Violence Forecasts
See hyperlocal (neighborhood/village) risk months ahead to pre-position aid, adjust access plans, and warn communities, while replacing broad “red zones” with precise grid-cell probabilities that cut false alarms, focus scarce resources, and defend decisions with transparent analytics.
What It is & Why It Matters
Our forecasts, updated monthly, predict locations and counts of battles, explosions/remote violence, protests/riots, acts of violence against civilians (VAC), VAC fatalities, and looting/property destruction incidents. We use industry-leading data (including country-specific custom datasets) and state-of-the-art forecasting technology. Dynamic Threat Mapper™ allows users to pinpoint outlier locations to relocate people and assets.
Operational decisions happen on routes, compounds, clinics, mine sites, and last-mile links—not at broad administrative polygons. Crisis Forecast delivers map-ready projections you can act on to achieve your objectives and protection.
Core Features
Predictions Made Useful
• Forecasts include 90%, 95%, and 99% confidence intervals—No guessing about how certain our predictions are.
• Stress-test quantile layers by grid cell and month—P50 (Baseline), P90–P95 (Adverse), P99–P99.5 (Extreme)—for budgeting, contingency planning, and solvency. Additional quantiles available upon request.
Explainability
• Explainable drivers: Feature importance highlights which factors and patterns at multiple spatial scales push risk up or down.
Dynamic Threat Mapper™
• Safe ↔ Danger Zone transitions: Identify boundaries between markedly different risk areas and locate nearby lower-risk corridors adjacent to hotspots—configurable by horizon and target.
• Rapid risk change views: Flag grid cells and clusters where risk is accelerating or easing over user-defined periods, enabling early action on emerging trends.
Workflow highlights
• Clear maps: Hybrid satellite basemap layer + labels make forecasted grid cell locations easy to identify.
• Intelligent comparisons: Hover or select single/multiple cells to compare forecasts with historical baselines across multiple time frames.
• Operational alerts: Create threshold- and condition-based alerts (e.g., notify when a selected area crosses a chosen quantile or shows a sustained increase).