WeatherNext is the AI weather model that predicted a Category-5 Caribbean hurricane five days in advance — and the same capability would change the calculus for every cyclone-prone coastline and food-bowl in Africa.
On 19 May 2026, Google DeepMind detailed how WeatherNext helped the United States National Hurricane Center predict the historic landfall of Hurricane Melissa in Jamaica in October 2025. Melissa hit Jamaica as the strongest hurricane on record to make landfall on the island, and tied for the strongest Atlantic hurricane ever recorded. WeatherNext forecast the Category-5 landfall five days in advance with 80% confidence, and close to 100% confidence three days out — the first time any model successfully predicted a Category-5 storm from initial wind speeds at Category 1. The National Hurricane Center's 2025 annual verification report named WeatherNext the top individual model for both track and intensity prediction across the season. The lesson is global, and the African coastline is one of the places where the lesson lands hardest.
What WeatherNext actually did differently
Traditional weather forecasting carries a long, structural weakness: physics-based models tend to be strong on track or strong on intensity, but rarely both at once. A model that predicts where a storm will go often misjudges how strong it will be by the time it arrives, and vice versa. The mismatch matters most for fast-intensifying storms — the ones where a Category-1 hurricane on Tuesday becomes a Category-5 catastrophe by Thursday. Melissa was exactly that kind of storm.
WeatherNext is an AI weather model trained on decades of global weather data and specialised extreme-tropical-cyclone datasets. The model runs ensemble forecasting — 50 "what-if" simulations of how the storm could evolve — and produces probabilistic risk assessments alongside deterministic track-and-intensity predictions. Where the physics models hedged on Melissa's intensity at landfall, WeatherNext's ensemble showed a clear high-probability path to Category-5 conditions days ahead of arrival. The Jamaican Meteorological Service, the United States National Hurricane Center, and the United States National Oceanic and Atmospheric Administration all relied on the WeatherNext forecast in the days before landfall.
Tropical storms and hurricanes can change very quickly in terms of their structure and their intensity, which makes them more challenging to predict than other types of weather systems.
— Michael Brennan, Director, US National Hurricane Center (https://deepmind.google/blog/how-weathernext-helped-the-national-hurricane-center-better-predict-hurricane-melissas-historic-landfall-in-jamaica/)
With early evacuation and better preparation, that reduction in harm really does make a difference to our people. It does actually save their lives.
— Evan Thompson, Meteorological Service Jamaica (https://deepmind.google/blog/how-weathernext-helped-the-national-hurricane-center-better-predict-hurricane-melissas-historic-landfall-in-jamaica/)
Why this lands hardest on Africa
Africa is the continent where weather forecasting matters most and is funded worst. Three connected facts make the point. First, the eastern and southern African coastlines — Mozambique, Madagascar, Tanzania, the Comoros, and South Africa's KwaZulu-Natal — are increasingly cyclone-prone as the southern Indian Ocean warms. Cyclone Idai in 2019 killed more than a thousand people across Mozambique, Zimbabwe, and Malawi; Cyclone Freddy in 2023 became the longest-lasting tropical cyclone on record and killed at least 1,400 across the same region. The next Idai-class storm is not a possibility; it is a scheduling question.
Second, the African agricultural belts — the Sahel, the maize belt across southern Africa, the cocoa belt across West Africa, the tea belt across East Africa — run on weather windows measured in days, not weeks. A two-week-ahead forecast of an early rainy season, a late dry spell, or an out-of-season storm is the difference between a harvest and a hungry season for tens of millions of smallholder farmers. Existing forecast resolution across most African meteorological services is too coarse to deliver that decision-grade information at the smallholder scale.
Third, the existing weather infrastructure across most African states is under-resourced. South Africa has SAWS; Kenya has KMD; a few other strong national services exist. Most countries depend on regional cooperation through ACMAD in Niamey and partnerships with the World Meteorological Organisation. None of these services have the compute or the staffing to run a WeatherNext-class model independently. The question is whether they can access one.
💡AI weather is climate justice
Hurricane Melissa is the Jamaican lesson. The Idai, Freddy, and the next-cyclone seasons are the southern African lesson. The smallholder farmer in Solwezi reading the rains is the Zambian lesson. AI weather forecasting is climate-justice infrastructure first, technology second.
What an African deployment would look like
WeatherNext is research; the operational version flows through Google's partnerships with national meteorological services. DeepMind already lists collaborations with the Philippines PAGASA, Taiwan CWA, Indonesia BMKG, and Vietnam VNMHA — a Pacific-and-Southeast-Asia cluster, plus the Jamaican Meteorological Service in the Caribbean. The African coastline is not in the announcement. The absence is consistent with the broader pattern visible in OpenAI for Singapore and Malta — sovereign-AI relationships are being signed where the diplomatic infrastructure already exists, and African ministries are not yet in the room at the speed the technology is moving.
The right structure for an African deployment is a regional one. SADC, the East African Community, ECOWAS, and the African Union have institutional weight that no individual African state can match alone. A continental partnership with DeepMind — or with any frontier AI weather lab — would bundle compute access, training data, and model deployment under terms that protect African data sovereignty and serve African forecasting priorities first. The model is the easy part; the institutional architecture is the slow part. Building it now, before the next Cyclone Idai, is the difference between a weather forecast and a delayed obituary.
There is a smaller, faster step available. African meteorological services can begin requesting WeatherNext access today, the same way the Caribbean and Pacific services already do. Cross-border data-sharing agreements through ACMAD already exist; layering WeatherNext access on top of those agreements does not require renegotiating the institutional foundation, only the data-and-compute terms. The work is unglamorous; the work is also operationally close to ready.
The agricultural dividend
Cyclone forecasting is the headline; the larger dividend lives in agriculture. African smallholder farmers feed roughly 70% of the continent on weather windows that get less reliable every year. The Sahelian rains arrive later and end earlier. Southern African mid-season dry spells extend further into what used to be reliable harvest months. East African flash floods catch farmers between planting and weeding. A model that gives a farmer in Solwezi, Kano, or Bahir Dar a fortnight's warning that the rains will fail is information that pays for itself in tonnes of maize, sorghum, and teff.
The infrastructure to deliver that information already exists in pieces. Mobile-phone penetration is high; vernacular-language voice messaging works at the village level; agricultural extension officers already run weekly visits in most strong agricultural regions. Bolting decision-grade AI forecasts onto that distribution chain is a project measured in months and millions of dollars, not years and billions. The pieces are operationally close together; what is missing is the convening will to put them in the same room. Read alongside Minerals for Lives — Zambia and the PEPFAR Bargain — the same dignity-first frame applies, and the same negotiating leverage is available if the African political class decides to pick it up.
Frequently Asked Questions
These are the questions African policy leaders, agricultural development professionals, and climate-justice readers have been asking since the WeatherNext announcement. Short answers follow, drawn from the DeepMind launch and the corroborating verification reports from the US National Hurricane Center.
What is WeatherNext?
In short, WeatherNext is an AI weather forecasting model from Google DeepMind and Google Research that predicts both tropical-cyclone tracks and intensities through ensemble forecasting trained on decades of global weather data. The answer, simply put, is the first model that handles both track and intensity well enough to be trusted on rapidly intensifying storms. The key is that the model uses 50 "what-if" scenarios to produce probabilistic risk assessments, which is how it forecast Hurricane Melissa's Category-5 landfall five days in advance with 80% confidence.
How did WeatherNext perform on Hurricane Melissa?
According to the National Hurricane Center's 2025 annual verification report, WeatherNext was the top individual model for both track and intensity prediction during the 2025 Atlantic hurricane season. Research from DeepMind's announcement shows the model forecast Melissa's Category-5 landfall in Jamaica five days in advance with 80% confidence and close to 100% confidence three days before landfall. Data from the verification report reveals this was the first successful prediction of a storm reaching Category-5 strength from initial wind speeds at Category 1.
Why does this matter for African coastlines and farmers?
African coastlines from Mozambique to KwaZulu-Natal are increasingly cyclone-prone, and Cyclones Idai (2019) and Freddy (2023) demonstrated the human cost of late or inaccurate forecasting. Analysis of African agricultural systems shows smallholder farmers across the Sahel, the southern African maize belt, and the East African highlands depend on weather windows measured in days. Evidence from the Caribbean and Pacific deployments of WeatherNext demonstrates that decision-grade forecasts cut harm meaningfully when paired with local meteorological services and effective public communication. The continent's exposure is high and the existing infrastructure is under-resourced.
Who is in the WeatherNext partner cohort today?
DeepMind lists collaborations with the United States National Hurricane Center, NOAA, the Meteorological Service Jamaica, the Philippines PAGASA, Taiwan's CWA, Indonesia's BMKG, and Vietnam's VNMHA. In other words, the cohort is concentrated in the Pacific, Southeast Asia, and the Caribbean. No African national meteorological service is named in the announcement, which is the conspicuous absence to fix.
What are the real risks of AI weather forecasting for vulnerable countries?
Analysis of AI-weather deployment reveals three durable risks. First, the data-sovereignty risk: training a model on a country's historical weather and atmospheric data without explicit terms can foreclose later sovereign options. Second, the dependency risk: a national meteorological service that depends on a frontier-lab model loses domestic capacity over time and faces real exposure if the partner withdraws. Third, the access-inequality risk: the partnership pattern visible today rewards countries with existing diplomatic infrastructure, leaving the most exposed populations the last to benefit. Each risk is structural, not cosmetic, and each is addressable through regional negotiation rather than bilateral haste.
