Functional approaches for predicting land use with the temporal evolution of coarse resolution remote sensing data

The sensor SPOT 4/Vegetation gives every day satellite images of Europe with medium spatial resolution, each pixel corresponding to an area of 1 km x 1 km. Such data are useful to characterize the development of the vegetation at a large scale. The pixels, named 'mixed' pixels, aggregate information of different crops and thus different themes of interest (wheat, corn, forest,...). We aim at estimating the land use when observing the temporal evolution of reflectances of mixed pixels. The statistical problem is to predict proportions with longitudinal covariates. We compared two functional approaches. The first relies on varying-time regression models and the second is an extension of the multilogit model for functional data. The comparison is achieved on a small area on which the land use is known. Satellite data were collected between March and August 1998. The functional multilogit model gives better predictions and the use of composite vegetation index is more efficient.