9. THE CONTRIBUTION OF REMOTE SENSING
TECHNOLOGIES AND ALGORITHMS TO LAND SURFACE
PROCESSES STUDIES
73
The ultimate objectives of remote sensing data interpretation are to
extract reliable and accurate information on the state of the system being
observed from the data, and, simultaneously, to reduce or eliminate all
undesirable variations present in the original data. These unwanted
variations can generally be attributed to the sensor itself (calibration) or to
the observation conditions (anisotropy effects), rather than intrinsic changes
in the target of interest (Verstraete et al. 1996). The quality and reliability of
this interpretation phase will directly affect the accuracy, sometimes even the
feasibility, of all products generated downstream by LSP models from these
data. The immediate benefit of a high performance interpretation scheme is
access to a large number of highly accurate and reliable end products.
The accurate and reliable estimate of Category 1 variables requires an
adequate sampling with respect to the spectral (visible, near-infrared and
mid-infrared) domain of interest and the angular domain of integration
(hemisphere). The spectral sampling is needed because the radiative
properties of land surfaces are changing significantly from visible
to near-infrared
The absorption of light in the visible
domain by plant leaves provides the energy required by the photosynthetic
activity and permits plant growth. On the other hand, light is very efficiently
scattered away in the near-infrared region, most probably as a result of
adaptation since the amount of energy per photon is not sufficient for the
synthesis of organic molecules in that spectral range (Gates 1980). An
adequate angular sampling is required to guarantee a sufficiently accurate
and reliable atmospheric decontamination (Martonchik et al. 1998a), and is
needed anyway to estimate surface albedo values. In fact the latter quantities
can be defined in various ways, directional hemispherical reflectance or
bihemispherical reflectance (Nicodemus et al. 1977) but they always
correspond to quantities integrated angularly over an hemisphere.
Obviously, the poorer the sampling, the more it is necessary to formulate
hypotheses and/or acquire data from additional independent sources. For
instance, without the angular sampling capability, simplifying assumptions
to convert radiances (bi-directional reflectance factors) into fluxes (albedos)
cannot be avoided. These assumptions can consist in using a Lambertian
hypothesis or in adding some a priori knowledge on the surface under study.
However the Lambertian hypothesis is not accurate enough (e.g., Deering et
al. 1995); and a priori knowledge on the typical anisotropy of land surfaces
is quite limited on a global scale because these features have not been
documented yet.
In the case of Category 2 variables, an extensive sampling in both the
spectral and angular domains is also a prerequisite to constrain the inversion
TECHNOLOGIES AND ALGORITHMS TO LAND SURFACE
PROCESSES STUDIES
73
The ultimate objectives of remote sensing data interpretation are to
extract reliable and accurate information on the state of the system being
observed from the data, and, simultaneously, to reduce or eliminate all
undesirable variations present in the original data. These unwanted
variations can generally be attributed to the sensor itself (calibration) or to
the observation conditions (anisotropy effects), rather than intrinsic changes
in the target of interest (Verstraete et al. 1996). The quality and reliability of
this interpretation phase will directly affect the accuracy, sometimes even the
feasibility, of all products generated downstream by LSP models from these
data. The immediate benefit of a high performance interpretation scheme is
access to a large number of highly accurate and reliable end products.
The accurate and reliable estimate of Category 1 variables requires an
adequate sampling with respect to the spectral (visible, near-infrared and
mid-infrared) domain of interest and the angular domain of integration
(hemisphere). The spectral sampling is needed because the radiative
properties of land surfaces are changing significantly from visible
to near-infrared
The absorption of light in the visible
domain by plant leaves provides the energy required by the photosynthetic
activity and permits plant growth. On the other hand, light is very efficiently
scattered away in the near-infrared region, most probably as a result of
adaptation since the amount of energy per photon is not sufficient for the
synthesis of organic molecules in that spectral range (Gates 1980). An
adequate angular sampling is required to guarantee a sufficiently accurate
and reliable atmospheric decontamination (Martonchik et al. 1998a), and is
needed anyway to estimate surface albedo values. In fact the latter quantities
can be defined in various ways, directional hemispherical reflectance or
bihemispherical reflectance (Nicodemus et al. 1977) but they always
correspond to quantities integrated angularly over an hemisphere.
Obviously, the poorer the sampling, the more it is necessary to formulate
hypotheses and/or acquire data from additional independent sources. For
instance, without the angular sampling capability, simplifying assumptions
to convert radiances (bi-directional reflectance factors) into fluxes (albedos)
cannot be avoided. These assumptions can consist in using a Lambertian
hypothesis or in adding some a priori knowledge on the surface under study.
However the Lambertian hypothesis is not accurate enough (e.g., Deering et
al. 1995); and a priori knowledge on the typical anisotropy of land surfaces
is quite limited on a global scale because these features have not been
documented yet.
In the case of Category 2 variables, an extensive sampling in both the
spectral and angular domains is also a prerequisite to constrain the inversion