A Bayesian network approach for selecting translocation sites for endangered island birds

Translocation has become increasingly important for conserving island species. Limited tools are available for guiding the selection of translocation sites, however, particularly when establishing rescue populations outside of historic ranges. We developed a Bayesian network model framework for translocation site selection for island birds. The model consisted of four primary components including ecological requirements for survival, anthropogenic threats at the population establishment site, effects the translocated species may have on native species, and operational support associated with the translocation process and ongoing management. We then used the model to identify potential sites for the establishment of a wild population of Guam Micronesian kingfishers (Todiramphus cinnamominus cinnamominus) on an island outside the bird’s historic range. Conditional probabilities that guided model evaluations were allocated using information from the literature, expert opinions, and a training set of islands outside the region under consideration for releases. The model was used to evaluate 239 islands where a translocation population of Micronesian kingfishers could be established. Five islands, all in the Federated States of Micronesia, were identified as being suitable for assisted colonization, including Kosrae, Yap, Faichuk, Weno and Fefan. Sensitivity analysis showed a correspondence between model variables and island characteristics indicated by the literature as being the most important for successful translocation. We found the Bayesian network model to be a useful tool for translocation site selection despite limited information on the natural history of the Guam Micronesian kingfisher and the factors that impact the success of translocations.

► We developed a Bayesian network model for avian translocation site selection.
► The Guam Micronesian kingfisher was used as an example of model application.
► Five of 239 islands were identified as suitable for kingfisher assisted colonization.
► Model structure was drawn from experts and through Bayesian learning.