The paper addresses land cover monitoring at large scale. It investigates an approach for extracting information from remote sensing data and evaluating the possibility of doing automated analysis to detect land use changes. A multi-resolution remote sensing data process is performed to obtain curves describing the temporal reflectance profiles related to land cover behaviour. Then a time-frequency transformation is used to extract relevant features characterizing these profiles. The features are then projected on a representation space improving discrimination between different land cover types. This approach can applied in a scenario dedicated to the detection of land use changes caused by erosion.