Spatial analysis of plant diversity and habitats in Italian coastal dune ecosystems to identify conservation priorities

Coastal sand dune ecosystems support unique plant diversity and provide essential ecosystem services, yet they are among the most threatened ecosystems worldwide due to pressures from urbanization, tourism, invasive alien species and coastal erosion. Thus, effective prioritization of conservation efforts and robust monitoring methods are urgently needed. This thesis investigates patterns of plant diversity and habitat distribution on coastal dunes by integrating different approaches, from the analysis of field-collected vegetation data to remote sensing techniques, with the goal of supporting coastal dune conservation. In the first part of the thesis (Chapter 1), I aimed to identify conservation priority hotspots in relation to an existing network of protected areas, focusing on the coastal dunes of Tuscany. The analysis of 𝛼-, 𝛽-, and 𝛾-diversity based on field-collected plant community data revealed compositionally unique sites as priority sites for conservation. However, they were only partially included within protected areas. The second part of the thesis examined the potential of remote sensing techniques for coastal dune habitat monitoring. Although these techniques provide broad spatial and temporal coverage, their application to coastal dunes is challenged by the small and fragmented nature of habitat patches relative to the spatial resolution of available imagery. To address this challenge, I first aimed to provide an effective technique for habitat monitoring from satellite data, by testing fuzzy approaches to image classification (Chapter 2). These approaches, which assign pixels probabilities of belonging to multiple habitats, provided a more realistic representation of vegetation patterns than traditional crisp approaches. Subsequently, I explored the use of Convolutional Neural Networks (CNNs), a promising technique that leverages both spectral and spatial information contained in images but still has limited application in habitat mapping. In a pilot study (Chapter 3), I evaluated CNN performance using spectral datasets with varying spatial resolutions, including Unmanned Aerial Vehicles (UAV), airborne, Google Earth and WorldView-3 imagery. High spatial resolution proved crucial for accurate habitat mapping, while additional spectral bands were beneficial only for coarser data. Finally, I extended the application of CNNs to multiple dune systems (Chapter 4). While UAV-based models showed promising generalization between two pilot sites, the broad-scale application of CNNs to a large set of sites along the Italian coastline showed that the high heterogeneity both in vegetation and imagery can limit transferability. In conclusion, the integration of approaches based on field data and remote sensing proved valuable for analyzing plant diversity patterns on coastal dunes, identifying conservation priorities and providing effective monitoring methods. Future progress toward broad-scale assessments will require further integration of multi-source, multi-scale and multi-temporal datasets with advanced methodologies, to support the conservation of these fragile ecosystems.