NinoWatch

NiñoWatch helps protect vulnerable communities from El Niño’s impacts by turning climate forecasts into local languages, verifying disaster news against independent evidence, and giving local broadcasters ready-to-air radio scripts and SMS alerts. Live at ninowatch.xyz. THE PROBLEM NOAA’s September 2026 outlook gave a greater than 90% chance of a very strong El Niño this fall and winter. The University of Chicago’s Climate Impact Lab projects about 451,000 additional heat deaths worldwide through February 2027, with Nigeria facing the highest national toll (31,400). These are modeled projections, not counts. The people most at risk, such as outdoor workers, smallholder farmers and rural communities, rarely see forecasts. They rely on local radio and chat-app forwards, where rumors spread, and heat deaths are poorly tracked. WHAT IT DOES Globe and region map: a spinning, zoomable globe shows risk across all configured regions. Selecting a region opens a heat map, a summary, and a region map with a legend for events, temperature, and variance from the average of previous years. Verified reports: local news is ingested, events are extracted, duplicates are clustered, and each report is labeled Confirmed, Corroborated, or Unverified from independent outlets, official statements, and weather plausibility. Rules in code decide the label, never the model. Evidence is shown with every label. Local languages: advisories are written once from verified facts, translated, and cached. Only languages a fluent reader has checked are offered, and machine translation is labeled. Broadcaster pack: a 30-second radio script and an SMS alert per advisory, so outlets can reach people with no internet. HOW WE USE META’S MODELS Muse Spark through the Meta Model API extracts events from local-language articles, matches reports to official statements, writes advisories, translates, and drafts the radio and SMS text. Specificaly, Muse Spark Open was used. STACK Next.js, TypeScript, Tailwind, TanStack Query, react-globe.gl; Python, FastAPI, Celery, Redis, PostgreSQL, Docker. WHAT’S LIVE The backend runs live with no mock data, and all 6 regions are configured: e.g. Nigeria and the Sahel, East Africa, Southeast Asia and others. LIMITS Heat-death figures are modeled projections. Labels show strength of evidence, not certainty. Machine-translated text is labeled. A report stays Unverified until independent evidence arrives. NEXT Community report submissions with a review queue, more languages after native review, and other hazards through the same region configuration.