Integrating animal tracking datasets at a continental scale for mapping Eurasian lynx habitat
Oeser, Julian; Heurich, Marco Dietmar; Kramer-Schadt, Stephanie A.; Mattisson, Jenny; Krofel, Miha; Krojerová-Prokešová, Jarmila; Zimmermann, Fridolin; Anders, Ole; Andrén, Henrik; Bagrade, Guna; Belotti, Elisa; Breitenmoser-Würsten, Christine; Bufka, Luděk; Cerne, Rok; Drouet-Hoguet, Nolwenn; Duľa, Martin; Fuxjäger, Christian; Gomerčić, Tomislav; Jędrzejewski, Włodzimierz; Kont, Raido; Koubek, Petr; Kowalczyk, Rafał; Kusak, Josip; Kubala, Jakub; Kutal, Miroslav; Linnell, John Durrus; Molinari-Jobin, Anja; Männil, Peep; Middelhoff, Tomma Lilli; Odden, John; Okarma, Henryk; Schmidt, Krzysztof; Signer, Sven; Tám, Branislav; Vogt, Kristina; Kuemmerle, Tobias
Peer reviewed, Journal article
Published version
Permanent lenke
https://hdl.handle.net/11250/3099059Utgivelsesdato
2023Metadata
Vis full innførselSamlinger
Originalversjon
10.1111/ddi.13784Sammendrag
Aim: The increasing availability of animal tracking datasets collected across many sites provides new opportunities to move beyond local assessments to enable de-tailed and consistent habitat mapping at biogeographical scales. However, integrating wildlife datasets across large areas and study sites is challenging, as species' varying responses to different environmental contexts must be reconciled. Here, we compare approaches for large-area habitat mapping and assess available habitat for a recolo-nizing large carnivore, the Eurasian lynx (Lynx lynx).Location: Europe.Methods: We use a continental-scale animal tracking database (450 individuals from 14 study sites) to systematically assess modelling approaches, comparing (1) global strategies that pool all data for training versus building local, site-specific models and combining them, (2) different approaches for incorporating regional variation in habi-tat selection and (3) different modelling algorithms, testing nonlinear mixed effects models as well as machine-learning algorithms.Results: Testing models on training sites and simulating model transfers, global and local modelling strategies achieved overall similar predictive performance. Model performance was the highest using flexible machine-learning algorithms and when incorporating variation in habitat selection as a function of environmental variation. Our best-performing model used a weighted combination of local, site-specific habi-tat models. Our habitat maps identified large areas of suitable, but currently unoccu-pied lynx habitat, with many of the most suitable unoccupied areas located in regions that could foster connectivity between currently isolated populations.Main Conclusions: We demonstrate that global and local modelling strategies can achieve robust habitat models at the continental scale and that considering regional variation in habitat selection improves broad-scale habitat mapping. More generally, we highlight the promise of large wildlife tracking databases for large-area habitat mapping. Our maps provide the first high-resolution, yet continental assessment of lynx habitat across Europe, providing a consistent basis for conservation planning for restoring the species within its former range.