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

Area of Science:

• Ecology
• Wildlife Management
• Predictive Modeling

Background:

• Traditional human-carnivore encounter management relies on mechanistic models needing extensive real-time data.
• These models offer limited operational decision-making support for short-term predictions.
• There is a need for effective, data-efficient systems for managing wildlife conflict risks.

Purpose of the Study:

• To develop and validate a decision-support system for predicting wildlife encounters using temporal dynamics.
• To assess the system's performance in predicting Asiatic black bear sightings in Japan.
• To demonstrate the system's adaptability for other wildlife conflicts and environmental hazards.

Main Methods:

• Developed an ensemble prediction system integrating sequential estimation (non-stationary Poisson processes), seasonal baselines with ratio corrections, and rule-based transitions.
• Applied the system to Asiatic black bear (Ursus thibetanus) sighting records from two Japanese regions with varying encounter frequencies.
• Utilized temporal dynamics without requiring mechanistic assumptions or extensive bear-specific covariates.

Main Results:

• The ensemble system achieved high correlations (≥0.75 from day 1, ≥0.96 by day 20) between predicted and observed monthly bear sightings.
• The system substantially outperformed a null model in prediction accuracy.
• Detected localized short-term clustering of bear sightings, with prior sightings increasing encounter probability within 500m for up to 3 days.

Conclusions:

• Temporal dynamics alone can achieve practical prediction limits for wildlife encounters, reducing reliance on complex mechanistic models.
• The developed system provides an implementable tool for wildlife managers to issue targeted warnings and reduce human-wildlife conflict.
• The system's adaptability to other incident time series makes it valuable for diverse environmental hazard prediction.