Monitoring habitat diversity with PRISMA hyperspectral and lidar-derived data in Natura 2000 sites: Case study from a Mediterranean forest

• Hyperspectral PRISMA is used to map a complex Mediterranean Natura 2000 forest. • High thematic resolution is obtained joining hyperspectral and canopy height data. • Random Forest and PLS-DA algorithms provided best accuracy. • Hyperspectral satellite data can improve the monitoring of biodiversity. The European Natura 2000 network is composed by >50 % of forest, for about 37.5 million of hectares, hosting unvaluable biodiversity. Reporting the conservation status of the Natura 2000 network is mandatory. but the monitoring of sites is based on a variable approach among different countries; accurate spatially explicit data are scarce, and often derived by manual photo interpretation and dated surveys. The increasing climate change impacts on forests and biodiversity, especially in the Mediterranean area, calls for improved monitoring and EU-harmonized procedures. Furthermore, assessing the spatial distribution and extent of natural habitats is another urgent requirement, that can be framed into the wider concept of Essential Biodiversity Variables. Here hyperspectral PRISMA data are used, together with canopy height information from lidar, to map the ecosystem diversity of a Mediterranean Natura 2000 forest site, at very high thematic resolution. The task is not trivial, considering the presence in the study area of different Quercus spp. dominated forest types. The classification tests were conducted with different algorithms and number of classes, to detect optimal solutions. Random Forests was capable to map 14 classes (overall accuracy >80 %) after input features reduction, similarly to Partial Least Squares Discriminant Analysis that instead ingested the full dataset. Even if characterized by higher spatial resolution, models based on Sentinel 2 data provided much lower accuracy than PRISMA. Considerations about the use of this satellite hyperspectral and lidar data, in the framework of improved ecosystem monitoring, were provided. This research illustrates the potential of using hyperspectral and lidar data to assess the forest habitat diversity in the Natura 2000 network, thus supporting the adoption of innovative data and approaches, based on remote sensing, to monitor natural resources and Essential Biodiversity Variables.