Co-registration of PRISMA Hyperspectral Imagery for Accurate Land Cover Classification

The precise and prompt categorization of land cover types holds significant importance in the realm of land resource planning and management, as well as in risk reduction. The utilization of hyperspectral satellite imagery, such as the imagery delivered by PRISMA, plays a vital role in analyzing environmental changes. Even though PRISMA products are distributed at Preprocessing Level 2D (radiometrically and geometrically calibrated), the images may exhibit registration errors on the order of a few hundred meters. Therefore, co-registration is a crucial preprocessing step before their utilization. This study utilized a local co-registration method based on the optical flow estimation technique to coregister the PRISMA images using Sentinel-2/Landsat 8–9 as references. The results showed that a careful selection of an appropriate reference image holds immense importance in the co-registration process, and the closer the acquisition time of the reference image is to the acquisition time of the image to be co-registered, the higher the quality of the co-registration results. By integrating cutting-edge machine learning techniques, the proposed co-registration approach further enhances the usability and accuracy of PRISMA products for land cover classification, and makes them a valuable source of information for applications in land management and thematic hazard studies, including scenarios such as flood monitoring and landslide analysis.