Know Today, Know Tomorrow: Ensemble Nowcasting of Bear Encounter Risk from Sighting Time Series

Managing near-term risks from human-carnivore encounters has traditionally relied on mechanistic models that require extensive real-time data on causal factors, offering limited support for operational decision-making when short-term predictions are needed. We developed a decision-support system that predicts monthly bear sightings from the start of each month by exploiting temporal dynamics without mechanistic assumptions. The ensemble integrates three components: sequential estimation via non-stationary Poisson processes, seasonal baselines with ratio corrections, and rule-based transitions as data accumulate. Applied to Asiatic black bear (Ursus thibetanus) sighting records from two Japanese regions differing 18-fold in encounter frequency (maximum monthly counts: 83 vs. 1490) and with contrasting seasonal peaks, the ensemble achieved correlations ≥0.75 between predicted and observed totals from day 1, rising to ≥0.96 by day 20, and substantially outperformed a null model. After controlling for baseline spatial and temporal risk, we detected localized short-term clustering: prior sightings increased encounter probability within 500 m for up to 3 days. This system demonstrates that temporal dynamics alone can approach practical prediction limits for wildlife encounters without any bear-specific covariates or detailed environmental predictors, and can be directly adapted to other wildlife conflicts and short-term environmental hazards wherever incident time series are available. By quantifying when (daily risk levels) and where (localized hotspots) encounters are most likely, it provides wildlife managers and residents with an immediately implementable tool for issuing targeted warnings, deploying patrols, and reducing human injuries in regions experiencing increasing human-wildlife conflict.