Massaciuccoli Lake Basin: risk hotspots of forest and grassland habitat degradation and biodiversity loss

This cartographic shows the spatial representation of the ecosystem risk hotspots of forest and grassland habitat degradation and biodiversity loss in the wetland of the Massaciuccoli Lake basin (Tuscany, Italy), together with the main environmental and anthropogenic stressors contributing to the risk. The risk hotspots are derived from the Multi K-means applied on Variational Autoencoder (MKMonVAE) index, a spatial analysis methodology designed to capture complex, non-linear relationships among environmental, ecological, and anthropogenic variables. The MKMonVAE method combines the deep learning capabilities of the Variational Autoencoder (VAE)—which extracts latent ecological structures from multivariate datasets—with the Multi K-means clustering algorithm, which identifies spatial clusters of similar ecological conditions and degradation risk. This map visually highlights the areas of higher vulnerability, where multiple pressures converge and the likelihood of ecosystem degradation is greater. The underlying environmental and socio-ecological stressors considered in the analysis include: Land imperviousness density change Tree cover density and change Grassland extent and change Land use and cover (Natura 2000 classes: 1120 Industrial, commercial and military units; 1310 Mineral extraction, dump and construction sites; 8120 Highly modified watercourses and canals) Plant phenology index (total productivity) Temperature (average absolute change) Potential evapotranspiration (relative change) Precipitation (cumulative relative change) Number of species (cumulative) Land use and cover change The map was produced within a QGIS project adopting the WGS84 (EPSG:4326) coordinate reference system and represents a derivative cartographic output of a broader comparative framework that included three methodologies (Overlap, Multi K-means, and MKMonVAE). This version focuses exclusively on the MKMonVAE model, which demonstrated a strong capacity to synthesize spatial variability and reveal ecological vulnerability patterns. All source datasets, methodological details, and the comparative analysis framework are available in the related dataset published on Zenodo (see Related Works section).