A novel method for multi-trajectory reconstruction based on LoMcT for avian migration in population level

Abstract

Migration is an essential ecological process, and is usually used to describe the seasonal movements between the breeding grounds and wintering areas. During migration, bird populations often disperse in groups and aggregate together due to geographical barriers, topography, seasonal climatic changes, species specific physiology, or other factors. Recording and reconstructing these diverse migratory routes are important for identifying major stopover sites as well as migration bottlenecks which may include key foraging grounds and resting areas, and ensuring high-quality habitat to provide adequate resources. However, good data including individual tracking data are only available for some regions and species large enough to carry a transmitter. Better approaches using observational data are needed to enable better understanding in less-studied regions. To reconstruct and visualize the long-distance avian migration routes with observations from the citizen-science dataset eBird, we developed an interpretive avian multi-trajectory reconstruction framework based on Level-order-Minimum-cost-Traversal (LoMcT) algorithm. This approach uses linear interpolation for missing records, spatial outlier detection for abnormal values, unsupervised clustering by density-based Mean-Shift algorithm for sub-group centroids, LoMcT algorithm based on the distances among centroids, and multi-trajectory reconstruction based on generalized additive models. We have verified the feasibility of our reconstruction method using 15 bird species, and analyzed the trends of the distribution density of birds' population during the long-distance migration cycle. Our analysis could help obtain the important gathering time points and sites in the moving process based on the multiple routes we reconstructed. These can be used in comparisons of multi-trajectory migration strategies between the transoceanic migratory birds and non-transoceanic ones, and provide the ability to understand how species are moving in the absence of individual tracking data to help target conservation better. We have demonstrated that the proposed approach is capable of reconstructing trajectories based on observational citizen-science data.