A multi-model ecosystem risk assessment methodology for the Massaciuccoli Lake basin wetland

Wetlands are among the ecosystems most exposed to degradation due to the concurrent action of anthropogenic pressures and climate-related stressors, particularly in highly anthropised landscapes. This study presents a multi-model comparative methodology for assessing and mapping the risk of forest and grassland habitat degradation and associated biodiversity loss in wetland environments. It integrates three complementary, unsupervised models applied to a harmonised geospatial dataset: (i) a deterministic cumulative overlap index, (ii) a multivariate clustering model, and (iii) a predictive, deep learning model based on a Variational Autoencoder combined with cluster analysis. The methodology is applied to the Massaciuccoli Lake basin (Tuscany, Italy), a Ramsar-listed wetland characterised by intense land-use pressure and high ecological value. Using 12 stressor variables describing ecological fragility and harmful driving forces at approximately 50 m spatial resolution, the models produced spatially explicit risk hotspot distribution maps consistent with previous localised studies. Results also highlight complementary aspects among the models: deterministic mapping identified areas of strong stressor co-occurrence; clustering revealed structurally homogeneous ecological conditions; and deep learning captured non-linear patterns and anomalous configurations associated with emerging ecosystem risks. Areas consistently classified as medium or high risk across multiple models delineated robust priority hotspots, while divergent classifications identified transitional or uncertain conditions requiring further investigation. The proposed methodology is potentially re-applicable to other wetlands to enhance the robustness and interpretability of ecosystem risk assessments. It provides a reproducible, scalable tool to support conservation planning, environmental management, and spatial decision-making in wetlands affected by complex, interacting stressors.