56
advances in computing capacities and newer optimal estimation (OE) approaches
for radiative transfer retrieval of atmospheric parameters are poised to transform
atmospheric correction in the 2020s (Thompson et al. 2018).
Following atmospheric correction, scene-dependent corrections are often
required, including corrections for different illumination and reflectance due to suntarget- sensor geometry, i.e., the bidirectional reflectance distribution function
(BRDF). Current methods to correct for across-track (and along-track) illumination
variation account for differences in vegetation structure and density, either through
continuous functions (Schläpfer et al. 2015; Weyermann et al. 2015) or using
land- cover stratification (Jensen et al. 2018). However, BRDF corrections are also
rapidly changing and likely will be improved by new OE methods. As well, methods
requiring land cover stratification are generally limited to local studies, whereas
broad-scale implementation across biomes and through time will be most stable as
long as scene-specific stratification is not required.
In addition to BRDF, corrections for topographic illumination are required
(Singh et al. 2015). However, such corrections can result in poor performance for
highly shaded slopes; they enhance noise on shaded slopes while suppressing signal
on illuminated slopes. In addition, differential illumination may still remain in
images due to multiple sensor artifacts as well as effects of vegetation structure
(Knyazikhin et al. 2013). These effects can be effectively addressed using vector
normalization (Feilhauer et al. 2010; Serbin et al. 2015) or continuum removal (e.g.,
Dahlin et al. 2013). Such approaches largely address structure-induced reflectance
effects of broadleaf and graminoid canopies, with minor variances remaining in
conifers. The residual effect of canopy structure on trait mapping largely relates to
an inability to fully account for within-canopy scattering of diffuse radiation,
especially in conifer forests.
Finally, when integrating data from multiple sources to map canopy traits, users
must address wavelength calibrations. Different sensors may have different band
centers, and these may change (on airborne devices) as they are recalibrated from
time to time. This requires image resampling, which is data and processing intensive and—to be done precisely—requires good knowledge of spectral response
functions or model recalibration to new wavelengths.
3.2 Linking Plant Functional Traits to Remote Sensing
Signatures
All materials interact with light energy in different and characteristic ways. With
respect to terrestrial ecosystems, spectroscopic RS leverages spectroradiometers,
which measure the intensity of light energy reflected from or transmitted through
leaves, plant canopies, or other materials (e.g., wood, soil, Fig. 3.3). The absorbing
and scattering properties of the individual elements (e.g., leaves, twigs, stems)
within the canopy or surface (soil) are defined by their physical and 3-D structure as
well as chemical constituents or bonds (Figs. 3.2 and 3.3), which drives the variS. P. Serbin and P. A. Townsend
advances in computing capacities and newer optimal estimation (OE) approaches
for radiative transfer retrieval of atmospheric parameters are poised to transform
atmospheric correction in the 2020s (Thompson et al. 2018).
Following atmospheric correction, scene-dependent corrections are often
required, including corrections for different illumination and reflectance due to suntarget- sensor geometry, i.e., the bidirectional reflectance distribution function
(BRDF). Current methods to correct for across-track (and along-track) illumination
variation account for differences in vegetation structure and density, either through
continuous functions (Schläpfer et al. 2015; Weyermann et al. 2015) or using
land- cover stratification (Jensen et al. 2018). However, BRDF corrections are also
rapidly changing and likely will be improved by new OE methods. As well, methods
requiring land cover stratification are generally limited to local studies, whereas
broad-scale implementation across biomes and through time will be most stable as
long as scene-specific stratification is not required.
In addition to BRDF, corrections for topographic illumination are required
(Singh et al. 2015). However, such corrections can result in poor performance for
highly shaded slopes; they enhance noise on shaded slopes while suppressing signal
on illuminated slopes. In addition, differential illumination may still remain in
images due to multiple sensor artifacts as well as effects of vegetation structure
(Knyazikhin et al. 2013). These effects can be effectively addressed using vector
normalization (Feilhauer et al. 2010; Serbin et al. 2015) or continuum removal (e.g.,
Dahlin et al. 2013). Such approaches largely address structure-induced reflectance
effects of broadleaf and graminoid canopies, with minor variances remaining in
conifers. The residual effect of canopy structure on trait mapping largely relates to
an inability to fully account for within-canopy scattering of diffuse radiation,
especially in conifer forests.
Finally, when integrating data from multiple sources to map canopy traits, users
must address wavelength calibrations. Different sensors may have different band
centers, and these may change (on airborne devices) as they are recalibrated from
time to time. This requires image resampling, which is data and processing intensive and—to be done precisely—requires good knowledge of spectral response
functions or model recalibration to new wavelengths.
3.2 Linking Plant Functional Traits to Remote Sensing
Signatures
All materials interact with light energy in different and characteristic ways. With
respect to terrestrial ecosystems, spectroscopic RS leverages spectroradiometers,
which measure the intensity of light energy reflected from or transmitted through
leaves, plant canopies, or other materials (e.g., wood, soil, Fig. 3.3). The absorbing
and scattering properties of the individual elements (e.g., leaves, twigs, stems)
within the canopy or surface (soil) are defined by their physical and 3-D structure as
well as chemical constituents or bonds (Figs. 3.2 and 3.3), which drives the variS. P. Serbin and P. A. Townsend