Accuracy Assessment of Pan-Sharpened PRISMA Hyperspectral Image for Seagrass Species Composition Mapping

Seagrass species mapping is challenging due to the high similarity of reflectance spectra among healthy seagrass species. In Indonesia, a single seagrass meadow is commonly occupied by different seagrass species, which adds to the complexity of seagrass species mapping. The availability of a newly introduced PRISMA hyperspectral sensor (30 m) with an additional panchromatic band (5 m) has opened a new possibility for effective and efficient seagrass species mapping in comparison with airborne hyperspectral imaging. Therefore, assessing the accuracy of pan-sharpened PRISMA hyperspectral seagrass species composition mapping in optically shallow tropical water is necessary. Karimunjawa Islands, Indonesia, was selected as our study area. The PRISMA hyperspectral image was obtained at level 2C, at-surface reflectance geocoded level. Smoothing filter-based intensity modulation (SFIM) pan-sharpening algorithm was used to pan-sharpen the 30 m hyperspectral bands into 5 m spatial resolution. LSMA was conducted to obtain the composition of seagrass species and other benthic covers from the pan-sharpened PRISMA hyperspectral image pixels based on the pure endmembers of six tropical seagrass species, bare substrates, and macroalgae. Field seagrass species data for the accuracy assessment reference was collected using photoquadrat and phototransect techniques. Our results indicated that using LSMA for seagrass species mapping using a pan-sharpened PRISMA hyperspectral image is ineffective. They produced low accuracy due to the similarity of spectral response between seagrass species.