Enhancing shallow water habitat mapping with high-resolution pansharpened PRISMA imagery data

Macrophytes are crucial elements in freshwater ecosystems and ecological processes, by supporting biodiversity, stabilizing sediments, and improving water quality. Nowadays, these ecosystems are threatened by environmental changes and the increase in anthropic pressures; moreover, their spatial distribution is often highly fragmented. Considering their importance, monitoring macrophytes’ spatio-temporal variability with non-invasive observations is fundamental. Hyperspectral remote sensing offers a valuable tool for mapping aquatic vegetation and benthic substrates accurately, although its application is often limited by spatial resolution, especially in environments characterized by patch distribution and mixed pixels. To address these limits, this study exploits the PRISMA panchromatic band for a PRISMA image acquired on 09/07/2024 on Lake Garda (Italy), combining high spectral resolution (66 band in the visible-near-infrared range) with enhanced spatial resolution (5 m). The pansharpening procedure is applied to PRISMA Level 2D images using a deep learning model trained through both supervised and unsupervised learning techniques. This study focuses on the optically shallow waters of the Sirmione Peninsula (southern portion of Lake Garda), characterized by the presence of macrophytes coverage, with the goal of classifying the benthic substrate into seven classes, and retrieving bottom depth. In addition to the panchromatic product with a spatial resolution of 5 m, aggregated products with resolutions of 10 m and 30 m were also generated. The results showed overall accuracies of 80%, 68%, and 48% for resolutions ranging from 5 m to 30 m in the classification of the benthic substrate. Similar results were obtained in the validation of bottom depths when compared to data collected by the Institute of the Italian Navy (IIM) and in situ measurements acquired during the field campaign (r = 0.82 for 5 m data, r = 0.78 for 10 m data, r = 0.75 for 30 m data). The outcomes enable more accurate monitoring of spatial-temporal dynamics in freshwater ecosystems and highlight the potential of pansharpened hyperspectral products to enhance the mapping and classification of benthic substrates, with higher accuracy in detecting small patches.