PRISMA Hyperspectral images for Tree Cover classification with Machine Learning algorithms a comparison with Sentinel-2

Remote sensing (RS) images are fundamental for earth observation and for the analysis of land cover and land cover change providing useful information for agroforestry planning and management. Multispectral data are the most common, such as those provided by the Sentinel-2 satellite, which inherits the legacy of the Landsat satellite. More recently, images from the Italian PRISMA satellite, which provides hyperspectral images, have opened new perspectives in land analysis due to an improved spectral resolution. The study area is situated in the Lazio region (Italy), it was selected for the presence of homogeneous and extensive wooded surfaces of large forest areas and orchards.  In this study an evaluation and a comparison of the results of Tree Cover classes classification obtained using images from different sources has been carried out. The PRISMA and Sentinel 2 images have been downloaded and preprocessed for comparison. The classification based on advanced machine learning techniques was carried out and the results have been compared by evaluating the achieved accuracy metrics with the different images. The study showed that the advantages represented by the higher spectral resolution are at least partially offset by the lower spatial resolution of PRISMA images. Due to the short time since the beginning of the PRISMA mission and the limited availability of images, the study represents one of the early examples of applying the potential of the PRISMA satellite.