Data-driven agent-based modeling for animal space use prediction – an applied example for grizzly bears in the southern interior of British Columbia

Despite its increasing popularity in ecological research, the application of Agent-Based Modeling (ABM) to the spatial prediction of habitat use is relatively underexplored. With empirical support of movement algorithms derived from correlative movement analysis, ABM can produce predictive hybrid models to inform conservation decisions in complex situations. In this study, we address this gap in model application by using an integrated simulative approach that combines agent-based movement models and data-driven resource selection analysis for animal space use prediction. Our methodology leverages the strengths of mechanistic and correlative approaches to dynamically project data-driven habitat selection patterns in new environments with similar ecosystem traits. We demonstrate our approach by modelling grizzly bear Ursus arctos horribilis space use for a study area in the southern interior of British Columbia, Canada. We developed our model using relatively simple animal movement rules informed by integrated Step Selection Analysis (iSSA) and evaluated the model results using independent data. We found that the model results transcended the scale of observation and led to the emergence of realistic population-level spatial patterns. Analyses of the model predictions yielded substantial support for the model’s predictive strength with ecological implications. Specifically, the model successfully predicted independent bear occurrence locations and showed association with a critical seasonal food resource for grizzly bear, the black huckleberry Vaccinium membranaceum. We recommend this methodology as a transferable alternative to existing approaches of mobile species modeling with potential for broader application in the field of conservation.