Emission and propagation of light in Earth Landscapes (i.e. Radiative Transfer) constricts the functioning of the biosphere on one hand, and the remote sensing data (satellite imagery, lidar signals). The measured signals depend on many factors, both experimentals and instrumentals. The RT modeling links remote sensing data to biophysical variables that characterize the landscape, in order to refine the data analysis or to assist in the technical specification of future sensors. This thesis presents the Discrete Anisotropic Radiative Transfer (DART) model and its recent improvements. This model represents the 3-dimensional landscapes in a matrix of cells, with consideration of the atmosphere. Vegetation, which is the main study, can be represented by a 1D, 2D or detailed 3D description using triangles, or by a classical statistical approach. Improvements of the model I made are of different types : improvement of the RT in the atmosphere, implementation of LIDAR modeling with a Monte Carlo approach and implementation of an original approach to model vegetation covers with varying degrees of realism. In addition to these theoretical works, I have strongly contributed to improve the code, with in particular transformation to a C++ code, making DART more robust, efficient, and easy to use to study earth surfaces by remote sensing.
