Pinpointing bird species vulnerability to free-ranging domestic cats using trait-based predictive models

We developed a method to assess bird species' vulnerability to predation by free-ranging domestic cats (Felis catus) using prey preference data from citizen science in Italy and the United Kingdom. By combining species traits and geographical range, we trained random forest models to predict prey preferences and identify missing prey species. Our analysis showed that including the geographical range significantly improved model accuracy and reduced prey detectability issues. Cross-validation confirmed that models trained in one country could effectively predict prey preferences in another, allowing for broader application. Shapley additive explanations values analysis revealed that small, generalist birds with a low hand-wing index and large geographical range were most likely to be preyed upon. We used these models to create vulnerability lists for United States bird species, which showed moderate overlap but high consistency with previous studies, highlighting their robustness. These results showed that this method could thus be used to improve our understanding of cat predation and inform targeted conservation strategies, with better citizen science data being crucial for further improvements.