Google WeatherNext 3 Brings Hourly, High-Resolution AI Forecasts to Search, Maps and Cloud

September 4, 2026

Google DeepMind has announced WeatherNext 3, a new artificial intelligence weather forecasting system that produces hourly predictions at substantially higher resolution. The model is designed to improve the weather information shown across Google Search, Gemini and Maps while giving developers and businesses access through Google Maps Platform and Google Cloud.

The release demonstrates a growing role for AI in scientific computing. Instead of relying only on traditional numerical simulations, modern forecasting systems can learn patterns from enormous historical datasets and generate predictions more quickly.

WeatherNext 3: key improvements

• Hourly forecasts powered by live satellite observations.

• Temperature and moisture predictions at roughly five-kilometer resolution.

• Other surface variables at approximately 10-kilometer resolution.

• Atmospheric variables at about 25-kilometer resolution.

• Google says the forecasts are roughly five times sharper than WeatherNext 2.

How WeatherNext 3 works

WeatherNext 3 incorporates current satellite data into an AI forecasting process. Live observations help the system estimate the present state of the atmosphere before predicting how conditions will develop hour by hour.

Traditional weather prediction solves complex physical equations on supercomputers. AI models learn statistical relationships from historical observations and forecasts, allowing them to generate predictions with different computational tradeoffs. The two approaches are increasingly complementary: AI can accelerate forecasting and expand the number of scenarios, while physics-based systems remain essential for scientific validation and operational resilience.

Why higher resolution matters

Weather can vary dramatically across short distances, especially near coastlines, mountains and urban areas. A sharper forecast grid can better represent local temperature, humidity and surface conditions that are smoothed out at lower resolution.

Google says WeatherNext 3 predicts temperature and moisture on a grid of about five kilometers, other surface conditions at around 10 kilometers and atmospheric variables at approximately 25 kilometers. Compared with WeatherNext 2, the output is roughly five times sharper.

Resolution alone does not guarantee accuracy. Forecast quality also depends on observations, training data, model design and the specific weather event. Local warnings from national meteorological agencies remain the authoritative source for emergency decisions.

Hourly AI forecasts across Google products

WeatherNext 3 will power information across GHow businesses can use probabilistic forecasts

The most valuable weather product is rarely a single prediction. Businesses need probabilities and alternative scenarios that connect to a decision. A logistics operator may reroute only when the likelihood of disruption crosses a threshold, while an energy company may prepare reserves across several possible temperature outcomes.

AI systems can generate larger forecast ensembles more efficiently, helping users estimate uncertainty instead of relying on one deterministic path. Applications should expose that uncertainty clearly and let customers choose thresholds appropriate to their risk.

Weather data for renewable energy

Solar and wind generation depend directly on atmospheric conditions. More frequent forecasts can help grid operators estimate supply, balance demand and schedule storage. Higher spatial resolution may be useful where cloud cover, terrain and coastal winds change quickly across short distances.

Forecast improvement does not remove the need for sensors and operational expertise. Energy planners should compare predictions with local measurements and maintain reserves for unexpected changes. Reliability comes from combining multiple signals.

Agriculture, insurance and local planning

Farmers could use hourly temperature, moisture and precipitation information to schedule irrigation, spraying and harvesting. Insurers may use improved historical and forecast data to understand exposure, although automated predictions should not become the sole basis for consequential coverage decisions.

Cities can combine forecasts with drainage, heat and transportation data. Local authorities might prepare cooling centers, adjust staffing or warn residents sooner. These uses require accessible communication because the people at greatest risk may not use advanced weather applications.

How developers should validate WeatherNext 3

A good evaluation compares the model with local observations and existing operational forecasts over time. Developers should score different variables, lead times, regions and event types rather than reporting one average. Extreme precipitation, tropical systems and rapidly developing storms deserve separate analysis.

Applications also need fallback providers and clear timestamps. A high-resolution forecast can be misleading if the latest update failed or the underlying observation stream is delayed. Users should always know when data was produced and which source supports it.

The relationship with public weather agencies

National meteorological services operate observing networks, issue official warnings and provide expert interpretation. Commercial AI forecasts should complement this public infrastructure. During dangerous events, applications should foreground authoritative alerts rather than creating competing emergency messages.

Collaboration can also improve science. Shared evaluation standards and transparent error analysis help researchers understand where machine learning adds value and where physics-based models remain stronger.

What to watch next

The key questions are accuracy during rare extremes, geographic consistency and the quality of uncertainty information. Developers will also watch API pricing, update reliability and the ease of combining WeatherNext with business data.

If those elements perform well, WeatherNext 3 could make sophisticated forecasting accessible to many more organizations. Its impact will be measured not by resolution alone, but by better decisions made before weather creates harm.oogle Search, Gemini and Maps. This wide distribution could make higher-resolution forecasts visible to consumers without requiring a specialist weather application.

Maps integration is particularly relevant for travel and location-based planning. Better hourly information may help people choose departure times, prepare for heat or rain and understand conditions around a destination. Gemini could also use forecast data in conversational planning, although users should verify high-impact advice.

Business and developer use cases

Google is making WeatherNext 3 available through Google Maps Platform and Google Cloud. That opens potential use cases in logistics, energy, agriculture, insurance, retail, construction and outdoor events.

A delivery company might combine hourly forecasts with route planning. Energy operators could estimate demand or renewable generation. Retailers could adjust inventory for extreme heat or heavy rain, while construction managers could schedule weather-sensitive work.

Developers should evaluate geographic coverage, update frequency, uncertainty information and service-level requirements before building critical operations around the data. Forecasts should be one input in a resilient decision process, not the only signal.

The wider AI weather race

Weather prediction has become one of the clearest examples of AI applied to physical science. Research groups and technology companies are competing to improve forecast speed, resolution and accuracy, while public agencies continue to provide the observations and operational expertise on which many systems depend.

The most useful progress will come from combining these strengths. Faster AI-generated ensembles could help forecasters explore more possible outcomes, while meteorologists interpret uncertainty and communicate risk to the public.

Limitations and responsible use

No forecast can eliminate uncertainty. Rare extremes, rapidly developing storms and gaps in observations remain difficult. Users should consult official warnings during severe weather, and businesses should design contingency plans for incorrect or delayed predictions.

Providers should also communicate confidence clearly. A precise-looking local forecast can create false certainty if the underlying probability is not visible. Transparent evaluation across regions and weather types will be crucial as AI forecasts become more common.

The bottom line

WeatherNext 3 brings AI weather forecasting closer to everyday decision-making. Its hourly updates and higher spatial resolution could improve consumer services and unlock new business tools. The largest benefit will come when faster AI predictions are paired with reliable observations, transparent uncertainty and expert meteorological judgment.

Source: Google DeepMind WeatherNext 3 announcement