Recognition of Animals’ Affiliation with Habitat Patches Using Machine Learning Techniques

Despite efforts to select optimal habitat patches for reintroduction programs, animals released to the wild rarely stay within such designated spots. Therefore, it is important to be able to evaluate their true landscape affiliation. Wisents upon having been reintroduced to the Bieszczady Mountains, have remained in two isolated subpopulations of slightly different habitats. Using movement data obtained through radio-tracking and employing machine learning (ML) techniques (notably, eXtreme Gradient Boosting (XGBoost)) for classification purposes, we were able to obtain a precise determination of preferred habitats within their home rangeswe analyzed 31,480 location records of wisent presence in the Bieszczady Mountains during the years 2002-2021 and correlated this data on land use and land cover obtained from the CORINE Land Cover (CLC) inventory. The machine-learning algorithm XGBoost was then applied to identify the affiliation of individual wisents to habitats within the home range of the given subpopulation. The results showed very high classification performance, with a recognition accuracy of 92% in the vegetative and 96% in the winter seasons. We thus propose a new approach to recognizing the affiliation of particular individuals to their home ranges. This amay considerably improve the decision-making process in conservation and management of wildlife populations.