Limited monitoring resources often constrain rigorous monitoring practices to species or populations of conservation concern. Insufficient monitoring can induce a tautology as lack of monitoring resources makes it difficult to determine whether a species or population deserves additional monitoring resources. When in-situ monitoring resources are limited, remote habitat monitoring could be a useful supplementary tool, as linking parameterized species distribution models to spatially explicit time-series of environmental correlates allows iterative prediction of population change. Yet the performance of predictive forecasts or hindcasts has been difficult to evaluate. We paired contemporary field data, historic population estimates, and a remotely-sensed archive of landscape change to evaluate predictions of American marten (Martes americana) population decline owing to habitat loss in Maine, USA. We estimated contemporary spatial patterns in marten density relative to landscape disturbance with spatial capture-recapture models. We compared current density estimates to historical density calculations to evaluate population decline, and compared historical calculations to habitat-based model predictions to evaluate the efficacy of habitat monitoring as a proxy for direct monitoring. Marten density was negatively associated with the proportion of surrounding regenerating forest, and point estimates within focal townships were 50–80% lower than historical calculations. Habitat-based hindcasts of marten density across our entire focal area interest suggested a smaller population decline (roughly 50%) within our focal area. Thus, although habitat-based predictions underpredicted marten decline, they provided correct directional inference. Habitat monitoring and predictions from species distribution models may provide useful inference about population changes given trends in habitat at limited expense when in-situ information is lacking.