Risk Assessment of Indicator Bird Species Under Climate and Land-Cover Change Using Ensemble Species Distribution Models

Bird populations reflect ecosystem health and serve as early warning signals for ecological changes. In Luxembourg, rising temperatures and increasing habitat loss and fragmentation threaten these indicator species and their functions. To monitor early ecosystem degradation, we selected six bird species representing forest, farmland, or open land habitats. However, with multiple interacting drivers of change, a single metric cannot assess future vulnerability. Starting with a risk-based approach allows evaluation of both climatic changes and whether habitat availability and protection amplify or reduce species exposure. While many studies have examined climate-driven range shifts, the combined effects of land-cover constraints and species-level risk assessments remain underexplored, despite their importance for conservation planning. Most existing studies describe future suitability but do not translate projections into measurable risk for decision-makers. Integrating a risk index allows us to identify which species face the greatest ecological threats and prioritise conservation actions accordingly.

To address these research gaps, we modelled current and future ranges for the six key bird species using an ensemble of machine-learning algorithms (Random Forest, XGBoost, LightGBM, CatBoost, and SVM). These models included historical climate and high-resolution land-cover data, ensuring both climate and landscape structure were considered. For future projections, we applied the SSP2-4.5 and SSP5-8.5 scenarios and filtered results with a 30 m habitat-specific mask to ensure ecological validity. Linking our predictions for 2021–2040 and 2041–2060 to the protected-area network, we incorporated these into a risk index. This index combined the percentage of habitat lost, climate stress under each scenario, and the proportion of remaining suitable area within protected areas. The resulting risk score is intended to guide national biodiversity prioritisation by highlighting species at the highest risk, especially those with limited conservation protection.

Our results demonstrate significant mid-century habitat losses under SSP5-8.5. For example, Lanius collurio, Oenanthe oenanthe, and Vanellus vanellus lose over 90% of their suitable area, while Ficedula hypoleuca and Accipiter gentilis lose 70–80%. Lullula arborea loses 47.9%. While forest refugia remain protected, open-land and ecotone species are highly vulnerable. These findings show that integrating high-resolution data, land-cover constraints, and risk assessments effectively targets climate-smart conservation efforts and advances BIOFIN-EU goals.