Classer la migration à l'ère du Big Data. Est-il possible d'identifier le comportement de migration par des routines automatiques ? Performance de 3 méthodes, MigrO, MigrateR et une approche de segmentation

Lucie Debeffe (INRAE, CEFS)

21 mai 2021

Migration remains a complex phenomenon, and previous work has shown the potential inconsistencies in the classification of movement. Here we aimed at evaluating the criticalities in the uninformed, automatic identification of ungulate migration with a test-case. Specifically, we first evaluate the robustness of different routines applied to the same datasets; and second, disentangle how the robustness of classification is affected by the routine applied, or, conversely, by the definition of the biological phenomenon that is then used to parametrise such routines. A dataset of 261 trajectories from 21 populations of one species distributed at the continental scale (red deer from Euromammals/Euroredeer database: was used. We classified each trajectory into migratory and non-migratory (resident and dispersal) movements with three unsupervised procedures that rely on spatio-temporal definition of seasonal ranges. Further, we compared the automatic classification output with visual classification from ecologists and wildlife biologists with a different degree of knowledge of migratory behaviour and red deer populations. By doing this, we aimed to specifically evaluate the consistency in identification of migration.;