Habitat selection is context dependent and varies across space and time. Therefore, habitat selection models built from declining or highly disturbed populations may not accurately represent the optimal behavior of species and would perform poorly predicting across space and time. Here, we performed an integrated step selection analysis to predict caribou habitat use in: 1) each of five jurisdictions in the western boreal forest of Canada and 2) one global model. We found the global model had better predictive performance across space than jurisdictional models and also performed well when projected across time. Furthermore, the difference in intensity of selection between jurisdictional models and the global model provided a metric of deviation from “optimal” habitat selection that corresponded with known herd population statuses. This result shows that by working together, jurisdictions can build better predictive models of both habitat and population trends to better inform strategic management.