Delineating relevant local populations of widely distributed species is a common challenge in conservation ecology. Caribou and reindeer (Rangifer tarandus) are in general decline throughout their global range, despite ongoing conservation efforts. In Canada, recovery actions for the threatened boreal population of woodland caribou (Rangifer tarandus caribou) are stratified by ‘local population units’ (LPUs) on ranges distributed across 2.4 × 10 km2 of the species’ geographic range. To estimate local population dynamics, LPUs are assumed to be geographically closed, though supporting evidence varies widely. We assembled an exceptionally large database of GPS telemetry locations (891,306 telemetry days, 1998–2020) from 1,586 adult female caribou across the 19 northwesternmost LPUs. We generated a many-to-many Gaussian Bayesian Network to identify candidate local populations at range-level extents, as well as subpopulations, termed ‘communities’ in network analysis. We detected local population boundaries that in some cases were consistent with accepted LPUs and consistent with the assumption of geographic closure. In other cases, local population boundaries did not map well to currently delineated LPUs. Several communities at smaller spatial extents were consistent with expert and local knowledge of caribou movements and support recovery planning and actions “stepped down” from entire ranges. Evidence consistent with population fragmentation was confirmed along the southern and southwestern boundaries of the species’ geographic range within the study area, as were more continuous distributions confirmed to the north. We suggest that network analysis can help to inform conservation planning for boreal caribou and other wide-ranging species that would benefit from data-driven characterizations of multiscale population spatial structure.