Age is not Habitat Persistence: Forecasting Forest-Habitat Support to Prioritize Biodiversity Resurveys in the Atlantic Forest

Historical biodiversity records guide forest restoration and environmental safeguards, but their continued validity is often inferred from record age rather than from observed habitat persistence. Existing age-based and binary land-cover heuristics can misrepresent site condition because they ignore recent habitat trajectories and are sensitive to the spatial scale of evaluation. We present an uncertainty-aware framework that links historical plant occurrences in the Atlantic Forest with annual land-cover trajectories to estimate habitat persistence and forecast near-term habitat loss. Positional uncertainty is propagated through Monte Carlo sampling, while LightGBM and logistic regression are evaluated with spatially blocked rolling-origin validation and a held-out temporal test. We show that a commonly used age-decay assumption understates observed habitat persistence, yet correcting it does not make record age a useful individual risk signal. Instead, current habitat state and recent dynamics provide substantially stronger predictive information, while the simpler logistic model remains competitive with gradient boosting for budgeted field decisions. The resulting ranking enriches confirmed habitat losses relative to random site selection, providing a practical way to direct limited biodiversity resurvey effort toward sites where field verification is most informative.