66
small compared to the full canopy medium, with no self-shading, and transmittance
is a function of optical properties and leaf area index (LAI). Additional canopy
parameters were added, including the hot-spot and canopy clumping to describe
sun-sensor illumination effects and the inhomogeneity of the canopy elements
(Kuusk 2018). Early SAIL models also fall into this classification (e.g., Verhoef
1984). On the other hand, heterogenous canopy RTM models, including 3-D models, address the fact that vegetation canopies are heterogenous (e.g., gaps between
crowns, spatial structure, differing canopy architectures) but range widely in their
complexity and implementations. These models provide enhanced detail in the
modeling of vegetation canopies but are necessarily more complex. Often these
models require additional information to model vegetation “scenes,” which can
include information on tree crown shape, stem location, and other properties (e.g.,
hot spot, clumping) in addition to leaf optical properties, sun-sensor geometry, and
LAI. These models range from 3-D Monte Carlo ray-tracing models, such as
FLIGHT (North 1996) and FLiES (Kobayashi and Iwabuchi 2008), to analytical
and hybrid approaches using a variety of canopy structure schemes including geometric optical (GO) representation of individual plants where tree placement follows a statistical distribution and leaf and stem scattering elements are homogenously
distributed (e.g., Kuusk and Nilson 2000; Nilson et al. 2003). For example, multiple
stream, including four-stream, two-layer models often utilize simplifying assumptions, to model canopies as homogenous and continuous (i.e., “slab canopies”), but
which are composed of a large number of small scattering elements (leaves, sometimes leaves and stems) with arbitrary inclination angles (e.g., 4SAIL2, Verhoef and
Bach 2007). The scattering elements and the soil can be prescribed with specific
optical properties using observed data or based on a leaf RTM, such as PROSPECT
(Jacquemoud et al. 2009). In addition, some models can divide complex scenes into
smaller cells to perform the radiative transfer calculations (e.g., DART,
Gastellu- Etchegorry et al. 2015) where the level of simulation detail is based on the
size of the cells and the degree of detail built into the model scene components. See
the review by Kuusk (2018) for more details regarding canopy RTMs and their
design, diversity, assumptions, and approaches.
The use of RTMs allows for the estimation of leaf and canopy traits using simulated canopy reflectance, without some of the limitations or challenges of empirical
approaches (3.3.1), such as the requirement of field sampling, scaling leaf traits to
the canopy, and other issues such as the timing of field and imagery collections.
Furthermore, RTMs can provide a more mechanistic connection between traits and
reflectance allowing for potentially broader application than empirical approaches
in areas were ground sampling may be sparse (e.g., remote regions such as the
Arctic or the tropics). In addition, RTMs provide the opportunity to prototype
inversion approaches across a range of remote sensing platforms and evaluate the
trade- offs between different sensor designs, spectral resolutions, and temporal
coverage (Shiklomanov et al. 2016), enabling the development of cross-platform
retrieval algorithms.
Depending on the application, and RTM complexity, inversion can be conducted
at the pixel or larger patch scales (i.e., collections of relatively homogenous areas of
S. P. Serbin and P. A. Townsend
small compared to the full canopy medium, with no self-shading, and transmittance
is a function of optical properties and leaf area index (LAI). Additional canopy
parameters were added, including the hot-spot and canopy clumping to describe
sun-sensor illumination effects and the inhomogeneity of the canopy elements
(Kuusk 2018). Early SAIL models also fall into this classification (e.g., Verhoef
1984). On the other hand, heterogenous canopy RTM models, including 3-D models, address the fact that vegetation canopies are heterogenous (e.g., gaps between
crowns, spatial structure, differing canopy architectures) but range widely in their
complexity and implementations. These models provide enhanced detail in the
modeling of vegetation canopies but are necessarily more complex. Often these
models require additional information to model vegetation “scenes,” which can
include information on tree crown shape, stem location, and other properties (e.g.,
hot spot, clumping) in addition to leaf optical properties, sun-sensor geometry, and
LAI. These models range from 3-D Monte Carlo ray-tracing models, such as
FLIGHT (North 1996) and FLiES (Kobayashi and Iwabuchi 2008), to analytical
and hybrid approaches using a variety of canopy structure schemes including geometric optical (GO) representation of individual plants where tree placement follows a statistical distribution and leaf and stem scattering elements are homogenously
distributed (e.g., Kuusk and Nilson 2000; Nilson et al. 2003). For example, multiple
stream, including four-stream, two-layer models often utilize simplifying assumptions, to model canopies as homogenous and continuous (i.e., “slab canopies”), but
which are composed of a large number of small scattering elements (leaves, sometimes leaves and stems) with arbitrary inclination angles (e.g., 4SAIL2, Verhoef and
Bach 2007). The scattering elements and the soil can be prescribed with specific
optical properties using observed data or based on a leaf RTM, such as PROSPECT
(Jacquemoud et al. 2009). In addition, some models can divide complex scenes into
smaller cells to perform the radiative transfer calculations (e.g., DART,
Gastellu- Etchegorry et al. 2015) where the level of simulation detail is based on the
size of the cells and the degree of detail built into the model scene components. See
the review by Kuusk (2018) for more details regarding canopy RTMs and their
design, diversity, assumptions, and approaches.
The use of RTMs allows for the estimation of leaf and canopy traits using simulated canopy reflectance, without some of the limitations or challenges of empirical
approaches (3.3.1), such as the requirement of field sampling, scaling leaf traits to
the canopy, and other issues such as the timing of field and imagery collections.
Furthermore, RTMs can provide a more mechanistic connection between traits and
reflectance allowing for potentially broader application than empirical approaches
in areas were ground sampling may be sparse (e.g., remote regions such as the
Arctic or the tropics). In addition, RTMs provide the opportunity to prototype
inversion approaches across a range of remote sensing platforms and evaluate the
trade- offs between different sensor designs, spectral resolutions, and temporal
coverage (Shiklomanov et al. 2016), enabling the development of cross-platform
retrieval algorithms.
Depending on the application, and RTM complexity, inversion can be conducted
at the pixel or larger patch scales (i.e., collections of relatively homogenous areas of
S. P. Serbin and P. A. Townsend