144
Chapter 16
Depending on the resolution of the LUT and on the accuracy of the
observations, one or more solutions can be obtained by the inverse
procedure. The output of this method is thus twofold: an ensemble of values
for the variables used by the radiative transfer model that created the LUT
and the range of variations for the values of these variables. This information
allows the production of land cover maps, together with the corresponding
level of uncertainties on the retrieved variable values.
The model used to create the LUT and the inversion algorithm itself will
be outlined. Preliminary results of the application of this technique on the
AVHRR/GVI/LASUR data set (Berthelot et al, 1997) at the global scale are
shown in the form of the maximum values of LAI retrieved for the months
of January and June 1989. The aim of this exercise is to demonstrate the
potential of such a new methodology for addressing land cover issues.
2.
OVERVIEW OF THE ALGORITHM
2.1
Modeling approach
Assessing the land cover type reduces to the qualitative description or the
quantitative characterization of the properties of the land surface. A series of
physical, chemical and biological properties are usually associated with a
given land cover type. Determining the latter therefore results in the
assignment of at least approximate values of these variables. The essence of
land cover classification on the basis of remote sensing data consists in
similarly associating certain radiative characteristics of the environment to
specific land cover types, so that the observation of the former leads to the
reliable identification of the latter. This is best achieved when the properties
of the environment are explicitly linked to the measurable observations
gathered in space, and the proposed approach thus hinges on the simulation
of top-of-atmosphere radiances typically measured by satellite sensors on the
basis of canopy variables. The set of such variables (or scenarios), together
with the resulting simulated reflectances, is then archived in a LUT, as these
computations are executed once and for all. Once this LUT is available, the
actual measurements gathered with the satellite sensors are compared with
the simulated reflectances of the LUT, and all table entries “close enough” to
the string of spectral and directional measurements are considered potential
solutions of the problem, i.e., possible descriptions of the actual
environment, as will be seen shortly. For the purpose of this paper, the LUT
was generated with the semi-discrete radiation transfer model of Gobron et
al. (1997) to represent the interaction of solar light with plant canopies. This
Chapter 16
Depending on the resolution of the LUT and on the accuracy of the
observations, one or more solutions can be obtained by the inverse
procedure. The output of this method is thus twofold: an ensemble of values
for the variables used by the radiative transfer model that created the LUT
and the range of variations for the values of these variables. This information
allows the production of land cover maps, together with the corresponding
level of uncertainties on the retrieved variable values.
The model used to create the LUT and the inversion algorithm itself will
be outlined. Preliminary results of the application of this technique on the
AVHRR/GVI/LASUR data set (Berthelot et al, 1997) at the global scale are
shown in the form of the maximum values of LAI retrieved for the months
of January and June 1989. The aim of this exercise is to demonstrate the
potential of such a new methodology for addressing land cover issues.
2.
OVERVIEW OF THE ALGORITHM
2.1
Modeling approach
Assessing the land cover type reduces to the qualitative description or the
quantitative characterization of the properties of the land surface. A series of
physical, chemical and biological properties are usually associated with a
given land cover type. Determining the latter therefore results in the
assignment of at least approximate values of these variables. The essence of
land cover classification on the basis of remote sensing data consists in
similarly associating certain radiative characteristics of the environment to
specific land cover types, so that the observation of the former leads to the
reliable identification of the latter. This is best achieved when the properties
of the environment are explicitly linked to the measurable observations
gathered in space, and the proposed approach thus hinges on the simulation
of top-of-atmosphere radiances typically measured by satellite sensors on the
basis of canopy variables. The set of such variables (or scenarios), together
with the resulting simulated reflectances, is then archived in a LUT, as these
computations are executed once and for all. Once this LUT is available, the
actual measurements gathered with the satellite sensors are compared with
the simulated reflectances of the LUT, and all table entries “close enough” to
the string of spectral and directional measurements are considered potential
solutions of the problem, i.e., possible descriptions of the actual
environment, as will be seen shortly. For the purpose of this paper, the LUT
was generated with the semi-discrete radiation transfer model of Gobron et
al. (1997) to represent the interaction of solar light with plant canopies. This