Worldwide, cetacean species have started to be protected, but they are still very vulnerable to accidental damage from an expanding range of human activities at sea. To properly manage these potential threats we need a detailed understanding of the seasonal distributions of these highly mobile populations. To achieve this goal, a growing effort has been underway to develop species distribution models (SDMs)that correctly describe and predict preferred species areas. However, accuracy is not always easy to achieve when physical barriers, such as islands, are present. Indeed, SDMs assume, if only implicitly, that the spatial effect is stationary, and that correlation is only dependent on the distance between observations and not on the direction or a spatial coordinates. The application of stationary SDMs in these cases could lead to incorrect predictions and, consequently, to uninformed decision making. In this study, we identify vulnerable habitats for the bottlenose dolphin in the Archipelago de La Maddalena, Northern Sardinia (Italy)using Bayesian hierarchical SDMs that account for the physical barriers issue and provide a full specification of the associated uncertainty. The approach we propose constitutes a major step forward in the understanding of cetacean species in many ecosystems where physical, geographical and topographical barriers are present.