Object-based Classification of Spot and Aster Data Complemented with Data Derived from Modis Vegetation Indices Time Series in a Mediterranean Test-site

A multi-scale object-based classification was carried out using data from three different sensors to map classes of interest in the framework of the EPISTIS project. This project aims to highlight the spatio-temporal patterns that underlie the epidemiology of certain diseases and more particularly of bluetongue in this case-study. A SPOT5 10m XS image of Sardinia taken in the springtime was segmented and the land-cover/land-use classes that are the most easily discriminated were mapped using a thresholding approach. Subsequently, a DEM was used as ancillary data to map the riparian vegetation. The remaining vegetation classes were then mapped using a nearest-neighbour algorithm. ASTER features, notably derived from the SWIR bands, were used in addition to SPOT in the feature space to improve vegetation discrimination. Images taken in the springtime allow for a good discrimination between semi-natural vegetation and arable land, which was the initial objective. However, project developments implied further discrimination within the arable land. Due to their spectral similarity at this resolution in a patchy Mediterranean landscape, a number of classes could not be sufficiently well classified even using additional textural and contextual features. Therefore, data derived from MODIS vegetation indices time series were included in the classification process so as to account for the vegetation dynamics and improve the classification results. * Corresponding author.