YonSci/forecast2action-ai

Forecast2Action AI An AI-powered early warning and anticipatory action platform for climate-driven disasters in Ethiopia. Forecast2Action AI transforms seasonal and subseasonal climate forecasts into actionable, ranked priority areas for disaster risk management and humanitarian response. # Forecast2Action AI **Forecast2Action AI** is an impact-based early warning and anticipatory action in Ethiopia. It turns a seasonal/subseasonal climate forecast into a ranked list of priority areas, an AI-generated integrated risk report, and audience-specific advisories (farmer, agro-pastoral, humanitarian, SMS/WhatsApp) with every number in the report computed deterministically before any AI model sees it. ## Live Demo - Frontend: forecast2action-ai.vercel.a… - Backend API: forecast2action-ai.onrender… - API docs (OpenAPI/Swagger): forecast2action-ai.onrender… --- ## 1. What it does Forecast2Action AI covers Ethiopia at the national scale (admin1/admin2/admin3), across seasonal and subseasonal forecast windows. For any selected period it: - computes hazard indices, exposure, and vulnerability per region from real, named datasets; - ranks regions into a priority queue using a deterministic, auditable formula; - classifies each region into an alert level (Trigger / Warning / Watch / No alert); - generates an AI-interpreted integrated risk report executive summary, national spatial overview, compound hazard interpretation, and per-area justifications; - produces audience-specific advisories (Farmer, Agro-pastoral, Humanitarian) and SMS/WhatsApp-ready community messages in English, Amharic, Oromifa/Afaan Oromo, Tigrinya, and Somali; - accepts community ground-truth reports that corroborate or contradict the forecast signal; - tracks recommended actions through to implementation, with CSV export; - answers follow-up questions through a grounded, streaming dashboard chat assistant (tuned for Disaster manager, Extension officer, NGO planner, or General audiences) that only cites real, already-computed evidence, never a freeform re-query of the model; - exports a self-contained HTML bulletin containing real climate-indicator, risk-layer, and priority-area maps (each with its own colorbar/legend and region boundaries) alongside …