Elaboration of multispectral models for heterogeneous media: Application to the LAI

Most of the models used in remote sensing link a ground parameter to one or more radiometric measurements. These models are defined and validated at local scale considering homogeneous surfaces; however, the application of these models at the satellite scale concerns pixels, which are always heterogeneous. The aim of this work is to propose a method, called spatialized method, which reduces the errors introduced by a change of scale. This method, based on the general results of Raffy (1992) requires the understanding of the models validated on the homogeneous surfaces constituting the pixel and the knowledge of the local spectral domain. The bounds of all possible actual in situ values are given on each pixel. A spatialized model is proposed. An application to the LAI/NDVI model is given with a simulated landscape, in which deciduous trees, conifers, bare soil and water are represented. This new method, valid for any models, surfaces and heterogeneities, appears to be accurate in the large majority of cases.