts to provide product uncertainties do exist. Serbin et al. (2014) and Singh et al. (2015) illustrate how
to incorporate data and modeling uncertainties at the leaf and canopy scales in the
mapping of plant functional traits. This approach captures the uncertainties stemming from the leaf-level estimation of traits (Serbin et al. 2014) and the modeling of
plot-level spectra and trait values (Singh et al. 2015) using a similar PLSR and
uncertainty analysis approach. The result is an ensemble of PLSR models to apply
to new RS data providing mean and error metrics for every pixel in the image.
However, even approaches such as these fail to incorporate and propagate the uncertainties stemming from the atmospheric correction workflow given the challenge of
extract the information needed to enable this on a pixel-by-pixel or even a scene-byscene basis. Future work will be required to focus on capturing this information and
providing it to the end-user who conducts the trait mapping efforts.
Uncertainty in RTM approaches have generally been derived based on inversion approaches applied to imagery. For example, as described in Sect. 3.2.2.2, a
commonly used approach to the inversion of RTM simulations for the RS of functional traits is the use of LUTs. Some LUT approaches provide results based on
the “best fit” of the model inversion results to the RS observations. However, this
only provides an assessment of error where field measurements can be used to
evaluate the retrieved values. Given the challenge of equifinality in RTM
approaches, later efforts have used an ensemble of best fit results to provide a
mean and distribution of values that provide a good fit of modeled reflectance to
observed (e.g., Weiss et al. 2000; Banskota et al. 2015). Using this approach
allows for the description of pixel-level uncertainty based on the best fit ensembles; however, these need to be combined with an accuracy assessment to get a
true uncertainty of the functional trait retrievals. More recent approaches have
leveraged Bayesian inversion approaches that provide output that is not a point
estimate for each parameter but rather the joint probability distribution that
includes estimates of parameter uncertainties and covariance structure
(Shiklomanov et al. 2016). Regardless of the approach, the key is that the derived
products provide a reasonable assessment of trait uncertainty across the spatial
and temporal domain (where appropriate).
3.3.3 Current and Future Opportunities in the Use of Remote
Sensing to Characterize Functional Traits
and Biodiversity
The ability to map foliar functional traits from imaging spectroscopy greatly
expands the potential for understanding patterns of vegetation function and functional diversity both locally and broadly across biomes, especially in comparison to
the challenges of fully characterizing spatial and temporal (across seasons and
between years) variation using field data (e.g., the TRY database). With forthcoming
spaceborne sensors (see Schimel et al., Chap. 19) and continental-scale experiments
3 Scaling Functional Traits from Leaves to Canopies