Classifying plant communities and mapping their potential distribution using very large, heterogeneous, georeferenced datasets: the case study of the forest vegetation in the Italian peninsula.

In recent years, the scientific community has made significant efforts in order to create geo­referenced vegetation databases (Global Index of Vegetation Plot Databases). These databases are useful to carry out assessments of the state of bio­ diversity, conservation of species and habitats, dispersion of alien species and as a support for the planning of land and protected areas. In this context, Italy has given birth to its own project called VegItaly, coordinated by the Italian Society of Vegetation Science and conducted by several research groups, in order to develop a uniform database for the entire country. Within this framework we have created a georeferenced database of phytosociological relevés of forest vegetation. This work was conducted in three main steps: digitalizing published phytosociological relevés, georeferencing them using indications on their location and topographical features, and updating species nomenclature according to the more recent checklist. In this way, we obtained a database of 5593 georeferenced relevés for insular and penin­ sular Italy. This database was then used to identify and map the main forest com­ munities in the study area. For the classification process we used a finite mixture model that is particularly suitable to this purpose since it is able to integrate in the classificatory process species composition, environmental variables and the spatial distribution of the relevés. The application of this method led to the identification of 20 forest communities that represent the remarkable biological diversity in pen­ insular and insular Italy. This variability is characterized by the presence of central temperate European communities, sub­Mediterranean ones with marked Balkan influences and typically Mediterranean ones. The final step was mapping the poten­ tial distribution of these forest communities using only the environmental variables and validating the map by means of the georeferenced relevés. Results indicate the sub­Mediterranean forest communities being characterized by a lower classification accuracy with respect to the other two macro­typologies. Despite the intrinsic limi­ tations of natural potential vegetation mapping, we believe that our approach can produce maps useful for small­scale landscape planning and nature conservation activities, being based on a replicable and standardized procedure that could be ap­ plied when georeferenced relevés are available.