Mapping urban waterbird habitats using machine learning and citizen science: A multi-scale analysis in Brussels,...

This study integrates citizen science (eBird) data, which provides the backbone of species with remote sensing variables to explore the spatial distribution of waterbird habitats in the urban region of Brussels. Machine learning, including XgBoost (XgB), Random Forest (RF), Extra Trees (ET), LightGBM (LGB), and CatBoost (CB), was applied to assess the influence of environmental factors at 5 scales (5, 10, 30, 50, and 100 m resolution). We found that landcover factors as highly correlated with species habitat suitability, with all models exhibiting high predictive accuracy. Notably, the 10 m resolution data combined with XgB and RF models demonstrated robust performance, making them promising tools for spatial modeling of waterbird habitats in urban or large-scale regions. Additionally, the inclusion of the impervious density map (IDM) improved prediction accuracy by reducing over estimations in highly urbanized areas. The study also highlighted the significant role of bio-climatic factors such as Bio 14 (precipitation seasonality), Bio 17 (precipitation of the wettest quarter), and Bio 18 (precipitation of the warmest quarter), as well as human activities, particularly NP (noise pressure), in shaping waterbird habitats. These results can directly guide urban biodiversity management in Brussels and highlight the necessity of developing targeted conservation strategies, which protect key habitats, enhance habitat connectivity, and mitigate the human-induced environmental stresses to ensure the long-term habitat sustainability.