In this paper, we tested a relatively new version of the Partial Least Square (PLS) method called the Sparse Partial Least Square Discriminant Analysis (SPLSDA). This improved method performs variable selection and classification in a one-step procedure and has been successfully applied in the field of bioinformatics (Lê Cao et al., 2011). We are applying the method on remote sensing data for the classification of natural and semi-natural habitats in a Natura 2000 site located in Southern France. The work has been performed in the framework of the MS.MONINA FP7 project which is using the potential of GMES for the development of a multi-scale mapping service aimed at monitoring European protected habitats and species at the local, regional and continental scales.