Predictive Migratory Movement Model illustrated through a Blue-Winged Teal (Anas discors) case study

Background/Question/Methods

Migration, relative to other portions of birds' annual cycle, is poorly understood due to its innate complexity of study. Study cost related to monitoring large-scale seasonal movement of large numbers of birds is the main hurdle to advancement. The main goal of this research is to present a novel approach, illustrated via a blue-winged teal (Anas discors) case study, to predict the movement of migratory birds. The presented process incorporates not only constraints on habitat, but also constraints on direction and distance. It allows for movement predictions to be made from under- or unsampled areas across large spatial scales.

The blue-winged teal case study encompasses Canada, the contiguous United States, and Mexico. The data used to build the model were the banding and recovery locations from the USGS’ Bird Banding Laboratory database. Specifically, the recovery locations from a banding site with multiple within-30-day recoveries was used to build a core maximum entropy model. The ancillary variables are measures of not only the environment, but also direction and distance. The core models were used to project probability of movement from starting locations that did not have sufficient species data for independent predictions. The final model for an unsampled area was based on an inverse-distance weighted averaged prediction from the three nearest core models. To illustrate this approach, we selected three unsampled locations to probabilistically predict where migratory blue-wing teals would stopover. These three selected locations, despite having no or little blue-winged teal data, are assumed to have populations.

Results/Conclusions

For the blue-winged teal case study, 104 suitable locations were identified to build the core models. These locations had a range of 20 to 228 within-30-day recoveries, and all core models had AUC scores greater than 0.80. From the appropriate core models, we built three projected models from unsampled points in Ontario, Montana, and New York.

It is concluded, based on validation of model predictions, that this approach to predicting migratory movement is well-founded. Additionally, this method can be employed to predict source location specific movement for multiple species.