Bear Activity Prediction Methodology
Learn how Kumamap calculates bear predictions using exponential decay. Understand our 2km grid risk methodology for hiking safety across Japan.
Our risk prediction model uses exponential decay to estimate relative risk based on how recently and how close bear activity has occurred. This page explains the methodology behind our bear activity predictions.
Grid-Based Data Collection
Japan is divided into 2km x 2km grid cells. Each cell tracks historical bear incidents from the past year (128,000+ incidents in our database). This granular approach allows us to identify local patterns while keeping file sizes manageable.
2km x 2km grid showing incident distribution
Exponential Decay Algorithm
We use exponential decay to model how risk decreases over time and distance. Recent, nearby activity contributes more to risk than older or distant activity. The decay uses a 60-day half-life for time (activity 60 days ago has ~37% weight) and a 5km half-life for distance (activity 5km away has ~37% weight).
Single Sighting
Risk Spread (Today)
Central sighting spreads risk to surrounding cells based on distance
Risk Score Calculation
- • Temporal decay: Activity today = 1.0, 30 days ago = 0.61, 60 days ago = 0.37
- • Spatial decay: At the location = 1.0, 3km away = 0.55, 5km away = 0.37
- • Combined score: Both factors multiply together (e.g., 30 days ago + 3km = 0.61 x 0.55 = 0.34)
- • Maximum aggregation: We use the highest score from any nearby sighting, not the sum
Risk Level Assignment
Each cell receives a continuous risk score from 0 to 1. Scores below 0.2 are considered safe and not displayed. Higher scores indicate greater risk from recent nearby activity.
Limitations
- This is a statistical model based on patterns, not a trained ML model
- Predictions reflect historical patterns, not real-time bear locations
- Areas without colored cells have no significant recent bear activity data
- Always exercise caution in bear country regardless of predictions